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 32k 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 65536
144 #define BPF_COMPLEXITY_LIMIT_STACK 1024
146 struct bpf_call_arg_meta {
147 struct bpf_map *map_ptr;
154 /* verbose verifier prints what it's seeing
155 * bpf_check() is called under lock, so no race to access these global vars
157 static u32 log_level, log_size, log_len;
158 static char *log_buf;
160 static DEFINE_MUTEX(bpf_verifier_lock);
162 /* log_level controls verbosity level of eBPF verifier.
163 * verbose() is used to dump the verification trace to the log, so the user
164 * can figure out what's wrong with the program
166 static __printf(1, 2) void verbose(const char *fmt, ...)
170 if (log_level == 0 || log_len >= log_size - 1)
174 log_len += vscnprintf(log_buf + log_len, log_size - log_len, fmt, args);
178 /* string representation of 'enum bpf_reg_type' */
179 static const char * const reg_type_str[] = {
181 [UNKNOWN_VALUE] = "inv",
182 [PTR_TO_CTX] = "ctx",
183 [CONST_PTR_TO_MAP] = "map_ptr",
184 [PTR_TO_MAP_VALUE] = "map_value",
185 [PTR_TO_MAP_VALUE_OR_NULL] = "map_value_or_null",
186 [PTR_TO_MAP_VALUE_ADJ] = "map_value_adj",
188 [PTR_TO_STACK] = "fp",
190 [PTR_TO_PACKET] = "pkt",
191 [PTR_TO_PACKET_END] = "pkt_end",
194 #define __BPF_FUNC_STR_FN(x) [BPF_FUNC_ ## x] = __stringify(bpf_ ## x)
195 static const char * const func_id_str[] = {
196 __BPF_FUNC_MAPPER(__BPF_FUNC_STR_FN)
198 #undef __BPF_FUNC_STR_FN
200 static const char *func_id_name(int id)
202 BUILD_BUG_ON(ARRAY_SIZE(func_id_str) != __BPF_FUNC_MAX_ID);
204 if (id >= 0 && id < __BPF_FUNC_MAX_ID && func_id_str[id])
205 return func_id_str[id];
210 static void print_verifier_state(struct bpf_verifier_state *state)
212 struct bpf_reg_state *reg;
216 for (i = 0; i < MAX_BPF_REG; i++) {
217 reg = &state->regs[i];
221 verbose(" R%d=%s", i, reg_type_str[t]);
222 if (t == CONST_IMM || t == PTR_TO_STACK)
223 verbose("%lld", reg->imm);
224 else if (t == PTR_TO_PACKET)
225 verbose("(id=%d,off=%d,r=%d)",
226 reg->id, reg->off, reg->range);
227 else if (t == UNKNOWN_VALUE && reg->imm)
228 verbose("%lld", reg->imm);
229 else if (t == CONST_PTR_TO_MAP || t == PTR_TO_MAP_VALUE ||
230 t == PTR_TO_MAP_VALUE_OR_NULL ||
231 t == PTR_TO_MAP_VALUE_ADJ)
232 verbose("(ks=%d,vs=%d,id=%u)",
233 reg->map_ptr->key_size,
234 reg->map_ptr->value_size,
236 if (reg->min_value != BPF_REGISTER_MIN_RANGE)
237 verbose(",min_value=%lld",
238 (long long)reg->min_value);
239 if (reg->max_value != BPF_REGISTER_MAX_RANGE)
240 verbose(",max_value=%llu",
241 (unsigned long long)reg->max_value);
243 for (i = 0; i < MAX_BPF_STACK; i += BPF_REG_SIZE) {
244 if (state->stack_slot_type[i] == STACK_SPILL)
245 verbose(" fp%d=%s", -MAX_BPF_STACK + i,
246 reg_type_str[state->spilled_regs[i / BPF_REG_SIZE].type]);
251 static const char *const bpf_class_string[] = {
259 [BPF_ALU64] = "alu64",
262 static const char *const bpf_alu_string[16] = {
263 [BPF_ADD >> 4] = "+=",
264 [BPF_SUB >> 4] = "-=",
265 [BPF_MUL >> 4] = "*=",
266 [BPF_DIV >> 4] = "/=",
267 [BPF_OR >> 4] = "|=",
268 [BPF_AND >> 4] = "&=",
269 [BPF_LSH >> 4] = "<<=",
270 [BPF_RSH >> 4] = ">>=",
271 [BPF_NEG >> 4] = "neg",
272 [BPF_MOD >> 4] = "%=",
273 [BPF_XOR >> 4] = "^=",
274 [BPF_MOV >> 4] = "=",
275 [BPF_ARSH >> 4] = "s>>=",
276 [BPF_END >> 4] = "endian",
279 static const char *const bpf_ldst_string[] = {
280 [BPF_W >> 3] = "u32",
281 [BPF_H >> 3] = "u16",
283 [BPF_DW >> 3] = "u64",
286 static const char *const bpf_jmp_string[16] = {
287 [BPF_JA >> 4] = "jmp",
288 [BPF_JEQ >> 4] = "==",
289 [BPF_JGT >> 4] = ">",
290 [BPF_JGE >> 4] = ">=",
291 [BPF_JSET >> 4] = "&",
292 [BPF_JNE >> 4] = "!=",
293 [BPF_JSGT >> 4] = "s>",
294 [BPF_JSGE >> 4] = "s>=",
295 [BPF_CALL >> 4] = "call",
296 [BPF_EXIT >> 4] = "exit",
299 static void print_bpf_insn(struct bpf_insn *insn)
301 u8 class = BPF_CLASS(insn->code);
303 if (class == BPF_ALU || class == BPF_ALU64) {
304 if (BPF_SRC(insn->code) == BPF_X)
305 verbose("(%02x) %sr%d %s %sr%d\n",
306 insn->code, class == BPF_ALU ? "(u32) " : "",
308 bpf_alu_string[BPF_OP(insn->code) >> 4],
309 class == BPF_ALU ? "(u32) " : "",
312 verbose("(%02x) %sr%d %s %s%d\n",
313 insn->code, class == BPF_ALU ? "(u32) " : "",
315 bpf_alu_string[BPF_OP(insn->code) >> 4],
316 class == BPF_ALU ? "(u32) " : "",
318 } else if (class == BPF_STX) {
319 if (BPF_MODE(insn->code) == BPF_MEM)
320 verbose("(%02x) *(%s *)(r%d %+d) = r%d\n",
322 bpf_ldst_string[BPF_SIZE(insn->code) >> 3],
324 insn->off, insn->src_reg);
325 else if (BPF_MODE(insn->code) == BPF_XADD)
326 verbose("(%02x) lock *(%s *)(r%d %+d) += r%d\n",
328 bpf_ldst_string[BPF_SIZE(insn->code) >> 3],
329 insn->dst_reg, insn->off,
332 verbose("BUG_%02x\n", insn->code);
333 } else if (class == BPF_ST) {
334 if (BPF_MODE(insn->code) != BPF_MEM) {
335 verbose("BUG_st_%02x\n", insn->code);
338 verbose("(%02x) *(%s *)(r%d %+d) = %d\n",
340 bpf_ldst_string[BPF_SIZE(insn->code) >> 3],
342 insn->off, insn->imm);
343 } else if (class == BPF_LDX) {
344 if (BPF_MODE(insn->code) != BPF_MEM) {
345 verbose("BUG_ldx_%02x\n", insn->code);
348 verbose("(%02x) r%d = *(%s *)(r%d %+d)\n",
349 insn->code, insn->dst_reg,
350 bpf_ldst_string[BPF_SIZE(insn->code) >> 3],
351 insn->src_reg, insn->off);
352 } else if (class == BPF_LD) {
353 if (BPF_MODE(insn->code) == BPF_ABS) {
354 verbose("(%02x) r0 = *(%s *)skb[%d]\n",
356 bpf_ldst_string[BPF_SIZE(insn->code) >> 3],
358 } else if (BPF_MODE(insn->code) == BPF_IND) {
359 verbose("(%02x) r0 = *(%s *)skb[r%d + %d]\n",
361 bpf_ldst_string[BPF_SIZE(insn->code) >> 3],
362 insn->src_reg, insn->imm);
363 } else if (BPF_MODE(insn->code) == BPF_IMM) {
364 verbose("(%02x) r%d = 0x%x\n",
365 insn->code, insn->dst_reg, insn->imm);
367 verbose("BUG_ld_%02x\n", insn->code);
370 } else if (class == BPF_JMP) {
371 u8 opcode = BPF_OP(insn->code);
373 if (opcode == BPF_CALL) {
374 verbose("(%02x) call %s#%d\n", insn->code,
375 func_id_name(insn->imm), insn->imm);
376 } else if (insn->code == (BPF_JMP | BPF_JA)) {
377 verbose("(%02x) goto pc%+d\n",
378 insn->code, insn->off);
379 } else if (insn->code == (BPF_JMP | BPF_EXIT)) {
380 verbose("(%02x) exit\n", insn->code);
381 } else if (BPF_SRC(insn->code) == BPF_X) {
382 verbose("(%02x) if r%d %s r%d goto pc%+d\n",
383 insn->code, insn->dst_reg,
384 bpf_jmp_string[BPF_OP(insn->code) >> 4],
385 insn->src_reg, insn->off);
387 verbose("(%02x) if r%d %s 0x%x goto pc%+d\n",
388 insn->code, insn->dst_reg,
389 bpf_jmp_string[BPF_OP(insn->code) >> 4],
390 insn->imm, insn->off);
393 verbose("(%02x) %s\n", insn->code, bpf_class_string[class]);
397 static int pop_stack(struct bpf_verifier_env *env, int *prev_insn_idx)
399 struct bpf_verifier_stack_elem *elem;
402 if (env->head == NULL)
405 memcpy(&env->cur_state, &env->head->st, sizeof(env->cur_state));
406 insn_idx = env->head->insn_idx;
408 *prev_insn_idx = env->head->prev_insn_idx;
409 elem = env->head->next;
416 static struct bpf_verifier_state *push_stack(struct bpf_verifier_env *env,
417 int insn_idx, int prev_insn_idx)
419 struct bpf_verifier_stack_elem *elem;
421 elem = kmalloc(sizeof(struct bpf_verifier_stack_elem), GFP_KERNEL);
425 memcpy(&elem->st, &env->cur_state, sizeof(env->cur_state));
426 elem->insn_idx = insn_idx;
427 elem->prev_insn_idx = prev_insn_idx;
428 elem->next = env->head;
431 if (env->stack_size > BPF_COMPLEXITY_LIMIT_STACK) {
432 verbose("BPF program is too complex\n");
437 /* pop all elements and return */
438 while (pop_stack(env, NULL) >= 0);
442 #define CALLER_SAVED_REGS 6
443 static const int caller_saved[CALLER_SAVED_REGS] = {
444 BPF_REG_0, BPF_REG_1, BPF_REG_2, BPF_REG_3, BPF_REG_4, BPF_REG_5
447 static void init_reg_state(struct bpf_reg_state *regs)
451 for (i = 0; i < MAX_BPF_REG; i++) {
452 regs[i].type = NOT_INIT;
454 regs[i].min_value = BPF_REGISTER_MIN_RANGE;
455 regs[i].max_value = BPF_REGISTER_MAX_RANGE;
459 regs[BPF_REG_FP].type = FRAME_PTR;
461 /* 1st arg to a function */
462 regs[BPF_REG_1].type = PTR_TO_CTX;
465 static void __mark_reg_unknown_value(struct bpf_reg_state *regs, u32 regno)
467 regs[regno].type = UNKNOWN_VALUE;
472 static void mark_reg_unknown_value(struct bpf_reg_state *regs, u32 regno)
474 BUG_ON(regno >= MAX_BPF_REG);
475 __mark_reg_unknown_value(regs, regno);
478 static void reset_reg_range_values(struct bpf_reg_state *regs, u32 regno)
480 regs[regno].min_value = BPF_REGISTER_MIN_RANGE;
481 regs[regno].max_value = BPF_REGISTER_MAX_RANGE;
484 static void mark_reg_unknown_value_and_range(struct bpf_reg_state *regs,
487 mark_reg_unknown_value(regs, regno);
488 reset_reg_range_values(regs, regno);
492 SRC_OP, /* register is used as source operand */
493 DST_OP, /* register is used as destination operand */
494 DST_OP_NO_MARK /* same as above, check only, don't mark */
497 static int check_reg_arg(struct bpf_reg_state *regs, u32 regno,
500 if (regno >= MAX_BPF_REG) {
501 verbose("R%d is invalid\n", regno);
506 /* check whether register used as source operand can be read */
507 if (regs[regno].type == NOT_INIT) {
508 verbose("R%d !read_ok\n", regno);
512 /* check whether register used as dest operand can be written to */
513 if (regno == BPF_REG_FP) {
514 verbose("frame pointer is read only\n");
518 mark_reg_unknown_value(regs, regno);
523 static int bpf_size_to_bytes(int bpf_size)
525 if (bpf_size == BPF_W)
527 else if (bpf_size == BPF_H)
529 else if (bpf_size == BPF_B)
531 else if (bpf_size == BPF_DW)
