1 /* Copyright (c) 2011-2014 PLUMgrid, http://plumgrid.com
2 * Copyright (c) 2016 Facebook
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of version 2 of the GNU General Public
6 * License as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful, but
9 * WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 #include <linux/kernel.h>
14 #include <linux/types.h>
15 #include <linux/slab.h>
16 #include <linux/bpf.h>
17 #include <linux/bpf_verifier.h>
18 #include <linux/filter.h>
19 #include <net/netlink.h>
20 #include <linux/file.h>
21 #include <linux/vmalloc.h>
22 #include <linux/stringify.h>
24 /* bpf_check() is a static code analyzer that walks eBPF program
25 * instruction by instruction and updates register/stack state.
26 * All paths of conditional branches are analyzed until 'bpf_exit' insn.
28 * The first pass is depth-first-search to check that the program is a DAG.
29 * It rejects the following programs:
30 * - larger than BPF_MAXINSNS insns
31 * - if loop is present (detected via back-edge)
32 * - unreachable insns exist (shouldn't be a forest. program = one function)
33 * - out of bounds or malformed jumps
34 * The second pass is all possible path descent from the 1st insn.
35 * Since it's analyzing all pathes through the program, the length of the
36 * analysis is limited to 64k insn, which may be hit even if total number of
37 * insn is less then 4K, but there are too many branches that change stack/regs.
38 * Number of 'branches to be analyzed' is limited to 1k
40 * On entry to each instruction, each register has a type, and the instruction
41 * changes the types of the registers depending on instruction semantics.
42 * If instruction is BPF_MOV64_REG(BPF_REG_1, BPF_REG_5), then type of R5 is
45 * All registers are 64-bit.
46 * R0 - return register
47 * R1-R5 argument passing registers
48 * R6-R9 callee saved registers
49 * R10 - frame pointer read-only
51 * At the start of BPF program the register R1 contains a pointer to bpf_context
52 * and has type PTR_TO_CTX.
54 * Verifier tracks arithmetic operations on pointers in case:
55 * BPF_MOV64_REG(BPF_REG_1, BPF_REG_10),
56 * BPF_ALU64_IMM(BPF_ADD, BPF_REG_1, -20),
57 * 1st insn copies R10 (which has FRAME_PTR) type into R1
58 * and 2nd arithmetic instruction is pattern matched to recognize
59 * that it wants to construct a pointer to some element within stack.
60 * So after 2nd insn, the register R1 has type PTR_TO_STACK
61 * (and -20 constant is saved for further stack bounds checking).
62 * Meaning that this reg is a pointer to stack plus known immediate constant.
64 * Most of the time the registers have UNKNOWN_VALUE type, which
65 * means the register has some value, but it's not a valid pointer.
66 * (like pointer plus pointer becomes UNKNOWN_VALUE type)
68 * When verifier sees load or store instructions the type of base register
69 * can be: PTR_TO_MAP_VALUE, PTR_TO_CTX, FRAME_PTR. These are three pointer
70 * types recognized by check_mem_access() function.
72 * PTR_TO_MAP_VALUE means that this register is pointing to 'map element value'
73 * and the range of [ptr, ptr + map's value_size) is accessible.
75 * registers used to pass values to function calls are checked against
76 * function argument constraints.
78 * ARG_PTR_TO_MAP_KEY is one of such argument constraints.
79 * It means that the register type passed to this function must be
80 * PTR_TO_STACK and it will be used inside the function as
81 * 'pointer to map element key'
83 * For example the argument constraints for bpf_map_lookup_elem():
84 * .ret_type = RET_PTR_TO_MAP_VALUE_OR_NULL,
85 * .arg1_type = ARG_CONST_MAP_PTR,
86 * .arg2_type = ARG_PTR_TO_MAP_KEY,
88 * ret_type says that this function returns 'pointer to map elem value or null'
89 * function expects 1st argument to be a const pointer to 'struct bpf_map' and
90 * 2nd argument should be a pointer to stack, which will be used inside
91 * the helper function as a pointer to map element key.
