]>
Commit | Line | Data |
---|---|---|
51580e79 | 1 | /* Copyright (c) 2011-2014 PLUMgrid, http://plumgrid.com |
969bf05e | 2 | * Copyright (c) 2016 Facebook |
51580e79 AS |
3 | * |
4 | * This program is free software; you can redistribute it and/or | |
5 | * modify it under the terms of version 2 of the GNU General Public | |
6 | * License as published by the Free Software Foundation. | |
7 | * | |
8 | * This program is distributed in the hope that it will be useful, but | |
9 | * WITHOUT ANY WARRANTY; without even the implied warranty of | |
10 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
11 | * General Public License for more details. | |
12 | */ | |
13 | #include <linux/kernel.h> | |
14 | #include <linux/types.h> | |
15 | #include <linux/slab.h> | |
16 | #include <linux/bpf.h> | |
58e2af8b | 17 | #include <linux/bpf_verifier.h> |
51580e79 AS |
18 | #include <linux/filter.h> |
19 | #include <net/netlink.h> | |
20 | #include <linux/file.h> | |
21 | #include <linux/vmalloc.h> | |
ebb676da | 22 | #include <linux/stringify.h> |
51580e79 AS |
23 | |
24 | /* bpf_check() is a static code analyzer that walks eBPF program | |
25 | * instruction by instruction and updates register/stack state. | |
26 | * All paths of conditional branches are analyzed until 'bpf_exit' insn. | |
27 | * | |
28 | * The first pass is depth-first-search to check that the program is a DAG. | |
29 | * It rejects the following programs: | |
30 | * - larger than BPF_MAXINSNS insns | |
31 | * - if loop is present (detected via back-edge) | |
32 | * - unreachable insns exist (shouldn't be a forest. program = one function) | |
33 | * - out of bounds or malformed jumps | |
34 | * The second pass is all possible path descent from the 1st insn. | |
35 | * Since it's analyzing all pathes through the program, the length of the | |
eba38a96 | 36 | * analysis is limited to 64k insn, which may be hit even if total number of |
51580e79 AS |
37 | * insn is less then 4K, but there are too many branches that change stack/regs. |
38 | * Number of 'branches to be analyzed' is limited to 1k | |
39 | * | |
40 | * On entry to each instruction, each register has a type, and the instruction | |
41 | * changes the types of the registers depending on instruction semantics. | |
42 | * If instruction is BPF_MOV64_REG(BPF_REG_1, BPF_REG_5), then type of R5 is | |
43 | * copied to R1. | |
44 | * | |
45 | * All registers are 64-bit. | |
46 | * R0 - return register | |
47 | * R1-R5 argument passing registers | |
48 | * R6-R9 callee saved registers | |
49 | * R10 - frame pointer read-only | |
50 | * | |
51 | * At the start of BPF program the register R1 contains a pointer to bpf_context | |
52 | * and has type PTR_TO_CTX. | |
53 | * | |
54 | * Verifier tracks arithmetic operations on pointers in case: | |
55 | * BPF_MOV64_REG(BPF_REG_1, BPF_REG_10), | |
56 | * BPF_ALU64_IMM(BPF_ADD, BPF_REG_1, -20), | |
57 | * 1st insn copies R10 (which has FRAME_PTR) type into R1 | |
58 | * and 2nd arithmetic instruction is pattern matched to recognize | |
59 | * that it wants to construct a pointer to some element within stack. | |
60 | * So after 2nd insn, the register R1 has type PTR_TO_STACK | |
61 | * (and -20 constant is saved for further stack bounds checking). | |
62 | * Meaning that this reg is a pointer to stack plus known immediate constant. | |
63 | * | |
f1174f77 | 64 | * Most of the time the registers have SCALAR_VALUE type, which |
51580e79 | 65 | * means the register has some value, but it's not a valid pointer. |
f1174f77 | 66 | * (like pointer plus pointer becomes SCALAR_VALUE type) |
51580e79 AS |
67 | * |
68 | * When verifier sees load or store instructions the type of base register | |
f1174f77 | 69 | * can be: PTR_TO_MAP_VALUE, PTR_TO_CTX, PTR_TO_STACK. These are three pointer |
51580e79 AS |
70 | * types recognized by check_mem_access() function. |
71 | * | |
72 | * PTR_TO_MAP_VALUE means that this register is pointing to 'map element value' | |
73 | * and the range of [ptr, ptr + map's value_size) is accessible. | |
74 | * | |
75 | * registers used to pass values to function calls are checked against | |
76 | * function argument constraints. | |
77 | * | |
78 | * ARG_PTR_TO_MAP_KEY is one of such argument constraints. | |
79 | * It means that the register type passed to this function must be | |
80 | * PTR_TO_STACK and it will be used inside the function as | |
81 | * 'pointer to map element key' | |
82 | * | |
83 | * For example the argument constraints for bpf_map_lookup_elem(): | |
84 | * .ret_type = RET_PTR_TO_MAP_VALUE_OR_NULL, | |
85 | * .arg1_type = ARG_CONST_MAP_PTR, | |
86 | * .arg2_type = ARG_PTR_TO_MAP_KEY, | |
87 | * | |
88 | * ret_type says that this function returns 'pointer to map elem value or null' | |
89 | * function expects 1st argument to be a const pointer to 'struct bpf_map' and | |
90 | * 2nd argument should be a pointer to stack, which will be used inside | |
91 | * the helper function as a pointer to map element key. | |
92 | * | |
93 | * On the kernel side the helper function looks like: | |
94 | * u64 bpf_map_lookup_elem(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5) | |
95 | * { | |
96 | * struct bpf_map *map = (struct bpf_map *) (unsigned long) r1; | |
97 | * void *key = (void *) (unsigned long) r2; | |
98 | * void *value; | |
99 | * | |
100 | * here kernel can access 'key' and 'map' pointers safely, knowing that | |
101 | * [key, key + map->key_size) bytes are valid and were initialized on | |
102 | * the stack of eBPF program. | |
103 | * } | |
104 | * | |
105 | * Corresponding eBPF program may look like: | |
106 | * BPF_MOV64_REG(BPF_REG_2, BPF_REG_10), // after this insn R2 type is FRAME_PTR | |
107 | * BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, -4), // after this insn R2 type is PTR_TO_STACK | |
108 | * BPF_LD_MAP_FD(BPF_REG_1, map_fd), // after this insn R1 type is CONST_PTR_TO_MAP | |
109 | * BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0, BPF_FUNC_map_lookup_elem), | |
110 | * here verifier looks at prototype of map_lookup_elem() and sees: | |
111 | * .arg1_type == ARG_CONST_MAP_PTR and R1->type == CONST_PTR_TO_MAP, which is ok, | |
112 | * Now verifier knows that this map has key of R1->map_ptr->key_size bytes | |
113 | * | |
114 | * Then .arg2_type == ARG_PTR_TO_MAP_KEY and R2->type == PTR_TO_STACK, ok so far, | |
115 | * Now verifier checks that [R2, R2 + map's key_size) are within stack limits | |
116 | * and were initialized prior to this call. | |
117 | * If it's ok, then verifier allows this BPF_CALL insn and looks at | |
118 | * .ret_type which is RET_PTR_TO_MAP_VALUE_OR_NULL, so it sets | |
119 | * R0->type = PTR_TO_MAP_VALUE_OR_NULL which means bpf_map_lookup_elem() function | |
120 | * returns ether pointer to map value or NULL. | |
121 | * | |
122 | * When type PTR_TO_MAP_VALUE_OR_NULL passes through 'if (reg != 0) goto +off' | |
123 | * insn, the register holding that pointer in the true branch changes state to | |
124 | * PTR_TO_MAP_VALUE and the same register changes state to CONST_IMM in the false | |
125 | * branch. See check_cond_jmp_op(). | |
126 | * | |
127 | * After the call R0 is set to return type of the function and registers R1-R5 | |
128 | * are set to NOT_INIT to indicate that they are no longer readable. | |
129 | */ | |
130 | ||
17a52670 | 131 | /* verifier_state + insn_idx are pushed to stack when branch is encountered */ |
58e2af8b | 132 | struct bpf_verifier_stack_elem { |
17a52670 AS |
133 | /* verifer state is 'st' |
134 | * before processing instruction 'insn_idx' | |
135 | * and after processing instruction 'prev_insn_idx' | |
136 | */ | |
58e2af8b | 137 | struct bpf_verifier_state st; |
17a52670 AS |
138 | int insn_idx; |
139 | int prev_insn_idx; | |
58e2af8b | 140 | struct bpf_verifier_stack_elem *next; |
cbd35700 AS |
141 | }; |
142 | ||
3c2ce60b | 143 | #define BPF_COMPLEXITY_LIMIT_INSNS 98304 |
07016151 DB |
144 | #define BPF_COMPLEXITY_LIMIT_STACK 1024 |
145 | ||
fad73a1a MKL |
146 | #define BPF_MAP_PTR_POISON ((void *)0xeB9F + POISON_POINTER_DELTA) |
147 | ||
33ff9823 DB |
148 | struct bpf_call_arg_meta { |
149 | struct bpf_map *map_ptr; | |
435faee1 | 150 | bool raw_mode; |
36bbef52 | 151 | bool pkt_access; |
435faee1 DB |
152 | int regno; |
153 | int access_size; | |
33ff9823 DB |
154 | }; |
155 | ||
cbd35700 AS |
156 | /* verbose verifier prints what it's seeing |
157 | * bpf_check() is called under lock, so no race to access these global vars | |
158 | */ | |
159 | static u32 log_level, log_size, log_len; | |
160 | static char *log_buf; | |
161 | ||
162 | static DEFINE_MUTEX(bpf_verifier_lock); | |
163 | ||
164 | /* log_level controls verbosity level of eBPF verifier. | |
165 | * verbose() is used to dump the verification trace to the log, so the user | |
166 | * can figure out what's wrong with the program | |
167 | */ | |
1d056d9c | 168 | static __printf(1, 2) void verbose(const char *fmt, ...) |
cbd35700 AS |
169 | { |
170 | va_list args; | |
171 | ||
172 | if (log_level == 0 || log_len >= log_size - 1) | |
173 | return; | |
174 | ||
175 | va_start(args, fmt); | |
176 | log_len += vscnprintf(log_buf + log_len, log_size - log_len, fmt, args); | |
177 | va_end(args); | |
178 | } | |
179 | ||
17a52670 AS |
180 | /* string representation of 'enum bpf_reg_type' */ |
181 | static const char * const reg_type_str[] = { | |
182 | [NOT_INIT] = "?", | |
f1174f77 | 183 | [SCALAR_VALUE] = "inv", |
17a52670 AS |
184 | [PTR_TO_CTX] = "ctx", |
185 | [CONST_PTR_TO_MAP] = "map_ptr", | |
186 | [PTR_TO_MAP_VALUE] = "map_value", | |
187 | [PTR_TO_MAP_VALUE_OR_NULL] = "map_value_or_null", | |
17a52670 | 188 | [PTR_TO_STACK] = "fp", |
969bf05e AS |
189 | [PTR_TO_PACKET] = "pkt", |
190 | [PTR_TO_PACKET_END] = "pkt_end", | |
17a52670 AS |
191 | }; |
192 | ||
ebb676da TG |
193 | #define __BPF_FUNC_STR_FN(x) [BPF_FUNC_ ## x] = __stringify(bpf_ ## x) |
194 | static const char * const func_id_str[] = { | |
195 | __BPF_FUNC_MAPPER(__BPF_FUNC_STR_FN) | |
196 | }; | |
197 | #undef __BPF_FUNC_STR_FN | |
198 | ||
199 | static const char *func_id_name(int id) | |
200 | { | |
201 | BUILD_BUG_ON(ARRAY_SIZE(func_id_str) != __BPF_FUNC_MAX_ID); | |
202 | ||
203 | if (id >= 0 && id < __BPF_FUNC_MAX_ID && func_id_str[id]) | |
204 | return func_id_str[id]; | |
205 | else | |
206 | return "unknown"; | |
207 | } | |
208 | ||
58e2af8b | 209 | static void print_verifier_state(struct bpf_verifier_state *state) |
17a52670 | 210 | { |
58e2af8b | 211 | struct bpf_reg_state *reg; |
17a52670 AS |
212 | enum bpf_reg_type t; |
213 | int i; | |
214 | ||
215 | for (i = 0; i < MAX_BPF_REG; i++) { | |
1a0dc1ac AS |
216 | reg = &state->regs[i]; |
217 | t = reg->type; | |
17a52670 AS |
218 | if (t == NOT_INIT) |
219 | continue; | |
220 | verbose(" R%d=%s", i, reg_type_str[t]); | |
f1174f77 EC |
221 | if ((t == SCALAR_VALUE || t == PTR_TO_STACK) && |
222 | tnum_is_const(reg->var_off)) { | |
223 | /* reg->off should be 0 for SCALAR_VALUE */ | |
224 | verbose("%lld", reg->var_off.value + reg->off); | |
225 | } else { | |
226 | verbose("(id=%d", reg->id); | |
227 | if (t != SCALAR_VALUE) | |
228 | verbose(",off=%d", reg->off); | |
229 | if (t == PTR_TO_PACKET) | |
230 | verbose(",r=%d", reg->range); | |
231 | else if (t == CONST_PTR_TO_MAP || | |
232 | t == PTR_TO_MAP_VALUE || | |
233 | t == PTR_TO_MAP_VALUE_OR_NULL) | |
234 | verbose(",ks=%d,vs=%d", | |
235 | reg->map_ptr->key_size, | |
236 | reg->map_ptr->value_size); | |
237 | if (reg->min_value != BPF_REGISTER_MIN_RANGE) | |
238 | verbose(",min_value=%lld", | |
239 | (long long)reg->min_value); | |
240 | if (reg->max_value != BPF_REGISTER_MAX_RANGE) | |
241 | verbose(",max_value=%llu", | |
242 | (unsigned long long)reg->max_value); | |
243 | if (!tnum_is_unknown(reg->var_off)) { | |
244 | char tn_buf[48]; | |
245 | ||
246 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); | |
247 | verbose(",var_off=%s", tn_buf); | |
248 | } | |
249 | verbose(")"); | |
250 | } | |
17a52670 | 251 | } |
9c399760 | 252 | for (i = 0; i < MAX_BPF_STACK; i += BPF_REG_SIZE) { |
1a0dc1ac | 253 | if (state->stack_slot_type[i] == STACK_SPILL) |
17a52670 | 254 | verbose(" fp%d=%s", -MAX_BPF_STACK + i, |
1a0dc1ac | 255 | reg_type_str[state->spilled_regs[i / BPF_REG_SIZE].type]); |
17a52670 AS |
256 | } |
257 | verbose("\n"); | |
258 | } | |
259 | ||
cbd35700 AS |
260 | static const char *const bpf_class_string[] = { |
261 | [BPF_LD] = "ld", | |
262 | [BPF_LDX] = "ldx", | |
263 | [BPF_ST] = "st", | |
264 | [BPF_STX] = "stx", | |
265 | [BPF_ALU] = "alu", | |
266 | [BPF_JMP] = "jmp", | |
267 | [BPF_RET] = "BUG", | |
268 | [BPF_ALU64] = "alu64", | |
269 | }; | |
270 | ||
687f0715 | 271 | static const char *const bpf_alu_string[16] = { |
cbd35700 AS |
272 | [BPF_ADD >> 4] = "+=", |
273 | [BPF_SUB >> 4] = "-=", | |
274 | [BPF_MUL >> 4] = "*=", | |
275 | [BPF_DIV >> 4] = "/=", | |
276 | [BPF_OR >> 4] = "|=", | |
277 | [BPF_AND >> 4] = "&=", | |
278 | [BPF_LSH >> 4] = "<<=", | |
279 | [BPF_RSH >> 4] = ">>=", | |
280 | [BPF_NEG >> 4] = "neg", | |
281 | [BPF_MOD >> 4] = "%=", | |
282 | [BPF_XOR >> 4] = "^=", | |
283 | [BPF_MOV >> 4] = "=", | |
284 | [BPF_ARSH >> 4] = "s>>=", | |
285 | [BPF_END >> 4] = "endian", | |
286 | }; | |
287 | ||
288 | static const char *const bpf_ldst_string[] = { | |
289 | [BPF_W >> 3] = "u32", | |
290 | [BPF_H >> 3] = "u16", | |
291 | [BPF_B >> 3] = "u8", | |
292 | [BPF_DW >> 3] = "u64", | |
293 | }; | |
294 | ||
687f0715 | 295 | static const char *const bpf_jmp_string[16] = { |
cbd35700 AS |
296 | [BPF_JA >> 4] = "jmp", |
297 | [BPF_JEQ >> 4] = "==", | |
298 | [BPF_JGT >> 4] = ">", | |
299 | [BPF_JGE >> 4] = ">=", | |
300 | [BPF_JSET >> 4] = "&", | |
301 | [BPF_JNE >> 4] = "!=", | |
302 | [BPF_JSGT >> 4] = "s>", | |
303 | [BPF_JSGE >> 4] = "s>=", | |
304 | [BPF_CALL >> 4] = "call", | |
305 | [BPF_EXIT >> 4] = "exit", | |
306 | }; | |
307 | ||
0d0e5769 DB |
308 | static void print_bpf_insn(const struct bpf_verifier_env *env, |
309 | const struct bpf_insn *insn) | |
cbd35700 AS |
310 | { |
311 | u8 class = BPF_CLASS(insn->code); | |
312 | ||
313 | if (class == BPF_ALU || class == BPF_ALU64) { | |
314 | if (BPF_SRC(insn->code) == BPF_X) | |
315 | verbose("(%02x) %sr%d %s %sr%d\n", | |
316 | insn->code, class == BPF_ALU ? "(u32) " : "", | |
317 | insn->dst_reg, | |
318 | bpf_alu_string[BPF_OP(insn->code) >> 4], | |
319 | class == BPF_ALU ? "(u32) " : "", | |
320 | insn->src_reg); | |
321 | else | |
322 | verbose("(%02x) %sr%d %s %s%d\n", | |
323 | insn->code, class == BPF_ALU ? "(u32) " : "", | |
324 | insn->dst_reg, | |
325 | bpf_alu_string[BPF_OP(insn->code) >> 4], | |
326 | class == BPF_ALU ? "(u32) " : "", | |
327 | insn->imm); | |
328 | } else if (class == BPF_STX) { | |
329 | if (BPF_MODE(insn->code) == BPF_MEM) | |
330 | verbose("(%02x) *(%s *)(r%d %+d) = r%d\n", | |
331 | insn->code, | |
332 | bpf_ldst_string[BPF_SIZE(insn->code) >> 3], | |
333 | insn->dst_reg, | |
334 | insn->off, insn->src_reg); | |
335 | else if (BPF_MODE(insn->code) == BPF_XADD) | |
336 | verbose("(%02x) lock *(%s *)(r%d %+d) += r%d\n", | |
337 | insn->code, | |
338 | bpf_ldst_string[BPF_SIZE(insn->code) >> 3], | |
339 | insn->dst_reg, insn->off, | |
340 | insn->src_reg); | |
341 | else | |
342 | verbose("BUG_%02x\n", insn->code); | |
343 | } else if (class == BPF_ST) { | |
344 | if (BPF_MODE(insn->code) != BPF_MEM) { | |
345 | verbose("BUG_st_%02x\n", insn->code); | |
346 | return; | |
347 | } | |
348 | verbose("(%02x) *(%s *)(r%d %+d) = %d\n", | |
349 | insn->code, | |
350 | bpf_ldst_string[BPF_SIZE(insn->code) >> 3], | |
351 | insn->dst_reg, | |
352 | insn->off, insn->imm); | |
353 | } else if (class == BPF_LDX) { | |
354 | if (BPF_MODE(insn->code) != BPF_MEM) { | |
355 | verbose("BUG_ldx_%02x\n", insn->code); | |
356 | return; | |
357 | } | |
358 | verbose("(%02x) r%d = *(%s *)(r%d %+d)\n", | |
359 | insn->code, insn->dst_reg, | |
360 | bpf_ldst_string[BPF_SIZE(insn->code) >> 3], | |
361 | insn->src_reg, insn->off); | |
362 | } else if (class == BPF_LD) { | |
363 | if (BPF_MODE(insn->code) == BPF_ABS) { | |
364 | verbose("(%02x) r0 = *(%s *)skb[%d]\n", | |
365 | insn->code, | |
366 | bpf_ldst_string[BPF_SIZE(insn->code) >> 3], | |
367 | insn->imm); | |
368 | } else if (BPF_MODE(insn->code) == BPF_IND) { | |
369 | verbose("(%02x) r0 = *(%s *)skb[r%d + %d]\n", | |
370 | insn->code, | |
371 | bpf_ldst_string[BPF_SIZE(insn->code) >> 3], | |
372 | insn->src_reg, insn->imm); | |
0d0e5769 DB |
373 | } else if (BPF_MODE(insn->code) == BPF_IMM && |
374 | BPF_SIZE(insn->code) == BPF_DW) { | |
375 | /* At this point, we already made sure that the second | |
376 | * part of the ldimm64 insn is accessible. | |
377 | */ | |
378 | u64 imm = ((u64)(insn + 1)->imm << 32) | (u32)insn->imm; | |
379 | bool map_ptr = insn->src_reg == BPF_PSEUDO_MAP_FD; | |
380 | ||
381 | if (map_ptr && !env->allow_ptr_leaks) | |
382 | imm = 0; | |
383 | ||
384 | verbose("(%02x) r%d = 0x%llx\n", insn->code, | |
385 | insn->dst_reg, (unsigned long long)imm); | |
cbd35700 AS |
386 | } else { |
387 | verbose("BUG_ld_%02x\n", insn->code); | |
388 | return; | |
389 | } | |
390 | } else if (class == BPF_JMP) { | |
391 | u8 opcode = BPF_OP(insn->code); | |
392 | ||
393 | if (opcode == BPF_CALL) { | |
ebb676da TG |
394 | verbose("(%02x) call %s#%d\n", insn->code, |
395 | func_id_name(insn->imm), insn->imm); | |
cbd35700 AS |
396 | } else if (insn->code == (BPF_JMP | BPF_JA)) { |
397 | verbose("(%02x) goto pc%+d\n", | |
398 | insn->code, insn->off); | |
399 | } else if (insn->code == (BPF_JMP | BPF_EXIT)) { | |
400 | verbose("(%02x) exit\n", insn->code); | |
401 | } else if (BPF_SRC(insn->code) == BPF_X) { | |
402 | verbose("(%02x) if r%d %s r%d goto pc%+d\n", | |
403 | insn->code, insn->dst_reg, | |
404 | bpf_jmp_string[BPF_OP(insn->code) >> 4], | |
405 | insn->src_reg, insn->off); | |
406 | } else { | |
407 | verbose("(%02x) if r%d %s 0x%x goto pc%+d\n", | |
408 | insn->code, insn->dst_reg, | |
409 | bpf_jmp_string[BPF_OP(insn->code) >> 4], | |
410 | insn->imm, insn->off); | |
411 | } | |
412 | } else { | |
413 | verbose("(%02x) %s\n", insn->code, bpf_class_string[class]); | |
414 | } | |
415 | } | |
416 | ||
58e2af8b | 417 | static int pop_stack(struct bpf_verifier_env *env, int *prev_insn_idx) |
17a52670 | 418 | { |
58e2af8b | 419 | struct bpf_verifier_stack_elem *elem; |
17a52670 AS |
420 | int insn_idx; |
421 | ||
422 | if (env->head == NULL) | |
423 | return -1; | |
424 | ||
425 | memcpy(&env->cur_state, &env->head->st, sizeof(env->cur_state)); | |
426 | insn_idx = env->head->insn_idx; | |
427 | if (prev_insn_idx) | |
428 | *prev_insn_idx = env->head->prev_insn_idx; | |
429 | elem = env->head->next; | |
430 | kfree(env->head); | |
431 | env->head = elem; | |
432 | env->stack_size--; | |
433 | return insn_idx; | |
434 | } | |
435 | ||
58e2af8b JK |
436 | static struct bpf_verifier_state *push_stack(struct bpf_verifier_env *env, |
437 | int insn_idx, int prev_insn_idx) | |
17a52670 | 438 | { |
58e2af8b | 439 | struct bpf_verifier_stack_elem *elem; |
17a52670 | 440 | |
58e2af8b | 441 | elem = kmalloc(sizeof(struct bpf_verifier_stack_elem), GFP_KERNEL); |
17a52670 AS |
442 | if (!elem) |
443 | goto err; | |
444 | ||
445 | memcpy(&elem->st, &env->cur_state, sizeof(env->cur_state)); | |
446 | elem->insn_idx = insn_idx; | |
447 | elem->prev_insn_idx = prev_insn_idx; | |
448 | elem->next = env->head; | |
449 | env->head = elem; | |
450 | env->stack_size++; | |
07016151 | 451 | if (env->stack_size > BPF_COMPLEXITY_LIMIT_STACK) { |
17a52670 AS |
452 | verbose("BPF program is too complex\n"); |
453 | goto err; | |
454 | } | |
455 | return &elem->st; | |
456 | err: | |
457 | /* pop all elements and return */ | |
458 | while (pop_stack(env, NULL) >= 0); | |
459 | return NULL; | |
460 | } | |
461 | ||
462 | #define CALLER_SAVED_REGS 6 | |
463 | static const int caller_saved[CALLER_SAVED_REGS] = { | |
464 | BPF_REG_0, BPF_REG_1, BPF_REG_2, BPF_REG_3, BPF_REG_4, BPF_REG_5 | |
465 | }; | |
466 | ||
f1174f77 EC |
467 | static void __mark_reg_not_init(struct bpf_reg_state *reg); |
468 | ||
469 | /* Mark the 'variable offset' part of a register as zero. This should be | |
470 | * used only on registers holding a pointer type. | |
471 | */ | |
472 | static void __mark_reg_known_zero(struct bpf_reg_state *reg) | |
a9789ef9 | 473 | { |
f1174f77 EC |
474 | reg->var_off = tnum_const(0); |
475 | reg->min_value = 0; | |
476 | reg->max_value = 0; | |
477 | } | |
a9789ef9 | 478 | |
f1174f77 EC |
479 | static void mark_reg_known_zero(struct bpf_reg_state *regs, u32 regno) |
480 | { | |
481 | if (WARN_ON(regno >= MAX_BPF_REG)) { | |
482 | verbose("mark_reg_known_zero(regs, %u)\n", regno); | |
483 | /* Something bad happened, let's kill all regs */ | |
484 | for (regno = 0; regno < MAX_BPF_REG; regno++) | |
485 | __mark_reg_not_init(regs + regno); | |
486 | return; | |
487 | } | |
