1 /* Copyright (c) 2011-2014 PLUMgrid, http://plumgrid.com
2 * Copyright (c) 2016 Facebook
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.
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.
13 #include <linux/kernel.h>
14 #include <linux/types.h>
15 #include <linux/slab.h>
16 #include <linux/bpf.h>
17 #include <linux/bpf_verifier.h>
18 #include <linux/filter.h>
19 #include <net/netlink.h>
20 #include <linux/file.h>
21 #include <linux/vmalloc.h>
22 #include <linux/stringify.h>
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.
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
36 * analysis is limited to 64k insn, which may be hit even if total number of
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
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
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
51 * At the start of BPF program the register R1 contains a pointer to bpf_context
52 * and has type PTR_TO_CTX.
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.
64 * Most of the time the registers have UNKNOWN_VALUE type, which
65 * means the register has some value, but it's not a valid pointer.
66 * (like pointer plus pointer becomes UNKNOWN_VALUE type)
68 * When verifier sees load or store instructions the type of base register
69 * can be: PTR_TO_MAP_VALUE, PTR_TO_CTX, FRAME_PTR. These are three pointer
70 * types recognized by check_mem_access() function.
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.
75 * registers used to pass values to function calls are checked against
76 * function argument constraints.
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'
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,
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.
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)
96 * struct bpf_map *map = (struct bpf_map *) (unsigned long) r1;
97 * void *key = (void *) (unsigned long) r2;
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.
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
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.
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().
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.
131 /* verifier_state + insn_idx are pushed to stack when branch is encountered */
132 struct bpf_verifier_stack_elem {
133 /* verifer state is 'st'
134 * before processing instruction 'insn_idx'
135 * and after processing instruction 'prev_insn_idx'
137 struct bpf_verifier_state st;
140 struct bpf_verifier_stack_elem *next;
143 #define BPF_COMPLEXITY_LIMIT_INSNS 98304
144 #define BPF_COMPLEXITY_LIMIT_STACK 1024
146 #define BPF_MAP_PTR_POISON ((void *)0xeB9F + POISON_POINTER_DELTA)
148 struct bpf_call_arg_meta {
149 struct bpf_map *map_ptr;
156 /* verbose verifier prints what it's seeing
157 * bpf_check() is called under lock, so no race to access these global vars
159 static u32 log_level, log_size, log_len;
160 static char *log_buf;
162 static DEFINE_MUTEX(bpf_verifier_lock);
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
168 static __printf(1, 2) void verbose(const char *fmt, ...)
172 if (log_level == 0 || log_len >= log_size - 1)
176 log_len += vscnprintf(log_buf + log_len, log_size - log_len, fmt, args);
180 /* string representation of 'enum bpf_reg_type' */
181 static const char * const reg_type_str[] = {
183 [UNKNOWN_VALUE] = "inv",
184 [PTR_TO_CTX] = "ctx",
185 [CONST_PTR_TO_MAP] = "map_ptr",
186 [PTR_TO_MAP_VALUE] = "map_value",
187 [PTR_TO_MAP_VALUE_OR_NULL] = "map_value_or_null",
188 [PTR_TO_MAP_VALUE_ADJ] = "map_value_adj",
190 [PTR_TO_STACK] = "fp",
192 [PTR_TO_PACKET] = "pkt",
193 [PTR_TO_PACKET_END] = "pkt_end",
196 #define __BPF_FUNC_STR_FN(x) [BPF_FUNC_ ## x] = __stringify(bpf_ ## x)
197 static const char * const func_id_str[] = {
198 __BPF_FUNC_MAPPER(__BPF_FUNC_STR_FN)
200 #undef __BPF_FUNC_STR_FN
202 static const char *func_id_name(int id)
204 BUILD_BUG_ON(ARRAY_SIZE(func_id_str) != __BPF_FUNC_MAX_ID);
206 if (id >= 0 && id < __BPF_FUNC_MAX_ID && func_id_str[id])
207 return func_id_str[id];
212 static void print_verifier_state(struct bpf_verifier_state *state)
214 struct bpf_reg_state *reg;
218 for (i = 0; i < MAX_BPF_REG; i++) {
219 reg = &state->regs[i];
223 verbose(" R%d=%s", i, reg_type_str[t]);
224 if (t == CONST_IMM || t == PTR_TO_STACK)
225 verbose("%lld", reg->imm);
226 else if (t == PTR_TO_PACKET)
227 verbose("(id=%d,off=%d,r=%d)",
228 reg->id, reg->off, reg->range);
229 else if (t == UNKNOWN_VALUE && reg->imm)
230 verbose("%lld", reg->imm);
231 else if (t == CONST_PTR_TO_MAP || t == PTR_TO_MAP_VALUE ||
232 t == PTR_TO_MAP_VALUE_OR_NULL ||
233 t == PTR_TO_MAP_VALUE_ADJ)
234 verbose("(ks=%d,vs=%d,id=%u)",
235 reg->map_ptr->key_size,
236 reg->map_ptr->value_size,
238 if (reg->min_value != BPF_REGISTER_MIN_RANGE)
239 verbose(",min_value=%lld",
240 (long long)reg->min_value);
241 if (reg->max_value != BPF_REGISTER_MAX_RANGE)
242 verbose(",max_value=%llu",
243 (unsigned long long)reg->max_value);
245 verbose(",min_align=%u", reg->min_align);
247 verbose(",aux_off=%u", reg->aux_off);
248 if (reg->aux_off_align)
249 verbose(",aux_off_align=%u", reg->aux_off_align);
251 for (i = 0; i < MAX_BPF_STACK; i += BPF_REG_SIZE) {
252 if (state->stack_slot_type[i] == STACK_SPILL)
253 verbose(" fp%d=%s", -MAX_BPF_STACK + i,
254 reg_type_str[state->spilled_regs[i / BPF_REG_SIZE].type]);
259 static const char *const bpf_class_string[] = {
267 [BPF_ALU64] = "alu64",
270 static const char *const bpf_alu_string[16] = {
271 [BPF_ADD >> 4] = "+=",
272 [BPF_SUB >> 4] = "-=",
273 [BPF_MUL >> 4] = "*=",
274 [BPF_DIV >> 4] = "/=",
275 [BPF_OR >> 4] = "|=",
276 [BPF_AND >> 4] = "&=",
277 [BPF_LSH >> 4] = "<<=",
278 [BPF_RSH >> 4] = ">>=",
279 [BPF_NEG >> 4] = "neg",
280 [BPF_MOD >> 4] = "%=",
281 [BPF_XOR >> 4] = "^=",
282 [BPF_MOV >> 4] = "=",
283 [BPF_ARSH >> 4] = "s>>=",
284 [BPF_END >> 4] = "endian",
287 static const char *const bpf_ldst_string[] = {
288 [BPF_W >> 3] = "u32",
289 [BPF_H >> 3] = "u16",
291 [BPF_DW >> 3] = "u64",
294 static const char *const bpf_jmp_string[16] = {
295 [BPF_JA >> 4] = "jmp",
296 [BPF_JEQ >> 4] = "==",
297 [BPF_JGT >> 4] = ">",
298 [BPF_JGE >> 4] = ">=",
299 [BPF_JSET >> 4] = "&",
300 [BPF_JNE >> 4] = "!=",
301 [BPF_JSGT >> 4] = "s>",
302 [BPF_JSGE >> 4] = "s>=",
303 [BPF_CALL >> 4] = "call",
304 [BPF_EXIT >> 4] = "exit",
307 static void print_bpf_insn(const struct bpf_verifier_env *env,
308 const struct bpf_insn *insn)
310 u8 class = BPF_CLASS(insn->code);
312 if (class == BPF_ALU || class == BPF_ALU64) {
313 if (BPF_SRC(insn->code) == BPF_X)
314 verbose("(%02x) %sr%d %s %sr%d\n",
315 insn->code, class == BPF_ALU ? "(u32) " : "",
317 bpf_alu_string[BPF_OP(insn->code) >> 4],
318 class == BPF_ALU ? "(u32) " : "",
321 verbose("(%02x) %sr%d %s %s%d\n",
322 insn->code, class == BPF_ALU ? "(u32) " : "",
324 bpf_alu_string[BPF_OP(insn->code) >> 4],
325 class == BPF_ALU ? "(u32) " : "",
327 } else if (class == BPF_STX) {
328 if (BPF_MODE(insn->code) == BPF_MEM)
329 verbose("(%02x) *(%s *)(r%d %+d) = r%d\n",
331 bpf_ldst_string[BPF_SIZE(insn->code) >> 3],
333 insn->off, insn->src_reg);
334 else if (BPF_MODE(insn->code) == BPF_XADD)
335 verbose("(%02x) lock *(%s *)(r%d %+d) += r%d\n",
337 bpf_ldst_string[BPF_SIZE(insn->code) >> 3],
338 insn->dst_reg, insn->off,
341 verbose("BUG_%02x\n", insn->code);
342 } else if (class == BPF_ST) {
343 if (BPF_MODE(insn->code) != BPF_MEM) {
344 verbose("BUG_st_%02x\n", insn->code);
347 verbose("(%02x) *(%s *)(r%d %+d) = %d\n",
349 bpf_ldst_string[BPF_SIZE(insn->code) >> 3],
351 insn->off, insn->imm);
352 } else if (class == BPF_LDX) {
353 if (BPF_MODE(insn->code) != BPF_MEM) {
354 verbose("BUG_ldx_%02x\n", insn->code);
357 verbose("(%02x) r%d = *(%s *)(r%d %+d)\n",
358 insn->code, insn->dst_reg,
359 bpf_ldst_string[BPF_SIZE(insn->code) >> 3],
360 insn->src_reg, insn->off);
361 } else if (class == BPF_LD) {
362 if (BPF_MODE(insn->code) == BPF_ABS) {
363 verbose("(%02x) r0 = *(%s *)skb[%d]\n",
365 bpf_ldst_string[BPF_SIZE(insn->code) >> 3],
367 } else if (BPF_MODE(insn->code) == BPF_IND) {
368 verbose("(%02x) r0 = *(%s *)skb[r%d + %d]\n",
370 bpf_ldst_string[BPF_SIZE(insn->code) >> 3],
371 insn->src_reg, insn->imm);
372 } else if (BPF_MODE(insn->code) == BPF_IMM &&
373 BPF_SIZE(insn->code) == BPF_DW) {
374 /* At this point, we already made sure that the second
375 * part of the ldimm64 insn is accessible.