93 * On the kernel side the helper function looks like:
94 * u64 bpf_map_lookup_elem(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5)
96 * struct bpf_map *map = (struct bpf_map *) (unsigned long) r1;
97 * void *key = (void *) (unsigned long) r2;
100 * here kernel can access 'key' and 'map' pointers safely, knowing that
101 * [key, key + map->key_size) bytes are valid and were initialized on
102 * the stack of eBPF program.
105 * Corresponding eBPF program may look like:
106 * BPF_MOV64_REG(BPF_REG_2, BPF_REG_10), // after this insn R2 type is FRAME_PTR
107 * BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, -4), // after this insn R2 type is PTR_TO_STACK
108 * BPF_LD_MAP_FD(BPF_REG_1, map_fd), // after this insn R1 type is CONST_PTR_TO_MAP
109 * BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0, BPF_FUNC_map_lookup_elem),
110 * here verifier looks at prototype of map_lookup_elem() and sees:
111 * .arg1_type == ARG_CONST_MAP_PTR and R1->type == CONST_PTR_TO_MAP, which is ok,
112 * Now verifier knows that this map has key of R1->map_ptr->key_size bytes
114 * Then .arg2_type == ARG_PTR_TO_MAP_KEY and R2->type == PTR_TO_STACK, ok so far,
115 * Now verifier checks that [R2, R2 + map's key_size) are within stack limits
116 * and were initialized prior to this call.
117 * If it's ok, then verifier allows this BPF_CALL insn and looks at
118 * .ret_type which is RET_PTR_TO_MAP_VALUE_OR_NULL, so it sets
119 * R0->type = PTR_TO_MAP_VALUE_OR_NULL which means bpf_map_lookup_elem() function
120 * returns ether pointer to map value or NULL.
122 * When type PTR_TO_MAP_VALUE_OR_NULL passes through 'if (reg != 0) goto +off'
123 * insn, the register holding that pointer in the true branch changes state to
124 * PTR_TO_MAP_VALUE and the same register changes state to CONST_IMM in the false
125 * branch. See check_cond_jmp_op().
127 * After the call R0 is set to return type of the function and registers R1-R5
128 * are set to NOT_INIT to indicate that they are no longer readable.
131 /* verifier_state + insn_idx are pushed to stack when branch is encountered */
132 struct bpf_verifier_stack_elem
{
133 /* verifer state is 'st'
134 * before processing instruction 'insn_idx'
135 * and after processing instruction 'prev_insn_idx'
137 struct bpf_verifier_state st
;
140 struct bpf_verifier_stack_elem
*next
;
143 #define BPF_COMPLEXITY_LIMIT_INSNS 98304
144 #define BPF_COMPLEXITY_LIMIT_STACK 1024
146 #define BPF_MAP_PTR_POISON ((void *)0xeB9F + POISON_POINTER_DELTA)
148 struct bpf_call_arg_meta
{
149 struct bpf_map
*map_ptr
;
156 /* verbose verifier prints what it's seeing
157 * bpf_check() is called under lock, so no race to access these global vars
159 static u32 log_level
, log_size
, log_len
;
160 static char *log_buf
;
162 static DEFINE_MUTEX(bpf_verifier_lock
);
164 /* log_level controls verbosity level of eBPF verifier.
165 * verbose() is used to dump the verification trace to the log, so the user
166 * can figure out what's wrong with the program
168 static __printf(1, 2) void verbose(const char *fmt
, ...)
