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>
23 /* bpf_check() is a static code analyzer that walks eBPF program
24 * instruction by instruction and updates register/stack state.
25 * All paths of conditional branches are analyzed until 'bpf_exit' insn.
27 * The first pass is depth-first-search to check that the program is a DAG.
28 * It rejects the following programs:
29 * - larger than BPF_MAXINSNS insns
30 * - if loop is present (detected via back-edge)
31 * - unreachable insns exist (shouldn't be a forest. program = one function)
32 * - out of bounds or malformed jumps
33 * The second pass is all possible path descent from the 1st insn.
34 * Since it's analyzing all pathes through the program, the length of the
35 * analysis is limited to 32k insn, which may be hit even if total number of
36 * insn is less then 4K, but there are too many branches that change stack/regs.
37 * Number of 'branches to be analyzed' is limited to 1k
39 * On entry to each instruction, each register has a type, and the instruction
40 * changes the types of the registers depending on instruction semantics.
41 * If instruction is BPF_MOV64_REG(BPF_REG_1, BPF_REG_5), then type of R5 is
44 * All registers are 64-bit.
45 * R0 - return register
46 * R1-R5 argument passing registers
47 * R6-R9 callee saved registers
48 * R10 - frame pointer read-only
50 * At the start of BPF program the register R1 contains a pointer to bpf_context
51 * and has type PTR_TO_CTX.
53 * Verifier tracks arithmetic operations on pointers in case:
54 * BPF_MOV64_REG(BPF_REG_1, BPF_REG_10),
55 * BPF_ALU64_IMM(BPF_ADD, BPF_REG_1, -20),
56 * 1st insn copies R10 (which has FRAME_PTR) type into R1
57 * and 2nd arithmetic instruction is pattern matched to recognize
58 * that it wants to construct a pointer to some element within stack.
59 * So after 2nd insn, the register R1 has type PTR_TO_STACK
60 * (and -20 constant is saved for further stack bounds checking).
61 * Meaning that this reg is a pointer to stack plus known immediate constant.
63 * Most of the time the registers have UNKNOWN_VALUE type, which
64 * means the register has some value, but it's not a valid pointer.
65 * (like pointer plus pointer becomes UNKNOWN_VALUE type)
67 * When verifier sees load or store instructions the type of base register
68 * can be: PTR_TO_MAP_VALUE, PTR_TO_CTX, FRAME_PTR. These are three pointer
69 * types recognized by check_mem_access() function.
71 * PTR_TO_MAP_VALUE means that this register is pointing to 'map element value'
72 * and the range of [ptr, ptr + map's value_size) is accessible.
74 * registers used to pass values to function calls are checked against
75 * function argument constraints.
77 * ARG_PTR_TO_MAP_KEY is one of such argument constraints.
78 * It means that the register type passed to this function must be
79 * PTR_TO_STACK and it will be used inside the function as
80 * 'pointer to map element key'
82 * For example the argument constraints for bpf_map_lookup_elem():
83 * .ret_type = RET_PTR_TO_MAP_VALUE_OR_NULL,
84 * .arg1_type = ARG_CONST_MAP_PTR,
85 * .arg2_type = ARG_PTR_TO_MAP_KEY,
87 * ret_type says that this function returns 'pointer to map elem value or null'
88 * function expects 1st argument to be a const pointer to 'struct bpf_map' and
89 * 2nd argument should be a pointer to stack, which will be used inside
90 * the helper function as a pointer to map element key.
92 * On the kernel side the helper function looks like:
93 * u64 bpf_map_lookup_elem(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5)
95 * struct bpf_map *map = (struct bpf_map *) (unsigned long) r1;
96 * void *key = (void *) (unsigned long) r2;
99 * here kernel can access 'key' and 'map' pointers safely, knowing that
100 * [key, key + map->key_size) bytes are valid and were initialized on
101 * the stack of eBPF program.
104 * Corresponding eBPF program may look like:
105 * BPF_MOV64_REG(BPF_REG_2, BPF_REG_10), // after this insn R2 type is FRAME_PTR
106 * BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, -4), // after this insn R2 type is PTR_TO_STACK
107 * BPF_LD_MAP_FD(BPF_REG_1, map_fd), // after this insn R1 type is CONST_PTR_TO_MAP
108 * BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0, BPF_FUNC_map_lookup_elem),
109 * here verifier looks at prototype of map_lookup_elem() and sees:
110 * .arg1_type == ARG_CONST_MAP_PTR and R1->type == CONST_PTR_TO_MAP, which is ok,
111 * Now verifier knows that this map has key of R1->map_ptr->key_size bytes
113 * Then .arg2_type == ARG_PTR_TO_MAP_KEY and R2->type == PTR_TO_STACK, ok so far,
114 * Now verifier checks that [R2, R2 + map's key_size) are within stack limits
115 * and were initialized prior to this call.
116 * If it's ok, then verifier allows this BPF_CALL insn and looks at
117 * .ret_type which is RET_PTR_TO_MAP_VALUE_OR_NULL, so it sets
118 * R0->type = PTR_TO_MAP_VALUE_OR_NULL which means bpf_map_lookup_elem() function
119 * returns ether pointer to map value or NULL.
121 * When type PTR_TO_MAP_VALUE_OR_NULL passes through 'if (reg != 0) goto +off'
122 * insn, the register holding that pointer in the true branch changes state to
123 * PTR_TO_MAP_VALUE and the same register changes state to CONST_IMM in the false
124 * branch. See check_cond_jmp_op().
126 * After the call R0 is set to return type of the function and registers R1-R5
127 * are set to NOT_INIT to indicate that they are no longer readable.
130 /* verifier_state + insn_idx are pushed to stack when branch is encountered */
131 struct bpf_verifier_stack_elem
{
132 /* verifer state is 'st'
133 * before processing instruction 'insn_idx'
134 * and after processing instruction 'prev_insn_idx'
136 struct bpf_verifier_state st
;
139 struct bpf_verifier_stack_elem
*next
;
142 #define BPF_COMPLEXITY_LIMIT_INSNS 65536
143 #define BPF_COMPLEXITY_LIMIT_STACK 1024
145 struct bpf_call_arg_meta
{
146 struct bpf_map
*map_ptr
;
153 /* verbose verifier prints what it's seeing
154 * bpf_check() is called under lock, so no race to access these global vars
156 static u32 log_level
, log_size
, log_len
;
157 static char *log_buf
;
159 static DEFINE_MUTEX(bpf_verifier_lock
);
161 /* log_level controls verbosity level of eBPF verifier.
162 * verbose() is used to dump the verification trace to the log, so the user
163 * can figure out what's wrong with the program
165 static __printf(1, 2) void verbose(const char *fmt
, ...)
169 if (log_level
== 0 || log_len
>= log_size
- 1)
173 log_len
+= vscnprintf(log_buf
+ log_len
, log_size
- log_len
, fmt
, args
);
177 /* string representation of 'enum bpf_reg_type' */
178 static const char * const reg_type_str
[] = {
180 [UNKNOWN_VALUE
] = "inv",
181 [PTR_TO_CTX
] = "ctx",
182 [CONST_PTR_TO_MAP
] = "map_ptr",
183 [PTR_TO_MAP_VALUE
] = "map_value",
184 [PTR_TO_MAP_VALUE_OR_NULL
] = "map_value_or_null",
185 [PTR_TO_MAP_VALUE_ADJ
] = "map_value_adj",
187 [PTR_TO_STACK
] = "fp",
189 [PTR_TO_PACKET
] = "pkt",
190 [PTR_TO_PACKET_END
] = "pkt_end",
193 static void print_verifier_state(struct bpf_verifier_state
*state
)
195 struct bpf_reg_state
*reg
;
199 for (i
= 0; i
< MAX_BPF_REG
; i
++) {
200 reg
= &state
->regs
[i
];
204 verbose(" R%d=%s", i
, reg_type_str
[t
]);
205 if (t
== CONST_IMM
|| t
== PTR_TO_STACK
)
206 verbose("%lld", reg
->imm
);
207 else if (t
== PTR_TO_PACKET
)
208 verbose("(id=%d,off=%d,r=%d)",
209 reg
->id
, reg
->off
, reg
->range
);
210 else if (t
== UNKNOWN_VALUE
&& reg
->imm
)
211 verbose("%lld", reg
->imm
);
212 else if (t
== CONST_PTR_TO_MAP
|| t
== PTR_TO_MAP_VALUE
||
213 t
== PTR_TO_MAP_VALUE_OR_NULL
||
214 t
== PTR_TO_MAP_VALUE_ADJ
)
215 verbose("(ks=%d,vs=%d)",
216 reg
->map_ptr
->key_size
,
217 reg
->map_ptr
->value_size
);
218 if (reg
->min_value
!= BPF_REGISTER_MIN_RANGE
)
219 verbose(",min_value=%lld",
220 (long long)reg
->min_value
);
221 if (reg
->max_value
!= BPF_REGISTER_MAX_RANGE
)
222 verbose(",max_value=%llu",
223 (unsigned long long)reg
->max_value
);
225 for (i
= 0; i
< MAX_BPF_STACK
; i
+= BPF_REG_SIZE
) {
226 if (state
->stack_slot_type
[i
] == STACK_SPILL
)
227 verbose(" fp%d=%s", -MAX_BPF_STACK
+ i
,
228 reg_type_str
[state
->spilled_regs
[i
/ BPF_REG_SIZE
].type
]);
233 static const char *const bpf_class_string
[] = {
241 [BPF_ALU64
] = "alu64",
244 static const char *const bpf_alu_string
[16] = {
245 [BPF_ADD
>> 4] = "+=",
246 [BPF_SUB
>> 4] = "-=",
247 [BPF_MUL
>> 4] = "*=",
248 [BPF_DIV
>> 4] = "/=",
249 [BPF_OR
>> 4] = "|=",
250 [BPF_AND
>> 4] = "&=",
251 [BPF_LSH
>> 4] = "<<=",
252 [BPF_RSH
>> 4] = ">>=",
253 [BPF_NEG
>> 4] = "neg",
254 [BPF_MOD
>> 4] = "%=",
255 [BPF_XOR
>> 4] = "^=",
256 [BPF_MOV
>> 4] = "=",
257 [BPF_ARSH
>> 4] = "s>>=",
258 [BPF_END
>> 4] = "endian",
261 static const char *const bpf_ldst_string
[] = {
262 [BPF_W
>> 3] = "u32",
263 [BPF_H
>> 3] = "u16",
265 [BPF_DW
>> 3] = "u64",
268 static const char *const bpf_jmp_string
[16] = {
269 [BPF_JA
>> 4] = "jmp",
270 [BPF_JEQ
>> 4] = "==",
271 [BPF_JGT
>> 4] = ">",
272 [BPF_JGE
>> 4] = ">=",
273 [BPF_JSET
>> 4] = "&",
274 [BPF_JNE
>> 4] = "!=",
275 [BPF_JSGT
>> 4] = "s>",
276 [BPF_JSGE
>> 4] = "s>=",
277 [BPF_CALL
>> 4] = "call",
278 [BPF_EXIT
>> 4] = "exit",
281 static void print_bpf_insn(struct bpf_insn
*insn
)
283 u8
class = BPF_CLASS(insn
->code
);
285 if (class == BPF_ALU
|| class == BPF_ALU64
) {
286 if (BPF_SRC(insn
->code
) == BPF_X
)
287 verbose("(%02x) %sr%d %s %sr%d\n",
288 insn
->code
, class == BPF_ALU
? "(u32) " : "",
290 bpf_alu_string
[BPF_OP(insn
->code
) >> 4],
291 class == BPF_ALU
? "(u32) " : "",
294 verbose("(%02x) %sr%d %s %s%d\n",
295 insn
->code
, class == BPF_ALU
? "(u32) " : "",
297 bpf_alu_string
[BPF_OP(insn
->code
) >> 4],
298 class == BPF_ALU
? "(u32) " : "",
300 } else if (class == BPF_STX
) {
301 if (BPF_MODE(insn
->code
) == BPF_MEM
)
302 verbose("(%02x) *(%s *)(r%d %+d) = r%d\n",
304 bpf_ldst_string
[BPF_SIZE(insn
->code
) >> 3],
306 insn
->off
, insn
->src_reg
);
307 else if (BPF_MODE(insn
->code
) == BPF_XADD
)
308 verbose("(%02x) lock *(%s *)(r%d %+d) += r%d\n",
310 bpf_ldst_string
[BPF_SIZE(insn
->code
) >> 3],
311 insn
->dst_reg
, insn
->off
,
314 verbose("BUG_%02x\n", insn
->code
);
315 } else if (class == BPF_ST
) {
316 if (BPF_MODE(insn
->code
) != BPF_MEM
) {
317 verbose("BUG_st_%02x\n", insn
->code
);
320 verbose("(%02x) *(%s *)(r%d %+d) = %d\n",
322 bpf_ldst_string
[BPF_SIZE(insn
->code
) >> 3],
324 insn
->off
, insn
->imm
);
325 } else if (class == BPF_LDX
) {
326 if (BPF_MODE(insn
->code
) != BPF_MEM
) {
327 verbose("BUG_ldx_%02x\n", insn
->code
);
330 verbose("(%02x) r%d = *(%s *)(r%d %+d)\n",
331 insn
->code
, insn
->dst_reg
,
332 bpf_ldst_string
[BPF_SIZE(insn
->code
) >> 3],
333 insn
->src_reg
, insn
->off
);
334 } else if (class == BPF_LD
) {
335 if (BPF_MODE(insn
->code
) == BPF_ABS
) {
