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 /* For platforms that do not have a Kconfig enabling
812 * CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS the value of
813 * NET_IP_ALIGN is universally set to '2'. And on platforms
814 * that do set CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS, we get
815 * to this code only in strict mode where we want to emulate
816 * the NET_IP_ALIGN==2 checking. Therefore use an
817 * unconditional IP align value of '2'.
820 if ((ip_align + reg_off + off) % size != 0) {
821 verbose("misaligned packet access off %d+%d+%d size %d\n",
822 ip_align, reg_off, off, size);
829 static int check_val_ptr_alignment(const struct bpf_reg_state *reg,
830 int size, bool strict)
832 if (strict && size != 1) {
833 verbose("Unknown alignment. Only byte-sized access allowed in value access.\n");
840 static int check_ptr_alignment(struct bpf_verifier_env *env,
841 const struct bpf_reg_state *reg,
844 bool strict = env->strict_alignment;
846 if (!IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS))
851 return check_pkt_ptr_alignment(reg, off, size, strict);
852 case PTR_TO_MAP_VALUE_ADJ:
853 return check_val_ptr_alignment(reg, size, strict);
855 if (off % size != 0) {
856 verbose("misaligned access off %d size %d\n",
865 /* check whether memory at (regno + off) is accessible for t = (read | write)
866 * if t==write, value_regno is a register which value is stored into memory
867 * if t==read, value_regno is a register which will receive the value from memory
868 * if t==write && value_regno==-1, some unknown value is stored into memory
869 * if t==read && value_regno==-1, don't care what we read from memory
871 static int check_mem_access(struct bpf_verifier_env *env, u32 regno, int off,
872 int bpf_size, enum bpf_access_type t,
875 struct bpf_verifier_state *state = &env->cur_state;
876 struct bpf_reg_state *reg = &state->regs[regno];
879 if (reg->type == PTR_TO_STACK)
882 size = bpf_size_to_bytes(bpf_size);
886 err = check_ptr_alignment(env, reg, off, size);
890 if (reg->type == PTR_TO_MAP_VALUE ||
891 reg->type == PTR_TO_MAP_VALUE_ADJ) {
892 if (t == BPF_WRITE && value_regno >= 0 &&
893 is_pointer_value(env, value_regno)) {
894 verbose("R%d leaks addr into map\n", value_regno);
898 if (reg->type == PTR_TO_MAP_VALUE_ADJ)
899 err = check_map_access_adj(env, regno, off, size);
901 err = check_map_access(env, regno, off, size);
902 if (!err && t == BPF_READ && value_regno >= 0)
903 mark_reg_unknown_value_and_range(state->regs,
906 } else if (reg->type == PTR_TO_CTX) {
907 enum bpf_reg_type reg_type = UNKNOWN_VALUE;
909 if (t == BPF_WRITE && value_regno >= 0 &&
910 is_pointer_value(env, value_regno)) {
911 verbose("R%d leaks addr into ctx\n", value_regno);
914 err = check_ctx_access(env, off, size, t, ®_type);
915 if (!err && t == BPF_READ && value_regno >= 0) {
916 mark_reg_unknown_value_and_range(state->regs,
918 /* note that reg.[id|off|range] == 0 */
919 state->regs[value_regno].type = reg_type;
920 state->regs[value_regno].aux_off = 0;
921 state->regs[value_regno].aux_off_align = 0;
924 } else if (reg->type == FRAME_PTR || reg->type == PTR_TO_STACK) {
925 if (off >= 0 || off < -MAX_BPF_STACK) {
926 verbose("invalid stack off=%d size=%d\n", off, size);
929 if (t == BPF_WRITE) {
930 if (!env->allow_ptr_leaks &&
931 state->stack_slot_type[MAX_BPF_STACK + off] == STACK_SPILL &&
932 size != BPF_REG_SIZE) {
933 verbose("attempt to corrupt spilled pointer on stack\n");
936 err = check_stack_write(state, off, size, value_regno);
938 err = check_stack_read(state, off, size, value_regno);
940 } else if (state->regs[regno].type == PTR_TO_PACKET) {
941 if (t == BPF_WRITE && !may_access_direct_pkt_data(env, NULL, t)) {
942 verbose("cannot write into packet\n");
945 if (t == BPF_WRITE && value_regno >= 0 &&
946 is_pointer_value(env, value_regno)) {
947 verbose("R%d leaks addr into packet\n", value_regno);
950 err = check_packet_access(env, regno, off, size);
951 if (!err && t == BPF_READ && value_regno >= 0)
952 mark_reg_unknown_value_and_range(state->regs,
955 verbose("R%d invalid mem access '%s'\n",
956 regno, reg_type_str[reg->type]);
960 if (!err && size <= 2 && value_regno >= 0 && env->allow_ptr_leaks &&
961 state->regs[value_regno].type == UNKNOWN_VALUE) {
962 /* 1 or 2 byte load zero-extends, determine the number of
963 * zero upper bits. Not doing it fo 4 byte load, since
964 * such values cannot be added to ptr_to_packet anyway.
966 state->regs[value_regno].imm = 64 - size * 8;
971 static int check_xadd(struct bpf_verifier_env *env, struct bpf_insn *insn)
973 struct bpf_reg_state *regs = env->cur_state.regs;
976 if ((BPF_SIZE(insn->code) != BPF_W && BPF_SIZE(insn->code) != BPF_DW) ||
978 verbose("BPF_XADD uses reserved fields\n");
982 /* check src1 operand */
983 err = check_reg_arg(regs, insn->src_reg, SRC_OP);
987 /* check src2 operand */
988 err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
992 /* check whether atomic_add can read the memory */
993 err = check_mem_access(env, insn->dst_reg, insn->off,
994 BPF_SIZE(insn->code), BPF_READ, -1);
998 /* check whether atomic_add can write into the same memory */
999 return check_mem_access(env, insn->dst_reg, insn->off,
1000 BPF_SIZE(insn->code), BPF_WRITE, -1);
1003 /* when register 'regno' is passed into function that will read 'access_size'
1004 * bytes from that pointer, make sure that it's within stack boundary
1005 * and all elements of stack are initialized
1007 static int check_stack_boundary(struct bpf_verifier_env *env, int regno,
1008 int access_size, bool zero_size_allowed,
1009 struct bpf_call_arg_meta *meta)
1011 struct bpf_verifier_state *state = &env->cur_state;
1012 struct bpf_reg_state *regs = state->regs;
1015 if (regs[regno].type != PTR_TO_STACK) {
1016 if (zero_size_allowed && access_size == 0 &&
1017 regs[regno].type == CONST_IMM &&
1018 regs[regno].imm == 0)
1021 verbose("R%d type=%s expected=%s\n", regno,
1022 reg_type_str[regs[regno].type],
1023 reg_type_str[PTR_TO_STACK]);
1027 off = regs[regno].imm;
1028 if (off >= 0 || off < -MAX_BPF_STACK || off + access_size > 0 ||
1030 verbose("invalid stack type R%d off=%d access_size=%d\n",
1031 regno, off, access_size);
1035 if (meta && meta->raw_mode) {
1036 meta->access_size = access_size;
1037 meta->regno = regno;
1041 for (i = 0; i < access_size; i++) {
1042 if (state->stack_slot_type[MAX_BPF_STACK + off + i] != STACK_MISC) {
1043 verbose("invalid indirect read from stack off %d+%d size %d\n",
1044 off, i, access_size);
1051 static int check_helper_mem_access(struct bpf_verifier_env *env, int regno,
1052 int access_size, bool zero_size_allowed,
1053 struct bpf_call_arg_meta *meta)
1055 struct bpf_reg_state *regs = env->cur_state.regs;
1057 switch (regs[regno].type) {
1059 return check_packet_access(env, regno, 0, access_size);
1060 case PTR_TO_MAP_VALUE:
1061 return check_map_access(env, regno, 0, access_size);
1062 case PTR_TO_MAP_VALUE_ADJ:
1063 return check_map_access_adj(env, regno, 0, access_size);
1064 default: /* const_imm|ptr_to_stack or invalid ptr */
1065 return check_stack_boundary(env, regno, access_size,
1066 zero_size_allowed, meta);
1070 static int check_func_arg(struct bpf_verifier_env *env, u32 regno,
1071 enum bpf_arg_type arg_type,
1072 struct bpf_call_arg_meta *meta)
1074 struct bpf_reg_state *regs = env->cur_state.regs, *reg = ®s[regno];
1075 enum bpf_reg_type expected_type, type = reg->type;
1078 if (arg_type == ARG_DONTCARE)
1081 if (type == NOT_INIT) {
1082 verbose("R%d !read_ok\n", regno);
1086 if (arg_type == ARG_ANYTHING) {
1087 if (is_pointer_value(env, regno)) {
1088 verbose("R%d leaks addr into helper function\n", regno);
1094 if (type == PTR_TO_PACKET &&
1095 !may_access_direct_pkt_data(env, meta, BPF_READ)) {
1096 verbose("helper access to the packet is not allowed\n");
1100 if (arg_type == ARG_PTR_TO_MAP_KEY ||
1101 arg_type == ARG_PTR_TO_MAP_VALUE) {
1102 expected_type = PTR_TO_STACK;
1103 if (type != PTR_TO_PACKET && type != expected_type)
1105 } else if (arg_type == ARG_CONST_SIZE ||
1106 arg_type == ARG_CONST_SIZE_OR_ZERO) {
1107 expected_type = CONST_IMM;
1108 /* One exception. Allow UNKNOWN_VALUE registers when the
1109 * boundaries are known and don't cause unsafe memory accesses
1111 if (type != UNKNOWN_VALUE && type != expected_type)
1113 } else if (arg_type == ARG_CONST_MAP_PTR) {
1114 expected_type = CONST_PTR_TO_MAP;
1115 if (type != expected_type)
1117 } else if (arg_type == ARG_PTR_TO_CTX) {
1118 expected_type = PTR_TO_CTX;
1119 if (type != expected_type)
1121 } else if (arg_type == ARG_PTR_TO_MEM ||
1122 arg_type == ARG_PTR_TO_UNINIT_MEM) {
1123 expected_type = PTR_TO_STACK;
1124 /* One exception here. In case function allows for NULL to be
1125 * passed in as argument, it's a CONST_IMM type. Final test
1126 * happens during stack boundary checking.