172 if (log_level
== 0 || log_len
>= log_size
- 1)
176 log_len
+= vscnprintf(log_buf
+ log_len
, log_size
- log_len
, fmt
, args
);
180 /* string representation of 'enum bpf_reg_type' */
181 static const char * const reg_type_str
[] = {
183 [UNKNOWN_VALUE
] = "inv",
184 [PTR_TO_CTX
] = "ctx",
185 [CONST_PTR_TO_MAP
] = "map_ptr",
186 [PTR_TO_MAP_VALUE
] = "map_value",
187 [PTR_TO_MAP_VALUE_OR_NULL
] = "map_value_or_null",
188 [PTR_TO_MAP_VALUE_ADJ
] = "map_value_adj",
190 [PTR_TO_STACK
] = "fp",
192 [PTR_TO_PACKET
] = "pkt",
193 [PTR_TO_PACKET_END
] = "pkt_end",
196 #define __BPF_FUNC_STR_FN(x) [BPF_FUNC_ ## x] = __stringify(bpf_ ## x)
197 static const char * const func_id_str
[] = {
198 __BPF_FUNC_MAPPER(__BPF_FUNC_STR_FN
)
200 #undef __BPF_FUNC_STR_FN
202 static const char *func_id_name(int id
)
204 BUILD_BUG_ON(ARRAY_SIZE(func_id_str
) != __BPF_FUNC_MAX_ID
);
206 if (id
>= 0 && id
< __BPF_FUNC_MAX_ID
&& func_id_str
[id
])
207 return func_id_str
[id
];
212 static void print_verifier_state(struct bpf_verifier_state
*state
)
214 struct bpf_reg_state
*reg
;
218 for (i
= 0; i
< MAX_BPF_REG
; i
++) {
219 reg
= &state
->regs
[i
];
223 verbose(" R%d=%s", i
, reg_type_str
[t
]);
224 if (t
== CONST_IMM
|| t
== PTR_TO_STACK
)
225 verbose("%lld", reg
->imm
);
226 else if (t
== PTR_TO_PACKET
)
227 verbose("(id=%d,off=%d,r=%d)",
228 reg
->id
, reg
->off
, reg
->range
);
229 else if (t
== UNKNOWN_VALUE
&& reg
->imm
)
230 verbose("%lld", reg
->imm
);
231 else if (t
== CONST_PTR_TO_MAP
|| t
== PTR_TO_MAP_VALUE
||
232 t
== PTR_TO_MAP_VALUE_OR_NULL
||
233 t
== PTR_TO_MAP_VALUE_ADJ
)
234 verbose("(ks=%d,vs=%d,id=%u)",
235 reg
->map_ptr
->key_size
,
236 reg
->map_ptr
->value_size
,
238 if (reg
->min_value
!= BPF_REGISTER_MIN_RANGE
)
239 verbose(",min_value=%lld",
240 (long long)reg
->min_value
);
241 if (reg
->max_value
!= BPF_REGISTER_MAX_RANGE
)
242 verbose(",max_value=%llu",
243 (unsigned long long)reg
->max_value
);
245 verbose(",min_align=%u", reg
->min_align
);
247 verbose(",aux_off=%u", reg
->aux_off
);
248 if (reg
->aux_off_align
)
249 verbose(",aux_off_align=%u", reg
->aux_off_align
);
251 for (i
= 0; i
< MAX_BPF_STACK
; i
+= BPF_REG_SIZE
) {
252 if (state
->stack_slot_type
[i
] == STACK_SPILL
)
253 verbose(" fp%d=%s", -MAX_BPF_STACK
+ i
,
254 reg_type_str
[state
->spilled_regs
[i
/ BPF_REG_SIZE
].type
]);
259 static const char *const bpf_class_string
[] = {
267 [BPF_ALU64
] = "alu64",
270 static const char *const bpf_alu_string
[16] = {
271 [BPF_ADD
>> 4] = "+=",
272 [BPF_SUB
>> 4] = "-=",
273 [BPF_MUL
>> 4] = "*=",
274 [BPF_DIV
>> 4] = "/=",
275 [BPF_OR
>> 4] = "|=",
276 [BPF_AND
>> 4] = "&=",
277 [BPF_LSH
>> 4] = "<<=",
278 [BPF_RSH
>> 4] = ">>=",
279 [BPF_NEG
>> 4] = "neg",
280 [BPF_MOD
