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=%llu",
238 (unsigned 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 BUG_ON(regno
>= MAX_BPF_REG
);
468 regs
[regno
].type
= UNKNOWN_VALUE
;
473 static void reset_reg_range_values(struct bpf_reg_state
*regs
, u32 regno
)
475 regs
[regno
].min_value
= BPF_REGISTER_MIN_RANGE
;
476 regs
[regno
].max_value
= BPF_REGISTER_MAX_RANGE
;
480 SRC_OP
, /* register is used as source operand */
481 DST_OP
, /* register is used as destination operand */
482 DST_OP_NO_MARK
/* same as above, check only, don't mark */
485 static int check_reg_arg(struct bpf_reg_state
*regs
, u32 regno
,
488 if (regno
>= MAX_BPF_REG
) {
489 verbose("R%d is invalid\n", regno
);
494 /* check whether register used as source operand can be read */
495 if (regs
[regno
].type
== NOT_INIT
) {
496 verbose("R%d !read_ok\n", regno
);
500 /* check whether register used as dest operand can be written to */
501 if (regno
== BPF_REG_FP
) {
502 verbose("frame pointer is read only\n");
506 mark_reg_unknown_value(regs
, regno
);
511 static int bpf_size_to_bytes(int bpf_size
)
513 if (bpf_size
== BPF_W
)
515 else if (bpf_size
== BPF_H
)
517 else if (bpf_size
== BPF_B
)
519 else if (bpf_size
== BPF_DW
)
525 static bool is_spillable_regtype(enum bpf_reg_type type
)
528 case PTR_TO_MAP_VALUE
:
529 case PTR_TO_MAP_VALUE_OR_NULL
:
533 case PTR_TO_PACKET_END
:
535 case CONST_PTR_TO_MAP
:
542 /* check_stack_read/write functions track spill/fill of registers,
543 * stack boundary and alignment are checked in check_mem_access()
545 static int check_stack_write(struct bpf_verifier_state
*state
, int off
,
546 int size
, int value_regno
)
549 /* caller checked that off % size == 0 and -MAX_BPF_STACK <= off < 0,
550 * so it's aligned access and [off, off + size) are within stack limits
553 if (value_regno
>= 0 &&
554 is_spillable_regtype(state
->regs
[value_regno
].type
)) {
556 /* register containing pointer is being spilled into stack */
557 if (size
!= BPF_REG_SIZE
) {
558 verbose("invalid size of register spill\n");
562 /* save register state */
563 state
->spilled_regs
[(MAX_BPF_STACK
+ off
) / BPF_REG_SIZE
] =
564 state
->regs
[value_regno
];
566 for (i
= 0; i
< BPF_REG_SIZE
; i
++)
567 state
->stack_slot_type
[MAX_BPF_STACK
+ off
+ i
] = STACK_SPILL
;
569 /* regular write of data into stack */
570 state
->spilled_regs
[(MAX_BPF_STACK
+ off
) / BPF_REG_SIZE
] =
571 (struct bpf_reg_state
) {};
573 for (i
= 0; i
< size
; i
++)
574 state
->stack_slot_type
[MAX_BPF_STACK
+ off
+ i
] = STACK_MISC
;
579 static int check_stack_read(struct bpf_verifier_state
*state
, int off
, int size
,
585 slot_type
= &state
->stack_slot_type
[MAX_BPF_STACK
+ off
];
587 if (slot_type
[0] == STACK_SPILL
) {
588 if (size
!= BPF_REG_SIZE
) {
589 verbose("invalid size of register spill\n");
592 for (i
= 1; i
< BPF_REG_SIZE
; i
++) {
593 if (slot_type
[i
] != STACK_SPILL
) {
594 verbose("corrupted spill memory\n");
599 if (value_regno
>= 0)
600 /* restore register state from stack */
601 state
->regs
[value_regno
] =
602 state
->spilled_regs
[(MAX_BPF_STACK
+ off
) / BPF_REG_SIZE
];
605 for (i
= 0; i
< size
; i
++) {
606 if (slot_type
[i
] != STACK_MISC
) {
607 verbose("invalid read from stack off %d+%d size %d\n",
612 if (value_regno
>= 0)
613 /* have read misc data from the stack */
614 mark_reg_unknown_value(state
->regs
, value_regno
);
619 /* check read/write into map element returned by bpf_map_lookup_elem() */
620 static int check_map_access(struct bpf_verifier_env
*env
, u32 regno
, int off
,
623 struct bpf_map
*map
= env
->cur_state
.regs
[regno
].map_ptr
;
625 if (off
< 0 || off
+ size
> map
->value_size
) {
626 verbose("invalid access to map value, value_size=%d off=%d size=%d\n",
627 map
->value_size
, off
, size
);
633 #define MAX_PACKET_OFF 0xffff
635 static bool may_access_direct_pkt_data(struct bpf_verifier_env
*env
,
636 const struct bpf_call_arg_meta
*meta
)
638 switch (env
->prog
->type
) {
639 case BPF_PROG_TYPE_SCHED_CLS
:
640 case BPF_PROG_TYPE_SCHED_ACT
:
641 case BPF_PROG_TYPE_XDP
:
643 return meta
->pkt_access
;
645 env
->seen_direct_write
= true;
652 static int check_packet_access(struct bpf_verifier_env
*env
, u32 regno
, int off
,
655 struct bpf_reg_state
*regs
= env
->cur_state
.regs
;
656 struct bpf_reg_state
*reg
= ®s
[regno
];
659 if (off
< 0 || size
<= 0 || off
+ size
> reg
->range
) {
660 verbose("invalid access to packet, off=%d size=%d, R%d(id=%d,off=%d,r=%d)\n",
661 off
, size
, regno
, reg
->id
, reg
->off
, reg
->range
);
667 /* check access to 'struct bpf_context' fields */
668 static int check_ctx_access(struct bpf_verifier_env
*env
, int off
, int size
,
669 enum bpf_access_type t
, enum bpf_reg_type
*reg_type
)
671 /* for analyzer ctx accesses are already validated and converted */
672 if (env
->analyzer_ops
)
675 if (env
->prog
->aux
->ops
->is_valid_access
&&
676 env
->prog
->aux
->ops
->is_valid_access(off
, size
, t
, reg_type
)) {
677 /* remember the offset of last byte accessed in ctx */
678 if (env
->prog
->aux
->max_ctx_offset
< off
+ size
)
679 env
->prog
->aux
->max_ctx_offset
= off
+ size
;
683 verbose("invalid bpf_context access off=%d size=%d\n", off
, size
);
687 static bool is_pointer_value(struct bpf_verifier_env
*env
, int regno
)
689 if (env
->allow_ptr_leaks
)
692 switch (env
->cur_state
.regs
[regno
].type
) {
701 static int check_ptr_alignment(struct bpf_verifier_env
*env
,
702 struct bpf_reg_state
*reg
, int off
, int size
)
704 if (reg
->type
!= PTR_TO_PACKET
&& reg
->type
!= PTR_TO_MAP_VALUE_ADJ
) {
705 if (off
% size
!= 0) {
706 verbose("misaligned access off %d size %d\n",
714 if (IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS
))
715 /* misaligned access to packet is ok on x86,arm,arm64 */
718 if (reg
->id
&& size
!= 1) {
719 verbose("Unknown packet alignment. Only byte-sized access allowed\n");
723 /* skb->data is NET_IP_ALIGN-ed */
724 if (reg
->type
== PTR_TO_PACKET
&&
725 (NET_IP_ALIGN
+ reg
->off
+ off
) % size
!= 0) {
726 verbose("misaligned packet access off %d+%d+%d size %d\n",
727 NET_IP_ALIGN
, reg
->off
, off
, size
);
733 /* check whether memory at (regno + off) is accessible for t = (read | write)
734 * if t==write, value_regno is a register which value is stored into memory
735 * if t==read, value_regno is a register which will receive the value from memory
736 * if t==write && value_regno==-1, some unknown value is stored into memory
737 * if t==read && value_regno==-1, don't care what we read from memory
739 static int check_mem_access(struct bpf_verifier_env
*env
, u32 regno
, int off
,
740 int bpf_size
, enum bpf_access_type t
,
743 struct bpf_verifier_state
*state
= &env
->cur_state
;
744 struct bpf_reg_state
*reg
= &state
->regs
[regno
];
747 if (reg
->type
== PTR_TO_STACK
)
750 size
= bpf_size_to_bytes(bpf_size
);
754 err
= check_ptr_alignment(env
, reg
, off
, size
);
758 if (reg
->type
== PTR_TO_MAP_VALUE
||
759 reg
->type
== PTR_TO_MAP_VALUE_ADJ
) {
760 if (t
== BPF_WRITE
&& value_regno
>= 0 &&
761 is_pointer_value(env
, value_regno
)) {
762 verbose("R%d leaks addr into map\n", value_regno
);
766 /* If we adjusted the register to this map value at all then we
767 * need to change off and size to min_value and max_value
768 * respectively to make sure our theoretical access will be
771 if (reg
->type
== PTR_TO_MAP_VALUE_ADJ
) {
773 print_verifier_state(state
);
774 env
->varlen_map_value_access
= true;
775 /* The minimum value is only important with signed
776 * comparisons where we can't assume the floor of a
777 * value is 0. If we are using signed variables for our
778 * index'es we need to make sure that whatever we use
779 * will have a set floor within our range.
781 if ((s64
)reg
->min_value
< 0) {
782 verbose("R%d min value is negative, either use unsigned index or do a if (index >=0) check.\n",
786 err
= check_map_access(env
, regno
, reg
->min_value
+ off
,
789 verbose("R%d min value is outside of the array range\n",
794 /* If we haven't set a max value then we need to bail
795 * since we can't be sure we won't do bad things.
797 if (reg
->max_value
== BPF_REGISTER_MAX_RANGE
) {
798 verbose("R%d unbounded memory access, make sure to bounds check any array access into a map\n",
802 off
+= reg
->max_value
;
804 err
= check_map_access(env
, regno
, off
, size
);
805 if (!err
&& t
== BPF_READ
&& value_regno
>= 0)
806 mark_reg_unknown_value(state
->regs
, value_regno
);
808 } else if (reg
->type
== PTR_TO_CTX
) {
809 enum bpf_reg_type reg_type
= UNKNOWN_VALUE
;
811 if (t
== BPF_WRITE
&& value_regno
>= 0 &&
812 is_pointer_value(env
, value_regno
)) {
813 verbose("R%d leaks addr into ctx\n", value_regno
);
816 err
= check_ctx_access(env
, off
, size
, t
, ®_type
);
817 if (!err
&& t
== BPF_READ
&& value_regno
>= 0) {
818 mark_reg_unknown_value(state
->regs
, value_regno
);
819 /* note that reg.[id|off|range] == 0 */
820 state
->regs
[value_regno
].type
= reg_type
;
823 } else if (reg
->type
== FRAME_PTR
|| reg
->type
== PTR_TO_STACK
) {
824 if (off
>= 0 || off
< -MAX_BPF_STACK
) {
825 verbose("invalid stack off=%d size=%d\n", off
, size
);
828 if (t
== BPF_WRITE
) {
829 if (!env
->allow_ptr_leaks
&&
830 state
->stack_slot_type
[MAX_BPF_STACK
+ off
] == STACK_SPILL
&&
831 size
!= BPF_REG_SIZE
) {
832 verbose("attempt to corrupt spilled pointer on stack\n");
835 err
= check_stack_write(state
, off
, size
, value_regno
);
837 err
= check_stack_read(state
, off
, size
, value_regno
);
839 } else if (state
->regs
[regno
].type
== PTR_TO_PACKET
) {
840 if (t
== BPF_WRITE
&& !may_access_direct_pkt_data(env
, NULL
)) {
841 verbose("cannot write into packet\n");
844 if (t
== BPF_WRITE
&& value_regno
>= 0 &&
845 is_pointer_value(env
, value_regno
)) {
846 verbose("R%d leaks addr into packet\n", value_regno
);
849 err
= check_packet_access(env
, regno
, off
, size
);
850 if (!err
&& t
== BPF_READ
&& value_regno
>= 0)
851 mark_reg_unknown_value(state
->regs
, value_regno
);
853 verbose("R%d invalid mem access '%s'\n",
854 regno
, reg_type_str
[reg
->type
]);
858 if (!err
&& size
<= 2 && value_regno
>= 0 && env
->allow_ptr_leaks
&&
859 state
->regs
[value_regno
].type
== UNKNOWN_VALUE
) {
860 /* 1 or 2 byte load zero-extends, determine the number of
861 * zero upper bits. Not doing it fo 4 byte load, since
862 * such values cannot be added to ptr_to_packet anyway.