377 u64 imm = ((u64)(insn + 1)->imm << 32) | (u32)insn->imm;
378 bool map_ptr = insn->src_reg == BPF_PSEUDO_MAP_FD;
380 if (map_ptr && !env->allow_ptr_leaks)
383 verbose("(%02x) r%d = 0x%llx\n", insn->code,
384 insn->dst_reg, (unsigned long long)imm);
386 verbose("BUG_ld_%02x\n", insn->code);
389 } else if (class == BPF_JMP) {
390 u8 opcode = BPF_OP(insn->code);
392 if (opcode == BPF_CALL) {
393 verbose("(%02x) call %s#%d\n", insn->code,
394 func_id_name(insn->imm), insn->imm);
395 } else if (insn->code == (BPF_JMP | BPF_JA)) {
396 verbose("(%02x) goto pc%+d\n",
397 insn->code, insn->off);
398 } else if (insn->code == (BPF_JMP | BPF_EXIT)) {
399 verbose("(%02x) exit\n", insn->code);
400 } else if (BPF_SRC(insn->code) == BPF_X) {
401 verbose("(%02x) if r%d %s r%d goto pc%+d\n",
402 insn->code, insn->dst_reg,
403 bpf_jmp_string[BPF_OP(insn->code) >> 4],
404 insn->src_reg, insn->off);
406 verbose("(%02x) if r%d %s 0x%x goto pc%+d\n",
407 insn->code, insn->dst_reg,
408 bpf_jmp_string[BPF_OP(insn->code) >> 4],
409 insn->imm, insn->off);
412 verbose("(%02x) %s\n", insn->code, bpf_class_string[class]);
416 static int pop_stack(struct bpf_verifier_env *env, int *prev_insn_idx)
418 struct bpf_verifier_stack_elem *elem;
421 if (env->head == NULL)
424 memcpy(&env->cur_state, &env->head->st, sizeof(env->cur_state));
425 insn_idx = env->head->insn_idx;
427 *prev_insn_idx = env->head->prev_insn_idx;
428 elem = env->head->next;
435 static struct bpf_verifier_state *push_stack(struct bpf_verifier_env *env,
436 int insn_idx, int prev_insn_idx)
438 struct bpf_verifier_stack_elem *elem;
440 elem = kmalloc(sizeof(struct bpf_verifier_stack_elem), GFP_KERNEL);
444 memcpy(&elem->st, &env->cur_state, sizeof(env->cur_state));
445 elem->insn_idx = insn_idx;
446 elem->prev_insn_idx = prev_insn_idx;
447 elem->next = env->head;
450 if (env->stack_size > BPF_COMPLEXITY_LIMIT_STACK) {
451 verbose("BPF program is too complex\n");
456 /* pop all elements and return */
457 while (pop_stack(env, NULL) >= 0);
461 #define CALLER_SAVED_REGS 6
462 static const int caller_saved[CALLER_SAVED_REGS] = {
463 BPF_REG_0, BPF_REG_1, BPF_REG_2, BPF_REG_3, BPF_REG_4, BPF_REG_5
466 static void init_reg_state(struct bpf_reg_state *regs)
470 for (i = 0; i < MAX_BPF_REG; i++) {
471 regs[i].type = NOT_INIT;
473 regs[i].min_value = BPF_REGISTER_MIN_RANGE;
474 regs[i].max_value = BPF_REGISTER_MAX_RANGE;
475 regs[i].min_align = 0;
477 regs[i].aux_off_align = 0;
481 regs[BPF_REG_FP].type = FRAME_PTR;
483 /* 1st arg to a function */
484 regs[BPF_REG_1].type = PTR_TO_CTX;
487 static void __mark_reg_unknown_value(struct bpf_reg_state *regs, u32 regno)
489 regs[regno].type = UNKNOWN_VALUE;
494 static void mark_reg_unknown_value(struct bpf_reg_state *regs, u32 regno)
496 BUG_ON(regno >= MAX_BPF_REG);
497 __mark_reg_unknown_value(regs, regno);
500 static void reset_reg_range_values(struct bpf_reg_state *regs, u32 regno)
502 regs[regno].min_value = BPF_REGISTER_MIN_RANGE;
503 regs[regno].max_value = BPF_REGISTER_MAX_RANGE;
504 regs[regno].min_align = 0;
507 static void mark_reg_unknown_value_and_range(struct bpf_reg_state *regs,
510 mark_reg_unknown_value(regs, regno);
511 reset_reg_range_values(regs, regno);
515 SRC_OP, /* register is used as source operand */
516 DST_OP, /* register is used as destination operand */
517 DST_OP_NO_MARK /* same as above, check only, don't mark */
520 static int check_reg_arg(struct bpf_reg_state *regs, u32 regno,
523 if (regno >= MAX_BPF_REG) {
524 verbose("R%d is invalid\n", regno);
529 /* check whether register used as source operand can be read */
530 if (regs[regno].type == NOT_INIT) {
531 verbose("R%d !read_ok\n", regno);
535 /* check whether register used as dest operand can be written to */
536 if (regno == BPF_REG_FP) {
537 verbose("frame pointer is read only\n");
541 mark_reg_unknown_value(regs, regno);
546 static int bpf_size_to_bytes(int bpf_size)
548 if (bpf_size == BPF_W)
550 else if (bpf_size == BPF_H)
552 else if (bpf_size == BPF_B)
554 else if (bpf_size == BPF_DW)
560 static bool is_spillable_regtype(enum bpf_reg_type type)
563 case PTR_TO_MAP_VALUE:
564 case PTR_TO_MAP_VALUE_OR_NULL:
565 case PTR_TO_MAP_VALUE_ADJ:
569 case PTR_TO_PACKET_END:
571 case CONST_PTR_TO_MAP:
578 /* check_stack_read/write functions track spill/fill of registers,
579 * stack boundary and alignment are checked in check_mem_access()
581 static int check_stack_write(struct bpf_verifier_state *state, int off,
582 int size, int value_regno)
585 /* caller checked that off % size == 0 and -MAX_BPF_STACK <= off < 0,
586 * so it's aligned access and [off, off + size) are within stack limits
589 if (value_regno >= 0 &&
590 is_spillable_regtype(state->regs[value_regno].type)) {
592 /* register containing pointer is being spilled into stack */
593 if (size != BPF_REG_SIZE) {
594 verbose("invalid size of register spill\n");
598 /* save register state */
599 state->spilled_regs[(MAX_BPF_STACK + off) / BPF_REG_SIZE] =
600 state->regs[value_regno];
602 for (i = 0; i < BPF_REG_SIZE; i++)
603 state->stack_slot_type[MAX_BPF_STACK + off + i] = STACK_SPILL;
605 /* regular write of data into stack */
606 state->spilled_regs[(MAX_BPF_STACK + off) / BPF_REG_SIZE] =
607 (struct bpf_reg_state) {};
609 for (i = 0; i < size; i++)
610 state->stack_slot_type[MAX_BPF_STACK + off + i] = STACK_MISC;
615 static int check_stack_read(struct bpf_verifier_state *state, int off, int size,
621 slot_type = &state->stack_slot_type[MAX_BPF_STACK + off];
623 if (slot_type[0] == STACK_SPILL) {
624 if (size != BPF_REG_SIZE) {
625 verbose("invalid size of register spill\n");
628 for (i = 1; i < BPF_REG_SIZE; i++) {
629 if (slot_type[i] != STACK_SPILL) {
630 verbose("corrupted spill memory\n");
635 if (value_regno >= 0)
636 /* restore register state from stack */
637 state->regs[value_regno] =
638 state->spilled_regs[(MAX_BPF_STACK + off) / BPF_REG_SIZE];
641 for (i = 0; i < size; i++) {
642 if (slot_type[i] != STACK_MISC) {
643 verbose("invalid read from stack off %d+%d size %d\n",
648 if (value_regno >= 0)
649 /* have read misc data from the stack */
650 mark_reg_unknown_value_and_range(state->regs,
656 /* check read/write into map element returned by bpf_map_lookup_elem() */
657 static int check_map_access(struct bpf_verifier_env *env, u32 regno, int off,
660 struct bpf_map *map = env->cur_state.regs[regno].map_ptr;
662 if (off < 0 || size <= 0 || off + size > map->value_size) {
663 verbose("invalid access to map value, value_size=%d off=%d size=%d\n",
664 map->value_size, off, size);
670 /* check read/write into an adjusted map element */
671 static int check_map_access_adj(struct bpf_verifier_env *env, u32 regno,
674 struct bpf_verifier_state *state = &env->cur_state;
675 struct bpf_reg_state *reg = &state->regs[regno];
678 /* We adjusted the register to this map value, so we
679 * need to change off and size to min_value and max_value
680 * respectively to make sure our theoretical access will be
684 print_verifier_state(state);
685 env->varlen_map_value_access = true;
686 /* The minimum value is only important with signed
687 * comparisons where we can't assume the floor of a
688 * value is 0. If we are using signed variables for our
689 * index'es we need to make sure that whatever we use
690 * will have a set floor within our range.
692 if (reg->min_value < 0) {
693 verbose("R%d min value is negative, either use unsigned index or do a if (index >=0) check.\n",
697 err = check_map_access(env, regno, reg->min_value + off, size);
699 verbose("R%d min value is outside of the array range\n",
704 /* If we haven't set a max value then we need to bail
705 * since we can't be sure we won't do bad things.
707 if (reg->max_value == BPF_REGISTER_MAX_RANGE) {
708 verbose("R%d unbounded memory access, make sure to bounds check any array access into a map\n",
712 return check_map_access(env, regno, reg->max_value + off, size);
715 #define MAX_PACKET_OFF 0xffff
717 static bool may_access_direct_pkt_data(struct bpf_verifier_env *env,
718 const struct bpf_call_arg_meta *meta,
719 enum bpf_access_type t)
721 switch (env->prog->type) {
722 case BPF_PROG_TYPE_LWT_IN:
723 case BPF_PROG_TYPE_LWT_OUT:
724 /* dst_input() and dst_output() can't write for now */
728 case BPF_PROG_TYPE_SCHED_CLS:
729 case BPF_PROG_TYPE_SCHED_ACT:
730 case BPF_PROG_TYPE_XDP:
731 case BPF_PROG_TYPE_LWT_XMIT:
733 return meta->pkt_access;
735 env->seen_direct_write = true;
742 static int check_packet_access(struct bpf_verifier_env *env, u32 regno, int off,
745 struct bpf_reg_state *regs = env->cur_state.regs;
746 struct bpf_reg_state *reg = ®s[regno];
749 if (off < 0 || size <= 0 || off + size > reg->range) {
750 verbose("invalid access to packet, off=%d size=%d, R%d(id=%d,off=%d,r=%d)\n",
751 off, size, regno, reg->id, reg->off, reg->range);
757 /* check access to 'struct bpf_context' fields */
758 static int check_ctx_access(struct bpf_verifier_env *env, int off, int size,
759 enum bpf_access_type t, enum bpf_reg_type *reg_type)
761 /* for analyzer ctx accesses are already validated and converted */
762 if (env->analyzer_ops)
765 if (env->prog->aux->ops->is_valid_access &&
766 env->prog->aux->ops->is_valid_access(off, size, t, reg_type)) {
767 /* remember the offset of last byte accessed in ctx */
768 if (env->prog->aux->max_ctx_offset < off + size)
769 env->prog->aux->max_ctx_offset = off + size;
773 verbose("invalid bpf_context access off=%d size=%d\n", off, size);
777 static bool is_pointer_value(struct bpf_verifier_env *env, int regno)
779 if (env->allow_ptr_leaks)
782 switch (env->cur_state.regs[regno].type) {
791 static int check_pkt_ptr_alignment(const struct bpf_reg_state *reg,
792 int off, int size, bool strict)
797 /* Byte size accesses are always allowed. */
798 if (!strict || size == 1)
803 if (reg->aux_off_align % size) {
804 verbose("Packet access is only %u byte aligned, %d byte access not allowed\n",
805 reg->aux_off_align, size);
808 reg_off += reg->aux_off;
811 /* skb->data is NET_IP_ALIGN-ed, but for strict alignment checking
812 * we force this to 2 which is universally what architectures use
813 * when they don't set CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS.
815 ip_align = strict ? 2 : NET_IP_ALIGN;
816 if ((ip_align + reg_off + off) % size != 0) {
817 verbose("misaligned packet access off %d+%d+%d size %d\n",
818 ip_align, reg_off, off, size);
825 static int check_val_ptr_alignment(const struct bpf_reg_state *reg,
826 int size, bool strict)
828 if (strict && size != 1) {
829 verbose("Unknown alignment. Only byte-sized access allowed in value access.\n");
836 static int check_ptr_alignment(struct bpf_verifier_env *env,
837 const struct bpf_reg_state *reg,
840 bool strict = env->strict_alignment;
842 if (!IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS))
847 return check_pkt_ptr_alignment(reg, off, size, strict);
848 case PTR_TO_MAP_VALUE_ADJ:
849 return check_val_ptr_alignment(reg, size, strict);
851 if (off % size != 0) {
852 verbose("misaligned access off %d size %d\n",
861 /* check whether memory at (regno + off) is accessible for t = (read | write)
862 * if t==write, value_regno is a register which value is stored into memory
863 * if t==read, value_regno is a register which will receive the value from memory
864 * if t==write && value_regno==-1, some unknown value is stored into memory
865 * if t==read && value_regno==-1, don't care what we read from memory
867 static int check_mem_access(struct bpf_verifier_env *env, u32 regno, int off,
868 int bpf_size, enum bpf_access_type t,
871 struct bpf_verifier_state *state = &env->cur_state;
872 struct bpf_reg_state *reg = &state->regs[regno];
875 if (reg->type == PTR_TO_STACK)
878 size = bpf_size_to_bytes(bpf_size);
882 err = check_ptr_alignment(env, reg, off, size);
886 if (reg->type == PTR_TO_MAP_VALUE ||
887 reg->type == PTR_TO_MAP_VALUE_ADJ) {
888 if (t == BPF_WRITE && value_regno >= 0 &&
889 is_pointer_value(env, value_regno)) {
890 verbose("R%d leaks addr into map\n", value_regno);
894 if (reg->type == PTR_TO_MAP_VALUE_ADJ)
895 err = check_map_access_adj(env, regno, off, size);
897 err = check_map_access(env, regno, off, size);
898 if (!err && t == BPF_READ && value_regno >= 0)
899 mark_reg_unknown_value_and_range(state->regs,
902 } else if (reg->type == PTR_TO_CTX) {
903 enum bpf_reg_type reg_type = UNKNOWN_VALUE;
905 if (t == BPF_WRITE && value_regno >= 0 &&
906 is_pointer_value(env, value_regno)) {
907 verbose("R%d leaks addr into ctx\n", value_regno);
910 err = check_ctx_access(env, off, size, t, ®_type);
911 if (!err && t == BPF_READ && value_regno >= 0) {
912 mark_reg_unknown_value_and_range(state->regs,
914 /* note that reg.[id|off|range] == 0 */
915 state->regs[value_regno].type = reg_type;
916 state->regs[value_regno].aux_off = 0;
917 state->regs[value_regno].aux_off_align = 0;
920 } else if (reg->type == FRAME_PTR || reg->type == PTR_TO_STACK) {
921 if (off >= 0 || off < -MAX_BPF_STACK) {
922 verbose("invalid stack off=%d size=%d\n", off, size);
925 if (t == BPF_WRITE) {
926 if (!env->allow_ptr_leaks &&
927 state->stack_slot_type[MAX_BPF_STACK + off] == STACK_SPILL &&
928 size != BPF_REG_SIZE) {
929 verbose("attempt to corrupt spilled pointer on stack\n");
932 err = check_stack_write(state, off, size, value_regno);
934 err = check_stack_read(state, off, size, value_regno);
936 } else if (state->regs[regno].type == PTR_TO_PACKET) {
937 if (t == BPF_WRITE && !may_access_direct_pkt_data(env, NULL, t)) {
938 verbose("cannot write into packet\n");
941 if (t == BPF_WRITE && value_regno >= 0 &&
942 is_pointer_value(env, value_regno)) {
943 verbose("R%d leaks addr into packet\n", value_regno);
946 err = check_packet_access(env, regno, off, size);
947 if (!err && t == BPF_READ && value_regno >= 0)
948 mark_reg_unknown_value_and_range(state->regs,
951 verbose("R%d invalid mem access '%s'\n",
952 regno, reg_type_str[reg->type]);
956 if (!err && size <= 2 && value_regno >= 0 && env->allow_ptr_leaks &&
957 state->regs[value_regno].type == UNKNOWN_VALUE) {
958 /* 1 or 2 byte load zero-extends, determine the number of
959 * zero upper bits. Not doing it fo 4 byte load, since
960 * such values cannot be added to ptr_to_packet anyway.