488 | __mark_reg_known_zero(regs + regno); | |
489 | } | |
490 | ||
491 | /* Mark a register as having a completely unknown (scalar) value. */ | |
492 | static void __mark_reg_unknown(struct bpf_reg_state *reg) | |
493 | { | |
494 | reg->type = SCALAR_VALUE; | |
495 | reg->id = 0; | |
496 | reg->off = 0; | |
497 | reg->var_off = tnum_unknown; | |
498 | reg->min_value = BPF_REGISTER_MIN_RANGE; | |
499 | reg->max_value = BPF_REGISTER_MAX_RANGE; | |
500 | } | |
501 | ||
502 | static void mark_reg_unknown(struct bpf_reg_state *regs, u32 regno) | |
503 | { | |
504 | if (WARN_ON(regno >= MAX_BPF_REG)) { | |
505 | verbose("mark_reg_unknown(regs, %u)\n", regno); | |
506 | /* Something bad happened, let's kill all regs */ | |
507 | for (regno = 0; regno < MAX_BPF_REG; regno++) | |
508 | __mark_reg_not_init(regs + regno); | |
509 | return; | |
510 | } | |
511 | __mark_reg_unknown(regs + regno); | |
512 | } | |
513 | ||
514 | static void __mark_reg_not_init(struct bpf_reg_state *reg) | |
515 | { | |
516 | __mark_reg_unknown(reg); | |
517 | reg->type = NOT_INIT; | |
518 | } | |
519 | ||
520 | static void mark_reg_not_init(struct bpf_reg_state *regs, u32 regno) | |
521 | { | |
522 | if (WARN_ON(regno >= MAX_BPF_REG)) { | |
523 | verbose("mark_reg_not_init(regs, %u)\n", regno); | |
524 | /* Something bad happened, let's kill all regs */ | |
525 | for (regno = 0; regno < MAX_BPF_REG; regno++) | |
526 | __mark_reg_not_init(regs + regno); | |
527 | return; | |
528 | } | |
529 | __mark_reg_not_init(regs + regno); | |
a9789ef9 DB |
530 | } |
531 | ||
58e2af8b | 532 | static void init_reg_state(struct bpf_reg_state *regs) |
17a52670 AS |
533 | { |
534 | int i; | |
535 | ||
a9789ef9 DB |
536 | for (i = 0; i < MAX_BPF_REG; i++) |
537 | mark_reg_not_init(regs, i); | |
17a52670 AS |
538 | |
539 | /* frame pointer */ | |
f1174f77 EC |
540 | regs[BPF_REG_FP].type = PTR_TO_STACK; |
541 | mark_reg_known_zero(regs, BPF_REG_FP); | |
17a52670 AS |
542 | |
543 | /* 1st arg to a function */ | |
544 | regs[BPF_REG_1].type = PTR_TO_CTX; | |
f1174f77 | 545 | mark_reg_known_zero(regs, BPF_REG_1); |
6760bf2d DB |
546 | } |
547 | ||
48461135 JB |
548 | static void reset_reg_range_values(struct bpf_reg_state *regs, u32 regno) |
549 | { | |
550 | regs[regno].min_value = BPF_REGISTER_MIN_RANGE; | |
551 | regs[regno].max_value = BPF_REGISTER_MAX_RANGE; | |
4cabc5b1 | 552 | regs[regno].value_from_signed = false; |
f0318d01 GB |
553 | } |
554 | ||
17a52670 AS |
555 | enum reg_arg_type { |
556 | SRC_OP, /* register is used as source operand */ | |
557 | DST_OP, /* register is used as destination operand */ | |
558 | DST_OP_NO_MARK /* same as above, check only, don't mark */ | |
559 | }; | |
560 | ||
58e2af8b | 561 | static int check_reg_arg(struct bpf_reg_state *regs, u32 regno, |
17a52670 AS |
562 | enum reg_arg_type t) |
563 | { | |
564 | if (regno >= MAX_BPF_REG) { | |
565 | verbose("R%d is invalid\n", regno); | |
566 | return -EINVAL; | |
567 | } | |
568 | ||
569 | if (t == SRC_OP) { | |
570 | /* check whether register used as source operand can be read */ | |
571 | if (regs[regno].type == NOT_INIT) { | |
572 | verbose("R%d !read_ok\n", regno); | |
573 | return -EACCES; | |
574 | } | |
575 | } else { | |
576 | /* check whether register used as dest operand can be written to */ | |
577 | if (regno == BPF_REG_FP) { | |
578 | verbose("frame pointer is read only\n"); | |
579 | return -EACCES; | |
580 | } | |
581 | if (t == DST_OP) | |
f1174f77 | 582 | mark_reg_unknown(regs, regno); |
17a52670 AS |
583 | } |
584 | return 0; | |
585 | } | |
586 | ||
1be7f75d AS |
587 | static bool is_spillable_regtype(enum bpf_reg_type type) |
588 | { | |
589 | switch (type) { | |
590 | case PTR_TO_MAP_VALUE: | |
591 | case PTR_TO_MAP_VALUE_OR_NULL: | |
592 | case PTR_TO_STACK: | |
593 | case PTR_TO_CTX: | |
969bf05e AS |
594 | case PTR_TO_PACKET: |
595 | case PTR_TO_PACKET_END: | |
1be7f75d AS |
596 | case CONST_PTR_TO_MAP: |
597 | return true; | |
598 | default: | |
599 | return false; | |
600 | } | |
601 | } | |
602 | ||
17a52670 AS |
603 | /* check_stack_read/write functions track spill/fill of registers, |
604 | * stack boundary and alignment are checked in check_mem_access() | |
605 | */ | |
58e2af8b JK |
606 | static int check_stack_write(struct bpf_verifier_state *state, int off, |
607 | int size, int value_regno) | |
17a52670 | 608 | { |
17a52670 | 609 | int i; |
9c399760 AS |
610 | /* caller checked that off % size == 0 and -MAX_BPF_STACK <= off < 0, |
611 | * so it's aligned access and [off, off + size) are within stack limits | |
612 | */ | |
17a52670 AS |
613 | |
614 | if (value_regno >= 0 && | |
1be7f75d | 615 | is_spillable_regtype(state->regs[value_regno].type)) { |
17a52670 AS |
616 | |
617 | /* register containing pointer is being spilled into stack */ | |
9c399760 | 618 | if (size != BPF_REG_SIZE) { |
17a52670 AS |
619 | verbose("invalid size of register spill\n"); |
620 | return -EACCES; | |
621 | } | |
622 | ||
17a52670 | 623 | /* save register state */ |
9c399760 AS |
624 | state->spilled_regs[(MAX_BPF_STACK + off) / BPF_REG_SIZE] = |
625 | state->regs[value_regno]; | |
17a52670 | 626 | |
9c399760 AS |
627 | for (i = 0; i < BPF_REG_SIZE; i++) |
628 | state->stack_slot_type[MAX_BPF_STACK + off + i] = STACK_SPILL; | |
629 | } else { | |
17a52670 | 630 | /* regular write of data into stack */ |
9c399760 | 631 | state->spilled_regs[(MAX_BPF_STACK + off) / BPF_REG_SIZE] = |
58e2af8b | 632 | (struct bpf_reg_state) {}; |
9c399760 AS |
633 | |
634 | for (i = 0; i < size; i++) | |
635 | state->stack_slot_type[MAX_BPF_STACK + off + i] = STACK_MISC; | |
17a52670 AS |
636 | } |
637 | return 0; | |
638 | } | |
639 | ||
58e2af8b | 640 | static int check_stack_read(struct bpf_verifier_state *state, int off, int size, |
17a52670 AS |
641 | int value_regno) |
642 | { | |
9c399760 | 643 | u8 *slot_type; |
17a52670 | 644 | int i; |
17a52670 | 645 | |
9c399760 | 646 | slot_type = &state->stack_slot_type[MAX_BPF_STACK + off]; |
17a52670 | 647 | |
9c399760 AS |
648 | if (slot_type[0] == STACK_SPILL) { |
649 | if (size != BPF_REG_SIZE) { | |
17a52670 AS |
650 | verbose("invalid size of register spill\n"); |
651 | return -EACCES; | |
652 | } | |
9c399760 AS |
653 | for (i = 1; i < BPF_REG_SIZE; i++) { |
654 | if (slot_type[i] != STACK_SPILL) { | |
17a52670 AS |
655 | verbose("corrupted spill memory\n"); |
656 | return -EACCES; | |
657 | } | |
658 | } | |
659 | ||
660 | if (value_regno >= 0) | |
661 | /* restore register state from stack */ | |
9c399760 AS |
662 | state->regs[value_regno] = |
663 | state->spilled_regs[(MAX_BPF_STACK + off) / BPF_REG_SIZE]; | |
17a52670 AS |
664 | return 0; |
665 | } else { | |
666 | for (i = 0; i < size; i++) { | |
9c399760 | 667 | if (slot_type[i] != STACK_MISC) { |
17a52670 AS |
668 | verbose("invalid read from stack off %d+%d size %d\n", |
669 | off, i, size); | |
670 | return -EACCES; | |
671 | } | |
672 | } | |
673 | if (value_regno >= 0) | |
674 | /* have read misc data from the stack */ | |
f1174f77 | 675 | mark_reg_unknown(state->regs, value_regno); |
17a52670 AS |
676 | return 0; |
677 | } | |
678 | } | |
679 | ||
680 | /* check read/write into map element returned by bpf_map_lookup_elem() */ | |
f1174f77 | 681 | static int __check_map_access(struct bpf_verifier_env *env, u32 regno, int off, |
17a52670 AS |
682 | int size) |
683 | { | |
684 | struct bpf_map *map = env->cur_state.regs[regno].map_ptr; | |
685 | ||
5722569b | 686 | if (off < 0 || size <= 0 || off + size > map->value_size) { |
17a52670 AS |
687 | verbose("invalid access to map value, value_size=%d off=%d size=%d\n", |
688 | map->value_size, off, size); | |
689 | return -EACCES; | |
690 | } | |
691 | return 0; | |
692 | } | |
693 | ||
f1174f77 EC |
694 | /* check read/write into a map element with possible variable offset */ |
695 | static int check_map_access(struct bpf_verifier_env *env, u32 regno, | |
dbcfe5f7 GB |
696 | int off, int size) |
697 | { | |
698 | struct bpf_verifier_state *state = &env->cur_state; | |
699 | struct bpf_reg_state *reg = &state->regs[regno]; | |
700 | int err; | |
701 | ||
f1174f77 EC |
702 | /* We may have adjusted the register to this map value, so we |
703 | * need to try adding each of min_value and max_value to off | |
704 | * to make sure our theoretical access will be safe. | |
dbcfe5f7 GB |
705 | */ |
706 | if (log_level) | |
707 | print_verifier_state(state); | |
f1174f77 EC |
708 | /* If the offset is variable, we will need to be stricter in state |
709 | * pruning from now on. | |
710 | */ | |
711 | if (!tnum_is_const(reg->var_off)) | |
712 | env->varlen_map_value_access = true; | |
dbcfe5f7 GB |
713 | /* The minimum value is only important with signed |
714 | * comparisons where we can't assume the floor of a | |
715 | * value is 0. If we are using signed variables for our | |
716 | * index'es we need to make sure that whatever we use | |
717 | * will have a set floor within our range. | |
718 | */ | |
719 | if (reg->min_value < 0) { | |
720 | verbose("R%d min value is negative, either use unsigned index or do a if (index >=0) check.\n", | |
721 | regno); | |
722 | return -EACCES; | |
723 | } | |
f1174f77 | 724 | err = __check_map_access(env, regno, reg->min_value + off, size); |
dbcfe5f7 | 725 | if (err) { |
f1174f77 | 726 | verbose("R%d min value is outside of the array range\n", regno); |
dbcfe5f7 GB |
727 | return err; |
728 | } | |
729 | ||
730 | /* If we haven't set a max value then we need to bail | |
731 | * since we can't be sure we won't do bad things. | |
732 | */ | |
733 | if (reg->max_value == BPF_REGISTER_MAX_RANGE) { | |
734 | verbose("R%d unbounded memory access, make sure to bounds check any array access into a map\n", | |
735 | regno); | |
736 | return -EACCES; | |
737 | } | |
f1174f77 EC |
738 | err = __check_map_access(env, regno, reg->max_value + off, size); |
739 | if (err) | |
740 | verbose("R%d max value is outside of the array range\n", regno); | |
741 | return err; | |
dbcfe5f7 GB |
742 | } |
743 | ||
969bf05e AS |
744 | #define MAX_PACKET_OFF 0xffff |
745 | ||
58e2af8b | 746 | static bool may_access_direct_pkt_data(struct bpf_verifier_env *env, |
3a0af8fd TG |
747 | const struct bpf_call_arg_meta *meta, |
748 | enum bpf_access_type t) | |
4acf6c0b | 749 | { |
36bbef52 | 750 | switch (env->prog->type) { |
3a0af8fd TG |
751 | case BPF_PROG_TYPE_LWT_IN: |
752 | case BPF_PROG_TYPE_LWT_OUT: | |
753 | /* dst_input() and dst_output() can't write for now */ | |
754 | if (t == BPF_WRITE) | |
755 | return false; | |
7e57fbb2 | 756 | /* fallthrough */ |
36bbef52 DB |
757 | case BPF_PROG_TYPE_SCHED_CLS: |
758 | case BPF_PROG_TYPE_SCHED_ACT: | |
4acf6c0b | 759 | case BPF_PROG_TYPE_XDP: |
3a0af8fd | 760 | case BPF_PROG_TYPE_LWT_XMIT: |
36bbef52 DB |
761 | if (meta) |
762 | return meta->pkt_access; | |
763 | ||
764 | env->seen_direct_write = true; | |
4acf6c0b BB |
765 | return true; |
766 | default: | |
767 | return false; | |
768 | } | |
769 | } | |
770 | ||
f1174f77 EC |
771 | static int __check_packet_access(struct bpf_verifier_env *env, u32 regno, |
772 | int off, int size) | |
969bf05e | 773 | { |
58e2af8b JK |
774 | struct bpf_reg_state *regs = env->cur_state.regs; |
775 | struct bpf_reg_state *reg = ®s[regno]; | |
969bf05e | 776 | |
f1174f77 | 777 | if (off < 0 || size <= 0 || (u64)off + size > reg->range) { |
d91b28ed AS |
778 | verbose("invalid access to packet, off=%d size=%d, R%d(id=%d,off=%d,r=%d)\n", |
779 | off, size, regno, reg->id, reg->off, reg->range); | |
969bf05e AS |
780 | return -EACCES; |
781 | } | |
782 | return 0; | |
783 | } | |
784 | ||
f1174f77 EC |
785 | static int check_packet_access(struct bpf_verifier_env *env, u32 regno, int off, |
786 | int size) | |
787 | { | |
788 | struct bpf_reg_state *regs = env->cur_state.regs; | |
789 | struct bpf_reg_state *reg = ®s[regno]; | |
790 | int err; | |
791 | ||
792 | /* We may have added a variable offset to the packet pointer; but any | |
793 | * reg->range we have comes after that. We are only checking the fixed | |
794 | * offset. | |
795 | */ | |
796 | ||
797 | /* We don't allow negative numbers, because we aren't tracking enough | |
798 | * detail to prove they're safe. | |
799 | */ | |
800 | if (reg->min_value < 0) { | |
801 | verbose("R%d min value is negative, either use unsigned index or do a if (index >=0) check.\n", | |
802 | regno); | |
803 | return -EACCES; | |
804 | } | |
805 | err = __check_packet_access(env, regno, off, size); | |
806 | if (err) { | |
807 | verbose("R%d offset is outside of the packet\n", regno); | |
808 | return err; | |
809 | } | |
810 | return err; | |
811 | } | |
812 | ||
813 | /* check access to 'struct bpf_context' fields. Supports fixed offsets only */ | |
31fd8581 | 814 | static int check_ctx_access(struct bpf_verifier_env *env, int insn_idx, int off, int size, |
19de99f7 | 815 | enum bpf_access_type t, enum bpf_reg_type *reg_type) |
17a52670 | 816 | { |
f96da094 DB |
817 | struct bpf_insn_access_aux info = { |
818 | .reg_type = *reg_type, | |
819 | }; | |
31fd8581 | 820 | |
13a27dfc JK |
821 | /* for analyzer ctx accesses are already validated and converted */ |
822 | if (env->analyzer_ops) | |
823 | return 0; | |
824 | ||
17a52670 | 825 | if (env->prog->aux->ops->is_valid_access && |
23994631 | 826 | env->prog->aux->ops->is_valid_access(off, size, t, &info)) { |
f96da094 DB |
827 | /* A non zero info.ctx_field_size indicates that this field is a |
828 | * candidate for later verifier transformation to load the whole | |
829 | * field and then apply a mask when accessed with a narrower | |
830 | * access than actual ctx access size. A zero info.ctx_field_size | |
831 | * will only allow for whole field access and rejects any other | |
832 | * type of narrower access. | |
31fd8581 | 833 | */ |
f96da094 | 834 | env->insn_aux_data[insn_idx].ctx_field_size = info.ctx_field_size; |
23994631 | 835 | *reg_type = info.reg_type; |
31fd8581 | 836 | |
32bbe007 AS |
837 | /* remember the offset of last byte accessed in ctx */ |
838 | if (env->prog->aux->max_ctx_offset < off + size) | |
839 | env->prog->aux->max_ctx_offset = off + size; | |
17a52670 | 840 | return 0; |
32bbe007 | 841 | } |
17a52670 AS |
842 | |
843 | verbose("invalid bpf_context access off=%d size=%d\n", off, size); | |
844 | return -EACCES; | |
845 | } | |
846 | ||
4cabc5b1 DB |
847 | static bool __is_pointer_value(bool allow_ptr_leaks, |
848 | const struct bpf_reg_state *reg) | |
1be7f75d | 849 | { |
4cabc5b1 | 850 | if (allow_ptr_leaks) |
1be7f75d AS |
851 | return false; |
852 | ||
f1174f77 | 853 | return reg->type != SCALAR_VALUE; |
1be7f75d AS |
854 | } |
855 | ||
4cabc5b1 DB |
856 | static bool is_pointer_value(struct bpf_verifier_env *env, int regno) |
857 | { | |
858 | return __is_pointer_value(env->allow_ptr_leaks, &env->cur_state.regs[regno]); | |
859 | } | |
860 | ||
79adffcd | 861 | static int check_pkt_ptr_alignment(const struct bpf_reg_state *reg, |
d1174416 | 862 | int off, int size, bool strict) |
969bf05e | 863 | { |
f1174f77 | 864 | struct tnum reg_off; |
e07b98d9 | 865 | int ip_align; |
d1174416 DM |
866 | |
867 | /* Byte size accesses are always allowed. */ | |
868 | if (!strict || size == 1) | |
869 | return 0; | |
870 | ||
e4eda884 DM |
871 | /* For platforms that do not have a Kconfig enabling |
872 | * CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS the value of | |
873 | * NET_IP_ALIGN is universally set to '2'. And on platforms | |
874 | * that do set CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS, we get | |
875 | * to this code only in strict mode where we want to emulate | |
876 | * the NET_IP_ALIGN==2 checking. Therefore use an | |
877 | * unconditional IP align value of '2'. | |
e07b98d9 | 878 | */ |
e4eda884 | 879 | ip_align = 2; |
f1174f77 EC |
880 | |
881 | reg_off = tnum_add(reg->var_off, tnum_const(ip_align + reg->off + off)); | |
882 | if (!tnum_is_aligned(reg_off, size)) { | |
883 | char tn_buf[48]; | |
884 | ||
885 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); | |
886 | verbose("misaligned packet access off %d+%s+%d+%d size %d\n", | |
887 | ip_align, tn_buf, reg->off, off, size); | |
969bf05e AS |
888 | return -EACCES; |
889 | } | |
79adffcd | 890 | |
969bf05e AS |
891 | return 0; |
892 | } | |
893 | ||
f1174f77 EC |
894 | static int check_generic_ptr_alignment(const struct bpf_reg_state *reg, |
895 | const char *pointer_desc, | |
896 | int off, int size, bool strict) | |
79adffcd | 897 | { |
f1174f77 EC |
898 | struct tnum reg_off; |
899 | ||
900 | /* Byte size accesses are always allowed. */ | |
901 | if (!strict || size == 1) | |
902 | return 0; | |
903 | ||
904 | reg_off = tnum_add(reg->var_off, tnum_const(reg->off + off)); | |
905 | if (!tnum_is_aligned(reg_off, size)) { | |
906 | char tn_buf[48]; | |
907 | ||
908 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); | |
909 | verbose("misaligned %saccess off %s+%d+%d size %d\n", | |
910 | pointer_desc, tn_buf, reg->off, off, size); | |
79adffcd DB |
911 | return -EACCES; |
912 | } | |
913 | ||
969bf05e AS |
914 | return 0; |
915 | } | |
916 | ||
e07b98d9 DM |
917 | static int check_ptr_alignment(struct bpf_verifier_env *env, |
918 | const struct bpf_reg_state *reg, | |
79adffcd DB |
919 | int off, int size) |
920 | { | |
e07b98d9 | 921 | bool strict = env->strict_alignment; |
f1174f77 | 922 | const char *pointer_desc = ""; |
d1174416 | 923 | |
79adffcd DB |
924 | switch (reg->type) { |
925 | case PTR_TO_PACKET: | |
f1174f77 | 926 | /* special case, because of NET_IP_ALIGN */ |
d1174416 | 927 | return check_pkt_ptr_alignment(reg, off, size, strict); |
f1174f77 EC |
928 | case PTR_TO_MAP_VALUE: |
929 | pointer_desc = "value "; | |
930 | break; | |
931 | case PTR_TO_CTX: | |
932 | pointer_desc = "context "; | |
933 | break; | |
934 | case PTR_TO_STACK: | |
935 | pointer_desc = "stack "; | |
936 | break; | |
79adffcd | 937 | default: |
f1174f77 | 938 | break; |
79adffcd | 939 | } |
f1174f77 | 940 | return check_generic_ptr_alignment(reg, pointer_desc, off, size, strict); |
79adffcd DB |
941 | } |
942 | ||
17a52670 AS |
943 | /* check whether memory at (regno + off) is accessible for t = (read | write) |
944 | * if t==write, value_regno is a register which value is stored into memory | |
945 | * if t==read, value_regno is a register which will receive the value from memory | |
946 | * if t==write && value_regno==-1, some unknown value is stored into memory | |
947 | * if t==read && value_regno==-1, don't care what we read from memory | |
948 | */ | |
31fd8581 | 949 | static int check_mem_access(struct bpf_verifier_env *env, int insn_idx, u32 regno, int off, |
17a52670 AS |
950 | int bpf_size, enum bpf_access_type t, |
951 | int value_regno) | |
952 | { | |
58e2af8b JK |
953 | struct bpf_verifier_state *state = &env->cur_state; |
954 | struct bpf_reg_state *reg = &state->regs[regno]; | |
17a52670 AS |
955 | int size, err = 0; |
956 | ||
957 | size = bpf_size_to_bytes(bpf_size); | |
958 | if (size < 0) | |
959 | return size; | |
960 | ||
f1174f77 | 961 | /* alignment checks will add in reg->off themselves */ |
e07b98d9 | 962 | err = check_ptr_alignment(env, reg, off, size); |
969bf05e AS |
963 | if (err) |
964 | return err; | |
17a52670 | 965 | |
f1174f77 EC |
966 | /* for access checks, reg->off is just part of off */ |
967 | off += reg->off; | |
968 | ||
969 | if (reg->type == PTR_TO_MAP_VALUE) { | |
1be7f75d AS |
970 | if (t == BPF_WRITE && value_regno >= 0 && |
971 | is_pointer_value(env, value_regno)) { | |
972 | verbose("R%d leaks addr into map\n", value_regno); | |
973 | return -EACCES; | |
974 | } | |
48461135 | 975 | |
f1174f77 | 976 | err = check_map_access(env, regno, off, size); |
17a52670 | 977 | if (!err && t == BPF_READ && value_regno >= 0) |
f1174f77 | 978 | mark_reg_unknown(state->regs, value_regno); |
17a52670 | 979 | |
1a0dc1ac | 980 | } else if (reg->type == PTR_TO_CTX) { |
f1174f77 | 981 | enum bpf_reg_type reg_type = SCALAR_VALUE; |
19de99f7 | 982 | |
1be7f75d AS |
983 | if (t == BPF_WRITE && value_regno >= 0 && |
984 | is_pointer_value(env, value_regno)) { | |
985 | verbose("R%d leaks addr into ctx\n", value_regno); | |
986 | return -EACCES; | |
987 | } | |
f1174f77 EC |
988 | /* ctx accesses must be at a fixed offset, so that we can |
989 | * determine what type of data were returned. | |
990 | */ | |
991 | if (!tnum_is_const(reg->var_off)) { | |
992 | char tn_buf[48]; | |
993 | ||
994 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); | |
995 | verbose("variable ctx access var_off=%s off=%d size=%d", | |
996 | tn_buf, off, size); | |
997 | return -EACCES; | |
998 | } | |
999 | off += reg->var_off.value; | |
31fd8581 | 1000 | err = check_ctx_access(env, insn_idx, off, size, t, ®_type); |
969bf05e | 1001 | if (!err && t == BPF_READ && value_regno >= 0) { |
f1174f77 EC |
1002 | /* ctx access returns either a scalar, or a |
1003 | * PTR_TO_PACKET[_END]. In the latter case, we know | |