864 state
->regs
[value_regno
].imm
= 64 - size
* 8;
869 static int check_xadd(struct bpf_verifier_env
*env
, struct bpf_insn
*insn
)
871 struct bpf_reg_state
*regs
= env
->cur_state
.regs
;
874 if ((BPF_SIZE(insn
->code
) != BPF_W
&& BPF_SIZE(insn
->code
) != BPF_DW
) ||
876 verbose("BPF_XADD uses reserved fields\n");
880 /* check src1 operand */
881 err
= check_reg_arg(regs
, insn
->src_reg
, SRC_OP
);
885 /* check src2 operand */
886 err
= check_reg_arg(regs
, insn
->dst_reg
, SRC_OP
);
890 /* check whether atomic_add can read the memory */
891 err
= check_mem_access(env
, insn
->dst_reg
, insn
->off
,
892 BPF_SIZE(insn
->code
), BPF_READ
, -1);
896 /* check whether atomic_add can write into the same memory */
897 return check_mem_access(env
, insn
->dst_reg
, insn
->off
,
898 BPF_SIZE(insn
->code
), BPF_WRITE
, -1);
901 /* when register 'regno' is passed into function that will read 'access_size'
902 * bytes from that pointer, make sure that it's within stack boundary
903 * and all elements of stack are initialized
905 static int check_stack_boundary(struct bpf_verifier_env
*env
, int regno
,
906 int access_size
, bool zero_size_allowed
,
907 struct bpf_call_arg_meta
*meta
)
909 struct bpf_verifier_state
*state
= &env
->cur_state
;
910 struct bpf_reg_state
*regs
= state
->regs
;
913 if (regs
[regno
].type
!= PTR_TO_STACK
) {
914 if (zero_size_allowed
&& access_size
== 0 &&
915 regs
[regno
].type
== CONST_IMM
&&
916 regs
[regno
].imm
== 0)
919 verbose("R%d type=%s expected=%s\n", regno
,
920 reg_type_str
[regs
[regno
].type
],
921 reg_type_str
[PTR_TO_STACK
]);
925 off
= regs
[regno
].imm
;
926 if (off
>= 0 || off
< -MAX_BPF_STACK
|| off
+ access_size
> 0 ||
928 verbose("invalid stack type R%d off=%d access_size=%d\n",
929 regno
, off
, access_size
);
933 if (meta
&& meta
->raw_mode
) {
934 meta
->access_size
= access_size
;
939 for (i
= 0; i
< access_size
; i
++) {
940 if (state
->stack_slot_type
[MAX_BPF_STACK
+ off
+ i
] != STACK_MISC
) {
941 verbose("invalid indirect read from stack off %d+%d size %d\n",
942 off
, i
, access_size
);
949 static int check_func_arg(struct bpf_verifier_env
*env
, u32 regno
,
950 enum bpf_arg_type arg_type
,
951 struct bpf_call_arg_meta
*meta
)
953 struct bpf_reg_state
*regs
= env
->cur_state
.regs
, *reg
= ®s
[regno
];
954 enum bpf_reg_type expected_type
, type
= reg
->type
;
957 if (arg_type
== ARG_DONTCARE
)
960 if (type
== NOT_INIT
) {
961 verbose("R%d !read_ok\n", regno
);
965 if (arg_type
== ARG_ANYTHING
) {
966 if (is_pointer_value(env
, regno
)) {
967 verbose("R%d leaks addr into helper function\n", regno
);
973 if (type
== PTR_TO_PACKET
&& !may_access_direct_pkt_data(env
, meta
)) {
974 verbose("helper access to the packet is not allowed\n");
978 if (arg_type
== ARG_PTR_TO_MAP_KEY
||
979 arg_type
== ARG_PTR_TO_MAP_VALUE
) {
980 expected_type
= PTR_TO_STACK
;
981 if (type
!= PTR_TO_PACKET
&& type
!= expected_type
)
983 } else if (arg_type
== ARG_CONST_STACK_SIZE
||
984 arg_type
== ARG_CONST_STACK_SIZE_OR_ZERO
) {
985 expected_type
= CONST_IMM
;
986 if (type
!= expected_type
)
988 } else if (arg_type
== ARG_CONST_MAP_PTR
) {
989 expected_type
= CONST_PTR_TO_MAP
;
990 if (type
!= expected_type
)
992 } else if (arg_type
== ARG_PTR_TO_CTX
) {
993 expected_type
= PTR_TO_CTX
;
994 if (type
!= expected_type
)
996 } else if (arg_type
== ARG_PTR_TO_STACK
||
997 arg_type
== ARG_PTR_TO_RAW_STACK
) {
998 expected_type
= PTR_TO_STACK
;
999 /* One exception here. In case function allows for NULL to be
1000 * passed in as argument, it's a CONST_IMM type. Final test
1001 * happens during stack boundary checking.
1003 if (type
== CONST_IMM
&& reg
->imm
== 0)
1004 /* final test in check_stack_boundary() */;
1005 else if (type
!= PTR_TO_PACKET
&& type
!= expected_type
)
1007 meta
->raw_mode
= arg_type
== ARG_PTR_TO_RAW_STACK
;
1009 verbose("unsupported arg_type %d\n", arg_type
);
1013 if (arg_type
== ARG_CONST_MAP_PTR
) {
1014 /* bpf_map_xxx(map_ptr) call: remember that map_ptr */
1015 meta
->map_ptr
= reg
->map_ptr
;
1016 } else if (arg_type
== ARG_PTR_TO_MAP_KEY
) {
1017 /* bpf_map_xxx(..., map_ptr, ..., key) call:
1018 * check that [key, key + map->key_size) are within
1019 * stack limits and initialized
1021 if (!meta
->map_ptr
) {
1022 /* in function declaration map_ptr must come before
1023 * map_key, so that it's verified and known before
1024 * we have to check map_key here. Otherwise it means
1025 * that kernel subsystem misconfigured verifier
1027 verbose("invalid map_ptr to access map->key\n");
1030 if (type
== PTR_TO_PACKET
)
1031 err
= check_packet_access(env
, regno
, 0,
1032 meta
->map_ptr
->key_size
);
1034 err
= check_stack_boundary(env
, regno
,
1035 meta
->map_ptr
->key_size
,
1037 } else if (arg_type
== ARG_PTR_TO_MAP_VALUE
) {
1038 /* bpf_map_xxx(..., map_ptr, ..., value) call:
1039 * check [value, value + map->value_size) validity
1041 if (!meta
->map_ptr
) {
1042 /* kernel subsystem misconfigured verifier */
1043 verbose("invalid map_ptr to access map->value\n");
1046 if (type
== PTR_TO_PACKET
)
1047 err
= check_packet_access(env
, regno
, 0,
1048 meta
->map_ptr
->value_size
);
1050 err
= check_stack_boundary(env
, regno
,
1051 meta
->map_ptr
->value_size
,
1053 } else if (arg_type
== ARG_CONST_STACK_SIZE
||
1054 arg_type
== ARG_CONST_STACK_SIZE_OR_ZERO
) {
1055 bool zero_size_allowed
= (arg_type
== ARG_CONST_STACK_SIZE_OR_ZERO
);
1057 /* bpf_xxx(..., buf, len) call will access 'len' bytes
1058 * from stack pointer 'buf'. Check it
1059 * note: regno == len, regno - 1 == buf
1062 /* kernel subsystem misconfigured verifier */
1063 verbose("ARG_CONST_STACK_SIZE cannot be first argument\n");
1066 if (regs
[regno
- 1].type
== PTR_TO_PACKET
)
1067 err
= check_packet_access(env
, regno
- 1, 0, reg
->imm
);
1069 err
= check_stack_boundary(env
, regno
- 1, reg
->imm
,
1070 zero_size_allowed
, meta
);
1075 verbose("R%d type=%s expected=%s\n", regno
,
1076 reg_type_str
[type
], reg_type_str
[expected_type
]);
1080 static int check_map_func_compatibility(struct bpf_map
*map
, int func_id
)
1085 /* We need a two way check, first is from map perspective ... */
1086 switch (map
->map_type
) {
1087 case BPF_MAP_TYPE_PROG_ARRAY
:
1088 if (func_id
!= BPF_FUNC_tail_call
)
1091 case BPF_MAP_TYPE_PERF_EVENT_ARRAY
:
1092 if (func_id
!= BPF_FUNC_perf_event_read
&&
1093 func_id
!= BPF_FUNC_perf_event_output
)
1096 case BPF_MAP_TYPE_STACK_TRACE
:
1097 if (func_id
!= BPF_FUNC_get_stackid
)
1100 case BPF_MAP_TYPE_CGROUP_ARRAY
:
1101 if (func_id
!= BPF_FUNC_skb_under_cgroup
&&
1102 func_id
!= BPF_FUNC_current_task_under_cgroup
)
1109 /* ... and second from the function itself. */
1111 case BPF_FUNC_tail_call
:
1112 if (map
->map_type
!= BPF_MAP_TYPE_PROG_ARRAY
)
1115 case BPF_FUNC_perf_event_read
:
1116 case BPF_FUNC_perf_event_output
:
1117 if (map
->map_type
!= BPF_MAP_TYPE_PERF_EVENT_ARRAY
)
1120 case BPF_FUNC_get_stackid
:
1121 if (map
->map_type
!= BPF_MAP_TYPE_STACK_TRACE
)
1124 case BPF_FUNC_current_task_under_cgroup
:
1125 case BPF_FUNC_skb_under_cgroup
:
1126 if (map
->map_type
!= BPF_MAP_TYPE_CGROUP_ARRAY
)
1135 verbose("cannot pass map_type %d into func %s#%d\n",
1136 map
->map_type
, func_id_name(func_id
), func_id
);
1140 static int check_raw_mode(const struct bpf_func_proto
*fn
)
1144 if (fn
->arg1_type
== ARG_PTR_TO_RAW_STACK
)
1146 if (fn
->arg2_type
== ARG_PTR_TO_RAW_STACK
)
1148 if (fn
->arg3_type
== ARG_PTR_TO_RAW_STACK
)
1150 if (fn
->arg4_type
== ARG_PTR_TO_RAW_STACK
)
1152 if (fn
->arg5_type
== ARG_PTR_TO_RAW_STACK
)
1155 return count
> 1 ? -EINVAL
: 0;
1158 static void clear_all_pkt_pointers(struct bpf_verifier_env
*env
)
1160 struct bpf_verifier_state
*state
= &env
->cur_state
;
1161 struct bpf_reg_state
*regs
= state
->regs
, *reg
;
1164 for (i
= 0; i
< MAX_BPF_REG
; i
++)
1165 if (regs
[i
].type
== PTR_TO_PACKET
||
1166 regs
[i
].type
== PTR_TO_PACKET_END
)
1167 mark_reg_unknown_value(regs
, i
);
1169 for (i
= 0; i
< MAX_BPF_STACK
; i
+= BPF_REG_SIZE
) {
1170 if (state
->stack_slot_type
[i
] != STACK_SPILL
)
1172 reg
= &state
->spilled_regs
[i
/ BPF_REG_SIZE
];
1173 if (reg
->type
!= PTR_TO_PACKET
&&
1174 reg
->type
!= PTR_TO_PACKET_END
)
1176 reg
->type
= UNKNOWN_VALUE
;
1181 static int check_call(struct bpf_verifier_env
*env
, int func_id
)
1183 struct bpf_verifier_state
*state
= &env
->cur_state
;
1184 const struct bpf_func_proto
*fn
= NULL
;
1185 struct bpf_reg_state
*regs
= state
->regs
;
1186 struct bpf_reg_state
*reg
;
1187 struct bpf_call_arg_meta meta
;
1191 /* find function prototype */
1192 if (func_id
< 0 || func_id
>= __BPF_FUNC_MAX_ID
) {
1193 verbose("invalid func %s#%d\n", func_id_name(func_id
), func_id
);
1197 if (env
->prog
->aux
->ops
->get_func_proto
)
1198 fn
= env
->prog
->aux
->ops
->get_func_proto(func_id
);
1201 verbose("unknown func %s#%d\n", func_id_name(func_id
), func_id
);
1205 /* eBPF programs must be GPL compatible to use GPL-ed functions */
1206 if (!env
->prog
->gpl_compatible
&& fn
->gpl_only
) {
1207 verbose("cannot call GPL only function from proprietary program\n");
1211 changes_data
= bpf_helper_changes_skb_data(fn
->func
);
1213 memset(&meta
, 0, sizeof(meta
));
1214 meta
.pkt_access
= fn
->pkt_access
;
1216 /* We only support one arg being in raw mode at the moment, which
1217 * is sufficient for the helper functions we have right now.