537 static bool is_spillable_regtype(enum bpf_reg_type type)
540 case PTR_TO_MAP_VALUE:
541 case PTR_TO_MAP_VALUE_OR_NULL:
542 case PTR_TO_MAP_VALUE_ADJ:
546 case PTR_TO_PACKET_END:
548 case CONST_PTR_TO_MAP:
555 /* check_stack_read/write functions track spill/fill of registers,
556 * stack boundary and alignment are checked in check_mem_access()
558 static int check_stack_write(struct bpf_verifier_state *state, int off,
559 int size, int value_regno)
562 /* caller checked that off % size == 0 and -MAX_BPF_STACK <= off < 0,
563 * so it's aligned access and [off, off + size) are within stack limits
566 if (value_regno >= 0 &&
567 is_spillable_regtype(state->regs[value_regno].type)) {
569 /* register containing pointer is being spilled into stack */
570 if (size != BPF_REG_SIZE) {
571 verbose("invalid size of register spill\n");
575 /* save register state */
576 state->spilled_regs[(MAX_BPF_STACK + off) / BPF_REG_SIZE] =
577 state->regs[value_regno];
579 for (i = 0; i < BPF_REG_SIZE; i++)
580 state->stack_slot_type[MAX_BPF_STACK + off + i] = STACK_SPILL;
582 /* regular write of data into stack */
583 state->spilled_regs[(MAX_BPF_STACK + off) / BPF_REG_SIZE] =
584 (struct bpf_reg_state) {};
586 for (i = 0; i < size; i++)
587 state->stack_slot_type[MAX_BPF_STACK + off + i] = STACK_MISC;
592 static int check_stack_read(struct bpf_verifier_state *state, int off, int size,
598 slot_type = &state->stack_slot_type[MAX_BPF_STACK + off];
600 if (slot_type[0] == STACK_SPILL) {
601 if (size != BPF_REG_SIZE) {
602 verbose("invalid size of register spill\n");
605 for (i = 1; i < BPF_REG_SIZE; i++) {
606 if (slot_type[i] != STACK_SPILL) {
607 verbose("corrupted spill memory\n");
612 if (value_regno >= 0)
613 /* restore register state from stack */
614 state->regs[value_regno] =
615 state->spilled_regs[(MAX_BPF_STACK + off) / BPF_REG_SIZE];
618 for (i = 0; i < size; i++) {
619 if (slot_type[i] != STACK_MISC) {
620 verbose("invalid read from stack off %d+%d size %d\n",
625 if (value_regno >= 0)
626 /* have read misc data from the stack */
627 mark_reg_unknown_value_and_range(state->regs,
633 /* check read/write into map element returned by bpf_map_lookup_elem() */
634 static int check_map_access(struct bpf_verifier_env *env, u32 regno, int off,
637 struct bpf_map *map = env->cur_state.regs[regno].map_ptr;
639 if (off < 0 || size <= 0 || off + size > map->value_size) {
640 verbose("invalid access to map value, value_size=%d off=%d size=%d\n",
641 map->value_size, off, size);
647 /* check read/write into an adjusted map element */
648 static int check_map_access_adj(struct bpf_verifier_env *env, u32 regno,
651 struct bpf_verifier_state *state = &env->cur_state;
652 struct bpf_reg_state *reg = &state->regs[regno];
655 /* We adjusted the register to this map value, so we
656 * need to change off and size to min_value and max_value
657 * respectively to make sure our theoretical access will be
661 print_verifier_state(state);
662 env->varlen_map_value_access = true;
663 /* The minimum value is only important with signed
664 * comparisons where we can't assume the floor of a
665 * value is 0. If we are using signed variables for our
666 * index'es we need to make sure that whatever we use
667 * will have a set floor within our range.
669 if (reg->min_value < 0) {
670 verbose("R%d min value is negative, either use unsigned index or do a if (index >=0) check.\n",
674 err = check_map_access(env, regno, reg->min_value + off, size);
676 verbose("R%d min value is outside of the array range\n",
681 /* If we haven't set a max value then we need to bail
682 * since we can't be sure we won't do bad things.
684 if (reg->max_value == BPF_REGISTER_MAX_RANGE) {
685 verbose("R%d unbounded memory access, make sure to bounds check any array access into a map\n",
689 return check_map_access(env, regno, reg->max_value + off, size);
692 #define MAX_PACKET_OFF 0xffff
694 static bool may_access_direct_pkt_data(struct bpf_verifier_env *env,
695 const struct bpf_call_arg_meta *meta,
696 enum bpf_access_type t)
698 switch (env->prog->type) {
699 case BPF_PROG_TYPE_LWT_IN:
700 case BPF_PROG_TYPE_LWT_OUT:
701 /* dst_input() and dst_output() can't write for now */
704 case BPF_PROG_TYPE_SCHED_CLS:
705 case BPF_PROG_TYPE_SCHED_ACT:
706 case BPF_PROG_TYPE_XDP:
707 case BPF_PROG_TYPE_LWT_XMIT:
709 return meta->pkt_access;
711 env->seen_direct_write = true;
718 static int check_packet_access(struct bpf_verifier_env *env, u32 regno, int off,
721 struct bpf_reg_state *regs = env->cur_state.regs;
722 struct bpf_reg_state *reg = ®s[regno];
725 if (off < 0 || size <= 0 || off + size > reg->range) {
726 verbose("invalid access to packet, off=%d size=%d, R%d(id=%d,off=%d,r=%d)\n",
727 off, size, regno, reg->id, reg->off, reg->range);
733 /* check access to 'struct bpf_context' fields */
734 static int check_ctx_access(struct bpf_verifier_env *env, int off, int size,
735 enum bpf_access_type t, enum bpf_reg_type *reg_type)
737 /* for analyzer ctx accesses are already validated and converted */
738 if (env->analyzer_ops)
741 if (env->prog->aux->ops->is_valid_access &&
742 env->prog->aux->ops->is_valid_access(off, size, t, reg_type)) {
743 /* remember the offset of last byte accessed in ctx */
744 if (env->prog->aux->max_ctx_offset < off + size)
745 env->prog->aux->max_ctx_offset = off + size;
749 verbose("invalid bpf_context access off=%d size=%d\n", off, size);
753 static bool is_pointer_value(struct bpf_verifier_env *env, int regno)
755 if (env->allow_ptr_leaks)
758 switch (env->cur_state.regs[regno].type) {
767 static int check_ptr_alignment(struct bpf_verifier_env *env,
768 struct bpf_reg_state *reg, int off, int size)
770 if (reg->type != PTR_TO_PACKET && reg->type != PTR_TO_MAP_VALUE_ADJ) {
771 if (off % size != 0) {
772 verbose("misaligned access off %d size %d\n",
780 if (IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS))
781 /* misaligned access to packet is ok on x86,arm,arm64 */
784 if (reg->id && size != 1) {
785 verbose("Unknown packet alignment. Only byte-sized access allowed\n");
789 /* skb->data is NET_IP_ALIGN-ed */
790 if (reg->type == PTR_TO_PACKET &&
791 (NET_IP_ALIGN + reg->off + off) % size != 0) {
792 verbose("misaligned packet access off %d+%d+%d size %d\n",
793 NET_IP_ALIGN, reg->off, off, size);
799 /* check whether memory at (regno + off) is accessible for t = (read | write)
800 * if t==write, value_regno is a register which value is stored into memory
801 * if t==read, value_regno is a register which will receive the value from memory
802 * if t==write && value_regno==-1, some unknown value is stored into memory
803 * if t==read && value_regno==-1, don't care what we read from memory
805 static int check_mem_access(struct bpf_verifier_env *env, u32 regno, int off,
806 int bpf_size, enum bpf_access_type t,
809 struct bpf_verifier_state *state = &env->cur_state;
810 struct bpf_reg_state *reg = &state->regs[regno];
813 if (reg->type == PTR_TO_STACK)
816 size = bpf_size_to_bytes(bpf_size);
820 err = check_ptr_alignment(env, reg, off, size);
824 if (reg->type == PTR_TO_MAP_VALUE ||
825 reg->type == PTR_TO_MAP_VALUE_ADJ) {
826 if (t == BPF_WRITE && value_regno >= 0 &&
827 is_pointer_value(env, value_regno)) {
828 verbose("R%d leaks addr into map\n", value_regno);
832 if (reg->type == PTR_TO_MAP_VALUE_ADJ)
833 err = check_map_access_adj(env, regno, off, size);
835 err = check_map_access(env, regno, off, size);
836 if (!err && t == BPF_READ && value_regno >= 0)
837 mark_reg_unknown_value_and_range(state->regs,
840 } else if (reg->type == PTR_TO_CTX) {
841 enum bpf_reg_type reg_type = UNKNOWN_VALUE;
843 if (t == BPF_WRITE && value_regno >= 0 &&
844 is_pointer_value(env, value_regno)) {
845 verbose("R%d leaks addr into ctx\n", value_regno);
848 err = check_ctx_access(env, off, size, t, ®_type);
849 if (!err && t == BPF_READ && value_regno >= 0) {
850 mark_reg_unknown_value_and_range(state->regs,
852 /* note that reg.[id|off|range] == 0 */
853 state->regs[value_regno].type = reg_type;
856 } else if (reg->type == FRAME_PTR || reg->type == PTR_TO_STACK) {
857 if (off >= 0 || off < -MAX_BPF_STACK) {
858 verbose("invalid stack off=%d size=%d\n", off, size);
861 if (t == BPF_WRITE) {
862 if (!env->allow_ptr_leaks &&
863 state->stack_slot_type[MAX_BPF_STACK + off] == STACK_SPILL &&
864 size != BPF_REG_SIZE) {
865 verbose("attempt to corrupt spilled pointer on stack\n");
868 err = check_stack_write(state, off, size, value_regno);
870 err = check_stack_read(state, off, size, value_regno);
872 } else if (state->regs[regno].type == PTR_TO_PACKET) {
873 if (t == BPF_WRITE && !may_access_direct_pkt_data(env, NULL, t)) {
874 verbose("cannot write into packet\n");
877 if (t == BPF_WRITE && value_regno >= 0 &&
878 is_pointer_value(env, value_regno)) {
879 verbose("R%d leaks addr into packet\n", value_regno);
882 err = check_packet_access(env, regno, off, size);
883 if (!err && t == BPF_READ && value_regno >= 0)
884 mark_reg_unknown_value_and_range(state->regs,
887 verbose("R%d invalid mem access '%s'\n",
888 regno, reg_type_str[reg->type]);
892 if (!err && size <= 2 && value_regno >= 0 && env->allow_ptr_leaks &&
893 state->regs[value_regno].type == UNKNOWN_VALUE) {
894 /* 1 or 2 byte load zero-extends, determine the number of
895 * zero upper bits. Not doing it fo 4 byte load, since
896 * such values cannot be added to ptr_to_packet anyway.