1128 if (type == CONST_IMM && reg->imm == 0)
1129 /* final test in check_stack_boundary() */;
1130 else if (type != PTR_TO_PACKET && type != PTR_TO_MAP_VALUE &&
1131 type != PTR_TO_MAP_VALUE_ADJ && type != expected_type)
1133 meta->raw_mode = arg_type == ARG_PTR_TO_UNINIT_MEM;
1135 verbose("unsupported arg_type %d\n", arg_type);
1139 if (arg_type == ARG_CONST_MAP_PTR) {
1140 /* bpf_map_xxx(map_ptr) call: remember that map_ptr */
1141 meta->map_ptr = reg->map_ptr;
1142 } else if (arg_type == ARG_PTR_TO_MAP_KEY) {
1143 /* bpf_map_xxx(..., map_ptr, ..., key) call:
1144 * check that [key, key + map->key_size) are within
1145 * stack limits and initialized
1147 if (!meta->map_ptr) {
1148 /* in function declaration map_ptr must come before
1149 * map_key, so that it's verified and known before
1150 * we have to check map_key here. Otherwise it means
1151 * that kernel subsystem misconfigured verifier
1153 verbose("invalid map_ptr to access map->key\n");
1156 if (type == PTR_TO_PACKET)
1157 err = check_packet_access(env, regno, 0,
1158 meta->map_ptr->key_size);
1160 err = check_stack_boundary(env, regno,
1161 meta->map_ptr->key_size,
1163 } else if (arg_type == ARG_PTR_TO_MAP_VALUE) {
1164 /* bpf_map_xxx(..., map_ptr, ..., value) call:
1165 * check [value, value + map->value_size) validity
1167 if (!meta->map_ptr) {
1168 /* kernel subsystem misconfigured verifier */
1169 verbose("invalid map_ptr to access map->value\n");
1172 if (type == PTR_TO_PACKET)
1173 err = check_packet_access(env, regno, 0,
1174 meta->map_ptr->value_size);
1176 err = check_stack_boundary(env, regno,
1177 meta->map_ptr->value_size,
1179 } else if (arg_type == ARG_CONST_SIZE ||
1180 arg_type == ARG_CONST_SIZE_OR_ZERO) {
1181 bool zero_size_allowed = (arg_type == ARG_CONST_SIZE_OR_ZERO);
1183 /* bpf_xxx(..., buf, len) call will access 'len' bytes
1184 * from stack pointer 'buf'. Check it
1185 * note: regno == len, regno - 1 == buf
1188 /* kernel subsystem misconfigured verifier */
1189 verbose("ARG_CONST_SIZE cannot be first argument\n");
1193 /* If the register is UNKNOWN_VALUE, the access check happens
1194 * using its boundaries. Otherwise, just use its imm
1196 if (type == UNKNOWN_VALUE) {
1197 /* For unprivileged variable accesses, disable raw
1198 * mode so that the program is required to
1199 * initialize all the memory that the helper could
1200 * just partially fill up.
1204 if (reg->min_value < 0) {
1205 verbose("R%d min value is negative, either use unsigned or 'var &= const'\n",
1210 if (reg->min_value == 0) {
1211 err = check_helper_mem_access(env, regno - 1, 0,
1218 if (reg->max_value == BPF_REGISTER_MAX_RANGE) {
1219 verbose("R%d unbounded memory access, use 'var &= const' or 'if (var < const)'\n",
1223 err = check_helper_mem_access(env, regno - 1,
1225 zero_size_allowed, meta);
1229 /* register is CONST_IMM */
1230 err = check_helper_mem_access(env, regno - 1, reg->imm,
1231 zero_size_allowed, meta);
1237 verbose("R%d type=%s expected=%s\n", regno,
1238 reg_type_str[type], reg_type_str[expected_type]);
1242 static int check_map_func_compatibility(struct bpf_map *map, int func_id)
1247 /* We need a two way check, first is from map perspective ... */
1248 switch (map->map_type) {
1249 case BPF_MAP_TYPE_PROG_ARRAY:
1250 if (func_id != BPF_FUNC_tail_call)
1253 case BPF_MAP_TYPE_PERF_EVENT_ARRAY:
1254 if (func_id != BPF_FUNC_perf_event_read &&
1255 func_id != BPF_FUNC_perf_event_output)
1258 case BPF_MAP_TYPE_STACK_TRACE:
1259 if (func_id != BPF_FUNC_get_stackid)
1262 case BPF_MAP_TYPE_CGROUP_ARRAY:
1263 if (func_id != BPF_FUNC_skb_under_cgroup &&
1264 func_id != BPF_FUNC_current_task_under_cgroup)
1267 case BPF_MAP_TYPE_ARRAY_OF_MAPS:
1268 case BPF_MAP_TYPE_HASH_OF_MAPS:
1269 if (func_id != BPF_FUNC_map_lookup_elem)
1275 /* ... and second from the function itself. */
1277 case BPF_FUNC_tail_call:
1278 if (map->map_type != BPF_MAP_TYPE_PROG_ARRAY)
1281 case BPF_FUNC_perf_event_read:
1282 case BPF_FUNC_perf_event_output:
1283 if (map->map_type != BPF_MAP_TYPE_PERF_EVENT_ARRAY)
1286 case BPF_FUNC_get_stackid:
1287 if (map->map_type != BPF_MAP_TYPE_STACK_TRACE)
1290 case BPF_FUNC_current_task_under_cgroup:
1291 case BPF_FUNC_skb_under_cgroup:
1292 if (map->map_type != BPF_MAP_TYPE_CGROUP_ARRAY)
1301 verbose("cannot pass map_type %d into func %s#%d\n",
1302 map->map_type, func_id_name(func_id), func_id);
1306 static int check_raw_mode(const struct bpf_func_proto *fn)
1310 if (fn->arg1_type == ARG_PTR_TO_UNINIT_MEM)
1312 if (fn->arg2_type == ARG_PTR_TO_UNINIT_MEM)
1314 if (fn->arg3_type == ARG_PTR_TO_UNINIT_MEM)
1316 if (fn->arg4_type == ARG_PTR_TO_UNINIT_MEM)
1318 if (fn->arg5_type == ARG_PTR_TO_UNINIT_MEM)
1321 return count > 1 ? -EINVAL : 0;
1324 static void clear_all_pkt_pointers(struct bpf_verifier_env *env)
1326 struct bpf_verifier_state *state = &env->cur_state;
1327 struct bpf_reg_state *regs = state->regs, *reg;
1330 for (i = 0; i < MAX_BPF_REG; i++)
1331 if (regs[i].type == PTR_TO_PACKET ||
1332 regs[i].type == PTR_TO_PACKET_END)
1333 mark_reg_unknown_value(regs, i);
1335 for (i = 0; i < MAX_BPF_STACK; i += BPF_REG_SIZE) {
1336 if (state->stack_slot_type[i] != STACK_SPILL)
1338 reg = &state->spilled_regs[i / BPF_REG_SIZE];
1339 if (reg->type != PTR_TO_PACKET &&
1340 reg->type != PTR_TO_PACKET_END)
1342 reg->type = UNKNOWN_VALUE;
1347 static int check_call(struct bpf_verifier_env *env, int func_id, int insn_idx)
1349 struct bpf_verifier_state *state = &env->cur_state;
1350 const struct bpf_func_proto *fn = NULL;
1351 struct bpf_reg_state *regs = state->regs;
1352 struct bpf_reg_state *reg;
1353 struct bpf_call_arg_meta meta;
1357 /* find function prototype */
1358 if (func_id < 0 || func_id >= __BPF_FUNC_MAX_ID) {
1359 verbose("invalid func %s#%d\n", func_id_name(func_id), func_id);
1363 if (env->prog->aux->ops->get_func_proto)
1364 fn = env->prog->aux->ops->get_func_proto(func_id);
1367 verbose("unknown func %s#%d\n", func_id_name(func_id), func_id);
1371 /* eBPF programs must be GPL compatible to use GPL-ed functions */
1372 if (!env->prog->gpl_compatible && fn->gpl_only) {
1373 verbose("cannot call GPL only function from proprietary program\n");
1377 changes_data = bpf_helper_changes_pkt_data(fn->func);
1379 memset(&meta, 0, sizeof(meta));
1380 meta.pkt_access = fn->pkt_access;
1382 /* We only support one arg being in raw mode at the moment, which
1383 * is sufficient for the helper functions we have right now.
1385 err = check_raw_mode(fn);
1387 verbose("kernel subsystem misconfigured func %s#%d\n",
1388 func_id_name(func_id), func_id);
1393 err = check_func_arg(env, BPF_REG_1, fn->arg1_type, &meta);
1396 err = check_func_arg(env, BPF_REG_2, fn->arg2_type, &meta);
1399 err = check_func_arg(env, BPF_REG_3, fn->arg3_type, &meta);
1402 err = check_func_arg(env, BPF_REG_4, fn->arg4_type, &meta);
1405 err = check_func_arg(env, BPF_REG_5, fn->arg5_type, &meta);
1409 /* Mark slots with STACK_MISC in case of raw mode, stack offset
1410 * is inferred from register state.
1412 for (i = 0; i < meta.access_size; i++) {
1413 err = check_mem_access(env, meta.regno, i, BPF_B, BPF_WRITE, -1);
1418 /* reset caller saved regs */
1419 for (i = 0; i < CALLER_SAVED_REGS; i++) {
1420 reg = regs + caller_saved[i];
1421 reg->type = NOT_INIT;
1425 /* update return register */
1426 if (fn->ret_type == RET_INTEGER) {
1427 regs[BPF_REG_0].type = UNKNOWN_VALUE;
1428 } else if (fn->ret_type == RET_VOID) {
1429 regs[BPF_REG_0].type = NOT_INIT;
1430 } else if (fn->ret_type == RET_PTR_TO_MAP_VALUE_OR_NULL) {
1431 struct bpf_insn_aux_data *insn_aux;
1433 regs[BPF_REG_0].type = PTR_TO_MAP_VALUE_OR_NULL;
1434 regs[BPF_REG_0].max_value = regs[BPF_REG_0].min_value = 0;
1435 /* remember map_ptr, so that check_map_access()
1436 * can check 'value_size' boundary of memory access
1437 * to map element returned from bpf_map_lookup_elem()
1439 if (meta.map_ptr == NULL) {
1440 verbose("kernel subsystem misconfigured verifier\n");
1443 regs[BPF_REG_0].map_ptr = meta.map_ptr;
1444 regs[BPF_REG_0].id = ++env->id_gen;
1445 insn_aux = &env->insn_aux_data[insn_idx];
1446 if (!insn_aux->map_ptr)
1447 insn_aux->map_ptr = meta.map_ptr;
1448 else if (insn_aux->map_ptr != meta.map_ptr)
1449 insn_aux->map_ptr = BPF_MAP_PTR_POISON;
1451 verbose("unknown return type %d of func %s#%d\n",
1452 fn->ret_type, func_id_name(func_id), func_id);
1456 err = check_map_func_compatibility(meta.map_ptr, func_id);
1461 clear_all_pkt_pointers(env);
1465 static int check_packet_ptr_add(struct bpf_verifier_env *env,
1466 struct bpf_insn *insn)
1468 struct bpf_reg_state *regs = env->cur_state.regs;
1469 struct bpf_reg_state *dst_reg = ®s[insn->dst_reg];
1470 struct bpf_reg_state *src_reg = ®s[insn->src_reg];
1471 struct bpf_reg_state tmp_reg;
1474 if (BPF_SRC(insn->code) == BPF_K) {
1475 /* pkt_ptr += imm */
1480 verbose("addition of negative constant to packet pointer is not allowed\n");
1483 if (imm >= MAX_PACKET_OFF ||
1484 imm + dst_reg->off >= MAX_PACKET_OFF) {
1485 verbose("constant %d is too large to add to packet pointer\n",
1489 /* a constant was added to pkt_ptr.