1004 | * the offset is zero. | |
1005 | */ | |
1006 | if (reg_type == SCALAR_VALUE) | |
1007 | mark_reg_unknown(state->regs, value_regno); | |
1008 | else | |
1009 | mark_reg_known_zero(state->regs, value_regno); | |
1010 | state->regs[value_regno].id = 0; | |
1011 | state->regs[value_regno].off = 0; | |
1012 | state->regs[value_regno].range = 0; | |
1955351d | 1013 | state->regs[value_regno].type = reg_type; |
969bf05e | 1014 | } |
17a52670 | 1015 | |
f1174f77 EC |
1016 | } else if (reg->type == PTR_TO_STACK) { |
1017 | /* stack accesses must be at a fixed offset, so that we can | |
1018 | * determine what type of data were returned. | |
1019 | * See check_stack_read(). | |
1020 | */ | |
1021 | if (!tnum_is_const(reg->var_off)) { | |
1022 | char tn_buf[48]; | |
1023 | ||
1024 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); | |
1025 | verbose("variable stack access var_off=%s off=%d size=%d", | |
1026 | tn_buf, off, size); | |
1027 | return -EACCES; | |
1028 | } | |
1029 | off += reg->var_off.value; | |
17a52670 AS |
1030 | if (off >= 0 || off < -MAX_BPF_STACK) { |
1031 | verbose("invalid stack off=%d size=%d\n", off, size); | |
1032 | return -EACCES; | |
1033 | } | |
8726679a AS |
1034 | |
1035 | if (env->prog->aux->stack_depth < -off) | |
1036 | env->prog->aux->stack_depth = -off; | |
1037 | ||
1be7f75d AS |
1038 | if (t == BPF_WRITE) { |
1039 | if (!env->allow_ptr_leaks && | |
1040 | state->stack_slot_type[MAX_BPF_STACK + off] == STACK_SPILL && | |
1041 | size != BPF_REG_SIZE) { | |
1042 | verbose("attempt to corrupt spilled pointer on stack\n"); | |
1043 | return -EACCES; | |
1044 | } | |
17a52670 | 1045 | err = check_stack_write(state, off, size, value_regno); |
1be7f75d | 1046 | } else { |
17a52670 | 1047 | err = check_stack_read(state, off, size, value_regno); |
1be7f75d | 1048 | } |
f1174f77 | 1049 | } else if (reg->type == PTR_TO_PACKET) { |
3a0af8fd | 1050 | if (t == BPF_WRITE && !may_access_direct_pkt_data(env, NULL, t)) { |
969bf05e AS |
1051 | verbose("cannot write into packet\n"); |
1052 | return -EACCES; | |
1053 | } | |
4acf6c0b BB |
1054 | if (t == BPF_WRITE && value_regno >= 0 && |
1055 | is_pointer_value(env, value_regno)) { | |
1056 | verbose("R%d leaks addr into packet\n", value_regno); | |
1057 | return -EACCES; | |
1058 | } | |
969bf05e AS |
1059 | err = check_packet_access(env, regno, off, size); |
1060 | if (!err && t == BPF_READ && value_regno >= 0) | |
f1174f77 | 1061 | mark_reg_unknown(state->regs, value_regno); |
17a52670 AS |
1062 | } else { |
1063 | verbose("R%d invalid mem access '%s'\n", | |
1a0dc1ac | 1064 | regno, reg_type_str[reg->type]); |
17a52670 AS |
1065 | return -EACCES; |
1066 | } | |
969bf05e | 1067 | |
f1174f77 EC |
1068 | if (!err && size < BPF_REG_SIZE && value_regno >= 0 && t == BPF_READ && |
1069 | state->regs[value_regno].type == SCALAR_VALUE) { | |
1070 | /* b/h/w load zero-extends, mark upper bits as known 0 */ | |
1071 | state->regs[value_regno].var_off = tnum_cast( | |
1072 | state->regs[value_regno].var_off, size); | |
1073 | /* sign bit is known zero, so we can bound the value */ | |
1074 | state->regs[value_regno].min_value = 0; | |
1075 | state->regs[value_regno].max_value = min_t(u64, | |
1076 | state->regs[value_regno].var_off.value | | |
1077 | state->regs[value_regno].var_off.mask, | |
1078 | BPF_REGISTER_MAX_RANGE); | |
969bf05e | 1079 | } |
17a52670 AS |
1080 | return err; |
1081 | } | |
1082 | ||
31fd8581 | 1083 | static int check_xadd(struct bpf_verifier_env *env, int insn_idx, struct bpf_insn *insn) |
17a52670 | 1084 | { |
58e2af8b | 1085 | struct bpf_reg_state *regs = env->cur_state.regs; |
17a52670 AS |
1086 | int err; |
1087 | ||
1088 | if ((BPF_SIZE(insn->code) != BPF_W && BPF_SIZE(insn->code) != BPF_DW) || | |
1089 | insn->imm != 0) { | |
1090 | verbose("BPF_XADD uses reserved fields\n"); | |
1091 | return -EINVAL; | |
1092 | } | |
1093 | ||
1094 | /* check src1 operand */ | |
1095 | err = check_reg_arg(regs, insn->src_reg, SRC_OP); | |
1096 | if (err) | |
1097 | return err; | |
1098 | ||
1099 | /* check src2 operand */ | |
1100 | err = check_reg_arg(regs, insn->dst_reg, SRC_OP); | |
1101 | if (err) | |
1102 | return err; | |
1103 | ||
6bdf6abc DB |
1104 | if (is_pointer_value(env, insn->src_reg)) { |
1105 | verbose("R%d leaks addr into mem\n", insn->src_reg); | |
1106 | return -EACCES; | |
1107 | } | |
1108 | ||
17a52670 | 1109 | /* check whether atomic_add can read the memory */ |
31fd8581 | 1110 | err = check_mem_access(env, insn_idx, insn->dst_reg, insn->off, |
17a52670 AS |
1111 | BPF_SIZE(insn->code), BPF_READ, -1); |
1112 | if (err) | |
1113 | return err; | |
1114 | ||
1115 | /* check whether atomic_add can write into the same memory */ | |
31fd8581 | 1116 | return check_mem_access(env, insn_idx, insn->dst_reg, insn->off, |
17a52670 AS |
1117 | BPF_SIZE(insn->code), BPF_WRITE, -1); |
1118 | } | |
1119 | ||
f1174f77 EC |
1120 | /* Does this register contain a constant zero? */ |
1121 | static bool register_is_null(struct bpf_reg_state reg) | |
1122 | { | |
1123 | return reg.type == SCALAR_VALUE && tnum_equals_const(reg.var_off, 0); | |
1124 | } | |
1125 | ||
17a52670 AS |
1126 | /* when register 'regno' is passed into function that will read 'access_size' |
1127 | * bytes from that pointer, make sure that it's within stack boundary | |
f1174f77 EC |
1128 | * and all elements of stack are initialized. |
1129 | * Unlike most pointer bounds-checking functions, this one doesn't take an | |
1130 | * 'off' argument, so it has to add in reg->off itself. | |
17a52670 | 1131 | */ |
58e2af8b | 1132 | static int check_stack_boundary(struct bpf_verifier_env *env, int regno, |
435faee1 DB |
1133 | int access_size, bool zero_size_allowed, |
1134 | struct bpf_call_arg_meta *meta) | |
17a52670 | 1135 | { |
58e2af8b JK |
1136 | struct bpf_verifier_state *state = &env->cur_state; |
1137 | struct bpf_reg_state *regs = state->regs; | |
17a52670 AS |
1138 | int off, i; |
1139 | ||
8e2fe1d9 | 1140 | if (regs[regno].type != PTR_TO_STACK) { |
f1174f77 | 1141 | /* Allow zero-byte read from NULL, regardless of pointer type */ |
8e2fe1d9 | 1142 | if (zero_size_allowed && access_size == 0 && |
f1174f77 | 1143 | register_is_null(regs[regno])) |
8e2fe1d9 DB |
1144 | return 0; |
1145 | ||
1146 | verbose("R%d type=%s expected=%s\n", regno, | |
1147 | reg_type_str[regs[regno].type], | |
1148 | reg_type_str[PTR_TO_STACK]); | |
17a52670 | 1149 | return -EACCES; |
8e2fe1d9 | 1150 | } |
17a52670 | 1151 | |
f1174f77 EC |
1152 | /* Only allow fixed-offset stack reads */ |
1153 | if (!tnum_is_const(regs[regno].var_off)) { | |
1154 | char tn_buf[48]; | |
1155 | ||
1156 | tnum_strn(tn_buf, sizeof(tn_buf), regs[regno].var_off); | |
1157 | verbose("invalid variable stack read R%d var_off=%s\n", | |
1158 | regno, tn_buf); | |
1159 | } | |
1160 | off = regs[regno].off + regs[regno].var_off.value; | |
17a52670 AS |
1161 | if (off >= 0 || off < -MAX_BPF_STACK || off + access_size > 0 || |
1162 | access_size <= 0) { | |
1163 | verbose("invalid stack type R%d off=%d access_size=%d\n", | |
1164 | regno, off, access_size); | |
1165 | return -EACCES; | |
1166 | } | |
1167 | ||
8726679a AS |
1168 | if (env->prog->aux->stack_depth < -off) |
1169 | env->prog->aux->stack_depth = -off; | |
1170 | ||
435faee1 DB |
1171 | if (meta && meta->raw_mode) { |
1172 | meta->access_size = access_size; | |
1173 | meta->regno = regno; | |
1174 | return 0; | |
1175 | } | |
1176 | ||
17a52670 | 1177 | for (i = 0; i < access_size; i++) { |
9c399760 | 1178 | if (state->stack_slot_type[MAX_BPF_STACK + off + i] != STACK_MISC) { |
17a52670 AS |
1179 | verbose("invalid indirect read from stack off %d+%d size %d\n", |
1180 | off, i, access_size); | |
1181 | return -EACCES; | |
1182 | } | |
1183 | } | |
1184 | return 0; | |
1185 | } | |
1186 | ||
06c1c049 GB |
1187 | static int check_helper_mem_access(struct bpf_verifier_env *env, int regno, |
1188 | int access_size, bool zero_size_allowed, | |
1189 | struct bpf_call_arg_meta *meta) | |
1190 | { | |
f1174f77 | 1191 | struct bpf_reg_state *regs = env->cur_state.regs, *reg = ®s[regno]; |
06c1c049 | 1192 | |
f1174f77 | 1193 | switch (reg->type) { |
06c1c049 | 1194 | case PTR_TO_PACKET: |
f1174f77 | 1195 | return check_packet_access(env, regno, reg->off, access_size); |
06c1c049 | 1196 | case PTR_TO_MAP_VALUE: |
f1174f77 EC |
1197 | return check_map_access(env, regno, reg->off, access_size); |
1198 | default: /* scalar_value|ptr_to_stack or invalid ptr */ | |
06c1c049 GB |
1199 | return check_stack_boundary(env, regno, access_size, |
1200 | zero_size_allowed, meta); | |
1201 | } | |
1202 | } | |
1203 | ||
58e2af8b | 1204 | static int check_func_arg(struct bpf_verifier_env *env, u32 regno, |
33ff9823 DB |
1205 | enum bpf_arg_type arg_type, |
1206 | struct bpf_call_arg_meta *meta) | |
17a52670 | 1207 | { |
58e2af8b | 1208 | struct bpf_reg_state *regs = env->cur_state.regs, *reg = ®s[regno]; |
6841de8b | 1209 | enum bpf_reg_type expected_type, type = reg->type; |
17a52670 AS |
1210 | int err = 0; |
1211 | ||
80f1d68c | 1212 | if (arg_type == ARG_DONTCARE) |
17a52670 AS |
1213 | return 0; |
1214 | ||
6841de8b | 1215 | if (type == NOT_INIT) { |
17a52670 AS |
1216 | verbose("R%d !read_ok\n", regno); |
1217 | return -EACCES; | |
1218 | } | |
1219 | ||
1be7f75d AS |
1220 | if (arg_type == ARG_ANYTHING) { |
1221 | if (is_pointer_value(env, regno)) { | |
1222 | verbose("R%d leaks addr into helper function\n", regno); | |
1223 | return -EACCES; | |
1224 | } | |
80f1d68c | 1225 | return 0; |
1be7f75d | 1226 | } |
80f1d68c | 1227 | |
3a0af8fd TG |
1228 | if (type == PTR_TO_PACKET && |
1229 | !may_access_direct_pkt_data(env, meta, BPF_READ)) { | |
36bbef52 | 1230 | verbose("helper access to the packet is not allowed\n"); |
6841de8b AS |
1231 | return -EACCES; |
1232 | } | |
1233 | ||
8e2fe1d9 | 1234 | if (arg_type == ARG_PTR_TO_MAP_KEY || |
17a52670 AS |
1235 | arg_type == ARG_PTR_TO_MAP_VALUE) { |
1236 | expected_type = PTR_TO_STACK; | |
6841de8b AS |
1237 | if (type != PTR_TO_PACKET && type != expected_type) |
1238 | goto err_type; | |
39f19ebb AS |
1239 | } else if (arg_type == ARG_CONST_SIZE || |
1240 | arg_type == ARG_CONST_SIZE_OR_ZERO) { | |
f1174f77 EC |
1241 | expected_type = SCALAR_VALUE; |
1242 | if (type != expected_type) | |
6841de8b | 1243 | goto err_type; |
17a52670 AS |
1244 | } else if (arg_type == ARG_CONST_MAP_PTR) { |
1245 | expected_type = CONST_PTR_TO_MAP; | |
6841de8b AS |
1246 | if (type != expected_type) |
1247 | goto err_type; | |
608cd71a AS |
1248 | } else if (arg_type == ARG_PTR_TO_CTX) { |
1249 | expected_type = PTR_TO_CTX; | |
6841de8b AS |
1250 | if (type != expected_type) |
1251 | goto err_type; | |
39f19ebb AS |
1252 | } else if (arg_type == ARG_PTR_TO_MEM || |
1253 | arg_type == ARG_PTR_TO_UNINIT_MEM) { | |
8e2fe1d9 DB |
1254 | expected_type = PTR_TO_STACK; |
1255 | /* One exception here. In case function allows for NULL to be | |
f1174f77 | 1256 | * passed in as argument, it's a SCALAR_VALUE type. Final test |
8e2fe1d9 DB |
1257 | * happens during stack boundary checking. |
1258 | */ | |
f1174f77 | 1259 | if (register_is_null(*reg)) |
6841de8b | 1260 | /* final test in check_stack_boundary() */; |
5722569b | 1261 | else if (type != PTR_TO_PACKET && type != PTR_TO_MAP_VALUE && |
f1174f77 | 1262 | type != expected_type) |
6841de8b | 1263 | goto err_type; |
39f19ebb | 1264 | meta->raw_mode = arg_type == ARG_PTR_TO_UNINIT_MEM; |
17a52670 AS |
1265 | } else { |
1266 | verbose("unsupported arg_type %d\n", arg_type); | |
1267 | return -EFAULT; | |
1268 | } | |
1269 | ||
17a52670 AS |
1270 | if (arg_type == ARG_CONST_MAP_PTR) { |
1271 | /* bpf_map_xxx(map_ptr) call: remember that map_ptr */ | |
33ff9823 | 1272 | meta->map_ptr = reg->map_ptr; |
17a52670 AS |
1273 | } else if (arg_type == ARG_PTR_TO_MAP_KEY) { |
1274 | /* bpf_map_xxx(..., map_ptr, ..., key) call: | |
1275 | * check that [key, key + map->key_size) are within | |
1276 | * stack limits and initialized | |
1277 | */ | |
33ff9823 | 1278 | if (!meta->map_ptr) { |
17a52670 AS |
1279 | /* in function declaration map_ptr must come before |
1280 | * map_key, so that it's verified and known before | |
1281 | * we have to check map_key here. Otherwise it means | |
1282 | * that kernel subsystem misconfigured verifier | |
1283 | */ | |
1284 | verbose("invalid map_ptr to access map->key\n"); | |
1285 | return -EACCES; | |
1286 | } | |
6841de8b | 1287 | if (type == PTR_TO_PACKET) |
f1174f77 | 1288 | err = check_packet_access(env, regno, reg->off, |
6841de8b AS |
1289 | meta->map_ptr->key_size); |
1290 | else | |
1291 | err = check_stack_boundary(env, regno, | |
1292 | meta->map_ptr->key_size, | |
1293 | false, NULL); | |
17a52670 AS |
1294 | } else if (arg_type == ARG_PTR_TO_MAP_VALUE) { |
1295 | /* bpf_map_xxx(..., map_ptr, ..., value) call: | |
1296 | * check [value, value + map->value_size) validity | |
1297 | */ | |
33ff9823 | 1298 | if (!meta->map_ptr) { |
17a52670 AS |
1299 | /* kernel subsystem misconfigured verifier */ |
1300 | verbose("invalid map_ptr to access map->value\n"); | |
1301 | return -EACCES; | |
1302 | } | |
6841de8b | 1303 | if (type == PTR_TO_PACKET) |
f1174f77 | 1304 | err = check_packet_access(env, regno, reg->off, |
6841de8b AS |
1305 | meta->map_ptr->value_size); |
1306 | else | |
1307 | err = check_stack_boundary(env, regno, | |
1308 | meta->map_ptr->value_size, | |
1309 | false, NULL); | |
39f19ebb AS |
1310 | } else if (arg_type == ARG_CONST_SIZE || |
1311 | arg_type == ARG_CONST_SIZE_OR_ZERO) { | |
1312 | bool zero_size_allowed = (arg_type == ARG_CONST_SIZE_OR_ZERO); | |
17a52670 | 1313 | |
17a52670 AS |
1314 | /* bpf_xxx(..., buf, len) call will access 'len' bytes |
1315 | * from stack pointer 'buf'. Check it | |
1316 | * note: regno == len, regno - 1 == buf | |
1317 | */ | |
1318 | if (regno == 0) { | |
1319 | /* kernel subsystem misconfigured verifier */ | |
39f19ebb | 1320 | verbose("ARG_CONST_SIZE cannot be first argument\n"); |
17a52670 AS |
1321 | return -EACCES; |
1322 | } | |
06c1c049 | 1323 | |
f1174f77 EC |
1324 | /* The register is SCALAR_VALUE; the access check |
1325 | * happens using its boundaries. | |
06c1c049 | 1326 | */ |
f1174f77 EC |
1327 | |
1328 | if (!tnum_is_const(reg->var_off)) | |
06c1c049 GB |
1329 | /* For unprivileged variable accesses, disable raw |
1330 | * mode so that the program is required to | |
1331 | * initialize all the memory that the helper could | |
1332 | * just partially fill up. | |
1333 | */ | |
1334 | meta = NULL; | |
1335 | ||
f1174f77 EC |
1336 | if (reg->min_value < 0) { |
1337 | verbose("R%d min value is negative, either use unsigned or 'var &= const'\n", | |
1338 | regno); | |
1339 | return -EACCES; | |
1340 | } | |
06c1c049 | 1341 | |
f1174f77 EC |
1342 | if (reg->min_value == 0) { |
1343 | err = check_helper_mem_access(env, regno - 1, 0, | |
1344 | zero_size_allowed, | |
1345 | meta); | |
06c1c049 GB |
1346 | if (err) |
1347 | return err; | |
06c1c049 | 1348 | } |
f1174f77 EC |
1349 | |
1350 | if (reg->max_value == BPF_REGISTER_MAX_RANGE) { | |
1351 | verbose("R%d unbounded memory access, use 'var &= const' or 'if (var < const)'\n", | |
1352 | regno); | |
1353 | return -EACCES; | |
1354 | } | |
1355 | err = check_helper_mem_access(env, regno - 1, | |
1356 | reg->max_value, | |
1357 | zero_size_allowed, meta); | |
17a52670 AS |
1358 | } |
1359 | ||
1360 | return err; | |
6841de8b AS |
1361 | err_type: |
1362 | verbose("R%d type=%s expected=%s\n", regno, | |
1363 | reg_type_str[type], reg_type_str[expected_type]); | |
1364 | return -EACCES; | |
17a52670 AS |
1365 | } |
1366 | ||
35578d79 KX |
1367 | static int check_map_func_compatibility(struct bpf_map *map, int func_id) |
1368 | { | |
35578d79 KX |
1369 | if (!map) |
1370 | return 0; | |
1371 | ||
6aff67c8 AS |
1372 | /* We need a two way check, first is from map perspective ... */ |
1373 | switch (map->map_type) { | |
1374 | case BPF_MAP_TYPE_PROG_ARRAY: | |
1375 | if (func_id != BPF_FUNC_tail_call) | |
1376 | goto error; | |
1377 | break; | |
1378 | case BPF_MAP_TYPE_PERF_EVENT_ARRAY: | |
1379 | if (func_id != BPF_FUNC_perf_event_read && | |
1380 | func_id != BPF_FUNC_perf_event_output) | |
1381 | goto error; | |
1382 | break; | |
1383 | case BPF_MAP_TYPE_STACK_TRACE: | |
1384 | if (func_id != BPF_FUNC_get_stackid) | |
1385 | goto error; | |
1386 | break; | |
4ed8ec52 | 1387 | case BPF_MAP_TYPE_CGROUP_ARRAY: |
60747ef4 | 1388 | if (func_id != BPF_FUNC_skb_under_cgroup && |
60d20f91 | 1389 | func_id != BPF_FUNC_current_task_under_cgroup) |
4a482f34 MKL |
1390 | goto error; |
1391 | break; | |
546ac1ff JF |
1392 | /* devmap returns a pointer to a live net_device ifindex that we cannot |
1393 | * allow to be modified from bpf side. So do not allow lookup elements | |
1394 | * for now. | |
1395 | */ | |
1396 | case BPF_MAP_TYPE_DEVMAP: | |
2ddf71e2 | 1397 | if (func_id != BPF_FUNC_redirect_map) |
546ac1ff JF |
1398 | goto error; |
1399 | break; | |
56f668df | 1400 | case BPF_MAP_TYPE_ARRAY_OF_MAPS: |
bcc6b1b7 | 1401 | case BPF_MAP_TYPE_HASH_OF_MAPS: |
56f668df MKL |
1402 | if (func_id != BPF_FUNC_map_lookup_elem) |
1403 | goto error; | |
6aff67c8 AS |
1404 | default: |
1405 | break; | |
1406 | } | |
1407 | ||
1408 | /* ... and second from the function itself. */ | |
1409 | switch (func_id) { | |
1410 | case BPF_FUNC_tail_call: | |
1411 | if (map->map_type != BPF_MAP_TYPE_PROG_ARRAY) | |
1412 | goto error; | |
1413 | break; | |
1414 | case BPF_FUNC_perf_event_read: | |
1415 | case BPF_FUNC_perf_event_output: | |
1416 | if (map->map_type != BPF_MAP_TYPE_PERF_EVENT_ARRAY) | |
1417 | goto error; | |
1418 | break; | |
1419 | case BPF_FUNC_get_stackid: | |
1420 | if (map->map_type != BPF_MAP_TYPE_STACK_TRACE) | |
1421 | goto error; | |
1422 | break; | |
60d20f91 | 1423 | case BPF_FUNC_current_task_under_cgroup: |
747ea55e | 1424 | case BPF_FUNC_skb_under_cgroup: |
4a482f34 MKL |
1425 | if (map->map_type != BPF_MAP_TYPE_CGROUP_ARRAY) |
1426 | goto error; | |
1427 | break; | |
97f91a7c JF |
1428 | case BPF_FUNC_redirect_map: |
1429 | if (map->map_type != BPF_MAP_TYPE_DEVMAP) | |
1430 | goto error; | |
1431 | break; | |
6aff67c8 AS |
1432 | default: |
1433 | break; | |
35578d79 KX |
1434 | } |
1435 | ||
1436 | return 0; | |
6aff67c8 | 1437 | error: |
ebb676da TG |
1438 | verbose("cannot pass map_type %d into func %s#%d\n", |
1439 | map->map_type, func_id_name(func_id), func_id); | |
6aff67c8 | 1440 | return -EINVAL; |
35578d79 KX |
1441 | } |
1442 | ||
435faee1 DB |
1443 | static int check_raw_mode(const struct bpf_func_proto *fn) |
1444 | { | |
1445 | int count = 0; | |
1446 | ||
39f19ebb | 1447 | if (fn->arg1_type == ARG_PTR_TO_UNINIT_MEM) |
435faee1 | 1448 | count++; |
39f19ebb | 1449 | if (fn->arg2_type == ARG_PTR_TO_UNINIT_MEM) |
435faee1 | 1450 | count++; |
39f19ebb | 1451 | if (fn->arg3_type == ARG_PTR_TO_UNINIT_MEM) |
435faee1 | 1452 | count++; |
39f19ebb | 1453 | if (fn->arg4_type == ARG_PTR_TO_UNINIT_MEM) |
435faee1 | 1454 | count++; |
39f19ebb | 1455 | if (fn->arg5_type == ARG_PTR_TO_UNINIT_MEM) |
435faee1 DB |
1456 | count++; |
1457 | ||
1458 | return count > 1 ? -EINVAL : 0; | |
1459 | } | |
1460 | ||
f1174f77 EC |
1461 | /* Packet data might have moved, any old PTR_TO_PACKET[_END] are now invalid, |
1462 | * so turn them into unknown SCALAR_VALUE. | |
1463 | */ | |
58e2af8b | 1464 | static void clear_all_pkt_pointers(struct bpf_verifier_env *env) |
969bf05e | 1465 | { |
58e2af8b JK |
1466 | struct bpf_verifier_state *state = &env->cur_state; |
1467 | struct bpf_reg_state *regs = state->regs, *reg; | |
969bf05e AS |
1468 | int i; |
1469 | ||
1470 | for (i = 0; i < MAX_BPF_REG; i++) | |
1471 | if (regs[i].type == PTR_TO_PACKET || | |
1472 | regs[i].type == PTR_TO_PACKET_END) | |
f1174f77 | 1473 | mark_reg_unknown(regs, i); |
969bf05e AS |
1474 | |
1475 | for (i = 0; i < MAX_BPF_STACK; i += BPF_REG_SIZE) { | |
1476 | if (state->stack_slot_type[i] != STACK_SPILL) | |
1477 | continue; | |
1478 | reg = &state->spilled_regs[i / BPF_REG_SIZE]; | |
1479 | if (reg->type != PTR_TO_PACKET && | |
1480 | reg->type != PTR_TO_PACKET_END) | |
1481 | continue; | |
f1174f77 | 1482 | __mark_reg_unknown(reg); |
969bf05e AS |
1483 | } |
1484 | } | |
1485 | ||
81ed18ab | 1486 | static int check_call(struct bpf_verifier_env *env, int func_id, int insn_idx) |
17a52670 | 1487 | { |
58e2af8b | 1488 | struct bpf_verifier_state *state = &env->cur_state; |
17a52670 | 1489 | const struct bpf_func_proto *fn = NULL; |
58e2af8b | 1490 | struct bpf_reg_state *regs = state->regs; |
33ff9823 | 1491 | struct bpf_call_arg_meta meta; |
969bf05e | 1492 | bool changes_data; |
17a52670 AS |
1493 | int i, err; |
1494 | ||
1495 | /* find function prototype */ | |
1496 | if (func_id < 0 || func_id >= __BPF_FUNC_MAX_ID) { | |
ebb676da | 1497 | verbose("invalid func %s#%d\n", func_id_name(func_id), func_id); |
17a52670 AS |
1498 | return -EINVAL; |