898 state->regs[value_regno].imm = 64 - size * 8;
903 static int check_xadd(struct bpf_verifier_env *env, struct bpf_insn *insn)
905 struct bpf_reg_state *regs = env->cur_state.regs;
908 if ((BPF_SIZE(insn->code) != BPF_W && BPF_SIZE(insn->code) != BPF_DW) ||
910 verbose("BPF_XADD uses reserved fields\n");
914 /* check src1 operand */
915 err = check_reg_arg(regs, insn->src_reg, SRC_OP);
919 /* check src2 operand */
920 err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
924 /* check whether atomic_add can read the memory */
925 err = check_mem_access(env, insn->dst_reg, insn->off,
926 BPF_SIZE(insn->code), BPF_READ, -1);
930 /* check whether atomic_add can write into the same memory */
931 return check_mem_access(env, insn->dst_reg, insn->off,
932 BPF_SIZE(insn->code), BPF_WRITE, -1);
935 /* when register 'regno' is passed into function that will read 'access_size'
936 * bytes from that pointer, make sure that it's within stack boundary
937 * and all elements of stack are initialized
939 static int check_stack_boundary(struct bpf_verifier_env *env, int regno,
940 int access_size, bool zero_size_allowed,
941 struct bpf_call_arg_meta *meta)
943 struct bpf_verifier_state *state = &env->cur_state;
944 struct bpf_reg_state *regs = state->regs;
947 if (regs[regno].type != PTR_TO_STACK) {
948 if (zero_size_allowed && access_size == 0 &&
949 regs[regno].type == CONST_IMM &&
950 regs[regno].imm == 0)
953 verbose("R%d type=%s expected=%s\n", regno,
954 reg_type_str[regs[regno].type],
955 reg_type_str[PTR_TO_STACK]);
959 off = regs[regno].imm;
960 if (off >= 0 || off < -MAX_BPF_STACK || off + access_size > 0 ||
962 verbose("invalid stack type R%d off=%d access_size=%d\n",
963 regno, off, access_size);
967 if (meta && meta->raw_mode) {
968 meta->access_size = access_size;
973 for (i = 0; i < access_size; i++) {
974 if (state->stack_slot_type[MAX_BPF_STACK + off + i] != STACK_MISC) {
975 verbose("invalid indirect read from stack off %d+%d size %d\n",
976 off, i, access_size);
983 static int check_helper_mem_access(struct bpf_verifier_env *env, int regno,
984 int access_size, bool zero_size_allowed,
985 struct bpf_call_arg_meta *meta)
987 struct bpf_reg_state *regs = env->cur_state.regs;
989 switch (regs[regno].type) {
991 return check_packet_access(env, regno, 0, access_size);
992 case PTR_TO_MAP_VALUE:
993 return check_map_access(env, regno, 0, access_size);
994 case PTR_TO_MAP_VALUE_ADJ:
995 return check_map_access_adj(env, regno, 0, access_size);
996 default: /* const_imm|ptr_to_stack or invalid ptr */
997 return check_stack_boundary(env, regno, access_size,
998 zero_size_allowed, meta);
1002 static int check_func_arg(struct bpf_verifier_env *env, u32 regno,
1003 enum bpf_arg_type arg_type,
1004 struct bpf_call_arg_meta *meta)
1006 struct bpf_reg_state *regs = env->cur_state.regs, *reg = ®s[regno];
1007 enum bpf_reg_type expected_type, type = reg->type;
1010 if (arg_type == ARG_DONTCARE)
1013 if (type == NOT_INIT) {
1014 verbose("R%d !read_ok\n", regno);
1018 if (arg_type == ARG_ANYTHING) {
1019 if (is_pointer_value(env, regno)) {
1020 verbose("R%d leaks addr into helper function\n", regno);
1026 if (type == PTR_TO_PACKET &&
1027 !may_access_direct_pkt_data(env, meta, BPF_READ)) {
1028 verbose("helper access to the packet is not allowed\n");
1032 if (arg_type == ARG_PTR_TO_MAP_KEY ||
1033 arg_type == ARG_PTR_TO_MAP_VALUE) {
1034 expected_type = PTR_TO_STACK;
1035 if (type != PTR_TO_PACKET && type != expected_type)
1037 } else if (arg_type == ARG_CONST_SIZE ||
1038 arg_type == ARG_CONST_SIZE_OR_ZERO) {
1039 expected_type = CONST_IMM;
1040 /* One exception. Allow UNKNOWN_VALUE registers when the
1041 * boundaries are known and don't cause unsafe memory accesses
1043 if (type != UNKNOWN_VALUE && type != expected_type)
1045 } else if (arg_type == ARG_CONST_MAP_PTR) {
1046 expected_type = CONST_PTR_TO_MAP;
1047 if (type != expected_type)
1049 } else if (arg_type == ARG_PTR_TO_CTX) {
1050 expected_type = PTR_TO_CTX;
1051 if (type != expected_type)
1053 } else if (arg_type == ARG_PTR_TO_MEM ||
1054 arg_type == ARG_PTR_TO_UNINIT_MEM) {
1055 expected_type = PTR_TO_STACK;
1056 /* One exception here. In case function allows for NULL to be
1057 * passed in as argument, it's a CONST_IMM type. Final test
1058 * happens during stack boundary checking.
1060 if (type == CONST_IMM && reg->imm == 0)
1061 /* final test in check_stack_boundary() */;
1062 else if (type != PTR_TO_PACKET && type != PTR_TO_MAP_VALUE &&
1063 type != PTR_TO_MAP_VALUE_ADJ && type != expected_type)
1065 meta->raw_mode = arg_type == ARG_PTR_TO_UNINIT_MEM;
1067 verbose("unsupported arg_type %d\n", arg_type);
1071 if (arg_type == ARG_CONST_MAP_PTR) {
1072 /* bpf_map_xxx(map_ptr) call: remember that map_ptr */
1073 meta->map_ptr = reg->map_ptr;
1074 } else if (arg_type == ARG_PTR_TO_MAP_KEY) {
1075 /* bpf_map_xxx(..., map_ptr, ..., key) call:
1076 * check that [key, key + map->key_size) are within
1077 * stack limits and initialized
1079 if (!meta->map_ptr) {
1080 /* in function declaration map_ptr must come before
1081 * map_key, so that it's verified and known before
1082 * we have to check map_key here. Otherwise it means
1083 * that kernel subsystem misconfigured verifier
1085 verbose("invalid map_ptr to access map->key\n");
1088 if (type == PTR_TO_PACKET)
1089 err = check_packet_access(env, regno, 0,
1090 meta->map_ptr->key_size);
1092 err = check_stack_boundary(env, regno,
1093 meta->map_ptr->key_size,
1095 } else if (arg_type == ARG_PTR_TO_MAP_VALUE) {
1096 /* bpf_map_xxx(..., map_ptr, ..., value) call:
1097 * check [value, value + map->value_size) validity
1099 if (!meta->map_ptr) {
1100 /* kernel subsystem misconfigured verifier */
1101 verbose("invalid map_ptr to access map->value\n");
1104 if (type == PTR_TO_PACKET)
1105 err = check_packet_access(env, regno, 0,
1106 meta->map_ptr->value_size);
1108 err = check_stack_boundary(env, regno,
1109 meta->map_ptr->value_size,
1111 } else if (arg_type == ARG_CONST_SIZE ||
1112 arg_type == ARG_CONST_SIZE_OR_ZERO) {
1113 bool zero_size_allowed = (arg_type == ARG_CONST_SIZE_OR_ZERO);
1115 /* bpf_xxx(..., buf, len) call will access 'len' bytes
1116 * from stack pointer 'buf'. Check it
1117 * note: regno == len, regno - 1 == buf
1120 /* kernel subsystem misconfigured verifier */
1121 verbose("ARG_CONST_SIZE cannot be first argument\n");
1125 /* If the register is UNKNOWN_VALUE, the access check happens
1126 * using its boundaries. Otherwise, just use its imm
1128 if (type == UNKNOWN_VALUE) {
1129 /* For unprivileged variable accesses, disable raw
1130 * mode so that the program is required to
1131 * initialize all the memory that the helper could
1132 * just partially fill up.
1136 if (reg->min_value < 0) {
1137 verbose("R%d min value is negative, either use unsigned or 'var &= const'\n",
1142 if (reg->min_value == 0) {
1143 err = check_helper_mem_access(env, regno - 1, 0,
1150 if (reg->max_value == BPF_REGISTER_MAX_RANGE) {
1151 verbose("R%d unbounded memory access, use 'var &= const' or 'if (var < const)'\n",
1155 err = check_helper_mem_access(env, regno - 1,
1157 zero_size_allowed, meta);
1161 /* register is CONST_IMM */
1162 err = check_helper_mem_access(env, regno - 1, reg->imm,
1163 zero_size_allowed, meta);
1169 verbose("R%d type=%s expected=%s\n", regno,
1170 reg_type_str[type], reg_type_str[expected_type]);
1174 static int check_map_func_compatibility(struct bpf_map *map, int func_id)
1179 /* We need a two way check, first is from map perspective ... */
1180 switch (map->map_type) {
1181 case BPF_MAP_TYPE_PROG_ARRAY:
1182 if (func_id != BPF_FUNC_tail_call)
1185 case BPF_MAP_TYPE_PERF_EVENT_ARRAY:
1186 if (func_id != BPF_FUNC_perf_event_read &&
1187 func_id != BPF_FUNC_perf_event_output)
1190 case BPF_MAP_TYPE_STACK_TRACE:
1191 if (func_id != BPF_FUNC_get_stackid)
1194 case BPF_MAP_TYPE_CGROUP_ARRAY:
1195 if (func_id != BPF_FUNC_skb_under_cgroup &&
1196 func_id != BPF_FUNC_current_task_under_cgroup)
1203 /* ... and second from the function itself. */
1205 case BPF_FUNC_tail_call:
1206 if (map->map_type != BPF_MAP_TYPE_PROG_ARRAY)
1209 case BPF_FUNC_perf_event_read:
1210 case BPF_FUNC_perf_event_output:
1211 if (map->map_type != BPF_MAP_TYPE_PERF_EVENT_ARRAY)
1214 case BPF_FUNC_get_stackid:
1215 if (map->map_type != BPF_MAP_TYPE_STACK_TRACE)
1218 case BPF_FUNC_current_task_under_cgroup:
1219 case BPF_FUNC_skb_under_cgroup:
1220 if (map->map_type != BPF_MAP_TYPE_CGROUP_ARRAY)
1229 verbose("cannot pass map_type %d into func %s#%d\n",
1230 map->map_type, func_id_name(func_id), func_id);
1234 static int check_raw_mode(const struct bpf_func_proto *fn)
1238 if (fn->arg1_type == ARG_PTR_TO_UNINIT_MEM)
1240 if (fn->arg2_type == ARG_PTR_TO_UNINIT_MEM)
1242 if (fn->arg3_type == ARG_PTR_TO_UNINIT_MEM)
1244 if (fn->arg4_type == ARG_PTR_TO_UNINIT_MEM)
1246 if (fn->arg5_type == ARG_PTR_TO_UNINIT_MEM)
1249 return count > 1 ? -EINVAL : 0;
1252 static void clear_all_pkt_pointers(struct bpf_verifier_env *env)
1254 struct bpf_verifier_state *state = &env->cur_state;
1255 struct bpf_reg_state *regs = state->regs, *reg;
1258 for (i = 0; i < MAX_BPF_REG; i++)
1259 if (regs[i].type == PTR_TO_PACKET ||
1260 regs[i].type == PTR_TO_PACKET_END)
1261 mark_reg_unknown_value(regs, i);
1263 for (i = 0; i < MAX_BPF_STACK; i += BPF_REG_SIZE) {
1264 if (state->stack_slot_type[i] != STACK_SPILL)
1266 reg = &state->spilled_regs[i / BPF_REG_SIZE];
1267 if (reg->type != PTR_TO_PACKET &&
1268 reg->type != PTR_TO_PACKET_END)
1270 reg->type = UNKNOWN_VALUE;
1275 static int check_call(struct bpf_verifier_env *env, int func_id)
1277 struct bpf_verifier_state *state = &env->cur_state;
1278 const struct bpf_func_proto *fn = NULL;
1279 struct bpf_reg_state *regs = state->regs;
1280 struct bpf_reg_state *reg;
1281 struct bpf_call_arg_meta meta;
1285 /* find function prototype */
1286 if (func_id < 0 || func_id >= __BPF_FUNC_MAX_ID) {
1287 verbose("invalid func %s#%d\n", func_id_name(func_id), func_id);
1291 if (env->prog->aux->ops->get_func_proto)
1292 fn = env->prog->aux->ops->get_func_proto(func_id);
1295 verbose("unknown func %s#%d\n", func_id_name(func_id), func_id);
1299 /* eBPF programs must be GPL compatible to use GPL-ed functions */
1300 if (!env->prog->gpl_compatible && fn->gpl_only) {
1301 verbose("cannot call GPL only function from proprietary program\n");
1305 changes_data = bpf_helper_changes_pkt_data(fn->func);
1307 memset(&meta, 0, sizeof(meta));
1308 meta.pkt_access = fn->pkt_access;
1310 /* We only support one arg being in raw mode at the moment, which
1311 * is sufficient for the helper functions we have right now.
1313 err = check_raw_mode(fn);
1315 verbose("kernel subsystem misconfigured func %s#%d\n",
1316 func_id_name(func_id), func_id);
1321 err = check_func_arg(env, BPF_REG_1, fn->arg1_type, &meta);
1324 err = check_func_arg(env, BPF_REG_2, fn->arg2_type, &meta);
1327 err = check_func_arg(env, BPF_REG_3, fn->arg3_type, &meta);
1330 err = check_func_arg(env, BPF_REG_4, fn->arg4_type, &meta);
1333 err = check_func_arg(env, BPF_REG_5, fn->arg5_type, &meta);
1337 /* Mark slots with STACK_MISC in case of raw mode, stack offset
1338 * is inferred from register state.