1490 * Remember it while keeping the same 'id'
1492 dst_reg->off += imm;
1496 if (src_reg->type == PTR_TO_PACKET) {
1497 /* R6=pkt(id=0,off=0,r=62) R7=imm22; r7 += r6 */
1498 tmp_reg = *dst_reg; /* save r7 state */
1499 *dst_reg = *src_reg; /* copy pkt_ptr state r6 into r7 */
1500 src_reg = &tmp_reg; /* pretend it's src_reg state */
1501 /* if the checks below reject it, the copy won't matter,
1502 * since we're rejecting the whole program. If all ok,
1503 * then imm22 state will be added to r7
1504 * and r7 will be pkt(id=0,off=22,r=62) while
1505 * r6 will stay as pkt(id=0,off=0,r=62)
1509 if (src_reg->type == CONST_IMM) {
1510 /* pkt_ptr += reg where reg is known constant */
1514 /* disallow pkt_ptr += reg
1515 * if reg is not uknown_value with guaranteed zero upper bits
1516 * otherwise pkt_ptr may overflow and addition will become
1517 * subtraction which is not allowed
1519 if (src_reg->type != UNKNOWN_VALUE) {
1520 verbose("cannot add '%s' to ptr_to_packet\n",
1521 reg_type_str[src_reg->type]);
1524 if (src_reg->imm < 48) {
1525 verbose("cannot add integer value with %lld upper zero bits to ptr_to_packet\n",
1530 had_id = (dst_reg->id != 0);
1532 /* dst_reg stays as pkt_ptr type and since some positive
1533 * integer value was added to the pointer, increment its 'id'
1535 dst_reg->id = ++env->id_gen;
1537 /* something was added to pkt_ptr, set range to zero */
1538 dst_reg->aux_off += dst_reg->off;
1542 dst_reg->aux_off_align = min(dst_reg->aux_off_align,
1543 src_reg->min_align);
1545 dst_reg->aux_off_align = src_reg->min_align;
1550 static int evaluate_reg_alu(struct bpf_verifier_env *env, struct bpf_insn *insn)
1552 struct bpf_reg_state *regs = env->cur_state.regs;
1553 struct bpf_reg_state *dst_reg = ®s[insn->dst_reg];
1554 u8 opcode = BPF_OP(insn->code);
1557 /* for type == UNKNOWN_VALUE:
1558 * imm > 0 -> number of zero upper bits
1559 * imm == 0 -> don't track which is the same as all bits can be non-zero
1562 if (BPF_SRC(insn->code) == BPF_X) {
1563 struct bpf_reg_state *src_reg = ®s[insn->src_reg];
1565 if (src_reg->type == UNKNOWN_VALUE && src_reg->imm > 0 &&
1566 dst_reg->imm && opcode == BPF_ADD) {
1568 * where both have zero upper bits. Adding them
1569 * can only result making one more bit non-zero
1570 * in the larger value.
1571 * Ex. 0xffff (imm=48) + 1 (imm=63) = 0x10000 (imm=47)
1572 * 0xffff (imm=48) + 0xffff = 0x1fffe (imm=47)
1574 dst_reg->imm = min(dst_reg->imm, src_reg->imm);
1578 if (src_reg->type == CONST_IMM && src_reg->imm > 0 &&
1579 dst_reg->imm && opcode == BPF_ADD) {
1581 * where dreg has zero upper bits and sreg is const.
1582 * Adding them can only result making one more bit
1583 * non-zero in the larger value.
1585 imm_log2 = __ilog2_u64((long long)src_reg->imm);
1586 dst_reg->imm = min(dst_reg->imm, 63 - imm_log2);
1590 /* all other cases non supported yet, just mark dst_reg */
1595 /* sign extend 32-bit imm into 64-bit to make sure that
1596 * negative values occupy bit 63. Note ilog2() would have
1597 * been incorrect, since sizeof(insn->imm) == 4
1599 imm_log2 = __ilog2_u64((long long)insn->imm);
1601 if (dst_reg->imm && opcode == BPF_LSH) {
1603 * if reg was a result of 2 byte load, then its imm == 48
1604 * which means that upper 48 bits are zero and shifting this reg
1605 * left by 4 would mean that upper 44 bits are still zero
1607 dst_reg->imm -= insn->imm;
1608 } else if (dst_reg->imm && opcode == BPF_MUL) {
1610 * if multiplying by 14 subtract 4
1611 * This is conservative calculation of upper zero bits.
1612 * It's not trying to special case insn->imm == 1 or 0 cases
1614 dst_reg->imm -= imm_log2 + 1;
1615 } else if (opcode == BPF_AND) {
1617 dst_reg->imm = 63 - imm_log2;
1618 } else if (dst_reg->imm && opcode == BPF_ADD) {
1620 dst_reg->imm = min(dst_reg->imm, 63 - imm_log2);
1622 } else if (opcode == BPF_RSH) {
1624 * which means that after right shift, upper bits will be zero
1625 * note that verifier already checked that
1626 * 0 <= imm < 64 for shift insn
1628 dst_reg->imm += insn->imm;
1629 if (unlikely(dst_reg->imm > 64))
1630 /* some dumb code did:
1633 * and all bits are zero now */
1636 /* all other alu ops, means that we don't know what will
1637 * happen to the value, mark it with unknown number of zero bits
1642 if (dst_reg->imm < 0) {
1643 /* all 64 bits of the register can contain non-zero bits
1644 * and such value cannot be added to ptr_to_packet, since it
1645 * may overflow, mark it as unknown to avoid further eval
1652 static int evaluate_reg_imm_alu(struct bpf_verifier_env *env,
1653 struct bpf_insn *insn)
1655 struct bpf_reg_state *regs = env->cur_state.regs;
1656 struct bpf_reg_state *dst_reg = ®s[insn->dst_reg];
1657 struct bpf_reg_state *src_reg = ®s[insn->src_reg];
1658 u8 opcode = BPF_OP(insn->code);
1659 u64 dst_imm = dst_reg->imm;
1661 /* dst_reg->type == CONST_IMM here. Simulate execution of insns
1662 * containing ALU ops. Don't care about overflow or negative
1663 * values, just add/sub/... them; registers are in u64.
1665 if (opcode == BPF_ADD && BPF_SRC(insn->code) == BPF_K) {
1666 dst_imm += insn->imm;
1667 } else if (opcode == BPF_ADD && BPF_SRC(insn->code) == BPF_X &&
1668 src_reg->type == CONST_IMM) {
1669 dst_imm += src_reg->imm;
1670 } else if (opcode == BPF_SUB && BPF_SRC(insn->code) == BPF_K) {
1671 dst_imm -= insn->imm;
1672 } else if (opcode == BPF_SUB && BPF_SRC(insn->code) == BPF_X &&
1673 src_reg->type == CONST_IMM) {
1674 dst_imm -= src_reg->imm;
1675 } else if (opcode == BPF_MUL && BPF_SRC(insn->code) == BPF_K) {
1676 dst_imm *= insn->imm;
1677 } else if (opcode == BPF_MUL && BPF_SRC(insn->code) == BPF_X &&
1678 src_reg->type == CONST_IMM) {
1679 dst_imm *= src_reg->imm;
1680 } else if (opcode == BPF_OR && BPF_SRC(insn->code) == BPF_K) {
1681 dst_imm |= insn->imm;
1682 } else if (opcode == BPF_OR && BPF_SRC(insn->code) == BPF_X &&
1683 src_reg->type == CONST_IMM) {
1684 dst_imm |= src_reg->imm;
1685 } else if (opcode == BPF_AND && BPF_SRC(insn->code) == BPF_K) {
1686 dst_imm &= insn->imm;
1687 } else if (opcode == BPF_AND && BPF_SRC(insn->code) == BPF_X &&
1688 src_reg->type == CONST_IMM) {
1689 dst_imm &= src_reg->imm;
1690 } else if (opcode == BPF_RSH && BPF_SRC(insn->code) == BPF_K) {
1691 dst_imm >>= insn->imm;
1692 } else if (opcode == BPF_RSH && BPF_SRC(insn->code) == BPF_X &&
1693 src_reg->type == CONST_IMM) {
1694 dst_imm >>= src_reg->imm;
1695 } else if (opcode == BPF_LSH && BPF_SRC(insn->code) == BPF_K) {
1696 dst_imm <<= insn->imm;
1697 } else if (opcode == BPF_LSH && BPF_SRC(insn->code) == BPF_X &&
1698 src_reg->type == CONST_IMM) {
1699 dst_imm <<= src_reg->imm;
1701 mark_reg_unknown_value(regs, insn->dst_reg);
1705 dst_reg->imm = dst_imm;
1710 static void check_reg_overflow(struct bpf_reg_state *reg)
1712 if (reg->max_value > BPF_REGISTER_MAX_RANGE)
1713 reg->max_value = BPF_REGISTER_MAX_RANGE;
1714 if (reg->min_value < BPF_REGISTER_MIN_RANGE ||
1715 reg->min_value > BPF_REGISTER_MAX_RANGE)
1716 reg->min_value = BPF_REGISTER_MIN_RANGE;
1719 static u32 calc_align(u32 imm)
1723 return imm - ((imm - 1) & imm);
1726 static void adjust_reg_min_max_vals(struct bpf_verifier_env *env,
1727 struct bpf_insn *insn)
1729 struct bpf_reg_state *regs = env->cur_state.regs, *dst_reg;
1730 s64 min_val = BPF_REGISTER_MIN_RANGE;
1731 u64 max_val = BPF_REGISTER_MAX_RANGE;
1732 u8 opcode = BPF_OP(insn->code);
1733 u32 dst_align, src_align;
1735 dst_reg = ®s[insn->dst_reg];
1737 if (BPF_SRC(insn->code) == BPF_X) {
1738 check_reg_overflow(®s[insn->src_reg]);
1739 min_val = regs[insn->src_reg].min_value;
1740 max_val = regs[insn->src_reg].max_value;
1742 /* If the source register is a random pointer then the
1743 * min_value/max_value values represent the range of the known
1744 * accesses into that value, not the actual min/max value of the
1745 * register itself. In this case we have to reset the reg range
1746 * values so we know it is not safe to look at.
1748 if (regs[insn->src_reg].type != CONST_IMM &&
1749 regs[insn->src_reg].type != UNKNOWN_VALUE) {
1750 min_val = BPF_REGISTER_MIN_RANGE;
1751 max_val = BPF_REGISTER_MAX_RANGE;
1754 src_align = regs[insn->src_reg].min_align;
1756 } else if (insn->imm < BPF_REGISTER_MAX_RANGE &&
1757 (s64)insn->imm > BPF_REGISTER_MIN_RANGE) {
1758 min_val = max_val = insn->imm;
1759 src_align = calc_align(insn->imm);
1762 dst_align = dst_reg->min_align;
1764 /* We don't know anything about what was done to this register, mark it
1767 if (min_val == BPF_REGISTER_MIN_RANGE &&
1768 max_val == BPF_REGISTER_MAX_RANGE) {
1769 reset_reg_range_values(regs, insn->dst_reg);
1773 /* If one of our values was at the end of our ranges then we can't just
1774 * do our normal operations to the register, we need to set the values
1775 * to the min/max since they are undefined.
1777 if (min_val == BPF_REGISTER_MIN_RANGE)
1778 dst_reg->min_value = BPF_REGISTER_MIN_RANGE;
1779 if (max_val == BPF_REGISTER_MAX_RANGE)
1780 dst_reg->max_value = BPF_REGISTER_MAX_RANGE;
1784 if (dst_reg->min_value != BPF_REGISTER_MIN_RANGE)
1785 dst_reg->min_value += min_val;
1786 if (dst_reg->max_value != BPF_REGISTER_MAX_RANGE)
1787 dst_reg->max_value += max_val;
1788 dst_reg->min_align = min(src_align, dst_align);
1791 if (dst_reg->min_value != BPF_REGISTER_MIN_RANGE)
1792 dst_reg->min_value -= min_val;
1793 if (dst_reg->max_value != BPF_REGISTER_MAX_RANGE)
1794 dst_reg->max_value -= max_val;
1795 dst_reg->min_align = min(src_align, dst_align);
1798 if (dst_reg->min_value != BPF_REGISTER_MIN_RANGE)
1799 dst_reg->min_value *= min_val;
1800 if (dst_reg->max_value != BPF_REGISTER_MAX_RANGE)
1801 dst_reg->max_value *= max_val;
1802 dst_reg->min_align = max(src_align, dst_align);
1805 /* Disallow AND'ing of negative numbers, ain't nobody got time
1806 * for that. Otherwise the minimum is 0 and the max is the max
1807 * value we could AND against.