1499 | } | |
1500 | ||
1501 | if (env->prog->aux->ops->get_func_proto) | |
1502 | fn = env->prog->aux->ops->get_func_proto(func_id); | |
1503 | ||
1504 | if (!fn) { | |
ebb676da | 1505 | verbose("unknown func %s#%d\n", func_id_name(func_id), func_id); |
17a52670 AS |
1506 | return -EINVAL; |
1507 | } | |
1508 | ||
1509 | /* eBPF programs must be GPL compatible to use GPL-ed functions */ | |
24701ece | 1510 | if (!env->prog->gpl_compatible && fn->gpl_only) { |
17a52670 AS |
1511 | verbose("cannot call GPL only function from proprietary program\n"); |
1512 | return -EINVAL; | |
1513 | } | |
1514 | ||
17bedab2 | 1515 | changes_data = bpf_helper_changes_pkt_data(fn->func); |
969bf05e | 1516 | |
33ff9823 | 1517 | memset(&meta, 0, sizeof(meta)); |
36bbef52 | 1518 | meta.pkt_access = fn->pkt_access; |
33ff9823 | 1519 | |
435faee1 DB |
1520 | /* We only support one arg being in raw mode at the moment, which |
1521 | * is sufficient for the helper functions we have right now. | |
1522 | */ | |
1523 | err = check_raw_mode(fn); | |
1524 | if (err) { | |
ebb676da TG |
1525 | verbose("kernel subsystem misconfigured func %s#%d\n", |
1526 | func_id_name(func_id), func_id); | |
435faee1 DB |
1527 | return err; |
1528 | } | |
1529 | ||
17a52670 | 1530 | /* check args */ |
33ff9823 | 1531 | err = check_func_arg(env, BPF_REG_1, fn->arg1_type, &meta); |
17a52670 AS |
1532 | if (err) |
1533 | return err; | |
33ff9823 | 1534 | err = check_func_arg(env, BPF_REG_2, fn->arg2_type, &meta); |
17a52670 AS |
1535 | if (err) |
1536 | return err; | |
33ff9823 | 1537 | err = check_func_arg(env, BPF_REG_3, fn->arg3_type, &meta); |
17a52670 AS |
1538 | if (err) |
1539 | return err; | |
33ff9823 | 1540 | err = check_func_arg(env, BPF_REG_4, fn->arg4_type, &meta); |
17a52670 AS |
1541 | if (err) |
1542 | return err; | |
33ff9823 | 1543 | err = check_func_arg(env, BPF_REG_5, fn->arg5_type, &meta); |
17a52670 AS |
1544 | if (err) |
1545 | return err; | |
1546 | ||
435faee1 DB |
1547 | /* Mark slots with STACK_MISC in case of raw mode, stack offset |
1548 | * is inferred from register state. | |
1549 | */ | |
1550 | for (i = 0; i < meta.access_size; i++) { | |
31fd8581 | 1551 | err = check_mem_access(env, insn_idx, meta.regno, i, BPF_B, BPF_WRITE, -1); |
435faee1 DB |
1552 | if (err) |
1553 | return err; | |
1554 | } | |
1555 | ||
17a52670 | 1556 | /* reset caller saved regs */ |
a9789ef9 DB |
1557 | for (i = 0; i < CALLER_SAVED_REGS; i++) |
1558 | mark_reg_not_init(regs, caller_saved[i]); | |
17a52670 AS |
1559 | |
1560 | /* update return register */ | |
1561 | if (fn->ret_type == RET_INTEGER) { | |
f1174f77 EC |
1562 | /* sets type to SCALAR_VALUE */ |
1563 | mark_reg_unknown(regs, BPF_REG_0); | |
17a52670 AS |
1564 | } else if (fn->ret_type == RET_VOID) { |
1565 | regs[BPF_REG_0].type = NOT_INIT; | |
1566 | } else if (fn->ret_type == RET_PTR_TO_MAP_VALUE_OR_NULL) { | |
fad73a1a MKL |
1567 | struct bpf_insn_aux_data *insn_aux; |
1568 | ||
17a52670 | 1569 | regs[BPF_REG_0].type = PTR_TO_MAP_VALUE_OR_NULL; |
f1174f77 EC |
1570 | /* There is no offset yet applied, variable or fixed */ |
1571 | mark_reg_known_zero(regs, BPF_REG_0); | |
1572 | regs[BPF_REG_0].off = 0; | |
17a52670 AS |
1573 | /* remember map_ptr, so that check_map_access() |
1574 | * can check 'value_size' boundary of memory access | |
1575 | * to map element returned from bpf_map_lookup_elem() | |
1576 | */ | |
33ff9823 | 1577 | if (meta.map_ptr == NULL) { |
17a52670 AS |
1578 | verbose("kernel subsystem misconfigured verifier\n"); |
1579 | return -EINVAL; | |
1580 | } | |
33ff9823 | 1581 | regs[BPF_REG_0].map_ptr = meta.map_ptr; |
57a09bf0 | 1582 | regs[BPF_REG_0].id = ++env->id_gen; |
fad73a1a MKL |
1583 | insn_aux = &env->insn_aux_data[insn_idx]; |
1584 | if (!insn_aux->map_ptr) | |
1585 | insn_aux->map_ptr = meta.map_ptr; | |
1586 | else if (insn_aux->map_ptr != meta.map_ptr) | |
1587 | insn_aux->map_ptr = BPF_MAP_PTR_POISON; | |
17a52670 | 1588 | } else { |
ebb676da TG |
1589 | verbose("unknown return type %d of func %s#%d\n", |
1590 | fn->ret_type, func_id_name(func_id), func_id); | |
17a52670 AS |
1591 | return -EINVAL; |
1592 | } | |
04fd61ab | 1593 | |
33ff9823 | 1594 | err = check_map_func_compatibility(meta.map_ptr, func_id); |
35578d79 KX |
1595 | if (err) |
1596 | return err; | |
04fd61ab | 1597 | |
969bf05e AS |
1598 | if (changes_data) |
1599 | clear_all_pkt_pointers(env); | |
1600 | return 0; | |
1601 | } | |
1602 | ||
f1174f77 | 1603 | static void check_reg_overflow(struct bpf_reg_state *reg) |
969bf05e | 1604 | { |
f1174f77 EC |
1605 | if (reg->max_value > BPF_REGISTER_MAX_RANGE) |
1606 | reg->max_value = BPF_REGISTER_MAX_RANGE; | |
1607 | if (reg->min_value < BPF_REGISTER_MIN_RANGE || | |
1608 | reg->min_value > BPF_REGISTER_MAX_RANGE) | |
1609 | reg->min_value = BPF_REGISTER_MIN_RANGE; | |
1610 | } | |
d1174416 | 1611 | |
f1174f77 EC |
1612 | static void coerce_reg_to_32(struct bpf_reg_state *reg) |
1613 | { | |
1614 | /* 32-bit values can't be negative as an s64 */ | |
1615 | if (reg->min_value < 0) | |
1616 | reg->min_value = 0; | |
1617 | /* clear high 32 bits */ | |
1618 | reg->var_off = tnum_cast(reg->var_off, 4); | |
1619 | /* Did value become known? Then update bounds */ | |
1620 | if (tnum_is_const(reg->var_off)) { | |
1621 | if ((s64)reg->var_off.value > BPF_REGISTER_MIN_RANGE) | |
1622 | reg->min_value = reg->var_off.value; | |
1623 | if (reg->var_off.value < BPF_REGISTER_MAX_RANGE) | |
1624 | reg->max_value = reg->var_off.value; | |
969bf05e | 1625 | } |
969bf05e AS |
1626 | } |
1627 | ||
f1174f77 EC |
1628 | /* Handles arithmetic on a pointer and a scalar: computes new min/max and var_off. |
1629 | * Caller must check_reg_overflow all argument regs beforehand. | |
1630 | * Caller should also handle BPF_MOV case separately. | |
1631 | * If we return -EACCES, caller may want to try again treating pointer as a | |
1632 | * scalar. So we only emit a diagnostic if !env->allow_ptr_leaks. | |
1633 | */ | |
1634 | static int adjust_ptr_min_max_vals(struct bpf_verifier_env *env, | |
1635 | struct bpf_insn *insn, | |
1636 | const struct bpf_reg_state *ptr_reg, | |
1637 | const struct bpf_reg_state *off_reg) | |
969bf05e | 1638 | { |
f1174f77 EC |
1639 | struct bpf_reg_state *regs = env->cur_state.regs, *dst_reg; |
1640 | bool known = tnum_is_const(off_reg->var_off); | |
1641 | s64 min_val = off_reg->min_value; | |
1642 | u64 max_val = off_reg->max_value; | |
969bf05e | 1643 | u8 opcode = BPF_OP(insn->code); |
f1174f77 | 1644 | u32 dst = insn->dst_reg; |
969bf05e | 1645 | |
f1174f77 | 1646 | dst_reg = ®s[dst]; |
969bf05e | 1647 | |
f1174f77 EC |
1648 | if (WARN_ON_ONCE(known && (min_val != max_val))) { |
1649 | print_verifier_state(&env->cur_state); | |
1650 | verbose("verifier internal error\n"); | |
1651 | return -EINVAL; | |
1652 | } | |
1653 | ||
1654 | if (BPF_CLASS(insn->code) != BPF_ALU64) { | |
1655 | /* 32-bit ALU ops on pointers produce (meaningless) scalars */ | |
1656 | if (!env->allow_ptr_leaks) | |
1657 | verbose("R%d 32-bit pointer arithmetic prohibited\n", | |
1658 | dst); | |
1659 | return -EACCES; | |
969bf05e AS |
1660 | } |
1661 | ||
f1174f77 EC |
1662 | if (ptr_reg->type == PTR_TO_MAP_VALUE_OR_NULL) { |
1663 | if (!env->allow_ptr_leaks) | |
1664 | verbose("R%d pointer arithmetic on PTR_TO_MAP_VALUE_OR_NULL prohibited, null-check it first\n", | |
1665 | dst); | |
1666 | return -EACCES; | |
1667 | } | |
1668 | if (ptr_reg->type == CONST_PTR_TO_MAP) { | |
1669 | if (!env->allow_ptr_leaks) | |
1670 | verbose("R%d pointer arithmetic on CONST_PTR_TO_MAP prohibited\n", | |
1671 | dst); | |
1672 | return -EACCES; | |
1673 | } | |
1674 | if (ptr_reg->type == PTR_TO_PACKET_END) { | |
1675 | if (!env->allow_ptr_leaks) | |
1676 | verbose("R%d pointer arithmetic on PTR_TO_PACKET_END prohibited\n", | |
1677 | dst); | |
1678 | return -EACCES; | |
1679 | } | |
1680 | ||
1681 | /* In case of 'scalar += pointer', dst_reg inherits pointer type and id. | |
1682 | * The id may be overwritten later if we create a new variable offset. | |
969bf05e | 1683 | */ |
f1174f77 EC |
1684 | dst_reg->type = ptr_reg->type; |
1685 | dst_reg->id = ptr_reg->id; | |
969bf05e | 1686 | |
f1174f77 EC |
1687 | switch (opcode) { |
1688 | case BPF_ADD: | |
1689 | /* We can take a fixed offset as long as it doesn't overflow | |
1690 | * the s32 'off' field | |
969bf05e | 1691 | */ |
f1174f77 EC |
1692 | if (known && (ptr_reg->off + min_val == |
1693 | (s64)(s32)(ptr_reg->off + min_val))) { | |
1694 | /* pointer += K. Accumulate it into fixed offset */ | |
1695 | dst_reg->min_value = ptr_reg->min_value; | |
1696 | dst_reg->max_value = ptr_reg->max_value; | |
1697 | dst_reg->var_off = ptr_reg->var_off; | |
1698 | dst_reg->off = ptr_reg->off + min_val; | |
1699 | dst_reg->range = ptr_reg->range; | |
1700 | break; | |
1701 | } | |
1702 | if (max_val == BPF_REGISTER_MAX_RANGE) { | |
1703 | if (!env->allow_ptr_leaks) | |
1704 | verbose("R%d tried to add unbounded value to pointer\n", | |
1705 | dst); | |
1706 | return -EACCES; | |
1707 | } | |
1708 | /* A new variable offset is created. Note that off_reg->off | |
1709 | * == 0, since it's a scalar. | |
1710 | * dst_reg gets the pointer type and since some positive | |
1711 | * integer value was added to the pointer, give it a new 'id' | |
1712 | * if it's a PTR_TO_PACKET. | |
1713 | * this creates a new 'base' pointer, off_reg (variable) gets | |
1714 | * added into the variable offset, and we copy the fixed offset | |
1715 | * from ptr_reg. | |
969bf05e | 1716 | */ |
f1174f77 EC |
1717 | if (min_val <= BPF_REGISTER_MIN_RANGE) |
1718 | dst_reg->min_value = BPF_REGISTER_MIN_RANGE; | |
1719 | if (dst_reg->min_value != BPF_REGISTER_MIN_RANGE) | |
1720 | dst_reg->min_value += min_val; | |
1721 | if (dst_reg->max_value != BPF_REGISTER_MAX_RANGE) | |
1722 | dst_reg->max_value += max_val; | |
1723 | dst_reg->var_off = tnum_add(ptr_reg->var_off, off_reg->var_off); | |
1724 | dst_reg->off = ptr_reg->off; | |
1725 | if (ptr_reg->type == PTR_TO_PACKET) { | |
1726 | dst_reg->id = ++env->id_gen; | |
1727 | /* something was added to pkt_ptr, set range to zero */ | |
1728 | dst_reg->range = 0; | |
1729 | } | |
1730 | break; | |
1731 | case BPF_SUB: | |
1732 | if (dst_reg == off_reg) { | |
1733 | /* scalar -= pointer. Creates an unknown scalar */ | |
1734 | if (!env->allow_ptr_leaks) | |
1735 | verbose("R%d tried to subtract pointer from scalar\n", | |
1736 | dst); | |
1737 | return -EACCES; | |
1738 | } | |
1739 | /* We don't allow subtraction from FP, because (according to | |
1740 | * test_verifier.c test "invalid fp arithmetic", JITs might not | |
1741 | * be able to deal with it. | |
969bf05e | 1742 | */ |
f1174f77 EC |
1743 | if (ptr_reg->type == PTR_TO_STACK) { |
1744 | if (!env->allow_ptr_leaks) | |
1745 | verbose("R%d subtraction from stack pointer prohibited\n", | |
1746 | dst); | |
1747 | return -EACCES; | |
1748 | } | |
1749 | if (known && (ptr_reg->off - min_val == | |
1750 | (s64)(s32)(ptr_reg->off - min_val))) { | |
1751 | /* pointer -= K. Subtract it from fixed offset */ | |
1752 | dst_reg->min_value = ptr_reg->min_value; | |
1753 | dst_reg->max_value = ptr_reg->max_value; | |
1754 | dst_reg->var_off = ptr_reg->var_off; | |
1755 | dst_reg->id = ptr_reg->id; | |
1756 | dst_reg->off = ptr_reg->off - min_val; | |
1757 | dst_reg->range = ptr_reg->range; | |
1758 | break; | |
1759 | } | |
1760 | /* Subtracting a negative value will just confuse everything. | |
1761 | * This can happen if off_reg is an immediate. | |
969bf05e | 1762 | */ |
f1174f77 EC |
1763 | if ((s64)max_val < 0) { |
1764 | if (!env->allow_ptr_leaks) | |
1765 | verbose("R%d tried to subtract negative max_val %lld from pointer\n", | |
1766 | dst, (s64)max_val); | |
1767 | return -EACCES; | |
1768 | } | |
1769 | /* A new variable offset is created. If the subtrahend is known | |
1770 | * nonnegative, then any reg->range we had before is still good. | |
969bf05e | 1771 | */ |
f1174f77 EC |
1772 | if (max_val >= BPF_REGISTER_MAX_RANGE) |
1773 | dst_reg->min_value = BPF_REGISTER_MIN_RANGE; | |
1774 | if (dst_reg->min_value != BPF_REGISTER_MIN_RANGE) | |
1775 | dst_reg->min_value -= max_val; | |
1776 | if (min_val <= BPF_REGISTER_MIN_RANGE) | |
1777 | dst_reg->max_value = BPF_REGISTER_MAX_RANGE; | |
1778 | if (dst_reg->max_value != BPF_REGISTER_MAX_RANGE) | |
1779 | dst_reg->max_value -= min_val; | |
1780 | dst_reg->var_off = tnum_sub(ptr_reg->var_off, off_reg->var_off); | |
1781 | dst_reg->off = ptr_reg->off; | |
1782 | if (ptr_reg->type == PTR_TO_PACKET) { | |
1783 | dst_reg->id = ++env->id_gen; | |
1784 | /* something was added to pkt_ptr, set range to zero */ | |
1785 | if (min_val < 0) | |
1786 | dst_reg->range = 0; | |
43188702 | 1787 | } |
f1174f77 EC |
1788 | break; |
1789 | case BPF_AND: | |
1790 | case BPF_OR: | |
1791 | case BPF_XOR: | |
1792 | /* bitwise ops on pointers are troublesome, prohibit for now. | |
1793 | * (However, in principle we could allow some cases, e.g. | |
1794 | * ptr &= ~3 which would reduce min_value by 3.) | |
1795 | */ | |
1796 | if (!env->allow_ptr_leaks) | |
1797 | verbose("R%d bitwise operator %s on pointer prohibited\n", | |
1798 | dst, bpf_alu_string[opcode >> 4]); | |
1799 | return -EACCES; | |
1800 | default: | |
1801 | /* other operators (e.g. MUL,LSH) produce non-pointer results */ | |
1802 | if (!env->allow_ptr_leaks) | |
1803 | verbose("R%d pointer arithmetic with %s operator prohibited\n", | |
1804 | dst, bpf_alu_string[opcode >> 4]); | |
1805 | return -EACCES; | |
43188702 JF |
1806 | } |
1807 | ||
f1174f77 | 1808 | check_reg_overflow(dst_reg); |
43188702 JF |
1809 | return 0; |
1810 | } | |
1811 | ||
f1174f77 EC |
1812 | static int adjust_scalar_min_max_vals(struct bpf_verifier_env *env, |
1813 | struct bpf_insn *insn, | |
1814 | struct bpf_reg_state *dst_reg, | |
1815 | struct bpf_reg_state src_reg) | |
969bf05e | 1816 | { |
58e2af8b | 1817 | struct bpf_reg_state *regs = env->cur_state.regs; |
f23cc643 JB |
1818 | s64 min_val = BPF_REGISTER_MIN_RANGE; |
1819 | u64 max_val = BPF_REGISTER_MAX_RANGE; | |
48461135 | 1820 | u8 opcode = BPF_OP(insn->code); |
f1174f77 | 1821 | bool src_known, dst_known; |
48461135 | 1822 | |
f1174f77 EC |
1823 | if (BPF_CLASS(insn->code) != BPF_ALU64) { |
1824 | /* 32-bit ALU ops are (32,32)->64 */ | |
1825 | coerce_reg_to_32(dst_reg); | |
1826 | coerce_reg_to_32(&src_reg); | |
9305706c | 1827 | } |
f1174f77 EC |
1828 | min_val = src_reg.min_value; |
1829 | max_val = src_reg.max_value; | |
1830 | src_known = tnum_is_const(src_reg.var_off); | |
1831 | dst_known = tnum_is_const(dst_reg->var_off); | |
f23cc643 | 1832 | |
48461135 JB |
1833 | switch (opcode) { |
1834 | case BPF_ADD: | |
f1174f77 EC |
1835 | if (min_val == BPF_REGISTER_MIN_RANGE) |
1836 | dst_reg->min_value = BPF_REGISTER_MIN_RANGE; | |
f23cc643 JB |
1837 | if (dst_reg->min_value != BPF_REGISTER_MIN_RANGE) |
1838 | dst_reg->min_value += min_val; | |
f1174f77 | 1839 | /* if max_val is MAX_RANGE, this will saturate dst->max */ |
f23cc643 JB |
1840 | if (dst_reg->max_value != BPF_REGISTER_MAX_RANGE) |
1841 | dst_reg->max_value += max_val; | |
f1174f77 | 1842 | dst_reg->var_off = tnum_add(dst_reg->var_off, src_reg.var_off); |
48461135 JB |
1843 | break; |
1844 | case BPF_SUB: | |
9305706c EC |
1845 | if (max_val == BPF_REGISTER_MAX_RANGE) |
1846 | dst_reg->min_value = BPF_REGISTER_MIN_RANGE; | |
f23cc643 | 1847 | if (dst_reg->min_value != BPF_REGISTER_MIN_RANGE) |
9305706c | 1848 | dst_reg->min_value -= max_val; |
f1174f77 EC |
1849 | if (min_val == BPF_REGISTER_MIN_RANGE) |
1850 | dst_reg->max_value = BPF_REGISTER_MAX_RANGE; | |
f23cc643 | 1851 | if (dst_reg->max_value != BPF_REGISTER_MAX_RANGE) |
9305706c | 1852 | dst_reg->max_value -= min_val; |
f1174f77 | 1853 | dst_reg->var_off = tnum_sub(dst_reg->var_off, src_reg.var_off); |
48461135 JB |
1854 | break; |
1855 | case BPF_MUL: | |
f1174f77 EC |
1856 | if (min_val < 0 || dst_reg->min_value < 0) { |
1857 | /* Ain't nobody got time to multiply that sign */ | |
1858 | __mark_reg_unknown(dst_reg); | |
1859 | break; | |
1860 | } | |
1861 | dst_reg->min_value *= min_val; | |
1862 | /* if max_val is MAX_RANGE, this will saturate dst->max. | |
1863 | * We know MAX_RANGE ** 2 won't overflow a u64, because | |
1864 | * MAX_RANGE itself fits in a u32. | |
1865 | */ | |
1866 | BUILD_BUG_ON(BPF_REGISTER_MAX_RANGE > (u32)-1); | |
f23cc643 JB |
1867 | if (dst_reg->max_value != BPF_REGISTER_MAX_RANGE) |
1868 | dst_reg->max_value *= max_val; | |
f1174f77 | 1869 | dst_reg->var_off = tnum_mul(dst_reg->var_off, src_reg.var_off); |
48461135 JB |
1870 | break; |
1871 | case BPF_AND: | |
f1174f77 EC |
1872 | if (src_known && dst_known) { |
1873 | u64 value = dst_reg->var_off.value & src_reg.var_off.value; | |
1874 | ||
1875 | dst_reg->var_off = tnum_const(value); | |
1876 | dst_reg->min_value = dst_reg->max_value = min_t(u64, | |
1877 | value, BPF_REGISTER_MAX_RANGE); | |
1878 | break; | |
1879 | } | |
1880 | /* Lose min_value when AND'ing negative numbers, ain't nobody | |
1881 | * got time for that. Otherwise we get our minimum from the | |
1882 | * var_off, since that's inherently bitwise. | |
1883 | * Our maximum is the minimum of the operands' maxima. | |
f23cc643 | 1884 | */ |
f1174f77 EC |
1885 | dst_reg->var_off = tnum_and(dst_reg->var_off, src_reg.var_off); |
1886 | if (min_val < 0 && dst_reg->min_value < 0) | |
f23cc643 JB |
1887 | dst_reg->min_value = BPF_REGISTER_MIN_RANGE; |
1888 | else | |
f1174f77 EC |
1889 | dst_reg->min_value = dst_reg->var_off.value; |
1890 | dst_reg->max_value = min(dst_reg->max_value, max_val); | |
1891 | break; | |
1892 | case BPF_OR: | |
1893 | if (src_known && dst_known) { | |
1894 | u64 value = dst_reg->var_off.value | src_reg.var_off.value; | |
1895 | ||
1896 | dst_reg->var_off = tnum_const(value); | |
1897 | dst_reg->min_value = dst_reg->max_value = min_t(u64, | |
1898 | value, BPF_REGISTER_MAX_RANGE); | |
1899 | break; | |
1900 | } | |
1901 | /* Lose ranges when OR'ing negative numbers, ain't nobody got | |
1902 | * time for that. Otherwise we get our maximum from the var_off, | |
1903 | * and our minimum is the maximum of the operands' minima. | |
1904 | */ | |
1905 | dst_reg->var_off = tnum_or(dst_reg->var_off, src_reg.var_off); | |
1906 | if (min_val < 0 || dst_reg->min_value < 0) { | |
1907 | dst_reg->min_value = BPF_REGISTER_MIN_RANGE; | |
1908 | dst_reg->max_value = BPF_REGISTER_MAX_RANGE; | |
1909 | } else { | |
1910 | dst_reg->min_value = max(dst_reg->min_value, min_val); | |
1911 | dst_reg->max_value = dst_reg->var_off.value | dst_reg->var_off.mask; | |
1912 | } | |
48461135 JB |
1913 | break; |
1914 | case BPF_LSH: | |
f1174f77 EC |
1915 | if (min_val < 0) { |
1916 | /* LSH by a negative number is undefined */ | |
1917 | mark_reg_unknown(regs, insn->dst_reg); | |
1918 | break; | |
1919 | } | |
48461135 JB |
1920 | /* Gotta have special overflow logic here, if we're shifting |
1921 | * more than MAX_RANGE then just assume we have an invalid | |
1922 | * range. | |
1923 | */ | |
d1174416 | 1924 | if (min_val > ilog2(BPF_REGISTER_MAX_RANGE)) { |
48461135 | 1925 | dst_reg->min_value = BPF_REGISTER_MIN_RANGE; |
f1174f77 | 1926 | dst_reg->var_off = tnum_unknown; |
d1174416 DM |
1927 | } else { |
1928 | if (dst_reg->min_value != BPF_REGISTER_MIN_RANGE) | |
1929 | dst_reg->min_value <<= min_val; | |
f1174f77 EC |
1930 | if (src_known) |
1931 | dst_reg->var_off = tnum_lshift(dst_reg->var_off, min_val); | |
1932 | else | |
1933 | dst_reg->var_off = tnum_lshift(tnum_unknown, min_val); | |
d1174416 | 1934 | } |
48461135 JB |
1935 | if (max_val > ilog2(BPF_REGISTER_MAX_RANGE)) |
1936 | dst_reg->max_value = BPF_REGISTER_MAX_RANGE; | |
f23cc643 | 1937 | else if (dst_reg->max_value != BPF_REGISTER_MAX_RANGE) |
48461135 JB |
1938 | dst_reg->max_value <<= max_val; |
1939 | break; | |
1940 | case BPF_RSH: | |
f1174f77 EC |
1941 | if (min_val < 0) { |
1942 | /* RSH by a negative number is undefined */ | |
1943 | mark_reg_unknown(regs, insn->dst_reg); | |
1944 | break; | |
1945 | } | |
1946 | /* BPF_RSH is an unsigned shift, so make the appropriate casts */ | |
1947 | if (dst_reg->min_value < 0) { | |
1948 | if (min_val) | |
1949 | /* Sign bit will be cleared */ | |
1950 | dst_reg->min_value = 0; | |
d1174416 | 1951 | } else { |
f23cc643 JB |
1952 | dst_reg->min_value = |
1953 | (u64)(dst_reg->min_value) >> min_val; | |
d1174416 | 1954 | } |
f1174f77 EC |
1955 | if (src_known) |
1956 | dst_reg->var_off = tnum_rshift(dst_reg->var_off, min_val); | |
1957 | else | |
1958 | dst_reg->var_off = tnum_rshift(tnum_unknown, min_val); | |
1959 | if (dst_reg->max_value == BPF_REGISTER_MAX_RANGE) | |
1960 | dst_reg->max_value = ~0; | |
1961 | dst_reg->max_value >>= max_val; | |
48461135 JB |
1962 | break; |
1963 | default: | |
f1174f77 | 1964 | mark_reg_unknown(regs, insn->dst_reg); |
48461135 JB |
1965 | break; |
1966 | } | |
1967 | ||