1340 for (i = 0; i < meta.access_size; i++) {
1341 err = check_mem_access(env, meta.regno, i, BPF_B, BPF_WRITE, -1);
1346 /* reset caller saved regs */
1347 for (i = 0; i < CALLER_SAVED_REGS; i++) {
1348 reg = regs + caller_saved[i];
1349 reg->type = NOT_INIT;
1353 /* update return register */
1354 if (fn->ret_type == RET_INTEGER) {
1355 regs[BPF_REG_0].type = UNKNOWN_VALUE;
1356 } else if (fn->ret_type == RET_VOID) {
1357 regs[BPF_REG_0].type = NOT_INIT;
1358 } else if (fn->ret_type == RET_PTR_TO_MAP_VALUE_OR_NULL) {
1359 regs[BPF_REG_0].type = PTR_TO_MAP_VALUE_OR_NULL;
1360 regs[BPF_REG_0].max_value = regs[BPF_REG_0].min_value = 0;
1361 /* remember map_ptr, so that check_map_access()
1362 * can check 'value_size' boundary of memory access
1363 * to map element returned from bpf_map_lookup_elem()
1365 if (meta.map_ptr == NULL) {
1366 verbose("kernel subsystem misconfigured verifier\n");
1369 regs[BPF_REG_0].map_ptr = meta.map_ptr;
1370 regs[BPF_REG_0].id = ++env->id_gen;
1372 verbose("unknown return type %d of func %s#%d\n",
1373 fn->ret_type, func_id_name(func_id), func_id);
1377 err = check_map_func_compatibility(meta.map_ptr, func_id);
1382 clear_all_pkt_pointers(env);
1386 static int check_packet_ptr_add(struct bpf_verifier_env *env,
1387 struct bpf_insn *insn)
1389 struct bpf_reg_state *regs = env->cur_state.regs;
1390 struct bpf_reg_state *dst_reg = ®s[insn->dst_reg];
1391 struct bpf_reg_state *src_reg = ®s[insn->src_reg];
1392 struct bpf_reg_state tmp_reg;
1395 if (BPF_SRC(insn->code) == BPF_K) {
1396 /* pkt_ptr += imm */
1401 verbose("addition of negative constant to packet pointer is not allowed\n");
1404 if (imm >= MAX_PACKET_OFF ||
1405 imm + dst_reg->off >= MAX_PACKET_OFF) {
1406 verbose("constant %d is too large to add to packet pointer\n",
1410 /* a constant was added to pkt_ptr.
1411 * Remember it while keeping the same 'id'
1413 dst_reg->off += imm;
1415 if (src_reg->type == PTR_TO_PACKET) {
1416 /* R6=pkt(id=0,off=0,r=62) R7=imm22; r7 += r6 */
1417 tmp_reg = *dst_reg; /* save r7 state */
1418 *dst_reg = *src_reg; /* copy pkt_ptr state r6 into r7 */
1419 src_reg = &tmp_reg; /* pretend it's src_reg state */
1420 /* if the checks below reject it, the copy won't matter,
1421 * since we're rejecting the whole program. If all ok,
1422 * then imm22 state will be added to r7
1423 * and r7 will be pkt(id=0,off=22,r=62) while
1424 * r6 will stay as pkt(id=0,off=0,r=62)
1428 if (src_reg->type == CONST_IMM) {
1429 /* pkt_ptr += reg where reg is known constant */
1433 /* disallow pkt_ptr += reg
1434 * if reg is not uknown_value with guaranteed zero upper bits
1435 * otherwise pkt_ptr may overflow and addition will become
1436 * subtraction which is not allowed
1438 if (src_reg->type != UNKNOWN_VALUE) {
1439 verbose("cannot add '%s' to ptr_to_packet\n",
1440 reg_type_str[src_reg->type]);
1443 if (src_reg->imm < 48) {
1444 verbose("cannot add integer value with %lld upper zero bits to ptr_to_packet\n",
1448 /* dst_reg stays as pkt_ptr type and since some positive
1449 * integer value was added to the pointer, increment its 'id'
1451 dst_reg->id = ++env->id_gen;
1453 /* something was added to pkt_ptr, set range and off to zero */
1460 static int evaluate_reg_alu(struct bpf_verifier_env *env, struct bpf_insn *insn)
1462 struct bpf_reg_state *regs = env->cur_state.regs;
1463 struct bpf_reg_state *dst_reg = ®s[insn->dst_reg];
1464 u8 opcode = BPF_OP(insn->code);
1467 /* for type == UNKNOWN_VALUE:
1468 * imm > 0 -> number of zero upper bits
1469 * imm == 0 -> don't track which is the same as all bits can be non-zero
1472 if (BPF_SRC(insn->code) == BPF_X) {
1473 struct bpf_reg_state *src_reg = ®s[insn->src_reg];
1475 if (src_reg->type == UNKNOWN_VALUE && src_reg->imm > 0 &&
1476 dst_reg->imm && opcode == BPF_ADD) {
1478 * where both have zero upper bits. Adding them
1479 * can only result making one more bit non-zero
1480 * in the larger value.
1481 * Ex. 0xffff (imm=48) + 1 (imm=63) = 0x10000 (imm=47)
1482 * 0xffff (imm=48) + 0xffff = 0x1fffe (imm=47)
1484 dst_reg->imm = min(dst_reg->imm, src_reg->imm);
1488 if (src_reg->type == CONST_IMM && src_reg->imm > 0 &&
1489 dst_reg->imm && opcode == BPF_ADD) {
1491 * where dreg has zero upper bits and sreg is const.
1492 * Adding them can only result making one more bit
1493 * non-zero in the larger value.
1495 imm_log2 = __ilog2_u64((long long)src_reg->imm);
1496 dst_reg->imm = min(dst_reg->imm, 63 - imm_log2);
1500 /* all other cases non supported yet, just mark dst_reg */
1505 /* sign extend 32-bit imm into 64-bit to make sure that
1506 * negative values occupy bit 63. Note ilog2() would have
1507 * been incorrect, since sizeof(insn->imm) == 4
1509 imm_log2 = __ilog2_u64((long long)insn->imm);
1511 if (dst_reg->imm && opcode == BPF_LSH) {
1513 * if reg was a result of 2 byte load, then its imm == 48
1514 * which means that upper 48 bits are zero and shifting this reg
1515 * left by 4 would mean that upper 44 bits are still zero
1517 dst_reg->imm -= insn->imm;
1518 } else if (dst_reg->imm && opcode == BPF_MUL) {
1520 * if multiplying by 14 subtract 4
1521 * This is conservative calculation of upper zero bits.
1522 * It's not trying to special case insn->imm == 1 or 0 cases
1524 dst_reg->imm -= imm_log2 + 1;
1525 } else if (opcode == BPF_AND) {
1527 dst_reg->imm = 63 - imm_log2;
1528 } else if (dst_reg->imm && opcode == BPF_ADD) {
1530 dst_reg->imm = min(dst_reg->imm, 63 - imm_log2);
1532 } else if (opcode == BPF_RSH) {
1534 * which means that after right shift, upper bits will be zero
1535 * note that verifier already checked that
1536 * 0 <= imm < 64 for shift insn
1538 dst_reg->imm += insn->imm;
1539 if (unlikely(dst_reg->imm > 64))
1540 /* some dumb code did:
1543 * and all bits are zero now */
1546 /* all other alu ops, means that we don't know what will
1547 * happen to the value, mark it with unknown number of zero bits
1552 if (dst_reg->imm < 0) {
1553 /* all 64 bits of the register can contain non-zero bits
1554 * and such value cannot be added to ptr_to_packet, since it
1555 * may overflow, mark it as unknown to avoid further eval
1562 static int evaluate_reg_imm_alu(struct bpf_verifier_env *env,
1563 struct bpf_insn *insn)
1565 struct bpf_reg_state *regs = env->cur_state.regs;
1566 struct bpf_reg_state *dst_reg = ®s[insn->dst_reg];
1567 struct bpf_reg_state *src_reg = ®s[insn->src_reg];
1568 u8 opcode = BPF_OP(insn->code);
1570 /* dst_reg->type == CONST_IMM here, simulate execution of 'add'/'or'
1571 * insn. Don't care about overflow or negative values, just add them
1573 if (opcode == BPF_ADD && BPF_SRC(insn->code) == BPF_K)
1574 dst_reg->imm += insn->imm;
1575 else if (opcode == BPF_ADD && BPF_SRC(insn->code) == BPF_X &&
1576 src_reg->type == CONST_IMM)
1577 dst_reg->imm += src_reg->imm;
1578 else if (opcode == BPF_OR && BPF_SRC(insn->code) == BPF_K)
1579 dst_reg->imm |= insn->imm;
1580 else if (opcode == BPF_OR && BPF_SRC(insn->code) == BPF_X &&
1581 src_reg->type == CONST_IMM)
1582 dst_reg->imm |= src_reg->imm;
1584 mark_reg_unknown_value(regs, insn->dst_reg);
1588 static void check_reg_overflow(struct bpf_reg_state *reg)
1590 if (reg->max_value > BPF_REGISTER_MAX_RANGE)
1591 reg->max_value = BPF_REGISTER_MAX_RANGE;
1592 if (reg->min_value < BPF_REGISTER_MIN_RANGE ||
1593 reg->min_value > BPF_REGISTER_MAX_RANGE)
1594 reg->min_value = BPF_REGISTER_MIN_RANGE;
1597 static void adjust_reg_min_max_vals(struct bpf_verifier_env *env,
1598 struct bpf_insn *insn)
1600 struct bpf_reg_state *regs = env->cur_state.regs, *dst_reg;
1601 s64 min_val = BPF_REGISTER_MIN_RANGE;
1602 u64 max_val = BPF_REGISTER_MAX_RANGE;
1603 u8 opcode = BPF_OP(insn->code);
1605 dst_reg = ®s[insn->dst_reg];
1606 if (BPF_SRC(insn->code) == BPF_X) {
1607 check_reg_overflow(®s[insn->src_reg]);
1608 min_val = regs[insn->src_reg].min_value;
1609 max_val = regs[insn->src_reg].max_value;
1611 /* If the source register is a random pointer then the
1612 * min_value/max_value values represent the range of the known
1613 * accesses into that value, not the actual min/max value of the
1614 * register itself. In this case we have to reset the reg range
1615 * values so we know it is not safe to look at.
1617 if (regs[insn->src_reg].type != CONST_IMM &&
1618 regs[insn->src_reg].type != UNKNOWN_VALUE) {
1619 min_val = BPF_REGISTER_MIN_RANGE;
1620 max_val = BPF_REGISTER_MAX_RANGE;
1622 } else if (insn->imm < BPF_REGISTER_MAX_RANGE &&
1623 (s64)insn->imm > BPF_REGISTER_MIN_RANGE) {
1624 min_val = max_val = insn->imm;
1627 /* We don't know anything about what was done to this register, mark it
1630 if (min_val == BPF_REGISTER_MIN_RANGE &&
1631 max_val == BPF_REGISTER_MAX_RANGE) {
1632 reset_reg_range_values(regs, insn->dst_reg);
1636 /* If one of our values was at the end of our ranges then we can't just
1637 * do our normal operations to the register, we need to set the values
1638 * to the min/max since they are undefined.
1640 if (min_val == BPF_REGISTER_MIN_RANGE)
1641 dst_reg->min_value = BPF_REGISTER_MIN_RANGE;
1642 if (max_val == BPF_REGISTER_MAX_RANGE)
1643 dst_reg->max_value = BPF_REGISTER_MAX_RANGE;
1647 if (dst_reg->min_value != BPF_REGISTER_MIN_RANGE)
1648 dst_reg->min_value += min_val;
1649 if (dst_reg->max_value != BPF_REGISTER_MAX_RANGE)
1650 dst_reg->max_value += max_val;
1653 if (dst_reg->min_value != BPF_REGISTER_MIN_RANGE)
1654 dst_reg->min_value -= min_val;
1655 if (dst_reg->max_value != BPF_REGISTER_MAX_RANGE)
1656 dst_reg->max_value -= max_val;
1659 if (dst_reg->min_value != BPF_REGISTER_MIN_RANGE)
1660 dst_reg->min_value *= min_val;
1661 if (dst_reg->max_value != BPF_REGISTER_MAX_RANGE)
1662 dst_reg->max_value *= max_val;
1665 /* Disallow AND'ing of negative numbers, ain't nobody got time
1666 * for that. Otherwise the minimum is 0 and the max is the max
1667 * value we could AND against.