962 state->regs[value_regno].imm = 64 - size * 8;
967 static int check_xadd(struct bpf_verifier_env *env, struct bpf_insn *insn)
969 struct bpf_reg_state *regs = env->cur_state.regs;
972 if ((BPF_SIZE(insn->code) != BPF_W && BPF_SIZE(insn->code) != BPF_DW) ||
974 verbose("BPF_XADD uses reserved fields\n");
978 /* check src1 operand */
979 err = check_reg_arg(regs, insn->src_reg, SRC_OP);
983 /* check src2 operand */
984 err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
988 /* check whether atomic_add can read the memory */
989 err = check_mem_access(env, insn->dst_reg, insn->off,
990 BPF_SIZE(insn->code), BPF_READ, -1);
994 /* check whether atomic_add can write into the same memory */
995 return check_mem_access(env, insn->dst_reg, insn->off,
996 BPF_SIZE(insn->code), BPF_WRITE, -1);
999 /* when register 'regno' is passed into function that will read 'access_size'
1000 * bytes from that pointer, make sure that it's within stack boundary
1001 * and all elements of stack are initialized
1003 static int check_stack_boundary(struct bpf_verifier_env *env, int regno,
1004 int access_size, bool zero_size_allowed,
1005 struct bpf_call_arg_meta *meta)
1007 struct bpf_verifier_state *state = &env->cur_state;
1008 struct bpf_reg_state *regs = state->regs;
1011 if (regs[regno].type != PTR_TO_STACK) {
1012 if (zero_size_allowed && access_size == 0 &&
1013 regs[regno].type == CONST_IMM &&
1014 regs[regno].imm == 0)
1017 verbose("R%d type=%s expected=%s\n", regno,
1018 reg_type_str[regs[regno].type],
1019 reg_type_str[PTR_TO_STACK]);
1023 off = regs[regno].imm;
1024 if (off >= 0 || off < -MAX_BPF_STACK || off + access_size > 0 ||
1026 verbose("invalid stack type R%d off=%d access_size=%d\n",
1027 regno, off, access_size);
1031 if (meta && meta->raw_mode) {
1032 meta->access_size = access_size;
1033 meta->regno = regno;
1037 for (i = 0; i < access_size; i++) {
1038 if (state->stack_slot_type[MAX_BPF_STACK + off + i] != STACK_MISC) {
1039 verbose("invalid indirect read from stack off %d+%d size %d\n",
1040 off, i, access_size);
1047 static int check_helper_mem_access(struct bpf_verifier_env *env, int regno,
1048 int access_size, bool zero_size_allowed,
1049 struct bpf_call_arg_meta *meta)
1051 struct bpf_reg_state *regs = env->cur_state.regs;
1053 switch (regs[regno].type) {
1055 return check_packet_access(env, regno, 0, access_size);
1056 case PTR_TO_MAP_VALUE:
1057 return check_map_access(env, regno, 0, access_size);
1058 case PTR_TO_MAP_VALUE_ADJ:
1059 return check_map_access_adj(env, regno, 0, access_size);
1060 default: /* const_imm|ptr_to_stack or invalid ptr */
1061 return check_stack_boundary(env, regno, access_size,
1062 zero_size_allowed, meta);
1066 static int check_func_arg(struct bpf_verifier_env *env, u32 regno,
1067 enum bpf_arg_type arg_type,
1068 struct bpf_call_arg_meta *meta)
1070 struct bpf_reg_state *regs = env->cur_state.regs, *reg = ®s[regno];
1071 enum bpf_reg_type expected_type, type = reg->type;
1074 if (arg_type == ARG_DONTCARE)
1077 if (type == NOT_INIT) {
1078 verbose("R%d !read_ok\n", regno);
1082 if (arg_type == ARG_ANYTHING) {
1083 if (is_pointer_value(env, regno)) {
1084 verbose("R%d leaks addr into helper function\n", regno);
1090 if (type == PTR_TO_PACKET &&
1091 !may_access_direct_pkt_data(env, meta, BPF_READ)) {
1092 verbose("helper access to the packet is not allowed\n");
1096 if (arg_type == ARG_PTR_TO_MAP_KEY ||
1097 arg_type == ARG_PTR_TO_MAP_VALUE) {
1098 expected_type = PTR_TO_STACK;
1099 if (type != PTR_TO_PACKET && type != expected_type)
1101 } else if (arg_type == ARG_CONST_SIZE ||
1102 arg_type == ARG_CONST_SIZE_OR_ZERO) {
1103 expected_type = CONST_IMM;
1104 /* One exception. Allow UNKNOWN_VALUE registers when the
1105 * boundaries are known and don't cause unsafe memory accesses
1107 if (type != UNKNOWN_VALUE && type != expected_type)
1109 } else if (arg_type == ARG_CONST_MAP_PTR) {
1110 expected_type = CONST_PTR_TO_MAP;
1111 if (type != expected_type)
1113 } else if (arg_type == ARG_PTR_TO_CTX) {
1114 expected_type = PTR_TO_CTX;
1115 if (type != expected_type)
1117 } else if (arg_type == ARG_PTR_TO_MEM ||
1118 arg_type == ARG_PTR_TO_UNINIT_MEM) {
1119 expected_type = PTR_TO_STACK;
1120 /* One exception here. In case function allows for NULL to be
1121 * passed in as argument, it's a CONST_IMM type. Final test
1122 * happens during stack boundary checking.
1124 if (type == CONST_IMM && reg->imm == 0)
1125 /* final test in check_stack_boundary() */;
1126 else if (type != PTR_TO_PACKET && type != PTR_TO_MAP_VALUE &&
1127 type != PTR_TO_MAP_VALUE_ADJ && type != expected_type)
1129 meta->raw_mode = arg_type == ARG_PTR_TO_UNINIT_MEM;
1131 verbose("unsupported arg_type %d\n", arg_type);
1135 if (arg_type == ARG_CONST_MAP_PTR) {
1136 /* bpf_map_xxx(map_ptr) call: remember that map_ptr */
1137 meta->map_ptr = reg->map_ptr;
1138 } else if (arg_type == ARG_PTR_TO_MAP_KEY) {
1139 /* bpf_map_xxx(..., map_ptr, ..., key) call:
1140 * check that [key, key + map->key_size) are within
1141 * stack limits and initialized
1143 if (!meta->map_ptr) {
1144 /* in function declaration map_ptr must come before
1145 * map_key, so that it's verified and known before
1146 * we have to check map_key here. Otherwise it means
1147 * that kernel subsystem misconfigured verifier
1149 verbose("invalid map_ptr to access map->key\n");
1152 if (type == PTR_TO_PACKET)
1153 err = check_packet_access(env, regno, 0,
1154 meta->map_ptr->key_size);
1156 err = check_stack_boundary(env, regno,
1157 meta->map_ptr->key_size,
1159 } else if (arg_type == ARG_PTR_TO_MAP_VALUE) {
1160 /* bpf_map_xxx(..., map_ptr, ..., value) call:
1161 * check [value, value + map->value_size) validity
1163 if (!meta->map_ptr) {
1164 /* kernel subsystem misconfigured verifier */
1165 verbose("invalid map_ptr to access map->value\n");
1168 if (type == PTR_TO_PACKET)
1169 err = check_packet_access(env, regno, 0,
1170 meta->map_ptr->value_size);
1172 err = check_stack_boundary(env, regno,
1173 meta->map_ptr->value_size,
1175 } else if (arg_type == ARG_CONST_SIZE ||
1176 arg_type == ARG_CONST_SIZE_OR_ZERO) {
1177 bool zero_size_allowed = (arg_type == ARG_CONST_SIZE_OR_ZERO);
1179 /* bpf_xxx(..., buf, len) call will access 'len' bytes
1180 * from stack pointer 'buf'. Check it
1181 * note: regno == len, regno - 1 == buf
1184 /* kernel subsystem misconfigured verifier */
1185 verbose("ARG_CONST_SIZE cannot be first argument\n");
1189 /* If the register is UNKNOWN_VALUE, the access check happens
1190 * using its boundaries. Otherwise, just use its imm
1192 if (type == UNKNOWN_VALUE) {
1193 /* For unprivileged variable accesses, disable raw
1194 * mode so that the program is required to
1195 * initialize all the memory that the helper could
1196 * just partially fill up.
1200 if (reg->min_value < 0) {
1201 verbose("R%d min value is negative, either use unsigned or 'var &= const'\n",
1206 if (reg->min_value == 0) {
1207 err = check_helper_mem_access(env, regno - 1, 0,
1214 if (reg->max_value == BPF_REGISTER_MAX_RANGE) {
1215 verbose("R%d unbounded memory access, use 'var &= const' or 'if (var < const)'\n",
1219 err = check_helper_mem_access(env, regno - 1,
1221 zero_size_allowed, meta);
1225 /* register is CONST_IMM */
1226 err = check_helper_mem_access(env, regno - 1, reg->imm,
1227 zero_size_allowed, meta);
1233 verbose("R%d type=%s expected=%s\n", regno,
1234 reg_type_str[type], reg_type_str[expected_type]);
1238 static int check_map_func_compatibility(struct bpf_map *map, int func_id)
1243 /* We need a two way check, first is from map perspective ... */
1244 switch (map->map_type) {
1245 case BPF_MAP_TYPE_PROG_ARRAY:
1246 if (func_id != BPF_FUNC_tail_call)
1249 case BPF_MAP_TYPE_PERF_EVENT_ARRAY:
1250 if (func_id != BPF_FUNC_perf_event_read &&
1251 func_id != BPF_FUNC_perf_event_output)
1254 case BPF_MAP_TYPE_STACK_TRACE:
1255 if (func_id != BPF_FUNC_get_stackid)
1258 case BPF_MAP_TYPE_CGROUP_ARRAY:
1259 if (func_id != BPF_FUNC_skb_under_cgroup &&
1260 func_id != BPF_FUNC_current_task_under_cgroup)
1263 case BPF_MAP_TYPE_ARRAY_OF_MAPS:
1264 case BPF_MAP_TYPE_HASH_OF_MAPS:
1265 if (func_id != BPF_FUNC_map_lookup_elem)
1271 /* ... and second from the function itself. */
1273 case BPF_FUNC_tail_call:
1274 if (map->map_type != BPF_MAP_TYPE_PROG_ARRAY)
1277 case BPF_FUNC_perf_event_read:
1278 case BPF_FUNC_perf_event_output:
1279 if (map->map_type != BPF_MAP_TYPE_PERF_EVENT_ARRAY)
1282 case BPF_FUNC_get_stackid:
1283 if (map->map_type != BPF_MAP_TYPE_STACK_TRACE)
1286 case BPF_FUNC_current_task_under_cgroup:
1287 case BPF_FUNC_skb_under_cgroup:
1288 if (map->map_type != BPF_MAP_TYPE_CGROUP_ARRAY)
1297 verbose("cannot pass map_type %d into func %s#%d\n",
1298 map->map_type, func_id_name(func_id), func_id);
1302 static int check_raw_mode(const struct bpf_func_proto *fn)
1306 if (fn->arg1_type == ARG_PTR_TO_UNINIT_MEM)
1308 if (fn->arg2_type == ARG_PTR_TO_UNINIT_MEM)
1310 if (fn->arg3_type == ARG_PTR_TO_UNINIT_MEM)
1312 if (fn->arg4_type == ARG_PTR_TO_UNINIT_MEM)
1314 if (fn->arg5_type == ARG_PTR_TO_UNINIT_MEM)
1317 return count > 1 ? -EINVAL : 0;
1320 static void clear_all_pkt_pointers(struct bpf_verifier_env *env)
1322 struct bpf_verifier_state *state = &env->cur_state;
1323 struct bpf_reg_state *regs = state->regs, *reg;
1326 for (i = 0; i < MAX_BPF_REG; i++)
1327 if (regs[i].type == PTR_TO_PACKET ||
1328 regs[i].type == PTR_TO_PACKET_END)
1329 mark_reg_unknown_value(regs, i);
1331 for (i = 0; i < MAX_BPF_STACK; i += BPF_REG_SIZE) {
1332 if (state->stack_slot_type[i] != STACK_SPILL)
1334 reg = &state->spilled_regs[i / BPF_REG_SIZE];
1335 if (reg->type != PTR_TO_PACKET &&
1336 reg->type != PTR_TO_PACKET_END)
1338 reg->type = UNKNOWN_VALUE;
1343 static int check_call(struct bpf_verifier_env *env, int func_id, int insn_idx)
1345 struct bpf_verifier_state *state = &env->cur_state;
1346 const struct bpf_func_proto *fn = NULL;
1347 struct bpf_reg_state *regs = state->regs;
1348 struct bpf_reg_state *reg;
1349 struct bpf_call_arg_meta meta;
1353 /* find function prototype */
1354 if (func_id < 0 || func_id >= __BPF_FUNC_MAX_ID) {
1355 verbose("invalid func %s#%d\n", func_id_name(func_id), func_id);
1359 if (env->prog->aux->ops->get_func_proto)
1360 fn = env->prog->aux->ops->get_func_proto(func_id);
1363 verbose("unknown func %s#%d\n", func_id_name(func_id), func_id);
1367 /* eBPF programs must be GPL compatible to use GPL-ed functions */
1368 if (!env->prog->gpl_compatible && fn->gpl_only) {
1369 verbose("cannot call GPL only function from proprietary program\n");
1373 changes_data = bpf_helper_changes_pkt_data(fn->func);
1375 memset(&meta, 0, sizeof(meta));
1376 meta.pkt_access = fn->pkt_access;
1378 /* We only support one arg being in raw mode at the moment, which
1379 * is sufficient for the helper functions we have right now.
1381 err = check_raw_mode(fn);
1383 verbose("kernel subsystem misconfigured func %s#%d\n",
1384 func_id_name(func_id), func_id);
1389 err = check_func_arg(env, BPF_REG_1, fn->arg1_type, &meta);
1392 err = check_func_arg(env, BPF_REG_2, fn->arg2_type, &meta);
1395 err = check_func_arg(env, BPF_REG_3, fn->arg3_type, &meta);
1398 err = check_func_arg(env, BPF_REG_4, fn->arg4_type, &meta);
1401 err = check_func_arg(env, BPF_REG_5, fn->arg5_type, &meta);
1405 /* Mark slots with STACK_MISC in case of raw mode, stack offset
1406 * is inferred from register state.