336 verbose("(%02x) r0 = *(%s *)skb[%d]\n",
338 bpf_ldst_string
[BPF_SIZE(insn
->code
) >> 3],
340 } else if (BPF_MODE(insn
->code
) == BPF_IND
) {
341 verbose("(%02x) r0 = *(%s *)skb[r%d + %d]\n",
343 bpf_ldst_string
[BPF_SIZE(insn
->code
) >> 3],
344 insn
->src_reg
, insn
->imm
);
345 } else if (BPF_MODE(insn
->code
) == BPF_IMM
) {
346 verbose("(%02x) r%d = 0x%x\n",
347 insn
->code
, insn
->dst_reg
, insn
->imm
);
349 verbose("BUG_ld_%02x\n", insn
->code
);
352 } else if (class == BPF_JMP
) {
353 u8 opcode
= BPF_OP(insn
->code
);
355 if (opcode
== BPF_CALL
) {
356 verbose("(%02x) call %d\n", insn
->code
, insn
->imm
);
357 } else if (insn
->code
== (BPF_JMP
| BPF_JA
)) {
358 verbose("(%02x) goto pc%+d\n",
359 insn
->code
, insn
->off
);
360 } else if (insn
->code
== (BPF_JMP
| BPF_EXIT
)) {
361 verbose("(%02x) exit\n", insn
->code
);
362 } else if (BPF_SRC(insn
->code
) == BPF_X
) {
363 verbose("(%02x) if r%d %s r%d goto pc%+d\n",
364 insn
->code
, insn
->dst_reg
,
365 bpf_jmp_string
[BPF_OP(insn
->code
) >> 4],
366 insn
->src_reg
, insn
->off
);
368 verbose("(%02x) if r%d %s 0x%x goto pc%+d\n",
369 insn
->code
, insn
->dst_reg
,
370 bpf_jmp_string
[BPF_OP(insn
->code
) >> 4],
371 insn
->imm
, insn
->off
);
374 verbose("(%02x) %s\n", insn
->code
, bpf_class_string
[class]);
378 static int pop_stack(struct bpf_verifier_env
*env
, int *prev_insn_idx
)
380 struct bpf_verifier_stack_elem
*elem
;
383 if (env
->head
== NULL
)
386 memcpy(&env
->cur_state
, &env
->head
->st
, sizeof(env
->cur_state
));
387 insn_idx
= env
->head
->insn_idx
;
389 *prev_insn_idx
= env
->head
->prev_insn_idx
;
390 elem
= env
->head
->next
;
397 static struct bpf_verifier_state
*push_stack(struct bpf_verifier_env
*env
,
398 int insn_idx
, int prev_insn_idx
)
400 struct bpf_verifier_stack_elem
*elem
;
402 elem
= kmalloc(sizeof(struct bpf_verifier_stack_elem
), GFP_KERNEL
);
406 memcpy(&elem
->st
, &env
->cur_state
, sizeof(env
->cur_state
));
407 elem
->insn_idx
= insn_idx
;
408 elem
->prev_insn_idx
= prev_insn_idx
;
409 elem
->next
= env
->head
;
412 if (env
->stack_size
> BPF_COMPLEXITY_LIMIT_STACK
) {
413 verbose("BPF program is too complex\n");
418 /* pop all elements and return */
419 while (pop_stack(env
, NULL
) >= 0);
423 #define CALLER_SAVED_REGS 6
424 static const int caller_saved
[CALLER_SAVED_REGS
] = {
425 BPF_REG_0
, BPF_REG_1
, BPF_REG_2
, BPF_REG_3
, BPF_REG_4
, BPF_REG_5
428 static void init_reg_state(struct bpf_reg_state
*regs
)
432 for (i
= 0; i
< MAX_BPF_REG
; i
++) {
433 regs
[i
].type
= NOT_INIT
;
435 regs
[i
].min_value
= BPF_REGISTER_MIN_RANGE
;
436 regs
[i
].max_value
= BPF_REGISTER_MAX_RANGE
;
440 regs
[BPF_REG_FP
].type
= FRAME_PTR
;
442 /* 1st arg to a function */
443 regs
[BPF_REG_1
].type
= PTR_TO_CTX
;
446 static void mark_reg_unknown_value(struct bpf_reg_state
*regs
, u32 regno
)
448 BUG_ON(regno
>= MAX_BPF_REG
);
449 regs
[regno
].type
= UNKNOWN_VALUE
;
453 static void reset_reg_range_values(struct bpf_reg_state
*regs
, u32 regno
)
455 regs
[regno
].min_value
= BPF_REGISTER_MIN_RANGE
;
456 regs
[regno
].max_value
= BPF_REGISTER_MAX_RANGE
;
460 SRC_OP
, /* register is used as source operand */
461 DST_OP
, /* register is used as destination operand */
462 DST_OP_NO_MARK
/* same as above, check only, don't mark */
465 static int check_reg_arg(struct bpf_reg_state
*regs
, u32 regno
,
468 if (regno
>= MAX_BPF_REG
) {
469 verbose("R%d is invalid\n", regno
);
474 /* check whether register used as source operand can be read */
475 if (regs
[regno
].type
== NOT_INIT
) {
476 verbose("R%d !read_ok\n", regno
);
480 /* check whether register used as dest operand can be written to */
481 if (regno
== BPF_REG_FP
) {
482 verbose("frame pointer is read only\n");
486 mark_reg_unknown_value(regs
, regno
);
491 static int bpf_size_to_bytes(int bpf_size
)
493 if (bpf_size
== BPF_W
)
495 else if (bpf_size
== BPF_H
)
497 else if (bpf_size
== BPF_B
)
499 else if (bpf_size
== BPF_DW
)
505 static bool is_spillable_regtype(enum bpf_reg_type type
)
508 case PTR_TO_MAP_VALUE
:
509 case PTR_TO_MAP_VALUE_OR_NULL
:
513 case PTR_TO_PACKET_END
:
515 case CONST_PTR_TO_MAP
:
522 /* check_stack_read/write functions track spill/fill of registers,
523 * stack boundary and alignment are checked in check_mem_access()
525 static int check_stack_write(struct bpf_verifier_state
*state
, int off
,
526 int size
, int value_regno
)
529 /* caller checked that off % size == 0 and -MAX_BPF_STACK <= off < 0,
530 * so it's aligned access and [off, off + size) are within stack limits
533 if (value_regno
>= 0 &&
534 is_spillable_regtype(state
->regs
[value_regno
].type
)) {
536 /* register containing pointer is being spilled into stack */
537 if (size
!= BPF_REG_SIZE
) {
538 verbose("invalid size of register spill\n");
542 /* save register state */
543 state
->spilled_regs
[(MAX_BPF_STACK
+ off
) / BPF_REG_SIZE
] =
544 state
->regs
[value_regno
];
546 for (i
= 0; i
< BPF_REG_SIZE
; i
++)
547 state
->stack_slot_type
[MAX_BPF_STACK
+ off
+ i
] = STACK_SPILL
;
549 /* regular write of data into stack */
550 state
->spilled_regs
[(MAX_BPF_STACK
+ off
) / BPF_REG_SIZE
] =
551 (struct bpf_reg_state
) {};
553 for (i
= 0; i
< size
; i
++)
554 state
->stack_slot_type
[MAX_BPF_STACK
+ off
+ i
] = STACK_MISC
;
559 static int check_stack_read(struct bpf_verifier_state
*state
, int off
, int size
,
565 slot_type
= &state
->stack_slot_type
[MAX_BPF_STACK
+ off
];
567 if (slot_type
[0] == STACK_SPILL
) {
568 if (size
!= BPF_REG_SIZE
) {
569 verbose("invalid size of register spill\n");
572 for (i
= 1; i
< BPF_REG_SIZE
; i
++) {
573 if (slot_type
[i
] != STACK_SPILL
) {
574 verbose("corrupted spill memory\n");
579 if (value_regno
>= 0)
580 /* restore register state from stack */
581 state
->regs
[value_regno
] =
582 state
->spilled_regs
[(MAX_BPF_STACK
+ off
) / BPF_REG_SIZE
];
585 for (i
= 0; i
< size
; i
++) {
586 if (slot_type
[i
] != STACK_MISC
) {
587 verbose("invalid read from stack off %d+%d size %d\n",
592 if (value_regno
>= 0)
593 /* have read misc data from the stack */
594 mark_reg_unknown_value(state
->regs
, value_regno
);
599 /* check read/write into map element returned by bpf_map_lookup_elem() */
600 static int check_map_access(struct bpf_verifier_env
*env
, u32 regno
, int off
,
603 struct bpf_map
*map
= env
->cur_state
.regs
[regno
].map_ptr
;
605 if (off
< 0 || off
+ size
> map
->value_size
) {
606 verbose("invalid access to map value, value_size=%d off=%d size=%d\n",
607 map
->value_size
, off
, size
);
613 #define MAX_PACKET_OFF 0xffff
615 static bool may_access_direct_pkt_data(struct bpf_verifier_env
*env
,
616 const struct bpf_call_arg_meta
*meta
)
618 switch (env
->prog
->type
) {
619 case BPF_PROG_TYPE_SCHED_CLS
:
620 case BPF_PROG_TYPE_SCHED_ACT
:
621 case BPF_PROG_TYPE_XDP
:
623 return meta
->pkt_access
;
625 env
->seen_direct_write
= true;
632 static int check_packet_access(struct bpf_verifier_env
*env
, u32 regno
, int off
,
635 struct bpf_reg_state
*regs
= env
->cur_state
.regs
;
636 struct bpf_reg_state
*reg
= ®s
[regno
];
639 if (off
< 0 || size
<= 0 || off
+ size
> reg
->range
) {
640 verbose("invalid access to packet, off=%d size=%d, R%d(id=%d,off=%d,r=%d)\n",
641 off
, size
, regno
, reg
->id
, reg
->off
, reg
->range
);
647 /* check access to 'struct bpf_context' fields */
648 static int check_ctx_access(struct bpf_verifier_env
*env
, int off
, int size
,
649 enum bpf_access_type t
, enum bpf_reg_type
*reg_type
)
651 /* for analyzer ctx accesses are already validated and converted */
652 if (env
->analyzer_ops
)
655 if (env
->prog
->aux
->ops
->is_valid_access
&&
656 env
->prog
->aux
->ops
->is_valid_access(off
, size
, t
, reg_type
)) {
657 /* remember the offset of last byte accessed in ctx */
658 if (env
->prog
->aux
->max_ctx_offset
< off
+ size
)
659 env
->prog
->aux
->max_ctx_offset
= off
+ size
;
663 verbose("invalid bpf_context access off=%d size=%d\n", off
, size
);
667 static bool is_pointer_value(struct bpf_verifier_env
*env
, int regno
)
669 if (env
->allow_ptr_leaks
)
672 switch (env
->cur_state
.regs
[regno
].type
) {
681 static int check_ptr_alignment(struct bpf_verifier_env
*env
,
682 struct bpf_reg_state
*reg
, int off
, int size
)
684 if (reg
->type
!= PTR_TO_PACKET
&& reg
->type
!= PTR_TO_MAP_VALUE_ADJ
) {
685 if (off
% size
!= 0) {
686 verbose("misaligned access off %d size %d\n",
694 if (IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS
))
695 /* misaligned access to packet is ok on x86,arm,arm64 */
698 if (reg
->id
&& size
!= 1) {
699 verbose("Unknown packet alignment. Only byte-sized access allowed\n");
703 /* skb->data is NET_IP_ALIGN-ed */
704 if (reg
->type
== PTR_TO_PACKET
&&
705 (NET_IP_ALIGN
+ reg
->off
+ off
) % size
!= 0) {
706 verbose("misaligned packet access off %d+%d+%d size %d\n",
707 NET_IP_ALIGN
, reg
->off
, off
, size
);
713 /* check whether memory at (regno + off) is accessible for t = (read | write)
714 * if t==write, value_regno is a register which value is stored into memory
715 * if t==read, value_regno is a register which will receive the value from memory
716 * if t==write && value_regno==-1, some unknown value is stored into memory
717 * if t==read && value_regno==-1, don't care what we read from memory
719 static int check_mem_access(struct bpf_verifier_env
*env
, u32 regno
, int off
,
720 int bpf_size
, enum bpf_access_type t
,
723 struct bpf_verifier_state
*state
= &env
->cur_state
;
724 struct bpf_reg_state
*reg
= &state
->regs
[regno
];
727 if (reg
->type
== PTR_TO_STACK
)
730 size
= bpf_size_to_bytes(bpf_size
);
734 err
= check_ptr_alignment(env
, reg
, off
, size
);
738 if (reg
->type
== PTR_TO_MAP_VALUE
||
739 reg
->type
== PTR_TO_MAP_VALUE_ADJ
) {
740 if (t
== BPF_WRITE
&& value_regno
>= 0 &&
741 is_pointer_value(env
, value_regno
)) {
742 verbose("R%d leaks addr into map\n", value_regno
);
746 /* If we adjusted the register to this map value at all then we
747 * need to change off and size to min_value and max_value
748 * respectively to make sure our theoretical access will be
751 if (reg
->type
== PTR_TO_MAP_VALUE_ADJ
) {
753 print_verifier_state(state
);
754 env
->varlen_map_value_access
= true;
755 /* The minimum value is only important with signed
756 * comparisons where we can't assume the floor of a
757 * value is 0. If we are using signed variables for our
758 * index'es we need to make sure that whatever we use
759 * will have a set floor within our range.
761 if (reg
->min_value
< 0) {
762 verbose("R%d min value is negative, either use unsigned index or do a if (index >=0) check.\n",
766 err
= check_map_access(env
, regno
, reg
->min_value
+ off
,
769 verbose("R%d min value is outside of the array range\n",
774 /* If we haven't set a max value then we need to bail
775 * since we can't be sure we won't do bad things.