1968 | check_reg_overflow(dst_reg); | |
f1174f77 EC |
1969 | return 0; |
1970 | } | |
1971 | ||
1972 | /* Handles ALU ops other than BPF_END, BPF_NEG and BPF_MOV: computes new min/max | |
1973 | * and var_off. | |
1974 | */ | |
1975 | static int adjust_reg_min_max_vals(struct bpf_verifier_env *env, | |
1976 | struct bpf_insn *insn) | |
1977 | { | |
1978 | struct bpf_reg_state *regs = env->cur_state.regs, *dst_reg, *src_reg; | |
1979 | struct bpf_reg_state *ptr_reg = NULL, off_reg = {0}; | |
1980 | u8 opcode = BPF_OP(insn->code); | |
1981 | int rc; | |
1982 | ||
1983 | dst_reg = ®s[insn->dst_reg]; | |
1984 | check_reg_overflow(dst_reg); | |
1985 | src_reg = NULL; | |
1986 | if (dst_reg->type != SCALAR_VALUE) | |
1987 | ptr_reg = dst_reg; | |
1988 | if (BPF_SRC(insn->code) == BPF_X) { | |
1989 | src_reg = ®s[insn->src_reg]; | |
1990 | check_reg_overflow(src_reg); | |
1991 | ||
1992 | if (src_reg->type != SCALAR_VALUE) { | |
1993 | if (dst_reg->type != SCALAR_VALUE) { | |
1994 | /* Combining two pointers by any ALU op yields | |
1995 | * an arbitrary scalar. | |
1996 | */ | |
1997 | if (!env->allow_ptr_leaks) { | |
1998 | verbose("R%d pointer %s pointer prohibited\n", | |
1999 | insn->dst_reg, | |
2000 | bpf_alu_string[opcode >> 4]); | |
2001 | return -EACCES; | |
2002 | } | |
2003 | mark_reg_unknown(regs, insn->dst_reg); | |
2004 | return 0; | |
2005 | } else { | |
2006 | /* scalar += pointer | |
2007 | * This is legal, but we have to reverse our | |
2008 | * src/dest handling in computing the range | |
2009 | */ | |
2010 | rc = adjust_ptr_min_max_vals(env, insn, | |
2011 | src_reg, dst_reg); | |
2012 | if (rc == -EACCES && env->allow_ptr_leaks) { | |
2013 | /* scalar += unknown scalar */ | |
2014 | __mark_reg_unknown(&off_reg); | |
2015 | return adjust_scalar_min_max_vals( | |
2016 | env, insn, | |
2017 | dst_reg, off_reg); | |
2018 | } | |
2019 | return rc; | |
2020 | } | |
2021 | } else if (ptr_reg) { | |
2022 | /* pointer += scalar */ | |
2023 | rc = adjust_ptr_min_max_vals(env, insn, | |
2024 | dst_reg, src_reg); | |
2025 | if (rc == -EACCES && env->allow_ptr_leaks) { | |
2026 | /* unknown scalar += scalar */ | |
2027 | __mark_reg_unknown(dst_reg); | |
2028 | return adjust_scalar_min_max_vals( | |
2029 | env, insn, dst_reg, *src_reg); | |
2030 | } | |
2031 | return rc; | |
2032 | } | |
2033 | } else { | |
2034 | /* Pretend the src is a reg with a known value, since we only | |
2035 | * need to be able to read from this state. | |
2036 | */ | |
2037 | off_reg.type = SCALAR_VALUE; | |
2038 | off_reg.var_off = tnum_const(insn->imm); | |
2039 | off_reg.min_value = insn->imm; | |
2040 | off_reg.max_value = insn->imm; | |
2041 | src_reg = &off_reg; | |
2042 | check_reg_overflow(src_reg); | |
2043 | if (ptr_reg) { /* pointer += K */ | |
2044 | rc = adjust_ptr_min_max_vals(env, insn, | |
2045 | ptr_reg, src_reg); | |
2046 | if (rc == -EACCES && env->allow_ptr_leaks) { | |
2047 | /* unknown scalar += K */ | |
2048 | __mark_reg_unknown(dst_reg); | |
2049 | return adjust_scalar_min_max_vals( | |
2050 | env, insn, dst_reg, off_reg); | |
2051 | } | |
2052 | return rc; | |
2053 | } | |
2054 | } | |
2055 | ||
2056 | /* Got here implies adding two SCALAR_VALUEs */ | |
2057 | if (WARN_ON_ONCE(ptr_reg)) { | |
2058 | print_verifier_state(&env->cur_state); | |
2059 | verbose("verifier internal error: unexpected ptr_reg\n"); | |
2060 | return -EINVAL; | |
2061 | } | |
2062 | if (WARN_ON(!src_reg)) { | |
2063 | print_verifier_state(&env->cur_state); | |
2064 | verbose("verifier internal error: no src_reg\n"); | |
2065 | return -EINVAL; | |
2066 | } | |
2067 | return adjust_scalar_min_max_vals(env, insn, dst_reg, *src_reg); | |
48461135 JB |
2068 | } |
2069 | ||
17a52670 | 2070 | /* check validity of 32-bit and 64-bit arithmetic operations */ |
58e2af8b | 2071 | static int check_alu_op(struct bpf_verifier_env *env, struct bpf_insn *insn) |
17a52670 | 2072 | { |
f1174f77 | 2073 | struct bpf_reg_state *regs = env->cur_state.regs; |
17a52670 AS |
2074 | u8 opcode = BPF_OP(insn->code); |
2075 | int err; | |
2076 | ||
2077 | if (opcode == BPF_END || opcode == BPF_NEG) { | |
2078 | if (opcode == BPF_NEG) { | |
2079 | if (BPF_SRC(insn->code) != 0 || | |
2080 | insn->src_reg != BPF_REG_0 || | |
2081 | insn->off != 0 || insn->imm != 0) { | |
2082 | verbose("BPF_NEG uses reserved fields\n"); | |
2083 | return -EINVAL; | |
2084 | } | |
2085 | } else { | |
2086 | if (insn->src_reg != BPF_REG_0 || insn->off != 0 || | |
2087 | (insn->imm != 16 && insn->imm != 32 && insn->imm != 64)) { | |
2088 | verbose("BPF_END uses reserved fields\n"); | |
2089 | return -EINVAL; | |
2090 | } | |
2091 | } | |
2092 | ||
2093 | /* check src operand */ | |
2094 | err = check_reg_arg(regs, insn->dst_reg, SRC_OP); | |
2095 | if (err) | |
2096 | return err; | |
2097 | ||
1be7f75d AS |
2098 | if (is_pointer_value(env, insn->dst_reg)) { |
2099 | verbose("R%d pointer arithmetic prohibited\n", | |
2100 | insn->dst_reg); | |
2101 | return -EACCES; | |
2102 | } | |
2103 | ||
17a52670 AS |
2104 | /* check dest operand */ |
2105 | err = check_reg_arg(regs, insn->dst_reg, DST_OP); | |
2106 | if (err) | |
2107 | return err; | |
2108 | ||
2109 | } else if (opcode == BPF_MOV) { | |
2110 | ||
2111 | if (BPF_SRC(insn->code) == BPF_X) { | |
2112 | if (insn->imm != 0 || insn->off != 0) { | |
2113 | verbose("BPF_MOV uses reserved fields\n"); | |
2114 | return -EINVAL; | |
2115 | } | |
2116 | ||
2117 | /* check src operand */ | |
2118 | err = check_reg_arg(regs, insn->src_reg, SRC_OP); | |
2119 | if (err) | |
2120 | return err; | |
2121 | } else { | |
2122 | if (insn->src_reg != BPF_REG_0 || insn->off != 0) { | |
2123 | verbose("BPF_MOV uses reserved fields\n"); | |
2124 | return -EINVAL; | |
2125 | } | |
2126 | } | |
2127 | ||
2128 | /* check dest operand */ | |
2129 | err = check_reg_arg(regs, insn->dst_reg, DST_OP); | |
2130 | if (err) | |
2131 | return err; | |
2132 | ||
2133 | if (BPF_SRC(insn->code) == BPF_X) { | |
2134 | if (BPF_CLASS(insn->code) == BPF_ALU64) { | |
2135 | /* case: R1 = R2 | |
2136 | * copy register state to dest reg | |
2137 | */ | |
2138 | regs[insn->dst_reg] = regs[insn->src_reg]; | |
2139 | } else { | |
f1174f77 | 2140 | /* R1 = (u32) R2 */ |
1be7f75d AS |
2141 | if (is_pointer_value(env, insn->src_reg)) { |
2142 | verbose("R%d partial copy of pointer\n", | |
2143 | insn->src_reg); | |
2144 | return -EACCES; | |
2145 | } | |
f1174f77 EC |
2146 | mark_reg_unknown(regs, insn->dst_reg); |
2147 | /* high 32 bits are known zero. But this is | |
2148 | * still out of range for max_value, so leave | |
2149 | * that. | |
2150 | */ | |
2151 | regs[insn->dst_reg].var_off = tnum_cast( | |
2152 | regs[insn->dst_reg].var_off, 4); | |
17a52670 AS |
2153 | } |
2154 | } else { | |
2155 | /* case: R = imm | |
2156 | * remember the value we stored into this reg | |
2157 | */ | |
f1174f77 EC |
2158 | regs[insn->dst_reg].type = SCALAR_VALUE; |
2159 | regs[insn->dst_reg].var_off = tnum_const(insn->imm); | |
48461135 JB |
2160 | regs[insn->dst_reg].max_value = insn->imm; |
2161 | regs[insn->dst_reg].min_value = insn->imm; | |
f1174f77 | 2162 | regs[insn->dst_reg].id = 0; |
17a52670 AS |
2163 | } |
2164 | ||
2165 | } else if (opcode > BPF_END) { | |
2166 | verbose("invalid BPF_ALU opcode %x\n", opcode); | |
2167 | return -EINVAL; | |
2168 | ||
2169 | } else { /* all other ALU ops: and, sub, xor, add, ... */ | |
2170 | ||
17a52670 AS |
2171 | if (BPF_SRC(insn->code) == BPF_X) { |
2172 | if (insn->imm != 0 || insn->off != 0) { | |
2173 | verbose("BPF_ALU uses reserved fields\n"); | |
2174 | return -EINVAL; | |
2175 | } | |
2176 | /* check src1 operand */ | |
2177 | err = check_reg_arg(regs, insn->src_reg, SRC_OP); | |
2178 | if (err) | |
2179 | return err; | |
2180 | } else { | |
2181 | if (insn->src_reg != BPF_REG_0 || insn->off != 0) { | |
2182 | verbose("BPF_ALU uses reserved fields\n"); | |
2183 | return -EINVAL; | |
2184 | } | |
2185 | } | |
2186 | ||
2187 | /* check src2 operand */ | |
2188 | err = check_reg_arg(regs, insn->dst_reg, SRC_OP); | |
2189 | if (err) | |
2190 | return err; | |
2191 | ||
2192 | if ((opcode == BPF_MOD || opcode == BPF_DIV) && | |
2193 | BPF_SRC(insn->code) == BPF_K && insn->imm == 0) { | |
2194 | verbose("div by zero\n"); | |
2195 | return -EINVAL; | |
2196 | } | |
2197 | ||
229394e8 RV |
2198 | if ((opcode == BPF_LSH || opcode == BPF_RSH || |
2199 | opcode == BPF_ARSH) && BPF_SRC(insn->code) == BPF_K) { | |
2200 | int size = BPF_CLASS(insn->code) == BPF_ALU64 ? 64 : 32; | |
2201 | ||
2202 | if (insn->imm < 0 || insn->imm >= size) { | |
2203 | verbose("invalid shift %d\n", insn->imm); | |
2204 | return -EINVAL; | |
2205 | } | |
2206 | } | |
2207 | ||
1a0dc1ac AS |
2208 | /* check dest operand */ |
2209 | err = check_reg_arg(regs, insn->dst_reg, DST_OP_NO_MARK); | |
2210 | if (err) | |
2211 | return err; | |
2212 | ||
f1174f77 | 2213 | return adjust_reg_min_max_vals(env, insn); |
17a52670 AS |
2214 | } |
2215 | ||
2216 | return 0; | |
2217 | } | |
2218 | ||
58e2af8b JK |
2219 | static void find_good_pkt_pointers(struct bpf_verifier_state *state, |
2220 | struct bpf_reg_state *dst_reg) | |
969bf05e | 2221 | { |
58e2af8b | 2222 | struct bpf_reg_state *regs = state->regs, *reg; |
969bf05e | 2223 | int i; |
2d2be8ca | 2224 | |
f1174f77 EC |
2225 | if (dst_reg->off < 0) |
2226 | /* This doesn't give us any range */ | |
2227 | return; | |
2228 | ||
2229 | if (dst_reg->max_value > MAX_PACKET_OFF || | |
2230 | dst_reg->max_value + dst_reg->off > MAX_PACKET_OFF) | |
2231 | /* Risk of overflow. For instance, ptr + (1<<63) may be less | |
2232 | * than pkt_end, but that's because it's also less than pkt. | |
2233 | */ | |
2234 | return; | |
2235 | ||
2d2be8ca DB |
2236 | /* LLVM can generate two kind of checks: |
2237 | * | |
2238 | * Type 1: | |
2239 | * | |
2240 | * r2 = r3; | |
2241 | * r2 += 8; | |
2242 | * if (r2 > pkt_end) goto <handle exception> | |
2243 | * <access okay> | |
2244 | * | |
2245 | * Where: | |
2246 | * r2 == dst_reg, pkt_end == src_reg | |
2247 | * r2=pkt(id=n,off=8,r=0) | |
2248 | * r3=pkt(id=n,off=0,r=0) | |
2249 | * | |
2250 | * Type 2: | |
2251 | * | |
2252 | * r2 = r3; | |
2253 | * r2 += 8; | |
2254 | * if (pkt_end >= r2) goto <access okay> | |
2255 | * <handle exception> | |
2256 | * | |
2257 | * Where: | |
2258 | * pkt_end == dst_reg, r2 == src_reg | |
2259 | * r2=pkt(id=n,off=8,r=0) | |
2260 | * r3=pkt(id=n,off=0,r=0) | |
2261 | * | |
2262 | * Find register r3 and mark its range as r3=pkt(id=n,off=0,r=8) | |
2263 | * so that range of bytes [r3, r3 + 8) is safe to access. | |
969bf05e | 2264 | */ |
2d2be8ca | 2265 | |
f1174f77 EC |
2266 | /* If our ids match, then we must have the same max_value. And we |
2267 | * don't care about the other reg's fixed offset, since if it's too big | |
2268 | * the range won't allow anything. | |
2269 | * dst_reg->off is known < MAX_PACKET_OFF, therefore it fits in a u16. | |
2270 | */ | |
969bf05e AS |
2271 | for (i = 0; i < MAX_BPF_REG; i++) |
2272 | if (regs[i].type == PTR_TO_PACKET && regs[i].id == dst_reg->id) | |
b1977682 | 2273 | /* keep the maximum range already checked */ |
f1174f77 | 2274 | regs[i].range = max_t(u16, regs[i].range, dst_reg->off); |
969bf05e AS |
2275 | |
2276 | for (i = 0; i < MAX_BPF_STACK; i += BPF_REG_SIZE) { | |
2277 | if (state->stack_slot_type[i] != STACK_SPILL) | |
2278 | continue; | |
2279 | reg = &state->spilled_regs[i / BPF_REG_SIZE]; | |
2280 | if (reg->type == PTR_TO_PACKET && reg->id == dst_reg->id) | |
f1174f77 | 2281 | reg->range = max_t(u16, reg->range, dst_reg->off); |
969bf05e AS |
2282 | } |
2283 | } | |
2284 | ||
48461135 JB |
2285 | /* Adjusts the register min/max values in the case that the dst_reg is the |
2286 | * variable register that we are working on, and src_reg is a constant or we're | |
2287 | * simply doing a BPF_K check. | |
f1174f77 | 2288 | * In JEQ/JNE cases we also adjust the var_off values. |
48461135 JB |
2289 | */ |
2290 | static void reg_set_min_max(struct bpf_reg_state *true_reg, | |
2291 | struct bpf_reg_state *false_reg, u64 val, | |
2292 | u8 opcode) | |
2293 | { | |
4cabc5b1 | 2294 | bool value_from_signed = true; |
f1174f77 EC |
2295 | |
2296 | /* If the dst_reg is a pointer, we can't learn anything about its | |
2297 | * variable offset from the compare (unless src_reg were a pointer into | |
2298 | * the same object, but we don't bother with that. | |
2299 | * Since false_reg and true_reg have the same type by construction, we | |
2300 | * only need to check one of them for pointerness. | |
2301 | */ | |
2302 | if (__is_pointer_value(false, false_reg)) | |
2303 | return; | |
4cabc5b1 | 2304 | |
48461135 JB |
2305 | switch (opcode) { |
2306 | case BPF_JEQ: | |
2307 | /* If this is false then we know nothing Jon Snow, but if it is | |
2308 | * true then we know for sure. | |
2309 | */ | |
2310 | true_reg->max_value = true_reg->min_value = val; | |
f1174f77 | 2311 | true_reg->var_off = tnum_const(val); |
48461135 JB |
2312 | break; |
2313 | case BPF_JNE: | |
2314 | /* If this is true we know nothing Jon Snow, but if it is false | |
2315 | * we know the value for sure; | |
2316 | */ | |
2317 | false_reg->max_value = false_reg->min_value = val; | |
f1174f77 | 2318 | false_reg->var_off = tnum_const(val); |
48461135 JB |
2319 | break; |
2320 | case BPF_JGT: | |
4cabc5b1 | 2321 | value_from_signed = false; |
7e57fbb2 | 2322 | /* fallthrough */ |
48461135 | 2323 | case BPF_JSGT: |
4cabc5b1 DB |
2324 | if (true_reg->value_from_signed != value_from_signed) |
2325 | reset_reg_range_values(true_reg, 0); | |
2326 | if (false_reg->value_from_signed != value_from_signed) | |
2327 | reset_reg_range_values(false_reg, 0); | |
2328 | if (opcode == BPF_JGT) { | |
2329 | /* Unsigned comparison, the minimum value is 0. */ | |
2330 | false_reg->min_value = 0; | |
2331 | } | |
48461135 JB |
2332 | /* If this is false then we know the maximum val is val, |
2333 | * otherwise we know the min val is val+1. | |
2334 | */ | |
2335 | false_reg->max_value = val; | |
4cabc5b1 | 2336 | false_reg->value_from_signed = value_from_signed; |
48461135 | 2337 | true_reg->min_value = val + 1; |
4cabc5b1 | 2338 | true_reg->value_from_signed = value_from_signed; |
48461135 JB |
2339 | break; |
2340 | case BPF_JGE: | |
4cabc5b1 | 2341 | value_from_signed = false; |
7e57fbb2 | 2342 | /* fallthrough */ |
48461135 | 2343 | case BPF_JSGE: |
4cabc5b1 DB |
2344 | if (true_reg->value_from_signed != value_from_signed) |
2345 | reset_reg_range_values(true_reg, 0); | |
2346 | if (false_reg->value_from_signed != value_from_signed) | |
2347 | reset_reg_range_values(false_reg, 0); | |
2348 | if (opcode == BPF_JGE) { | |
2349 | /* Unsigned comparison, the minimum value is 0. */ | |
2350 | false_reg->min_value = 0; | |
2351 | } | |
48461135 JB |
2352 | /* If this is false then we know the maximum value is val - 1, |
2353 | * otherwise we know the mimimum value is val. | |
2354 | */ | |
2355 | false_reg->max_value = val - 1; | |
4cabc5b1 | 2356 | false_reg->value_from_signed = value_from_signed; |
48461135 | 2357 | true_reg->min_value = val; |
4cabc5b1 | 2358 | true_reg->value_from_signed = value_from_signed; |
48461135 JB |
2359 | break; |
2360 | default: | |
2361 | break; | |
2362 | } | |
2363 | ||
2364 | check_reg_overflow(false_reg); | |
2365 | check_reg_overflow(true_reg); | |
2366 | } | |
2367 | ||
f1174f77 EC |
2368 | /* Same as above, but for the case that dst_reg holds a constant and src_reg is |
2369 | * the variable reg. | |
48461135 JB |
2370 | */ |
2371 | static void reg_set_min_max_inv(struct bpf_reg_state *true_reg, | |
2372 | struct bpf_reg_state *false_reg, u64 val, | |
2373 | u8 opcode) | |
2374 | { | |
4cabc5b1 | 2375 | bool value_from_signed = true; |
f1174f77 EC |
2376 | |
2377 | if (__is_pointer_value(false, false_reg)) | |
2378 | return; | |
4cabc5b1 | 2379 | |
48461135 JB |
2380 | switch (opcode) { |
2381 | case BPF_JEQ: | |
2382 | /* If this is false then we know nothing Jon Snow, but if it is | |
2383 | * true then we know for sure. | |
2384 | */ | |
2385 | true_reg->max_value = true_reg->min_value = val; | |
f1174f77 | 2386 | true_reg->var_off = tnum_const(val); |
48461135 JB |
2387 | break; |
2388 | case BPF_JNE: | |
2389 | /* If this is true we know nothing Jon Snow, but if it is false | |
2390 | * we know the value for sure; | |
2391 | */ | |
2392 | false_reg->max_value = false_reg->min_value = val; | |
f1174f77 | 2393 | false_reg->var_off = tnum_const(val); |
48461135 JB |
2394 | break; |
2395 | case BPF_JGT: | |
4cabc5b1 | 2396 | value_from_signed = false; |
7e57fbb2 | 2397 | /* fallthrough */ |
48461135 | 2398 | case BPF_JSGT: |
4cabc5b1 DB |
2399 | if (true_reg->value_from_signed != value_from_signed) |
2400 | reset_reg_range_values(true_reg, 0); | |
2401 | if (false_reg->value_from_signed != value_from_signed) | |
2402 | reset_reg_range_values(false_reg, 0); | |
2403 | if (opcode == BPF_JGT) { | |
2404 | /* Unsigned comparison, the minimum value is 0. */ | |
2405 | true_reg->min_value = 0; | |
2406 | } | |
48461135 JB |
2407 | /* |
2408 | * If this is false, then the val is <= the register, if it is | |
2409 | * true the register <= to the val. | |
2410 | */ | |
2411 | false_reg->min_value = val; | |
4cabc5b1 | 2412 | false_reg->value_from_signed = value_from_signed; |
48461135 | 2413 | true_reg->max_value = val - 1; |
4cabc5b1 | 2414 | true_reg->value_from_signed = value_from_signed; |
48461135 JB |
2415 | break; |
2416 | case BPF_JGE: | |
4cabc5b1 | 2417 | value_from_signed = false; |
7e57fbb2 | 2418 | /* fallthrough */ |
48461135 | 2419 | case BPF_JSGE: |
4cabc5b1 DB |
2420 | if (true_reg->value_from_signed != value_from_signed) |
2421 | reset_reg_range_values(true_reg, 0); | |
2422 | if (false_reg->value_from_signed != value_from_signed) | |
2423 | reset_reg_range_values(false_reg, 0); | |
2424 | if (opcode == BPF_JGE) { | |
2425 | /* Unsigned comparison, the minimum value is 0. */ | |
2426 | true_reg->min_value = 0; | |
2427 | } | |
48461135 JB |
2428 | /* If this is false then constant < register, if it is true then |
2429 | * the register < constant. | |
2430 | */ | |
2431 | false_reg->min_value = val + 1; | |
4cabc5b1 | 2432 | false_reg->value_from_signed = value_from_signed; |
48461135 | 2433 | true_reg->max_value = val; |
4cabc5b1 | 2434 | true_reg->value_from_signed = value_from_signed; |
48461135 JB |
2435 | break; |
2436 | default: | |
2437 | break; | |
2438 | } | |
2439 | ||
2440 | check_reg_overflow(false_reg); | |
2441 | check_reg_overflow(true_reg); | |
f1174f77 EC |
2442 | } |
2443 | ||
2444 | /* Regs are known to be equal, so intersect their min/max/var_off */ | |
2445 | static void __reg_combine_min_max(struct bpf_reg_state *src_reg, | |
2446 | struct bpf_reg_state *dst_reg) | |
2447 | { | |
2448 | src_reg->min_value = dst_reg->min_value = max(src_reg->min_value, | |
2449 | dst_reg->min_value); | |
2450 | src_reg->max_value = dst_reg->max_value = min(src_reg->max_value, | |
2451 | dst_reg->max_value); | |
2452 | src_reg->var_off = dst_reg->var_off = tnum_intersect(src_reg->var_off, | |
2453 | dst_reg->var_off); | |
2454 | check_reg_overflow(src_reg); | |
2455 | check_reg_overflow(dst_reg); | |
2456 | } | |
2457 | ||
2458 | static void reg_combine_min_max(struct bpf_reg_state *true_src, | |
2459 | struct bpf_reg_state *true_dst, | |
2460 | struct bpf_reg_state *false_src, | |
2461 | struct bpf_reg_state *false_dst, | |
2462 | u8 opcode) | |
2463 | { | |
2464 | switch (opcode) { | |
2465 | case BPF_JEQ: | |
2466 | __reg_combine_min_max(true_src, true_dst); | |
2467 | break; | |
2468 | case BPF_JNE: | |
2469 | __reg_combine_min_max(false_src, false_dst); | |
4cabc5b1 | 2470 | } |
48461135 JB |
2471 | } |
2472 | ||
57a09bf0 | 2473 | static void mark_map_reg(struct bpf_reg_state *regs, u32 regno, u32 id, |
f1174f77 | 2474 | bool is_null) |
57a09bf0 TG |
2475 | { |
2476 | struct bpf_reg_state *reg = ®s[regno]; | |
2477 | ||
2478 | if (reg->type == PTR_TO_MAP_VALUE_OR_NULL && reg->id == id) { | |
f1174f77 EC |
2479 | /* Old offset (both fixed and variable parts) should |
2480 | * have been known-zero, because we don't allow pointer | |
2481 | * arithmetic on pointers that might be NULL. | |
2482 | */ | |
2483 | if (WARN_ON_ONCE(reg->min_value || reg->max_value || | |
2484 | reg->var_off.value || reg->var_off.mask || | |
2485 | reg->off)) { | |
2486 | reg->min_value = reg->max_value = reg->off = 0; | |
2487 | reg->var_off = tnum_const(0); | |
2488 | } | |
2489 | if (is_null) { | |
2490 | reg->type = SCALAR_VALUE; | |
56f668df MKL |
2491 | } else if (reg->map_ptr->inner_map_meta) { |
2492 | reg->type = CONST_PTR_TO_MAP; | |
2493 | reg->map_ptr = reg->map_ptr->inner_map_meta; | |
2494 | } else { | |
f1174f77 | 2495 | reg->type = PTR_TO_MAP_VALUE; |
56f668df | 2496 | } |
a08dd0da DB |
2497 | /* We don't need id from this point onwards anymore, thus we |
2498 | * should better reset it, so that state pruning has chances | |