1219 err
= check_raw_mode(fn
);
1221 verbose("kernel subsystem misconfigured func %s#%d\n",
1222 func_id_name(func_id
), func_id
);
1227 err
= check_func_arg(env
, BPF_REG_1
, fn
->arg1_type
, &meta
);
1230 err
= check_func_arg(env
, BPF_REG_2
, fn
->arg2_type
, &meta
);
1233 err
= check_func_arg(env
, BPF_REG_3
, fn
->arg3_type
, &meta
);
1236 err
= check_func_arg(env
, BPF_REG_4
, fn
->arg4_type
, &meta
);
1239 err
= check_func_arg(env
, BPF_REG_5
, fn
->arg5_type
, &meta
);
1243 /* Mark slots with STACK_MISC in case of raw mode, stack offset
1244 * is inferred from register state.
1246 for (i
= 0; i
< meta
.access_size
; i
++) {
1247 err
= check_mem_access(env
, meta
.regno
, i
, BPF_B
, BPF_WRITE
, -1);
1252 /* reset caller saved regs */
1253 for (i
= 0; i
< CALLER_SAVED_REGS
; i
++) {
1254 reg
= regs
+ caller_saved
[i
];
1255 reg
->type
= NOT_INIT
;
1259 /* update return register */
1260 if (fn
->ret_type
== RET_INTEGER
) {
1261 regs
[BPF_REG_0
].type
= UNKNOWN_VALUE
;
1262 } else if (fn
->ret_type
== RET_VOID
) {
1263 regs
[BPF_REG_0
].type
= NOT_INIT
;
1264 } else if (fn
->ret_type
== RET_PTR_TO_MAP_VALUE_OR_NULL
) {
1265 regs
[BPF_REG_0
].type
= PTR_TO_MAP_VALUE_OR_NULL
;
1266 regs
[BPF_REG_0
].max_value
= regs
[BPF_REG_0
].min_value
= 0;
1267 /* remember map_ptr, so that check_map_access()
1268 * can check 'value_size' boundary of memory access
1269 * to map element returned from bpf_map_lookup_elem()
1271 if (meta
.map_ptr
== NULL
) {
1272 verbose("kernel subsystem misconfigured verifier\n");
1275 regs
[BPF_REG_0
].map_ptr
= meta
.map_ptr
;
1276 regs
[BPF_REG_0
].id
= ++env
->id_gen
;
1278 verbose("unknown return type %d of func %s#%d\n",
1279 fn
->ret_type
, func_id_name(func_id
), func_id
);
1283 err
= check_map_func_compatibility(meta
.map_ptr
, func_id
);
1288 clear_all_pkt_pointers(env
);
1292 static int check_packet_ptr_add(struct bpf_verifier_env
*env
,
1293 struct bpf_insn
*insn
)
1295 struct bpf_reg_state
*regs
= env
->cur_state
.regs
;
1296 struct bpf_reg_state
*dst_reg
= ®s
[insn
->dst_reg
];
1297 struct bpf_reg_state
*src_reg
= ®s
[insn
->src_reg
];
1298 struct bpf_reg_state tmp_reg
;
1301 if (BPF_SRC(insn
->code
) == BPF_K
) {
1302 /* pkt_ptr += imm */
1307 verbose("addition of negative constant to packet pointer is not allowed\n");
1310 if (imm
>= MAX_PACKET_OFF
||
1311 imm
+ dst_reg
->off
>= MAX_PACKET_OFF
) {
1312 verbose("constant %d is too large to add to packet pointer\n",
1316 /* a constant was added to pkt_ptr.
1317 * Remember it while keeping the same 'id'
1319 dst_reg
->off
+= imm
;
1321 if (src_reg
->type
== PTR_TO_PACKET
) {
1322 /* R6=pkt(id=0,off=0,r=62) R7=imm22; r7 += r6 */
1323 tmp_reg
= *dst_reg
; /* save r7 state */
1324 *dst_reg
= *src_reg
; /* copy pkt_ptr state r6 into r7 */
1325 src_reg
= &tmp_reg
; /* pretend it's src_reg state */
1326 /* if the checks below reject it, the copy won't matter,
1327 * since we're rejecting the whole program. If all ok,
1328 * then imm22 state will be added to r7
1329 * and r7 will be pkt(id=0,off=22,r=62) while
1330 * r6 will stay as pkt(id=0,off=0,r=62)
1334 if (src_reg
->type
== CONST_IMM
) {
1335 /* pkt_ptr += reg where reg is known constant */
1339 /* disallow pkt_ptr += reg
1340 * if reg is not uknown_value with guaranteed zero upper bits
1341 * otherwise pkt_ptr may overflow and addition will become
1342 * subtraction which is not allowed
1344 if (src_reg
->type
!= UNKNOWN_VALUE
) {
1345 verbose("cannot add '%s' to ptr_to_packet\n",
1346 reg_type_str
[src_reg
->type
]);
1349 if (src_reg
->imm
< 48) {
1350 verbose("cannot add integer value with %lld upper zero bits to ptr_to_packet\n",
1354 /* dst_reg stays as pkt_ptr type and since some positive
1355 * integer value was added to the pointer, increment its 'id'
1357 dst_reg
->id
= ++env
->id_gen
;
1359 /* something was added to pkt_ptr, set range and off to zero */
1366 static int evaluate_reg_alu(struct bpf_verifier_env
*env
, struct bpf_insn
*insn
)
1368 struct bpf_reg_state
*regs
= env
->cur_state
.regs
;
1369 struct bpf_reg_state
*dst_reg
= ®s
[insn
->dst_reg
];
1370 u8 opcode
= BPF_OP(insn
->code
);
1373 /* for type == UNKNOWN_VALUE:
1374 * imm > 0 -> number of zero upper bits
1375 * imm == 0 -> don't track which is the same as all bits can be non-zero
1378 if (BPF_SRC(insn
->code
) == BPF_X
) {
1379 struct bpf_reg_state
*src_reg
= ®s
[insn
->src_reg
];
1381 if (src_reg
->type
== UNKNOWN_VALUE
&& src_reg
->imm
> 0 &&
1382 dst_reg
->imm
&& opcode
== BPF_ADD
) {
1384 * where both have zero upper bits. Adding them
1385 * can only result making one more bit non-zero
1386 * in the larger value.
1387 * Ex. 0xffff (imm=48) + 1 (imm=63) = 0x10000 (imm=47)
1388 * 0xffff (imm=48) + 0xffff = 0x1fffe (imm=47)
1390 dst_reg
->imm
= min(dst_reg
->imm
, src_reg
->imm
);
1394 if (src_reg
->type
== CONST_IMM
&& src_reg
->imm
> 0 &&
1395 dst_reg
->imm
&& opcode
== BPF_ADD
) {
1397 * where dreg has zero upper bits and sreg is const.
1398 * Adding them can only result making one more bit
1399 * non-zero in the larger value.
1401 imm_log2
= __ilog2_u64((long long)src_reg
->imm
);
1402 dst_reg
->imm
= min(dst_reg
->imm
, 63 - imm_log2
);
1406 /* all other cases non supported yet, just mark dst_reg */
1411 /* sign extend 32-bit imm into 64-bit to make sure that
1412 * negative values occupy bit 63. Note ilog2() would have
1413 * been incorrect, since sizeof(insn->imm) == 4
1415 imm_log2
= __ilog2_u64((long long)insn
->imm
);
1417 if (dst_reg
->imm
&& opcode
== BPF_LSH
) {
1419 * if reg was a result of 2 byte load, then its imm == 48
1420 * which means that upper 48 bits are zero and shifting this reg
1421 * left by 4 would mean that upper 44 bits are still zero
1423 dst_reg
->imm
-= insn
->imm
;
1424 } else if (dst_reg
->imm
&& opcode
== BPF_MUL
) {
1426 * if multiplying by 14 subtract 4
1427 * This is conservative calculation of upper zero bits.
1428 * It's not trying to special case insn->imm == 1 or 0 cases
1430 dst_reg
->imm
-= imm_log2
+ 1;
1431 } else if (opcode
== BPF_AND
) {
1433 dst_reg
->imm
= 63 - imm_log2
;
1434 } else if (dst_reg
->imm
&& opcode
== BPF_ADD
) {
1436 dst_reg
->imm
= min(dst_reg
->imm
, 63 - imm_log2
);
1438 } else if (opcode
== BPF_RSH
) {
1440 * which means that after right shift, upper bits will be zero
1441 * note that verifier already checked that
1442 * 0 <= imm < 64 for shift insn
1444 dst_reg
->imm
+= insn
->imm
;
1445 if (unlikely(dst_reg
->imm
> 64))
1446 /* some dumb code did:
1449 * and all bits are zero now */
1452 /* all other alu ops, means that we don't know what will
1453 * happen to the value, mark it with unknown number of zero bits
1458 if (dst_reg
->imm
< 0) {
1459 /* all 64 bits of the register can contain non-zero bits
1460 * and such value cannot be added to ptr_to_packet, since it
1461 * may overflow, mark it as unknown to avoid further eval
1468 static int evaluate_reg_imm_alu(struct bpf_verifier_env
*env
,
1469 struct bpf_insn
*insn
)
1471 struct bpf_reg_state
*regs
= env
->cur_state
.regs
;
1472 struct bpf_reg_state
*dst_reg
= ®s
[insn
->dst_reg
];
1473 struct bpf_reg_state
*src_reg
= ®s
[insn
->src_reg
];
1474 u8 opcode
= BPF_OP(insn
->code
);
1476 /* dst_reg->type == CONST_IMM here, simulate execution of 'add' insn.