1670 dst_reg->min_value = BPF_REGISTER_MIN_RANGE;
1672 dst_reg->min_value = 0;
1673 dst_reg->max_value = max_val;
1676 /* Gotta have special overflow logic here, if we're shifting
1677 * more than MAX_RANGE then just assume we have an invalid
1680 if (min_val > ilog2(BPF_REGISTER_MAX_RANGE))
1681 dst_reg->min_value = BPF_REGISTER_MIN_RANGE;
1682 else if (dst_reg->min_value != BPF_REGISTER_MIN_RANGE)
1683 dst_reg->min_value <<= min_val;
1685 if (max_val > ilog2(BPF_REGISTER_MAX_RANGE))
1686 dst_reg->max_value = BPF_REGISTER_MAX_RANGE;
1687 else if (dst_reg->max_value != BPF_REGISTER_MAX_RANGE)
1688 dst_reg->max_value <<= max_val;
1691 /* RSH by a negative number is undefined, and the BPF_RSH is an
1692 * unsigned shift, so make the appropriate casts.
1694 if (min_val < 0 || dst_reg->min_value < 0)
1695 dst_reg->min_value = BPF_REGISTER_MIN_RANGE;
1697 dst_reg->min_value =
1698 (u64)(dst_reg->min_value) >> min_val;
1699 if (dst_reg->max_value != BPF_REGISTER_MAX_RANGE)
1700 dst_reg->max_value >>= max_val;
1703 reset_reg_range_values(regs, insn->dst_reg);
1707 check_reg_overflow(dst_reg);
1710 /* check validity of 32-bit and 64-bit arithmetic operations */
1711 static int check_alu_op(struct bpf_verifier_env *env, struct bpf_insn *insn)
1713 struct bpf_reg_state *regs = env->cur_state.regs, *dst_reg;
1714 u8 opcode = BPF_OP(insn->code);
1717 if (opcode == BPF_END || opcode == BPF_NEG) {
1718 if (opcode == BPF_NEG) {
1719 if (BPF_SRC(insn->code) != 0 ||
1720 insn->src_reg != BPF_REG_0 ||
1721 insn->off != 0 || insn->imm != 0) {
1722 verbose("BPF_NEG uses reserved fields\n");
1726 if (insn->src_reg != BPF_REG_0 || insn->off != 0 ||
1727 (insn->imm != 16 && insn->imm != 32 && insn->imm != 64)) {
1728 verbose("BPF_END uses reserved fields\n");
1733 /* check src operand */
1734 err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
1738 if (is_pointer_value(env, insn->dst_reg)) {
1739 verbose("R%d pointer arithmetic prohibited\n",
1744 /* check dest operand */
1745 err = check_reg_arg(regs, insn->dst_reg, DST_OP);
1749 } else if (opcode == BPF_MOV) {
1751 if (BPF_SRC(insn->code) == BPF_X) {
1752 if (insn->imm != 0 || insn->off != 0) {
1753 verbose("BPF_MOV uses reserved fields\n");
1757 /* check src operand */
1758 err = check_reg_arg(regs, insn->src_reg, SRC_OP);
1762 if (insn->src_reg != BPF_REG_0 || insn->off != 0) {
1763 verbose("BPF_MOV uses reserved fields\n");
1768 /* check dest operand */
1769 err = check_reg_arg(regs, insn->dst_reg, DST_OP);
1773 /* we are setting our register to something new, we need to
1774 * reset its range values.
1776 reset_reg_range_values(regs, insn->dst_reg);
1778 if (BPF_SRC(insn->code) == BPF_X) {
1779 if (BPF_CLASS(insn->code) == BPF_ALU64) {
1781 * copy register state to dest reg
1783 regs[insn->dst_reg] = regs[insn->src_reg];
1785 if (is_pointer_value(env, insn->src_reg)) {
1786 verbose("R%d partial copy of pointer\n",
1790 mark_reg_unknown_value(regs, insn->dst_reg);
1794 * remember the value we stored into this reg
1796 regs[insn->dst_reg].type = CONST_IMM;
1797 regs[insn->dst_reg].imm = insn->imm;
1798 regs[insn->dst_reg].max_value = insn->imm;
1799 regs[insn->dst_reg].min_value = insn->imm;
1802 } else if (opcode > BPF_END) {
1803 verbose("invalid BPF_ALU opcode %x\n", opcode);
1806 } else { /* all other ALU ops: and, sub, xor, add, ... */
1808 if (BPF_SRC(insn->code) == BPF_X) {
1809 if (insn->imm != 0 || insn->off != 0) {
1810 verbose("BPF_ALU uses reserved fields\n");
1813 /* check src1 operand */
1814 err = check_reg_arg(regs, insn->src_reg, SRC_OP);
1818 if (insn->src_reg != BPF_REG_0 || insn->off != 0) {
1819 verbose("BPF_ALU uses reserved fields\n");
1824 /* check src2 operand */
1825 err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
1829 if ((opcode == BPF_MOD || opcode == BPF_DIV) &&
1830 BPF_SRC(insn->code) == BPF_K && insn->imm == 0) {
1831 verbose("div by zero\n");
1835 if ((opcode == BPF_LSH || opcode == BPF_RSH ||
1836 opcode == BPF_ARSH) && BPF_SRC(insn->code) == BPF_K) {
1837 int size = BPF_CLASS(insn->code) == BPF_ALU64 ? 64 : 32;
1839 if (insn->imm < 0 || insn->imm >= size) {
1840 verbose("invalid shift %d\n", insn->imm);
1845 /* check dest operand */
1846 err = check_reg_arg(regs, insn->dst_reg, DST_OP_NO_MARK);
1850 dst_reg = ®s[insn->dst_reg];
1852 /* first we want to adjust our ranges. */
1853 adjust_reg_min_max_vals(env, insn);
1855 /* pattern match 'bpf_add Rx, imm' instruction */
1856 if (opcode == BPF_ADD && BPF_CLASS(insn->code) == BPF_ALU64 &&
1857 dst_reg->type == FRAME_PTR && BPF_SRC(insn->code) == BPF_K) {
1858 dst_reg->type = PTR_TO_STACK;
1859 dst_reg->imm = insn->imm;
1861 } else if (opcode == BPF_ADD &&
1862 BPF_CLASS(insn->code) == BPF_ALU64 &&
1863 (dst_reg->type == PTR_TO_PACKET ||
1864 (BPF_SRC(insn->code) == BPF_X &&
1865 regs[insn->src_reg].type == PTR_TO_PACKET))) {
1866 /* ptr_to_packet += K|X */
1867 return check_packet_ptr_add(env, insn);
1868 } else if (BPF_CLASS(insn->code) == BPF_ALU64 &&
1869 dst_reg->type == UNKNOWN_VALUE &&
1870 env->allow_ptr_leaks) {
1871 /* unknown += K|X */
1872 return evaluate_reg_alu(env, insn);
1873 } else if (BPF_CLASS(insn->code) == BPF_ALU64 &&
1874 dst_reg->type == CONST_IMM &&
1875 env->allow_ptr_leaks) {
1876 /* reg_imm += K|X */
1877 return evaluate_reg_imm_alu(env, insn);
1878 } else if (is_pointer_value(env, insn->dst_reg)) {
1879 verbose("R%d pointer arithmetic prohibited\n",
1882 } else if (BPF_SRC(insn->code) == BPF_X &&
1883 is_pointer_value(env, insn->src_reg)) {
1884 verbose("R%d pointer arithmetic prohibited\n",
1889 /* If we did pointer math on a map value then just set it to our
1890 * PTR_TO_MAP_VALUE_ADJ type so we can deal with any stores or
1891 * loads to this register appropriately, otherwise just mark the
1892 * register as unknown.
1894 if (env->allow_ptr_leaks &&
1895 (dst_reg->type == PTR_TO_MAP_VALUE ||
1896 dst_reg->type == PTR_TO_MAP_VALUE_ADJ))
1897 dst_reg->type = PTR_TO_MAP_VALUE_ADJ;
1899 mark_reg_unknown_value(regs, insn->dst_reg);
1905 static void find_good_pkt_pointers(struct bpf_verifier_state *state,
1906 struct bpf_reg_state *dst_reg)
1908 struct bpf_reg_state *regs = state->regs, *reg;
1911 /* LLVM can generate two kind of checks:
1917 * if (r2 > pkt_end) goto <handle exception>
1921 * r2 == dst_reg, pkt_end == src_reg
1922 * r2=pkt(id=n,off=8,r=0)
1923 * r3=pkt(id=n,off=0,r=0)
1929 * if (pkt_end >= r2) goto <access okay>
1930 * <handle exception>
1933 * pkt_end == dst_reg, r2 == src_reg
1934 * r2=pkt(id=n,off=8,r=0)
1935 * r3=pkt(id=n,off=0,r=0)
1937 * Find register r3 and mark its range as r3=pkt(id=n,off=0,r=8)
1938 * so that range of bytes [r3, r3 + 8) is safe to access.
1941 for (i = 0; i < MAX_BPF_REG; i++)
1942 if (regs[i].type == PTR_TO_PACKET && regs[i].id == dst_reg->id)
1943 regs[i].range = dst_reg->off;
1945 for (i = 0; i < MAX_BPF_STACK; i += BPF_REG_SIZE) {
1946 if (state->stack_slot_type[i] != STACK_SPILL)
1948 reg = &state->spilled_regs[i / BPF_REG_SIZE];
1949 if (reg->type == PTR_TO_PACKET && reg->id == dst_reg->id)
1950 reg->range = dst_reg->off;
1954 /* Adjusts the register min/max values in the case that the dst_reg is the
1955 * variable register that we are working on, and src_reg is a constant or we're
1956 * simply doing a BPF_K check.
1958 static void reg_set_min_max(struct bpf_reg_state *true_reg,
1959 struct bpf_reg_state *false_reg, u64 val,
1964 /* If this is false then we know nothing Jon Snow, but if it is
1965 * true then we know for sure.
1967 true_reg->max_value = true_reg->min_value = val;
1970 /* If this is true we know nothing Jon Snow, but if it is false
1971 * we know the value for sure;
1973 false_reg->max_value = false_reg->min_value = val;
1976 /* Unsigned comparison, the minimum value is 0. */
1977 false_reg->min_value = 0;
1979 /* If this is false then we know the maximum val is val,
1980 * otherwise we know the min val is val+1.
1982 false_reg->max_value = val;
1983 true_reg->min_value = val + 1;
1986 /* Unsigned comparison, the minimum value is 0. */
1987 false_reg->min_value = 0;
1989 /* If this is false then we know the maximum value is val - 1,
1990 * otherwise we know the mimimum value is val.
1992 false_reg->max_value = val - 1;
1993 true_reg->min_value = val;
1999 check_reg_overflow(false_reg);
2000 check_reg_overflow(true_reg);
2003 /* Same as above, but for the case that dst_reg is a CONST_IMM reg and src_reg
2004 * is the variable reg.
2006 static void reg_set_min_max_inv(struct bpf_reg_state *true_reg,
2007 struct bpf_reg_state *false_reg, u64 val,
2012 /* If this is false then we know nothing Jon Snow, but if it is
2013 * true then we know for sure.
2015 true_reg->max_value = true_reg->min_value = val;
2018 /* If this is true we know nothing Jon Snow, but if it is false
2019 * we know the value for sure;
2021 false_reg->max_value = false_reg->min_value = val;
2024 /* Unsigned comparison, the minimum value is 0. */
2025 true_reg->min_value = 0;
2028 * If this is false, then the val is <= the register, if it is
2029 * true the register <= to the val.
2031 false_reg->min_value = val;
2032 true_reg->max_value = val - 1;
2035 /* Unsigned comparison, the minimum value is 0. */
2036 true_reg->min_value = 0;
2038 /* If this is false then constant < register, if it is true then
2039 * the register < constant.
2041 false_reg->min_value = val + 1;
2042 true_reg->max_value = val;
2048 check_reg_overflow(false_reg);
2049 check_reg_overflow(true_reg);
2052 static void mark_map_reg(struct bpf_reg_state *regs, u32 regno, u32 id,
2053 enum bpf_reg_type type)
2055 struct bpf_reg_state *reg = ®s[regno];
2057 if (reg->type == PTR_TO_MAP_VALUE_OR_NULL && reg->id == id) {
2059 /* We don't need id from this point onwards anymore, thus we
2060 * should better reset it, so that state pruning has chances
2064 if (type == UNKNOWN_VALUE)
2065 __mark_reg_unknown_value(regs, regno);
2069 /* The logic is similar to find_good_pkt_pointers(), both could eventually
2070 * be folded together at some point.