1408 for (i = 0; i < meta.access_size; i++) {
1409 err = check_mem_access(env, meta.regno, i, BPF_B, BPF_WRITE, -1);
1414 /* reset caller saved regs */
1415 for (i = 0; i < CALLER_SAVED_REGS; i++) {
1416 reg = regs + caller_saved[i];
1417 reg->type = NOT_INIT;
1421 /* update return register */
1422 if (fn->ret_type == RET_INTEGER) {
1423 regs[BPF_REG_0].type = UNKNOWN_VALUE;
1424 } else if (fn->ret_type == RET_VOID) {
1425 regs[BPF_REG_0].type = NOT_INIT;
1426 } else if (fn->ret_type == RET_PTR_TO_MAP_VALUE_OR_NULL) {
1427 struct bpf_insn_aux_data *insn_aux;
1429 regs[BPF_REG_0].type = PTR_TO_MAP_VALUE_OR_NULL;
1430 regs[BPF_REG_0].max_value = regs[BPF_REG_0].min_value = 0;
1431 /* remember map_ptr, so that check_map_access()
1432 * can check 'value_size' boundary of memory access
1433 * to map element returned from bpf_map_lookup_elem()
1435 if (meta.map_ptr == NULL) {
1436 verbose("kernel subsystem misconfigured verifier\n");
1439 regs[BPF_REG_0].map_ptr = meta.map_ptr;
1440 regs[BPF_REG_0].id = ++env->id_gen;
1441 insn_aux = &env->insn_aux_data[insn_idx];
1442 if (!insn_aux->map_ptr)
1443 insn_aux->map_ptr = meta.map_ptr;
1444 else if (insn_aux->map_ptr != meta.map_ptr)
1445 insn_aux->map_ptr = BPF_MAP_PTR_POISON;
1447 verbose("unknown return type %d of func %s#%d\n",
1448 fn->ret_type, func_id_name(func_id), func_id);
1452 err = check_map_func_compatibility(meta.map_ptr, func_id);
1457 clear_all_pkt_pointers(env);
1461 static int check_packet_ptr_add(struct bpf_verifier_env *env,
1462 struct bpf_insn *insn)
1464 struct bpf_reg_state *regs = env->cur_state.regs;
1465 struct bpf_reg_state *dst_reg = ®s[insn->dst_reg];
1466 struct bpf_reg_state *src_reg = ®s[insn->src_reg];
1467 struct bpf_reg_state tmp_reg;
1470 if (BPF_SRC(insn->code) == BPF_K) {
1471 /* pkt_ptr += imm */
1476 verbose("addition of negative constant to packet pointer is not allowed\n");
1479 if (imm >= MAX_PACKET_OFF ||
1480 imm + dst_reg->off >= MAX_PACKET_OFF) {
1481 verbose("constant %d is too large to add to packet pointer\n",
1485 /* a constant was added to pkt_ptr.
1486 * Remember it while keeping the same 'id'
1488 dst_reg->off += imm;
1492 if (src_reg->type == PTR_TO_PACKET) {
1493 /* R6=pkt(id=0,off=0,r=62) R7=imm22; r7 += r6 */
1494 tmp_reg = *dst_reg; /* save r7 state */
1495 *dst_reg = *src_reg; /* copy pkt_ptr state r6 into r7 */
1496 src_reg = &tmp_reg; /* pretend it's src_reg state */
1497 /* if the checks below reject it, the copy won't matter,
1498 * since we're rejecting the whole program. If all ok,
1499 * then imm22 state will be added to r7
1500 * and r7 will be pkt(id=0,off=22,r=62) while
1501 * r6 will stay as pkt(id=0,off=0,r=62)
1505 if (src_reg->type == CONST_IMM) {
1506 /* pkt_ptr += reg where reg is known constant */
1510 /* disallow pkt_ptr += reg
1511 * if reg is not uknown_value with guaranteed zero upper bits
1512 * otherwise pkt_ptr may overflow and addition will become
1513 * subtraction which is not allowed
1515 if (src_reg->type != UNKNOWN_VALUE) {
1516 verbose("cannot add '%s' to ptr_to_packet\n",
1517 reg_type_str[src_reg->type]);
1520 if (src_reg->imm < 48) {
1521 verbose("cannot add integer value with %lld upper zero bits to ptr_to_packet\n",
1526 had_id = (dst_reg->id != 0);
1528 /* dst_reg stays as pkt_ptr type and since some positive
1529 * integer value was added to the pointer, increment its 'id'
1531 dst_reg->id = ++env->id_gen;
1533 /* something was added to pkt_ptr, set range to zero */
1534 dst_reg->aux_off += dst_reg->off;
1538 dst_reg->aux_off_align = min(dst_reg->aux_off_align,
1539 src_reg->min_align);
1541 dst_reg->aux_off_align = src_reg->min_align;
1546 static int evaluate_reg_alu(struct bpf_verifier_env *env, struct bpf_insn *insn)
1548 struct bpf_reg_state *regs = env->cur_state.regs;
1549 struct bpf_reg_state *dst_reg = ®s[insn->dst_reg];
1550 u8 opcode = BPF_OP(insn->code);
1553 /* for type == UNKNOWN_VALUE:
1554 * imm > 0 -> number of zero upper bits
1555 * imm == 0 -> don't track which is the same as all bits can be non-zero
1558 if (BPF_SRC(insn->code) == BPF_X) {
1559 struct bpf_reg_state *src_reg = ®s[insn->src_reg];
1561 if (src_reg->type == UNKNOWN_VALUE && src_reg->imm > 0 &&
1562 dst_reg->imm && opcode == BPF_ADD) {
1564 * where both have zero upper bits. Adding them
1565 * can only result making one more bit non-zero
1566 * in the larger value.
1567 * Ex. 0xffff (imm=48) + 1 (imm=63) = 0x10000 (imm=47)
1568 * 0xffff (imm=48) + 0xffff = 0x1fffe (imm=47)
1570 dst_reg->imm = min(dst_reg->imm, src_reg->imm);
1574 if (src_reg->type == CONST_IMM && src_reg->imm > 0 &&
1575 dst_reg->imm && opcode == BPF_ADD) {
1577 * where dreg has zero upper bits and sreg is const.
1578 * Adding them can only result making one more bit
1579 * non-zero in the larger value.
1581 imm_log2 = __ilog2_u64((long long)src_reg->imm);
1582 dst_reg->imm = min(dst_reg->imm, 63 - imm_log2);
1586 /* all other cases non supported yet, just mark dst_reg */
1591 /* sign extend 32-bit imm into 64-bit to make sure that
1592 * negative values occupy bit 63. Note ilog2() would have
1593 * been incorrect, since sizeof(insn->imm) == 4
1595 imm_log2 = __ilog2_u64((long long)insn->imm);
1597 if (dst_reg->imm && opcode == BPF_LSH) {
1599 * if reg was a result of 2 byte load, then its imm == 48
1600 * which means that upper 48 bits are zero and shifting this reg
1601 * left by 4 would mean that upper 44 bits are still zero
1603 dst_reg->imm -= insn->imm;
1604 } else if (dst_reg->imm && opcode == BPF_MUL) {
1606 * if multiplying by 14 subtract 4
1607 * This is conservative calculation of upper zero bits.
1608 * It's not trying to special case insn->imm == 1 or 0 cases
1610 dst_reg->imm -= imm_log2 + 1;
1611 } else if (opcode == BPF_AND) {
1613 dst_reg->imm = 63 - imm_log2;
1614 } else if (dst_reg->imm && opcode == BPF_ADD) {
1616 dst_reg->imm = min(dst_reg->imm, 63 - imm_log2);
1618 } else if (opcode == BPF_RSH) {
1620 * which means that after right shift, upper bits will be zero
1621 * note that verifier already checked that
1622 * 0 <= imm < 64 for shift insn
1624 dst_reg->imm += insn->imm;
1625 if (unlikely(dst_reg->imm > 64))
1626 /* some dumb code did:
1629 * and all bits are zero now */
1632 /* all other alu ops, means that we don't know what will
1633 * happen to the value, mark it with unknown number of zero bits
1638 if (dst_reg->imm < 0) {
1639 /* all 64 bits of the register can contain non-zero bits
1640 * and such value cannot be added to ptr_to_packet, since it
1641 * may overflow, mark it as unknown to avoid further eval
1648 static int evaluate_reg_imm_alu(struct bpf_verifier_env *env,
1649 struct bpf_insn *insn)
1651 struct bpf_reg_state *regs = env->cur_state.regs;
1652 struct bpf_reg_state *dst_reg = ®s[insn->dst_reg];
1653 struct bpf_reg_state *src_reg = ®s[insn->src_reg];
1654 u8 opcode = BPF_OP(insn->code);
1655 u64 dst_imm = dst_reg->imm;
1657 /* dst_reg->type == CONST_IMM here. Simulate execution of insns
1658 * containing ALU ops. Don't care about overflow or negative
1659 * values, just add/sub/... them; registers are in u64.
1661 if (opcode == BPF_ADD && BPF_SRC(insn->code) == BPF_K) {
1662 dst_imm += insn->imm;
1663 } else if (opcode == BPF_ADD && BPF_SRC(insn->code) == BPF_X &&
1664 src_reg->type == CONST_IMM) {
1665 dst_imm += src_reg->imm;
1666 } else if (opcode == BPF_SUB && BPF_SRC(insn->code) == BPF_K) {
1667 dst_imm -= insn->imm;
1668 } else if (opcode == BPF_SUB && BPF_SRC(insn->code) == BPF_X &&
1669 src_reg->type == CONST_IMM) {
1670 dst_imm -= src_reg->imm;
1671 } else if (opcode == BPF_MUL && BPF_SRC(insn->code) == BPF_K) {
1672 dst_imm *= insn->imm;
1673 } else if (opcode == BPF_MUL && BPF_SRC(insn->code) == BPF_X &&
1674 src_reg->type == CONST_IMM) {
1675 dst_imm *= src_reg->imm;
1676 } else if (opcode == BPF_OR && BPF_SRC(insn->code) == BPF_K) {
1677 dst_imm |= insn->imm;
1678 } else if (opcode == BPF_OR && BPF_SRC(insn->code) == BPF_X &&
1679 src_reg->type == CONST_IMM) {
1680 dst_imm |= src_reg->imm;
1681 } else if (opcode == BPF_AND && BPF_SRC(insn->code) == BPF_K) {
1682 dst_imm &= insn->imm;
1683 } else if (opcode == BPF_AND && BPF_SRC(insn->code) == BPF_X &&
1684 src_reg->type == CONST_IMM) {
1685 dst_imm &= src_reg->imm;
1686 } else if (opcode == BPF_RSH && BPF_SRC(insn->code) == BPF_K) {
1687 dst_imm >>= insn->imm;
1688 } else if (opcode == BPF_RSH && BPF_SRC(insn->code) == BPF_X &&
1689 src_reg->type == CONST_IMM) {
1690 dst_imm >>= src_reg->imm;
1691 } else if (opcode == BPF_LSH && BPF_SRC(insn->code) == BPF_K) {
1692 dst_imm <<= insn->imm;
1693 } else if (opcode == BPF_LSH && BPF_SRC(insn->code) == BPF_X &&
1694 src_reg->type == CONST_IMM) {
1695 dst_imm <<= src_reg->imm;
1697 mark_reg_unknown_value(regs, insn->dst_reg);
1701 dst_reg->imm = dst_imm;
1706 static void check_reg_overflow(struct bpf_reg_state *reg)
1708 if (reg->max_value > BPF_REGISTER_MAX_RANGE)
1709 reg->max_value = BPF_REGISTER_MAX_RANGE;
1710 if (reg->min_value < BPF_REGISTER_MIN_RANGE ||
1711 reg->min_value > BPF_REGISTER_MAX_RANGE)
1712 reg->min_value = BPF_REGISTER_MIN_RANGE;
1715 static u32 calc_align(u32 imm)
1719 return imm - ((imm - 1) & imm);
1722 static void adjust_reg_min_max_vals(struct bpf_verifier_env *env,
1723 struct bpf_insn *insn)
1725 struct bpf_reg_state *regs = env->cur_state.regs, *dst_reg;
1726 s64 min_val = BPF_REGISTER_MIN_RANGE;
1727 u64 max_val = BPF_REGISTER_MAX_RANGE;
1728 u8 opcode = BPF_OP(insn->code);
1729 u32 dst_align, src_align;
1731 dst_reg = ®s[insn->dst_reg];
1733 if (BPF_SRC(insn->code) == BPF_X) {
1734 check_reg_overflow(®s[insn->src_reg]);
1735 min_val = regs[insn->src_reg].min_value;
1736 max_val = regs[insn->src_reg].max_value;
1738 /* If the source register is a random pointer then the
1739 * min_value/max_value values represent the range of the known
1740 * accesses into that value, not the actual min/max value of the
1741 * register itself. In this case we have to reset the reg range
1742 * values so we know it is not safe to look at.
1744 if (regs[insn->src_reg].type != CONST_IMM &&
1745 regs[insn->src_reg].type != UNKNOWN_VALUE) {
1746 min_val = BPF_REGISTER_MIN_RANGE;
1747 max_val = BPF_REGISTER_MAX_RANGE;
1750 src_align = regs[insn->src_reg].min_align;
1752 } else if (insn->imm < BPF_REGISTER_MAX_RANGE &&
1753 (s64)insn->imm > BPF_REGISTER_MIN_RANGE) {
1754 min_val = max_val = insn->imm;
1755 src_align = calc_align(insn->imm);
1758 dst_align = dst_reg->min_align;
1760 /* We don't know anything about what was done to this register, mark it
1763 if (min_val == BPF_REGISTER_MIN_RANGE &&
1764 max_val == BPF_REGISTER_MAX_RANGE) {
1765 reset_reg_range_values(regs, insn->dst_reg);
1769 /* If one of our values was at the end of our ranges then we can't just
1770 * do our normal operations to the register, we need to set the values
1771 * to the min/max since they are undefined.
1773 if (min_val == BPF_REGISTER_MIN_RANGE)
1774 dst_reg->min_value = BPF_REGISTER_MIN_RANGE;
1775 if (max_val == BPF_REGISTER_MAX_RANGE)
1776 dst_reg->max_value = BPF_REGISTER_MAX_RANGE;
1780 if (dst_reg->min_value != BPF_REGISTER_MIN_RANGE)
1781 dst_reg->min_value += min_val;
1782 if (dst_reg->max_value != BPF_REGISTER_MAX_RANGE)
1783 dst_reg->max_value += max_val;
1784 dst_reg->min_align = min(src_align, dst_align);
1787 if (dst_reg->min_value != BPF_REGISTER_MIN_RANGE)
1788 dst_reg->min_value -= min_val;
1789 if (dst_reg->max_value != BPF_REGISTER_MAX_RANGE)
1790 dst_reg->max_value -= max_val;
1791 dst_reg->min_align = min(src_align, dst_align);
1794 if (dst_reg->min_value != BPF_REGISTER_MIN_RANGE)
1795 dst_reg->min_value *= min_val;
1796 if (dst_reg->max_value != BPF_REGISTER_MAX_RANGE)
1797 dst_reg->max_value *= max_val;
1798 dst_reg->min_align = max(src_align, dst_align);
1801 /* Disallow AND'ing of negative numbers, ain't nobody got time
1802 * for that. Otherwise the minimum is 0 and the max is the max
1803 * value we could AND against.