1810 dst_reg->min_value = BPF_REGISTER_MIN_RANGE;
1812 dst_reg->min_value = 0;
1813 dst_reg->max_value = max_val;
1814 dst_reg->min_align = max(src_align, dst_align);
1817 /* Gotta have special overflow logic here, if we're shifting
1818 * more than MAX_RANGE then just assume we have an invalid
1821 if (min_val > ilog2(BPF_REGISTER_MAX_RANGE)) {
1822 dst_reg->min_value = BPF_REGISTER_MIN_RANGE;
1823 dst_reg->min_align = 1;
1825 if (dst_reg->min_value != BPF_REGISTER_MIN_RANGE)
1826 dst_reg->min_value <<= min_val;
1827 if (!dst_reg->min_align)
1828 dst_reg->min_align = 1;
1829 dst_reg->min_align <<= min_val;
1831 if (max_val > ilog2(BPF_REGISTER_MAX_RANGE))
1832 dst_reg->max_value = BPF_REGISTER_MAX_RANGE;
1833 else if (dst_reg->max_value != BPF_REGISTER_MAX_RANGE)
1834 dst_reg->max_value <<= max_val;
1837 /* RSH by a negative number is undefined, and the BPF_RSH is an
1838 * unsigned shift, so make the appropriate casts.
1840 if (min_val < 0 || dst_reg->min_value < 0) {
1841 dst_reg->min_value = BPF_REGISTER_MIN_RANGE;
1843 dst_reg->min_value =
1844 (u64)(dst_reg->min_value) >> min_val;
1847 dst_reg->min_align = 1;
1849 dst_reg->min_align >>= (u64) min_val;
1850 if (!dst_reg->min_align)
1851 dst_reg->min_align = 1;
1853 if (dst_reg->max_value != BPF_REGISTER_MAX_RANGE)
1854 dst_reg->max_value >>= max_val;
1857 reset_reg_range_values(regs, insn->dst_reg);
1861 check_reg_overflow(dst_reg);
1864 /* check validity of 32-bit and 64-bit arithmetic operations */
1865 static int check_alu_op(struct bpf_verifier_env *env, struct bpf_insn *insn)
1867 struct bpf_reg_state *regs = env->cur_state.regs, *dst_reg;
1868 u8 opcode = BPF_OP(insn->code);
1871 if (opcode == BPF_END || opcode == BPF_NEG) {
1872 if (opcode == BPF_NEG) {
1873 if (BPF_SRC(insn->code) != 0 ||
1874 insn->src_reg != BPF_REG_0 ||
1875 insn->off != 0 || insn->imm != 0) {
1876 verbose("BPF_NEG uses reserved fields\n");
1880 if (insn->src_reg != BPF_REG_0 || insn->off != 0 ||
1881 (insn->imm != 16 && insn->imm != 32 && insn->imm != 64)) {
1882 verbose("BPF_END uses reserved fields\n");
1887 /* check src operand */
1888 err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
1892 if (is_pointer_value(env, insn->dst_reg)) {
1893 verbose("R%d pointer arithmetic prohibited\n",
1898 /* check dest operand */
1899 err = check_reg_arg(regs, insn->dst_reg, DST_OP);
1903 } else if (opcode == BPF_MOV) {
1905 if (BPF_SRC(insn->code) == BPF_X) {
1906 if (insn->imm != 0 || insn->off != 0) {
1907 verbose("BPF_MOV uses reserved fields\n");
1911 /* check src operand */
1912 err = check_reg_arg(regs, insn->src_reg, SRC_OP);
1916 if (insn->src_reg != BPF_REG_0 || insn->off != 0) {
1917 verbose("BPF_MOV uses reserved fields\n");
1922 /* check dest operand */
1923 err = check_reg_arg(regs, insn->dst_reg, DST_OP);
1927 /* we are setting our register to something new, we need to
1928 * reset its range values.
1930 reset_reg_range_values(regs, insn->dst_reg);
1932 if (BPF_SRC(insn->code) == BPF_X) {
1933 if (BPF_CLASS(insn->code) == BPF_ALU64) {
1935 * copy register state to dest reg
1937 regs[insn->dst_reg] = regs[insn->src_reg];
1939 if (is_pointer_value(env, insn->src_reg)) {
1940 verbose("R%d partial copy of pointer\n",
1944 mark_reg_unknown_value(regs, insn->dst_reg);
1948 * remember the value we stored into this reg
1950 regs[insn->dst_reg].type = CONST_IMM;
1951 regs[insn->dst_reg].imm = insn->imm;
1952 regs[insn->dst_reg].max_value = insn->imm;
1953 regs[insn->dst_reg].min_value = insn->imm;
1954 regs[insn->dst_reg].min_align = calc_align(insn->imm);
1957 } else if (opcode > BPF_END) {
1958 verbose("invalid BPF_ALU opcode %x\n", opcode);
1961 } else { /* all other ALU ops: and, sub, xor, add, ... */
1963 if (BPF_SRC(insn->code) == BPF_X) {
1964 if (insn->imm != 0 || insn->off != 0) {
1965 verbose("BPF_ALU uses reserved fields\n");
1968 /* check src1 operand */
1969 err = check_reg_arg(regs, insn->src_reg, SRC_OP);
1973 if (insn->src_reg != BPF_REG_0 || insn->off != 0) {
1974 verbose("BPF_ALU uses reserved fields\n");
1979 /* check src2 operand */
1980 err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
1984 if ((opcode == BPF_MOD || opcode == BPF_DIV) &&
1985 BPF_SRC(insn->code) == BPF_K && insn->imm == 0) {
1986 verbose("div by zero\n");
1990 if ((opcode == BPF_LSH || opcode == BPF_RSH ||
1991 opcode == BPF_ARSH) && BPF_SRC(insn->code) == BPF_K) {
1992 int size = BPF_CLASS(insn->code) == BPF_ALU64 ? 64 : 32;
1994 if (insn->imm < 0 || insn->imm >= size) {
1995 verbose("invalid shift %d\n", insn->imm);
2000 /* check dest operand */
2001 err = check_reg_arg(regs, insn->dst_reg, DST_OP_NO_MARK);
2005 dst_reg = ®s[insn->dst_reg];
2007 /* first we want to adjust our ranges. */
2008 adjust_reg_min_max_vals(env, insn);
2010 /* pattern match 'bpf_add Rx, imm' instruction */
2011 if (opcode == BPF_ADD && BPF_CLASS(insn->code) == BPF_ALU64 &&
2012 dst_reg->type == FRAME_PTR && BPF_SRC(insn->code) == BPF_K) {
2013 dst_reg->type = PTR_TO_STACK;
2014 dst_reg->imm = insn->imm;
2016 } else if (opcode == BPF_ADD &&
2017 BPF_CLASS(insn->code) == BPF_ALU64 &&
2018 dst_reg->type == PTR_TO_STACK &&
2019 ((BPF_SRC(insn->code) == BPF_X &&
2020 regs[insn->src_reg].type == CONST_IMM) ||
2021 BPF_SRC(insn->code) == BPF_K)) {
2022 if (BPF_SRC(insn->code) == BPF_X)
2023 dst_reg->imm += regs[insn->src_reg].imm;
2025 dst_reg->imm += insn->imm;
2027 } else if (opcode == BPF_ADD &&
2028 BPF_CLASS(insn->code) == BPF_ALU64 &&
2029 (dst_reg->type == PTR_TO_PACKET ||
2030 (BPF_SRC(insn->code) == BPF_X &&
2031 regs[insn->src_reg].type == PTR_TO_PACKET))) {
2032 /* ptr_to_packet += K|X */
2033 return check_packet_ptr_add(env, insn);
2034 } else if (BPF_CLASS(insn->code) == BPF_ALU64 &&
2035 dst_reg->type == UNKNOWN_VALUE &&
2036 env->allow_ptr_leaks) {
2037 /* unknown += K|X */
2038 return evaluate_reg_alu(env, insn);
2039 } else if (BPF_CLASS(insn->code) == BPF_ALU64 &&
2040 dst_reg->type == CONST_IMM &&
2041 env->allow_ptr_leaks) {
2042 /* reg_imm += K|X */
2043 return evaluate_reg_imm_alu(env, insn);
2044 } else if (is_pointer_value(env, insn->dst_reg)) {
2045 verbose("R%d pointer arithmetic prohibited\n",
2048 } else if (BPF_SRC(insn->code) == BPF_X &&
2049 is_pointer_value(env, insn->src_reg)) {
2050 verbose("R%d pointer arithmetic prohibited\n",
2055 /* If we did pointer math on a map value then just set it to our
2056 * PTR_TO_MAP_VALUE_ADJ type so we can deal with any stores or
2057 * loads to this register appropriately, otherwise just mark the
2058 * register as unknown.
2060 if (env->allow_ptr_leaks &&
2061 BPF_CLASS(insn->code) == BPF_ALU64 && opcode == BPF_ADD &&
2062 (dst_reg->type == PTR_TO_MAP_VALUE ||
2063 dst_reg->type == PTR_TO_MAP_VALUE_ADJ))
2064 dst_reg->type = PTR_TO_MAP_VALUE_ADJ;
2066 mark_reg_unknown_value(regs, insn->dst_reg);
2072 static void find_good_pkt_pointers(struct bpf_verifier_state *state,
2073 struct bpf_reg_state *dst_reg)
2075 struct bpf_reg_state *regs = state->regs, *reg;
2078 /* LLVM can generate two kind of checks:
2084 * if (r2 > pkt_end) goto <handle exception>
2088 * r2 == dst_reg, pkt_end == src_reg
2089 * r2=pkt(id=n,off=8,r=0)
2090 * r3=pkt(id=n,off=0,r=0)
2096 * if (pkt_end >= r2) goto <access okay>
2097 * <handle exception>
2100 * pkt_end == dst_reg, r2 == src_reg
2101 * r2=pkt(id=n,off=8,r=0)
2102 * r3=pkt(id=n,off=0,r=0)
2104 * Find register r3 and mark its range as r3=pkt(id=n,off=0,r=8)
2105 * so that range of bytes [r3, r3 + 8) is safe to access.
2108 for (i = 0; i < MAX_BPF_REG; i++)
2109 if (regs[i].type == PTR_TO_PACKET && regs[i].id == dst_reg->id)
2110 /* keep the maximum range already checked */
2111 regs[i].range = max(regs[i].range, dst_reg->off);
2113 for (i = 0; i < MAX_BPF_STACK; i += BPF_REG_SIZE) {
2114 if (state->stack_slot_type[i] != STACK_SPILL)
2116 reg = &state->spilled_regs[i / BPF_REG_SIZE];
2117 if (reg->type == PTR_TO_PACKET && reg->id == dst_reg->id)
2118 reg->range = max(reg->range, dst_reg->off);
2122 /* Adjusts the register min/max values in the case that the dst_reg is the
2123 * variable register that we are working on, and src_reg is a constant or we're
2124 * simply doing a BPF_K check.
2126 static void reg_set_min_max(struct bpf_reg_state *true_reg,
2127 struct bpf_reg_state *false_reg, u64 val,
2132 /* If this is false then we know nothing Jon Snow, but if it is
2133 * true then we know for sure.
2135 true_reg->max_value = true_reg->min_value = val;
2138 /* If this is true we know nothing Jon Snow, but if it is false
2139 * we know the value for sure;
2141 false_reg->max_value = false_reg->min_value = val;
2144 /* Unsigned comparison, the minimum value is 0. */
2145 false_reg->min_value = 0;
2148 /* If this is false then we know the maximum val is val,
2149 * otherwise we know the min val is val+1.