777 if (reg
->max_value
== BPF_REGISTER_MAX_RANGE
) {
778 verbose("R%d unbounded memory access, make sure to bounds check any array access into a map\n",
782 off
+= reg
->max_value
;
784 err
= check_map_access(env
, regno
, off
, size
);
785 if (!err
&& t
== BPF_READ
&& value_regno
>= 0)
786 mark_reg_unknown_value(state
->regs
, value_regno
);
788 } else if (reg
->type
== PTR_TO_CTX
) {
789 enum bpf_reg_type reg_type
= UNKNOWN_VALUE
;
791 if (t
== BPF_WRITE
&& value_regno
>= 0 &&
792 is_pointer_value(env
, value_regno
)) {
793 verbose("R%d leaks addr into ctx\n", value_regno
);
796 err
= check_ctx_access(env
, off
, size
, t
, ®_type
);
797 if (!err
&& t
== BPF_READ
&& value_regno
>= 0) {
798 mark_reg_unknown_value(state
->regs
, value_regno
);
799 /* note that reg.[id|off|range] == 0 */
800 state
->regs
[value_regno
].type
= reg_type
;
803 } else if (reg
->type
== FRAME_PTR
|| reg
->type
== PTR_TO_STACK
) {
804 if (off
>= 0 || off
< -MAX_BPF_STACK
) {
805 verbose("invalid stack off=%d size=%d\n", off
, size
);
808 if (t
== BPF_WRITE
) {
809 if (!env
->allow_ptr_leaks
&&
810 state
->stack_slot_type
[MAX_BPF_STACK
+ off
] == STACK_SPILL
&&
811 size
!= BPF_REG_SIZE
) {
812 verbose("attempt to corrupt spilled pointer on stack\n");
815 err
= check_stack_write(state
, off
, size
, value_regno
);
817 err
= check_stack_read(state
, off
, size
, value_regno
);
819 } else if (state
->regs
[regno
].type
== PTR_TO_PACKET
) {
820 if (t
== BPF_WRITE
&& !may_access_direct_pkt_data(env
, NULL
)) {
821 verbose("cannot write into packet\n");
824 if (t
== BPF_WRITE
&& value_regno
>= 0 &&
825 is_pointer_value(env
, value_regno
)) {
826 verbose("R%d leaks addr into packet\n", value_regno
);
829 err
= check_packet_access(env
, regno
, off
, size
);
830 if (!err
&& t
== BPF_READ
&& value_regno
>= 0)
831 mark_reg_unknown_value(state
->regs
, value_regno
);
833 verbose("R%d invalid mem access '%s'\n",
834 regno
, reg_type_str
[reg
->type
]);
838 if (!err
&& size
<= 2 && value_regno
>= 0 && env
->allow_ptr_leaks
&&
839 state
->regs
[value_regno
].type
== UNKNOWN_VALUE
) {
840 /* 1 or 2 byte load zero-extends, determine the number of
841 * zero upper bits. Not doing it fo 4 byte load, since
842 * such values cannot be added to ptr_to_packet anyway.
844 state
->regs
[value_regno
].imm
= 64 - size
* 8;
849 static int check_xadd(struct bpf_verifier_env
*env
, struct bpf_insn
*insn
)
851 struct bpf_reg_state
*regs
= env
->cur_state
.regs
;
854 if ((BPF_SIZE(insn
->code
) != BPF_W
&& BPF_SIZE(insn
->code
) != BPF_DW
) ||
856 verbose("BPF_XADD uses reserved fields\n");
860 /* check src1 operand */
861 err
= check_reg_arg(regs
, insn
->src_reg
, SRC_OP
);
865 /* check src2 operand */
866 err
= check_reg_arg(regs
, insn
->dst_reg
, SRC_OP
);
870 /* check whether atomic_add can read the memory */
871 err
= check_mem_access(env
, insn
->dst_reg
, insn
->off
,
872 BPF_SIZE(insn
->code
), BPF_READ
, -1);
876 /* check whether atomic_add can write into the same memory */
877 return check_mem_access(env
, insn
->dst_reg
, insn
->off
,
878 BPF_SIZE(insn
->code
), BPF_WRITE
, -1);
881 /* when register 'regno' is passed into function that will read 'access_size'
882 * bytes from that pointer, make sure that it's within stack boundary
883 * and all elements of stack are initialized
885 static int check_stack_boundary(struct bpf_verifier_env
*env
, int regno
,
886 int access_size
, bool zero_size_allowed
,
887 struct bpf_call_arg_meta
*meta
)
889 struct bpf_verifier_state
*state
= &env
->cur_state
;
890 struct bpf_reg_state
*regs
= state
->regs
;
893 if (regs
[regno
].type
!= PTR_TO_STACK
) {
894 if (zero_size_allowed
&& access_size
== 0 &&
895 regs
[regno
].type
== CONST_IMM
&&
896 regs
[regno
].imm
== 0)
899 verbose("R%d type=%s expected=%s\n", regno
,
900 reg_type_str
[regs
[regno
].type
],
901 reg_type_str
[PTR_TO_STACK
]);
905 off
= regs
[regno
].imm
;
906 if (off
>= 0 || off
< -MAX_BPF_STACK
|| off
+ access_size
> 0 ||
908 verbose("invalid stack type R%d off=%d access_size=%d\n",
909 regno
, off
, access_size
);
913 if (meta
&& meta
->raw_mode
) {
914 meta
->access_size
= access_size
;
919 for (i
= 0; i
< access_size
; i
++) {
920 if (state
->stack_slot_type
[MAX_BPF_STACK
+ off
+ i
] != STACK_MISC
) {
921 verbose("invalid indirect read from stack off %d+%d size %d\n",
922 off
, i
, access_size
);
929 static int check_func_arg(struct bpf_verifier_env
*env
, u32 regno
,
930 enum bpf_arg_type arg_type
,
931 struct bpf_call_arg_meta
*meta
)
933 struct bpf_reg_state
*regs
= env
->cur_state
.regs
, *reg
= ®s
[regno
];
934 enum bpf_reg_type expected_type
, type
= reg
->type
;
937 if (arg_type
== ARG_DONTCARE
)
940 if (type
== NOT_INIT
) {
941 verbose("R%d !read_ok\n", regno
);
945 if (arg_type
== ARG_ANYTHING
) {
946 if (is_pointer_value(env
, regno
)) {
947 verbose("R%d leaks addr into helper function\n", regno
);
953 if (type
== PTR_TO_PACKET
&& !may_access_direct_pkt_data(env
, meta
)) {
954 verbose("helper access to the packet is not allowed\n");
958 if (arg_type
== ARG_PTR_TO_MAP_KEY
||
959 arg_type
== ARG_PTR_TO_MAP_VALUE
) {
960 expected_type
= PTR_TO_STACK
;
961 if (type
!= PTR_TO_PACKET
&& type
!= expected_type
)
963 } else if (arg_type
== ARG_CONST_STACK_SIZE
||
964 arg_type
== ARG_CONST_STACK_SIZE_OR_ZERO
) {
965 expected_type
= CONST_IMM
;
966 if (type
!= expected_type
)
968 } else if (arg_type
== ARG_CONST_MAP_PTR
) {
969 expected_type
= CONST_PTR_TO_MAP
;
970 if (type
!= expected_type
)
972 } else if (arg_type
== ARG_PTR_TO_CTX
) {
973 expected_type
= PTR_TO_CTX
;
974 if (type
!= expected_type
)
976 } else if (arg_type
== ARG_PTR_TO_STACK
||
977 arg_type
== ARG_PTR_TO_RAW_STACK
) {
978 expected_type
= PTR_TO_STACK
;
979 /* One exception here. In case function allows for NULL to be
980 * passed in as argument, it's a CONST_IMM type. Final test
981 * happens during stack boundary checking.
983 if (type
== CONST_IMM
&& reg
->imm
== 0)
984 /* final test in check_stack_boundary() */;
985 else if (type
!= PTR_TO_PACKET
&& type
!= expected_type
)
987 meta
->raw_mode
= arg_type
== ARG_PTR_TO_RAW_STACK
;
989 verbose("unsupported arg_type %d\n", arg_type
);
993 if (arg_type
== ARG_CONST_MAP_PTR
) {
994 /* bpf_map_xxx(map_ptr) call: remember that map_ptr */
995 meta
->map_ptr
= reg
->map_ptr
;
996 } else if (arg_type
== ARG_PTR_TO_MAP_KEY
) {
997 /* bpf_map_xxx(..., map_ptr, ..., key) call:
998 * check that [key, key + map->key_size) are within
999 * stack limits and initialized
1001 if (!meta
->map_ptr
) {
1002 /* in function declaration map_ptr must come before
1003 * map_key, so that it's verified and known before
1004 * we have to check map_key here. Otherwise it means
1005 * that kernel subsystem misconfigured verifier
1007 verbose("invalid map_ptr to access map->key\n");
1010 if (type
== PTR_TO_PACKET
)
1011 err
= check_packet_access(env
, regno
, 0,
1012 meta
->map_ptr
->key_size
);
1014 err
= check_stack_boundary(env
, regno
,
1015 meta
->map_ptr
->key_size
,
1017 } else if (arg_type
== ARG_PTR_TO_MAP_VALUE
) {
1018 /* bpf_map_xxx(..., map_ptr, ..., value) call:
1019 * check [value, value + map->value_size) validity
1021 if (!meta
->map_ptr
) {
1022 /* kernel subsystem misconfigured verifier */
1023 verbose("invalid map_ptr to access map->value\n");
1026 if (type
== PTR_TO_PACKET
)
1027 err
= check_packet_access(env
, regno
, 0,
1028 meta
->map_ptr
->value_size
);
1030 err
= check_stack_boundary(env
, regno
,
1031 meta
->map_ptr
->value_size
,
1033 } else if (arg_type
== ARG_CONST_STACK_SIZE
||
1034 arg_type
== ARG_CONST_STACK_SIZE_OR_ZERO
) {
1035 bool zero_size_allowed
= (arg_type
== ARG_CONST_STACK_SIZE_OR_ZERO
);
1037 /* bpf_xxx(..., buf, len) call will access 'len' bytes
1038 * from stack pointer 'buf'. Check it
1039 * note: regno == len, regno - 1 == buf
1042 /* kernel subsystem misconfigured verifier */
1043 verbose("ARG_CONST_STACK_SIZE cannot be first argument\n");
1046 if (regs
[regno
- 1].type
== PTR_TO_PACKET
)
1047 err
= check_packet_access(env
, regno
- 1, 0, reg
->imm
);
1049 err
= check_stack_boundary(env
, regno
- 1, reg
->imm
,
1050 zero_size_allowed
, meta
);
1055 verbose("R%d type=%s expected=%s\n", regno
,
1056 reg_type_str
[type
], reg_type_str
[expected_type
]);
1060 static int check_map_func_compatibility(struct bpf_map
*map
, int func_id
)
1065 /* We need a two way check, first is from map perspective ... */
1066 switch (map
->map_type
) {
1067 case BPF_MAP_TYPE_PROG_ARRAY
:
1068 if (func_id
!= BPF_FUNC_tail_call
)
1071 case BPF_MAP_TYPE_PERF_EVENT_ARRAY
:
1072 if (func_id
!= BPF_FUNC_perf_event_read
&&
1073 func_id
!= BPF_FUNC_perf_event_output
)
1076 case BPF_MAP_TYPE_STACK_TRACE
:
1077 if (func_id
!= BPF_FUNC_get_stackid
)
1080 case BPF_MAP_TYPE_CGROUP_ARRAY
:
1081 if (func_id
!= BPF_FUNC_skb_under_cgroup
&&
1082 func_id
!= BPF_FUNC_current_task_under_cgroup
)
1089 /* ... and second from the function itself. */
1091 case BPF_FUNC_tail_call
:
1092 if (map
->map_type
!= BPF_MAP_TYPE_PROG_ARRAY
)
1095 case BPF_FUNC_perf_event_read
:
1096 case BPF_FUNC_perf_event_output
:
1097 if (map
->map_type
!= BPF_MAP_TYPE_PERF_EVENT_ARRAY
)
1100 case BPF_FUNC_get_stackid
:
1101 if (map
->map_type
!= BPF_MAP_TYPE_STACK_TRACE
)
1104 case BPF_FUNC_current_task_under_cgroup
:
1105 case BPF_FUNC_skb_under_cgroup
:
1106 if (map
->map_type
!= BPF_MAP_TYPE_CGROUP_ARRAY
)
1115 verbose("cannot pass map_type %d into func %d\n",
1116 map
->map_type
, func_id
);
1120 static int check_raw_mode(const struct bpf_func_proto
*fn
)
1124 if (fn
->arg1_type
== ARG_PTR_TO_RAW_STACK
)
1126 if (fn
->arg2_type
== ARG_PTR_TO_RAW_STACK
)
1128 if (fn
->arg3_type
== ARG_PTR_TO_RAW_STACK
)
1130 if (fn
->arg4_type
== ARG_PTR_TO_RAW_STACK
)
1132 if (fn
->arg5_type
== ARG_PTR_TO_RAW_STACK
)
1135 return count
> 1 ? -EINVAL
: 0;
1138 static void clear_all_pkt_pointers(struct bpf_verifier_env
*env
)
1140 struct bpf_verifier_state
*state
= &env
->cur_state
;
1141 struct bpf_reg_state
*regs
= state
->regs
, *reg
;
1144 for (i
= 0; i
< MAX_BPF_REG
; i
++)
1145 if (regs
[i
].type
== PTR_TO_PACKET
||
1146 regs
[i
].type
== PTR_TO_PACKET_END
)
1147 mark_reg_unknown_value(regs
, i
);
1149 for (i
= 0; i
< MAX_BPF_STACK
; i
+= BPF_REG_SIZE
) {
1150 if (state
->stack_slot_type
[i
] != STACK_SPILL
)
1152 reg
= &state
->spilled_regs
[i
/ BPF_REG_SIZE
];
1153 if (reg
->type
!= PTR_TO_PACKET
&&
1154 reg
->type
!= PTR_TO_PACKET_END
)
1156 reg
->type
= UNKNOWN_VALUE
;
1161 static int check_call(struct bpf_verifier_env
*env
, int func_id
)
1163 struct bpf_verifier_state
*state
= &env
->cur_state
;
1164 const struct bpf_func_proto
*fn
= NULL
;
1165 struct bpf_reg_state
*regs
= state
->regs
;
1166 struct bpf_reg_state
*reg
;
1167 struct bpf_call_arg_meta meta
;
1171 /* find function prototype */
1172 if (func_id
< 0 || func_id
>= __BPF_FUNC_MAX_ID
) {
1173 verbose("invalid func %d\n", func_id
);
1177 if (env
->prog
->aux
->ops
->get_func_proto
)
1178 fn
= env
->prog
->aux
->ops
->get_func_proto(func_id
);
1181 verbose("unknown func %d\n", func_id
);
1185 /* eBPF programs must be GPL compatible to use GPL-ed functions */
1186 if (!env
->prog
->gpl_compatible
&& fn
->gpl_only
) {
1187 verbose("cannot call GPL only function from proprietary program\n");
1191 changes_data
= bpf_helper_changes_skb_data(fn
->func
);
1193 memset(&meta
, 0, sizeof(meta
));
1194 meta
.pkt_access
= fn
->pkt_access
;
1196 /* We only support one arg being in raw mode at the moment, which
1197 * is sufficient for the helper functions we have right now.