1806 dst_reg->min_value = BPF_REGISTER_MIN_RANGE;
1808 dst_reg->min_value = 0;
1809 dst_reg->max_value = max_val;
1810 dst_reg->min_align = max(src_align, dst_align);
1813 /* Gotta have special overflow logic here, if we're shifting
1814 * more than MAX_RANGE then just assume we have an invalid
1817 if (min_val > ilog2(BPF_REGISTER_MAX_RANGE)) {
1818 dst_reg->min_value = BPF_REGISTER_MIN_RANGE;
1819 dst_reg->min_align = 1;
1821 if (dst_reg->min_value != BPF_REGISTER_MIN_RANGE)
1822 dst_reg->min_value <<= min_val;
1823 if (!dst_reg->min_align)
1824 dst_reg->min_align = 1;
1825 dst_reg->min_align <<= min_val;
1827 if (max_val > ilog2(BPF_REGISTER_MAX_RANGE))
1828 dst_reg->max_value = BPF_REGISTER_MAX_RANGE;
1829 else if (dst_reg->max_value != BPF_REGISTER_MAX_RANGE)
1830 dst_reg->max_value <<= max_val;
1833 /* RSH by a negative number is undefined, and the BPF_RSH is an
1834 * unsigned shift, so make the appropriate casts.
1836 if (min_val < 0 || dst_reg->min_value < 0) {
1837 dst_reg->min_value = BPF_REGISTER_MIN_RANGE;
1839 dst_reg->min_value =
1840 (u64)(dst_reg->min_value) >> min_val;
1843 dst_reg->min_align = 1;
1845 dst_reg->min_align >>= (u64) min_val;
1846 if (!dst_reg->min_align)
1847 dst_reg->min_align = 1;
1849 if (dst_reg->max_value != BPF_REGISTER_MAX_RANGE)
1850 dst_reg->max_value >>= max_val;
1853 reset_reg_range_values(regs, insn->dst_reg);
1857 check_reg_overflow(dst_reg);
1860 /* check validity of 32-bit and 64-bit arithmetic operations */
1861 static int check_alu_op(struct bpf_verifier_env *env, struct bpf_insn *insn)
1863 struct bpf_reg_state *regs = env->cur_state.regs, *dst_reg;
1864 u8 opcode = BPF_OP(insn->code);
1867 if (opcode == BPF_END || opcode == BPF_NEG) {
1868 if (opcode == BPF_NEG) {
1869 if (BPF_SRC(insn->code) != 0 ||
1870 insn->src_reg != BPF_REG_0 ||
1871 insn->off != 0 || insn->imm != 0) {
1872 verbose("BPF_NEG uses reserved fields\n");
1876 if (insn->src_reg != BPF_REG_0 || insn->off != 0 ||
1877 (insn->imm != 16 && insn->imm != 32 && insn->imm != 64)) {
1878 verbose("BPF_END uses reserved fields\n");
1883 /* check src operand */
1884 err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
1888 if (is_pointer_value(env, insn->dst_reg)) {
1889 verbose("R%d pointer arithmetic prohibited\n",
1894 /* check dest operand */
1895 err = check_reg_arg(regs, insn->dst_reg, DST_OP);
1899 } else if (opcode == BPF_MOV) {
1901 if (BPF_SRC(insn->code) == BPF_X) {
1902 if (insn->imm != 0 || insn->off != 0) {
1903 verbose("BPF_MOV uses reserved fields\n");
1907 /* check src operand */
1908 err = check_reg_arg(regs, insn->src_reg, SRC_OP);
1912 if (insn->src_reg != BPF_REG_0 || insn->off != 0) {
1913 verbose("BPF_MOV uses reserved fields\n");
1918 /* check dest operand */
1919 err = check_reg_arg(regs, insn->dst_reg, DST_OP);
1923 /* we are setting our register to something new, we need to
1924 * reset its range values.
1926 reset_reg_range_values(regs, insn->dst_reg);
1928 if (BPF_SRC(insn->code) == BPF_X) {
1929 if (BPF_CLASS(insn->code) == BPF_ALU64) {
1931 * copy register state to dest reg
1933 regs[insn->dst_reg] = regs[insn->src_reg];
1935 if (is_pointer_value(env, insn->src_reg)) {
1936 verbose("R%d partial copy of pointer\n",
1940 mark_reg_unknown_value(regs, insn->dst_reg);
1944 * remember the value we stored into this reg
1946 regs[insn->dst_reg].type = CONST_IMM;
1947 regs[insn->dst_reg].imm = insn->imm;
1948 regs[insn->dst_reg].max_value = insn->imm;
1949 regs[insn->dst_reg].min_value = insn->imm;
1950 regs[insn->dst_reg].min_align = calc_align(insn->imm);
1953 } else if (opcode > BPF_END) {
1954 verbose("invalid BPF_ALU opcode %x\n", opcode);
1957 } else { /* all other ALU ops: and, sub, xor, add, ... */
1959 if (BPF_SRC(insn->code) == BPF_X) {
1960 if (insn->imm != 0 || insn->off != 0) {
1961 verbose("BPF_ALU uses reserved fields\n");
1964 /* check src1 operand */
1965 err = check_reg_arg(regs, insn->src_reg, SRC_OP);
1969 if (insn->src_reg != BPF_REG_0 || insn->off != 0) {
1970 verbose("BPF_ALU uses reserved fields\n");
1975 /* check src2 operand */
1976 err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
1980 if ((opcode == BPF_MOD || opcode == BPF_DIV) &&
1981 BPF_SRC(insn->code) == BPF_K && insn->imm == 0) {
1982 verbose("div by zero\n");
1986 if ((opcode == BPF_LSH || opcode == BPF_RSH ||
1987 opcode == BPF_ARSH) && BPF_SRC(insn->code) == BPF_K) {
1988 int size = BPF_CLASS(insn->code) == BPF_ALU64 ? 64 : 32;
1990 if (insn->imm < 0 || insn->imm >= size) {
1991 verbose("invalid shift %d\n", insn->imm);
1996 /* check dest operand */
1997 err = check_reg_arg(regs, insn->dst_reg, DST_OP_NO_MARK);
2001 dst_reg = ®s[insn->dst_reg];
2003 /* first we want to adjust our ranges. */
2004 adjust_reg_min_max_vals(env, insn);
2006 /* pattern match 'bpf_add Rx, imm' instruction */
2007 if (opcode == BPF_ADD && BPF_CLASS(insn->code) == BPF_ALU64 &&
2008 dst_reg->type == FRAME_PTR && BPF_SRC(insn->code) == BPF_K) {
2009 dst_reg->type = PTR_TO_STACK;
2010 dst_reg->imm = insn->imm;
2012 } else if (opcode == BPF_ADD &&
2013 BPF_CLASS(insn->code) == BPF_ALU64 &&
2014 dst_reg->type == PTR_TO_STACK &&
2015 ((BPF_SRC(insn->code) == BPF_X &&
2016 regs[insn->src_reg].type == CONST_IMM) ||
2017 BPF_SRC(insn->code) == BPF_K)) {
2018 if (BPF_SRC(insn->code) == BPF_X)
2019 dst_reg->imm += regs[insn->src_reg].imm;
2021 dst_reg->imm += insn->imm;
2023 } else if (opcode == BPF_ADD &&
2024 BPF_CLASS(insn->code) == BPF_ALU64 &&
2025 (dst_reg->type == PTR_TO_PACKET ||
2026 (BPF_SRC(insn->code) == BPF_X &&
2027 regs[insn->src_reg].type == PTR_TO_PACKET))) {
2028 /* ptr_to_packet += K|X */
2029 return check_packet_ptr_add(env, insn);
2030 } else if (BPF_CLASS(insn->code) == BPF_ALU64 &&
2031 dst_reg->type == UNKNOWN_VALUE &&
2032 env->allow_ptr_leaks) {
2033 /* unknown += K|X */
2034 return evaluate_reg_alu(env, insn);
2035 } else if (BPF_CLASS(insn->code) == BPF_ALU64 &&
2036 dst_reg->type == CONST_IMM &&
2037 env->allow_ptr_leaks) {
2038 /* reg_imm += K|X */
2039 return evaluate_reg_imm_alu(env, insn);
2040 } else if (is_pointer_value(env, insn->dst_reg)) {
2041 verbose("R%d pointer arithmetic prohibited\n",
2044 } else if (BPF_SRC(insn->code) == BPF_X &&
2045 is_pointer_value(env, insn->src_reg)) {
2046 verbose("R%d pointer arithmetic prohibited\n",
2051 /* If we did pointer math on a map value then just set it to our
2052 * PTR_TO_MAP_VALUE_ADJ type so we can deal with any stores or
2053 * loads to this register appropriately, otherwise just mark the
2054 * register as unknown.
2056 if (env->allow_ptr_leaks &&
2057 BPF_CLASS(insn->code) == BPF_ALU64 && opcode == BPF_ADD &&
2058 (dst_reg->type == PTR_TO_MAP_VALUE ||
2059 dst_reg->type == PTR_TO_MAP_VALUE_ADJ))
2060 dst_reg->type = PTR_TO_MAP_VALUE_ADJ;
2062 mark_reg_unknown_value(regs, insn->dst_reg);
2068 static void find_good_pkt_pointers(struct bpf_verifier_state *state,
2069 struct bpf_reg_state *dst_reg)
2071 struct bpf_reg_state *regs = state->regs, *reg;
2074 /* LLVM can generate two kind of checks:
2080 * if (r2 > pkt_end) goto <handle exception>
2084 * r2 == dst_reg, pkt_end == src_reg
2085 * r2=pkt(id=n,off=8,r=0)
2086 * r3=pkt(id=n,off=0,r=0)
2092 * if (pkt_end >= r2) goto <access okay>
2093 * <handle exception>
2096 * pkt_end == dst_reg, r2 == src_reg
2097 * r2=pkt(id=n,off=8,r=0)
2098 * r3=pkt(id=n,off=0,r=0)
2100 * Find register r3 and mark its range as r3=pkt(id=n,off=0,r=8)
2101 * so that range of bytes [r3, r3 + 8) is safe to access.
2104 for (i = 0; i < MAX_BPF_REG; i++)
2105 if (regs[i].type == PTR_TO_PACKET && regs[i].id == dst_reg->id)
2106 /* keep the maximum range already checked */
2107 regs[i].range = max(regs[i].range, dst_reg->off);
2109 for (i = 0; i < MAX_BPF_STACK; i += BPF_REG_SIZE) {
2110 if (state->stack_slot_type[i] != STACK_SPILL)
2112 reg = &state->spilled_regs[i / BPF_REG_SIZE];
2113 if (reg->type == PTR_TO_PACKET && reg->id == dst_reg->id)
2114 reg->range = max(reg->range, dst_reg->off);
2118 /* Adjusts the register min/max values in the case that the dst_reg is the
2119 * variable register that we are working on, and src_reg is a constant or we're
2120 * simply doing a BPF_K check.
2122 static void reg_set_min_max(struct bpf_reg_state *true_reg,
2123 struct bpf_reg_state *false_reg, u64 val,
2128 /* If this is false then we know nothing Jon Snow, but if it is
2129 * true then we know for sure.
2131 true_reg->max_value = true_reg->min_value = val;
2134 /* If this is true we know nothing Jon Snow, but if it is false
2135 * we know the value for sure;
2137 false_reg->max_value = false_reg->min_value = val;
2140 /* Unsigned comparison, the minimum value is 0. */
2141 false_reg->min_value = 0;
2144 /* If this is false then we know the maximum val is val,
2145 * otherwise we know the min val is val+1.
2147 false_reg->max_value = val;
2148 true_reg->min_value = val + 1;
2151 /* Unsigned comparison, the minimum value is 0. */
2152 false_reg->min_value = 0;
2155 /* If this is false then we know the maximum value is val - 1,
2156 * otherwise we know the mimimum value is val.
2158 false_reg->max_value = val - 1;
2159 true_reg->min_value = val;
2165 check_reg_overflow(false_reg);
2166 check_reg_overflow(true_reg);
2169 /* Same as above, but for the case that dst_reg is a CONST_IMM reg and src_reg
2170 * is the variable reg.
2172 static void reg_set_min_max_inv(struct bpf_reg_state *true_reg,
2173 struct bpf_reg_state *false_reg, u64 val,
2178 /* If this is false then we know nothing Jon Snow, but if it is
2179 * true then we know for sure.
2181 true_reg->max_value = true_reg->min_value = val;
2184 /* If this is true we know nothing Jon Snow, but if it is false
2185 * we know the value for sure;
2187 false_reg->max_value = false_reg->min_value = val;
2190 /* Unsigned comparison, the minimum value is 0. */
2191 true_reg->min_value = 0;
2195 * If this is false, then the val is <= the register, if it is
2196 * true the register <= to the val.
2198 false_reg->min_value = val;
2199 true_reg->max_value = val - 1;
2202 /* Unsigned comparison, the minimum value is 0. */
2203 true_reg->min_value = 0;
2206 /* If this is false then constant < register, if it is true then
2207 * the register < constant.
2209 false_reg->min_value = val + 1;
2210 true_reg->max_value = val;
2216 check_reg_overflow(false_reg);
2217 check_reg_overflow(true_reg);
2220 static void mark_map_reg(struct bpf_reg_state *regs, u32 regno, u32 id,
2221 enum bpf_reg_type type)
2223 struct bpf_reg_state *reg = ®s[regno];
2225 if (reg->type == PTR_TO_MAP_VALUE_OR_NULL && reg->id == id) {
2226 if (type == UNKNOWN_VALUE) {
2227 __mark_reg_unknown_value(regs, regno);
2228 } else if (reg->map_ptr->inner_map_meta) {
2229 reg->type = CONST_PTR_TO_MAP;
2230 reg->map_ptr = reg->map_ptr->inner_map_meta;
2234 /* We don't need id from this point onwards anymore, thus we
2235 * should better reset it, so that state pruning has chances
2242 /* The logic is similar to find_good_pkt_pointers(), both could eventually
2243 * be folded together at some point.