2151 false_reg->max_value = val;
2152 true_reg->min_value = val + 1;
2155 /* Unsigned comparison, the minimum value is 0. */
2156 false_reg->min_value = 0;
2159 /* If this is false then we know the maximum value is val - 1,
2160 * otherwise we know the mimimum value is val.
2162 false_reg->max_value = val - 1;
2163 true_reg->min_value = val;
2169 check_reg_overflow(false_reg);
2170 check_reg_overflow(true_reg);
2173 /* Same as above, but for the case that dst_reg is a CONST_IMM reg and src_reg
2174 * is the variable reg.
2176 static void reg_set_min_max_inv(struct bpf_reg_state *true_reg,
2177 struct bpf_reg_state *false_reg, u64 val,
2182 /* If this is false then we know nothing Jon Snow, but if it is
2183 * true then we know for sure.
2185 true_reg->max_value = true_reg->min_value = val;
2188 /* If this is true we know nothing Jon Snow, but if it is false
2189 * we know the value for sure;
2191 false_reg->max_value = false_reg->min_value = val;
2194 /* Unsigned comparison, the minimum value is 0. */
2195 true_reg->min_value = 0;
2199 * If this is false, then the val is <= the register, if it is
2200 * true the register <= to the val.
2202 false_reg->min_value = val;
2203 true_reg->max_value = val - 1;
2206 /* Unsigned comparison, the minimum value is 0. */
2207 true_reg->min_value = 0;
2210 /* If this is false then constant < register, if it is true then
2211 * the register < constant.
2213 false_reg->min_value = val + 1;
2214 true_reg->max_value = val;
2220 check_reg_overflow(false_reg);
2221 check_reg_overflow(true_reg);
2224 static void mark_map_reg(struct bpf_reg_state *regs, u32 regno, u32 id,
2225 enum bpf_reg_type type)
2227 struct bpf_reg_state *reg = ®s[regno];
2229 if (reg->type == PTR_TO_MAP_VALUE_OR_NULL && reg->id == id) {
2230 if (type == UNKNOWN_VALUE) {
2231 __mark_reg_unknown_value(regs, regno);
2232 } else if (reg->map_ptr->inner_map_meta) {
2233 reg->type = CONST_PTR_TO_MAP;
2234 reg->map_ptr = reg->map_ptr->inner_map_meta;
2238 /* We don't need id from this point onwards anymore, thus we
2239 * should better reset it, so that state pruning has chances
2246 /* The logic is similar to find_good_pkt_pointers(), both could eventually
2247 * be folded together at some point.
2249 static void mark_map_regs(struct bpf_verifier_state *state, u32 regno,
2250 enum bpf_reg_type type)
2252 struct bpf_reg_state *regs = state->regs;
2253 u32 id = regs[regno].id;
2256 for (i = 0; i < MAX_BPF_REG; i++)
2257 mark_map_reg(regs, i, id, type);
2259 for (i = 0; i < MAX_BPF_STACK; i += BPF_REG_SIZE) {
2260 if (state->stack_slot_type[i] != STACK_SPILL)
2262 mark_map_reg(state->spilled_regs, i / BPF_REG_SIZE, id, type);
2266 static int check_cond_jmp_op(struct bpf_verifier_env *env,
2267 struct bpf_insn *insn, int *insn_idx)
2269 struct bpf_verifier_state *other_branch, *this_branch = &env->cur_state;
2270 struct bpf_reg_state *regs = this_branch->regs, *dst_reg;
2271 u8 opcode = BPF_OP(insn->code);
2274 if (opcode > BPF_EXIT) {
2275 verbose("invalid BPF_JMP opcode %x\n", opcode);
2279 if (BPF_SRC(insn->code) == BPF_X) {
2280 if (insn->imm != 0) {
2281 verbose("BPF_JMP uses reserved fields\n");
2285 /* check src1 operand */
2286 err = check_reg_arg(regs, insn->src_reg, SRC_OP);
2290 if (is_pointer_value(env, insn->src_reg)) {
2291 verbose("R%d pointer comparison prohibited\n",
2296 if (insn->src_reg != BPF_REG_0) {
2297 verbose("BPF_JMP uses reserved fields\n");
2302 /* check src2 operand */
2303 err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
2307 dst_reg = ®s[insn->dst_reg];
2309 /* detect if R == 0 where R was initialized to zero earlier */
2310 if (BPF_SRC(insn->code) == BPF_K &&
2311 (opcode == BPF_JEQ || opcode == BPF_JNE) &&
2312 dst_reg->type == CONST_IMM && dst_reg->imm == insn->imm) {
2313 if (opcode == BPF_JEQ) {
2314 /* if (imm == imm) goto pc+off;
2315 * only follow the goto, ignore fall-through
2317 *insn_idx += insn->off;
2320 /* if (imm != imm) goto pc+off;
2321 * only follow fall-through branch, since
2322 * that's where the program will go
2328 other_branch = push_stack(env, *insn_idx + insn->off + 1, *insn_idx);
2332 /* detect if we are comparing against a constant value so we can adjust
2333 * our min/max values for our dst register.
2335 if (BPF_SRC(insn->code) == BPF_X) {
2336 if (regs[insn->src_reg].type == CONST_IMM)
2337 reg_set_min_max(&other_branch->regs[insn->dst_reg],
2338 dst_reg, regs[insn->src_reg].imm,
2340 else if (dst_reg->type == CONST_IMM)
2341 reg_set_min_max_inv(&other_branch->regs[insn->src_reg],
2342 ®s[insn->src_reg], dst_reg->imm,
2345 reg_set_min_max(&other_branch->regs[insn->dst_reg],
2346 dst_reg, insn->imm, opcode);
2349 /* detect if R == 0 where R is returned from bpf_map_lookup_elem() */
2350 if (BPF_SRC(insn->code) == BPF_K &&
2351 insn->imm == 0 && (opcode == BPF_JEQ || opcode == BPF_JNE) &&
2352 dst_reg->type == PTR_TO_MAP_VALUE_OR_NULL) {
2353 /* Mark all identical map registers in each branch as either
2354 * safe or unknown depending R == 0 or R != 0 conditional.
2356 mark_map_regs(this_branch, insn->dst_reg,
2357 opcode == BPF_JEQ ? PTR_TO_MAP_VALUE : UNKNOWN_VALUE);
2358 mark_map_regs(other_branch, insn->dst_reg,
2359 opcode == BPF_JEQ ? UNKNOWN_VALUE : PTR_TO_MAP_VALUE);
2360 } else if (BPF_SRC(insn->code) == BPF_X && opcode == BPF_JGT &&
2361 dst_reg->type == PTR_TO_PACKET &&
2362 regs[insn->src_reg].type == PTR_TO_PACKET_END) {
2363 find_good_pkt_pointers(this_branch, dst_reg);
2364 } else if (BPF_SRC(insn->code) == BPF_X && opcode == BPF_JGE &&
2365 dst_reg->type == PTR_TO_PACKET_END &&
2366 regs[insn->src_reg].type == PTR_TO_PACKET) {
2367 find_good_pkt_pointers(other_branch, ®s[insn->src_reg]);
2368 } else if (is_pointer_value(env, insn->dst_reg)) {
2369 verbose("R%d pointer comparison prohibited\n", insn->dst_reg);
2373 print_verifier_state(this_branch);
2377 /* return the map pointer stored inside BPF_LD_IMM64 instruction */
2378 static struct bpf_map *ld_imm64_to_map_ptr(struct bpf_insn *insn)
2380 u64 imm64 = ((u64) (u32) insn[0].imm) | ((u64) (u32) insn[1].imm) << 32;
2382 return (struct bpf_map *) (unsigned long) imm64;
2385 /* verify BPF_LD_IMM64 instruction */
2386 static int check_ld_imm(struct bpf_verifier_env *env, struct bpf_insn *insn)
2388 struct bpf_reg_state *regs = env->cur_state.regs;
2391 if (BPF_SIZE(insn->code) != BPF_DW) {
2392 verbose("invalid BPF_LD_IMM insn\n");
2395 if (insn->off != 0) {
2396 verbose("BPF_LD_IMM64 uses reserved fields\n");
2400 err = check_reg_arg(regs, insn->dst_reg, DST_OP);
2404 if (insn->src_reg == 0) {
2405 u64 imm = ((u64)(insn + 1)->imm << 32) | (u32)insn->imm;
2407 regs[insn->dst_reg].type = CONST_IMM;
2408 regs[insn->dst_reg].imm = imm;
2412 /* replace_map_fd_with_map_ptr() should have caught bad ld_imm64 */
2413 BUG_ON(insn->src_reg != BPF_PSEUDO_MAP_FD);
2415 regs[insn->dst_reg].type = CONST_PTR_TO_MAP;
2416 regs[insn->dst_reg].map_ptr = ld_imm64_to_map_ptr(insn);
2420 static bool may_access_skb(enum bpf_prog_type type)
2423 case BPF_PROG_TYPE_SOCKET_FILTER:
2424 case BPF_PROG_TYPE_SCHED_CLS:
2425 case BPF_PROG_TYPE_SCHED_ACT:
2432 /* verify safety of LD_ABS|LD_IND instructions:
2433 * - they can only appear in the programs where ctx == skb
2434 * - since they are wrappers of function calls, they scratch R1-R5 registers,
2435 * preserve R6-R9, and store return value into R0
2438 * ctx == skb == R6 == CTX
2441 * SRC == any register
2442 * IMM == 32-bit immediate
2445 * R0 - 8/16/32-bit skb data converted to cpu endianness
2447 static int check_ld_abs(struct bpf_verifier_env *env, struct bpf_insn *insn)
2449 struct bpf_reg_state *regs = env->cur_state.regs;
2450 u8 mode = BPF_MODE(insn->code);
2451 struct bpf_reg_state *reg;
2454 if (!may_access_skb(env->prog->type)) {
2455 verbose("BPF_LD_[ABS|IND] instructions not allowed for this program type\n");
2459 if (insn->dst_reg != BPF_REG_0 || insn->off != 0 ||
2460 BPF_SIZE(insn->code) == BPF_DW ||
2461 (mode == BPF_ABS && insn->src_reg != BPF_REG_0)) {
2462 verbose("BPF_LD_[ABS|IND] uses reserved fields\n");
2466 /* check whether implicit source operand (register R6) is readable */
2467 err = check_reg_arg(regs, BPF_REG_6, SRC_OP);
2471 if (regs[BPF_REG_6].type != PTR_TO_CTX) {
2472 verbose("at the time of BPF_LD_ABS|IND R6 != pointer to skb\n");
2476 if (mode == BPF_IND) {
2477 /* check explicit source operand */
2478 err = check_reg_arg(regs, insn->src_reg, SRC_OP);
2483 /* reset caller saved regs to unreadable */
2484 for (i = 0; i < CALLER_SAVED_REGS; i++) {
2485 reg = regs + caller_saved[i];
2486 reg->type = NOT_INIT;
2490 /* mark destination R0 register as readable, since it contains
2491 * the value fetched from the packet
2493 regs[BPF_REG_0].type = UNKNOWN_VALUE;
2497 /* non-recursive DFS pseudo code
2498 * 1 procedure DFS-iterative(G,v):
2499 * 2 label v as discovered
2500 * 3 let S be a stack
2502 * 5 while S is not empty
2504 * 7 if t is what we're looking for:
2506 * 9 for all edges e in G.adjacentEdges(t) do
2507 * 10 if edge e is already labelled
2508 * 11 continue with the next edge
2509 * 12 w <- G.adjacentVertex(t,e)
2510 * 13 if vertex w is not discovered and not explored
2511 * 14 label e as tree-edge
2512 * 15 label w as discovered
2515 * 18 else if vertex w is discovered
2516 * 19 label e as back-edge