1477 * Don't care about overflow or negative values, just add them
1479 if (opcode
== BPF_ADD
&& BPF_SRC(insn
->code
) == BPF_K
)
1480 dst_reg
->imm
+= insn
->imm
;
1481 else if (opcode
== BPF_ADD
&& BPF_SRC(insn
->code
) == BPF_X
&&
1482 src_reg
->type
== CONST_IMM
)
1483 dst_reg
->imm
+= src_reg
->imm
;
1485 mark_reg_unknown_value(regs
, insn
->dst_reg
);
1489 static void check_reg_overflow(struct bpf_reg_state
*reg
)
1491 if (reg
->max_value
> BPF_REGISTER_MAX_RANGE
)
1492 reg
->max_value
= BPF_REGISTER_MAX_RANGE
;
1493 if ((s64
)reg
->min_value
< BPF_REGISTER_MIN_RANGE
)
1494 reg
->min_value
= BPF_REGISTER_MIN_RANGE
;
1497 static void adjust_reg_min_max_vals(struct bpf_verifier_env
*env
,
1498 struct bpf_insn
*insn
)
1500 struct bpf_reg_state
*regs
= env
->cur_state
.regs
, *dst_reg
;
1501 u64 min_val
= BPF_REGISTER_MIN_RANGE
, max_val
= BPF_REGISTER_MAX_RANGE
;
1502 u8 opcode
= BPF_OP(insn
->code
);
1504 dst_reg
= ®s
[insn
->dst_reg
];
1505 if (BPF_SRC(insn
->code
) == BPF_X
) {
1506 check_reg_overflow(®s
[insn
->src_reg
]);
1507 min_val
= regs
[insn
->src_reg
].min_value
;
1508 max_val
= regs
[insn
->src_reg
].max_value
;
1510 /* If the source register is a random pointer then the
1511 * min_value/max_value values represent the range of the known
1512 * accesses into that value, not the actual min/max value of the
1513 * register itself. In this case we have to reset the reg range
1514 * values so we know it is not safe to look at.
1516 if (regs
[insn
->src_reg
].type
!= CONST_IMM
&&
1517 regs
[insn
->src_reg
].type
!= UNKNOWN_VALUE
) {
1518 min_val
= BPF_REGISTER_MIN_RANGE
;
1519 max_val
= BPF_REGISTER_MAX_RANGE
;
1521 } else if (insn
->imm
< BPF_REGISTER_MAX_RANGE
&&
1522 (s64
)insn
->imm
> BPF_REGISTER_MIN_RANGE
) {
1523 min_val
= max_val
= insn
->imm
;
1526 /* We don't know anything about what was done to this register, mark it
1529 if (min_val
== BPF_REGISTER_MIN_RANGE
&&
1530 max_val
== BPF_REGISTER_MAX_RANGE
) {
1531 reset_reg_range_values(regs
, insn
->dst_reg
);
1537 dst_reg
->min_value
+= min_val
;
1538 dst_reg
->max_value
+= max_val
;
1541 dst_reg
->min_value
-= min_val
;
1542 dst_reg
->max_value
-= max_val
;
1545 dst_reg
->min_value
*= min_val
;
1546 dst_reg
->max_value
*= max_val
;
1549 /* & is special since it could end up with 0 bits set. */
1550 dst_reg
->min_value
&= min_val
;
1551 dst_reg
->max_value
= max_val
;
1554 /* Gotta have special overflow logic here, if we're shifting
1555 * more than MAX_RANGE then just assume we have an invalid
1558 if (min_val
> ilog2(BPF_REGISTER_MAX_RANGE
))
1559 dst_reg
->min_value
= BPF_REGISTER_MIN_RANGE
;
1561 dst_reg
->min_value
<<= min_val
;
1563 if (max_val
> ilog2(BPF_REGISTER_MAX_RANGE
))
1564 dst_reg
->max_value
= BPF_REGISTER_MAX_RANGE
;
1566 dst_reg
->max_value
<<= max_val
;
1569 dst_reg
->min_value
>>= min_val
;
1570 dst_reg
->max_value
>>= max_val
;
1573 /* % is special since it is an unsigned modulus, so the floor
1576 dst_reg
->min_value
= 0;
1577 dst_reg
->max_value
= max_val
- 1;
1580 reset_reg_range_values(regs
, insn
->dst_reg
);
1584 check_reg_overflow(dst_reg
);
1587 /* check validity of 32-bit and 64-bit arithmetic operations */
1588 static int check_alu_op(struct bpf_verifier_env
*env
, struct bpf_insn
*insn
)
1590 struct bpf_reg_state
*regs
= env
->cur_state
.regs
, *dst_reg
;
1591 u8 opcode
= BPF_OP(insn
->code
);
1594 if (opcode
== BPF_END
|| opcode
== BPF_NEG
) {
1595 if (opcode
== BPF_NEG
) {
1596 if (BPF_SRC(insn
->code
) != 0 ||
1597 insn
->src_reg
!= BPF_REG_0
||
1598 insn
->off
!= 0 || insn
->imm
!= 0) {
1599 verbose("BPF_NEG uses reserved fields\n");
1603 if (insn
->src_reg
!= BPF_REG_0
|| insn
->off
!= 0 ||
1604 (insn
->imm
!= 16 && insn
->imm
!= 32 && insn
->imm
!= 64)) {
1605 verbose("BPF_END uses reserved fields\n");
1610 /* check src operand */
1611 err
= check_reg_arg(regs
, insn
->dst_reg
, SRC_OP
);
1615 if (is_pointer_value(env
, insn
->dst_reg
)) {
1616 verbose("R%d pointer arithmetic prohibited\n",
1621 /* check dest operand */
1622 err
= check_reg_arg(regs
, insn
->dst_reg
, DST_OP
);
1626 } else if (opcode
== BPF_MOV
) {
1628 if (BPF_SRC(insn
->code
) == BPF_X
) {
1629 if (insn
->imm
!= 0 || insn
->off
!= 0) {
1630 verbose("BPF_MOV uses reserved fields\n");
1634 /* check src operand */
1635 err
= check_reg_arg(regs
, insn
->src_reg
, SRC_OP
);
1639 if (insn
->src_reg
!= BPF_REG_0
|| insn
->off
!= 0) {
1640 verbose("BPF_MOV uses reserved fields\n");
1645 /* check dest operand */
1646 err
= check_reg_arg(regs
, insn
->dst_reg
, DST_OP
);
1650 /* we are setting our register to something new, we need to
1651 * reset its range values.
1653 reset_reg_range_values(regs
, insn
->dst_reg
);
1655 if (BPF_SRC(insn
->code
) == BPF_X
) {
1656 if (BPF_CLASS(insn
->code
) == BPF_ALU64
) {
1658 * copy register state to dest reg
1660 regs
[insn
->dst_reg
] = regs
[insn
->src_reg
];
1662 if (is_pointer_value(env
, insn
->src_reg
)) {
1663 verbose("R%d partial copy of pointer\n",
1667 mark_reg_unknown_value(regs
, insn
->dst_reg
);
1671 * remember the value we stored into this reg
1673 regs
[insn
->dst_reg
].type
= CONST_IMM
;
1674 regs
[insn
->dst_reg
].imm
= insn
->imm
;
1675 regs
[insn
->dst_reg
].max_value
= insn
->imm
;
1676 regs
[insn
->dst_reg
].min_value
= insn
->imm
;
1679 } else if (opcode
> BPF_END
) {
1680 verbose("invalid BPF_ALU opcode %x\n", opcode
);
1683 } else { /* all other ALU ops: and, sub, xor, add, ... */
1685 if (BPF_SRC(insn
->code
) == BPF_X
) {
1686 if (insn
->imm
!= 0 || insn
->off
!= 0) {
1687 verbose("BPF_ALU uses reserved fields\n");
1690 /* check src1 operand */
1691 err
= check_reg_arg(regs
, insn
->src_reg
, SRC_OP
);
1695 if (insn
->src_reg
!= BPF_REG_0
|| insn
->off
!= 0) {
1696 verbose("BPF_ALU uses reserved fields\n");
1701 /* check src2 operand */
1702 err
= check_reg_arg(regs
, insn
->dst_reg
, SRC_OP
);
1706 if ((opcode
== BPF_MOD
|| opcode
== BPF_DIV
) &&
1707 BPF_SRC(insn
->code
) == BPF_K
&& insn
->imm
== 0) {
1708 verbose("div by zero\n");
1712 if ((opcode
== BPF_LSH
|| opcode
== BPF_RSH
||
1713 opcode
== BPF_ARSH
) && BPF_SRC(insn
->code
) == BPF_K
) {
1714 int size
= BPF_CLASS(insn
->code
) == BPF_ALU64
? 64 : 32;
1716 if (insn
->imm
< 0 || insn
->imm
>= size
) {
1717 verbose("invalid shift %d\n", insn
->imm
);
1722 /* check dest operand */
1723 err
= check_reg_arg(regs
, insn
->dst_reg
, DST_OP_NO_MARK
);
1727 dst_reg
= ®s
[insn
->dst_reg
];
1729 /* first we want to adjust our ranges. */
1730 adjust_reg_min_max_vals(env
, insn
);
1732 /* pattern match 'bpf_add Rx, imm' instruction */
1733 if (opcode
== BPF_ADD
&& BPF_CLASS(insn
->code
) == BPF_ALU64
&&
1734 dst_reg
->type
== FRAME_PTR
&& BPF_SRC(insn
->code
) == BPF_K
) {
1735 dst_reg
->type
= PTR_TO_STACK
;
1736 dst_reg
->imm
= insn
->imm
;
1738 } else if (opcode
== BPF_ADD
&&
1739 BPF_CLASS(insn
->code
) == BPF_ALU64
&&
1740 (dst_reg
->type
== PTR_TO_PACKET
||
1741 (BPF_SRC(insn
->code
) == BPF_X
&&
1742 regs
[insn
->src_reg
].type
== PTR_TO_PACKET
))) {
1743 /* ptr_to_packet += K|X */
1744 return check_packet_ptr_add(env
, insn
);
1745 } else if (BPF_CLASS(insn
->code
) == BPF_ALU64
&&
1746 dst_reg
->type
== UNKNOWN_VALUE
&&
1747 env
->allow_ptr_leaks
) {
1748 /* unknown += K|X */
1749 return evaluate_reg_alu(env
, insn
);
1750 } else if (BPF_CLASS(insn
->code
) == BPF_ALU64
&&
1751 dst_reg
->type
== CONST_IMM
&&
1752 env
->allow_ptr_leaks
) {
1753 /* reg_imm += K|X */
1754 return evaluate_reg_imm_alu(env
, insn
);
1755 } else if (is_pointer_value(env
, insn
->dst_reg
)) {
1756 verbose("R%d pointer arithmetic prohibited\n",
1759 } else if (BPF_SRC(insn
->code
) == BPF_X
&&
1760 is_pointer_value(env
, insn
->src_reg
)) {
1761 verbose("R%d pointer arithmetic prohibited\n",
1766 /* If we did pointer math on a map value then just set it to our
1767 * PTR_TO_MAP_VALUE_ADJ type so we can deal with any stores or
1768 * loads to this register appropriately, otherwise just mark the
1769 * register as unknown.