1199 err
= check_raw_mode(fn
);
1201 verbose("kernel subsystem misconfigured func %d\n", func_id
);
1206 err
= check_func_arg(env
, BPF_REG_1
, fn
->arg1_type
, &meta
);
1209 err
= check_func_arg(env
, BPF_REG_2
, fn
->arg2_type
, &meta
);
1212 err
= check_func_arg(env
, BPF_REG_3
, fn
->arg3_type
, &meta
);
1215 err
= check_func_arg(env
, BPF_REG_4
, fn
->arg4_type
, &meta
);
1218 err
= check_func_arg(env
, BPF_REG_5
, fn
->arg5_type
, &meta
);
1222 /* Mark slots with STACK_MISC in case of raw mode, stack offset
1223 * is inferred from register state.
1225 for (i
= 0; i
< meta
.access_size
; i
++) {
1226 err
= check_mem_access(env
, meta
.regno
, i
, BPF_B
, BPF_WRITE
, -1);
1231 /* reset caller saved regs */
1232 for (i
= 0; i
< CALLER_SAVED_REGS
; i
++) {
1233 reg
= regs
+ caller_saved
[i
];
1234 reg
->type
= NOT_INIT
;
1238 /* update return register */
1239 if (fn
->ret_type
== RET_INTEGER
) {
1240 regs
[BPF_REG_0
].type
= UNKNOWN_VALUE
;
1241 } else if (fn
->ret_type
== RET_VOID
) {
1242 regs
[BPF_REG_0
].type
= NOT_INIT
;
1243 } else if (fn
->ret_type
== RET_PTR_TO_MAP_VALUE_OR_NULL
) {
1244 regs
[BPF_REG_0
].type
= PTR_TO_MAP_VALUE_OR_NULL
;
1245 regs
[BPF_REG_0
].max_value
= regs
[BPF_REG_0
].min_value
= 0;
1246 /* remember map_ptr, so that check_map_access()
1247 * can check 'value_size' boundary of memory access
1248 * to map element returned from bpf_map_lookup_elem()
1250 if (meta
.map_ptr
== NULL
) {
1251 verbose("kernel subsystem misconfigured verifier\n");
1254 regs
[BPF_REG_0
].map_ptr
= meta
.map_ptr
;
1256 verbose("unknown return type %d of func %d\n",
1257 fn
->ret_type
, func_id
);
1261 err
= check_map_func_compatibility(meta
.map_ptr
, func_id
);
1266 clear_all_pkt_pointers(env
);
1270 static int check_packet_ptr_add(struct bpf_verifier_env
*env
,
1271 struct bpf_insn
*insn
)
1273 struct bpf_reg_state
*regs
= env
->cur_state
.regs
;
1274 struct bpf_reg_state
*dst_reg
= ®s
[insn
->dst_reg
];
1275 struct bpf_reg_state
*src_reg
= ®s
[insn
->src_reg
];
1276 struct bpf_reg_state tmp_reg
;
1279 if (BPF_SRC(insn
->code
) == BPF_K
) {
1280 /* pkt_ptr += imm */
1285 verbose("addition of negative constant to packet pointer is not allowed\n");
1288 if (imm
>= MAX_PACKET_OFF
||
1289 imm
+ dst_reg
->off
>= MAX_PACKET_OFF
) {
1290 verbose("constant %d is too large to add to packet pointer\n",
1294 /* a constant was added to pkt_ptr.
1295 * Remember it while keeping the same 'id'
1297 dst_reg
->off
+= imm
;
1299 if (src_reg
->type
== PTR_TO_PACKET
) {
1300 /* R6=pkt(id=0,off=0,r=62) R7=imm22; r7 += r6 */
1301 tmp_reg
= *dst_reg
; /* save r7 state */
1302 *dst_reg
= *src_reg
; /* copy pkt_ptr state r6 into r7 */
1303 src_reg
= &tmp_reg
; /* pretend it's src_reg state */
1304 /* if the checks below reject it, the copy won't matter,
1305 * since we're rejecting the whole program. If all ok,
1306 * then imm22 state will be added to r7
1307 * and r7 will be pkt(id=0,off=22,r=62) while
1308 * r6 will stay as pkt(id=0,off=0,r=62)
1312 if (src_reg
->type
== CONST_IMM
) {
1313 /* pkt_ptr += reg where reg is known constant */
1317 /* disallow pkt_ptr += reg
1318 * if reg is not uknown_value with guaranteed zero upper bits
1319 * otherwise pkt_ptr may overflow and addition will become
1320 * subtraction which is not allowed
1322 if (src_reg
->type
!= UNKNOWN_VALUE
) {
1323 verbose("cannot add '%s' to ptr_to_packet\n",
1324 reg_type_str
[src_reg
->type
]);
1327 if (src_reg
->imm
< 48) {
1328 verbose("cannot add integer value with %lld upper zero bits to ptr_to_packet\n",
1332 /* dst_reg stays as pkt_ptr type and since some positive
1333 * integer value was added to the pointer, increment its 'id'
1335 dst_reg
->id
= ++env
->id_gen
;
1337 /* something was added to pkt_ptr, set range and off to zero */
1344 static int evaluate_reg_alu(struct bpf_verifier_env
*env
, struct bpf_insn
*insn
)
1346 struct bpf_reg_state
*regs
= env
->cur_state
.regs
;
1347 struct bpf_reg_state
*dst_reg
= ®s
[insn
->dst_reg
];
1348 u8 opcode
= BPF_OP(insn
->code
);
1351 /* for type == UNKNOWN_VALUE:
1352 * imm > 0 -> number of zero upper bits
1353 * imm == 0 -> don't track which is the same as all bits can be non-zero
1356 if (BPF_SRC(insn
->code
) == BPF_X
) {
1357 struct bpf_reg_state
*src_reg
= ®s
[insn
->src_reg
];
1359 if (src_reg
->type
== UNKNOWN_VALUE
&& src_reg
->imm
> 0 &&
1360 dst_reg
->imm
&& opcode
== BPF_ADD
) {
1362 * where both have zero upper bits. Adding them
1363 * can only result making one more bit non-zero
1364 * in the larger value.
1365 * Ex. 0xffff (imm=48) + 1 (imm=63) = 0x10000 (imm=47)
1366 * 0xffff (imm=48) + 0xffff = 0x1fffe (imm=47)
1368 dst_reg
->imm
= min(dst_reg
->imm
, src_reg
->imm
);
1372 if (src_reg
->type
== CONST_IMM
&& src_reg
->imm
> 0 &&
1373 dst_reg
->imm
&& opcode
== BPF_ADD
) {
1375 * where dreg has zero upper bits and sreg is const.
1376 * Adding them can only result making one more bit
1377 * non-zero in the larger value.
1379 imm_log2
= __ilog2_u64((long long)src_reg
->imm
);
1380 dst_reg
->imm
= min(dst_reg
->imm
, 63 - imm_log2
);
1384 /* all other cases non supported yet, just mark dst_reg */
1389 /* sign extend 32-bit imm into 64-bit to make sure that
1390 * negative values occupy bit 63. Note ilog2() would have
1391 * been incorrect, since sizeof(insn->imm) == 4
1393 imm_log2
= __ilog2_u64((long long)insn
->imm
);
1395 if (dst_reg
->imm
&& opcode
== BPF_LSH
) {
1397 * if reg was a result of 2 byte load, then its imm == 48
1398 * which means that upper 48 bits are zero and shifting this reg
1399 * left by 4 would mean that upper 44 bits are still zero
1401 dst_reg
->imm
-= insn
->imm
;
1402 } else if (dst_reg
->imm
&& opcode
== BPF_MUL
) {
1404 * if multiplying by 14 subtract 4
1405 * This is conservative calculation of upper zero bits.
1406 * It's not trying to special case insn->imm == 1 or 0 cases
1408 dst_reg
->imm
-= imm_log2
+ 1;
1409 } else if (opcode
== BPF_AND
) {
1411 dst_reg
->imm
= 63 - imm_log2
;
1412 } else if (dst_reg
->imm
&& opcode
== BPF_ADD
) {
1414 dst_reg
->imm
= min(dst_reg
->imm
, 63 - imm_log2
);
1416 } else if (opcode
== BPF_RSH
) {
1418 * which means that after right shift, upper bits will be zero
1419 * note that verifier already checked that
1420 * 0 <= imm < 64 for shift insn
1422 dst_reg
->imm
+= insn
->imm
;
1423 if (unlikely(dst_reg
->imm
> 64))
1424 /* some dumb code did:
1427 * and all bits are zero now */
1430 /* all other alu ops, means that we don't know what will
1431 * happen to the value, mark it with unknown number of zero bits
1436 if (dst_reg
->imm
< 0) {
1437 /* all 64 bits of the register can contain non-zero bits
1438 * and such value cannot be added to ptr_to_packet, since it
1439 * may overflow, mark it as unknown to avoid further eval
1446 static int evaluate_reg_imm_alu(struct bpf_verifier_env
*env
,
1447 struct bpf_insn
*insn
)
1449 struct bpf_reg_state
*regs
= env
->cur_state
.regs
;
1450 struct bpf_reg_state
*dst_reg
= ®s
[insn
->dst_reg
];
1451 struct bpf_reg_state
*src_reg
= ®s
[insn
->src_reg
];
1452 u8 opcode
= BPF_OP(insn
->code
);
1454 /* dst_reg->type == CONST_IMM here, simulate execution of 'add' insn.
1455 * Don't care about overflow or negative values, just add them
1457 if (opcode
== BPF_ADD
&& BPF_SRC(insn
->code
) == BPF_K
)
1458 dst_reg
->imm
+= insn
->imm
;
1459 else if (opcode
== BPF_ADD
&& BPF_SRC(insn
->code
) == BPF_X
&&
1460 src_reg
->type
== CONST_IMM
)
1461 dst_reg
->imm
+= src_reg
->imm
;
1463 mark_reg_unknown_value(regs
, insn
->dst_reg
);
1467 static void check_reg_overflow(struct bpf_reg_state
*reg
)
1469 if (reg
->max_value
> BPF_REGISTER_MAX_RANGE
)
1470 reg
->max_value
= BPF_REGISTER_MAX_RANGE
;
1471 if (reg
->min_value
< BPF_REGISTER_MIN_RANGE
||
1472 reg
->min_value
> BPF_REGISTER_MAX_RANGE
)
1473 reg
->min_value
= BPF_REGISTER_MIN_RANGE
;
1476 static void adjust_reg_min_max_vals(struct bpf_verifier_env
*env
,
1477 struct bpf_insn
*insn
)
1479 struct bpf_reg_state
*regs
= env
->cur_state
.regs
, *dst_reg
;
1480 s64 min_val
= BPF_REGISTER_MIN_RANGE
;
1481 u64 max_val
= BPF_REGISTER_MAX_RANGE
;
1482 bool min_set
= false, max_set
= false;
1483 u8 opcode
= BPF_OP(insn
->code
);
1485 dst_reg
= ®s
[insn
->dst_reg
];
1486 if (BPF_SRC(insn
->code
) == BPF_X
) {
1487 check_reg_overflow(®s
[insn
->src_reg
]);
1488 min_val
= regs
[insn
->src_reg
].min_value
;
1489 max_val
= regs
[insn
->src_reg
].max_value
;
1491 /* If the source register is a random pointer then the
1492 * min_value/max_value values represent the range of the known
1493 * accesses into that value, not the actual min/max value of the
1494 * register itself. In this case we have to reset the reg range
1495 * values so we know it is not safe to look at.
1497 if (regs
[insn
->src_reg
].type
!= CONST_IMM
&&
1498 regs
[insn
->src_reg
].type
!= UNKNOWN_VALUE
) {
1499 min_val
= BPF_REGISTER_MIN_RANGE
;
1500 max_val
= BPF_REGISTER_MAX_RANGE
;
1502 } else if (insn
->imm
< BPF_REGISTER_MAX_RANGE
&&
1503 (s64
)insn
->imm
> BPF_REGISTER_MIN_RANGE
) {
1504 min_val
= max_val
= insn
->imm
;
1505 min_set
= max_set
= true;
1508 /* We don't know anything about what was done to this register, mark it
1511 if (min_val
== BPF_REGISTER_MIN_RANGE
&&
1512 max_val
== BPF_REGISTER_MAX_RANGE
) {
1513 reset_reg_range_values(regs
, insn
->dst_reg
);
1517 /* If one of our values was at the end of our ranges then we can't just
1518 * do our normal operations to the register, we need to set the values
1519 * to the min/max since they are undefined.