2499 | * to take effect. | |
2500 | */ | |
2501 | reg->id = 0; | |
57a09bf0 TG |
2502 | } |
2503 | } | |
2504 | ||
2505 | /* The logic is similar to find_good_pkt_pointers(), both could eventually | |
2506 | * be folded together at some point. | |
2507 | */ | |
2508 | static void mark_map_regs(struct bpf_verifier_state *state, u32 regno, | |
f1174f77 | 2509 | bool is_null) |
57a09bf0 TG |
2510 | { |
2511 | struct bpf_reg_state *regs = state->regs; | |
a08dd0da | 2512 | u32 id = regs[regno].id; |
57a09bf0 TG |
2513 | int i; |
2514 | ||
2515 | for (i = 0; i < MAX_BPF_REG; i++) | |
f1174f77 | 2516 | mark_map_reg(regs, i, id, is_null); |
57a09bf0 TG |
2517 | |
2518 | for (i = 0; i < MAX_BPF_STACK; i += BPF_REG_SIZE) { | |
2519 | if (state->stack_slot_type[i] != STACK_SPILL) | |
2520 | continue; | |
f1174f77 | 2521 | mark_map_reg(state->spilled_regs, i / BPF_REG_SIZE, id, is_null); |
57a09bf0 TG |
2522 | } |
2523 | } | |
2524 | ||
58e2af8b | 2525 | static int check_cond_jmp_op(struct bpf_verifier_env *env, |
17a52670 AS |
2526 | struct bpf_insn *insn, int *insn_idx) |
2527 | { | |
58e2af8b JK |
2528 | struct bpf_verifier_state *other_branch, *this_branch = &env->cur_state; |
2529 | struct bpf_reg_state *regs = this_branch->regs, *dst_reg; | |
17a52670 AS |
2530 | u8 opcode = BPF_OP(insn->code); |
2531 | int err; | |
2532 | ||
2533 | if (opcode > BPF_EXIT) { | |
2534 | verbose("invalid BPF_JMP opcode %x\n", opcode); | |
2535 | return -EINVAL; | |
2536 | } | |
2537 | ||
2538 | if (BPF_SRC(insn->code) == BPF_X) { | |
2539 | if (insn->imm != 0) { | |
2540 | verbose("BPF_JMP uses reserved fields\n"); | |
2541 | return -EINVAL; | |
2542 | } | |
2543 | ||
2544 | /* check src1 operand */ | |
2545 | err = check_reg_arg(regs, insn->src_reg, SRC_OP); | |
2546 | if (err) | |
2547 | return err; | |
1be7f75d AS |
2548 | |
2549 | if (is_pointer_value(env, insn->src_reg)) { | |
2550 | verbose("R%d pointer comparison prohibited\n", | |
2551 | insn->src_reg); | |
2552 | return -EACCES; | |
2553 | } | |
17a52670 AS |
2554 | } else { |
2555 | if (insn->src_reg != BPF_REG_0) { | |
2556 | verbose("BPF_JMP uses reserved fields\n"); | |
2557 | return -EINVAL; | |
2558 | } | |
2559 | } | |
2560 | ||
2561 | /* check src2 operand */ | |
2562 | err = check_reg_arg(regs, insn->dst_reg, SRC_OP); | |
2563 | if (err) | |
2564 | return err; | |
2565 | ||
1a0dc1ac AS |
2566 | dst_reg = ®s[insn->dst_reg]; |
2567 | ||
17a52670 AS |
2568 | /* detect if R == 0 where R was initialized to zero earlier */ |
2569 | if (BPF_SRC(insn->code) == BPF_K && | |
2570 | (opcode == BPF_JEQ || opcode == BPF_JNE) && | |
f1174f77 EC |
2571 | dst_reg->type == SCALAR_VALUE && |
2572 | tnum_equals_const(dst_reg->var_off, insn->imm)) { | |
17a52670 AS |
2573 | if (opcode == BPF_JEQ) { |
2574 | /* if (imm == imm) goto pc+off; | |
2575 | * only follow the goto, ignore fall-through | |
2576 | */ | |
2577 | *insn_idx += insn->off; | |
2578 | return 0; | |
2579 | } else { | |
2580 | /* if (imm != imm) goto pc+off; | |
2581 | * only follow fall-through branch, since | |
2582 | * that's where the program will go | |
2583 | */ | |
2584 | return 0; | |
2585 | } | |
2586 | } | |
2587 | ||
2588 | other_branch = push_stack(env, *insn_idx + insn->off + 1, *insn_idx); | |
2589 | if (!other_branch) | |
2590 | return -EFAULT; | |
2591 | ||
48461135 JB |
2592 | /* detect if we are comparing against a constant value so we can adjust |
2593 | * our min/max values for our dst register. | |
f1174f77 EC |
2594 | * this is only legit if both are scalars (or pointers to the same |
2595 | * object, I suppose, but we don't support that right now), because | |
2596 | * otherwise the different base pointers mean the offsets aren't | |
2597 | * comparable. | |
48461135 JB |
2598 | */ |
2599 | if (BPF_SRC(insn->code) == BPF_X) { | |
f1174f77 EC |
2600 | if (dst_reg->type == SCALAR_VALUE && |
2601 | regs[insn->src_reg].type == SCALAR_VALUE) { | |
2602 | if (tnum_is_const(regs[insn->src_reg].var_off)) | |
2603 | reg_set_min_max(&other_branch->regs[insn->dst_reg], | |
2604 | dst_reg, regs[insn->src_reg].var_off.value, | |
2605 | opcode); | |
2606 | else if (tnum_is_const(dst_reg->var_off)) | |
2607 | reg_set_min_max_inv(&other_branch->regs[insn->src_reg], | |
2608 | ®s[insn->src_reg], | |
2609 | dst_reg->var_off.value, opcode); | |
2610 | else if (opcode == BPF_JEQ || opcode == BPF_JNE) | |
2611 | /* Comparing for equality, we can combine knowledge */ | |
2612 | reg_combine_min_max(&other_branch->regs[insn->src_reg], | |
2613 | &other_branch->regs[insn->dst_reg], | |
2614 | ®s[insn->src_reg], | |
2615 | ®s[insn->dst_reg], opcode); | |
2616 | } | |
2617 | } else if (dst_reg->type == SCALAR_VALUE) { | |
48461135 JB |
2618 | reg_set_min_max(&other_branch->regs[insn->dst_reg], |
2619 | dst_reg, insn->imm, opcode); | |
2620 | } | |
2621 | ||
58e2af8b | 2622 | /* detect if R == 0 where R is returned from bpf_map_lookup_elem() */ |
17a52670 | 2623 | if (BPF_SRC(insn->code) == BPF_K && |
1a0dc1ac AS |
2624 | insn->imm == 0 && (opcode == BPF_JEQ || opcode == BPF_JNE) && |
2625 | dst_reg->type == PTR_TO_MAP_VALUE_OR_NULL) { | |
57a09bf0 TG |
2626 | /* Mark all identical map registers in each branch as either |
2627 | * safe or unknown depending R == 0 or R != 0 conditional. | |
2628 | */ | |
f1174f77 EC |
2629 | mark_map_regs(this_branch, insn->dst_reg, opcode == BPF_JNE); |
2630 | mark_map_regs(other_branch, insn->dst_reg, opcode == BPF_JEQ); | |
969bf05e AS |
2631 | } else if (BPF_SRC(insn->code) == BPF_X && opcode == BPF_JGT && |
2632 | dst_reg->type == PTR_TO_PACKET && | |
2633 | regs[insn->src_reg].type == PTR_TO_PACKET_END) { | |
2d2be8ca DB |
2634 | find_good_pkt_pointers(this_branch, dst_reg); |
2635 | } else if (BPF_SRC(insn->code) == BPF_X && opcode == BPF_JGE && | |
2636 | dst_reg->type == PTR_TO_PACKET_END && | |
2637 | regs[insn->src_reg].type == PTR_TO_PACKET) { | |
2638 | find_good_pkt_pointers(other_branch, ®s[insn->src_reg]); | |
1be7f75d AS |
2639 | } else if (is_pointer_value(env, insn->dst_reg)) { |
2640 | verbose("R%d pointer comparison prohibited\n", insn->dst_reg); | |
2641 | return -EACCES; | |
17a52670 AS |
2642 | } |
2643 | if (log_level) | |
2d2be8ca | 2644 | print_verifier_state(this_branch); |
17a52670 AS |
2645 | return 0; |
2646 | } | |
2647 | ||
0246e64d AS |
2648 | /* return the map pointer stored inside BPF_LD_IMM64 instruction */ |
2649 | static struct bpf_map *ld_imm64_to_map_ptr(struct bpf_insn *insn) | |
2650 | { | |
2651 | u64 imm64 = ((u64) (u32) insn[0].imm) | ((u64) (u32) insn[1].imm) << 32; | |
2652 | ||
2653 | return (struct bpf_map *) (unsigned long) imm64; | |
2654 | } | |
2655 | ||
17a52670 | 2656 | /* verify BPF_LD_IMM64 instruction */ |
58e2af8b | 2657 | static int check_ld_imm(struct bpf_verifier_env *env, struct bpf_insn *insn) |
17a52670 | 2658 | { |
58e2af8b | 2659 | struct bpf_reg_state *regs = env->cur_state.regs; |
17a52670 AS |
2660 | int err; |
2661 | ||
2662 | if (BPF_SIZE(insn->code) != BPF_DW) { | |
2663 | verbose("invalid BPF_LD_IMM insn\n"); | |
2664 | return -EINVAL; | |
2665 | } | |
2666 | if (insn->off != 0) { | |
2667 | verbose("BPF_LD_IMM64 uses reserved fields\n"); | |
2668 | return -EINVAL; | |
2669 | } | |
2670 | ||
2671 | err = check_reg_arg(regs, insn->dst_reg, DST_OP); | |
2672 | if (err) | |
2673 | return err; | |
2674 | ||
6b173873 | 2675 | if (insn->src_reg == 0) { |
6b173873 JK |
2676 | u64 imm = ((u64)(insn + 1)->imm << 32) | (u32)insn->imm; |
2677 | ||
f1174f77 EC |
2678 | regs[insn->dst_reg].type = SCALAR_VALUE; |
2679 | regs[insn->dst_reg].min_value = imm; | |
2680 | regs[insn->dst_reg].max_value = imm; | |
2681 | check_reg_overflow(®s[insn->dst_reg]); | |
2682 | regs[insn->dst_reg].var_off = tnum_const(imm); | |
4a2ff55a | 2683 | regs[insn->dst_reg].id = 0; |
17a52670 | 2684 | return 0; |
6b173873 | 2685 | } |
17a52670 AS |
2686 | |
2687 | /* replace_map_fd_with_map_ptr() should have caught bad ld_imm64 */ | |
2688 | BUG_ON(insn->src_reg != BPF_PSEUDO_MAP_FD); | |
2689 | ||
2690 | regs[insn->dst_reg].type = CONST_PTR_TO_MAP; | |
2691 | regs[insn->dst_reg].map_ptr = ld_imm64_to_map_ptr(insn); | |
2692 | return 0; | |
2693 | } | |
2694 | ||
96be4325 DB |
2695 | static bool may_access_skb(enum bpf_prog_type type) |
2696 | { | |
2697 | switch (type) { | |
2698 | case BPF_PROG_TYPE_SOCKET_FILTER: | |
2699 | case BPF_PROG_TYPE_SCHED_CLS: | |
94caee8c | 2700 | case BPF_PROG_TYPE_SCHED_ACT: |
96be4325 DB |
2701 | return true; |
2702 | default: | |
2703 | return false; | |
2704 | } | |
2705 | } | |
2706 | ||
ddd872bc AS |
2707 | /* verify safety of LD_ABS|LD_IND instructions: |
2708 | * - they can only appear in the programs where ctx == skb | |
2709 | * - since they are wrappers of function calls, they scratch R1-R5 registers, | |
2710 | * preserve R6-R9, and store return value into R0 | |
2711 | * | |
2712 | * Implicit input: | |
2713 | * ctx == skb == R6 == CTX | |
2714 | * | |
2715 | * Explicit input: | |
2716 | * SRC == any register | |
2717 | * IMM == 32-bit immediate | |
2718 | * | |
2719 | * Output: | |
2720 | * R0 - 8/16/32-bit skb data converted to cpu endianness | |
2721 | */ | |
58e2af8b | 2722 | static int check_ld_abs(struct bpf_verifier_env *env, struct bpf_insn *insn) |
ddd872bc | 2723 | { |
58e2af8b | 2724 | struct bpf_reg_state *regs = env->cur_state.regs; |
ddd872bc | 2725 | u8 mode = BPF_MODE(insn->code); |
ddd872bc AS |
2726 | int i, err; |
2727 | ||
24701ece | 2728 | if (!may_access_skb(env->prog->type)) { |
1a0dc1ac | 2729 | verbose("BPF_LD_[ABS|IND] instructions not allowed for this program type\n"); |
ddd872bc AS |
2730 | return -EINVAL; |
2731 | } | |
2732 | ||
2733 | if (insn->dst_reg != BPF_REG_0 || insn->off != 0 || | |
d82bccc6 | 2734 | BPF_SIZE(insn->code) == BPF_DW || |
ddd872bc | 2735 | (mode == BPF_ABS && insn->src_reg != BPF_REG_0)) { |
1a0dc1ac | 2736 | verbose("BPF_LD_[ABS|IND] uses reserved fields\n"); |
ddd872bc AS |
2737 | return -EINVAL; |
2738 | } | |
2739 | ||
2740 | /* check whether implicit source operand (register R6) is readable */ | |
2741 | err = check_reg_arg(regs, BPF_REG_6, SRC_OP); | |
2742 | if (err) | |
2743 | return err; | |
2744 | ||
2745 | if (regs[BPF_REG_6].type != PTR_TO_CTX) { | |
2746 | verbose("at the time of BPF_LD_ABS|IND R6 != pointer to skb\n"); | |
2747 | return -EINVAL; | |
2748 | } | |
2749 | ||
2750 | if (mode == BPF_IND) { | |
2751 | /* check explicit source operand */ | |
2752 | err = check_reg_arg(regs, insn->src_reg, SRC_OP); | |
2753 | if (err) | |
2754 | return err; | |
2755 | } | |
2756 | ||
2757 | /* reset caller saved regs to unreadable */ | |
a9789ef9 DB |
2758 | for (i = 0; i < CALLER_SAVED_REGS; i++) |
2759 | mark_reg_not_init(regs, caller_saved[i]); | |
ddd872bc AS |
2760 | |
2761 | /* mark destination R0 register as readable, since it contains | |
2762 | * the value fetched from the packet | |
2763 | */ | |
f1174f77 | 2764 | mark_reg_unknown(regs, BPF_REG_0); |
ddd872bc AS |
2765 | return 0; |
2766 | } | |
2767 | ||
475fb78f AS |
2768 | /* non-recursive DFS pseudo code |
2769 | * 1 procedure DFS-iterative(G,v): | |
2770 | * 2 label v as discovered | |
2771 | * 3 let S be a stack | |
2772 | * 4 S.push(v) | |
2773 | * 5 while S is not empty | |
2774 | * 6 t <- S.pop() | |
2775 | * 7 if t is what we're looking for: | |
2776 | * 8 return t | |
2777 | * 9 for all edges e in G.adjacentEdges(t) do | |
2778 | * 10 if edge e is already labelled | |
2779 | * 11 continue with the next edge | |
2780 | * 12 w <- G.adjacentVertex(t,e) | |
2781 | * 13 if vertex w is not discovered and not explored | |
2782 | * 14 label e as tree-edge | |
2783 | * 15 label w as discovered | |
2784 | * 16 S.push(w) | |
2785 | * 17 continue at 5 | |
2786 | * 18 else if vertex w is discovered | |
2787 | * 19 label e as back-edge | |
2788 | * 20 else | |
2789 | * 21 // vertex w is explored | |
2790 | * 22 label e as forward- or cross-edge | |
2791 | * 23 label t as explored | |
2792 | * 24 S.pop() | |
2793 | * | |
2794 | * convention: | |
2795 | * 0x10 - discovered | |
2796 | * 0x11 - discovered and fall-through edge labelled | |
2797 | * 0x12 - discovered and fall-through and branch edges labelled | |
2798 | * 0x20 - explored | |
2799 | */ | |
2800 | ||
2801 | enum { | |
2802 | DISCOVERED = 0x10, | |
2803 | EXPLORED = 0x20, | |
2804 | FALLTHROUGH = 1, | |
2805 | BRANCH = 2, | |
2806 | }; | |
2807 | ||
58e2af8b | 2808 | #define STATE_LIST_MARK ((struct bpf_verifier_state_list *) -1L) |
f1bca824 | 2809 | |
475fb78f AS |
2810 | static int *insn_stack; /* stack of insns to process */ |
2811 | static int cur_stack; /* current stack index */ | |
2812 | static int *insn_state; | |
2813 | ||
2814 | /* t, w, e - match pseudo-code above: | |
2815 | * t - index of current instruction | |
2816 | * w - next instruction | |
2817 | * e - edge | |
2818 | */ | |
58e2af8b | 2819 | static int push_insn(int t, int w, int e, struct bpf_verifier_env *env) |
475fb78f AS |
2820 | { |
2821 | if (e == FALLTHROUGH && insn_state[t] >= (DISCOVERED | FALLTHROUGH)) | |
2822 | return 0; | |
2823 | ||
2824 | if (e == BRANCH && insn_state[t] >= (DISCOVERED | BRANCH)) | |
2825 | return 0; | |
2826 | ||
2827 | if (w < 0 || w >= env->prog->len) { | |
2828 | verbose("jump out of range from insn %d to %d\n", t, w); | |
2829 | return -EINVAL; | |
2830 | } | |
2831 | ||
f1bca824 AS |
2832 | if (e == BRANCH) |
2833 | /* mark branch target for state pruning */ | |
2834 | env->explored_states[w] = STATE_LIST_MARK; | |
2835 | ||
475fb78f AS |
2836 | if (insn_state[w] == 0) { |
2837 | /* tree-edge */ | |
2838 | insn_state[t] = DISCOVERED | e; | |
2839 | insn_state[w] = DISCOVERED; | |
2840 | if (cur_stack >= env->prog->len) | |
2841 | return -E2BIG; | |
2842 | insn_stack[cur_stack++] = w; | |
2843 | return 1; | |
2844 | } else if ((insn_state[w] & 0xF0) == DISCOVERED) { | |
2845 | verbose("back-edge from insn %d to %d\n", t, w); | |
2846 | return -EINVAL; | |
2847 | } else if (insn_state[w] == EXPLORED) { | |
2848 | /* forward- or cross-edge */ | |
2849 | insn_state[t] = DISCOVERED | e; | |
2850 | } else { | |
2851 | verbose("insn state internal bug\n"); | |
2852 | return -EFAULT; | |
2853 | } | |
2854 | return 0; | |
2855 | } | |
2856 | ||
2857 | /* non-recursive depth-first-search to detect loops in BPF program | |
2858 | * loop == back-edge in directed graph | |
2859 | */ | |
58e2af8b | 2860 | static int check_cfg(struct bpf_verifier_env *env) |
475fb78f AS |
2861 | { |
2862 | struct bpf_insn *insns = env->prog->insnsi; | |
2863 | int insn_cnt = env->prog->len; | |
2864 | int ret = 0; | |
2865 | int i, t; | |
2866 | ||
2867 | insn_state = kcalloc(insn_cnt, sizeof(int), GFP_KERNEL); | |
2868 | if (!insn_state) | |
2869 | return -ENOMEM; | |
2870 | ||
2871 | insn_stack = kcalloc(insn_cnt, sizeof(int), GFP_KERNEL); | |
2872 | if (!insn_stack) { | |
2873 | kfree(insn_state); | |
2874 | return -ENOMEM; | |
2875 | } | |
2876 | ||
2877 | insn_state[0] = DISCOVERED; /* mark 1st insn as discovered */ | |
2878 | insn_stack[0] = 0; /* 0 is the first instruction */ | |
2879 | cur_stack = 1; | |
2880 | ||
2881 | peek_stack: | |
2882 | if (cur_stack == 0) | |
2883 | goto check_state; | |
2884 | t = insn_stack[cur_stack - 1]; | |
2885 | ||
2886 | if (BPF_CLASS(insns[t].code) == BPF_JMP) { | |
2887 | u8 opcode = BPF_OP(insns[t].code); | |
2888 | ||
2889 | if (opcode == BPF_EXIT) { | |
2890 | goto mark_explored; | |
2891 | } else if (opcode == BPF_CALL) { | |
2892 | ret = push_insn(t, t + 1, FALLTHROUGH, env); | |
2893 | if (ret == 1) | |
2894 | goto peek_stack; | |
2895 | else if (ret < 0) | |
2896 | goto err_free; | |
07016151 DB |
2897 | if (t + 1 < insn_cnt) |
2898 | env->explored_states[t + 1] = STATE_LIST_MARK; | |
475fb78f AS |
2899 | } else if (opcode == BPF_JA) { |
2900 | if (BPF_SRC(insns[t].code) != BPF_K) { | |
2901 | ret = -EINVAL; | |
2902 | goto err_free; | |
2903 | } | |
2904 | /* unconditional jump with single edge */ | |
2905 | ret = push_insn(t, t + insns[t].off + 1, | |
2906 | FALLTHROUGH, env); | |
2907 | if (ret == 1) | |
2908 | goto peek_stack; | |
2909 | else if (ret < 0) | |
2910 | goto err_free; | |
f1bca824 AS |
2911 | /* tell verifier to check for equivalent states |
2912 | * after every call and jump | |
2913 | */ | |
c3de6317 AS |
2914 | if (t + 1 < insn_cnt) |
2915 | env->explored_states[t + 1] = STATE_LIST_MARK; | |
475fb78f AS |
2916 | } else { |
2917 | /* conditional jump with two edges */ | |
3c2ce60b | 2918 | env->explored_states[t] = STATE_LIST_MARK; |
475fb78f AS |
2919 | ret = push_insn(t, t + 1, FALLTHROUGH, env); |
2920 | if (ret == 1) | |
2921 | goto peek_stack; | |
2922 | else if (ret < 0) | |
2923 | goto err_free; | |
2924 | ||
2925 | ret = push_insn(t, t + insns[t].off + 1, BRANCH, env); | |
2926 | if (ret == 1) | |
2927 | goto peek_stack; | |
2928 | else if (ret < 0) | |
2929 | goto err_free; | |
2930 | } | |
2931 | } else { | |
2932 | /* all other non-branch instructions with single | |
2933 | * fall-through edge | |
2934 | */ | |
2935 | ret = push_insn(t, t + 1, FALLTHROUGH, env); | |
2936 | if (ret == 1) | |
2937 | goto peek_stack; | |
2938 | else if (ret < 0) | |
2939 | goto err_free; | |
2940 | } | |
2941 | ||
2942 | mark_explored: | |
2943 | insn_state[t] = EXPLORED; | |
2944 | if (cur_stack-- <= 0) { | |
2945 | verbose("pop stack internal bug\n"); | |
2946 | ret = -EFAULT; | |
2947 | goto err_free; | |
2948 | } | |
2949 | goto peek_stack; | |
2950 | ||
2951 | check_state: | |
2952 | for (i = 0; i < insn_cnt; i++) { | |
2953 | if (insn_state[i] != EXPLORED) { | |
2954 | verbose("unreachable insn %d\n", i); | |
2955 | ret = -EINVAL; | |
2956 | goto err_free; | |
2957 | } | |
2958 | } | |
2959 | ret = 0; /* cfg looks good */ | |
2960 | ||
2961 | err_free: | |
2962 | kfree(insn_state); | |
2963 | kfree(insn_stack); | |
2964 | return ret; | |
2965 | } | |
2966 | ||
f1174f77 EC |
2967 | /* check %cur's range satisfies %old's */ |
2968 | static bool range_within(struct bpf_reg_state *old, | |
2969 | struct bpf_reg_state *cur) | |
2970 | { | |
2971 | return old->min_value <= cur->min_value && | |
2972 | old->max_value >= cur->max_value; | |
2973 | } | |
2974 | ||
2975 | /* Maximum number of register states that can exist at once */ | |
2976 | #define ID_MAP_SIZE (MAX_BPF_REG + MAX_BPF_STACK / BPF_REG_SIZE) | |
2977 | struct idpair { | |
2978 | u32 old; | |
2979 | u32 cur; | |
2980 | }; | |
2981 | ||
2982 | /* If in the old state two registers had the same id, then they need to have | |
2983 | * the same id in the new state as well. But that id could be different from | |
2984 | * the old state, so we need to track the mapping from old to new ids. | |
2985 | * Once we have seen that, say, a reg with old id 5 had new id 9, any subsequent | |
2986 | * regs with old id 5 must also have new id 9 for the new state to be safe. But | |
2987 | * regs with a different old id could still have new id 9, we don't care about | |
2988 | * that. | |
2989 | * So we look through our idmap to see if this old id has been seen before. If | |
2990 | * so, we require the new id to match; otherwise, we add the id pair to the map. | |
969bf05e | 2991 | */ |
f1174f77 | 2992 | static bool check_ids(u32 old_id, u32 cur_id, struct idpair *idmap) |
969bf05e | 2993 | { |
f1174f77 | 2994 | unsigned int i; |
969bf05e | 2995 | |
f1174f77 EC |
2996 | for (i = 0; i < ID_MAP_SIZE; i++) { |
2997 | if (!idmap[i].old) { | |
2998 | /* Reached an empty slot; haven't seen this id before */ | |
2999 | idmap[i].old = old_id; | |
3000 | idmap[i].cur = cur_id; | |
3001 | return true; | |
3002 | } | |
3003 | if (idmap[i].old == old_id) | |
3004 | return idmap[i].cur == cur_id; | |
3005 | } | |
3006 | /* We ran out of idmap slots, which should be impossible */ | |
3007 | WARN_ON_ONCE(1); | |
3008 | return false; | |
3009 | } | |
3010 | ||
3011 | /* Returns true if (rold safe implies rcur safe) */ | |
3012 | static bool regsafe(struct bpf_reg_state *rold, | |
3013 | struct bpf_reg_state *rcur, | |
3014 | bool varlen_map_access, struct idpair *idmap) | |
3015 | { | |
3016 | if (memcmp(rold, rcur, sizeof(*rold)) == 0) | |
969bf05e AS |
3017 | return true; |
3018 | ||
f1174f77 EC |
3019 | if (rold->type == NOT_INIT) |
3020 | /* explored state can't have used this */ | |
969bf05e | 3021 | return true; |
f1174f77 EC |
3022 | if (rcur->type == NOT_INIT) |
3023 | return false; | |
3024 | switch (rold->type) { | |
3025 | case SCALAR_VALUE: | |
3026 | if (rcur->type == SCALAR_VALUE) { | |
3027 | /* new val must satisfy old val knowledge */ | |
3028 | return range_within(rold, rcur) && | |
3029 | tnum_in(rold->var_off, rcur->var_off); | |
3030 | } else { | |
3031 | /* if we knew anything about the old value, we're not | |
3032 | * equal, because we can't know anything about the | |
3033 | * scalar value of the pointer in the new value. | |
3034 | */ | |
3035 | return rold->min_value == BPF_REGISTER_MIN_RANGE && | |