2518 * 21 // vertex w is explored
2519 * 22 label e as forward- or cross-edge
2520 * 23 label t as explored
2525 * 0x11 - discovered and fall-through edge labelled
2526 * 0x12 - discovered and fall-through and branch edges labelled
2537 #define STATE_LIST_MARK ((struct bpf_verifier_state_list *) -1L)
2539 static int *insn_stack; /* stack of insns to process */
2540 static int cur_stack; /* current stack index */
2541 static int *insn_state;
2543 /* t, w, e - match pseudo-code above:
2544 * t - index of current instruction
2545 * w - next instruction
2548 static int push_insn(int t, int w, int e, struct bpf_verifier_env *env)
2550 if (e == FALLTHROUGH && insn_state[t] >= (DISCOVERED | FALLTHROUGH))
2553 if (e == BRANCH && insn_state[t] >= (DISCOVERED | BRANCH))
2556 if (w < 0 || w >= env->prog->len) {
2557 verbose("jump out of range from insn %d to %d\n", t, w);
2562 /* mark branch target for state pruning */
2563 env->explored_states[w] = STATE_LIST_MARK;
2565 if (insn_state[w] == 0) {
2567 insn_state[t] = DISCOVERED | e;
2568 insn_state[w] = DISCOVERED;
2569 if (cur_stack >= env->prog->len)
2571 insn_stack[cur_stack++] = w;
2573 } else if ((insn_state[w] & 0xF0) == DISCOVERED) {
2574 verbose("back-edge from insn %d to %d\n", t, w);
2576 } else if (insn_state[w] == EXPLORED) {
2577 /* forward- or cross-edge */
2578 insn_state[t] = DISCOVERED | e;
2580 verbose("insn state internal bug\n");
2586 /* non-recursive depth-first-search to detect loops in BPF program
2587 * loop == back-edge in directed graph
2589 static int check_cfg(struct bpf_verifier_env *env)
2591 struct bpf_insn *insns = env->prog->insnsi;
2592 int insn_cnt = env->prog->len;
2596 insn_state = kcalloc(insn_cnt, sizeof(int), GFP_KERNEL);
2600 insn_stack = kcalloc(insn_cnt, sizeof(int), GFP_KERNEL);
2606 insn_state[0] = DISCOVERED; /* mark 1st insn as discovered */
2607 insn_stack[0] = 0; /* 0 is the first instruction */
2613 t = insn_stack[cur_stack - 1];
2615 if (BPF_CLASS(insns[t].code) == BPF_JMP) {
2616 u8 opcode = BPF_OP(insns[t].code);
2618 if (opcode == BPF_EXIT) {
2620 } else if (opcode == BPF_CALL) {
2621 ret = push_insn(t, t + 1, FALLTHROUGH, env);
2626 if (t + 1 < insn_cnt)
2627 env->explored_states[t + 1] = STATE_LIST_MARK;
2628 } else if (opcode == BPF_JA) {
2629 if (BPF_SRC(insns[t].code) != BPF_K) {
2633 /* unconditional jump with single edge */
2634 ret = push_insn(t, t + insns[t].off + 1,
2640 /* tell verifier to check for equivalent states
2641 * after every call and jump
2643 if (t + 1 < insn_cnt)
2644 env->explored_states[t + 1] = STATE_LIST_MARK;
2646 /* conditional jump with two edges */
2647 env->explored_states[t] = STATE_LIST_MARK;
2648 ret = push_insn(t, t + 1, FALLTHROUGH, env);
2654 ret = push_insn(t, t + insns[t].off + 1, BRANCH, env);
2661 /* all other non-branch instructions with single
2664 ret = push_insn(t, t + 1, FALLTHROUGH, env);
2672 insn_state[t] = EXPLORED;
2673 if (cur_stack-- <= 0) {
2674 verbose("pop stack internal bug\n");
2681 for (i = 0; i < insn_cnt; i++) {
2682 if (insn_state[i] != EXPLORED) {
2683 verbose("unreachable insn %d\n", i);
2688 ret = 0; /* cfg looks good */
2696 /* the following conditions reduce the number of explored insns
2697 * from ~140k to ~80k for ultra large programs that use a lot of ptr_to_packet
2699 static bool compare_ptrs_to_packet(struct bpf_reg_state *old,
2700 struct bpf_reg_state *cur)
2702 if (old->id != cur->id)
2705 /* old ptr_to_packet is more conservative, since it allows smaller
2707 * old(off=0,r=10) is equal to cur(off=0,r=20), because
2708 * old(off=0,r=10) means that with range=10 the verifier proceeded
2709 * further and found no issues with the program. Now we're in the same
2710 * spot with cur(off=0,r=20), so we're safe too, since anything further
2711 * will only be looking at most 10 bytes after this pointer.
2713 if (old->off == cur->off && old->range < cur->range)
2716 /* old(off=20,r=10) is equal to cur(off=22,re=22 or 5 or 0)
2717 * since both cannot be used for packet access and safe(old)
2718 * pointer has smaller off that could be used for further
2719 * 'if (ptr > data_end)' check
2721 * old(off=20,r=10) and cur(off=22,r=22) and cur(off=22,r=0) mean
2722 * that we cannot access the packet.
2723 * The safe range is:
2724 * [ptr, ptr + range - off)
2725 * so whenever off >=range, it means no safe bytes from this pointer.
2726 * When comparing old->off <= cur->off, it means that older code
2727 * went with smaller offset and that offset was later
2728 * used to figure out the safe range after 'if (ptr > data_end)' check
2729 * Say, 'old' state was explored like:
2730 * ... R3(off=0, r=0)
2732 * ... now R4(off=20,r=0) <-- here
2733 * if (R4 > data_end)
2734 * ... R4(off=20,r=20), R3(off=0,r=20) and R3 can be used to access.
2735 * ... the code further went all the way to bpf_exit.
2736 * Now the 'cur' state at the mark 'here' has R4(off=30,r=0).
2737 * old_R4(off=20,r=0) equal to cur_R4(off=30,r=0), since if the verifier
2738 * goes further, such cur_R4 will give larger safe packet range after
2739 * 'if (R4 > data_end)' and all further insn were already good with r=20,
2740 * so they will be good with r=30 and we can prune the search.
2742 if (old->off <= cur->off &&
2743 old->off >= old->range && cur->off >= cur->range)
2749 /* compare two verifier states
2751 * all states stored in state_list are known to be valid, since
2752 * verifier reached 'bpf_exit' instruction through them
2754 * this function is called when verifier exploring different branches of
2755 * execution popped from the state stack. If it sees an old state that has
2756 * more strict register state and more strict stack state then this execution
2757 * branch doesn't need to be explored further, since verifier already
2758 * concluded that more strict state leads to valid finish.
2760 * Therefore two states are equivalent if register state is more conservative
2761 * and explored stack state is more conservative than the current one.
2764 * (slot1=INV slot2=MISC) == (slot1=MISC slot2=MISC)
2765 * (slot1=MISC slot2=MISC) != (slot1=INV slot2=MISC)
2767 * In other words if current stack state (one being explored) has more
2768 * valid slots than old one that already passed validation, it means
2769 * the verifier can stop exploring and conclude that current state is valid too
2771 * Similarly with registers. If explored state has register type as invalid
2772 * whereas register type in current state is meaningful, it means that
2773 * the current state will reach 'bpf_exit' instruction safely
2775 static bool states_equal(struct bpf_verifier_env *env,
2776 struct bpf_verifier_state *old,
2777 struct bpf_verifier_state *cur)
2779 bool varlen_map_access = env->varlen_map_value_access;
2780 struct bpf_reg_state *rold, *rcur;
2783 for (i = 0; i < MAX_BPF_REG; i++) {
2784 rold = &old->regs[i];
2785 rcur = &cur->regs[i];
2787 if (memcmp(rold, rcur, sizeof(*rold)) == 0)
2790 /* If the ranges were not the same, but everything else was and
2791 * we didn't do a variable access into a map then we are a-ok.
2793 if (!varlen_map_access &&
2794 memcmp(rold, rcur, offsetofend(struct bpf_reg_state, id)) == 0)
2797 /* If we didn't map access then again we don't care about the
2798 * mismatched range values and it's ok if our old type was
2799 * UNKNOWN and we didn't go to a NOT_INIT'ed reg.
2801 if (rold->type == NOT_INIT ||
2802 (!varlen_map_access && rold->type == UNKNOWN_VALUE &&
2803 rcur->type != NOT_INIT))
2806 /* Don't care about the reg->id in this case. */
2807 if (rold->type == PTR_TO_MAP_VALUE_OR_NULL &&
2808 rcur->type == PTR_TO_MAP_VALUE_OR_NULL &&
2809 rold->map_ptr == rcur->map_ptr)
2812 if (rold->type == PTR_TO_PACKET && rcur->type == PTR_TO_PACKET &&
2813 compare_ptrs_to_packet(rold, rcur))
2819 for (i = 0; i < MAX_BPF_STACK; i++) {
2820 if (old->stack_slot_type[i] == STACK_INVALID)
2822 if (old->stack_slot_type[i] != cur->stack_slot_type[i])
2823 /* Ex: old explored (safe) state has STACK_SPILL in
2824 * this stack slot, but current has has STACK_MISC ->
2825 * this verifier states are not equivalent,
2826 * return false to continue verification of this path
2829 if (i % BPF_REG_SIZE)
2831 if (memcmp(&old->spilled_regs[i / BPF_REG_SIZE],
2832 &cur->spilled_regs[i / BPF_REG_SIZE],
2833 sizeof(old->spilled_regs[0])))
2834 /* when explored and current stack slot types are
2835 * the same, check that stored pointers types
2836 * are the same as well.
2837 * Ex: explored safe path could have stored
2838 * (bpf_reg_state) {.type = PTR_TO_STACK, .imm = -8}
2839 * but current path has stored:
2840 * (bpf_reg_state) {.type = PTR_TO_STACK, .imm = -16}
2841 * such verifier states are not equivalent.
2842 * return false to continue verification of this path
2851 static int is_state_visited(struct bpf_verifier_env *env, int insn_idx)
2853 struct bpf_verifier_state_list *new_sl;
2854 struct bpf_verifier_state_list *sl;
2856 sl = env->explored_states[insn_idx];
2858 /* this 'insn_idx' instruction wasn't marked, so we will not
2859 * be doing state search here
2863 while (sl != STATE_LIST_MARK) {
2864 if (states_equal(env, &sl->state, &env->cur_state))
2865 /* reached equivalent register/stack state,
2872 /* there were no equivalent states, remember current one.