1521 if (min_val
== BPF_REGISTER_MIN_RANGE
)
1522 dst_reg
->min_value
= BPF_REGISTER_MIN_RANGE
;
1523 if (max_val
== BPF_REGISTER_MAX_RANGE
)
1524 dst_reg
->max_value
= BPF_REGISTER_MAX_RANGE
;
1528 if (dst_reg
->min_value
!= BPF_REGISTER_MIN_RANGE
)
1529 dst_reg
->min_value
+= min_val
;
1530 if (dst_reg
->max_value
!= BPF_REGISTER_MAX_RANGE
)
1531 dst_reg
->max_value
+= max_val
;
1534 if (dst_reg
->min_value
!= BPF_REGISTER_MIN_RANGE
)
1535 dst_reg
->min_value
-= min_val
;
1536 if (dst_reg
->max_value
!= BPF_REGISTER_MAX_RANGE
)
1537 dst_reg
->max_value
-= max_val
;
1540 if (dst_reg
->min_value
!= BPF_REGISTER_MIN_RANGE
)
1541 dst_reg
->min_value
*= min_val
;
1542 if (dst_reg
->max_value
!= BPF_REGISTER_MAX_RANGE
)
1543 dst_reg
->max_value
*= max_val
;
1546 /* Disallow AND'ing of negative numbers, ain't nobody got time
1547 * for that. Otherwise the minimum is 0 and the max is the max
1548 * value we could AND against.
1551 dst_reg
->min_value
= BPF_REGISTER_MIN_RANGE
;
1553 dst_reg
->min_value
= 0;
1554 dst_reg
->max_value
= max_val
;
1557 /* Gotta have special overflow logic here, if we're shifting
1558 * more than MAX_RANGE then just assume we have an invalid
1561 if (min_val
> ilog2(BPF_REGISTER_MAX_RANGE
))
1562 dst_reg
->min_value
= BPF_REGISTER_MIN_RANGE
;
1563 else if (dst_reg
->min_value
!= BPF_REGISTER_MIN_RANGE
)
1564 dst_reg
->min_value
<<= min_val
;
1566 if (max_val
> ilog2(BPF_REGISTER_MAX_RANGE
))
1567 dst_reg
->max_value
= BPF_REGISTER_MAX_RANGE
;
1568 else if (dst_reg
->max_value
!= BPF_REGISTER_MAX_RANGE
)
1569 dst_reg
->max_value
<<= max_val
;
1572 /* RSH by a negative number is undefined, and the BPF_RSH is an
1573 * unsigned shift, so make the appropriate casts.
1575 if (min_val
< 0 || dst_reg
->min_value
< 0)
1576 dst_reg
->min_value
= BPF_REGISTER_MIN_RANGE
;
1578 dst_reg
->min_value
=
1579 (u64
)(dst_reg
->min_value
) >> min_val
;
1580 if (dst_reg
->max_value
!= BPF_REGISTER_MAX_RANGE
)
1581 dst_reg
->max_value
>>= max_val
;
1584 reset_reg_range_values(regs
, insn
->dst_reg
);
1588 check_reg_overflow(dst_reg
);
1591 /* check validity of 32-bit and 64-bit arithmetic operations */
1592 static int check_alu_op(struct bpf_verifier_env
*env
, struct bpf_insn
*insn
)
1594 struct bpf_reg_state
*regs
= env
->cur_state
.regs
, *dst_reg
;
1595 u8 opcode
= BPF_OP(insn
->code
);
1598 if (opcode
== BPF_END
|| opcode
== BPF_NEG
) {
1599 if (opcode
== BPF_NEG
) {
1600 if (BPF_SRC(insn
->code
) != 0 ||
1601 insn
->src_reg
!= BPF_REG_0
||
1602 insn
->off
!= 0 || insn
->imm
!= 0) {
1603 verbose("BPF_NEG uses reserved fields\n");
1607 if (insn
->src_reg
!= BPF_REG_0
|| insn
->off
!= 0 ||
1608 (insn
->imm
!= 16 && insn
->imm
!= 32 && insn
->imm
!= 64)) {
1609 verbose("BPF_END uses reserved fields\n");
1614 /* check src operand */
1615 err
= check_reg_arg(regs
, insn
->dst_reg
, SRC_OP
);
1619 if (is_pointer_value(env
, insn
->dst_reg
)) {
1620 verbose("R%d pointer arithmetic prohibited\n",
1625 /* check dest operand */
1626 err
= check_reg_arg(regs
, insn
->dst_reg
, DST_OP
);
1630 } else if (opcode
== BPF_MOV
) {
1632 if (BPF_SRC(insn
->code
) == BPF_X
) {
1633 if (insn
->imm
!= 0 || insn
->off
!= 0) {
1634 verbose("BPF_MOV uses reserved fields\n");
1638 /* check src operand */
1639 err
= check_reg_arg(regs
, insn
->src_reg
, SRC_OP
);
1643 if (insn
->src_reg
!= BPF_REG_0
|| insn
->off
!= 0) {
1644 verbose("BPF_MOV uses reserved fields\n");
1649 /* check dest operand */
1650 err
= check_reg_arg(regs
, insn
->dst_reg
, DST_OP
);
1654 /* we are setting our register to something new, we need to
1655 * reset its range values.
1657 reset_reg_range_values(regs
, insn
->dst_reg
);
1659 if (BPF_SRC(insn
->code
) == BPF_X
) {
1660 if (BPF_CLASS(insn
->code
) == BPF_ALU64
) {
1662 * copy register state to dest reg
1664 regs
[insn
->dst_reg
] = regs
[insn
->src_reg
];
1666 if (is_pointer_value(env
, insn
->src_reg
)) {
1667 verbose("R%d partial copy of pointer\n",
1671 regs
[insn
->dst_reg
].type
= UNKNOWN_VALUE
;
1672 regs
[insn
->dst_reg
].map_ptr
= NULL
;
1676 * remember the value we stored into this reg
1678 regs
[insn
->dst_reg
].type
= CONST_IMM
;
1679 regs
[insn
->dst_reg
].imm
= insn
->imm
;
1680 regs
[insn
->dst_reg
].max_value
= insn
->imm
;
1681 regs
[insn
->dst_reg
].min_value
= insn
->imm
;
1684 } else if (opcode
> BPF_END
) {
1685 verbose("invalid BPF_ALU opcode %x\n", opcode
);
1688 } else { /* all other ALU ops: and, sub, xor, add, ... */
1690 if (BPF_SRC(insn
->code
) == BPF_X
) {
1691 if (insn
->imm
!= 0 || insn
->off
!= 0) {
1692 verbose("BPF_ALU uses reserved fields\n");
1695 /* check src1 operand */
1696 err
= check_reg_arg(regs
, insn
->src_reg
, SRC_OP
);
1700 if (insn
->src_reg
!= BPF_REG_0
|| insn
->off
!= 0) {
1701 verbose("BPF_ALU uses reserved fields\n");
1706 /* check src2 operand */
1707 err
= check_reg_arg(regs
, insn
->dst_reg
, SRC_OP
);
1711 if ((opcode
== BPF_MOD
|| opcode
== BPF_DIV
) &&
1712 BPF_SRC(insn
->code
) == BPF_K
&& insn
->imm
== 0) {
1713 verbose("div by zero\n");
1717 if ((opcode
== BPF_LSH
|| opcode
== BPF_RSH
||
1718 opcode
== BPF_ARSH
) && BPF_SRC(insn
->code
) == BPF_K
) {
1719 int size
= BPF_CLASS(insn
->code
) == BPF_ALU64
? 64 : 32;
1721 if (insn
->imm
< 0 || insn
->imm
>= size
) {
1722 verbose("invalid shift %d\n", insn
->imm
);
1727 /* check dest operand */
1728 err
= check_reg_arg(regs
, insn
->dst_reg
, DST_OP_NO_MARK
);
1732 dst_reg
= ®s
[insn
->dst_reg
];
1734 /* first we want to adjust our ranges. */
1735 adjust_reg_min_max_vals(env
, insn
);
1737 /* pattern match 'bpf_add Rx, imm' instruction */
1738 if (opcode
== BPF_ADD
&& BPF_CLASS(insn
->code
) == BPF_ALU64
&&
1739 dst_reg
->type
== FRAME_PTR
&& BPF_SRC(insn
->code
) == BPF_K
) {
1740 dst_reg
->type
= PTR_TO_STACK
;
1741 dst_reg
->imm
= insn
->imm
;
1743 } else if (opcode
== BPF_ADD
&&
1744 BPF_CLASS(insn
->code
) == BPF_ALU64
&&
1745 (dst_reg
->type
== PTR_TO_PACKET
||
1746 (BPF_SRC(insn
->code
) == BPF_X
&&
1747 regs
[insn
->src_reg
].type
== PTR_TO_PACKET
))) {
1748 /* ptr_to_packet += K|X */
1749 return check_packet_ptr_add(env
, insn
);
1750 } else if (BPF_CLASS(insn
->code
) == BPF_ALU64
&&
1751 dst_reg
->type
== UNKNOWN_VALUE
&&
1752 env
->allow_ptr_leaks
) {
1753 /* unknown += K|X */
1754 return evaluate_reg_alu(env
, insn
);
1755 } else if (BPF_CLASS(insn
->code
) == BPF_ALU64
&&
1756 dst_reg
->type
== CONST_IMM
&&
1757 env
->allow_ptr_leaks
) {
1758 /* reg_imm += K|X */
1759 return evaluate_reg_imm_alu(env
, insn
);
1760 } else if (is_pointer_value(env
, insn
->dst_reg
)) {
1761 verbose("R%d pointer arithmetic prohibited\n",
1764 } else if (BPF_SRC(insn
->code
) == BPF_X
&&
1765 is_pointer_value(env
, insn
->src_reg
)) {
1766 verbose("R%d pointer arithmetic prohibited\n",
1771 /* If we did pointer math on a map value then just set it to our
1772 * PTR_TO_MAP_VALUE_ADJ type so we can deal with any stores or
1773 * loads to this register appropriately, otherwise just mark the
1774 * register as unknown.
1776 if (env
->allow_ptr_leaks
&&
1777 (dst_reg
->type
== PTR_TO_MAP_VALUE
||
1778 dst_reg
->type
== PTR_TO_MAP_VALUE_ADJ
))
1779 dst_reg
->type
= PTR_TO_MAP_VALUE_ADJ
;
1781 mark_reg_unknown_value(regs
, insn
->dst_reg
);
1787 static void find_good_pkt_pointers(struct bpf_verifier_state
*state
,
1788 struct bpf_reg_state
*dst_reg
)
1790 struct bpf_reg_state
*regs
= state
->regs
, *reg
;
1793 /* LLVM can generate two kind of checks:
1799 * if (r2 > pkt_end) goto <handle exception>
1803 * r2 == dst_reg, pkt_end == src_reg
1804 * r2=pkt(id=n,off=8,r=0)
1805 * r3=pkt(id=n,off=0,r=0)
1811 * if (pkt_end >= r2) goto <access okay>
1812 * <handle exception>
1815 * pkt_end == dst_reg, r2 == src_reg
1816 * r2=pkt(id=n,off=8,r=0)
1817 * r3=pkt(id=n,off=0,r=0)
1819 * Find register r3 and mark its range as r3=pkt(id=n,off=0,r=8)
1820 * so that range of bytes [r3, r3 + 8) is safe to access.
1823 for (i
= 0; i
< MAX_BPF_REG
; i
++)
1824 if (regs
[i
].type
== PTR_TO_PACKET
&& regs
[i
].id
== dst_reg
->id
)
1825 regs
[i
].range
= dst_reg
->off
;
1827 for (i
= 0; i
< MAX_BPF_STACK
; i
+= BPF_REG_SIZE
) {
1828 if (state
->stack_slot_type
[i
] != STACK_SPILL
)
1830 reg
= &state
->spilled_regs
[i
/ BPF_REG_SIZE
];
1831 if (reg
->type
== PTR_TO_PACKET
&& reg
->id
== dst_reg
->id
)
1832 reg
->range
= dst_reg
->off
;
1836 /* Adjusts the register min/max values in the case that the dst_reg is the
1837 * variable register that we are working on, and src_reg is a constant or we're
1838 * simply doing a BPF_K check.
1840 static void reg_set_min_max(struct bpf_reg_state
*true_reg
,
1841 struct bpf_reg_state
*false_reg
, u64 val
,
1846 /* If this is false then we know nothing Jon Snow, but if it is
1847 * true then we know for sure.
1849 true_reg
->max_value
= true_reg
->min_value
= val
;
1852 /* If this is true we know nothing Jon Snow, but if it is false
1853 * we know the value for sure;
1855 false_reg
->max_value
= false_reg
->min_value
= val
;
1858 /* Unsigned comparison, the minimum value is 0. */
1859 false_reg
->min_value
= 0;
1861 /* If this is false then we know the maximum val is val,
1862 * otherwise we know the min val is val+1.
1864 false_reg
->max_value
= val
;
1865 true_reg
->min_value
= val
+ 1;
1868 /* Unsigned comparison, the minimum value is 0. */
1869 false_reg
->min_value
= 0;
1871 /* If this is false then we know the maximum value is val - 1,
1872 * otherwise we know the mimimum value is val.
1874 false_reg
->max_value
= val
- 1;
1875 true_reg
->min_value
= val
;
1881 check_reg_overflow(false_reg
);
1882 check_reg_overflow(true_reg
);
1885 /* Same as above, but for the case that dst_reg is a CONST_IMM reg and src_reg
1886 * is the variable reg.
1888 static void reg_set_min_max_inv(struct bpf_reg_state
*true_reg
,
1889 struct bpf_reg_state
*false_reg
, u64 val
,
1894 /* If this is false then we know nothing Jon Snow, but if it is
1895 * true then we know for sure.
1897 true_reg
->max_value
= true_reg
->min_value
= val
;
1900 /* If this is true we know nothing Jon Snow, but if it is false
1901 * we know the value for sure;
1903 false_reg
->max_value
= false_reg
->min_value
= val
;
1906 /* Unsigned comparison, the minimum value is 0. */
1907 true_reg
->min_value
= 0;
1910 * If this is false, then the val is <= the register, if it is
1911 * true the register <= to the val.