1771 if (env
->allow_ptr_leaks
&&
1772 (dst_reg
->type
== PTR_TO_MAP_VALUE
||
1773 dst_reg
->type
== PTR_TO_MAP_VALUE_ADJ
))
1774 dst_reg
->type
= PTR_TO_MAP_VALUE_ADJ
;
1776 mark_reg_unknown_value(regs
, insn
->dst_reg
);
1782 static void find_good_pkt_pointers(struct bpf_verifier_state
*state
,
1783 struct bpf_reg_state
*dst_reg
)
1785 struct bpf_reg_state
*regs
= state
->regs
, *reg
;
1788 /* LLVM can generate two kind of checks:
1794 * if (r2 > pkt_end) goto <handle exception>
1798 * r2 == dst_reg, pkt_end == src_reg
1799 * r2=pkt(id=n,off=8,r=0)
1800 * r3=pkt(id=n,off=0,r=0)
1806 * if (pkt_end >= r2) goto <access okay>
1807 * <handle exception>
1810 * pkt_end == dst_reg, r2 == src_reg
1811 * r2=pkt(id=n,off=8,r=0)
1812 * r3=pkt(id=n,off=0,r=0)
1814 * Find register r3 and mark its range as r3=pkt(id=n,off=0,r=8)
1815 * so that range of bytes [r3, r3 + 8) is safe to access.
1818 for (i
= 0; i
< MAX_BPF_REG
; i
++)
1819 if (regs
[i
].type
== PTR_TO_PACKET
&& regs
[i
].id
== dst_reg
->id
)
1820 regs
[i
].range
= dst_reg
->off
;
1822 for (i
= 0; i
< MAX_BPF_STACK
; i
+= BPF_REG_SIZE
) {
1823 if (state
->stack_slot_type
[i
] != STACK_SPILL
)
1825 reg
= &state
->spilled_regs
[i
/ BPF_REG_SIZE
];
1826 if (reg
->type
== PTR_TO_PACKET
&& reg
->id
== dst_reg
->id
)
1827 reg
->range
= dst_reg
->off
;
1831 /* Adjusts the register min/max values in the case that the dst_reg is the
1832 * variable register that we are working on, and src_reg is a constant or we're
1833 * simply doing a BPF_K check.
1835 static void reg_set_min_max(struct bpf_reg_state
*true_reg
,
1836 struct bpf_reg_state
*false_reg
, u64 val
,
1841 /* If this is false then we know nothing Jon Snow, but if it is
1842 * true then we know for sure.
1844 true_reg
->max_value
= true_reg
->min_value
= val
;
1847 /* If this is true we know nothing Jon Snow, but if it is false
1848 * we know the value for sure;
1850 false_reg
->max_value
= false_reg
->min_value
= val
;
1853 /* Unsigned comparison, the minimum value is 0. */
1854 false_reg
->min_value
= 0;
1856 /* If this is false then we know the maximum val is val,
1857 * otherwise we know the min val is val+1.
1859 false_reg
->max_value
= val
;
1860 true_reg
->min_value
= val
+ 1;
1863 /* Unsigned comparison, the minimum value is 0. */
1864 false_reg
->min_value
= 0;
1866 /* If this is false then we know the maximum value is val - 1,
1867 * otherwise we know the mimimum value is val.
1869 false_reg
->max_value
= val
- 1;
1870 true_reg
->min_value
= val
;
1876 check_reg_overflow(false_reg
);
1877 check_reg_overflow(true_reg
);
1880 /* Same as above, but for the case that dst_reg is a CONST_IMM reg and src_reg
1881 * is the variable reg.
1883 static void reg_set_min_max_inv(struct bpf_reg_state
*true_reg
,
1884 struct bpf_reg_state
*false_reg
, u64 val
,
1889 /* If this is false then we know nothing Jon Snow, but if it is
1890 * true then we know for sure.
1892 true_reg
->max_value
= true_reg
->min_value
= val
;
1895 /* If this is true we know nothing Jon Snow, but if it is false
1896 * we know the value for sure;
1898 false_reg
->max_value
= false_reg
->min_value
= val
;
1901 /* Unsigned comparison, the minimum value is 0. */
1902 true_reg
->min_value
= 0;
1905 * If this is false, then the val is <= the register, if it is
1906 * true the register <= to the val.
1908 false_reg
->min_value
= val
;
1909 true_reg
->max_value
= val
- 1;
1912 /* Unsigned comparison, the minimum value is 0. */
1913 true_reg
->min_value
= 0;
1915 /* If this is false then constant < register, if it is true then
1916 * the register < constant.
1918 false_reg
->min_value
= val
+ 1;
1919 true_reg
->max_value
= val
;
1925 check_reg_overflow(false_reg
);
1926 check_reg_overflow(true_reg
);
1929 static void mark_map_reg(struct bpf_reg_state
*regs
, u32 regno
, u32 id
,
1930 enum bpf_reg_type type
)
1932 struct bpf_reg_state
*reg
= ®s
[regno
];
1934 if (reg
->type
== PTR_TO_MAP_VALUE_OR_NULL
&& reg
->id
== id
) {
1936 if (type
== UNKNOWN_VALUE
)
1937 mark_reg_unknown_value(regs
, regno
);
1941 /* The logic is similar to find_good_pkt_pointers(), both could eventually
1942 * be folded together at some point.
1944 static void mark_map_regs(struct bpf_verifier_state
*state
, u32 regno
,
1945 enum bpf_reg_type type
)
1947 struct bpf_reg_state
*regs
= state
->regs
;
1950 for (i
= 0; i
< MAX_BPF_REG
; i
++)
1951 mark_map_reg(regs
, i
, regs
[regno
].id
, type
);
1953 for (i
= 0; i
< MAX_BPF_STACK
; i
+= BPF_REG_SIZE
) {
1954 if (state
->stack_slot_type
[i
] != STACK_SPILL
)
1956 mark_map_reg(state
->spilled_regs
, i
/ BPF_REG_SIZE
,
1957 regs
[regno
].id
, type
);
1961 static int check_cond_jmp_op(struct bpf_verifier_env
*env
,
1962 struct bpf_insn
*insn
, int *insn_idx
)
1964 struct bpf_verifier_state
*other_branch
, *this_branch
= &env
->cur_state
;
1965 struct bpf_reg_state
*regs
= this_branch
->regs
, *dst_reg
;
1966 u8 opcode
= BPF_OP(insn
->code
);
1969 if (opcode
> BPF_EXIT
) {
1970 verbose("invalid BPF_JMP opcode %x\n", opcode
);
1974 if (BPF_SRC(insn
->code
) == BPF_X
) {
1975 if (insn
->imm
!= 0) {
1976 verbose("BPF_JMP uses reserved fields\n");
1980 /* check src1 operand */
1981 err
= check_reg_arg(regs
, insn
->src_reg
, SRC_OP
);
1985 if (is_pointer_value(env
, insn
->src_reg
)) {
1986 verbose("R%d pointer comparison prohibited\n",
1991 if (insn
->src_reg
!= BPF_REG_0
) {
1992 verbose("BPF_JMP uses reserved fields\n");
1997 /* check src2 operand */
1998 err
= check_reg_arg(regs
, insn
->dst_reg
, SRC_OP
);
2002 dst_reg
= ®s
[insn
->dst_reg
];
2004 /* detect if R == 0 where R was initialized to zero earlier */
2005 if (BPF_SRC(insn
->code
) == BPF_K
&&
2006 (opcode
== BPF_JEQ
|| opcode
== BPF_JNE
) &&
2007 dst_reg
->type
== CONST_IMM
&& dst_reg
->imm
== insn
->imm
) {
2008 if (opcode
== BPF_JEQ
) {
2009 /* if (imm == imm) goto pc+off;
2010 * only follow the goto, ignore fall-through
2012 *insn_idx
+= insn
->off
;
2015 /* if (imm != imm) goto pc+off;
2016 * only follow fall-through branch, since
2017 * that's where the program will go
2023 other_branch
= push_stack(env
, *insn_idx
+ insn
->off
+ 1, *insn_idx
);
2027 /* detect if we are comparing against a constant value so we can adjust
2028 * our min/max values for our dst register.
2030 if (BPF_SRC(insn
->code
) == BPF_X
) {
2031 if (regs
[insn
->src_reg
].type
== CONST_IMM
)
2032 reg_set_min_max(&other_branch
->regs
[insn
->dst_reg
],
2033 dst_reg
, regs
[insn
->src_reg
].imm
,
2035 else if (dst_reg
->type
== CONST_IMM
)
2036 reg_set_min_max_inv(&other_branch
->regs
[insn
->src_reg
],
2037 ®s
[insn
->src_reg
], dst_reg
->imm
,
2040 reg_set_min_max(&other_branch
->regs
[insn
->dst_reg
],
2041 dst_reg
, insn
->imm
, opcode
);
2044 /* detect if R == 0 where R is returned from bpf_map_lookup_elem() */
2045 if (BPF_SRC(insn
->code
) == BPF_K
&&
2046 insn
->imm
== 0 && (opcode
== BPF_JEQ
|| opcode
== BPF_JNE
) &&
2047 dst_reg
->type
== PTR_TO_MAP_VALUE_OR_NULL
) {
2048 /* Mark all identical map registers in each branch as either
2049 * safe or unknown depending R == 0 or R != 0 conditional.
2051 mark_map_regs(this_branch
, insn
->dst_reg
,
2052 opcode
== BPF_JEQ
? PTR_TO_MAP_VALUE
: UNKNOWN_VALUE
);
2053 mark_map_regs(other_branch
, insn
->dst_reg
,
2054 opcode
== BPF_JEQ
? UNKNOWN_VALUE
: PTR_TO_MAP_VALUE
);
2055 } else if (BPF_SRC(insn
->code
) == BPF_X
&& opcode
== BPF_JGT
&&
2056 dst_reg
->type
== PTR_TO_PACKET
&&
2057 regs
[insn
->src_reg
].type
== PTR_TO_PACKET_END
) {
2058 find_good_pkt_pointers(this_branch
, dst_reg
);
2059 } else if (BPF_SRC(insn
->code
) == BPF_X
&& opcode
== BPF_JGE
&&
2060 dst_reg
->type
== PTR_TO_PACKET_END
&&
2061 regs
[insn
->src_reg
].type
== PTR_TO_PACKET
) {
2062 find_good_pkt_pointers(other_branch
, ®s
[insn
->src_reg
]);
2063 } else if (is_pointer_value(env
, insn
->dst_reg
)) {
2064 verbose("R%d pointer comparison prohibited\n", insn
->dst_reg
);
2068 print_verifier_state(this_branch
);
2072 /* return the map pointer stored inside BPF_LD_IMM64 instruction */
2073 static struct bpf_map
*ld_imm64_to_map_ptr(struct bpf_insn
*insn
)
2075 u64 imm64
= ((u64
) (u32
) insn
[0].imm
) | ((u64
) (u32
) insn
[1].imm
) << 32;
2077 return (struct bpf_map
*) (unsigned long) imm64
;
2080 /* verify BPF_LD_IMM64 instruction */
2081 static int check_ld_imm(struct bpf_verifier_env
*env
, struct bpf_insn
*insn
)
2083 struct bpf_reg_state
*regs
= env
->cur_state
.regs
;
2086 if (BPF_SIZE(insn
->code
) != BPF_DW
) {
2087 verbose("invalid BPF_LD_IMM insn\n");
2090 if (insn
->off
!= 0) {
2091 verbose("BPF_LD_IMM64 uses reserved fields\n");
2095 err
= check_reg_arg(regs
, insn
->dst_reg
, DST_OP
);
2099 if (insn
->src_reg
== 0) {
2100 /* generic move 64-bit immediate into a register,
2101 * only analyzer needs to collect the ld_imm value.