2072 static void mark_map_regs(struct bpf_verifier_state *state, u32 regno,
2073 enum bpf_reg_type type)
2075 struct bpf_reg_state *regs = state->regs;
2076 u32 id = regs[regno].id;
2079 for (i = 0; i < MAX_BPF_REG; i++)
2080 mark_map_reg(regs, i, id, type);
2082 for (i = 0; i < MAX_BPF_STACK; i += BPF_REG_SIZE) {
2083 if (state->stack_slot_type[i] != STACK_SPILL)
2085 mark_map_reg(state->spilled_regs, i / BPF_REG_SIZE, id, type);
2089 static int check_cond_jmp_op(struct bpf_verifier_env *env,
2090 struct bpf_insn *insn, int *insn_idx)
2092 struct bpf_verifier_state *other_branch, *this_branch = &env->cur_state;
2093 struct bpf_reg_state *regs = this_branch->regs, *dst_reg;
2094 u8 opcode = BPF_OP(insn->code);
2097 if (opcode > BPF_EXIT) {
2098 verbose("invalid BPF_JMP opcode %x\n", opcode);
2102 if (BPF_SRC(insn->code) == BPF_X) {
2103 if (insn->imm != 0) {
2104 verbose("BPF_JMP uses reserved fields\n");
2108 /* check src1 operand */
2109 err = check_reg_arg(regs, insn->src_reg, SRC_OP);
2113 if (is_pointer_value(env, insn->src_reg)) {
2114 verbose("R%d pointer comparison prohibited\n",
2119 if (insn->src_reg != BPF_REG_0) {
2120 verbose("BPF_JMP uses reserved fields\n");
2125 /* check src2 operand */
2126 err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
2130 dst_reg = ®s[insn->dst_reg];
2132 /* detect if R == 0 where R was initialized to zero earlier */
2133 if (BPF_SRC(insn->code) == BPF_K &&
2134 (opcode == BPF_JEQ || opcode == BPF_JNE) &&
2135 dst_reg->type == CONST_IMM && dst_reg->imm == insn->imm) {
2136 if (opcode == BPF_JEQ) {
2137 /* if (imm == imm) goto pc+off;
2138 * only follow the goto, ignore fall-through
2140 *insn_idx += insn->off;
2143 /* if (imm != imm) goto pc+off;
2144 * only follow fall-through branch, since
2145 * that's where the program will go
2151 other_branch = push_stack(env, *insn_idx + insn->off + 1, *insn_idx);
2155 /* detect if we are comparing against a constant value so we can adjust
2156 * our min/max values for our dst register.
2158 if (BPF_SRC(insn->code) == BPF_X) {
2159 if (regs[insn->src_reg].type == CONST_IMM)
2160 reg_set_min_max(&other_branch->regs[insn->dst_reg],
2161 dst_reg, regs[insn->src_reg].imm,
2163 else if (dst_reg->type == CONST_IMM)
2164 reg_set_min_max_inv(&other_branch->regs[insn->src_reg],
2165 ®s[insn->src_reg], dst_reg->imm,
2168 reg_set_min_max(&other_branch->regs[insn->dst_reg],
2169 dst_reg, insn->imm, opcode);
2172 /* detect if R == 0 where R is returned from bpf_map_lookup_elem() */
2173 if (BPF_SRC(insn->code) == BPF_K &&
2174 insn->imm == 0 && (opcode == BPF_JEQ || opcode == BPF_JNE) &&
2175 dst_reg->type == PTR_TO_MAP_VALUE_OR_NULL) {
2176 /* Mark all identical map registers in each branch as either
2177 * safe or unknown depending R == 0 or R != 0 conditional.
2179 mark_map_regs(this_branch, insn->dst_reg,
2180 opcode == BPF_JEQ ? PTR_TO_MAP_VALUE : UNKNOWN_VALUE);
2181 mark_map_regs(other_branch, insn->dst_reg,
2182 opcode == BPF_JEQ ? UNKNOWN_VALUE : PTR_TO_MAP_VALUE);
2183 } else if (BPF_SRC(insn->code) == BPF_X && opcode == BPF_JGT &&
2184 dst_reg->type == PTR_TO_PACKET &&
2185 regs[insn->src_reg].type == PTR_TO_PACKET_END) {
2186 find_good_pkt_pointers(this_branch, dst_reg);
2187 } else if (BPF_SRC(insn->code) == BPF_X && opcode == BPF_JGE &&
2188 dst_reg->type == PTR_TO_PACKET_END &&
2189 regs[insn->src_reg].type == PTR_TO_PACKET) {
2190 find_good_pkt_pointers(other_branch, ®s[insn->src_reg]);
2191 } else if (is_pointer_value(env, insn->dst_reg)) {
2192 verbose("R%d pointer comparison prohibited\n", insn->dst_reg);
2196 print_verifier_state(this_branch);
2200 /* return the map pointer stored inside BPF_LD_IMM64 instruction */
2201 static struct bpf_map *ld_imm64_to_map_ptr(struct bpf_insn *insn)
2203 u64 imm64 = ((u64) (u32) insn[0].imm) | ((u64) (u32) insn[1].imm) << 32;
2205 return (struct bpf_map *) (unsigned long) imm64;
2208 /* verify BPF_LD_IMM64 instruction */
2209 static int check_ld_imm(struct bpf_verifier_env *env, struct bpf_insn *insn)
2211 struct bpf_reg_state *regs = env->cur_state.regs;
2214 if (BPF_SIZE(insn->code) != BPF_DW) {
2215 verbose("invalid BPF_LD_IMM insn\n");
2218 if (insn->off != 0) {
2219 verbose("BPF_LD_IMM64 uses reserved fields\n");
2223 err = check_reg_arg(regs, insn->dst_reg, DST_OP);
2227 if (insn->src_reg == 0) {
2228 /* generic move 64-bit immediate into a register,
2229 * only analyzer needs to collect the ld_imm value.
2231 u64 imm = ((u64)(insn + 1)->imm << 32) | (u32)insn->imm;
2233 if (!env->analyzer_ops)
2236 regs[insn->dst_reg].type = CONST_IMM;
2237 regs[insn->dst_reg].imm = imm;
2241 /* replace_map_fd_with_map_ptr() should have caught bad ld_imm64 */
2242 BUG_ON(insn->src_reg != BPF_PSEUDO_MAP_FD);
2244 regs[insn->dst_reg].type = CONST_PTR_TO_MAP;
2245 regs[insn->dst_reg].map_ptr = ld_imm64_to_map_ptr(insn);
2249 static bool may_access_skb(enum bpf_prog_type type)
2252 case BPF_PROG_TYPE_SOCKET_FILTER:
2253 case BPF_PROG_TYPE_SCHED_CLS:
2254 case BPF_PROG_TYPE_SCHED_ACT:
2261 /* verify safety of LD_ABS|LD_IND instructions:
2262 * - they can only appear in the programs where ctx == skb
2263 * - since they are wrappers of function calls, they scratch R1-R5 registers,
2264 * preserve R6-R9, and store return value into R0
2267 * ctx == skb == R6 == CTX
2270 * SRC == any register
2271 * IMM == 32-bit immediate
2274 * R0 - 8/16/32-bit skb data converted to cpu endianness
2276 static int check_ld_abs(struct bpf_verifier_env *env, struct bpf_insn *insn)
2278 struct bpf_reg_state *regs = env->cur_state.regs;
2279 u8 mode = BPF_MODE(insn->code);
2280 struct bpf_reg_state *reg;
2283 if (!may_access_skb(env->prog->type)) {
2284 verbose("BPF_LD_[ABS|IND] instructions not allowed for this program type\n");
2288 if (insn->dst_reg != BPF_REG_0 || insn->off != 0 ||
2289 BPF_SIZE(insn->code) == BPF_DW ||
2290 (mode == BPF_ABS && insn->src_reg != BPF_REG_0)) {
2291 verbose("BPF_LD_[ABS|IND] uses reserved fields\n");
2295 /* check whether implicit source operand (register R6) is readable */
2296 err = check_reg_arg(regs, BPF_REG_6, SRC_OP);
2300 if (regs[BPF_REG_6].type != PTR_TO_CTX) {
2301 verbose("at the time of BPF_LD_ABS|IND R6 != pointer to skb\n");
2305 if (mode == BPF_IND) {
2306 /* check explicit source operand */
2307 err = check_reg_arg(regs, insn->src_reg, SRC_OP);
2312 /* reset caller saved regs to unreadable */
2313 for (i = 0; i < CALLER_SAVED_REGS; i++) {
2314 reg = regs + caller_saved[i];
2315 reg->type = NOT_INIT;
2319 /* mark destination R0 register as readable, since it contains
2320 * the value fetched from the packet
2322 regs[BPF_REG_0].type = UNKNOWN_VALUE;
2326 /* non-recursive DFS pseudo code
2327 * 1 procedure DFS-iterative(G,v):
2328 * 2 label v as discovered
2329 * 3 let S be a stack
2331 * 5 while S is not empty
2333 * 7 if t is what we're looking for:
2335 * 9 for all edges e in G.adjacentEdges(t) do
2336 * 10 if edge e is already labelled
2337 * 11 continue with the next edge
2338 * 12 w <- G.adjacentVertex(t,e)
2339 * 13 if vertex w is not discovered and not explored
2340 * 14 label e as tree-edge
2341 * 15 label w as discovered
2344 * 18 else if vertex w is discovered
2345 * 19 label e as back-edge
2347 * 21 // vertex w is explored
2348 * 22 label e as forward- or cross-edge
2349 * 23 label t as explored
2354 * 0x11 - discovered and fall-through edge labelled
2355 * 0x12 - discovered and fall-through and branch edges labelled
2366 #define STATE_LIST_MARK ((struct bpf_verifier_state_list *) -1L)
2368 static int *insn_stack; /* stack of insns to process */
2369 static int cur_stack; /* current stack index */
2370 static int *insn_state;
2372 /* t, w, e - match pseudo-code above:
2373 * t - index of current instruction
2374 * w - next instruction
2377 static int push_insn(int t, int w, int e, struct bpf_verifier_env *env)
2379 if (e == FALLTHROUGH && insn_state[t] >= (DISCOVERED | FALLTHROUGH))
2382 if (e == BRANCH && insn_state[t] >= (DISCOVERED | BRANCH))
2385 if (w < 0 || w >= env->prog->len) {
2386 verbose("jump out of range from insn %d to %d\n", t, w);
2391 /* mark branch target for state pruning */
2392 env->explored_states[w] = STATE_LIST_MARK;
2394 if (insn_state[w] == 0) {
2396 insn_state[t] = DISCOVERED | e;
2397 insn_state[w] = DISCOVERED;
2398 if (cur_stack >= env->prog->len)
2400 insn_stack[cur_stack++] = w;
2402 } else if ((insn_state[w] & 0xF0) == DISCOVERED) {
2403 verbose("back-edge from insn %d to %d\n", t, w);
2405 } else if (insn_state[w] == EXPLORED) {
2406 /* forward- or cross-edge */
2407 insn_state[t] = DISCOVERED | e;
2409 verbose("insn state internal bug\n");
2415 /* non-recursive depth-first-search to detect loops in BPF program
2416 * loop == back-edge in directed graph
2418 static int check_cfg(struct bpf_verifier_env *env)
2420 struct bpf_insn *insns = env->prog->insnsi;
2421 int insn_cnt = env->prog->len;
2425 insn_state = kcalloc(insn_cnt, sizeof(int), GFP_KERNEL);
2429 insn_stack = kcalloc(insn_cnt, sizeof(int), GFP_KERNEL);
2435 insn_state[0] = DISCOVERED; /* mark 1st insn as discovered */
2436 insn_stack[0] = 0; /* 0 is the first instruction */
2442 t = insn_stack[cur_stack - 1];
2444 if (BPF_CLASS(insns[t].code) == BPF_JMP) {
2445 u8 opcode = BPF_OP(insns[t].code);
2447 if (opcode == BPF_EXIT) {
2449 } else if (opcode == BPF_CALL) {
2450 ret = push_insn(t, t + 1, FALLTHROUGH, env);
2455 if (t + 1 < insn_cnt)
2456 env->explored_states[t + 1] = STATE_LIST_MARK;
2457 } else if (opcode == BPF_JA) {
2458 if (BPF_SRC(insns[t].code) != BPF_K) {
2462 /* unconditional jump with single edge */
2463 ret = push_insn(t, t + insns[t].off + 1,
2469 /* tell verifier to check for equivalent states
2470 * after every call and jump
2472 if (t + 1 < insn_cnt)
2473 env->explored_states[t + 1] = STATE_LIST_MARK;
2475 /* conditional jump with two edges */
2476 ret = push_insn(t, t + 1, FALLTHROUGH, env);
2482 ret = push_insn(t, t + insns[t].off + 1, BRANCH, env);
2489 /* all other non-branch instructions with single
2492 ret = push_insn(t, t + 1, FALLTHROUGH, env);
2500 insn_state[t] = EXPLORED;
2501 if (cur_stack-- <= 0) {
2502 verbose("pop stack internal bug\n");
2509 for (i = 0; i < insn_cnt; i++) {
2510 if (insn_state[i] != EXPLORED) {
2511 verbose("unreachable insn %d\n", i);
2516 ret = 0; /* cfg looks good */
2524 /* the following conditions reduce the number of explored insns
2525 * from ~140k to ~80k for ultra large programs that use a lot of ptr_to_packet
2527 static bool compare_ptrs_to_packet(struct bpf_reg_state *old,
2528 struct bpf_reg_state *cur)
2530 if (old->id != cur->id)
2533 /* old ptr_to_packet is more conservative, since it allows smaller
2535 * old(off=0,r=10) is equal to cur(off=0,r=20), because
2536 * old(off=0,r=10) means that with range=10 the verifier proceeded
2537 * further and found no issues with the program. Now we're in the same
2538 * spot with cur(off=0,r=20), so we're safe too, since anything further
2539 * will only be looking at most 10 bytes after this pointer.