1913 false_reg
->min_value
= val
;
1914 true_reg
->max_value
= val
- 1;
1917 /* Unsigned comparison, the minimum value is 0. */
1918 true_reg
->min_value
= 0;
1920 /* If this is false then constant < register, if it is true then
1921 * the register < constant.
1923 false_reg
->min_value
= val
+ 1;
1924 true_reg
->max_value
= val
;
1930 check_reg_overflow(false_reg
);
1931 check_reg_overflow(true_reg
);
1934 static int check_cond_jmp_op(struct bpf_verifier_env
*env
,
1935 struct bpf_insn
*insn
, int *insn_idx
)
1937 struct bpf_verifier_state
*other_branch
, *this_branch
= &env
->cur_state
;
1938 struct bpf_reg_state
*regs
= this_branch
->regs
, *dst_reg
;
1939 u8 opcode
= BPF_OP(insn
->code
);
1942 if (opcode
> BPF_EXIT
) {
1943 verbose("invalid BPF_JMP opcode %x\n", opcode
);
1947 if (BPF_SRC(insn
->code
) == BPF_X
) {
1948 if (insn
->imm
!= 0) {
1949 verbose("BPF_JMP uses reserved fields\n");
1953 /* check src1 operand */
1954 err
= check_reg_arg(regs
, insn
->src_reg
, SRC_OP
);
1958 if (is_pointer_value(env
, insn
->src_reg
)) {
1959 verbose("R%d pointer comparison prohibited\n",
1964 if (insn
->src_reg
!= BPF_REG_0
) {
1965 verbose("BPF_JMP uses reserved fields\n");
1970 /* check src2 operand */
1971 err
= check_reg_arg(regs
, insn
->dst_reg
, SRC_OP
);
1975 dst_reg
= ®s
[insn
->dst_reg
];
1977 /* detect if R == 0 where R was initialized to zero earlier */
1978 if (BPF_SRC(insn
->code
) == BPF_K
&&
1979 (opcode
== BPF_JEQ
|| opcode
== BPF_JNE
) &&
1980 dst_reg
->type
== CONST_IMM
&& dst_reg
->imm
== insn
->imm
) {
1981 if (opcode
== BPF_JEQ
) {
1982 /* if (imm == imm) goto pc+off;
1983 * only follow the goto, ignore fall-through
1985 *insn_idx
+= insn
->off
;
1988 /* if (imm != imm) goto pc+off;
1989 * only follow fall-through branch, since
1990 * that's where the program will go
1996 other_branch
= push_stack(env
, *insn_idx
+ insn
->off
+ 1, *insn_idx
);
2000 /* detect if we are comparing against a constant value so we can adjust
2001 * our min/max values for our dst register.
2003 if (BPF_SRC(insn
->code
) == BPF_X
) {
2004 if (regs
[insn
->src_reg
].type
== CONST_IMM
)
2005 reg_set_min_max(&other_branch
->regs
[insn
->dst_reg
],
2006 dst_reg
, regs
[insn
->src_reg
].imm
,
2008 else if (dst_reg
->type
== CONST_IMM
)
2009 reg_set_min_max_inv(&other_branch
->regs
[insn
->src_reg
],
2010 ®s
[insn
->src_reg
], dst_reg
->imm
,
2013 reg_set_min_max(&other_branch
->regs
[insn
->dst_reg
],
2014 dst_reg
, insn
->imm
, opcode
);
2017 /* detect if R == 0 where R is returned from bpf_map_lookup_elem() */
2018 if (BPF_SRC(insn
->code
) == BPF_K
&&
2019 insn
->imm
== 0 && (opcode
== BPF_JEQ
|| opcode
== BPF_JNE
) &&
2020 dst_reg
->type
== PTR_TO_MAP_VALUE_OR_NULL
) {
2021 if (opcode
== BPF_JEQ
) {
2022 /* next fallthrough insn can access memory via
2025 regs
[insn
->dst_reg
].type
= PTR_TO_MAP_VALUE
;
2026 /* branch targer cannot access it, since reg == 0 */
2027 mark_reg_unknown_value(other_branch
->regs
,
2030 other_branch
->regs
[insn
->dst_reg
].type
= PTR_TO_MAP_VALUE
;
2031 mark_reg_unknown_value(regs
, insn
->dst_reg
);
2033 } else if (BPF_SRC(insn
->code
) == BPF_X
&& opcode
== BPF_JGT
&&
2034 dst_reg
->type
== PTR_TO_PACKET
&&
2035 regs
[insn
->src_reg
].type
== PTR_TO_PACKET_END
) {
2036 find_good_pkt_pointers(this_branch
, dst_reg
);
2037 } else if (BPF_SRC(insn
->code
) == BPF_X
&& opcode
== BPF_JGE
&&
2038 dst_reg
->type
== PTR_TO_PACKET_END
&&
2039 regs
[insn
->src_reg
].type
== PTR_TO_PACKET
) {
2040 find_good_pkt_pointers(other_branch
, ®s
[insn
->src_reg
]);
2041 } else if (is_pointer_value(env
, insn
->dst_reg
)) {
2042 verbose("R%d pointer comparison prohibited\n", insn
->dst_reg
);
2046 print_verifier_state(this_branch
);
2050 /* return the map pointer stored inside BPF_LD_IMM64 instruction */
2051 static struct bpf_map
*ld_imm64_to_map_ptr(struct bpf_insn
*insn
)
2053 u64 imm64
= ((u64
) (u32
) insn
[0].imm
) | ((u64
) (u32
) insn
[1].imm
) << 32;
2055 return (struct bpf_map
*) (unsigned long) imm64
;
2058 /* verify BPF_LD_IMM64 instruction */
2059 static int check_ld_imm(struct bpf_verifier_env
*env
, struct bpf_insn
*insn
)
2061 struct bpf_reg_state
*regs
= env
->cur_state
.regs
;
2064 if (BPF_SIZE(insn
->code
) != BPF_DW
) {
2065 verbose("invalid BPF_LD_IMM insn\n");
2068 if (insn
->off
!= 0) {
2069 verbose("BPF_LD_IMM64 uses reserved fields\n");
2073 err
= check_reg_arg(regs
, insn
->dst_reg
, DST_OP
);
2077 if (insn
->src_reg
== 0) {
2078 /* generic move 64-bit immediate into a register,
2079 * only analyzer needs to collect the ld_imm value.
2081 u64 imm
= ((u64
)(insn
+ 1)->imm
<< 32) | (u32
)insn
->imm
;
2083 if (!env
->analyzer_ops
)
2086 regs
[insn
->dst_reg
].type
= CONST_IMM
;
2087 regs
[insn
->dst_reg
].imm
= imm
;
2091 /* replace_map_fd_with_map_ptr() should have caught bad ld_imm64 */
2092 BUG_ON(insn
->src_reg
!= BPF_PSEUDO_MAP_FD
);
2094 regs
[insn
->dst_reg
].type
= CONST_PTR_TO_MAP
;
2095 regs
[insn
->dst_reg
].map_ptr
= ld_imm64_to_map_ptr(insn
);
2099 static bool may_access_skb(enum bpf_prog_type type
)
2102 case BPF_PROG_TYPE_SOCKET_FILTER
:
2103 case BPF_PROG_TYPE_SCHED_CLS
:
2104 case BPF_PROG_TYPE_SCHED_ACT
:
2111 /* verify safety of LD_ABS|LD_IND instructions:
2112 * - they can only appear in the programs where ctx == skb
2113 * - since they are wrappers of function calls, they scratch R1-R5 registers,
2114 * preserve R6-R9, and store return value into R0
2117 * ctx == skb == R6 == CTX
2120 * SRC == any register
2121 * IMM == 32-bit immediate
2124 * R0 - 8/16/32-bit skb data converted to cpu endianness
2126 static int check_ld_abs(struct bpf_verifier_env
*env
, struct bpf_insn
*insn
)
2128 struct bpf_reg_state
*regs
= env
->cur_state
.regs
;
2129 u8 mode
= BPF_MODE(insn
->code
);
2130 struct bpf_reg_state
*reg
;
2133 if (!may_access_skb(env
->prog
->type
)) {
2134 verbose("BPF_LD_[ABS|IND] instructions not allowed for this program type\n");
2138 if (insn
->dst_reg
!= BPF_REG_0
|| insn
->off
!= 0 ||
2139 BPF_SIZE(insn
->code
) == BPF_DW
||
2140 (mode
== BPF_ABS
&& insn
->src_reg
!= BPF_REG_0
)) {
2141 verbose("BPF_LD_[ABS|IND] uses reserved fields\n");
2145 /* check whether implicit source operand (register R6) is readable */
2146 err
= check_reg_arg(regs
, BPF_REG_6
, SRC_OP
);
2150 if (regs
[BPF_REG_6
].type
!= PTR_TO_CTX
) {
2151 verbose("at the time of BPF_LD_ABS|IND R6 != pointer to skb\n");
2155 if (mode
== BPF_IND
) {
2156 /* check explicit source operand */
2157 err
= check_reg_arg(regs
, insn
->src_reg
, SRC_OP
);
2162 /* reset caller saved regs to unreadable */
2163 for (i
= 0; i
< CALLER_SAVED_REGS
; i
++) {
2164 reg
= regs
+ caller_saved
[i
];
2165 reg
->type
= NOT_INIT
;
2169 /* mark destination R0 register as readable, since it contains
2170 * the value fetched from the packet
2172 regs
[BPF_REG_0
].type
= UNKNOWN_VALUE
;
2176 /* non-recursive DFS pseudo code
2177 * 1 procedure DFS-iterative(G,v):
2178 * 2 label v as discovered
2179 * 3 let S be a stack
2181 * 5 while S is not empty
2183 * 7 if t is what we're looking for:
2185 * 9 for all edges e in G.adjacentEdges(t) do
2186 * 10 if edge e is already labelled
2187 * 11 continue with the next edge
2188 * 12 w <- G.adjacentVertex(t,e)
2189 * 13 if vertex w is not discovered and not explored
2190 * 14 label e as tree-edge
2191 * 15 label w as discovered
2194 * 18 else if vertex w is discovered
2195 * 19 label e as back-edge
2197 * 21 // vertex w is explored
2198 * 22 label e as forward- or cross-edge
2199 * 23 label t as explored
2204 * 0x11 - discovered and fall-through edge labelled
2205 * 0x12 - discovered and fall-through and branch edges labelled
2216 #define STATE_LIST_MARK ((struct bpf_verifier_state_list *) -1L)
2218 static int *insn_stack
; /* stack of insns to process */
2219 static int cur_stack
; /* current stack index */
2220 static int *insn_state
;
2222 /* t, w, e - match pseudo-code above:
2223 * t - index of current instruction
2224 * w - next instruction
2227 static int push_insn(int t
, int w
, int e
, struct bpf_verifier_env
*env
)
2229 if (e
== FALLTHROUGH
&& insn_state
[t
] >= (DISCOVERED
| FALLTHROUGH
))
2232 if (e
== BRANCH
&& insn_state
[t
] >= (DISCOVERED
| BRANCH
))
2235 if (w
< 0 || w
>= env
->prog
->len
) {
2236 verbose("jump out of range from insn %d to %d\n", t
, w
);
2241 /* mark branch target for state pruning */
2242 env
->explored_states
[w
] = STATE_LIST_MARK
;
2244 if (insn_state
[w
] == 0) {
2246 insn_state
[t
] = DISCOVERED
| e
;
2247 insn_state
[w
] = DISCOVERED
;
2248 if (cur_stack
>= env
->prog
->len
)
2250 insn_stack
[cur_stack
++] = w
;
2252 } else if ((insn_state
[w
] & 0xF0) == DISCOVERED
) {
2253 verbose("back-edge from insn %d to %d\n", t
, w
);
2255 } else if (insn_state
[w
] == EXPLORED
) {
2256 /* forward- or cross-edge */
2257 insn_state
[t
] = DISCOVERED
| e
;
2259 verbose("insn state internal bug\n");
2265 /* non-recursive depth-first-search to detect loops in BPF program
2266 * loop == back-edge in directed graph
2268 static int check_cfg(struct bpf_verifier_env
*env
)
2270 struct bpf_insn
*insns
= env
->prog
->insnsi
;
2271 int insn_cnt
= env
->prog
->len
;
2275 insn_state
= kcalloc(insn_cnt
, sizeof(int), GFP_KERNEL
);
2279 insn_stack
= kcalloc(insn_cnt
, sizeof(int), GFP_KERNEL
);
2285 insn_state
[0] = DISCOVERED
; /* mark 1st insn as discovered */
2286 insn_stack
[0] = 0; /* 0 is the first instruction */
2292 t
= insn_stack
[cur_stack
- 1];
2294 if (BPF_CLASS(insns
[t
].code
) == BPF_JMP
) {
2295 u8 opcode
= BPF_OP(insns
[t
].code
);
2297 if (opcode
== BPF_EXIT
) {
2299 } else if (opcode
== BPF_CALL
) {
2300 ret
= push_insn(t
, t
+ 1, FALLTHROUGH
, env
);
2305 if (t
+ 1 < insn_cnt
)
2306 env
->explored_states
[t
+ 1] = STATE_LIST_MARK
;
2307 } else if (opcode
== BPF_JA
) {
2308 if (BPF_SRC(insns
[t
].code
) != BPF_K
) {
2312 /* unconditional jump with single edge */
2313 ret
= push_insn(t
, t
+ insns
[t
].off
+ 1,
2319 /* tell verifier to check for equivalent states
2320 * after every call and jump
2322 if (t
+ 1 < insn_cnt
)
2323 env
->explored_states
[t
+ 1] = STATE_LIST_MARK
;
2325 /* conditional jump with two edges */
2326 ret
= push_insn(t
, t
+ 1, FALLTHROUGH
, env
);
2332 ret
= push_insn(t
, t
+ insns
[t
].off
+ 1, BRANCH
, env
);
2339 /* all other non-branch instructions with single
2342 ret
= push_insn(t
, t
+ 1, FALLTHROUGH
, env
);
2350 insn_state
[t
] = EXPLORED
;
2351 if (cur_stack
-- <= 0) {
2352 verbose("pop stack internal bug\n");
2359 for (i
= 0; i
< insn_cnt
; i
++) {
2360 if (insn_state
[i
] != EXPLORED
) {
2361 verbose("unreachable insn %d\n", i
);
2366 ret
= 0; /* cfg looks good */
2374 /* the following conditions reduce the number of explored insns
2375 * from ~140k to ~80k for ultra large programs that use a lot of ptr_to_packet
2377 static bool compare_ptrs_to_packet(struct bpf_reg_state
*old
,
2378 struct bpf_reg_state
*cur
)
2380 if (old
->id
!= cur
->id
)
2383 /* old ptr_to_packet is more conservative, since it allows smaller
2385 * old(off=0,r=10) is equal to cur(off=0,r=20), because
2386 * old(off=0,r=10) means that with range=10 the verifier proceeded
2387 * further and found no issues with the program. Now we're in the same
2388 * spot with cur(off=0,r=20), so we're safe too, since anything further
2389 * will only be looking at most 10 bytes after this pointer.