2103 u64 imm
= ((u64
)(insn
+ 1)->imm
<< 32) | (u32
)insn
->imm
;
2105 if (!env
->analyzer_ops
)
2108 regs
[insn
->dst_reg
].type
= CONST_IMM
;
2109 regs
[insn
->dst_reg
].imm
= imm
;
2113 /* replace_map_fd_with_map_ptr() should have caught bad ld_imm64 */
2114 BUG_ON(insn
->src_reg
!= BPF_PSEUDO_MAP_FD
);
2116 regs
[insn
->dst_reg
].type
= CONST_PTR_TO_MAP
;
2117 regs
[insn
->dst_reg
].map_ptr
= ld_imm64_to_map_ptr(insn
);
2121 static bool may_access_skb(enum bpf_prog_type type
)
2124 case BPF_PROG_TYPE_SOCKET_FILTER
:
2125 case BPF_PROG_TYPE_SCHED_CLS
:
2126 case BPF_PROG_TYPE_SCHED_ACT
:
2133 /* verify safety of LD_ABS|LD_IND instructions:
2134 * - they can only appear in the programs where ctx == skb
2135 * - since they are wrappers of function calls, they scratch R1-R5 registers,
2136 * preserve R6-R9, and store return value into R0
2139 * ctx == skb == R6 == CTX
2142 * SRC == any register
2143 * IMM == 32-bit immediate
2146 * R0 - 8/16/32-bit skb data converted to cpu endianness
2148 static int check_ld_abs(struct bpf_verifier_env
*env
, struct bpf_insn
*insn
)
2150 struct bpf_reg_state
*regs
= env
->cur_state
.regs
;
2151 u8 mode
= BPF_MODE(insn
->code
);
2152 struct bpf_reg_state
*reg
;
2155 if (!may_access_skb(env
->prog
->type
)) {
2156 verbose("BPF_LD_[ABS|IND] instructions not allowed for this program type\n");
2160 if (insn
->dst_reg
!= BPF_REG_0
|| insn
->off
!= 0 ||
2161 BPF_SIZE(insn
->code
) == BPF_DW
||
2162 (mode
== BPF_ABS
&& insn
->src_reg
!= BPF_REG_0
)) {
2163 verbose("BPF_LD_[ABS|IND] uses reserved fields\n");
2167 /* check whether implicit source operand (register R6) is readable */
2168 err
= check_reg_arg(regs
, BPF_REG_6
, SRC_OP
);
2172 if (regs
[BPF_REG_6
].type
!= PTR_TO_CTX
) {
2173 verbose("at the time of BPF_LD_ABS|IND R6 != pointer to skb\n");
2177 if (mode
== BPF_IND
) {
2178 /* check explicit source operand */
2179 err
= check_reg_arg(regs
, insn
->src_reg
, SRC_OP
);
2184 /* reset caller saved regs to unreadable */
2185 for (i
= 0; i
< CALLER_SAVED_REGS
; i
++) {
2186 reg
= regs
+ caller_saved
[i
];
2187 reg
->type
= NOT_INIT
;
2191 /* mark destination R0 register as readable, since it contains
2192 * the value fetched from the packet
2194 regs
[BPF_REG_0
].type
= UNKNOWN_VALUE
;
2198 /* non-recursive DFS pseudo code
2199 * 1 procedure DFS-iterative(G,v):
2200 * 2 label v as discovered
2201 * 3 let S be a stack
2203 * 5 while S is not empty
2205 * 7 if t is what we're looking for:
2207 * 9 for all edges e in G.adjacentEdges(t) do
2208 * 10 if edge e is already labelled
2209 * 11 continue with the next edge
2210 * 12 w <- G.adjacentVertex(t,e)
2211 * 13 if vertex w is not discovered and not explored
2212 * 14 label e as tree-edge
2213 * 15 label w as discovered
2216 * 18 else if vertex w is discovered
2217 * 19 label e as back-edge
2219 * 21 // vertex w is explored
2220 * 22 label e as forward- or cross-edge
2221 * 23 label t as explored
2226 * 0x11 - discovered and fall-through edge labelled
2227 * 0x12 - discovered and fall-through and branch edges labelled
2238 #define STATE_LIST_MARK ((struct bpf_verifier_state_list *) -1L)
2240 static int *insn_stack
; /* stack of insns to process */
2241 static int cur_stack
; /* current stack index */
2242 static int *insn_state
;
2244 /* t, w, e - match pseudo-code above:
2245 * t - index of current instruction
2246 * w - next instruction
2249 static int push_insn(int t
, int w
, int e
, struct bpf_verifier_env
*env
)
2251 if (e
== FALLTHROUGH
&& insn_state
[t
] >= (DISCOVERED
| FALLTHROUGH
))
2254 if (e
== BRANCH
&& insn_state
[t
] >= (DISCOVERED
| BRANCH
))
2257 if (w
< 0 || w
>= env
->prog
->len
) {
2258 verbose("jump out of range from insn %d to %d\n", t
, w
);
2263 /* mark branch target for state pruning */
2264 env
->explored_states
[w
] = STATE_LIST_MARK
;
2266 if (insn_state
[w
] == 0) {
2268 insn_state
[t
] = DISCOVERED
| e
;
2269 insn_state
[w
] = DISCOVERED
;
2270 if (cur_stack
>= env
->prog
->len
)
2272 insn_stack
[cur_stack
++] = w
;
2274 } else if ((insn_state
[w
] & 0xF0) == DISCOVERED
) {
2275 verbose("back-edge from insn %d to %d\n", t
, w
);
2277 } else if (insn_state
[w
] == EXPLORED
) {
2278 /* forward- or cross-edge */
2279 insn_state
[t
] = DISCOVERED
| e
;
2281 verbose("insn state internal bug\n");
2287 /* non-recursive depth-first-search to detect loops in BPF program
2288 * loop == back-edge in directed graph
2290 static int check_cfg(struct bpf_verifier_env
*env
)
2292 struct bpf_insn
*insns
= env
->prog
->insnsi
;
2293 int insn_cnt
= env
->prog
->len
;
2297 insn_state
= kcalloc(insn_cnt
, sizeof(int), GFP_KERNEL
);
2301 insn_stack
= kcalloc(insn_cnt
, sizeof(int), GFP_KERNEL
);
2307 insn_state
[0] = DISCOVERED
; /* mark 1st insn as discovered */
2308 insn_stack
[0] = 0; /* 0 is the first instruction */
2314 t
= insn_stack
[cur_stack
- 1];
2316 if (BPF_CLASS(insns
[t
].code
) == BPF_JMP
) {
2317 u8 opcode
= BPF_OP(insns
[t
].code
);
2319 if (opcode
== BPF_EXIT
) {
2321 } else if (opcode
== BPF_CALL
) {
2322 ret
= push_insn(t
, t
+ 1, FALLTHROUGH
, env
);
2327 if (t
+ 1 < insn_cnt
)
2328 env
->explored_states
[t
+ 1] = STATE_LIST_MARK
;
2329 } else if (opcode
== BPF_JA
) {
2330 if (BPF_SRC(insns
[t
].code
) != BPF_K
) {
2334 /* unconditional jump with single edge */
2335 ret
= push_insn(t
, t
+ insns
[t
].off
+ 1,
2341 /* tell verifier to check for equivalent states
2342 * after every call and jump
2344 if (t
+ 1 < insn_cnt
)
2345 env
->explored_states
[t
+ 1] = STATE_LIST_MARK
;
2347 /* conditional jump with two edges */
2348 ret
= push_insn(t
, t
+ 1, FALLTHROUGH
, env
);
2354 ret
= push_insn(t
, t
+ insns
[t
].off
+ 1, BRANCH
, env
);
2361 /* all other non-branch instructions with single
2364 ret
= push_insn(t
, t
+ 1, FALLTHROUGH
, env
);
2372 insn_state
[t
] = EXPLORED
;
2373 if (cur_stack
-- <= 0) {
2374 verbose("pop stack internal bug\n");
2381 for (i
= 0; i
< insn_cnt
; i
++) {
2382 if (insn_state
[i
] != EXPLORED
) {
2383 verbose("unreachable insn %d\n", i
);
2388 ret
= 0; /* cfg looks good */
2396 /* the following conditions reduce the number of explored insns
2397 * from ~140k to ~80k for ultra large programs that use a lot of ptr_to_packet
2399 static bool compare_ptrs_to_packet(struct bpf_reg_state
*old
,
2400 struct bpf_reg_state
*cur
)
2402 if (old
->id
!= cur
->id
)
2405 /* old ptr_to_packet is more conservative, since it allows smaller
2407 * old(off=0,r=10) is equal to cur(off=0,r=20), because
2408 * old(off=0,r=10) means that with range=10 the verifier proceeded
2409 * further and found no issues with the program. Now we're in the same
2410 * spot with cur(off=0,r=20), so we're safe too, since anything further
2411 * will only be looking at most 10 bytes after this pointer.
2413 if (old
->off
== cur
->off
&& old
->range
< cur
->range
)
2416 /* old(off=20,r=10) is equal to cur(off=22,re=22 or 5 or 0)
2417 * since both cannot be used for packet access and safe(old)
2418 * pointer has smaller off that could be used for further
2419 * 'if (ptr > data_end)' check
2421 * old(off=20,r=10) and cur(off=22,r=22) and cur(off=22,r=0) mean
2422 * that we cannot access the packet.
2423 * The safe range is:
2424 * [ptr, ptr + range - off)
2425 * so whenever off >=range, it means no safe bytes from this pointer.
2426 * When comparing old->off <= cur->off, it means that older code
2427 * went with smaller offset and that offset was later
2428 * used to figure out the safe range after 'if (ptr > data_end)' check
2429 * Say, 'old' state was explored like:
2430 * ... R3(off=0, r=0)
2432 * ... now R4(off=20,r=0) <-- here
2433 * if (R4 > data_end)
2434 * ... R4(off=20,r=20), R3(off=0,r=20) and R3 can be used to access.
2435 * ... the code further went all the way to bpf_exit.
2436 * Now the 'cur' state at the mark 'here' has R4(off=30,r=0).
2437 * old_R4(off=20,r=0) equal to cur_R4(off=30,r=0), since if the verifier
2438 * goes further, such cur_R4 will give larger safe packet range after
2439 * 'if (R4 > data_end)' and all further insn were already good with r=20,
2440 * so they will be good with r=30 and we can prune the search.
2442 if (old
->off
<= cur
->off
&&
2443 old
->off
>= old
->range
&& cur
->off
>= cur
->range
)
2449 /* compare two verifier states
2451 * all states stored in state_list are known to be valid, since
2452 * verifier reached 'bpf_exit' instruction through them
2454 * this function is called when verifier exploring different branches of
2455 * execution popped from the state stack. If it sees an old state that has
2456 * more strict register state and more strict stack state then this execution
2457 * branch doesn't need to be explored further, since verifier already
2458 * concluded that more strict state leads to valid finish.
2460 * Therefore two states are equivalent if register state is more conservative
2461 * and explored stack state is more conservative than the current one.
2464 * (slot1=INV slot2=MISC) == (slot1=MISC slot2=MISC)
2465 * (slot1=MISC slot2=MISC) != (slot1=INV slot2=MISC)
2467 * In other words if current stack state (one being explored) has more
2468 * valid slots than old one that already passed validation, it means
2469 * the verifier can stop exploring and conclude that current state is valid too
2471 * Similarly with registers. If explored state has register type as invalid
2472 * whereas register type in current state is meaningful, it means that
2473 * the current state will reach 'bpf_exit' instruction safely
2475 static bool states_equal(struct bpf_verifier_env
*env
,
2476 struct bpf_verifier_state
*old
,
2477 struct bpf_verifier_state
*cur
)
2479 struct bpf_reg_state
*rold
, *rcur
;
2482 for (i
= 0; i
< MAX_BPF_REG
; i
++) {
2483 rold
= &old
->regs
[i
];
2484 rcur
= &cur
->regs
[i
];
2486 if (memcmp(rold
, rcur
, sizeof(*rold
)) == 0)
2489 /* If the ranges were not the same, but everything else was and
2490 * we didn't do a variable access into a map then we are a-ok.