2541 if (old->off == cur->off && old->range < cur->range)
2544 /* old(off=20,r=10) is equal to cur(off=22,re=22 or 5 or 0)
2545 * since both cannot be used for packet access and safe(old)
2546 * pointer has smaller off that could be used for further
2547 * 'if (ptr > data_end)' check
2549 * old(off=20,r=10) and cur(off=22,r=22) and cur(off=22,r=0) mean
2550 * that we cannot access the packet.
2551 * The safe range is:
2552 * [ptr, ptr + range - off)
2553 * so whenever off >=range, it means no safe bytes from this pointer.
2554 * When comparing old->off <= cur->off, it means that older code
2555 * went with smaller offset and that offset was later
2556 * used to figure out the safe range after 'if (ptr > data_end)' check
2557 * Say, 'old' state was explored like:
2558 * ... R3(off=0, r=0)
2560 * ... now R4(off=20,r=0) <-- here
2561 * if (R4 > data_end)
2562 * ... R4(off=20,r=20), R3(off=0,r=20) and R3 can be used to access.
2563 * ... the code further went all the way to bpf_exit.
2564 * Now the 'cur' state at the mark 'here' has R4(off=30,r=0).
2565 * old_R4(off=20,r=0) equal to cur_R4(off=30,r=0), since if the verifier
2566 * goes further, such cur_R4 will give larger safe packet range after
2567 * 'if (R4 > data_end)' and all further insn were already good with r=20,
2568 * so they will be good with r=30 and we can prune the search.
2570 if (old->off <= cur->off &&
2571 old->off >= old->range && cur->off >= cur->range)
2577 /* compare two verifier states
2579 * all states stored in state_list are known to be valid, since
2580 * verifier reached 'bpf_exit' instruction through them
2582 * this function is called when verifier exploring different branches of
2583 * execution popped from the state stack. If it sees an old state that has
2584 * more strict register state and more strict stack state then this execution
2585 * branch doesn't need to be explored further, since verifier already
2586 * concluded that more strict state leads to valid finish.
2588 * Therefore two states are equivalent if register state is more conservative
2589 * and explored stack state is more conservative than the current one.
2592 * (slot1=INV slot2=MISC) == (slot1=MISC slot2=MISC)
2593 * (slot1=MISC slot2=MISC) != (slot1=INV slot2=MISC)
2595 * In other words if current stack state (one being explored) has more
2596 * valid slots than old one that already passed validation, it means
2597 * the verifier can stop exploring and conclude that current state is valid too
2599 * Similarly with registers. If explored state has register type as invalid
2600 * whereas register type in current state is meaningful, it means that
2601 * the current state will reach 'bpf_exit' instruction safely
2603 static bool states_equal(struct bpf_verifier_env *env,
2604 struct bpf_verifier_state *old,
2605 struct bpf_verifier_state *cur)
2607 bool varlen_map_access = env->varlen_map_value_access;
2608 struct bpf_reg_state *rold, *rcur;
2611 for (i = 0; i < MAX_BPF_REG; i++) {
2612 rold = &old->regs[i];
2613 rcur = &cur->regs[i];
2615 if (memcmp(rold, rcur, sizeof(*rold)) == 0)
2618 /* If the ranges were not the same, but everything else was and
2619 * we didn't do a variable access into a map then we are a-ok.
2621 if (!varlen_map_access &&
2622 memcmp(rold, rcur, offsetofend(struct bpf_reg_state, id)) == 0)
2625 /* If we didn't map access then again we don't care about the
2626 * mismatched range values and it's ok if our old type was
2627 * UNKNOWN and we didn't go to a NOT_INIT'ed reg.
2629 if (rold->type == NOT_INIT ||
2630 (!varlen_map_access && rold->type == UNKNOWN_VALUE &&
2631 rcur->type != NOT_INIT))
2634 if (rold->type == PTR_TO_PACKET && rcur->type == PTR_TO_PACKET &&
2635 compare_ptrs_to_packet(rold, rcur))
2641 for (i = 0; i < MAX_BPF_STACK; i++) {
2642 if (old->stack_slot_type[i] == STACK_INVALID)
2644 if (old->stack_slot_type[i] != cur->stack_slot_type[i])
2645 /* Ex: old explored (safe) state has STACK_SPILL in
2646 * this stack slot, but current has has STACK_MISC ->
2647 * this verifier states are not equivalent,
2648 * return false to continue verification of this path
2651 if (i % BPF_REG_SIZE)
2653 if (memcmp(&old->spilled_regs[i / BPF_REG_SIZE],
2654 &cur->spilled_regs[i / BPF_REG_SIZE],
2655 sizeof(old->spilled_regs[0])))
2656 /* when explored and current stack slot types are
2657 * the same, check that stored pointers types
2658 * are the same as well.
2659 * Ex: explored safe path could have stored
2660 * (bpf_reg_state) {.type = PTR_TO_STACK, .imm = -8}
2661 * but current path has stored:
2662 * (bpf_reg_state) {.type = PTR_TO_STACK, .imm = -16}
2663 * such verifier states are not equivalent.
2664 * return false to continue verification of this path
2673 static int is_state_visited(struct bpf_verifier_env *env, int insn_idx)
2675 struct bpf_verifier_state_list *new_sl;
2676 struct bpf_verifier_state_list *sl;
2678 sl = env->explored_states[insn_idx];
2680 /* this 'insn_idx' instruction wasn't marked, so we will not
2681 * be doing state search here
2685 while (sl != STATE_LIST_MARK) {
2686 if (states_equal(env, &sl->state, &env->cur_state))
2687 /* reached equivalent register/stack state,
2694 /* there were no equivalent states, remember current one.
2695 * technically the current state is not proven to be safe yet,
2696 * but it will either reach bpf_exit (which means it's safe) or
2697 * it will be rejected. Since there are no loops, we won't be
2698 * seeing this 'insn_idx' instruction again on the way to bpf_exit
2700 new_sl = kmalloc(sizeof(struct bpf_verifier_state_list), GFP_USER);
2704 /* add new state to the head of linked list */
2705 memcpy(&new_sl->state, &env->cur_state, sizeof(env->cur_state));
2706 new_sl->next = env->explored_states[insn_idx];
2707 env->explored_states[insn_idx] = new_sl;
2711 static int ext_analyzer_insn_hook(struct bpf_verifier_env *env,
2712 int insn_idx, int prev_insn_idx)
2714 if (!env->analyzer_ops || !env->analyzer_ops->insn_hook)
2717 return env->analyzer_ops->insn_hook(env, insn_idx, prev_insn_idx);
2720 static int do_check(struct bpf_verifier_env *env)
2722 struct bpf_verifier_state *state = &env->cur_state;
2723 struct bpf_insn *insns = env->prog->insnsi;
2724 struct bpf_reg_state *regs = state->regs;
2725 int insn_cnt = env->prog->len;
2726 int insn_idx, prev_insn_idx = 0;
2727 int insn_processed = 0;
2728 bool do_print_state = false;
2730 init_reg_state(regs);
2732 env->varlen_map_value_access = false;
2734 struct bpf_insn *insn;
2738 if (insn_idx >= insn_cnt) {
2739 verbose("invalid insn idx %d insn_cnt %d\n",
2740 insn_idx, insn_cnt);
2744 insn = &insns[insn_idx];
2745 class = BPF_CLASS(insn->code);
2747 if (++insn_processed > BPF_COMPLEXITY_LIMIT_INSNS) {
2748 verbose("BPF program is too large. Proccessed %d insn\n",
2753 err = is_state_visited(env, insn_idx);
2757 /* found equivalent state, can prune the search */
2760 verbose("\nfrom %d to %d: safe\n",
2761 prev_insn_idx, insn_idx);
2763 verbose("%d: safe\n", insn_idx);
2765 goto process_bpf_exit;
2768 if (log_level && do_print_state) {
2769 verbose("\nfrom %d to %d:", prev_insn_idx, insn_idx);
2770 print_verifier_state(&env->cur_state);
2771 do_print_state = false;
2775 verbose("%d: ", insn_idx);
2776 print_bpf_insn(insn);
2779 err = ext_analyzer_insn_hook(env, insn_idx, prev_insn_idx);
2783 if (class == BPF_ALU || class == BPF_ALU64) {
2784 err = check_alu_op(env, insn);
2788 } else if (class == BPF_LDX) {
2789 enum bpf_reg_type *prev_src_type, src_reg_type;
2791 /* check for reserved fields is already done */
2793 /* check src operand */
2794 err = check_reg_arg(regs, insn->src_reg, SRC_OP);
2798 err = check_reg_arg(regs, insn->dst_reg, DST_OP_NO_MARK);
2802 src_reg_type = regs[insn->src_reg].type;
2804 /* check that memory (src_reg + off) is readable,
2805 * the state of dst_reg will be updated by this func
2807 err = check_mem_access(env, insn->src_reg, insn->off,
2808 BPF_SIZE(insn->code), BPF_READ,
2813 if (BPF_SIZE(insn->code) != BPF_W &&
2814 BPF_SIZE(insn->code) != BPF_DW) {
2819 prev_src_type = &env->insn_aux_data[insn_idx].ptr_type;
2821 if (*prev_src_type == NOT_INIT) {
2823 * dst_reg = *(u32 *)(src_reg + off)
2824 * save type to validate intersecting paths
2826 *prev_src_type = src_reg_type;
2828 } else if (src_reg_type != *prev_src_type &&
2829 (src_reg_type == PTR_TO_CTX ||
2830 *prev_src_type == PTR_TO_CTX)) {
2831 /* ABuser program is trying to use the same insn
2832 * dst_reg = *(u32*) (src_reg + off)
2833 * with different pointer types:
2834 * src_reg == ctx in one branch and
2835 * src_reg == stack|map in some other branch.
2838 verbose("same insn cannot be used with different pointers\n");
2842 } else if (class == BPF_STX) {
2843 enum bpf_reg_type *prev_dst_type, dst_reg_type;
2845 if (BPF_MODE(insn->code) == BPF_XADD) {
2846 err = check_xadd(env, insn);
2853 /* check src1 operand */
2854 err = check_reg_arg(regs, insn->src_reg, SRC_OP);
2857 /* check src2 operand */
2858 err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
2862 dst_reg_type = regs[insn->dst_reg].type;
2864 /* check that memory (dst_reg + off) is writeable */
2865 err = check_mem_access(env, insn->dst_reg, insn->off,
2866 BPF_SIZE(insn->code), BPF_WRITE,
2871 prev_dst_type = &env->insn_aux_data[insn_idx].ptr_type;
2873 if (*prev_dst_type == NOT_INIT) {
2874 *prev_dst_type = dst_reg_type;
2875 } else if (dst_reg_type != *prev_dst_type &&
2876 (dst_reg_type == PTR_TO_CTX ||
2877 *prev_dst_type == PTR_TO_CTX)) {
2878 verbose("same insn cannot be used with different pointers\n");
2882 } else if (class == BPF_ST) {
2883 if (BPF_MODE(insn->code) != BPF_MEM ||
2884 insn->src_reg != BPF_REG_0) {
2885 verbose("BPF_ST uses reserved fields\n");
2888 /* check src operand */
2889 err = check_reg_arg(regs, insn->dst_reg, SRC_OP);
2893 /* check that memory (dst_reg + off) is writeable */
2894 err = check_mem_access(env, insn->dst_reg, insn->off,
2895 BPF_SIZE(insn->code), BPF_WRITE,
2900 } else if (class == BPF_JMP) {
2901 u8 opcode = BPF_OP(insn->code);
2903 if (opcode == BPF_CALL) {
2904 if (BPF_SRC(insn->code) != BPF_K ||
2906 insn->src_reg != BPF_REG_0 ||
2907 insn->dst_reg != BPF_REG_0) {
2908 verbose("BPF_CALL uses reserved fields\n");
2912 err = check_call(env, insn->imm);
2916 } else if (opcode == BPF_JA) {
2917 if (BPF_SRC(insn->code) != BPF_K ||
2919 insn->src_reg != BPF_REG_0 ||
2920 insn->dst_reg != BPF_REG_0) {
2921 verbose("BPF_JA uses reserved fields\n");
2925 insn_idx += insn->off + 1;
2928 } else if (opcode == BPF_EXIT) {
2929 if (BPF_SRC(insn->code) != BPF_K ||
2931 insn->src_reg != BPF_REG_0 ||
2932 insn->dst_reg != BPF_REG_0) {
2933 verbose("BPF_EXIT uses reserved fields\n");
2937 /* eBPF calling convetion is such that R0 is used
2938 * to return the value from eBPF program.