2873 * technically the current state is not proven to be safe yet,
2874 * but it will either reach bpf_exit (which means it's safe) or
2875 * it will be rejected. Since there are no loops, we won't be
2876 * seeing this 'insn_idx' instruction again on the way to bpf_exit
2878 new_sl = kmalloc(sizeof(struct bpf_verifier_state_list), GFP_USER);
2882 /* add new state to the head of linked list */
2883 memcpy(&new_sl->state, &env->cur_state, sizeof(env->cur_state));
2884 new_sl->next = env->explored_states[insn_idx];
2885 env->explored_states[insn_idx] = new_sl;
2889 static int ext_analyzer_insn_hook(struct bpf_verifier_env *env,
2890 int insn_idx, int prev_insn_idx)
2892 if (!env->analyzer_ops || !env->analyzer_ops->insn_hook)
2895 return env->analyzer_ops->insn_hook(env, insn_idx, prev_insn_idx);
2898 static int do_check(struct bpf_verifier_env *env)
2900 struct bpf_verifier_state *state = &env->cur_state;
2901 struct bpf_insn *insns = env->prog->insnsi;
2902 struct bpf_reg_state *regs = state->regs;
2903 int insn_cnt = env->prog->len;
2904 int insn_idx, prev_insn_idx = 0;
2905 int insn_processed = 0;
2906 bool do_print_state = false;
2908 init_reg_state(regs);
2910 env->varlen_map_value_access = false;
2912 struct bpf_insn *insn;
2916 if (insn_idx >= insn_cnt) {
2917 verbose("invalid insn idx %d insn_cnt %d\n",
2918 insn_idx, insn_cnt);
2922 insn = &insns[insn_idx];
2923 class = BPF_CLASS(insn->code);
2925 if (++insn_processed > BPF_COMPLEXITY_LIMIT_INSNS) {
2926 verbose("BPF program is too large. Processed %d insn\n",
2931 err = is_state_visited(env, insn_idx);
2935 /* found equivalent state, can prune the search */
2938 verbose("\nfrom %d to %d: safe\n",
2939 prev_insn_idx, insn_idx);
2941 verbose("%d: safe\n", insn_idx);
2943 goto process_bpf_exit;
2949 if (log_level > 1 || (log_level && do_print_state)) {
2951 verbose("%d:", insn_idx);
2953 verbose("\nfrom %d to %d:",
2954 prev_insn_idx, insn_idx);
2955 print_verifier_state(&env->cur_state);
2956 do_print_state = false;
2960 verbose("%d: ", insn_idx);
2961 print_bpf_insn(env, insn);
2964 err = ext_analyzer_insn_hook(env, insn_idx, prev_insn_idx);
2968 if (class == BPF_ALU || class == BPF_ALU64) {
2969 err = check_alu_op(env, insn);
2973 } else if (class == BPF_LDX) {
2974 enum bpf_reg_type *prev_src_type, src_reg_type;
2976 /* check for reserved fields is already done */
2978 /* check src operand */
2979 err = check_reg_arg(regs, insn->src_reg, SRC_OP);
2983 err = check_reg_arg(regs, insn->dst_reg, DST_OP_NO_MARK);
2987 src_reg_type = regs[insn->src_reg].type;
2989 /* check that memory (src_reg + off) is readable,
2990 * the state of dst_reg will be updated by this func
2992 err = check_mem_access(env, insn->src_reg, insn->off,
2993 BPF_SIZE(insn->code), BPF_READ,
2998 if (BPF_SIZE(insn->code) != BPF_W &&
2999 BPF_SIZE(insn->code) != BPF_DW) {
3004 prev_src_type = &env->insn_aux_data[insn_idx].ptr_type;
3006 if (*prev_src_type == NOT_INIT) {
3008 * dst_reg = *(u32 *)(src_reg + off)
3009 * save type to validate intersecting paths
3011 *prev_src_type = src_reg_type;
3013 } else if (src_reg_type != *prev_src_type &&
3014 (src_reg_type == PTR_TO_CTX ||
3015 *prev_src_type == PTR_TO_CTX)) {
3016 /* ABuser program is trying to use the same insn
3017 * dst_reg = *(u32*) (src_reg + off)
3018 * with different pointer types:
3019 * src_reg == ctx in one branch and
3020 * src_reg == stack|map in some other branch.
3023 verbose("same insn cannot be used with different pointers\n");
3027 } else if (class == BPF_STX) {
3028 enum bpf_reg_type *prev_dst_type, dst_reg_type;
3030 if (BPF_MODE(insn->code) == BPF_XADD) {
3031 err = check_xadd(env, insn);
3038 /* check src1 operand */
3039 err = check_reg_arg(regs, insn->src_reg, SRC_OP);
3042 /* check src2 operand */
3043 err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
3047 dst_reg_type = regs[insn->dst_reg].type;
3049 /* check that memory (dst_reg + off) is writeable */
3050 err = check_mem_access(env, insn->dst_reg, insn->off,
3051 BPF_SIZE(insn->code), BPF_WRITE,
3056 prev_dst_type = &env->insn_aux_data[insn_idx].ptr_type;
3058 if (*prev_dst_type == NOT_INIT) {
3059 *prev_dst_type = dst_reg_type;
3060 } else if (dst_reg_type != *prev_dst_type &&
3061 (dst_reg_type == PTR_TO_CTX ||
3062 *prev_dst_type == PTR_TO_CTX)) {
3063 verbose("same insn cannot be used with different pointers\n");
3067 } else if (class == BPF_ST) {
3068 if (BPF_MODE(insn->code) != BPF_MEM ||
3069 insn->src_reg != BPF_REG_0) {
3070 verbose("BPF_ST uses reserved fields\n");
3073 /* check src operand */
3074 err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
3078 /* check that memory (dst_reg + off) is writeable */
3079 err = check_mem_access(env, insn->dst_reg, insn->off,
3080 BPF_SIZE(insn->code), BPF_WRITE,
3085 } else if (class == BPF_JMP) {
3086 u8 opcode = BPF_OP(insn->code);
3088 if (opcode == BPF_CALL) {
3089 if (BPF_SRC(insn->code) != BPF_K ||
3091 insn->src_reg != BPF_REG_0 ||
3092 insn->dst_reg != BPF_REG_0) {
3093 verbose("BPF_CALL uses reserved fields\n");
3097 err = check_call(env, insn->imm, insn_idx);
3101 } else if (opcode == BPF_JA) {
3102 if (BPF_SRC(insn->code) != BPF_K ||
3104 insn->src_reg != BPF_REG_0 ||
3105 insn->dst_reg != BPF_REG_0) {
3106 verbose("BPF_JA uses reserved fields\n");
3110 insn_idx += insn->off + 1;
3113 } else if (opcode == BPF_EXIT) {
3114 if (BPF_SRC(insn->code) != BPF_K ||
3116 insn->src_reg != BPF_REG_0 ||
3117 insn->dst_reg != BPF_REG_0) {
3118 verbose("BPF_EXIT uses reserved fields\n");
3122 /* eBPF calling convetion is such that R0 is used
3123 * to return the value from eBPF program.
3124 * Make sure that it's readable at this time
3125 * of bpf_exit, which means that program wrote
3126 * something into it earlier
3128 err = check_reg_arg(regs, BPF_REG_0, SRC_OP);
3132 if (is_pointer_value(env, BPF_REG_0)) {
3133 verbose("R0 leaks addr as return value\n");
3138 insn_idx = pop_stack(env, &prev_insn_idx);
3142 do_print_state = true;
3146 err = check_cond_jmp_op(env, insn, &insn_idx);
3150 } else if (class == BPF_LD) {
3151 u8 mode = BPF_MODE(insn->code);
3153 if (mode == BPF_ABS || mode == BPF_IND) {
3154 err = check_ld_abs(env, insn);
3158 } else if (mode == BPF_IMM) {
3159 err = check_ld_imm(env, insn);
3165 verbose("invalid BPF_LD mode\n");
3168 reset_reg_range_values(regs, insn->dst_reg);
3170 verbose("unknown insn class %d\n", class);
3177 verbose("processed %d insns\n", insn_processed);
3181 static int check_map_prealloc(struct bpf_map *map)
3183 return (map->map_type != BPF_MAP_TYPE_HASH &&
3184 map->map_type != BPF_MAP_TYPE_PERCPU_HASH &&
3185 map->map_type != BPF_MAP_TYPE_HASH_OF_MAPS) ||
3186 !(map->map_flags & BPF_F_NO_PREALLOC);
3189 static int check_map_prog_compatibility(struct bpf_map *map,
3190 struct bpf_prog *prog)
3193 /* Make sure that BPF_PROG_TYPE_PERF_EVENT programs only use
3194 * preallocated hash maps, since doing memory allocation
3195 * in overflow_handler can crash depending on where nmi got
3198 if (prog->type == BPF_PROG_TYPE_PERF_EVENT) {
3199 if (!check_map_prealloc(map)) {
3200 verbose("perf_event programs can only use preallocated hash map\n");
3203 if (map->inner_map_meta &&
3204 !check_map_prealloc(map->inner_map_meta)) {
3205 verbose("perf_event programs can only use preallocated inner hash map\n");
3212 /* look for pseudo eBPF instructions that access map FDs and
3213 * replace them with actual map pointers
3215 static int replace_map_fd_with_map_ptr(struct bpf_verifier_env *env)
3217 struct bpf_insn *insn = env->prog->insnsi;
3218 int insn_cnt = env->prog->len;
3221 err = bpf_prog_calc_tag(env->prog);
3225 for (i = 0; i < insn_cnt; i++, insn++) {
3226 if (BPF_CLASS(insn->code) == BPF_LDX &&
3227 (BPF_MODE(insn->code) != BPF_MEM || insn->imm != 0)) {
3228 verbose("BPF_LDX uses reserved fields\n");
3232 if (BPF_CLASS(insn->code) == BPF_STX &&
3233 ((BPF_MODE(insn->code) != BPF_MEM &&
3234 BPF_MODE(insn->code) != BPF_XADD) || insn->imm != 0)) {
3235 verbose("BPF_STX uses reserved fields\n");
3239 if (insn[0].code == (BPF_LD | BPF_IMM | BPF_DW)) {
3240 struct bpf_map *map;
3243 if (i == insn_cnt - 1 || insn[1].code != 0 ||
3244 insn[1].dst_reg != 0 || insn[1].src_reg != 0 ||
3246 verbose("invalid bpf_ld_imm64 insn\n");
3250 if (insn->src_reg == 0)
3251 /* valid generic load 64-bit imm */
3254 if (insn->src_reg != BPF_PSEUDO_MAP_FD) {
3255 verbose("unrecognized bpf_ld_imm64 insn\n");
3259 f = fdget(insn->imm);
3260 map = __bpf_map_get(f);
3262 verbose("fd %d is not pointing to valid bpf_map\n",
3264 return PTR_ERR(map);
3267 err = check_map_prog_compatibility(map, env->prog);
3273 /* store map pointer inside BPF_LD_IMM64 instruction */
3274 insn[0].imm = (u32) (unsigned long) map;
3275 insn[1].imm = ((u64) (unsigned long) map) >> 32;
3277 /* check whether we recorded this map already */
3278 for (j = 0; j < env->used_map_cnt; j++)
3279 if (env->used_maps[j] == map) {
3284 if (env->used_map_cnt >= MAX_USED_MAPS) {
3289 /* hold the map. If the program is rejected by verifier,
3290 * the map will be released by release_maps() or it
3291 * will be used by the valid program until it's unloaded
3292 * and all maps are released in free_bpf_prog_info()
3294 map = bpf_map_inc(map, false);
3297 return PTR_ERR(map);
3299 env->used_maps[env->used_map_cnt++] = map;
3308 /* now all pseudo BPF_LD_IMM64 instructions load valid
3309 * 'struct bpf_map *' into a register instead of user map_fd.
3310 * These pointers will be used later by verifier to validate map access.