2492 if (!env
->varlen_map_value_access
&&
2493 rold
->type
== rcur
->type
&& rold
->imm
== rcur
->imm
)
2496 if (rold
->type
== NOT_INIT
||
2497 (rold
->type
== UNKNOWN_VALUE
&& rcur
->type
!= NOT_INIT
))
2500 if (rold
->type
== PTR_TO_PACKET
&& rcur
->type
== PTR_TO_PACKET
&&
2501 compare_ptrs_to_packet(rold
, rcur
))
2507 for (i
= 0; i
< MAX_BPF_STACK
; i
++) {
2508 if (old
->stack_slot_type
[i
] == STACK_INVALID
)
2510 if (old
->stack_slot_type
[i
] != cur
->stack_slot_type
[i
])
2511 /* Ex: old explored (safe) state has STACK_SPILL in
2512 * this stack slot, but current has has STACK_MISC ->
2513 * this verifier states are not equivalent,
2514 * return false to continue verification of this path
2517 if (i
% BPF_REG_SIZE
)
2519 if (memcmp(&old
->spilled_regs
[i
/ BPF_REG_SIZE
],
2520 &cur
->spilled_regs
[i
/ BPF_REG_SIZE
],
2521 sizeof(old
->spilled_regs
[0])))
2522 /* when explored and current stack slot types are
2523 * the same, check that stored pointers types
2524 * are the same as well.
2525 * Ex: explored safe path could have stored
2526 * (bpf_reg_state) {.type = PTR_TO_STACK, .imm = -8}
2527 * but current path has stored:
2528 * (bpf_reg_state) {.type = PTR_TO_STACK, .imm = -16}
2529 * such verifier states are not equivalent.
2530 * return false to continue verification of this path
2539 static int is_state_visited(struct bpf_verifier_env
*env
, int insn_idx
)
2541 struct bpf_verifier_state_list
*new_sl
;
2542 struct bpf_verifier_state_list
*sl
;
2544 sl
= env
->explored_states
[insn_idx
];
2546 /* this 'insn_idx' instruction wasn't marked, so we will not
2547 * be doing state search here
2551 while (sl
!= STATE_LIST_MARK
) {
2552 if (states_equal(env
, &sl
->state
, &env
->cur_state
))
2553 /* reached equivalent register/stack state,
2560 /* there were no equivalent states, remember current one.
2561 * technically the current state is not proven to be safe yet,
2562 * but it will either reach bpf_exit (which means it's safe) or
2563 * it will be rejected. Since there are no loops, we won't be
2564 * seeing this 'insn_idx' instruction again on the way to bpf_exit
2566 new_sl
= kmalloc(sizeof(struct bpf_verifier_state_list
), GFP_USER
);
2570 /* add new state to the head of linked list */
2571 memcpy(&new_sl
->state
, &env
->cur_state
, sizeof(env
->cur_state
));
2572 new_sl
->next
= env
->explored_states
[insn_idx
];
2573 env
->explored_states
[insn_idx
] = new_sl
;
2577 static int ext_analyzer_insn_hook(struct bpf_verifier_env
*env
,
2578 int insn_idx
, int prev_insn_idx
)
2580 if (!env
->analyzer_ops
|| !env
->analyzer_ops
->insn_hook
)
2583 return env
->analyzer_ops
->insn_hook(env
, insn_idx
, prev_insn_idx
);
2586 static int do_check(struct bpf_verifier_env
*env
)
2588 struct bpf_verifier_state
*state
= &env
->cur_state
;
2589 struct bpf_insn
*insns
= env
->prog
->insnsi
;
2590 struct bpf_reg_state
*regs
= state
->regs
;
2591 int insn_cnt
= env
->prog
->len
;
2592 int insn_idx
, prev_insn_idx
= 0;
2593 int insn_processed
= 0;
2594 bool do_print_state
= false;
2596 init_reg_state(regs
);
2598 env
->varlen_map_value_access
= false;
2600 struct bpf_insn
*insn
;
2604 if (insn_idx
>= insn_cnt
) {
2605 verbose("invalid insn idx %d insn_cnt %d\n",
2606 insn_idx
, insn_cnt
);
2610 insn
= &insns
[insn_idx
];
2611 class = BPF_CLASS(insn
->code
);
2613 if (++insn_processed
> BPF_COMPLEXITY_LIMIT_INSNS
) {
2614 verbose("BPF program is too large. Proccessed %d insn\n",
2619 err
= is_state_visited(env
, insn_idx
);
2623 /* found equivalent state, can prune the search */
2626 verbose("\nfrom %d to %d: safe\n",
2627 prev_insn_idx
, insn_idx
);
2629 verbose("%d: safe\n", insn_idx
);
2631 goto process_bpf_exit
;
2634 if (log_level
&& do_print_state
) {
2635 verbose("\nfrom %d to %d:", prev_insn_idx
, insn_idx
);
2636 print_verifier_state(&env
->cur_state
);
2637 do_print_state
= false;
2641 verbose("%d: ", insn_idx
);
2642 print_bpf_insn(insn
);
2645 err
= ext_analyzer_insn_hook(env
, insn_idx
, prev_insn_idx
);
2649 if (class == BPF_ALU
|| class == BPF_ALU64
) {
2650 err
= check_alu_op(env
, insn
);
2654 } else if (class == BPF_LDX
) {
2655 enum bpf_reg_type
*prev_src_type
, src_reg_type
;
2657 /* check for reserved fields is already done */
2659 /* check src operand */
2660 err
= check_reg_arg(regs
, insn
->src_reg
, SRC_OP
);
2664 err
= check_reg_arg(regs
, insn
->dst_reg
, DST_OP_NO_MARK
);
2668 src_reg_type
= regs
[insn
->src_reg
].type
;
2670 /* check that memory (src_reg + off) is readable,
2671 * the state of dst_reg will be updated by this func
2673 err
= check_mem_access(env
, insn
->src_reg
, insn
->off
,
2674 BPF_SIZE(insn
->code
), BPF_READ
,
2679 reset_reg_range_values(regs
, insn
->dst_reg
);
2680 if (BPF_SIZE(insn
->code
) != BPF_W
&&
2681 BPF_SIZE(insn
->code
) != BPF_DW
) {
2686 prev_src_type
= &env
->insn_aux_data
[insn_idx
].ptr_type
;
2688 if (*prev_src_type
== NOT_INIT
) {
2690 * dst_reg = *(u32 *)(src_reg + off)
2691 * save type to validate intersecting paths
2693 *prev_src_type
= src_reg_type
;
2695 } else if (src_reg_type
!= *prev_src_type
&&
2696 (src_reg_type
== PTR_TO_CTX
||
2697 *prev_src_type
== PTR_TO_CTX
)) {
2698 /* ABuser program is trying to use the same insn
2699 * dst_reg = *(u32*) (src_reg + off)
2700 * with different pointer types:
2701 * src_reg == ctx in one branch and
2702 * src_reg == stack|map in some other branch.
2705 verbose("same insn cannot be used with different pointers\n");
2709 } else if (class == BPF_STX
) {
2710 enum bpf_reg_type
*prev_dst_type
, dst_reg_type
;
2712 if (BPF_MODE(insn
->code
) == BPF_XADD
) {
2713 err
= check_xadd(env
, insn
);
2720 /* check src1 operand */
2721 err
= check_reg_arg(regs
, insn
->src_reg
, SRC_OP
);
2724 /* check src2 operand */
2725 err
= check_reg_arg(regs
, insn
->dst_reg
, SRC_OP
);
2729 dst_reg_type
= regs
[insn
->dst_reg
].type
;
2731 /* check that memory (dst_reg + off) is writeable */
2732 err
= check_mem_access(env
, insn
->dst_reg
, insn
->off
,
2733 BPF_SIZE(insn
->code
), BPF_WRITE
,
2738 prev_dst_type
= &env
->insn_aux_data
[insn_idx
].ptr_type
;
2740 if (*prev_dst_type
== NOT_INIT
) {
2741 *prev_dst_type
= dst_reg_type
;
2742 } else if (dst_reg_type
!= *prev_dst_type
&&
2743 (dst_reg_type
== PTR_TO_CTX
||
2744 *prev_dst_type
== PTR_TO_CTX
)) {
2745 verbose("same insn cannot be used with different pointers\n");
2749 } else if (class == BPF_ST
) {
2750 if (BPF_MODE(insn
->code
) != BPF_MEM
||
2751 insn
->src_reg
!= BPF_REG_0
) {
2752 verbose("BPF_ST uses reserved fields\n");
2755 /* check src operand */
2756 err
= check_reg_arg(regs
, insn
->dst_reg
, SRC_OP
);
2760 /* check that memory (dst_reg + off) is writeable */
2761 err
= check_mem_access(env
, insn
->dst_reg
, insn
->off
,
2762 BPF_SIZE(insn
->code
), BPF_WRITE
,
2767 } else if (class == BPF_JMP
) {
2768 u8 opcode
= BPF_OP(insn
->code
);
2770 if (opcode
== BPF_CALL
) {
2771 if (BPF_SRC(insn
->code
) != BPF_K
||
2773 insn
->src_reg
!= BPF_REG_0
||
2774 insn
->dst_reg
!= BPF_REG_0
) {
2775 verbose("BPF_CALL uses reserved fields\n");
2779 err
= check_call(env
, insn
->imm
);
2783 } else if (opcode
== BPF_JA
) {
2784 if (BPF_SRC(insn
->code
) != BPF_K
||
2786 insn
->src_reg
!= BPF_REG_0
||
2787 insn
->dst_reg
!= BPF_REG_0
) {
2788 verbose("BPF_JA uses reserved fields\n");
2792 insn_idx
+= insn
->off
+ 1;
2795 } else if (opcode
== BPF_EXIT
) {
2796 if (BPF_SRC(insn
->code
) != BPF_K
||
2798 insn
->src_reg
!= BPF_REG_0
||
2799 insn
->dst_reg
!= BPF_REG_0
) {
2800 verbose("BPF_EXIT uses reserved fields\n");
2804 /* eBPF calling convetion is such that R0 is used
2805 * to return the value from eBPF program.