2939 * Make sure that it's readable at this time
2940 * of bpf_exit, which means that program wrote
2941 * something into it earlier
2943 err = check_reg_arg(regs, BPF_REG_0, SRC_OP);
2947 if (is_pointer_value(env, BPF_REG_0)) {
2948 verbose("R0 leaks addr as return value\n");
2953 insn_idx = pop_stack(env, &prev_insn_idx);
2957 do_print_state = true;
2961 err = check_cond_jmp_op(env, insn, &insn_idx);
2965 } else if (class == BPF_LD) {
2966 u8 mode = BPF_MODE(insn->code);
2968 if (mode == BPF_ABS || mode == BPF_IND) {
2969 err = check_ld_abs(env, insn);
2973 } else if (mode == BPF_IMM) {
2974 err = check_ld_imm(env, insn);
2980 verbose("invalid BPF_LD mode\n");
2983 reset_reg_range_values(regs, insn->dst_reg);
2985 verbose("unknown insn class %d\n", class);
2992 verbose("processed %d insns\n", insn_processed);
2996 static int check_map_prog_compatibility(struct bpf_map *map,
2997 struct bpf_prog *prog)
3000 if (prog->type == BPF_PROG_TYPE_PERF_EVENT &&
3001 (map->map_type == BPF_MAP_TYPE_HASH ||
3002 map->map_type == BPF_MAP_TYPE_PERCPU_HASH) &&
3003 (map->map_flags & BPF_F_NO_PREALLOC)) {
3004 verbose("perf_event programs can only use preallocated hash map\n");
3010 /* look for pseudo eBPF instructions that access map FDs and
3011 * replace them with actual map pointers
3013 static int replace_map_fd_with_map_ptr(struct bpf_verifier_env *env)
3015 struct bpf_insn *insn = env->prog->insnsi;
3016 int insn_cnt = env->prog->len;
3019 err = bpf_prog_calc_digest(env->prog);
3023 for (i = 0; i < insn_cnt; i++, insn++) {
3024 if (BPF_CLASS(insn->code) == BPF_LDX &&
3025 (BPF_MODE(insn->code) != BPF_MEM || insn->imm != 0)) {
3026 verbose("BPF_LDX uses reserved fields\n");
3030 if (BPF_CLASS(insn->code) == BPF_STX &&
3031 ((BPF_MODE(insn->code) != BPF_MEM &&
3032 BPF_MODE(insn->code) != BPF_XADD) || insn->imm != 0)) {
3033 verbose("BPF_STX uses reserved fields\n");
3037 if (insn[0].code == (BPF_LD | BPF_IMM | BPF_DW)) {
3038 struct bpf_map *map;
3041 if (i == insn_cnt - 1 || insn[1].code != 0 ||
3042 insn[1].dst_reg != 0 || insn[1].src_reg != 0 ||
3044 verbose("invalid bpf_ld_imm64 insn\n");
3048 if (insn->src_reg == 0)
3049 /* valid generic load 64-bit imm */
3052 if (insn->src_reg != BPF_PSEUDO_MAP_FD) {
3053 verbose("unrecognized bpf_ld_imm64 insn\n");
3057 f = fdget(insn->imm);
3058 map = __bpf_map_get(f);
3060 verbose("fd %d is not pointing to valid bpf_map\n",
3062 return PTR_ERR(map);
3065 err = check_map_prog_compatibility(map, env->prog);
3071 /* store map pointer inside BPF_LD_IMM64 instruction */
3072 insn[0].imm = (u32) (unsigned long) map;
3073 insn[1].imm = ((u64) (unsigned long) map) >> 32;
3075 /* check whether we recorded this map already */
3076 for (j = 0; j < env->used_map_cnt; j++)
3077 if (env->used_maps[j] == map) {
3082 if (env->used_map_cnt >= MAX_USED_MAPS) {
3087 /* hold the map. If the program is rejected by verifier,
3088 * the map will be released by release_maps() or it
3089 * will be used by the valid program until it's unloaded
3090 * and all maps are released in free_bpf_prog_info()
3092 map = bpf_map_inc(map, false);
3095 return PTR_ERR(map);
3097 env->used_maps[env->used_map_cnt++] = map;
3106 /* now all pseudo BPF_LD_IMM64 instructions load valid
3107 * 'struct bpf_map *' into a register instead of user map_fd.
3108 * These pointers will be used later by verifier to validate map access.
3113 /* drop refcnt of maps used by the rejected program */
3114 static void release_maps(struct bpf_verifier_env *env)
3118 for (i = 0; i < env->used_map_cnt; i++)
3119 bpf_map_put(env->used_maps[i]);
3122 /* convert pseudo BPF_LD_IMM64 into generic BPF_LD_IMM64 */
3123 static void convert_pseudo_ld_imm64(struct bpf_verifier_env *env)
3125 struct bpf_insn *insn = env->prog->insnsi;
3126 int insn_cnt = env->prog->len;
3129 for (i = 0; i < insn_cnt; i++, insn++)
3130 if (insn->code == (BPF_LD | BPF_IMM | BPF_DW))
3134 /* convert load instructions that access fields of 'struct __sk_buff'
3135 * into sequence of instructions that access fields of 'struct sk_buff'
3137 static int convert_ctx_accesses(struct bpf_verifier_env *env)
3139 const struct bpf_verifier_ops *ops = env->prog->aux->ops;
3140 const int insn_cnt = env->prog->len;
3141 struct bpf_insn insn_buf[16], *insn;
3142 struct bpf_prog *new_prog;
3143 enum bpf_access_type type;
3144 int i, cnt, delta = 0;
3146 if (ops->gen_prologue) {
3147 cnt = ops->gen_prologue(insn_buf, env->seen_direct_write,
3149 if (cnt >= ARRAY_SIZE(insn_buf)) {
3150 verbose("bpf verifier is misconfigured\n");
3153 new_prog = bpf_patch_insn_single(env->prog, 0,
3157 env->prog = new_prog;
3162 if (!ops->convert_ctx_access)
3165 insn = env->prog->insnsi + delta;
3167 for (i = 0; i < insn_cnt; i++, insn++) {
3168 if (insn->code == (BPF_LDX | BPF_MEM | BPF_W) ||
3169 insn->code == (BPF_LDX | BPF_MEM | BPF_DW))
3171 else if (insn->code == (BPF_STX | BPF_MEM | BPF_W) ||
3172 insn->code == (BPF_STX | BPF_MEM | BPF_DW))
3177 if (env->insn_aux_data[i].ptr_type != PTR_TO_CTX)
3180 cnt = ops->convert_ctx_access(type, insn, insn_buf, env->prog);
3181 if (cnt == 0 || cnt >= ARRAY_SIZE(insn_buf)) {
3182 verbose("bpf verifier is misconfigured\n");
3186 new_prog = bpf_patch_insn_single(env->prog, i + delta, insn_buf,
3193 /* keep walking new program and skip insns we just inserted */
3194 env->prog = new_prog;
3195 insn = new_prog->insnsi + i + delta;
3201 static void free_states(struct bpf_verifier_env *env)
3203 struct bpf_verifier_state_list *sl, *sln;
3206 if (!env->explored_states)
3209 for (i = 0; i < env->prog->len; i++) {
3210 sl = env->explored_states[i];
3213 while (sl != STATE_LIST_MARK) {
3220 kfree(env->explored_states);
3223 int bpf_check(struct bpf_prog **prog, union bpf_attr *attr)
3225 char __user *log_ubuf = NULL;
3226 struct bpf_verifier_env *env;
3229 /* 'struct bpf_verifier_env' can be global, but since it's not small,
3230 * allocate/free it every time bpf_check() is called
3232 env = kzalloc(sizeof(struct bpf_verifier_env), GFP_KERNEL);
3236 env->insn_aux_data = vzalloc(sizeof(struct bpf_insn_aux_data) *
3239 if (!env->insn_aux_data)
3243 /* grab the mutex to protect few globals used by verifier */
3244 mutex_lock(&bpf_verifier_lock);
3246 if (attr->log_level || attr->log_buf || attr->log_size) {
3247 /* user requested verbose verifier output
3248 * and supplied buffer to store the verification trace
3250 log_level = attr->log_level;
3251 log_ubuf = (char __user *) (unsigned long) attr->log_buf;
3252 log_size = attr->log_size;
3256 /* log_* values have to be sane */
3257 if (log_size < 128 || log_size > UINT_MAX >> 8 ||
3258 log_level == 0 || log_ubuf == NULL)
3262 log_buf = vmalloc(log_size);
3269 ret = replace_map_fd_with_map_ptr(env);
3271 goto skip_full_check;
3273 env->explored_states = kcalloc(env->prog->len,
3274 sizeof(struct bpf_verifier_state_list *),
3277 if (!env->explored_states)
3278 goto skip_full_check;
3280 ret = check_cfg(env);
3282 goto skip_full_check;
3284 env->allow_ptr_leaks = capable(CAP_SYS_ADMIN);
3286 ret = do_check(env);
3289 while (pop_stack(env, NULL) >= 0);
3293 /* program is valid, convert *(u32*)(ctx + off) accesses */
3294 ret = convert_ctx_accesses(env);
3296 if (log_level && log_len >= log_size - 1) {
3297 BUG_ON(log_len >= log_size);
3298 /* verifier log exceeded user supplied buffer */
3300 /* fall through to return what was recorded */
3303 /* copy verifier log back to user space including trailing zero */
3304 if (log_level && copy_to_user(log_ubuf, log_buf, log_len + 1) != 0) {
3309 if (ret == 0 && env->used_map_cnt) {
3310 /* if program passed verifier, update used_maps in bpf_prog_info */
3311 env->prog->aux->used_maps = kmalloc_array(env->used_map_cnt,
3312 sizeof(env->used_maps[0]),
3315 if (!env->prog->aux->used_maps) {
3320 memcpy(env->prog->aux->used_maps, env->used_maps,
3321 sizeof(env->used_maps[0]) * env->used_map_cnt);
3322 env->prog->aux->used_map_cnt = env->used_map_cnt;
3324 /* program is valid. Convert pseudo bpf_ld_imm64 into generic
3325 * bpf_ld_imm64 instructions
3327 convert_pseudo_ld_imm64(env);
3333 if (!env->prog->aux->used_maps)
3334 /* if we didn't copy map pointers into bpf_prog_info, release
3335 * them now. Otherwise free_bpf_prog_info() will release them.
3340 mutex_unlock(&bpf_verifier_lock);
3341 vfree(env->insn_aux_data);
3347 int bpf_analyzer(struct bpf_prog *prog, const struct bpf_ext_analyzer_ops *ops,
3350 struct bpf_verifier_env *env;
3353 env = kzalloc(sizeof(struct bpf_verifier_env), GFP_KERNEL);
3357 env->insn_aux_data = vzalloc(sizeof(struct bpf_insn_aux_data) *
3360 if (!env->insn_aux_data)
3363 env->analyzer_ops = ops;
3364 env->analyzer_priv = priv;
3366 /* grab the mutex to protect few globals used by verifier */
3367 mutex_lock(&bpf_verifier_lock);
3371 env->explored_states = kcalloc(env->prog->len,
3372 sizeof(struct bpf_verifier_state_list *),
3375 if (!env->explored_states)
3376 goto skip_full_check;
3378 ret = check_cfg(env);
3380 goto skip_full_check;
3382 env->allow_ptr_leaks = capable(CAP_SYS_ADMIN);
3384 ret = do_check(env);
3387 while (pop_stack(env, NULL) >= 0);
3390 mutex_unlock(&bpf_verifier_lock);
3391 vfree(env->insn_aux_data);
3396 EXPORT_SYMBOL_GPL(bpf_analyzer);