3315 /* drop refcnt of maps used by the rejected program */
3316 static void release_maps(struct bpf_verifier_env *env)
3320 for (i = 0; i < env->used_map_cnt; i++)
3321 bpf_map_put(env->used_maps[i]);
3324 /* convert pseudo BPF_LD_IMM64 into generic BPF_LD_IMM64 */
3325 static void convert_pseudo_ld_imm64(struct bpf_verifier_env *env)
3327 struct bpf_insn *insn = env->prog->insnsi;
3328 int insn_cnt = env->prog->len;
3331 for (i = 0; i < insn_cnt; i++, insn++)
3332 if (insn->code == (BPF_LD | BPF_IMM | BPF_DW))
3336 /* single env->prog->insni[off] instruction was replaced with the range
3337 * insni[off, off + cnt). Adjust corresponding insn_aux_data by copying
3338 * [0, off) and [off, end) to new locations, so the patched range stays zero
3340 static int adjust_insn_aux_data(struct bpf_verifier_env *env, u32 prog_len,
3343 struct bpf_insn_aux_data *new_data, *old_data = env->insn_aux_data;
3347 new_data = vzalloc(sizeof(struct bpf_insn_aux_data) * prog_len);
3350 memcpy(new_data, old_data, sizeof(struct bpf_insn_aux_data) * off);
3351 memcpy(new_data + off + cnt - 1, old_data + off,
3352 sizeof(struct bpf_insn_aux_data) * (prog_len - off - cnt + 1));
3353 env->insn_aux_data = new_data;
3358 static struct bpf_prog *bpf_patch_insn_data(struct bpf_verifier_env *env, u32 off,
3359 const struct bpf_insn *patch, u32 len)
3361 struct bpf_prog *new_prog;
3363 new_prog = bpf_patch_insn_single(env->prog, off, patch, len);
3366 if (adjust_insn_aux_data(env, new_prog->len, off, len))
3371 /* convert load instructions that access fields of 'struct __sk_buff'
3372 * into sequence of instructions that access fields of 'struct sk_buff'
3374 static int convert_ctx_accesses(struct bpf_verifier_env *env)
3376 const struct bpf_verifier_ops *ops = env->prog->aux->ops;
3377 const int insn_cnt = env->prog->len;
3378 struct bpf_insn insn_buf[16], *insn;
3379 struct bpf_prog *new_prog;
3380 enum bpf_access_type type;
3381 int i, cnt, delta = 0;
3383 if (ops->gen_prologue) {
3384 cnt = ops->gen_prologue(insn_buf, env->seen_direct_write,
3386 if (cnt >= ARRAY_SIZE(insn_buf)) {
3387 verbose("bpf verifier is misconfigured\n");
3390 new_prog = bpf_patch_insn_data(env, 0, insn_buf, cnt);
3394 env->prog = new_prog;
3399 if (!ops->convert_ctx_access)
3402 insn = env->prog->insnsi + delta;
3404 for (i = 0; i < insn_cnt; i++, insn++) {
3405 if (insn->code == (BPF_LDX | BPF_MEM | BPF_B) ||
3406 insn->code == (BPF_LDX | BPF_MEM | BPF_H) ||
3407 insn->code == (BPF_LDX | BPF_MEM | BPF_W) ||
3408 insn->code == (BPF_LDX | BPF_MEM | BPF_DW))
3410 else if (insn->code == (BPF_STX | BPF_MEM | BPF_B) ||
3411 insn->code == (BPF_STX | BPF_MEM | BPF_H) ||
3412 insn->code == (BPF_STX | BPF_MEM | BPF_W) ||
3413 insn->code == (BPF_STX | BPF_MEM | BPF_DW))
3418 if (env->insn_aux_data[i + delta].ptr_type != PTR_TO_CTX)
3421 cnt = ops->convert_ctx_access(type, insn, insn_buf, env->prog);
3422 if (cnt == 0 || cnt >= ARRAY_SIZE(insn_buf)) {
3423 verbose("bpf verifier is misconfigured\n");
3427 new_prog = bpf_patch_insn_data(env, i + delta, insn_buf, cnt);
3433 /* keep walking new program and skip insns we just inserted */
3434 env->prog = new_prog;
3435 insn = new_prog->insnsi + i + delta;
3441 /* fixup insn->imm field of bpf_call instructions
3442 * and inline eligible helpers as explicit sequence of BPF instructions
3444 * this function is called after eBPF program passed verification
3446 static int fixup_bpf_calls(struct bpf_verifier_env *env)
3448 struct bpf_prog *prog = env->prog;
3449 struct bpf_insn *insn = prog->insnsi;
3450 const struct bpf_func_proto *fn;
3451 const int insn_cnt = prog->len;
3452 struct bpf_insn insn_buf[16];
3453 struct bpf_prog *new_prog;
3454 struct bpf_map *map_ptr;
3455 int i, cnt, delta = 0;
3457 for (i = 0; i < insn_cnt; i++, insn++) {
3458 if (insn->code != (BPF_JMP | BPF_CALL))
3461 if (insn->imm == BPF_FUNC_get_route_realm)
3462 prog->dst_needed = 1;
3463 if (insn->imm == BPF_FUNC_get_prandom_u32)
3464 bpf_user_rnd_init_once();
3465 if (insn->imm == BPF_FUNC_tail_call) {
3466 /* If we tail call into other programs, we
3467 * cannot make any assumptions since they can
3468 * be replaced dynamically during runtime in
3469 * the program array.
3471 prog->cb_access = 1;
3473 /* mark bpf_tail_call as different opcode to avoid
3474 * conditional branch in the interpeter for every normal
3475 * call and to prevent accidental JITing by JIT compiler
3476 * that doesn't support bpf_tail_call yet
3479 insn->code |= BPF_X;
3483 if (ebpf_jit_enabled() && insn->imm == BPF_FUNC_map_lookup_elem) {
3484 map_ptr = env->insn_aux_data[i + delta].map_ptr;
3485 if (map_ptr == BPF_MAP_PTR_POISON ||
3486 !map_ptr->ops->map_gen_lookup)
3487 goto patch_call_imm;
3489 cnt = map_ptr->ops->map_gen_lookup(map_ptr, insn_buf);
3490 if (cnt == 0 || cnt >= ARRAY_SIZE(insn_buf)) {
3491 verbose("bpf verifier is misconfigured\n");
3495 new_prog = bpf_patch_insn_data(env, i + delta, insn_buf,
3502 /* keep walking new program and skip insns we just inserted */
3503 env->prog = prog = new_prog;
3504 insn = new_prog->insnsi + i + delta;
3509 fn = prog->aux->ops->get_func_proto(insn->imm);
3510 /* all functions that have prototype and verifier allowed
3511 * programs to call them, must be real in-kernel functions
3514 verbose("kernel subsystem misconfigured func %s#%d\n",
3515 func_id_name(insn->imm), insn->imm);
3518 insn->imm = fn->func - __bpf_call_base;
3524 static void free_states(struct bpf_verifier_env *env)
3526 struct bpf_verifier_state_list *sl, *sln;
3529 if (!env->explored_states)
3532 for (i = 0; i < env->prog->len; i++) {
3533 sl = env->explored_states[i];
3536 while (sl != STATE_LIST_MARK) {
3543 kfree(env->explored_states);
3546 int bpf_check(struct bpf_prog **prog, union bpf_attr *attr)
3548 char __user *log_ubuf = NULL;
3549 struct bpf_verifier_env *env;
3552 /* 'struct bpf_verifier_env' can be global, but since it's not small,
3553 * allocate/free it every time bpf_check() is called
3555 env = kzalloc(sizeof(struct bpf_verifier_env), GFP_KERNEL);
3559 env->insn_aux_data = vzalloc(sizeof(struct bpf_insn_aux_data) *
3562 if (!env->insn_aux_data)
3566 /* grab the mutex to protect few globals used by verifier */
3567 mutex_lock(&bpf_verifier_lock);
3569 if (attr->log_level || attr->log_buf || attr->log_size) {
3570 /* user requested verbose verifier output
3571 * and supplied buffer to store the verification trace
3573 log_level = attr->log_level;
3574 log_ubuf = (char __user *) (unsigned long) attr->log_buf;
3575 log_size = attr->log_size;
3579 /* log_* values have to be sane */
3580 if (log_size < 128 || log_size > UINT_MAX >> 8 ||
3581 log_level == 0 || log_ubuf == NULL)
3585 log_buf = vmalloc(log_size);
3591 if (attr->prog_flags & BPF_F_STRICT_ALIGNMENT)
3592 env->strict_alignment = true;
3594 env->strict_alignment = false;
3596 ret = replace_map_fd_with_map_ptr(env);
3598 goto skip_full_check;
3600 env->explored_states = kcalloc(env->prog->len,
3601 sizeof(struct bpf_verifier_state_list *),
3604 if (!env->explored_states)
3605 goto skip_full_check;
3607 ret = check_cfg(env);
3609 goto skip_full_check;
3611 env->allow_ptr_leaks = capable(CAP_SYS_ADMIN);
3613 ret = do_check(env);
3616 while (pop_stack(env, NULL) >= 0);
3620 /* program is valid, convert *(u32*)(ctx + off) accesses */
3621 ret = convert_ctx_accesses(env);
3624 ret = fixup_bpf_calls(env);
3626 if (log_level && log_len >= log_size - 1) {
3627 BUG_ON(log_len >= log_size);
3628 /* verifier log exceeded user supplied buffer */
3630 /* fall through to return what was recorded */
3633 /* copy verifier log back to user space including trailing zero */
3634 if (log_level && copy_to_user(log_ubuf, log_buf, log_len + 1) != 0) {
3639 if (ret == 0 && env->used_map_cnt) {
3640 /* if program passed verifier, update used_maps in bpf_prog_info */
3641 env->prog->aux->used_maps = kmalloc_array(env->used_map_cnt,
3642 sizeof(env->used_maps[0]),
3645 if (!env->prog->aux->used_maps) {
3650 memcpy(env->prog->aux->used_maps, env->used_maps,
3651 sizeof(env->used_maps[0]) * env->used_map_cnt);
3652 env->prog->aux->used_map_cnt = env->used_map_cnt;
3654 /* program is valid. Convert pseudo bpf_ld_imm64 into generic
3655 * bpf_ld_imm64 instructions
3657 convert_pseudo_ld_imm64(env);
3663 if (!env->prog->aux->used_maps)
3664 /* if we didn't copy map pointers into bpf_prog_info, release
3665 * them now. Otherwise free_bpf_prog_info() will release them.
3670 mutex_unlock(&bpf_verifier_lock);
3671 vfree(env->insn_aux_data);
3677 int bpf_analyzer(struct bpf_prog *prog, const struct bpf_ext_analyzer_ops *ops,
3680 struct bpf_verifier_env *env;
3683 env = kzalloc(sizeof(struct bpf_verifier_env), GFP_KERNEL);
3687 env->insn_aux_data = vzalloc(sizeof(struct bpf_insn_aux_data) *
3690 if (!env->insn_aux_data)
3693 env->analyzer_ops = ops;
3694 env->analyzer_priv = priv;
3696 /* grab the mutex to protect few globals used by verifier */
3697 mutex_lock(&bpf_verifier_lock);
3700 env->strict_alignment = false;
3702 env->explored_states = kcalloc(env->prog->len,
3703 sizeof(struct bpf_verifier_state_list *),
3706 if (!env->explored_states)
3707 goto skip_full_check;
3709 ret = check_cfg(env);
3711 goto skip_full_check;
3713 env->allow_ptr_leaks = capable(CAP_SYS_ADMIN);
3715 ret = do_check(env);
3718 while (pop_stack(env, NULL) >= 0);
3721 mutex_unlock(&bpf_verifier_lock);
3722 vfree(env->insn_aux_data);
3727 EXPORT_SYMBOL_GPL(bpf_analyzer);