>> 4] = "%=",
281 [BPF_XOR
>> 4] = "^=",
282 [BPF_MOV
>> 4] = "=",
283 [BPF_ARSH
>> 4] = "s>>=",
284 [BPF_END
>> 4] = "endian",
287 static const char *const bpf_ldst_string
[] = {
288 [BPF_W
>> 3] = "u32",
289 [BPF_H
>> 3] = "u16",
291 [BPF_DW
>> 3] = "u64",
294 static const char *const bpf_jmp_string
[16] = {
295 [BPF_JA
>> 4] = "jmp",
296 [BPF_JEQ
>> 4] = "==",
297 [BPF_JGT
>> 4] = ">",
298 [BPF_JGE
>> 4] = ">=",
299 [BPF_JSET
>> 4] = "&",
300 [BPF_JNE
>> 4] = "!=",
301 [BPF_JSGT
>> 4] = "s>",
302 [BPF_JSGE
>> 4] = "s>=",
303 [BPF_CALL
>> 4] = "call",
304 [BPF_EXIT
>> 4] = "exit",
307 static void print_bpf_insn(const struct bpf_verifier_env
*env
,
308 const struct bpf_insn
*insn
)
310 u8
class = BPF_CLASS(insn
->code
);
312 if (class == BPF_ALU
|| class == BPF_ALU64
) {
313 if (BPF_SRC(insn
->code
) == BPF_X
)
314 verbose("(%02x) %sr%d %s %sr%d\n",
315 insn
->code
, class == BPF_ALU
? "(u32) " : "",
317 bpf_alu_string
[BPF_OP(insn
->code
) >> 4],
318 class == BPF_ALU
? "(u32) " : "",
321 verbose("(%02x) %sr%d %s %s%d\n",
322 insn
->code
, class == BPF_ALU
? "(u32) " : "",
324 bpf_alu_string
[BPF_OP(insn
->code
) >> 4],
325 class == BPF_ALU
? "(u32) " : "",
327 } else if (class == BPF_STX
) {
328 if (BPF_MODE(insn
->code
) == BPF_MEM
)
329 verbose("(%02x) *(%s *)(r%d %+d) = r%d\n",
331 bpf_ldst_string
[BPF_SIZE(insn
->code
) >> 3],
333 insn
->off
, insn
->src_reg
);
334 else if (BPF_MODE(insn
->code
) == BPF_XADD
)
335 verbose("(%02x) lock *(%s *)(r%d %+d) += r%d\n",
337 bpf_ldst_string
[BPF_SIZE(insn
->code
) >> 3],
338 insn
->dst_reg
, insn
->off
,
341 verbose("BUG_%02x\n", insn
->code
);
342 } else if (class == BPF_ST
) {
343 if (BPF_MODE(insn
->code
) != BPF_MEM
) {
344 verbose("BUG_st_%02x\n", insn
->code
);
347 verbose("(%02x) *(%s *)(r%d %+d) = %d\n",
349 bpf_ldst_string
[BPF_SIZE(insn
->code
) >> 3],
351 insn
->off
, insn
->imm
);
352 } else if (class == BPF_LDX
) {
353 if (BPF_MODE(insn
->code
) != BPF_MEM
) {
354 verbose("BUG_ldx_%02x\n", insn
->code
);
357 verbose("(%02x) r%d = *(%s *)(r%d %+d)\n",
358 insn
->code
, insn
->dst_reg
,
359 bpf_ldst_string
[BPF_SIZE(insn
->code
) >> 3],
360 insn
->src_reg
, insn
->off
);
361 } else if (class == BPF_LD
) {
362 if (BPF_MODE(insn
->code
) == BPF_ABS
) {
363 verbose("(%02x) r0 = *(%s *)skb[%d]\n",
365 bpf_ldst_string
[BPF_SIZE(insn
->code
) >> 3],
367 } else if (BPF_MODE(insn
->code
) == BPF_IND
) {
368 verbose("(%02x) r0 = *(%s *)skb[r%d + %d]\n",
370 bpf_ldst_string
[BPF_SIZE(insn
->code
) >> 3],
371 insn
->src_reg
, insn
->imm
);
372 } else if (BPF_MODE(insn
->code
) == BPF_IMM
&&
373 BPF_SIZE(insn
->code
) == BPF_DW
) {
374 /* At this point, we already made sure that the second
375 * part of the ldimm64 insn is accessible.