3036 | rold->max_value == BPF_REGISTER_MAX_RANGE && | |
3037 | tnum_is_unknown(rold->var_off); | |
3038 | } | |
3039 | case PTR_TO_MAP_VALUE: | |
3040 | if (varlen_map_access) { | |
3041 | /* If the new min/max/var_off satisfy the old ones and | |
3042 | * everything else matches, we are OK. | |
3043 | * We don't care about the 'id' value, because nothing | |
3044 | * uses it for PTR_TO_MAP_VALUE (only for ..._OR_NULL) | |
3045 | */ | |
3046 | return memcmp(rold, rcur, offsetof(struct bpf_reg_state, id)) == 0 && | |
3047 | range_within(rold, rcur) && | |
3048 | tnum_in(rold->var_off, rcur->var_off); | |
3049 | } else { | |
3050 | /* If the ranges/var_off were not the same, but | |
3051 | * everything else was and we didn't do a variable | |
3052 | * access into a map then we are a-ok. | |
3053 | */ | |
3054 | return memcmp(rold, rcur, offsetof(struct bpf_reg_state, id)) == 0; | |
3055 | } | |
3056 | case PTR_TO_MAP_VALUE_OR_NULL: | |
3057 | /* a PTR_TO_MAP_VALUE could be safe to use as a | |
3058 | * PTR_TO_MAP_VALUE_OR_NULL into the same map. | |
3059 | * However, if the old PTR_TO_MAP_VALUE_OR_NULL then got NULL- | |
3060 | * checked, doing so could have affected others with the same | |
3061 | * id, and we can't check for that because we lost the id when | |
3062 | * we converted to a PTR_TO_MAP_VALUE. | |
3063 | */ | |
3064 | if (rcur->type != PTR_TO_MAP_VALUE_OR_NULL) | |
3065 | return false; | |
3066 | if (memcmp(rold, rcur, offsetof(struct bpf_reg_state, id))) | |
3067 | return false; | |
3068 | /* Check our ids match any regs they're supposed to */ | |
3069 | return check_ids(rold->id, rcur->id, idmap); | |
3070 | case PTR_TO_PACKET: | |
3071 | if (rcur->type != PTR_TO_PACKET) | |
3072 | return false; | |
3073 | /* We must have at least as much range as the old ptr | |
3074 | * did, so that any accesses which were safe before are | |
3075 | * still safe. This is true even if old range < old off, | |
3076 | * since someone could have accessed through (ptr - k), or | |
3077 | * even done ptr -= k in a register, to get a safe access. | |
3078 | */ | |
3079 | if (rold->range > rcur->range) | |
3080 | return false; | |
3081 | /* If the offsets don't match, we can't trust our alignment; | |
3082 | * nor can we be sure that we won't fall out of range. | |
3083 | */ | |
3084 | if (rold->off != rcur->off) | |
3085 | return false; | |
3086 | /* id relations must be preserved */ | |
3087 | if (rold->id && !check_ids(rold->id, rcur->id, idmap)) | |
3088 | return false; | |
3089 | /* new val must satisfy old val knowledge */ | |
3090 | return range_within(rold, rcur) && | |
3091 | tnum_in(rold->var_off, rcur->var_off); | |
3092 | case PTR_TO_CTX: | |
3093 | case CONST_PTR_TO_MAP: | |
3094 | case PTR_TO_STACK: | |
3095 | case PTR_TO_PACKET_END: | |
3096 | /* Only valid matches are exact, which memcmp() above | |
3097 | * would have accepted | |
3098 | */ | |
3099 | default: | |
3100 | /* Don't know what's going on, just say it's not safe */ | |
3101 | return false; | |
3102 | } | |
969bf05e | 3103 | |
f1174f77 EC |
3104 | /* Shouldn't get here; if we do, say it's not safe */ |
3105 | WARN_ON_ONCE(1); | |
969bf05e AS |
3106 | return false; |
3107 | } | |
3108 | ||
f1bca824 AS |
3109 | /* compare two verifier states |
3110 | * | |
3111 | * all states stored in state_list are known to be valid, since | |
3112 | * verifier reached 'bpf_exit' instruction through them | |
3113 | * | |
3114 | * this function is called when verifier exploring different branches of | |
3115 | * execution popped from the state stack. If it sees an old state that has | |
3116 | * more strict register state and more strict stack state then this execution | |
3117 | * branch doesn't need to be explored further, since verifier already | |
3118 | * concluded that more strict state leads to valid finish. | |
3119 | * | |
3120 | * Therefore two states are equivalent if register state is more conservative | |
3121 | * and explored stack state is more conservative than the current one. | |
3122 | * Example: | |
3123 | * explored current | |
3124 | * (slot1=INV slot2=MISC) == (slot1=MISC slot2=MISC) | |
3125 | * (slot1=MISC slot2=MISC) != (slot1=INV slot2=MISC) | |
3126 | * | |
3127 | * In other words if current stack state (one being explored) has more | |
3128 | * valid slots than old one that already passed validation, it means | |
3129 | * the verifier can stop exploring and conclude that current state is valid too | |
3130 | * | |
3131 | * Similarly with registers. If explored state has register type as invalid | |
3132 | * whereas register type in current state is meaningful, it means that | |
3133 | * the current state will reach 'bpf_exit' instruction safely | |
3134 | */ | |
48461135 JB |
3135 | static bool states_equal(struct bpf_verifier_env *env, |
3136 | struct bpf_verifier_state *old, | |
58e2af8b | 3137 | struct bpf_verifier_state *cur) |
f1bca824 | 3138 | { |
e2d2afe1 | 3139 | bool varlen_map_access = env->varlen_map_value_access; |
f1174f77 EC |
3140 | struct idpair *idmap; |
3141 | bool ret = false; | |
f1bca824 AS |
3142 | int i; |
3143 | ||
f1174f77 EC |
3144 | idmap = kcalloc(ID_MAP_SIZE, sizeof(struct idpair), GFP_KERNEL); |
3145 | /* If we failed to allocate the idmap, just say it's not safe */ | |
3146 | if (!idmap) | |
1a0dc1ac | 3147 | return false; |
f1174f77 EC |
3148 | |
3149 | for (i = 0; i < MAX_BPF_REG; i++) { | |
3150 | if (!regsafe(&old->regs[i], &cur->regs[i], varlen_map_access, | |
3151 | idmap)) | |
3152 | goto out_free; | |
f1bca824 AS |
3153 | } |
3154 | ||
3155 | for (i = 0; i < MAX_BPF_STACK; i++) { | |
9c399760 AS |
3156 | if (old->stack_slot_type[i] == STACK_INVALID) |
3157 | continue; | |
3158 | if (old->stack_slot_type[i] != cur->stack_slot_type[i]) | |
3159 | /* Ex: old explored (safe) state has STACK_SPILL in | |
3160 | * this stack slot, but current has has STACK_MISC -> | |
3161 | * this verifier states are not equivalent, | |
3162 | * return false to continue verification of this path | |
3163 | */ | |
f1174f77 | 3164 | goto out_free; |
9c399760 AS |
3165 | if (i % BPF_REG_SIZE) |
3166 | continue; | |
d25da6ca DB |
3167 | if (old->stack_slot_type[i] != STACK_SPILL) |
3168 | continue; | |
f1174f77 EC |
3169 | if (!regsafe(&old->spilled_regs[i / BPF_REG_SIZE], |
3170 | &cur->spilled_regs[i / BPF_REG_SIZE], | |
3171 | varlen_map_access, idmap)) | |
3172 | /* when explored and current stack slot are both storing | |
3173 | * spilled registers, check that stored pointers types | |
9c399760 AS |
3174 | * are the same as well. |
3175 | * Ex: explored safe path could have stored | |
f1174f77 | 3176 | * (bpf_reg_state) {.type = PTR_TO_STACK, .off = -8} |
9c399760 | 3177 | * but current path has stored: |
f1174f77 | 3178 | * (bpf_reg_state) {.type = PTR_TO_STACK, .off = -16} |
9c399760 AS |
3179 | * such verifier states are not equivalent. |
3180 | * return false to continue verification of this path | |
3181 | */ | |
f1174f77 | 3182 | goto out_free; |
9c399760 AS |
3183 | else |
3184 | continue; | |
f1bca824 | 3185 | } |
f1174f77 EC |
3186 | ret = true; |
3187 | out_free: | |
3188 | kfree(idmap); | |
3189 | return ret; | |
f1bca824 AS |
3190 | } |
3191 | ||
58e2af8b | 3192 | static int is_state_visited(struct bpf_verifier_env *env, int insn_idx) |
f1bca824 | 3193 | { |
58e2af8b JK |
3194 | struct bpf_verifier_state_list *new_sl; |
3195 | struct bpf_verifier_state_list *sl; | |
f1bca824 AS |
3196 | |
3197 | sl = env->explored_states[insn_idx]; | |
3198 | if (!sl) | |
3199 | /* this 'insn_idx' instruction wasn't marked, so we will not | |
3200 | * be doing state search here | |
3201 | */ | |
3202 | return 0; | |
3203 | ||
3204 | while (sl != STATE_LIST_MARK) { | |
48461135 | 3205 | if (states_equal(env, &sl->state, &env->cur_state)) |
f1bca824 AS |
3206 | /* reached equivalent register/stack state, |
3207 | * prune the search | |
3208 | */ | |
3209 | return 1; | |
3210 | sl = sl->next; | |
3211 | } | |
3212 | ||
3213 | /* there were no equivalent states, remember current one. | |
3214 | * technically the current state is not proven to be safe yet, | |
3215 | * but it will either reach bpf_exit (which means it's safe) or | |
3216 | * it will be rejected. Since there are no loops, we won't be | |
3217 | * seeing this 'insn_idx' instruction again on the way to bpf_exit | |
3218 | */ | |
58e2af8b | 3219 | new_sl = kmalloc(sizeof(struct bpf_verifier_state_list), GFP_USER); |
f1bca824 AS |
3220 | if (!new_sl) |
3221 | return -ENOMEM; | |
3222 | ||
3223 | /* add new state to the head of linked list */ | |
3224 | memcpy(&new_sl->state, &env->cur_state, sizeof(env->cur_state)); | |
3225 | new_sl->next = env->explored_states[insn_idx]; | |
3226 | env->explored_states[insn_idx] = new_sl; | |
3227 | return 0; | |
3228 | } | |
3229 | ||
13a27dfc JK |
3230 | static int ext_analyzer_insn_hook(struct bpf_verifier_env *env, |
3231 | int insn_idx, int prev_insn_idx) | |
3232 | { | |
3233 | if (!env->analyzer_ops || !env->analyzer_ops->insn_hook) | |
3234 | return 0; | |
3235 | ||
3236 | return env->analyzer_ops->insn_hook(env, insn_idx, prev_insn_idx); | |
3237 | } | |
3238 | ||
58e2af8b | 3239 | static int do_check(struct bpf_verifier_env *env) |
17a52670 | 3240 | { |
58e2af8b | 3241 | struct bpf_verifier_state *state = &env->cur_state; |
17a52670 | 3242 | struct bpf_insn *insns = env->prog->insnsi; |
58e2af8b | 3243 | struct bpf_reg_state *regs = state->regs; |
17a52670 AS |
3244 | int insn_cnt = env->prog->len; |
3245 | int insn_idx, prev_insn_idx = 0; | |
3246 | int insn_processed = 0; | |
3247 | bool do_print_state = false; | |
3248 | ||
3249 | init_reg_state(regs); | |
3250 | insn_idx = 0; | |
48461135 | 3251 | env->varlen_map_value_access = false; |
17a52670 AS |
3252 | for (;;) { |
3253 | struct bpf_insn *insn; | |
3254 | u8 class; | |
3255 | int err; | |
3256 | ||
3257 | if (insn_idx >= insn_cnt) { | |
3258 | verbose("invalid insn idx %d insn_cnt %d\n", | |
3259 | insn_idx, insn_cnt); | |
3260 | return -EFAULT; | |
3261 | } | |
3262 | ||
3263 | insn = &insns[insn_idx]; | |
3264 | class = BPF_CLASS(insn->code); | |
3265 | ||
07016151 | 3266 | if (++insn_processed > BPF_COMPLEXITY_LIMIT_INSNS) { |
bc1750f3 | 3267 | verbose("BPF program is too large. Processed %d insn\n", |
17a52670 AS |
3268 | insn_processed); |
3269 | return -E2BIG; | |
3270 | } | |
3271 | ||
f1bca824 AS |
3272 | err = is_state_visited(env, insn_idx); |
3273 | if (err < 0) | |
3274 | return err; | |
3275 | if (err == 1) { | |
3276 | /* found equivalent state, can prune the search */ | |
3277 | if (log_level) { | |
3278 | if (do_print_state) | |
3279 | verbose("\nfrom %d to %d: safe\n", | |
3280 | prev_insn_idx, insn_idx); | |
3281 | else | |
3282 | verbose("%d: safe\n", insn_idx); | |
3283 | } | |
3284 | goto process_bpf_exit; | |
3285 | } | |
3286 | ||
3c2ce60b DB |
3287 | if (need_resched()) |
3288 | cond_resched(); | |
3289 | ||
c5fc9692 DM |
3290 | if (log_level > 1 || (log_level && do_print_state)) { |
3291 | if (log_level > 1) | |
3292 | verbose("%d:", insn_idx); | |
3293 | else | |
3294 | verbose("\nfrom %d to %d:", | |
3295 | prev_insn_idx, insn_idx); | |
1a0dc1ac | 3296 | print_verifier_state(&env->cur_state); |
17a52670 AS |
3297 | do_print_state = false; |
3298 | } | |
3299 | ||
3300 | if (log_level) { | |
3301 | verbose("%d: ", insn_idx); | |
0d0e5769 | 3302 | print_bpf_insn(env, insn); |
17a52670 AS |
3303 | } |
3304 | ||
13a27dfc JK |
3305 | err = ext_analyzer_insn_hook(env, insn_idx, prev_insn_idx); |
3306 | if (err) | |
3307 | return err; | |
3308 | ||
17a52670 | 3309 | if (class == BPF_ALU || class == BPF_ALU64) { |
1be7f75d | 3310 | err = check_alu_op(env, insn); |
17a52670 AS |
3311 | if (err) |
3312 | return err; | |
3313 | ||
3314 | } else if (class == BPF_LDX) { | |
3df126f3 | 3315 | enum bpf_reg_type *prev_src_type, src_reg_type; |
9bac3d6d AS |
3316 | |
3317 | /* check for reserved fields is already done */ | |
3318 | ||
17a52670 AS |
3319 | /* check src operand */ |
3320 | err = check_reg_arg(regs, insn->src_reg, SRC_OP); | |
3321 | if (err) | |
3322 | return err; | |
3323 | ||
3324 | err = check_reg_arg(regs, insn->dst_reg, DST_OP_NO_MARK); | |
3325 | if (err) | |
3326 | return err; | |
3327 | ||
725f9dcd AS |
3328 | src_reg_type = regs[insn->src_reg].type; |
3329 | ||
17a52670 AS |
3330 | /* check that memory (src_reg + off) is readable, |
3331 | * the state of dst_reg will be updated by this func | |
3332 | */ | |
31fd8581 | 3333 | err = check_mem_access(env, insn_idx, insn->src_reg, insn->off, |
17a52670 AS |
3334 | BPF_SIZE(insn->code), BPF_READ, |
3335 | insn->dst_reg); | |
3336 | if (err) | |
3337 | return err; | |
3338 | ||
3df126f3 JK |
3339 | prev_src_type = &env->insn_aux_data[insn_idx].ptr_type; |
3340 | ||
3341 | if (*prev_src_type == NOT_INIT) { | |
9bac3d6d AS |
3342 | /* saw a valid insn |
3343 | * dst_reg = *(u32 *)(src_reg + off) | |
3df126f3 | 3344 | * save type to validate intersecting paths |
9bac3d6d | 3345 | */ |
3df126f3 | 3346 | *prev_src_type = src_reg_type; |
9bac3d6d | 3347 | |
3df126f3 | 3348 | } else if (src_reg_type != *prev_src_type && |
9bac3d6d | 3349 | (src_reg_type == PTR_TO_CTX || |
3df126f3 | 3350 | *prev_src_type == PTR_TO_CTX)) { |
9bac3d6d AS |
3351 | /* ABuser program is trying to use the same insn |
3352 | * dst_reg = *(u32*) (src_reg + off) | |
3353 | * with different pointer types: | |
3354 | * src_reg == ctx in one branch and | |
3355 | * src_reg == stack|map in some other branch. | |
3356 | * Reject it. | |
3357 | */ | |
3358 | verbose("same insn cannot be used with different pointers\n"); | |
3359 | return -EINVAL; | |
3360 | } | |
3361 | ||
17a52670 | 3362 | } else if (class == BPF_STX) { |
3df126f3 | 3363 | enum bpf_reg_type *prev_dst_type, dst_reg_type; |
d691f9e8 | 3364 | |
17a52670 | 3365 | if (BPF_MODE(insn->code) == BPF_XADD) { |
31fd8581 | 3366 | err = check_xadd(env, insn_idx, insn); |
17a52670 AS |
3367 | if (err) |
3368 | return err; | |
3369 | insn_idx++; | |
3370 | continue; | |
3371 | } | |
3372 | ||
17a52670 AS |
3373 | /* check src1 operand */ |
3374 | err = check_reg_arg(regs, insn->src_reg, SRC_OP); | |
3375 | if (err) | |
3376 | return err; | |
3377 | /* check src2 operand */ | |
3378 | err = check_reg_arg(regs, insn->dst_reg, SRC_OP); | |
3379 | if (err) | |
3380 | return err; | |
3381 | ||
d691f9e8 AS |
3382 | dst_reg_type = regs[insn->dst_reg].type; |
3383 | ||
17a52670 | 3384 | /* check that memory (dst_reg + off) is writeable */ |
31fd8581 | 3385 | err = check_mem_access(env, insn_idx, insn->dst_reg, insn->off, |
17a52670 AS |
3386 | BPF_SIZE(insn->code), BPF_WRITE, |
3387 | insn->src_reg); | |
3388 | if (err) | |
3389 | return err; | |
3390 | ||
3df126f3 JK |
3391 | prev_dst_type = &env->insn_aux_data[insn_idx].ptr_type; |
3392 | ||
3393 | if (*prev_dst_type == NOT_INIT) { | |
3394 | *prev_dst_type = dst_reg_type; | |
3395 | } else if (dst_reg_type != *prev_dst_type && | |
d691f9e8 | 3396 | (dst_reg_type == PTR_TO_CTX || |
3df126f3 | 3397 | *prev_dst_type == PTR_TO_CTX)) { |
d691f9e8 AS |
3398 | verbose("same insn cannot be used with different pointers\n"); |
3399 | return -EINVAL; | |
3400 | } | |
3401 | ||
17a52670 AS |
3402 | } else if (class == BPF_ST) { |
3403 | if (BPF_MODE(insn->code) != BPF_MEM || | |
3404 | insn->src_reg != BPF_REG_0) { | |
3405 | verbose("BPF_ST uses reserved fields\n"); | |
3406 | return -EINVAL; | |
3407 | } | |
3408 | /* check src operand */ | |
3409 | err = check_reg_arg(regs, insn->dst_reg, SRC_OP); | |
3410 | if (err) | |
3411 | return err; | |
3412 | ||
3413 | /* check that memory (dst_reg + off) is writeable */ | |
31fd8581 | 3414 | err = check_mem_access(env, insn_idx, insn->dst_reg, insn->off, |
17a52670 AS |
3415 | BPF_SIZE(insn->code), BPF_WRITE, |
3416 | -1); | |
3417 | if (err) | |
3418 | return err; | |
3419 | ||
3420 | } else if (class == BPF_JMP) { | |
3421 | u8 opcode = BPF_OP(insn->code); | |
3422 | ||
3423 | if (opcode == BPF_CALL) { | |
3424 | if (BPF_SRC(insn->code) != BPF_K || | |
3425 | insn->off != 0 || | |
3426 | insn->src_reg != BPF_REG_0 || | |
3427 | insn->dst_reg != BPF_REG_0) { | |
3428 | verbose("BPF_CALL uses reserved fields\n"); | |
3429 | return -EINVAL; | |
3430 | } | |
3431 | ||
81ed18ab | 3432 | err = check_call(env, insn->imm, insn_idx); |
17a52670 AS |
3433 | if (err) |
3434 | return err; | |
3435 | ||
3436 | } else if (opcode == BPF_JA) { | |
3437 | if (BPF_SRC(insn->code) != BPF_K || | |
3438 | insn->imm != 0 || | |
3439 | insn->src_reg != BPF_REG_0 || | |
3440 | insn->dst_reg != BPF_REG_0) { | |
3441 | verbose("BPF_JA uses reserved fields\n"); | |
3442 | return -EINVAL; | |
3443 | } | |
3444 | ||
3445 | insn_idx += insn->off + 1; | |
3446 | continue; | |
3447 | ||
3448 | } else if (opcode == BPF_EXIT) { | |
3449 | if (BPF_SRC(insn->code) != BPF_K || | |
3450 | insn->imm != 0 || | |
3451 | insn->src_reg != BPF_REG_0 || | |
3452 | insn->dst_reg != BPF_REG_0) { | |
3453 | verbose("BPF_EXIT uses reserved fields\n"); | |
3454 | return -EINVAL; | |
3455 | } | |
3456 | ||
3457 | /* eBPF calling convetion is such that R0 is used | |
3458 | * to return the value from eBPF program. | |
3459 | * Make sure that it's readable at this time | |
3460 | * of bpf_exit, which means that program wrote | |
3461 | * something into it earlier | |
3462 | */ | |
3463 | err = check_reg_arg(regs, BPF_REG_0, SRC_OP); | |
3464 | if (err) | |
3465 | return err; | |
3466 | ||
1be7f75d AS |
3467 | if (is_pointer_value(env, BPF_REG_0)) { |
3468 | verbose("R0 leaks addr as return value\n"); | |
3469 | return -EACCES; | |
3470 | } | |
3471 | ||
f1bca824 | 3472 | process_bpf_exit: |
17a52670 AS |
3473 | insn_idx = pop_stack(env, &prev_insn_idx); |
3474 | if (insn_idx < 0) { | |
3475 | break; | |
3476 | } else { | |
3477 | do_print_state = true; | |
3478 | continue; | |
3479 | } | |
3480 | } else { | |
3481 | err = check_cond_jmp_op(env, insn, &insn_idx); | |
3482 | if (err) | |
3483 | return err; | |
3484 | } | |
3485 | } else if (class == BPF_LD) { | |
3486 | u8 mode = BPF_MODE(insn->code); | |
3487 | ||
3488 | if (mode == BPF_ABS || mode == BPF_IND) { | |
ddd872bc AS |
3489 | err = check_ld_abs(env, insn); |
3490 | if (err) | |
3491 | return err; | |
3492 | ||
17a52670 AS |
3493 | } else if (mode == BPF_IMM) { |
3494 | err = check_ld_imm(env, insn); | |
3495 | if (err) | |
3496 | return err; | |
3497 | ||
3498 | insn_idx++; | |
3499 | } else { | |
3500 | verbose("invalid BPF_LD mode\n"); | |
3501 | return -EINVAL; | |
3502 | } | |
3503 | } else { | |
3504 | verbose("unknown insn class %d\n", class); | |
3505 | return -EINVAL; | |
3506 | } | |
3507 | ||
3508 | insn_idx++; | |
3509 | } | |
3510 | ||
8726679a AS |
3511 | verbose("processed %d insns, stack depth %d\n", |
3512 | insn_processed, env->prog->aux->stack_depth); | |
17a52670 AS |
3513 | return 0; |
3514 | } | |
3515 | ||
56f668df MKL |
3516 | static int check_map_prealloc(struct bpf_map *map) |
3517 | { | |
3518 | return (map->map_type != BPF_MAP_TYPE_HASH && | |
bcc6b1b7 MKL |
3519 | map->map_type != BPF_MAP_TYPE_PERCPU_HASH && |
3520 | map->map_type != BPF_MAP_TYPE_HASH_OF_MAPS) || | |
56f668df MKL |
3521 | !(map->map_flags & BPF_F_NO_PREALLOC); |
3522 | } | |
3523 | ||
fdc15d38 AS |
3524 | static int check_map_prog_compatibility(struct bpf_map *map, |
3525 | struct bpf_prog *prog) | |
3526 | ||
3527 | { | |
56f668df MKL |
3528 | /* Make sure that BPF_PROG_TYPE_PERF_EVENT programs only use |
3529 | * preallocated hash maps, since doing memory allocation | |
3530 | * in overflow_handler can crash depending on where nmi got | |
3531 | * triggered. | |
3532 | */ | |
3533 | if (prog->type == BPF_PROG_TYPE_PERF_EVENT) { | |
3534 | if (!check_map_prealloc(map)) { | |
3535 | verbose("perf_event programs can only use preallocated hash map\n"); | |
3536 | return -EINVAL; | |
3537 | } | |
3538 | if (map->inner_map_meta && | |
3539 | !check_map_prealloc(map->inner_map_meta)) { | |
3540 | verbose("perf_event programs can only use preallocated inner hash map\n"); | |
3541 | return -EINVAL; | |
3542 | } | |
fdc15d38 AS |
3543 | } |
3544 | return 0; | |
3545 | } | |
3546 | ||
0246e64d AS |
3547 | /* look for pseudo eBPF instructions that access map FDs and |
3548 | * replace them with actual map pointers | |
3549 | */ | |
58e2af8b | 3550 | static int replace_map_fd_with_map_ptr(struct bpf_verifier_env *env) |
0246e64d AS |
3551 | { |
3552 | struct bpf_insn *insn = env->prog->insnsi; | |
3553 | int insn_cnt = env->prog->len; | |
fdc15d38 | 3554 | int i, j, err; |
0246e64d | 3555 | |
f1f7714e | 3556 | err = bpf_prog_calc_tag(env->prog); |
aafe6ae9 DB |
3557 | if (err) |
3558 | return err; | |
3559 | ||
0246e64d | 3560 | for (i = 0; i < insn_cnt; i++, insn++) { |
9bac3d6d | 3561 | if (BPF_CLASS(insn->code) == BPF_LDX && |
d691f9e8 | 3562 | (BPF_MODE(insn->code) != BPF_MEM || insn->imm != 0)) { |
9bac3d6d AS |