2391 if (old
->off
== cur
->off
&& old
->range
< cur
->range
)
2394 /* old(off=20,r=10) is equal to cur(off=22,re=22 or 5 or 0)
2395 * since both cannot be used for packet access and safe(old)
2396 * pointer has smaller off that could be used for further
2397 * 'if (ptr > data_end)' check
2399 * old(off=20,r=10) and cur(off=22,r=22) and cur(off=22,r=0) mean
2400 * that we cannot access the packet.
2401 * The safe range is:
2402 * [ptr, ptr + range - off)
2403 * so whenever off >=range, it means no safe bytes from this pointer.
2404 * When comparing old->off <= cur->off, it means that older code
2405 * went with smaller offset and that offset was later
2406 * used to figure out the safe range after 'if (ptr > data_end)' check
2407 * Say, 'old' state was explored like:
2408 * ... R3(off=0, r=0)
2410 * ... now R4(off=20,r=0) <-- here
2411 * if (R4 > data_end)
2412 * ... R4(off=20,r=20), R3(off=0,r=20) and R3 can be used to access.
2413 * ... the code further went all the way to bpf_exit.
2414 * Now the 'cur' state at the mark 'here' has R4(off=30,r=0).
2415 * old_R4(off=20,r=0) equal to cur_R4(off=30,r=0), since if the verifier
2416 * goes further, such cur_R4 will give larger safe packet range after
2417 * 'if (R4 > data_end)' and all further insn were already good with r=20,
2418 * so they will be good with r=30 and we can prune the search.
2420 if (old
->off
<= cur
->off
&&
2421 old
->off
>= old
->range
&& cur
->off
>= cur
->range
)
2427 /* compare two verifier states
2429 * all states stored in state_list are known to be valid, since
2430 * verifier reached 'bpf_exit' instruction through them
2432 * this function is called when verifier exploring different branches of
2433 * execution popped from the state stack. If it sees an old state that has
2434 * more strict register state and more strict stack state then this execution
2435 * branch doesn't need to be explored further, since verifier already
2436 * concluded that more strict state leads to valid finish.
2438 * Therefore two states are equivalent if register state is more conservative
2439 * and explored stack state is more conservative than the current one.
2442 * (slot1=INV slot2=MISC) == (slot1=MISC slot2=MISC)
2443 * (slot1=MISC slot2=MISC) != (slot1=INV slot2=MISC)
2445 * In other words if current stack state (one being explored) has more
2446 * valid slots than old one that already passed validation, it means
2447 * the verifier can stop exploring and conclude that current state is valid too
2449 * Similarly with registers. If explored state has register type as invalid
2450 * whereas register type in current state is meaningful, it means that
2451 * the current state will reach 'bpf_exit' instruction safely
2453 static bool states_equal(struct bpf_verifier_env
*env
,
2454 struct bpf_verifier_state
*old
,
2455 struct bpf_verifier_state
*cur
)
2457 bool varlen_map_access
= env
->varlen_map_value_access
;
2458 struct bpf_reg_state
*rold
, *rcur
;
2461 for (i
= 0; i
< MAX_BPF_REG
; i
++) {
2462 rold
= &old
->regs
[i
];
2463 rcur
= &cur
->regs
[i
];
2465 if (memcmp(rold
, rcur
, sizeof(*rold
)) == 0)
2468 /* If the ranges were not the same, but everything else was and
2469 * we didn't do a variable access into a map then we are a-ok.
2471 if (!varlen_map_access
&&
2472 rold
->type
== rcur
->type
&& rold
->imm
== rcur
->imm
)
2475 /* If we didn't map access then again we don't care about the
2476 * mismatched range values and it's ok if our old type was
2477 * UNKNOWN and we didn't go to a NOT_INIT'ed reg.
2479 if (rold
->type
== NOT_INIT
||
2480 (!varlen_map_access
&& rold
->type
== UNKNOWN_VALUE
&&
2481 rcur
->type
!= NOT_INIT
))
2484 if (rold
->type
== PTR_TO_PACKET
&& rcur
->type
== PTR_TO_PACKET
&&
2485 compare_ptrs_to_packet(rold
, rcur
))
2491 for (i
= 0; i
< MAX_BPF_STACK
; i
++) {
2492 if (old
->stack_slot_type
[i
] == STACK_INVALID
)
2494 if (old
->stack_slot_type
[i
] != cur
->stack_slot_type
[i
])
2495 /* Ex: old explored (safe) state has STACK_SPILL in
2496 * this stack slot, but current has has STACK_MISC ->
2497 * this verifier states are not equivalent,
2498 * return false to continue verification of this path
2501 if (i
% BPF_REG_SIZE
)
2503 if (memcmp(&old
->spilled_regs
[i
/ BPF_REG_SIZE
],
2504 &cur
->spilled_regs
[i
/ BPF_REG_SIZE
],
2505 sizeof(old
->spilled_regs
[0])))
2506 /* when explored and current stack slot types are
2507 * the same, check that stored pointers types
2508 * are the same as well.
2509 * Ex: explored safe path could have stored
2510 * (bpf_reg_state) {.type = PTR_TO_STACK, .imm = -8}
2511 * but current path has stored:
2512 * (bpf_reg_state) {.type = PTR_TO_STACK, .imm = -16}
2513 * such verifier states are not equivalent.
2514 * return false to continue verification of this path
2523 static int is_state_visited(struct bpf_verifier_env
*env
, int insn_idx
)
2525 struct bpf_verifier_state_list
*new_sl
;
2526 struct bpf_verifier_state_list
*sl
;
2528 sl
= env
->explored_states
[insn_idx
];
2530 /* this 'insn_idx' instruction wasn't marked, so we will not
2531 * be doing state search here
2535 while (sl
!= STATE_LIST_MARK
) {
2536 if (states_equal(env
, &sl
->state
, &env
->cur_state
))
2537 /* reached equivalent register/stack state,
2544 /* there were no equivalent states, remember current one.
2545 * technically the current state is not proven to be safe yet,
2546 * but it will either reach bpf_exit (which means it's safe) or
2547 * it will be rejected. Since there are no loops, we won't be
2548 * seeing this 'insn_idx' instruction again on the way to bpf_exit
2550 new_sl
= kmalloc(sizeof(struct bpf_verifier_state_list
), GFP_USER
);
2554 /* add new state to the head of linked list */
2555 memcpy(&new_sl
->state
, &env
->cur_state
, sizeof(env
->cur_state
));
2556 new_sl
->next
= env
->explored_states
[insn_idx
];
2557 env
->explored_states
[insn_idx
] = new_sl
;
2561 static int ext_analyzer_insn_hook(struct bpf_verifier_env
*env
,
2562 int insn_idx
, int prev_insn_idx
)
2564 if (!env
->analyzer_ops
|| !env
->analyzer_ops
->insn_hook
)
2567 return env
->analyzer_ops
->insn_hook(env
, insn_idx
, prev_insn_idx
);
2570 static int do_check(struct bpf_verifier_env
*env
)
2572 struct bpf_verifier_state
*state
= &env
->cur_state
;
2573 struct bpf_insn
*insns
= env
->prog
->insnsi
;
2574 struct bpf_reg_state
*regs
= state
->regs
;
2575 int insn_cnt
= env
->prog
->len
;
2576 int insn_idx
, prev_insn_idx
= 0;
2577 int insn_processed
= 0;
2578 bool do_print_state
= false;
2580 init_reg_state(regs
);
2582 env
->varlen_map_value_access
= false;
2584 struct bpf_insn
*insn
;
2588 if (insn_idx
>= insn_cnt
) {
2589 verbose("invalid insn idx %d insn_cnt %d\n",
2590 insn_idx
, insn_cnt
);
2594 insn
= &insns
[insn_idx
];
2595 class = BPF_CLASS(insn
->code
);
2597 if (++insn_processed
> BPF_COMPLEXITY_LIMIT_INSNS
) {
2598 verbose("BPF program is too large. Proccessed %d insn\n",
2603 err
= is_state_visited(env
, insn_idx
);
2607 /* found equivalent state, can prune the search */
2610 verbose("\nfrom %d to %d: safe\n",
2611 prev_insn_idx
, insn_idx
);
2613 verbose("%d: safe\n", insn_idx
);
2615 goto process_bpf_exit
;
2618 if (log_level
&& do_print_state
) {
2619 verbose("\nfrom %d to %d:", prev_insn_idx
, insn_idx
);
2620 print_verifier_state(&env
->cur_state
);
2621 do_print_state
= false;
2625 verbose("%d: ", insn_idx
);
2626 print_bpf_insn(insn
);
2629 err
= ext_analyzer_insn_hook(env
, insn_idx
, prev_insn_idx
);
2633 if (class == BPF_ALU
|| class == BPF_ALU64
) {
2634 err
= check_alu_op(env
, insn
);
2638 } else if (class == BPF_LDX
) {
2639 enum bpf_reg_type
*prev_src_type
, src_reg_type
;
2641 /* check for reserved fields is already done */
2643 /* check src operand */
2644 err
= check_reg_arg(regs
, insn
->src_reg
, SRC_OP
);
2648 err
= check_reg_arg(regs
, insn
->dst_reg
, DST_OP_NO_MARK
);
2652 src_reg_type
= regs
[insn
->src_reg
].type
;
2654 /* check that memory (src_reg + off) is readable,
2655 * the state of dst_reg will be updated by this func
2657 err
= check_mem_access(env
, insn
->src_reg
, insn
->off
,
2658 BPF_SIZE(insn
->code
), BPF_READ
,
2663 reset_reg_range_values(regs
, insn
->dst_reg
);
2664 if (BPF_SIZE(insn
->code
) != BPF_W
&&
2665 BPF_SIZE(insn
->code
) != BPF_DW
) {
2670 prev_src_type
= &env
->insn_aux_data
[insn_idx
].ptr_type
;
2672 if (*prev_src_type
== NOT_INIT
) {
2674 * dst_reg = *(u32 *)(src_reg + off)
2675 * save type to validate intersecting paths
2677 *prev_src_type
= src_reg_type
;
2679 } else if (src_reg_type
!= *prev_src_type
&&
2680 (src_reg_type
== PTR_TO_CTX
||
2681 *prev_src_type
== PTR_TO_CTX
)) {
2682 /* ABuser program is trying to use the same insn
2683 * dst_reg = *(u32*) (src_reg + off)
2684 * with different pointer types:
2685 * src_reg == ctx in one branch and
2686 * src_reg == stack|map in some other branch.
2689 verbose("same insn cannot be used with different pointers\n");
2693 } else if (class == BPF_STX
) {
2694 enum bpf_reg_type
*prev_dst_type
, dst_reg_type
;
2696 if (BPF_MODE(insn
->code
) == BPF_XADD
) {
2697 err
= check_xadd(env
, insn
);
2704 /* check src1 operand */
2705 err
= check_reg_arg(regs
, insn
->src_reg
, SRC_OP
);
2708 /* check src2 operand */
2709 err
= check_reg_arg(regs
, insn
->dst_reg
, SRC_OP
);
2713 dst_reg_type
= regs
[insn
->dst_reg
].type
;
2715 /* check that memory (dst_reg + off) is writeable */
2716 err
= check_mem_access(env
, insn
->dst_reg
, insn
->off
,
2717 BPF_SIZE(insn
->code
), BPF_WRITE
,
2722 prev_dst_type
= &env
->insn_aux_data
[insn_idx
].ptr_type
;
2724 if (*prev_dst_type
== NOT_INIT
) {
2725 *prev_dst_type
= dst_reg_type
;
2726 } else if (dst_reg_type
!= *prev_dst_type
&&
2727 (dst_reg_type
== PTR_TO_CTX
||
2728 *prev_dst_type
== PTR_TO_CTX
)) {
2729 verbose("same insn cannot be used with different pointers\n");
2733 } else if (class == BPF_ST
) {
2734 if (BPF_MODE(insn
->code
) != BPF_MEM
||
2735 insn
->src_reg
!= BPF_REG_0
) {
2736 verbose("BPF_ST uses reserved fields\n");
2739 /* check src operand */
2740 err
= check_reg_arg(regs
, insn
->dst_reg
, SRC_OP
);
2744 /* check that memory (dst_reg + off) is writeable */
2745 err
= check_mem_access(env
, insn
->dst_reg
, insn
->off
,
2746 BPF_SIZE(insn
->code
), BPF_WRITE
,
2751 } else if (class == BPF_JMP
) {
2752 u8 opcode
= BPF_OP(insn
->code
);
2754 if (opcode
== BPF_CALL
) {
2755 if (BPF_SRC(insn
->code
) != BPF_K
||
2757 insn
->src_reg
!= BPF_REG_0
||
2758 insn
->dst_reg
!= BPF_REG_0
) {
2759 verbose("BPF_CALL uses reserved fields\n");
2763 err
= check_call(env
, insn
->imm
);
2767 } else if (opcode
== BPF_JA
) {
2768 if (BPF_SRC(insn
->code
) != BPF_K
||
2770 insn
->src_reg
!= BPF_REG_0
||
2771 insn
->dst_reg
!= BPF_REG_0
) {
2772 verbose("BPF_JA uses reserved fields\n");
2776 insn_idx
+= insn
->off
+ 1;
2779 } else if (opcode
== BPF_EXIT
) {
2780 if (BPF_SRC(insn
->code
) != BPF_K
||
2782 insn
->src_reg
!= BPF_REG_0
||
2783 insn
->dst_reg
!= BPF_REG_0
) {
2784 verbose("BPF_EXIT uses reserved fields\n");
2788 /* eBPF calling convetion is such that R0 is used
2789 * to return the value from eBPF program.