377 u64 imm
= ((u64
)(insn
+ 1)->imm
<< 32) | (u32
)insn
->imm
;
378 bool map_ptr
= insn
->src_reg
== BPF_PSEUDO_MAP_FD
;
380 if (map_ptr
&& !env
->allow_ptr_leaks
)
383 verbose("(%02x) r%d = 0x%llx\n", insn
->code
,
384 insn
->dst_reg
, (unsigned long long)imm
);
386 verbose("BUG_ld_%02x\n", insn
->code
);
389 } else if (class == BPF_JMP
) {
390 u8 opcode
= BPF_OP(insn
->code
);
392 if (opcode
== BPF_CALL
) {
393 verbose("(%02x) call %s#%d\n", insn
->code
,
394 func_id_name(insn
->imm
), insn
->imm
);
395 } else if (insn
->code
== (BPF_JMP
| BPF_JA
)) {
396 verbose("(%02x) goto pc%+d\n",
397 insn
->code
, insn
->off
);
398 } else if (insn
->code
== (BPF_JMP
| BPF_EXIT
)) {
399 verbose("(%02x) exit\n", insn
->code
);
400 } else if (BPF_SRC(insn
->code
) == BPF_X
) {
401 verbose("(%02x) if r%d %s r%d goto pc%+d\n",
402 insn
->code
, insn
->dst_reg
,
403 bpf_jmp_string
[BPF_OP(insn
->code
) >> 4],
404 insn
->src_reg
, insn
->off
);
406 verbose("(%02x) if r%d %s 0x%x goto pc%+d\n",
407 insn
->code
, insn
->dst_reg
,
408 bpf_jmp_string
[BPF_OP(insn
->code
) >> 4],
409 insn
->imm
, insn
->off
);
412 verbose("(%02x) %s\n", insn
->code
, bpf_class_string
[class]);
416 static int pop_stack(struct bpf_verifier_env
*env
, int *prev_insn_idx
)
418 struct bpf_verifier_stack_elem
*elem
;
421 if (env
->head
== NULL
)
424 memcpy(&env
->cur_state
, &env
->head
->st
, sizeof(env
->cur_state
));
425 insn_idx
= env
->head
->insn_idx
;
427 *prev_insn_idx
= env
->head
->prev_insn_idx
;
428 elem
= env
->head
->next
;
435 static struct bpf_verifier_state
*push_stack(struct bpf_verifier_env
*env
,
436 int insn_idx
, int prev_insn_idx
)
438 struct bpf_verifier_stack_elem
*elem
;
440 elem
= kmalloc(sizeof(struct bpf_verifier_stack_elem
), GFP_KERNEL
);
444 memcpy(&elem
->st
, &env
->cur_state
, sizeof(env
->cur_state
));
445 elem
->insn_idx
= insn_idx
;
446 elem
->prev_insn_idx
= prev_insn_idx
;
447 elem
->next
= env
->head
;
450 if (env
->stack_size
> BPF_COMPLEXITY_LIMIT_STACK
) {
451 verbose("BPF program is too complex\n");
456 /* pop all elements and return */
457 while (pop_stack(env
, NULL
) >= 0);
461 #define CALLER_SAVED_REGS 6
462 static const int caller_saved
[CALLER_SAVED_REGS
] = {
463 BPF_REG_0
, BPF_REG_1
, BPF_REG_2
, BPF_REG_3
, BPF_REG_4
, BPF_REG_5
466 static void mark_reg_not_init(struct bpf_reg_state
*regs
, u32 regno
)
468 BUG_ON(regno
>= MAX_BPF_REG
);
470 memset(®s
[regno
], 0, sizeof(regs
[regno
]));
471 regs
[regno
].type
= NOT_INIT
;
472 regs
[regno
].min_value
= BPF_REGISTER_MIN_RANGE
;
473 regs
[regno
].max_value
= BPF_REGISTER_MAX_RANGE
;
476 static void init_reg_state(struct bpf_reg_state
*regs
)
480 for (i
= 0; i
< MAX_BPF_REG
; i
++)
481 mark_reg_not_init(regs
, i
);
484 regs
[BPF_REG_FP
].type
= FRAME_PTR
;
486 /* 1st arg to a function */
487 regs
[BPF_REG_1
].type
= PTR_TO_CTX
;
490 static void __mark_reg_unknown_value(struct bpf_reg_state
*regs
, u32 regno
)
492 regs
[regno
].type
= UNKNOWN_VALUE
;