3563 | verbose("BPF_LDX uses reserved fields\n"); |
3564 | return -EINVAL; | |
3565 | } | |
3566 | ||
d691f9e8 AS |
3567 | if (BPF_CLASS(insn->code) == BPF_STX && |
3568 | ((BPF_MODE(insn->code) != BPF_MEM && | |
3569 | BPF_MODE(insn->code) != BPF_XADD) || insn->imm != 0)) { | |
3570 | verbose("BPF_STX uses reserved fields\n"); | |
3571 | return -EINVAL; | |
3572 | } | |
3573 | ||
0246e64d AS |
3574 | if (insn[0].code == (BPF_LD | BPF_IMM | BPF_DW)) { |
3575 | struct bpf_map *map; | |
3576 | struct fd f; | |
3577 | ||
3578 | if (i == insn_cnt - 1 || insn[1].code != 0 || | |
3579 | insn[1].dst_reg != 0 || insn[1].src_reg != 0 || | |
3580 | insn[1].off != 0) { | |
3581 | verbose("invalid bpf_ld_imm64 insn\n"); | |
3582 | return -EINVAL; | |
3583 | } | |
3584 | ||
3585 | if (insn->src_reg == 0) | |
3586 | /* valid generic load 64-bit imm */ | |
3587 | goto next_insn; | |
3588 | ||
3589 | if (insn->src_reg != BPF_PSEUDO_MAP_FD) { | |
3590 | verbose("unrecognized bpf_ld_imm64 insn\n"); | |
3591 | return -EINVAL; | |
3592 | } | |
3593 | ||
3594 | f = fdget(insn->imm); | |
c2101297 | 3595 | map = __bpf_map_get(f); |
0246e64d AS |
3596 | if (IS_ERR(map)) { |
3597 | verbose("fd %d is not pointing to valid bpf_map\n", | |
3598 | insn->imm); | |
0246e64d AS |
3599 | return PTR_ERR(map); |
3600 | } | |
3601 | ||
fdc15d38 AS |
3602 | err = check_map_prog_compatibility(map, env->prog); |
3603 | if (err) { | |
3604 | fdput(f); | |
3605 | return err; | |
3606 | } | |
3607 | ||
0246e64d AS |
3608 | /* store map pointer inside BPF_LD_IMM64 instruction */ |
3609 | insn[0].imm = (u32) (unsigned long) map; | |
3610 | insn[1].imm = ((u64) (unsigned long) map) >> 32; | |
3611 | ||
3612 | /* check whether we recorded this map already */ | |
3613 | for (j = 0; j < env->used_map_cnt; j++) | |
3614 | if (env->used_maps[j] == map) { | |
3615 | fdput(f); | |
3616 | goto next_insn; | |
3617 | } | |
3618 | ||
3619 | if (env->used_map_cnt >= MAX_USED_MAPS) { | |
3620 | fdput(f); | |
3621 | return -E2BIG; | |
3622 | } | |
3623 | ||
0246e64d AS |
3624 | /* hold the map. If the program is rejected by verifier, |
3625 | * the map will be released by release_maps() or it | |
3626 | * will be used by the valid program until it's unloaded | |
3627 | * and all maps are released in free_bpf_prog_info() | |
3628 | */ | |
92117d84 AS |
3629 | map = bpf_map_inc(map, false); |
3630 | if (IS_ERR(map)) { | |
3631 | fdput(f); | |
3632 | return PTR_ERR(map); | |
3633 | } | |
3634 | env->used_maps[env->used_map_cnt++] = map; | |
3635 | ||
0246e64d AS |
3636 | fdput(f); |
3637 | next_insn: | |
3638 | insn++; | |
3639 | i++; | |
3640 | } | |
3641 | } | |
3642 | ||
3643 | /* now all pseudo BPF_LD_IMM64 instructions load valid | |
3644 | * 'struct bpf_map *' into a register instead of user map_fd. | |
3645 | * These pointers will be used later by verifier to validate map access. | |
3646 | */ | |
3647 | return 0; | |
3648 | } | |
3649 | ||
3650 | /* drop refcnt of maps used by the rejected program */ | |
58e2af8b | 3651 | static void release_maps(struct bpf_verifier_env *env) |
0246e64d AS |
3652 | { |
3653 | int i; | |
3654 | ||
3655 | for (i = 0; i < env->used_map_cnt; i++) | |
3656 | bpf_map_put(env->used_maps[i]); | |
3657 | } | |
3658 | ||
3659 | /* convert pseudo BPF_LD_IMM64 into generic BPF_LD_IMM64 */ | |
58e2af8b | 3660 | static void convert_pseudo_ld_imm64(struct bpf_verifier_env *env) |
0246e64d AS |
3661 | { |
3662 | struct bpf_insn *insn = env->prog->insnsi; | |
3663 | int insn_cnt = env->prog->len; | |
3664 | int i; | |
3665 | ||
3666 | for (i = 0; i < insn_cnt; i++, insn++) | |
3667 | if (insn->code == (BPF_LD | BPF_IMM | BPF_DW)) | |
3668 | insn->src_reg = 0; | |
3669 | } | |
3670 | ||
8041902d AS |
3671 | /* single env->prog->insni[off] instruction was replaced with the range |
3672 | * insni[off, off + cnt). Adjust corresponding insn_aux_data by copying | |
3673 | * [0, off) and [off, end) to new locations, so the patched range stays zero | |
3674 | */ | |
3675 | static int adjust_insn_aux_data(struct bpf_verifier_env *env, u32 prog_len, | |
3676 | u32 off, u32 cnt) | |
3677 | { | |
3678 | struct bpf_insn_aux_data *new_data, *old_data = env->insn_aux_data; | |
3679 | ||
3680 | if (cnt == 1) | |
3681 | return 0; | |
3682 | new_data = vzalloc(sizeof(struct bpf_insn_aux_data) * prog_len); | |
3683 | if (!new_data) | |
3684 | return -ENOMEM; | |
3685 | memcpy(new_data, old_data, sizeof(struct bpf_insn_aux_data) * off); | |
3686 | memcpy(new_data + off + cnt - 1, old_data + off, | |
3687 | sizeof(struct bpf_insn_aux_data) * (prog_len - off - cnt + 1)); | |
3688 | env->insn_aux_data = new_data; | |
3689 | vfree(old_data); | |
3690 | return 0; | |
3691 | } | |
3692 | ||
3693 | static struct bpf_prog *bpf_patch_insn_data(struct bpf_verifier_env *env, u32 off, | |
3694 | const struct bpf_insn *patch, u32 len) | |
3695 | { | |
3696 | struct bpf_prog *new_prog; | |
3697 | ||
3698 | new_prog = bpf_patch_insn_single(env->prog, off, patch, len); | |
3699 | if (!new_prog) | |
3700 | return NULL; | |
3701 | if (adjust_insn_aux_data(env, new_prog->len, off, len)) | |
3702 | return NULL; | |
3703 | return new_prog; | |
3704 | } | |
3705 | ||
9bac3d6d AS |
3706 | /* convert load instructions that access fields of 'struct __sk_buff' |
3707 | * into sequence of instructions that access fields of 'struct sk_buff' | |
3708 | */ | |
58e2af8b | 3709 | static int convert_ctx_accesses(struct bpf_verifier_env *env) |
9bac3d6d | 3710 | { |
36bbef52 | 3711 | const struct bpf_verifier_ops *ops = env->prog->aux->ops; |
f96da094 | 3712 | int i, cnt, size, ctx_field_size, delta = 0; |
3df126f3 | 3713 | const int insn_cnt = env->prog->len; |
36bbef52 | 3714 | struct bpf_insn insn_buf[16], *insn; |
9bac3d6d | 3715 | struct bpf_prog *new_prog; |
d691f9e8 | 3716 | enum bpf_access_type type; |
f96da094 DB |
3717 | bool is_narrower_load; |
3718 | u32 target_size; | |
9bac3d6d | 3719 | |
36bbef52 DB |
3720 | if (ops->gen_prologue) { |
3721 | cnt = ops->gen_prologue(insn_buf, env->seen_direct_write, | |
3722 | env->prog); | |
3723 | if (cnt >= ARRAY_SIZE(insn_buf)) { | |
3724 | verbose("bpf verifier is misconfigured\n"); | |
3725 | return -EINVAL; | |
3726 | } else if (cnt) { | |
8041902d | 3727 | new_prog = bpf_patch_insn_data(env, 0, insn_buf, cnt); |
36bbef52 DB |
3728 | if (!new_prog) |
3729 | return -ENOMEM; | |
8041902d | 3730 | |
36bbef52 | 3731 | env->prog = new_prog; |
3df126f3 | 3732 | delta += cnt - 1; |
36bbef52 DB |
3733 | } |
3734 | } | |
3735 | ||
3736 | if (!ops->convert_ctx_access) | |
9bac3d6d AS |
3737 | return 0; |
3738 | ||
3df126f3 | 3739 | insn = env->prog->insnsi + delta; |
36bbef52 | 3740 | |
9bac3d6d | 3741 | for (i = 0; i < insn_cnt; i++, insn++) { |
62c7989b DB |
3742 | if (insn->code == (BPF_LDX | BPF_MEM | BPF_B) || |
3743 | insn->code == (BPF_LDX | BPF_MEM | BPF_H) || | |
3744 | insn->code == (BPF_LDX | BPF_MEM | BPF_W) || | |
ea2e7ce5 | 3745 | insn->code == (BPF_LDX | BPF_MEM | BPF_DW)) |
d691f9e8 | 3746 | type = BPF_READ; |
62c7989b DB |
3747 | else if (insn->code == (BPF_STX | BPF_MEM | BPF_B) || |
3748 | insn->code == (BPF_STX | BPF_MEM | BPF_H) || | |
3749 | insn->code == (BPF_STX | BPF_MEM | BPF_W) || | |
ea2e7ce5 | 3750 | insn->code == (BPF_STX | BPF_MEM | BPF_DW)) |
d691f9e8 AS |
3751 | type = BPF_WRITE; |
3752 | else | |
9bac3d6d AS |
3753 | continue; |
3754 | ||
8041902d | 3755 | if (env->insn_aux_data[i + delta].ptr_type != PTR_TO_CTX) |
9bac3d6d | 3756 | continue; |
9bac3d6d | 3757 | |
31fd8581 | 3758 | ctx_field_size = env->insn_aux_data[i + delta].ctx_field_size; |
f96da094 | 3759 | size = BPF_LDST_BYTES(insn); |
31fd8581 YS |
3760 | |
3761 | /* If the read access is a narrower load of the field, | |
3762 | * convert to a 4/8-byte load, to minimum program type specific | |
3763 | * convert_ctx_access changes. If conversion is successful, | |
3764 | * we will apply proper mask to the result. | |
3765 | */ | |
f96da094 | 3766 | is_narrower_load = size < ctx_field_size; |
31fd8581 | 3767 | if (is_narrower_load) { |
f96da094 DB |
3768 | u32 off = insn->off; |
3769 | u8 size_code; | |
3770 | ||
3771 | if (type == BPF_WRITE) { | |
3772 | verbose("bpf verifier narrow ctx access misconfigured\n"); | |
3773 | return -EINVAL; | |
3774 | } | |
31fd8581 | 3775 | |
f96da094 | 3776 | size_code = BPF_H; |
31fd8581 YS |
3777 | if (ctx_field_size == 4) |
3778 | size_code = BPF_W; | |
3779 | else if (ctx_field_size == 8) | |
3780 | size_code = BPF_DW; | |
f96da094 | 3781 | |
31fd8581 YS |
3782 | insn->off = off & ~(ctx_field_size - 1); |
3783 | insn->code = BPF_LDX | BPF_MEM | size_code; | |
3784 | } | |
f96da094 DB |
3785 | |
3786 | target_size = 0; | |
3787 | cnt = ops->convert_ctx_access(type, insn, insn_buf, env->prog, | |
3788 | &target_size); | |
3789 | if (cnt == 0 || cnt >= ARRAY_SIZE(insn_buf) || | |
3790 | (ctx_field_size && !target_size)) { | |
9bac3d6d AS |
3791 | verbose("bpf verifier is misconfigured\n"); |
3792 | return -EINVAL; | |
3793 | } | |
f96da094 DB |
3794 | |
3795 | if (is_narrower_load && size < target_size) { | |
31fd8581 YS |
3796 | if (ctx_field_size <= 4) |
3797 | insn_buf[cnt++] = BPF_ALU32_IMM(BPF_AND, insn->dst_reg, | |
f96da094 | 3798 | (1 << size * 8) - 1); |
31fd8581 YS |
3799 | else |
3800 | insn_buf[cnt++] = BPF_ALU64_IMM(BPF_AND, insn->dst_reg, | |
f96da094 | 3801 | (1 << size * 8) - 1); |
31fd8581 | 3802 | } |
9bac3d6d | 3803 | |
8041902d | 3804 | new_prog = bpf_patch_insn_data(env, i + delta, insn_buf, cnt); |
9bac3d6d AS |
3805 | if (!new_prog) |
3806 | return -ENOMEM; | |
3807 | ||
3df126f3 | 3808 | delta += cnt - 1; |
9bac3d6d AS |
3809 | |
3810 | /* keep walking new program and skip insns we just inserted */ | |
3811 | env->prog = new_prog; | |
3df126f3 | 3812 | insn = new_prog->insnsi + i + delta; |
9bac3d6d AS |
3813 | } |
3814 | ||
3815 | return 0; | |
3816 | } | |
3817 | ||
79741b3b | 3818 | /* fixup insn->imm field of bpf_call instructions |
81ed18ab | 3819 | * and inline eligible helpers as explicit sequence of BPF instructions |
e245c5c6 AS |
3820 | * |
3821 | * this function is called after eBPF program passed verification | |
3822 | */ | |
79741b3b | 3823 | static int fixup_bpf_calls(struct bpf_verifier_env *env) |
e245c5c6 | 3824 | { |
79741b3b AS |
3825 | struct bpf_prog *prog = env->prog; |
3826 | struct bpf_insn *insn = prog->insnsi; | |
e245c5c6 | 3827 | const struct bpf_func_proto *fn; |
79741b3b | 3828 | const int insn_cnt = prog->len; |
81ed18ab AS |
3829 | struct bpf_insn insn_buf[16]; |
3830 | struct bpf_prog *new_prog; | |
3831 | struct bpf_map *map_ptr; | |
3832 | int i, cnt, delta = 0; | |
e245c5c6 | 3833 | |
79741b3b AS |
3834 | for (i = 0; i < insn_cnt; i++, insn++) { |
3835 | if (insn->code != (BPF_JMP | BPF_CALL)) | |
3836 | continue; | |
e245c5c6 | 3837 | |
79741b3b AS |
3838 | if (insn->imm == BPF_FUNC_get_route_realm) |
3839 | prog->dst_needed = 1; | |
3840 | if (insn->imm == BPF_FUNC_get_prandom_u32) | |
3841 | bpf_user_rnd_init_once(); | |
79741b3b | 3842 | if (insn->imm == BPF_FUNC_tail_call) { |
7b9f6da1 DM |
3843 | /* If we tail call into other programs, we |
3844 | * cannot make any assumptions since they can | |
3845 | * be replaced dynamically during runtime in | |
3846 | * the program array. | |
3847 | */ | |
3848 | prog->cb_access = 1; | |
80a58d02 | 3849 | env->prog->aux->stack_depth = MAX_BPF_STACK; |
7b9f6da1 | 3850 | |
79741b3b AS |
3851 | /* mark bpf_tail_call as different opcode to avoid |
3852 | * conditional branch in the interpeter for every normal | |
3853 | * call and to prevent accidental JITing by JIT compiler | |
3854 | * that doesn't support bpf_tail_call yet | |
e245c5c6 | 3855 | */ |
79741b3b | 3856 | insn->imm = 0; |
71189fa9 | 3857 | insn->code = BPF_JMP | BPF_TAIL_CALL; |
79741b3b AS |
3858 | continue; |
3859 | } | |
e245c5c6 | 3860 | |
81ed18ab AS |
3861 | if (ebpf_jit_enabled() && insn->imm == BPF_FUNC_map_lookup_elem) { |
3862 | map_ptr = env->insn_aux_data[i + delta].map_ptr; | |
fad73a1a MKL |
3863 | if (map_ptr == BPF_MAP_PTR_POISON || |
3864 | !map_ptr->ops->map_gen_lookup) | |
81ed18ab AS |
3865 | goto patch_call_imm; |
3866 | ||
3867 | cnt = map_ptr->ops->map_gen_lookup(map_ptr, insn_buf); | |
3868 | if (cnt == 0 || cnt >= ARRAY_SIZE(insn_buf)) { | |
3869 | verbose("bpf verifier is misconfigured\n"); | |
3870 | return -EINVAL; | |
3871 | } | |
3872 | ||
3873 | new_prog = bpf_patch_insn_data(env, i + delta, insn_buf, | |
3874 | cnt); | |
3875 | if (!new_prog) | |
3876 | return -ENOMEM; | |
3877 | ||
3878 | delta += cnt - 1; | |
3879 | ||
3880 | /* keep walking new program and skip insns we just inserted */ | |
3881 | env->prog = prog = new_prog; | |
3882 | insn = new_prog->insnsi + i + delta; | |
3883 | continue; | |
3884 | } | |
3885 | ||
3886 | patch_call_imm: | |
79741b3b AS |
3887 | fn = prog->aux->ops->get_func_proto(insn->imm); |
3888 | /* all functions that have prototype and verifier allowed | |
3889 | * programs to call them, must be real in-kernel functions | |
3890 | */ | |
3891 | if (!fn->func) { | |
3892 | verbose("kernel subsystem misconfigured func %s#%d\n", | |
3893 | func_id_name(insn->imm), insn->imm); | |
3894 | return -EFAULT; | |
e245c5c6 | 3895 | } |
79741b3b | 3896 | insn->imm = fn->func - __bpf_call_base; |
e245c5c6 | 3897 | } |
e245c5c6 | 3898 | |
79741b3b AS |
3899 | return 0; |
3900 | } | |
e245c5c6 | 3901 | |
58e2af8b | 3902 | static void free_states(struct bpf_verifier_env *env) |
f1bca824 | 3903 | { |
58e2af8b | 3904 | struct bpf_verifier_state_list *sl, *sln; |
f1bca824 AS |
3905 | int i; |
3906 | ||
3907 | if (!env->explored_states) | |
3908 | return; | |
3909 | ||
3910 | for (i = 0; i < env->prog->len; i++) { | |
3911 | sl = env->explored_states[i]; | |
3912 | ||
3913 | if (sl) | |
3914 | while (sl != STATE_LIST_MARK) { | |
3915 | sln = sl->next; | |
3916 | kfree(sl); | |
3917 | sl = sln; | |
3918 | } | |
3919 | } | |
3920 | ||
3921 | kfree(env->explored_states); | |
3922 | } | |
3923 | ||
9bac3d6d | 3924 | int bpf_check(struct bpf_prog **prog, union bpf_attr *attr) |
51580e79 | 3925 | { |
cbd35700 | 3926 | char __user *log_ubuf = NULL; |
58e2af8b | 3927 | struct bpf_verifier_env *env; |
51580e79 AS |
3928 | int ret = -EINVAL; |
3929 | ||
58e2af8b | 3930 | /* 'struct bpf_verifier_env' can be global, but since it's not small, |
cbd35700 AS |
3931 | * allocate/free it every time bpf_check() is called |
3932 | */ | |
58e2af8b | 3933 | env = kzalloc(sizeof(struct bpf_verifier_env), GFP_KERNEL); |
cbd35700 AS |
3934 | if (!env) |
3935 | return -ENOMEM; | |
3936 | ||
3df126f3 JK |
3937 | env->insn_aux_data = vzalloc(sizeof(struct bpf_insn_aux_data) * |
3938 | (*prog)->len); | |
3939 | ret = -ENOMEM; | |
3940 | if (!env->insn_aux_data) | |
3941 | goto err_free_env; | |
9bac3d6d | 3942 | env->prog = *prog; |
0246e64d | 3943 | |
cbd35700 AS |
3944 | /* grab the mutex to protect few globals used by verifier */ |
3945 | mutex_lock(&bpf_verifier_lock); | |
3946 | ||
3947 | if (attr->log_level || attr->log_buf || attr->log_size) { | |
3948 | /* user requested verbose verifier output | |
3949 | * and supplied buffer to store the verification trace | |
3950 | */ | |
3951 | log_level = attr->log_level; | |
3952 | log_ubuf = (char __user *) (unsigned long) attr->log_buf; | |
3953 | log_size = attr->log_size; | |
3954 | log_len = 0; | |
3955 | ||
3956 | ret = -EINVAL; | |
3957 | /* log_* values have to be sane */ | |
3958 | if (log_size < 128 || log_size > UINT_MAX >> 8 || | |
3959 | log_level == 0 || log_ubuf == NULL) | |
3df126f3 | 3960 | goto err_unlock; |
cbd35700 AS |
3961 | |
3962 | ret = -ENOMEM; | |
3963 | log_buf = vmalloc(log_size); | |
3964 | if (!log_buf) | |
3df126f3 | 3965 | goto err_unlock; |
cbd35700 AS |
3966 | } else { |
3967 | log_level = 0; | |
3968 | } | |
1ad2f583 DB |
3969 | |
3970 | env->strict_alignment = !!(attr->prog_flags & BPF_F_STRICT_ALIGNMENT); | |
3971 | if (!IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS)) | |
e07b98d9 | 3972 | env->strict_alignment = true; |
cbd35700 | 3973 | |
0246e64d AS |
3974 | ret = replace_map_fd_with_map_ptr(env); |
3975 | if (ret < 0) | |
3976 | goto skip_full_check; | |
3977 | ||
9bac3d6d | 3978 | env->explored_states = kcalloc(env->prog->len, |
58e2af8b | 3979 | sizeof(struct bpf_verifier_state_list *), |
f1bca824 AS |
3980 | GFP_USER); |
3981 | ret = -ENOMEM; | |
3982 | if (!env->explored_states) | |
3983 | goto skip_full_check; | |
3984 | ||
475fb78f AS |
3985 | ret = check_cfg(env); |
3986 | if (ret < 0) | |
3987 | goto skip_full_check; | |
3988 | ||
1be7f75d AS |
3989 | env->allow_ptr_leaks = capable(CAP_SYS_ADMIN); |
3990 | ||
17a52670 | 3991 | ret = do_check(env); |
cbd35700 | 3992 | |
0246e64d | 3993 | skip_full_check: |
17a52670 | 3994 | while (pop_stack(env, NULL) >= 0); |
f1bca824 | 3995 | free_states(env); |
0246e64d | 3996 | |
9bac3d6d AS |
3997 | if (ret == 0) |
3998 | /* program is valid, convert *(u32*)(ctx + off) accesses */ | |
3999 | ret = convert_ctx_accesses(env); | |
4000 | ||
e245c5c6 | 4001 | if (ret == 0) |
79741b3b | 4002 | ret = fixup_bpf_calls(env); |
e245c5c6 | 4003 | |
cbd35700 AS |
4004 | if (log_level && log_len >= log_size - 1) { |
4005 | BUG_ON(log_len >= log_size); | |
4006 | /* verifier log exceeded user supplied buffer */ | |
4007 | ret = -ENOSPC; | |
4008 | /* fall through to return what was recorded */ | |
4009 | } | |
4010 | ||
4011 | /* copy verifier log back to user space including trailing zero */ | |
4012 | if (log_level && copy_to_user(log_ubuf, log_buf, log_len + 1) != 0) { | |
4013 | ret = -EFAULT; | |
4014 | goto free_log_buf; | |
4015 | } | |
4016 | ||
0246e64d AS |
4017 | if (ret == 0 && env->used_map_cnt) { |
4018 | /* if program passed verifier, update used_maps in bpf_prog_info */ | |
9bac3d6d AS |
4019 | env->prog->aux->used_maps = kmalloc_array(env->used_map_cnt, |
4020 | sizeof(env->used_maps[0]), | |
4021 | GFP_KERNEL); | |
0246e64d | 4022 | |
9bac3d6d | 4023 | if (!env->prog->aux->used_maps) { |
0246e64d AS |
4024 | ret = -ENOMEM; |
4025 | goto free_log_buf; | |
4026 | } | |
4027 | ||
9bac3d6d | 4028 | memcpy(env->prog->aux->used_maps, env->used_maps, |
0246e64d | 4029 | sizeof(env->used_maps[0]) * env->used_map_cnt); |
9bac3d6d | 4030 | env->prog->aux->used_map_cnt = env->used_map_cnt; |
0246e64d AS |
4031 | |
4032 | /* program is valid. Convert pseudo bpf_ld_imm64 into generic | |
4033 | * bpf_ld_imm64 instructions | |
4034 | */ | |
4035 | convert_pseudo_ld_imm64(env); | |
4036 | } | |
cbd35700 AS |
4037 | |
4038 | free_log_buf: | |
4039 | if (log_level) | |
4040 | vfree(log_buf); | |
9bac3d6d | 4041 | if (!env->prog->aux->used_maps) |
0246e64d AS |
4042 | /* if we didn't copy map pointers into bpf_prog_info, release |
4043 | * them now. Otherwise free_bpf_prog_info() will release them. | |
4044 | */ | |
4045 | release_maps(env); | |
9bac3d6d | 4046 | *prog = env->prog; |
3df126f3 | 4047 | err_unlock: |
cbd35700 | 4048 | mutex_unlock(&bpf_verifier_lock); |
3df126f3 JK |
4049 | vfree(env->insn_aux_data); |
4050 | err_free_env: | |
4051 | kfree(env); | |
51580e79 AS |
4052 | return ret; |
4053 | } | |
13a27dfc JK |
4054 | |
4055 | int bpf_analyzer(struct bpf_prog *prog, const struct bpf_ext_analyzer_ops *ops, | |
4056 | void *priv) | |
4057 | { | |
4058 | struct bpf_verifier_env *env; | |
4059 | int ret; | |
4060 | ||
4061 | env = kzalloc(sizeof(struct bpf_verifier_env), GFP_KERNEL); | |
4062 | if (!env) | |
4063 | return -ENOMEM; | |
4064 | ||
4065 | env->insn_aux_data = vzalloc(sizeof(struct bpf_insn_aux_data) * | |
4066 | prog->len); | |
4067 | ret = -ENOMEM; | |
4068 | if (!env->insn_aux_data) | |
4069 | goto err_free_env; | |
4070 | env->prog = prog; | |
4071 | env->analyzer_ops = ops; | |
4072 | env->analyzer_priv = priv; | |
4073 | ||
4074 | /* grab the mutex to protect few globals used by verifier */ | |
4075 | mutex_lock(&bpf_verifier_lock); | |
4076 | ||
4077 | log_level = 0; | |
1ad2f583 | 4078 | |
e07b98d9 | 4079 | env->strict_alignment = false; |
1ad2f583 DB |
4080 | if (!IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS)) |
4081 | env->strict_alignment = true; | |
13a27dfc JK |
4082 | |
4083 | env->explored_states = kcalloc(env->prog->len, | |
4084 | sizeof(struct bpf_verifier_state_list *), | |
4085 | GFP_KERNEL); | |
4086 | ret = -ENOMEM; | |
4087 | if (!env->explored_states) | |
4088 | goto skip_full_check; | |
4089 | ||
4090 | ret = check_cfg(env); | |
4091 | if (ret < 0) | |
4092 | goto skip_full_check; | |
4093 | ||
4094 | env->allow_ptr_leaks = capable(CAP_SYS_ADMIN); | |
4095 | ||
4096 | ret = do_check(env); | |
4097 | ||
4098 | skip_full_check: | |
4099 | while (pop_stack(env, NULL) >= 0); | |
4100 | free_states(env); | |
4101 | ||
4102 | mutex_unlock(&bpf_verifier_lock); | |
4103 | vfree(env->insn_aux_data); | |
4104 | err_free_env: | |
4105 | kfree(env); | |
4106 | return ret; | |
4107 | } | |
4108 | EXPORT_SYMBOL_GPL(bpf_analyzer); |