2245 static void mark_map_regs(struct bpf_verifier_state *state, u32 regno,
2246 enum bpf_reg_type type)
2248 struct bpf_reg_state *regs = state->regs;
2249 u32 id = regs[regno].id;
2252 for (i = 0; i < MAX_BPF_REG; i++)
2253 mark_map_reg(regs, i, id, type);
2255 for (i = 0; i < MAX_BPF_STACK; i += BPF_REG_SIZE) {
2256 if (state->stack_slot_type[i] != STACK_SPILL)
2258 mark_map_reg(state->spilled_regs, i / BPF_REG_SIZE, id, type);
2262 static int check_cond_jmp_op(struct bpf_verifier_env *env,
2263 struct bpf_insn *insn, int *insn_idx)
2265 struct bpf_verifier_state *other_branch, *this_branch = &env->cur_state;
2266 struct bpf_reg_state *regs = this_branch->regs, *dst_reg;
2267 u8 opcode = BPF_OP(insn->code);
2270 if (opcode > BPF_EXIT) {
2271 verbose("invalid BPF_JMP opcode %x\n", opcode);
2275 if (BPF_SRC(insn->code) == BPF_X) {
2276 if (insn->imm != 0) {
2277 verbose("BPF_JMP uses reserved fields\n");
2281 /* check src1 operand */
2282 err = check_reg_arg(regs, insn->src_reg, SRC_OP);
2286 if (is_pointer_value(env, insn->src_reg)) {
2287 verbose("R%d pointer comparison prohibited\n",
2292 if (insn->src_reg != BPF_REG_0) {
2293 verbose("BPF_JMP uses reserved fields\n");
2298 /* check src2 operand */
2299 err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
2303 dst_reg = ®s[insn->dst_reg];
2305 /* detect if R == 0 where R was initialized to zero earlier */
2306 if (BPF_SRC(insn->code) == BPF_K &&
2307 (opcode == BPF_JEQ || opcode == BPF_JNE) &&
2308 dst_reg->type == CONST_IMM && dst_reg->imm == insn->imm) {
2309 if (opcode == BPF_JEQ) {
2310 /* if (imm == imm) goto pc+off;
2311 * only follow the goto, ignore fall-through
2313 *insn_idx += insn->off;
2316 /* if (imm != imm) goto pc+off;
2317 * only follow fall-through branch, since
2318 * that's where the program will go
2324 other_branch = push_stack(env, *insn_idx + insn->off + 1, *insn_idx);
2328 /* detect if we are comparing against a constant value so we can adjust
2329 * our min/max values for our dst register.
2331 if (BPF_SRC(insn->code) == BPF_X) {
2332 if (regs[insn->src_reg].type == CONST_IMM)
2333 reg_set_min_max(&other_branch->regs[insn->dst_reg],
2334 dst_reg, regs[insn->src_reg].imm,
2336 else if (dst_reg->type == CONST_IMM)
2337 reg_set_min_max_inv(&other_branch->regs[insn->src_reg],
2338 ®s[insn->src_reg], dst_reg->imm,
2341 reg_set_min_max(&other_branch->regs[insn->dst_reg],
2342 dst_reg, insn->imm, opcode);
2345 /* detect if R == 0 where R is returned from bpf_map_lookup_elem() */
2346 if (BPF_SRC(insn->code) == BPF_K &&
2347 insn->imm == 0 && (opcode == BPF_JEQ || opcode == BPF_JNE) &&
2348 dst_reg->type == PTR_TO_MAP_VALUE_OR_NULL) {
2349 /* Mark all identical map registers in each branch as either
2350 * safe or unknown depending R == 0 or R != 0 conditional.
2352 mark_map_regs(this_branch, insn->dst_reg,
2353 opcode == BPF_JEQ ? PTR_TO_MAP_VALUE : UNKNOWN_VALUE);
2354 mark_map_regs(other_branch, insn->dst_reg,
2355 opcode == BPF_JEQ ? UNKNOWN_VALUE : PTR_TO_MAP_VALUE);
2356 } else if (BPF_SRC(insn->code) == BPF_X && opcode == BPF_JGT &&
2357 dst_reg->type == PTR_TO_PACKET &&
2358 regs[insn->src_reg].type == PTR_TO_PACKET_END) {
2359 find_good_pkt_pointers(this_branch, dst_reg);
2360 } else if (BPF_SRC(insn->code) == BPF_X && opcode == BPF_JGE &&
2361 dst_reg->type == PTR_TO_PACKET_END &&
2362 regs[insn->src_reg].type == PTR_TO_PACKET) {
2363 find_good_pkt_pointers(other_branch, ®s[insn->src_reg]);
2364 } else if (is_pointer_value(env, insn->dst_reg)) {
2365 verbose("R%d pointer comparison prohibited\n", insn->dst_reg);
2369 print_verifier_state(this_branch);
2373 /* return the map pointer stored inside BPF_LD_IMM64 instruction */
2374 static struct bpf_map *ld_imm64_to_map_ptr(struct bpf_insn *insn)
2376 u64 imm64 = ((u64) (u32) insn[0].imm) | ((u64) (u32) insn[1].imm) << 32;
2378 return (struct bpf_map *) (unsigned long) imm64;
2381 /* verify BPF_LD_IMM64 instruction */
2382 static int check_ld_imm(struct bpf_verifier_env *env, struct bpf_insn *insn)
2384 struct bpf_reg_state *regs = env->cur_state.regs;
2387 if (BPF_SIZE(insn->code) != BPF_DW) {
2388 verbose("invalid BPF_LD_IMM insn\n");
2391 if (insn->off != 0) {
2392 verbose("BPF_LD_IMM64 uses reserved fields\n");
2396 err = check_reg_arg(regs, insn->dst_reg, DST_OP);
2400 if (insn->src_reg == 0) {
2401 u64 imm = ((u64)(insn + 1)->imm << 32) | (u32)insn->imm;
2403 regs[insn->dst_reg].type = CONST_IMM;
2404 regs[insn->dst_reg].imm = imm;
2408 /* replace_map_fd_with_map_ptr() should have caught bad ld_imm64 */
2409 BUG_ON(insn->src_reg != BPF_PSEUDO_MAP_FD);
2411 regs[insn->dst_reg].type = CONST_PTR_TO_MAP;
2412 regs[insn->dst_reg].map_ptr = ld_imm64_to_map_ptr(insn);
2416 static bool may_access_skb(enum bpf_prog_type type)
2419 case BPF_PROG_TYPE_SOCKET_FILTER:
2420 case BPF_PROG_TYPE_SCHED_CLS:
2421 case BPF_PROG_TYPE_SCHED_ACT:
2428 /* verify safety of LD_ABS|LD_IND instructions:
2429 * - they can only appear in the programs where ctx == skb
2430 * - since they are wrappers of function calls, they scratch R1-R5 registers,
2431 * preserve R6-R9, and store return value into R0
2434 * ctx == skb == R6 == CTX
2437 * SRC == any register
2438 * IMM == 32-bit immediate
2441 * R0 - 8/16/32-bit skb data converted to cpu endianness
2443 static int check_ld_abs(struct bpf_verifier_env *env, struct bpf_insn *insn)
2445 struct bpf_reg_state *regs = env->cur_state.regs;
2446 u8 mode = BPF_MODE(insn->code);
2447 struct bpf_reg_state *reg;
2450 if (!may_access_skb(env->prog->type)) {
2451 verbose("BPF_LD_[ABS|IND] instructions not allowed for this program type\n");
2455 if (insn->dst_reg != BPF_REG_0 || insn->off != 0 ||
2456 BPF_SIZE(insn->code) == BPF_DW ||
2457 (mode == BPF_ABS && insn->src_reg != BPF_REG_0)) {
2458 verbose("BPF_LD_[ABS|IND] uses reserved fields\n");
2462 /* check whether implicit source operand (register R6) is readable */
2463 err = check_reg_arg(regs, BPF_REG_6, SRC_OP);
2467 if (regs[BPF_REG_6].type != PTR_TO_CTX) {
2468 verbose("at the time of BPF_LD_ABS|IND R6 != pointer to skb\n");
2472 if (mode == BPF_IND) {
2473 /* check explicit source operand */
2474 err = check_reg_arg(regs, insn->src_reg, SRC_OP);
2479 /* reset caller saved regs to unreadable */
2480 for (i = 0; i < CALLER_SAVED_REGS; i++) {
2481 reg = regs + caller_saved[i];
2482 reg->type = NOT_INIT;
2486 /* mark destination R0 register as readable, since it contains
2487 * the value fetched from the packet
2489 regs[BPF_REG_0].type = UNKNOWN_VALUE;
2493 /* non-recursive DFS pseudo code
2494 * 1 procedure DFS-iterative(G,v):
2495 * 2 label v as discovered
2496 * 3 let S be a stack
2498 * 5 while S is not empty
2500 * 7 if t is what we're looking for:
2502 * 9 for all edges e in G.adjacentEdges(t) do
2503 * 10 if edge e is already labelled
2504 * 11 continue with the next edge
2505 * 12 w <- G.adjacentVertex(t,e)
2506 * 13 if vertex w is not discovered and not explored
2507 * 14 label e as tree-edge
2508 * 15 label w as discovered
2511 * 18 else if vertex w is discovered
2512 * 19 label e as back-edge
2514 * 21 // vertex w is explored
2515 * 22 label e as forward- or cross-edge
2516 * 23 label t as explored
2521 * 0x11 - discovered and fall-through edge labelled
2522 * 0x12 - discovered and fall-through and branch edges labelled
2533 #define STATE_LIST_MARK ((struct bpf_verifier_state_list *) -1L)
2535 static int *insn_stack; /* stack of insns to process */
2536 static int cur_stack; /* current stack index */
2537 static int *insn_state;
2539 /* t, w, e - match pseudo-code above:
2540 * t - index of current instruction
2541 * w - next instruction
2544 static int push_insn(int t, int w, int e, struct bpf_verifier_env *env)
2546 if (e == FALLTHROUGH && insn_state[t] >= (DISCOVERED | FALLTHROUGH))
2549 if (e == BRANCH && insn_state[t] >= (DISCOVERED | BRANCH))
2552 if (w < 0 || w >= env->prog->len) {
2553 verbose("jump out of range from insn %d to %d\n", t, w);
2558 /* mark branch target for state pruning */
2559 env->explored_states[w] = STATE_LIST_MARK;
2561 if (insn_state[w] == 0) {
2563 insn_state[t] = DISCOVERED | e;
2564 insn_state[w] = DISCOVERED;
2565 if (cur_stack >= env->prog->len)
2567 insn_stack[cur_stack++] = w;
2569 } else if ((insn_state[w] & 0xF0) == DISCOVERED) {
2570 verbose("back-edge from insn %d to %d\n", t, w);
2572 } else if (insn_state[w] == EXPLORED) {
2573 /* forward- or cross-edge */
2574 insn_state[t] = DISCOVERED | e;
2576 verbose("insn state internal bug\n");
2582 /* non-recursive depth-first-search to detect loops in BPF program
2583 * loop == back-edge in directed graph
2585 static int check_cfg(struct bpf_verifier_env *env)
2587 struct bpf_insn *insns = env->prog->insnsi;
2588 int insn_cnt = env->prog->len;
2592 insn_state = kcalloc(insn_cnt, sizeof(int), GFP_KERNEL);
2596 insn_stack = kcalloc(insn_cnt, sizeof(int), GFP_KERNEL);
2602 insn_state[0] = DISCOVERED; /* mark 1st insn as discovered */
2603 insn_stack[0] = 0; /* 0 is the first instruction */
2609 t = insn_stack[cur_stack - 1];
2611 if (BPF_CLASS(insns[t].code) == BPF_JMP) {
2612 u8 opcode = BPF_OP(insns[t].code);
2614 if (opcode == BPF_EXIT) {
2616 } else if (opcode == BPF_CALL) {
2617 ret = push_insn(t, t + 1, FALLTHROUGH, env);
2622 if (t + 1 < insn_cnt)
2623 env->explored_states[t + 1] = STATE_LIST_MARK;
2624 } else if (opcode == BPF_JA) {
2625 if (BPF_SRC(insns[t].code) != BPF_K) {
2629 /* unconditional jump with single edge */
2630 ret = push_insn(t, t + insns[t].off + 1,
2636 /* tell verifier to check for equivalent states
2637 * after every call and jump
2639 if (t + 1 < insn_cnt)
2640 env->explored_states[t + 1] = STATE_LIST_MARK;
2642 /* conditional jump with two edges */
2643 env->explored_states[t] = STATE_LIST_MARK;
2644 ret = push_insn(t, t + 1, FALLTHROUGH, env);
2650 ret = push_insn(t, t + insns[t].off + 1, BRANCH, env);
2657 /* all other non-branch instructions with single
2660 ret = push_insn(t, t + 1, FALLTHROUGH, env);
2668 insn_state[t] = EXPLORED;
2669 if (cur_stack-- <= 0) {
2670 verbose("pop stack internal bug\n");
2677 for (i = 0; i < insn_cnt; i++) {
2678 if (insn_state[i] != EXPLORED) {
2679 verbose("unreachable insn %d\n", i);
2684 ret = 0; /* cfg looks good */
2692 /* the following conditions reduce the number of explored insns
2693 * from ~140k to ~80k for ultra large programs that use a lot of ptr_to_packet
2695 static bool compare_ptrs_to_packet(struct bpf_reg_state *old,
2696 struct bpf_reg_state *cur)
2698 if (old->id != cur->id)
2701 /* old ptr_to_packet is more conservative, since it allows smaller
2703 * old(off=0,r=10) is equal to cur(off=0,r=20), because
2704 * old(off=0,r=10) means that with range=10 the verifier proceeded
2705 * further and found no issues with the program. Now we're in the same
2706 * spot with cur(off=0,r=20), so we're safe too, since anything further
2707 * will only be looking at most 10 bytes after this pointer.
2709 if (old->off == cur->off && old->range < cur->range)
2712 /* old(off=20,r=10) is equal to cur(off=22,re=22 or 5 or 0)
2713 * since both cannot be used for packet access and safe(old)
2714 * pointer has smaller off that could be used for further
2715 * 'if (ptr > data_end)' check
2717 * old(off=20,r=10) and cur(off=22,r=22) and cur(off=22,r=0) mean
2718 * that we cannot access the packet.
2719 * The safe range is:
2720 * [ptr, ptr + range - off)
2721 * so whenever off >=range, it means no safe bytes from this pointer.
2722 * When comparing old->off <= cur->off, it means that older code
2723 * went with smaller offset and that offset was later
2724 * used to figure out the safe range after 'if (ptr > data_end)' check
2725 * Say, 'old' state was explored like:
2726 * ... R3(off=0, r=0)
2728 * ... now R4(off=20,r=0) <-- here
2729 * if (R4 > data_end)
2730 * ... R4(off=20,r=20), R3(off=0,r=20) and R3 can be used to access.