497 static void mark_reg_unknown_value(struct bpf_reg_state
*regs
, u32 regno
)
499 BUG_ON(regno
>= MAX_BPF_REG
);
500 __mark_reg_unknown_value(regs
, regno
);
503 static void reset_reg_range_values(struct bpf_reg_state
*regs
, u32 regno
)
505 regs
[regno
].min_value
= BPF_REGISTER_MIN_RANGE
;
506 regs
[regno
].max_value
= BPF_REGISTER_MAX_RANGE
;
507 regs
[regno
].min_align
= 0;
510 static void mark_reg_unknown_value_and_range(struct bpf_reg_state
*regs
,
513 mark_reg_unknown_value(regs
, regno
);
514 reset_reg_range_values(regs
, regno
);
518 SRC_OP
, /* register is used as source operand */
519 DST_OP
, /* register is used as destination operand */
520 DST_OP_NO_MARK
/* same as above, check only, don't mark */
523 static int check_reg_arg(struct bpf_reg_state
*regs
, u32 regno
,
526 if (regno
>= MAX_BPF_REG
) {
527 verbose("R%d is invalid\n", regno
);
532 /* check whether register used as source operand can be read */
533 if (regs
[regno
].type
== NOT_INIT
) {
534 verbose("R%d !read_ok\n", regno
);
538 /* check whether register used as dest operand can be written to */
539 if (regno
== BPF_REG_FP
) {
540 verbose("frame pointer is read only\n");
544 mark_reg_unknown_value(regs
, regno
);
549 static int bpf_size_to_bytes(int bpf_size
)
551 if (bpf_size
== BPF_W
)
553 else if (bpf_size
== BPF_H
)
555 else if (bpf_size
== BPF_B
)
557 else if (bpf_size
== BPF_DW
)
563 static bool is_spillable_regtype(enum bpf_reg_type type
)
566 case PTR_TO_MAP_VALUE
:
567 case PTR_TO_MAP_VALUE_OR_NULL
:
568 case PTR_TO_MAP_VALUE_ADJ
:
572 case PTR_TO_PACKET_END
:
574 case CONST_PTR_TO_MAP
:
581 /* check_stack_read/write functions track spill/fill of registers,
582 * stack boundary and alignment are checked in check_mem_access()
584 static int check_stack_write(struct bpf_verifier_state
*state
, int off
,
585 int size
, int value_regno
)
588 /* caller checked that off % size == 0 and -MAX_BPF_STACK <= off < 0,
589 * so it's aligned access and [off, off + size) are within stack limits
592 if (value_regno
>= 0 &&
593 is_spillable_regtype(state
->regs
[value_regno
].type
)) {
595 /* register containing pointer is being spilled into stack */
596 if (size
!= BPF_REG_SIZE
) {
597 verbose("invalid size of register spill\n");
601 /* save register state */
602 state
->spilled_regs
[(MAX_BPF_STACK
+ off
) / BPF_REG_SIZE
] =
603 state
->regs
[value_regno
];
605 for (i
= 0; i
< BPF_REG_SIZE
; i
++)
606 state
->stack_slot_type
[MAX_BPF_STACK
+ off
+ i
] = STACK_SPILL
;
608 /* regular write of data into stack */
609 state
->spilled_regs
[(MAX_BPF_STACK
+ off
) / BPF_REG_SIZE
] =
610 (struct bpf_reg_state
) {};
612 for (i
= 0; i
< size
; i
++)
613 state
->stack_slot_type
[MAX_BPF_STACK
+ off
+ i
] = STACK_MISC
;
618 static int check_stack_read(struct bpf_verifier_state
*state
, int off
, int size
,
624 slot_type
= &state
->stack_slot_type
[MAX_BPF_STACK
+ off
];
626 if (slot_type
[0] == STACK_SPILL
) {
627 if (size
!= BPF_REG_SIZE
) {
628 verbose("invalid size of register spill\n");