2806 * Make sure that it's readable at this time
2807 * of bpf_exit, which means that program wrote
2808 * something into it earlier
2810 err
= check_reg_arg(regs
, BPF_REG_0
, SRC_OP
);
2814 if (is_pointer_value(env
, BPF_REG_0
)) {
2815 verbose("R0 leaks addr as return value\n");
2820 insn_idx
= pop_stack(env
, &prev_insn_idx
);
2824 do_print_state
= true;
2828 err
= check_cond_jmp_op(env
, insn
, &insn_idx
);
2832 } else if (class == BPF_LD
) {
2833 u8 mode
= BPF_MODE(insn
->code
);
2835 if (mode
== BPF_ABS
|| mode
== BPF_IND
) {
2836 err
= check_ld_abs(env
, insn
);
2840 } else if (mode
== BPF_IMM
) {
2841 err
= check_ld_imm(env
, insn
);
2847 verbose("invalid BPF_LD mode\n");
2850 reset_reg_range_values(regs
, insn
->dst_reg
);
2852 verbose("unknown insn class %d\n", class);
2859 verbose("processed %d insns\n", insn_processed
);
2863 static int check_map_prog_compatibility(struct bpf_map
*map
,
2864 struct bpf_prog
*prog
)
2867 if (prog
->type
== BPF_PROG_TYPE_PERF_EVENT
&&
2868 (map
->map_type
== BPF_MAP_TYPE_HASH
||
2869 map
->map_type
== BPF_MAP_TYPE_PERCPU_HASH
) &&
2870 (map
->map_flags
& BPF_F_NO_PREALLOC
)) {
2871 verbose("perf_event programs can only use preallocated hash map\n");
2877 /* look for pseudo eBPF instructions that access map FDs and
2878 * replace them with actual map pointers
2880 static int replace_map_fd_with_map_ptr(struct bpf_verifier_env
*env
)
2882 struct bpf_insn
*insn
= env
->prog
->insnsi
;
2883 int insn_cnt
= env
->prog
->len
;
2886 for (i
= 0; i
< insn_cnt
; i
++, insn
++) {
2887 if (BPF_CLASS(insn
->code
) == BPF_LDX
&&
2888 (BPF_MODE(insn
->code
) != BPF_MEM
|| insn
->imm
!= 0)) {
2889 verbose("BPF_LDX uses reserved fields\n");
2893 if (BPF_CLASS(insn
->code
) == BPF_STX
&&
2894 ((BPF_MODE(insn
->code
) != BPF_MEM
&&
2895 BPF_MODE(insn
->code
) != BPF_XADD
) || insn
->imm
!= 0)) {
2896 verbose("BPF_STX uses reserved fields\n");
2900 if (insn
[0].code
== (BPF_LD
| BPF_IMM
| BPF_DW
)) {
2901 struct bpf_map
*map
;
2904 if (i
== insn_cnt
- 1 || insn
[1].code
!= 0 ||
2905 insn
[1].dst_reg
!= 0 || insn
[1].src_reg
!= 0 ||
2907 verbose("invalid bpf_ld_imm64 insn\n");
2911 if (insn
->src_reg
== 0)
2912 /* valid generic load 64-bit imm */
2915 if (insn
->src_reg
!= BPF_PSEUDO_MAP_FD
) {
2916 verbose("unrecognized bpf_ld_imm64 insn\n");
2920 f
= fdget(insn
->imm
);
2921 map
= __bpf_map_get(f
);
2923 verbose("fd %d is not pointing to valid bpf_map\n",
2925 return PTR_ERR(map
);
2928 err
= check_map_prog_compatibility(map
, env
->prog
);
2934 /* store map pointer inside BPF_LD_IMM64 instruction */
2935 insn
[0].imm
= (u32
) (unsigned long) map
;
2936 insn
[1].imm
= ((u64
) (unsigned long) map
) >> 32;
2938 /* check whether we recorded this map already */
2939 for (j
= 0; j
< env
->used_map_cnt
; j
++)
2940 if (env
->used_maps
[j
] == map
) {
2945 if (env
->used_map_cnt
>= MAX_USED_MAPS
) {
2950 /* hold the map. If the program is rejected by verifier,
2951 * the map will be released by release_maps() or it
2952 * will be used by the valid program until it's unloaded
2953 * and all maps are released in free_bpf_prog_info()
2955 map
= bpf_map_inc(map
, false);
2958 return PTR_ERR(map
);
2960 env
->used_maps
[env
->used_map_cnt
++] = map
;
2969 /* now all pseudo BPF_LD_IMM64 instructions load valid
2970 * 'struct bpf_map *' into a register instead of user map_fd.
2971 * These pointers will be used later by verifier to validate map access.
2976 /* drop refcnt of maps used by the rejected program */
2977 static void release_maps(struct bpf_verifier_env
*env
)
2981 for (i
= 0; i
< env
->used_map_cnt
; i
++)
2982 bpf_map_put(env
->used_maps
[i
]);
2985 /* convert pseudo BPF_LD_IMM64 into generic BPF_LD_IMM64 */
2986 static void convert_pseudo_ld_imm64(struct bpf_verifier_env
*env
)
2988 struct bpf_insn
*insn
= env
->prog
->insnsi
;
2989 int insn_cnt
= env
->prog
->len
;
2992 for (i
= 0; i
< insn_cnt
; i
++, insn
++)
2993 if (insn
->code
== (BPF_LD
| BPF_IMM
| BPF_DW
))
2997 /* convert load instructions that access fields of 'struct __sk_buff'
2998 * into sequence of instructions that access fields of 'struct sk_buff'
3000 static int convert_ctx_accesses(struct bpf_verifier_env
*env
)
3002 const struct bpf_verifier_ops
*ops
= env
->prog
->aux
->ops
;
3003 const int insn_cnt
= env
->prog
->len
;
3004 struct bpf_insn insn_buf
[16], *insn
;
3005 struct bpf_prog
*new_prog
;
3006 enum bpf_access_type type
;
3007 int i
, cnt
, delta
= 0;
3009 if (ops
->gen_prologue
) {
3010 cnt
= ops
->gen_prologue(insn_buf
, env
->seen_direct_write
,
3012 if (cnt
>= ARRAY_SIZE(insn_buf
)) {
3013 verbose("bpf verifier is misconfigured\n");
3016 new_prog
= bpf_patch_insn_single(env
->prog
, 0,
3020 env
->prog
= new_prog
;
3025 if (!ops
->convert_ctx_access
)
3028 insn
= env
->prog
->insnsi
+ delta
;
3030 for (i
= 0; i
< insn_cnt
; i
++, insn
++) {
3031 if (insn
->code
== (BPF_LDX
| BPF_MEM
| BPF_W
) ||
3032 insn
->code
== (BPF_LDX
| BPF_MEM
| BPF_DW
))
3034 else if (insn
->code
== (BPF_STX
| BPF_MEM
| BPF_W
) ||
3035 insn
->code
== (BPF_STX
| BPF_MEM
| BPF_DW
))
3040 if (env
->insn_aux_data
[i
].ptr_type
!= PTR_TO_CTX
)
3043 cnt
= ops
->convert_ctx_access(type
, insn
->dst_reg
, insn
->src_reg
,
3044 insn
->off
, insn_buf
, env
->prog
);
3045 if (cnt
== 0 || cnt
>= ARRAY_SIZE(insn_buf
)) {
3046 verbose("bpf verifier is misconfigured\n");
3050 new_prog
= bpf_patch_insn_single(env
->prog
, i
+ delta
, insn_buf
,
3057 /* keep walking new program and skip insns we just inserted */
3058 env
->prog
= new_prog
;
3059 insn
= new_prog
->insnsi
+ i
+ delta
;
3065 static void free_states(struct bpf_verifier_env
*env
)
3067 struct bpf_verifier_state_list
*sl
, *sln
;
3070 if (!env
->explored_states
)
3073 for (i
= 0; i
< env
->prog
->len
; i
++) {
3074 sl
= env
->explored_states
[i
];
3077 while (sl
!= STATE_LIST_MARK
) {
3084 kfree(env
->explored_states
);
3087 int bpf_check(struct bpf_prog
**prog
, union bpf_attr
*attr
)
3089 char __user
*log_ubuf
= NULL
;
3090 struct bpf_verifier_env
*env
;
3093 if ((*prog
)->len
<= 0 || (*prog
)->len
> BPF_MAXINSNS
)
3096 /* 'struct bpf_verifier_env' can be global, but since it's not small,
3097 * allocate/free it every time bpf_check() is called
3099 env
= kzalloc(sizeof(struct bpf_verifier_env
), GFP_KERNEL
);
3103 env
->insn_aux_data
= vzalloc(sizeof(struct bpf_insn_aux_data
) *
3106 if (!env
->insn_aux_data
)
3110 /* grab the mutex to protect few globals used by verifier */
3111 mutex_lock(&bpf_verifier_lock
);
3113 if (attr
->log_level
|| attr
->log_buf
|| attr
->log_size
) {
3114 /* user requested verbose verifier output
3115 * and supplied buffer to store the verification trace
3117 log_level
= attr
->log_level
;
3118 log_ubuf
= (char __user
*) (unsigned long) attr
->log_buf
;
3119 log_size
= attr
->log_size
;
3123 /* log_* values have to be sane */
3124 if (log_size
< 128 || log_size
> UINT_MAX
>> 8 ||
3125 log_level
== 0 || log_ubuf
== NULL
)
3129 log_buf
= vmalloc(log_size
);
3136 ret
= replace_map_fd_with_map_ptr(env
);
3138 goto skip_full_check
;
3140 env
->explored_states
= kcalloc(env
->prog
->len
,
3141 sizeof(struct bpf_verifier_state_list
*),
3144 if (!env
->explored_states
)
3145 goto skip_full_check
;
3147 ret
= check_cfg(env
);
3149 goto skip_full_check
;
3151 env
->allow_ptr_leaks
= capable(CAP_SYS_ADMIN
);
3153 ret
= do_check(env
);
3156 while (pop_stack(env
, NULL
) >= 0);
3160 /* program is valid, convert *(u32*)(ctx + off) accesses */
3161 ret
= convert_ctx_accesses(env
);
3163 if (log_level
&& log_len
>= log_size
- 1) {
3164 BUG_ON(log_len
>= log_size
);
3165 /* verifier log exceeded user supplied buffer */
3167 /* fall through to return what was recorded */
3170 /* copy verifier log back to user space including trailing zero */
3171 if (log_level
&& copy_to_user(log_ubuf
, log_buf
, log_len
+ 1) != 0) {
3176 if (ret
== 0 && env
->used_map_cnt
) {
3177 /* if program passed verifier, update used_maps in bpf_prog_info */
3178 env
->prog
->aux
->used_maps
= kmalloc_array(env
->used_map_cnt
,
3179 sizeof(env
->used_maps
[0]),
3182 if (!env
->prog
->aux
->used_maps
) {
3187 memcpy(env
->prog
->aux
->used_maps
, env
->used_maps
,
3188 sizeof(env
->used_maps
[0]) * env
->used_map_cnt
);
3189 env
->prog
->aux
->used_map_cnt
= env
->used_map_cnt
;
3191 /* program is valid. Convert pseudo bpf_ld_imm64 into generic
3192 * bpf_ld_imm64 instructions
3194 convert_pseudo_ld_imm64(env
);
3200 if (!env
->prog
->aux
->used_maps
)
3201 /* if we didn't copy map pointers into bpf_prog_info, release
3202 * them now. Otherwise free_bpf_prog_info() will release them.
3207 mutex_unlock(&bpf_verifier_lock
);
3208 vfree(env
->insn_aux_data
);
3214 int bpf_analyzer(struct bpf_prog
*prog
, const struct bpf_ext_analyzer_ops
*ops
,
3217 struct bpf_verifier_env
*env
;
3220 env
= kzalloc(sizeof(struct bpf_verifier_env
), GFP_KERNEL
);
3224 env
->insn_aux_data
= vzalloc(sizeof(struct bpf_insn_aux_data
) *
3227 if (!env
->insn_aux_data
)
3230 env
->analyzer_ops
= ops
;
3231 env
->analyzer_priv
= priv
;
3233 /* grab the mutex to protect few globals used by verifier */
3234 mutex_lock(&bpf_verifier_lock
);
3238 env
->explored_states
= kcalloc(env
->prog
->len
,
3239 sizeof(struct bpf_verifier_state_list
*),
3242 if (!env
->explored_states
)
3243 goto skip_full_check
;
3245 ret
= check_cfg(env
);
3247 goto skip_full_check
;
3249 env
->allow_ptr_leaks
= capable(CAP_SYS_ADMIN
);
3251 ret
= do_check(env
);
3254 while (pop_stack(env
, NULL
) >= 0);
3257 mutex_unlock(&bpf_verifier_lock
);
3258 vfree(env
->insn_aux_data
);
3263 EXPORT_SYMBOL_GPL(bpf_analyzer
);