2731 * ... the code further went all the way to bpf_exit.
2732 * Now the 'cur' state at the mark 'here' has R4(off=30,r=0).
2733 * old_R4(off=20,r=0) equal to cur_R4(off=30,r=0), since if the verifier
2734 * goes further, such cur_R4 will give larger safe packet range after
2735 * 'if (R4 > data_end)' and all further insn were already good with r=20,
2736 * so they will be good with r=30 and we can prune the search.
2738 if (old->off <= cur->off &&
2739 old->off >= old->range && cur->off >= cur->range)
2745 /* compare two verifier states
2747 * all states stored in state_list are known to be valid, since
2748 * verifier reached 'bpf_exit' instruction through them
2750 * this function is called when verifier exploring different branches of
2751 * execution popped from the state stack. If it sees an old state that has
2752 * more strict register state and more strict stack state then this execution
2753 * branch doesn't need to be explored further, since verifier already
2754 * concluded that more strict state leads to valid finish.
2756 * Therefore two states are equivalent if register state is more conservative
2757 * and explored stack state is more conservative than the current one.
2760 * (slot1=INV slot2=MISC) == (slot1=MISC slot2=MISC)
2761 * (slot1=MISC slot2=MISC) != (slot1=INV slot2=MISC)
2763 * In other words if current stack state (one being explored) has more
2764 * valid slots than old one that already passed validation, it means
2765 * the verifier can stop exploring and conclude that current state is valid too
2767 * Similarly with registers. If explored state has register type as invalid
2768 * whereas register type in current state is meaningful, it means that
2769 * the current state will reach 'bpf_exit' instruction safely
2771 static bool states_equal(struct bpf_verifier_env *env,
2772 struct bpf_verifier_state *old,
2773 struct bpf_verifier_state *cur)
2775 bool varlen_map_access = env->varlen_map_value_access;
2776 struct bpf_reg_state *rold, *rcur;
2779 for (i = 0; i < MAX_BPF_REG; i++) {
2780 rold = &old->regs[i];
2781 rcur = &cur->regs[i];
2783 if (memcmp(rold, rcur, sizeof(*rold)) == 0)
2786 /* If the ranges were not the same, but everything else was and
2787 * we didn't do a variable access into a map then we are a-ok.
2789 if (!varlen_map_access &&
2790 memcmp(rold, rcur, offsetofend(struct bpf_reg_state, id)) == 0)
2793 /* If we didn't map access then again we don't care about the
2794 * mismatched range values and it's ok if our old type was
2795 * UNKNOWN and we didn't go to a NOT_INIT'ed reg.
2797 if (rold->type == NOT_INIT ||
2798 (!varlen_map_access && rold->type == UNKNOWN_VALUE &&
2799 rcur->type != NOT_INIT))
2802 /* Don't care about the reg->id in this case. */
2803 if (rold->type == PTR_TO_MAP_VALUE_OR_NULL &&
2804 rcur->type == PTR_TO_MAP_VALUE_OR_NULL &&
2805 rold->map_ptr == rcur->map_ptr)
2808 if (rold->type == PTR_TO_PACKET && rcur->type == PTR_TO_PACKET &&
2809 compare_ptrs_to_packet(rold, rcur))
2815 for (i = 0; i < MAX_BPF_STACK; i++) {
2816 if (old->stack_slot_type[i] == STACK_INVALID)
2818 if (old->stack_slot_type[i] != cur->stack_slot_type[i])
2819 /* Ex: old explored (safe) state has STACK_SPILL in
2820 * this stack slot, but current has has STACK_MISC ->
2821 * this verifier states are not equivalent,
2822 * return false to continue verification of this path
2825 if (i % BPF_REG_SIZE)
2827 if (memcmp(&old->spilled_regs[i / BPF_REG_SIZE],
2828 &cur->spilled_regs[i / BPF_REG_SIZE],
2829 sizeof(old->spilled_regs[0])))
2830 /* when explored and current stack slot types are
2831 * the same, check that stored pointers types
2832 * are the same as well.
2833 * Ex: explored safe path could have stored
2834 * (bpf_reg_state) {.type = PTR_TO_STACK, .imm = -8}
2835 * but current path has stored:
2836 * (bpf_reg_state) {.type = PTR_TO_STACK, .imm = -16}
2837 * such verifier states are not equivalent.
2838 * return false to continue verification of this path
2847 static int is_state_visited(struct bpf_verifier_env *env, int insn_idx)
2849 struct bpf_verifier_state_list *new_sl;
2850 struct bpf_verifier_state_list *sl;
2852 sl = env->explored_states[insn_idx];
2854 /* this 'insn_idx' instruction wasn't marked, so we will not
2855 * be doing state search here
2859 while (sl != STATE_LIST_MARK) {
2860 if (states_equal(env, &sl->state, &env->cur_state))
2861 /* reached equivalent register/stack state,
2868 /* there were no equivalent states, remember current one.
2869 * technically the current state is not proven to be safe yet,
2870 * but it will either reach bpf_exit (which means it's safe) or
2871 * it will be rejected. Since there are no loops, we won't be
2872 * seeing this 'insn_idx' instruction again on the way to bpf_exit
2874 new_sl = kmalloc(sizeof(struct bpf_verifier_state_list), GFP_USER);
2878 /* add new state to the head of linked list */
2879 memcpy(&new_sl->state, &env->cur_state, sizeof(env->cur_state));
2880 new_sl->next = env->explored_states[insn_idx];
2881 env->explored_states[insn_idx] = new_sl;
2885 static int ext_analyzer_insn_hook(struct bpf_verifier_env *env,
2886 int insn_idx, int prev_insn_idx)
2888 if (!env->analyzer_ops || !env->analyzer_ops->insn_hook)
2891 return env->analyzer_ops->insn_hook(env, insn_idx, prev_insn_idx);
2894 static int do_check(struct bpf_verifier_env *env)
2896 struct bpf_verifier_state *state = &env->cur_state;
2897 struct bpf_insn *insns = env->prog->insnsi;
2898 struct bpf_reg_state *regs = state->regs;
2899 int insn_cnt = env->prog->len;
2900 int insn_idx, prev_insn_idx = 0;
2901 int insn_processed = 0;
2902 bool do_print_state = false;
2904 init_reg_state(regs);
2906 env->varlen_map_value_access = false;
2908 struct bpf_insn *insn;
2912 if (insn_idx >= insn_cnt) {
2913 verbose("invalid insn idx %d insn_cnt %d\n",
2914 insn_idx, insn_cnt);
2918 insn = &insns[insn_idx];
2919 class = BPF_CLASS(insn->code);
2921 if (++insn_processed > BPF_COMPLEXITY_LIMIT_INSNS) {
2922 verbose("BPF program is too large. Processed %d insn\n",
2927 err = is_state_visited(env, insn_idx);
2931 /* found equivalent state, can prune the search */
2934 verbose("\nfrom %d to %d: safe\n",
2935 prev_insn_idx, insn_idx);
2937 verbose("%d: safe\n", insn_idx);
2939 goto process_bpf_exit;
2945 if (log_level > 1 || (log_level && do_print_state)) {
2947 verbose("%d:", insn_idx);
2949 verbose("\nfrom %d to %d:",
2950 prev_insn_idx, insn_idx);
2951 print_verifier_state(&env->cur_state);
2952 do_print_state = false;
2956 verbose("%d: ", insn_idx);
2957 print_bpf_insn(env, insn);
2960 err = ext_analyzer_insn_hook(env, insn_idx, prev_insn_idx);
2964 if (class == BPF_ALU || class == BPF_ALU64) {
2965 err = check_alu_op(env, insn);
2969 } else if (class == BPF_LDX) {
2970 enum bpf_reg_type *prev_src_type, src_reg_type;
2972 /* check for reserved fields is already done */
2974 /* check src operand */
2975 err = check_reg_arg(regs, insn->src_reg, SRC_OP);
2979 err = check_reg_arg(regs, insn->dst_reg, DST_OP_NO_MARK);
2983 src_reg_type = regs[insn->src_reg].type;
2985 /* check that memory (src_reg + off) is readable,
2986 * the state of dst_reg will be updated by this func
2988 err = check_mem_access(env, insn->src_reg, insn->off,
2989 BPF_SIZE(insn->code), BPF_READ,
2994 if (BPF_SIZE(insn->code) != BPF_W &&
2995 BPF_SIZE(insn->code) != BPF_DW) {
3000 prev_src_type = &env->insn_aux_data[insn_idx].ptr_type;
3002 if (*prev_src_type == NOT_INIT) {
3004 * dst_reg = *(u32 *)(src_reg + off)
3005 * save type to validate intersecting paths
3007 *prev_src_type = src_reg_type;
3009 } else if (src_reg_type != *prev_src_type &&
3010 (src_reg_type == PTR_TO_CTX ||
3011 *prev_src_type == PTR_TO_CTX)) {
3012 /* ABuser program is trying to use the same insn
3013 * dst_reg = *(u32*) (src_reg + off)
3014 * with different pointer types:
3015 * src_reg == ctx in one branch and
3016 * src_reg == stack|map in some other branch.
3019 verbose("same insn cannot be used with different pointers\n");
3023 } else if (class == BPF_STX) {
3024 enum bpf_reg_type *prev_dst_type, dst_reg_type;
3026 if (BPF_MODE(insn->code) == BPF_XADD) {
3027 err = check_xadd(env, insn);
3034 /* check src1 operand */
3035 err = check_reg_arg(regs, insn->src_reg, SRC_OP);
3038 /* check src2 operand */
3039 err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
3043 dst_reg_type = regs[insn->dst_reg].type;
3045 /* check that memory (dst_reg + off) is writeable */
3046 err = check_mem_access(env, insn->dst_reg, insn->off,
3047 BPF_SIZE(insn->code), BPF_WRITE,
3052 prev_dst_type = &env->insn_aux_data[insn_idx].ptr_type;
3054 if (*prev_dst_type == NOT_INIT) {
3055 *prev_dst_type = dst_reg_type;
3056 } else if (dst_reg_type != *prev_dst_type &&
3057 (dst_reg_type == PTR_TO_CTX ||
3058 *prev_dst_type == PTR_TO_CTX)) {
3059 verbose("same insn cannot be used with different pointers\n");
3063 } else if (class == BPF_ST) {
3064 if (BPF_MODE(insn->code) != BPF_MEM ||
3065 insn->src_reg != BPF_REG_0) {
3066 verbose("BPF_ST uses reserved fields\n");
3069 /* check src operand */
3070 err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
3074 /* check that memory (dst_reg + off) is writeable */
3075 err = check_mem_access(env, insn->dst_reg, insn->off,
3076 BPF_SIZE(insn->code), BPF_WRITE,
3081 } else if (class == BPF_JMP) {
3082 u8 opcode = BPF_OP(insn->code);
3084 if (opcode == BPF_CALL) {
3085 if (BPF_SRC(insn->code) != BPF_K ||
3087 insn->src_reg != BPF_REG_0 ||
3088 insn->dst_reg != BPF_REG_0) {
3089 verbose("BPF_CALL uses reserved fields\n");
3093 err = check_call(env, insn->imm, insn_idx);
3097 } else if (opcode == BPF_JA) {
3098 if (BPF_SRC(insn->code) != BPF_K ||
3100 insn->src_reg != BPF_REG_0 ||
3101 insn->dst_reg != BPF_REG_0) {
3102 verbose("BPF_JA uses reserved fields\n");
3106 insn_idx += insn->off + 1;
3109 } else if (opcode == BPF_EXIT) {
3110 if (BPF_SRC(insn->code) != BPF_K ||
3112 insn->src_reg != BPF_REG_0 ||
3113 insn->dst_reg != BPF_REG_0) {
3114 verbose("BPF_EXIT uses reserved fields\n");
3118 /* eBPF calling convetion is such that R0 is used
3119 * to return the value from eBPF program.