2790 * Make sure that it's readable at this time
2791 * of bpf_exit, which means that program wrote
2792 * something into it earlier
2794 err
= check_reg_arg(regs
, BPF_REG_0
, SRC_OP
);
2798 if (is_pointer_value(env
, BPF_REG_0
)) {
2799 verbose("R0 leaks addr as return value\n");
2804 insn_idx
= pop_stack(env
, &prev_insn_idx
);
2808 do_print_state
= true;
2812 err
= check_cond_jmp_op(env
, insn
, &insn_idx
);
2816 } else if (class == BPF_LD
) {
2817 u8 mode
= BPF_MODE(insn
->code
);
2819 if (mode
== BPF_ABS
|| mode
== BPF_IND
) {
2820 err
= check_ld_abs(env
, insn
);
2824 } else if (mode
== BPF_IMM
) {
2825 err
= check_ld_imm(env
, insn
);
2831 verbose("invalid BPF_LD mode\n");
2834 reset_reg_range_values(regs
, insn
->dst_reg
);
2836 verbose("unknown insn class %d\n", class);
2843 verbose("processed %d insns\n", insn_processed
);
2847 static int check_map_prog_compatibility(struct bpf_map
*map
,
2848 struct bpf_prog
*prog
)
2851 if (prog
->type
== BPF_PROG_TYPE_PERF_EVENT
&&
2852 (map
->map_type
== BPF_MAP_TYPE_HASH
||
2853 map
->map_type
== BPF_MAP_TYPE_PERCPU_HASH
) &&
2854 (map
->map_flags
& BPF_F_NO_PREALLOC
)) {
2855 verbose("perf_event programs can only use preallocated hash map\n");
2861 /* look for pseudo eBPF instructions that access map FDs and
2862 * replace them with actual map pointers
2864 static int replace_map_fd_with_map_ptr(struct bpf_verifier_env
*env
)
2866 struct bpf_insn
*insn
= env
->prog
->insnsi
;
2867 int insn_cnt
= env
->prog
->len
;
2870 for (i
= 0; i
< insn_cnt
; i
++, insn
++) {
2871 if (BPF_CLASS(insn
->code
) == BPF_LDX
&&
2872 (BPF_MODE(insn
->code
) != BPF_MEM
|| insn
->imm
!= 0)) {
2873 verbose("BPF_LDX uses reserved fields\n");
2877 if (BPF_CLASS(insn
->code
) == BPF_STX
&&
2878 ((BPF_MODE(insn
->code
) != BPF_MEM
&&
2879 BPF_MODE(insn
->code
) != BPF_XADD
) || insn
->imm
!= 0)) {
2880 verbose("BPF_STX uses reserved fields\n");
2884 if (insn
[0].code
== (BPF_LD
| BPF_IMM
| BPF_DW
)) {
2885 struct bpf_map
*map
;
2888 if (i
== insn_cnt
- 1 || insn
[1].code
!= 0 ||
2889 insn
[1].dst_reg
!= 0 || insn
[1].src_reg
!= 0 ||
2891 verbose("invalid bpf_ld_imm64 insn\n");
2895 if (insn
->src_reg
== 0)
2896 /* valid generic load 64-bit imm */
2899 if (insn
->src_reg
!= BPF_PSEUDO_MAP_FD
) {
2900 verbose("unrecognized bpf_ld_imm64 insn\n");
2904 f
= fdget(insn
->imm
);
2905 map
= __bpf_map_get(f
);
2907 verbose("fd %d is not pointing to valid bpf_map\n",
2909 return PTR_ERR(map
);
2912 err
= check_map_prog_compatibility(map
, env
->prog
);
2918 /* store map pointer inside BPF_LD_IMM64 instruction */
2919 insn
[0].imm
= (u32
) (unsigned long) map
;
2920 insn
[1].imm
= ((u64
) (unsigned long) map
) >> 32;
2922 /* check whether we recorded this map already */
2923 for (j
= 0; j
< env
->used_map_cnt
; j
++)
2924 if (env
->used_maps
[j
] == map
) {
2929 if (env
->used_map_cnt
>= MAX_USED_MAPS
) {
2934 /* hold the map. If the program is rejected by verifier,
2935 * the map will be released by release_maps() or it
2936 * will be used by the valid program until it's unloaded
2937 * and all maps are released in free_bpf_prog_info()
2939 map
= bpf_map_inc(map
, false);
2942 return PTR_ERR(map
);
2944 env
->used_maps
[env
->used_map_cnt
++] = map
;
2953 /* now all pseudo BPF_LD_IMM64 instructions load valid
2954 * 'struct bpf_map *' into a register instead of user map_fd.
2955 * These pointers will be used later by verifier to validate map access.
2960 /* drop refcnt of maps used by the rejected program */
2961 static void release_maps(struct bpf_verifier_env
*env
)
2965 for (i
= 0; i
< env
->used_map_cnt
; i
++)
2966 bpf_map_put(env
->used_maps
[i
]);
2969 /* convert pseudo BPF_LD_IMM64 into generic BPF_LD_IMM64 */
2970 static void convert_pseudo_ld_imm64(struct bpf_verifier_env
*env
)
2972 struct bpf_insn
*insn
= env
->prog
->insnsi
;
2973 int insn_cnt
= env
->prog
->len
;
2976 for (i
= 0; i
< insn_cnt
; i
++, insn
++)
2977 if (insn
->code
== (BPF_LD
| BPF_IMM
| BPF_DW
))
2981 /* convert load instructions that access fields of 'struct __sk_buff'
2982 * into sequence of instructions that access fields of 'struct sk_buff'
2984 static int convert_ctx_accesses(struct bpf_verifier_env
*env
)
2986 const struct bpf_verifier_ops
*ops
= env
->prog
->aux
->ops
;
2987 const int insn_cnt
= env
->prog
->len
;
2988 struct bpf_insn insn_buf
[16], *insn
;
2989 struct bpf_prog
*new_prog
;
2990 enum bpf_access_type type
;
2991 int i
, cnt
, delta
= 0;
2993 if (ops
->gen_prologue
) {
2994 cnt
= ops
->gen_prologue(insn_buf
, env
->seen_direct_write
,
2996 if (cnt
>= ARRAY_SIZE(insn_buf
)) {
2997 verbose("bpf verifier is misconfigured\n");
3000 new_prog
= bpf_patch_insn_single(env
->prog
, 0,
3004 env
->prog
= new_prog
;
3009 if (!ops
->convert_ctx_access
)
3012 insn
= env
->prog
->insnsi
+ delta
;
3014 for (i
= 0; i
< insn_cnt
; i
++, insn
++) {
3015 if (insn
->code
== (BPF_LDX
| BPF_MEM
| BPF_W
) ||
3016 insn
->code
== (BPF_LDX
| BPF_MEM
| BPF_DW
))
3018 else if (insn
->code
== (BPF_STX
| BPF_MEM
| BPF_W
) ||
3019 insn
->code
== (BPF_STX
| BPF_MEM
| BPF_DW
))
3024 if (env
->insn_aux_data
[i
].ptr_type
!= PTR_TO_CTX
)
3027 cnt
= ops
->convert_ctx_access(type
, insn
->dst_reg
, insn
->src_reg
,
3028 insn
->off
, insn_buf
, env
->prog
);
3029 if (cnt
== 0 || cnt
>= ARRAY_SIZE(insn_buf
)) {
3030 verbose("bpf verifier is misconfigured\n");
3034 new_prog
= bpf_patch_insn_single(env
->prog
, i
+ delta
, insn_buf
,
3041 /* keep walking new program and skip insns we just inserted */
3042 env
->prog
= new_prog
;
3043 insn
= new_prog
->insnsi
+ i
+ delta
;
3049 static void free_states(struct bpf_verifier_env
*env
)
3051 struct bpf_verifier_state_list
*sl
, *sln
;
3054 if (!env
->explored_states
)
3057 for (i
= 0; i
< env
->prog
->len
; i
++) {
3058 sl
= env
->explored_states
[i
];
3061 while (sl
!= STATE_LIST_MARK
) {
3068 kfree(env
->explored_states
);
3071 int bpf_check(struct bpf_prog
**prog
, union bpf_attr
*attr
)
3073 char __user
*log_ubuf
= NULL
;
3074 struct bpf_verifier_env
*env
;
3077 if ((*prog
)->len
<= 0 || (*prog
)->len
> BPF_MAXINSNS
)
3080 /* 'struct bpf_verifier_env' can be global, but since it's not small,
3081 * allocate/free it every time bpf_check() is called
3083 env
= kzalloc(sizeof(struct bpf_verifier_env
), GFP_KERNEL
);
3087 env
->insn_aux_data
= vzalloc(sizeof(struct bpf_insn_aux_data
) *
3090 if (!env
->insn_aux_data
)
3094 /* grab the mutex to protect few globals used by verifier */
3095 mutex_lock(&bpf_verifier_lock
);
3097 if (attr
->log_level
|| attr
->log_buf
|| attr
->log_size
) {
3098 /* user requested verbose verifier output
3099 * and supplied buffer to store the verification trace
3101 log_level
= attr
->log_level
;
3102 log_ubuf
= (char __user
*) (unsigned long) attr
->log_buf
;
3103 log_size
= attr
->log_size
;
3107 /* log_* values have to be sane */
3108 if (log_size
< 128 || log_size
> UINT_MAX
>> 8 ||
3109 log_level
== 0 || log_ubuf
== NULL
)
3113 log_buf
= vmalloc(log_size
);
3120 ret
= replace_map_fd_with_map_ptr(env
);
3122 goto skip_full_check
;
3124 env
->explored_states
= kcalloc(env
->prog
->len
,
3125 sizeof(struct bpf_verifier_state_list
*),
3128 if (!env
->explored_states
)
3129 goto skip_full_check
;
3131 ret
= check_cfg(env
);
3133 goto skip_full_check
;
3135 env
->allow_ptr_leaks
= capable(CAP_SYS_ADMIN
);
3137 ret
= do_check(env
);
3140 while (pop_stack(env
, NULL
) >= 0);
3144 /* program is valid, convert *(u32*)(ctx + off) accesses */
3145 ret
= convert_ctx_accesses(env
);
3147 if (log_level
&& log_len
>= log_size
- 1) {
3148 BUG_ON(log_len
>= log_size
);
3149 /* verifier log exceeded user supplied buffer */
3151 /* fall through to return what was recorded */
3154 /* copy verifier log back to user space including trailing zero */
3155 if (log_level
&& copy_to_user(log_ubuf
, log_buf
, log_len
+ 1) != 0) {
3160 if (ret
== 0 && env
->used_map_cnt
) {
3161 /* if program passed verifier, update used_maps in bpf_prog_info */
3162 env
->prog
->aux
->used_maps
= kmalloc_array(env
->used_map_cnt
,
3163 sizeof(env
->used_maps
[0]),
3166 if (!env
->prog
->aux
->used_maps
) {
3171 memcpy(env
->prog
->aux
->used_maps
, env
->used_maps
,
3172 sizeof(env
->used_maps
[0]) * env
->used_map_cnt
);
3173 env
->prog
->aux
->used_map_cnt
= env
->used_map_cnt
;
3175 /* program is valid. Convert pseudo bpf_ld_imm64 into generic
3176 * bpf_ld_imm64 instructions
3178 convert_pseudo_ld_imm64(env
);
3184 if (!env
->prog
->aux
->used_maps
)
3185 /* if we didn't copy map pointers into bpf_prog_info, release
3186 * them now. Otherwise free_bpf_prog_info() will release them.
3191 mutex_unlock(&bpf_verifier_lock
);
3192 vfree(env
->insn_aux_data
);
3198 int bpf_analyzer(struct bpf_prog
*prog
, const struct bpf_ext_analyzer_ops
*ops
,
3201 struct bpf_verifier_env
*env
;
3204 env
= kzalloc(sizeof(struct bpf_verifier_env
), GFP_KERNEL
);
3208 env
->insn_aux_data
= vzalloc(sizeof(struct bpf_insn_aux_data
) *
3211 if (!env
->insn_aux_data
)
3214 env
->analyzer_ops
= ops
;
3215 env
->analyzer_priv
= priv
;
3217 /* grab the mutex to protect few globals used by verifier */
3218 mutex_lock(&bpf_verifier_lock
);
3222 env
->explored_states
= kcalloc(env
->prog
->len
,
3223 sizeof(struct bpf_verifier_state_list
*),
3226 if (!env
->explored_states
)
3227 goto skip_full_check
;
3229 ret
= check_cfg(env
);
3231 goto skip_full_check
;
3233 env
->allow_ptr_leaks
= capable(CAP_SYS_ADMIN
);
3235 ret
= do_check(env
);
3238 while (pop_stack(env
, NULL
) >= 0);
3241 mutex_unlock(&bpf_verifier_lock
);
3242 vfree(env
->insn_aux_data
);
3247 EXPORT_SYMBOL_GPL(bpf_analyzer
);