3120 * Make sure that it's readable at this time
3121 * of bpf_exit, which means that program wrote
3122 * something into it earlier
3124 err = check_reg_arg(regs, BPF_REG_0, SRC_OP);
3128 if (is_pointer_value(env, BPF_REG_0)) {
3129 verbose("R0 leaks addr as return value\n");
3134 insn_idx = pop_stack(env, &prev_insn_idx);
3138 do_print_state = true;
3142 err = check_cond_jmp_op(env, insn, &insn_idx);
3146 } else if (class == BPF_LD) {
3147 u8 mode = BPF_MODE(insn->code);
3149 if (mode == BPF_ABS || mode == BPF_IND) {
3150 err = check_ld_abs(env, insn);
3154 } else if (mode == BPF_IMM) {
3155 err = check_ld_imm(env, insn);
3161 verbose("invalid BPF_LD mode\n");
3164 reset_reg_range_values(regs, insn->dst_reg);
3166 verbose("unknown insn class %d\n", class);
3173 verbose("processed %d insns\n", insn_processed);
3177 static int check_map_prealloc(struct bpf_map *map)
3179 return (map->map_type != BPF_MAP_TYPE_HASH &&
3180 map->map_type != BPF_MAP_TYPE_PERCPU_HASH &&
3181 map->map_type != BPF_MAP_TYPE_HASH_OF_MAPS) ||
3182 !(map->map_flags & BPF_F_NO_PREALLOC);
3185 static int check_map_prog_compatibility(struct bpf_map *map,
3186 struct bpf_prog *prog)
3189 /* Make sure that BPF_PROG_TYPE_PERF_EVENT programs only use
3190 * preallocated hash maps, since doing memory allocation
3191 * in overflow_handler can crash depending on where nmi got
3194 if (prog->type == BPF_PROG_TYPE_PERF_EVENT) {
3195 if (!check_map_prealloc(map)) {
3196 verbose("perf_event programs can only use preallocated hash map\n");
3199 if (map->inner_map_meta &&
3200 !check_map_prealloc(map->inner_map_meta)) {
3201 verbose("perf_event programs can only use preallocated inner hash map\n");
3208 /* look for pseudo eBPF instructions that access map FDs and
3209 * replace them with actual map pointers
3211 static int replace_map_fd_with_map_ptr(struct bpf_verifier_env *env)
3213 struct bpf_insn *insn = env->prog->insnsi;
3214 int insn_cnt = env->prog->len;
3217 err = bpf_prog_calc_tag(env->prog);
3221 for (i = 0; i < insn_cnt; i++, insn++) {
3222 if (BPF_CLASS(insn->code) == BPF_LDX &&
3223 (BPF_MODE(insn->code) != BPF_MEM || insn->imm != 0)) {
3224 verbose("BPF_LDX uses reserved fields\n");
3228 if (BPF_CLASS(insn->code) == BPF_STX &&
3229 ((BPF_MODE(insn->code) != BPF_MEM &&
3230 BPF_MODE(insn->code) != BPF_XADD) || insn->imm != 0)) {
3231 verbose("BPF_STX uses reserved fields\n");
3235 if (insn[0].code == (BPF_LD | BPF_IMM | BPF_DW)) {
3236 struct bpf_map *map;
3239 if (i == insn_cnt - 1 || insn[1].code != 0 ||
3240 insn[1].dst_reg != 0 || insn[1].src_reg != 0 ||
3242 verbose("invalid bpf_ld_imm64 insn\n");
3246 if (insn->src_reg == 0)
3247 /* valid generic load 64-bit imm */
3250 if (insn->src_reg != BPF_PSEUDO_MAP_FD) {
3251 verbose("unrecognized bpf_ld_imm64 insn\n");
3255 f = fdget(insn->imm);
3256 map = __bpf_map_get(f);
3258 verbose("fd %d is not pointing to valid bpf_map\n",
3260 return PTR_ERR(map);
3263 err = check_map_prog_compatibility(map, env->prog);
3269 /* store map pointer inside BPF_LD_IMM64 instruction */
3270 insn[0].imm = (u32) (unsigned long) map;
3271 insn[1].imm = ((u64) (unsigned long) map) >> 32;
3273 /* check whether we recorded this map already */
3274 for (j = 0; j < env->used_map_cnt; j++)
3275 if (env->used_maps[j] == map) {
3280 if (env->used_map_cnt >= MAX_USED_MAPS) {
3285 /* hold the map. If the program is rejected by verifier,
3286 * the map will be released by release_maps() or it
3287 * will be used by the valid program until it's unloaded
3288 * and all maps are released in free_bpf_prog_info()
3290 map = bpf_map_inc(map, false);
3293 return PTR_ERR(map);
3295 env->used_maps[env->used_map_cnt++] = map;
3304 /* now all pseudo BPF_LD_IMM64 instructions load valid
3305 * 'struct bpf_map *' into a register instead of user map_fd.
3306 * These pointers will be used later by verifier to validate map access.
3311 /* drop refcnt of maps used by the rejected program */
3312 static void release_maps(struct bpf_verifier_env *env)
3316 for (i = 0; i < env->used_map_cnt; i++)
3317 bpf_map_put(env->used_maps[i]);
3320 /* convert pseudo BPF_LD_IMM64 into generic BPF_LD_IMM64 */
3321 static void convert_pseudo_ld_imm64(struct bpf_verifier_env *env)
3323 struct bpf_insn *insn = env->prog->insnsi;
3324 int insn_cnt = env->prog->len;
3327 for (i = 0; i < insn_cnt; i++, insn++)
3328 if (insn->code == (BPF_LD | BPF_IMM | BPF_DW))
3332 /* single env->prog->insni[off] instruction was replaced with the range
3333 * insni[off, off + cnt). Adjust corresponding insn_aux_data by copying
3334 * [0, off) and [off, end) to new locations, so the patched range stays zero
3336 static int adjust_insn_aux_data(struct bpf_verifier_env *env, u32 prog_len,
3339 struct bpf_insn_aux_data *new_data, *old_data = env->insn_aux_data;
3343 new_data = vzalloc(sizeof(struct bpf_insn_aux_data) * prog_len);
3346 memcpy(new_data, old_data, sizeof(struct bpf_insn_aux_data) * off);
3347 memcpy(new_data + off + cnt - 1, old_data + off,
3348 sizeof(struct bpf_insn_aux_data) * (prog_len - off - cnt + 1));
3349 env->insn_aux_data = new_data;
3354 static struct bpf_prog *bpf_patch_insn_data(struct bpf_verifier_env *env, u32 off,
3355 const struct bpf_insn *patch, u32 len)
3357 struct bpf_prog *new_prog;
3359 new_prog = bpf_patch_insn_single(env->prog, off, patch, len);
3362 if (adjust_insn_aux_data(env, new_prog->len, off, len))
3367 /* convert load instructions that access fields of 'struct __sk_buff'
3368 * into sequence of instructions that access fields of 'struct sk_buff'
3370 static int convert_ctx_accesses(struct bpf_verifier_env *env)
3372 const struct bpf_verifier_ops *ops = env->prog->aux->ops;
3373 const int insn_cnt = env->prog->len;
3374 struct bpf_insn insn_buf[16], *insn;
3375 struct bpf_prog *new_prog;
3376 enum bpf_access_type type;
3377 int i, cnt, delta = 0;
3379 if (ops->gen_prologue) {
3380 cnt = ops->gen_prologue(insn_buf, env->seen_direct_write,
3382 if (cnt >= ARRAY_SIZE(insn_buf)) {
3383 verbose("bpf verifier is misconfigured\n");
3386 new_prog = bpf_patch_insn_data(env, 0, insn_buf, cnt);
3390 env->prog = new_prog;
3395 if (!ops->convert_ctx_access)
3398 insn = env->prog->insnsi + delta;
3400 for (i = 0; i < insn_cnt; i++, insn++) {
3401 if (insn->code == (BPF_LDX | BPF_MEM | BPF_B) ||
3402 insn->code == (BPF_LDX | BPF_MEM | BPF_H) ||
3403 insn->code == (BPF_LDX | BPF_MEM | BPF_W) ||
3404 insn->code == (BPF_LDX | BPF_MEM | BPF_DW))
3406 else if (insn->code == (BPF_STX | BPF_MEM | BPF_B) ||
3407 insn->code == (BPF_STX | BPF_MEM | BPF_H) ||
3408 insn->code == (BPF_STX | BPF_MEM | BPF_W) ||
3409 insn->code == (BPF_STX | BPF_MEM | BPF_DW))
3414 if (env->insn_aux_data[i + delta].ptr_type != PTR_TO_CTX)
3417 cnt = ops->convert_ctx_access(type, insn, insn_buf, env->prog);
3418 if (cnt == 0 || cnt >= ARRAY_SIZE(insn_buf)) {
3419 verbose("bpf verifier is misconfigured\n");
3423 new_prog = bpf_patch_insn_data(env, i + delta, insn_buf, cnt);
3429 /* keep walking new program and skip insns we just inserted */
3430 env->prog = new_prog;
3431 insn = new_prog->insnsi + i + delta;
3437 /* fixup insn->imm field of bpf_call instructions
3438 * and inline eligible helpers as explicit sequence of BPF instructions
3440 * this function is called after eBPF program passed verification
3442 static int fixup_bpf_calls(struct bpf_verifier_env *env)
3444 struct bpf_prog *prog = env->prog;
3445 struct bpf_insn *insn = prog->insnsi;
3446 const struct bpf_func_proto *fn;
3447 const int insn_cnt = prog->len;
3448 struct bpf_insn insn_buf[16];
3449 struct bpf_prog *new_prog;
3450 struct bpf_map *map_ptr;
3451 int i, cnt, delta = 0;
3453 for (i = 0; i < insn_cnt; i++, insn++) {
3454 if (insn->code != (BPF_JMP | BPF_CALL))
3457 if (insn->imm == BPF_FUNC_get_route_realm)
3458 prog->dst_needed = 1;
3459 if (insn->imm == BPF_FUNC_get_prandom_u32)
3460 bpf_user_rnd_init_once();
3461 if (insn->imm == BPF_FUNC_tail_call) {
3462 /* If we tail call into other programs, we
3463 * cannot make any assumptions since they can
3464 * be replaced dynamically during runtime in
3465 * the program array.
3467 prog->cb_access = 1;
3469 /* mark bpf_tail_call as different opcode to avoid
3470 * conditional branch in the interpeter for every normal
3471 * call and to prevent accidental JITing by JIT compiler
3472 * that doesn't support bpf_tail_call yet
3475 insn->code |= BPF_X;
3479 if (ebpf_jit_enabled() && insn->imm == BPF_FUNC_map_lookup_elem) {
3480 map_ptr = env->insn_aux_data[i + delta].map_ptr;
3481 if (map_ptr == BPF_MAP_PTR_POISON ||
3482 !map_ptr->ops->map_gen_lookup)
3483 goto patch_call_imm;
3485 cnt = map_ptr->ops->map_gen_lookup(map_ptr, insn_buf);
3486 if (cnt == 0 || cnt >= ARRAY_SIZE(insn_buf)) {
3487 verbose("bpf verifier is misconfigured\n");
3491 new_prog = bpf_patch_insn_data(env, i + delta, insn_buf,
3498 /* keep walking new program and skip insns we just inserted */
3499 env->prog = prog = new_prog;
3500 insn = new_prog->insnsi + i + delta;
3505 fn = prog->aux->ops->get_func_proto(insn->imm);
3506 /* all functions that have prototype and verifier allowed
3507 * programs to call them, must be real in-kernel functions
3510 verbose("kernel subsystem misconfigured func %s#%d\n",
3511 func_id_name(insn->imm), insn->imm);
3514 insn->imm = fn->func - __bpf_call_base;
3520 static void free_states(struct bpf_verifier_env *env)
3522 struct bpf_verifier_state_list *sl, *sln;
3525 if (!env->explored_states)
3528 for (i = 0; i < env->prog->len; i++) {
3529 sl = env->explored_states[i];
3532 while (sl != STATE_LIST_MARK) {
3539 kfree(env->explored_states);
3542 int bpf_check(struct bpf_prog **prog, union bpf_attr *attr)
3544 char __user *log_ubuf = NULL;
3545 struct bpf_verifier_env *env;
3548 /* 'struct bpf_verifier_env' can be global, but since it's not small,
3549 * allocate/free it every time bpf_check() is called
3551 env = kzalloc(sizeof(struct bpf_verifier_env), GFP_KERNEL);
3555 env->insn_aux_data = vzalloc(sizeof(struct bpf_insn_aux_data) *
3558 if (!env->insn_aux_data)
3562 /* grab the mutex to protect few globals used by verifier */
3563 mutex_lock(&bpf_verifier_lock);
3565 if (attr->log_level || attr->log_buf || attr->log_size) {
3566 /* user requested verbose verifier output
3567 * and supplied buffer to store the verification trace
3569 log_level = attr->log_level;
3570 log_ubuf = (char __user *) (unsigned long) attr->log_buf;
3571 log_size = attr->log_size;
3575 /* log_* values have to be sane */
3576 if (log_size < 128 || log_size > UINT_MAX >> 8 ||
3577 log_level == 0 || log_ubuf == NULL)
3581 log_buf = vmalloc(log_size);
3587 if (attr->prog_flags & BPF_F_STRICT_ALIGNMENT)
3588 env->strict_alignment = true;
3590 env->strict_alignment = false;
3592 ret = replace_map_fd_with_map_ptr(env);
3594 goto skip_full_check;
3596 env->explored_states = kcalloc(env->prog->len,
3597 sizeof(struct bpf_verifier_state_list *),
3600 if (!env->explored_states)
3601 goto skip_full_check;
3603 ret = check_cfg(env);
3605 goto skip_full_check;
3607 env->allow_ptr_leaks = capable(CAP_SYS_ADMIN);
3609 ret = do_check(env);
3612 while (pop_stack(env, NULL) >= 0);
3616 /* program is valid, convert *(u32*)(ctx + off) accesses */
3617 ret = convert_ctx_accesses(env);
3620 ret = fixup_bpf_calls(env);
3622 if (log_level && log_len >= log_size - 1) {
3623 BUG_ON(log_len >= log_size);
3624 /* verifier log exceeded user supplied buffer */
3626 /* fall through to return what was recorded */
3629 /* copy verifier log back to user space including trailing zero */
3630 if (log_level && copy_to_user(log_ubuf, log_buf, log_len + 1) != 0) {
3635 if (ret == 0 && env->used_map_cnt) {
3636 /* if program passed verifier, update used_maps in bpf_prog_info */
3637 env->prog->aux->used_maps = kmalloc_array(env->used_map_cnt,
3638 sizeof(env->used_maps[0]),
3641 if (!env->prog->aux->used_maps) {
3646 memcpy(env->prog->aux->used_maps, env->used_maps,
3647 sizeof(env->used_maps[0]) * env->used_map_cnt);
3648 env->prog->aux->used_map_cnt = env->used_map_cnt;
3650 /* program is valid. Convert pseudo bpf_ld_imm64 into generic
3651 * bpf_ld_imm64 instructions
3653 convert_pseudo_ld_imm64(env);
3659 if (!env->prog->aux->used_maps)
3660 /* if we didn't copy map pointers into bpf_prog_info, release
3661 * them now. Otherwise free_bpf_prog_info() will release them.
3666 mutex_unlock(&bpf_verifier_lock);
3667 vfree(env->insn_aux_data);
3673 int bpf_analyzer(struct bpf_prog *prog, const struct bpf_ext_analyzer_ops *ops,
3676 struct bpf_verifier_env *env;
3679 env = kzalloc(sizeof(struct bpf_verifier_env), GFP_KERNEL);
3683 env->insn_aux_data = vzalloc(sizeof(struct bpf_insn_aux_data) *
3686 if (!env->insn_aux_data)
3689 env->analyzer_ops = ops;
3690 env->analyzer_priv = priv;
3692 /* grab the mutex to protect few globals used by verifier */
3693 mutex_lock(&bpf_verifier_lock);
3696 env->strict_alignment = false;
3698 env->explored_states = kcalloc(env->prog->len,
3699 sizeof(struct bpf_verifier_state_list *),
3702 if (!env->explored_states)
3703 goto skip_full_check;
3705 ret = check_cfg(env);
3707 goto skip_full_check;
3709 env->allow_ptr_leaks = capable(CAP_SYS_ADMIN);
3711 ret = do_check(env);
3714 while (pop_stack(env, NULL) >= 0);
3717 mutex_unlock(&bpf_verifier_lock);
3718 vfree(env->insn_aux_data);
3723 EXPORT_SYMBOL_GPL(bpf_analyzer);