631 for (i
= 1; i
< BPF_REG_SIZE
; i
++) {
632 if (slot_type
[i
] != STACK_SPILL
) {
633 verbose("corrupted spill memory\n");
638 if (value_regno
>= 0)
639 /* restore register state from stack */
640 state
->regs
[value_regno
] =
641 state
->spilled_regs
[(MAX_BPF_STACK
+ off
) / BPF_REG_SIZE
];
644 for (i
= 0; i
< size
; i
++) {
645 if (slot_type
[i
] != STACK_MISC
) {
646 verbose("invalid read from stack off %d+%d size %d\n",
651 if (value_regno
>= 0)
652 /* have read misc data from the stack */
653 mark_reg_unknown_value_and_range(state
->regs
,
659 /* check read/write into map element returned by bpf_map_lookup_elem() */
660 static int check_map_access(struct bpf_verifier_env
*env
, u32 regno
, int off
,
663 struct bpf_map
*map
= env
->cur_state
.regs
[regno
].map_ptr
;
665 if (off
< 0 || size
<= 0 || off
+ size
> map
->value_size
) {
666 verbose("invalid access to map value, value_size=%d off=%d size=%d\n",
667 map
->value_size
, off
, size
);
673 /* check read/write into an adjusted map element */
674 static int check_map_access_adj(struct bpf_verifier_env
*env
, u32 regno
,
677 struct bpf_verifier_state
*state
= &env
->cur_state
;
678 struct bpf_reg_state
*reg
= &state
->regs
[regno
];
681 /* We adjusted the register to this map value, so we
682 * need to change off and size to min_value and max_value
683 * respectively to make sure our theoretical access will be
687 print_verifier_state(state
);
688 env
->varlen_map_value_access
= true;
689 /* The minimum value is only important with signed
690 * comparisons where we can't assume the floor of a
691 * value is 0. If we are using signed variables for our
692 * index'es we need to make sure that whatever we use
693 * will have a set floor within our range.
695 if (reg
->min_value
< 0) {
696 verbose("R%d min value is negative, either use unsigned index or do a if (index >=0) check.\n",
700 err
= check_map_access(env
, regno
, reg
->min_value
+ off
, size
);
702 verbose("R%d min value is outside of the array range\n",
707 /* If we haven't set a max value then we need to bail
708 * since we can't be sure we won't do bad things.
710 if (reg
->max_value
== BPF_REGISTER_MAX_RANGE
) {
711 verbose("R%d unbounded memory access, make sure to bounds check any array access into a map\n",
715 return check_map_access(env
, regno
, reg
->max_value
+ off
, size
);
718 #define MAX_PACKET_OFF 0xffff
720 static bool may_access_direct_pkt_data(struct bpf_verifier_env
*env
,
721 const struct bpf_call_arg_meta
*meta
,
722 enum bpf_access_type t
)
724 switch (env
->prog
->type
) {
725 case BPF_PROG_TYPE_LWT_IN
:
726 case BPF_PROG_TYPE_LWT_OUT
:
727 /* dst_input() and dst_output() can't write for now */
731 case BPF_PROG_TYPE_SCHED_CLS
:
732 case BPF_PROG_TYPE_SCHED_ACT
:
733 case BPF_PROG_TYPE_XDP
:
734 case BPF_PROG_TYPE_LWT_XMIT
:
736 return meta
->pkt_access
;
738 env
->seen_direct_write
= true;
745 static int check_packet_access(struct bpf_verifier_env
*env
, u32 regno
, int off
,
748 struct bpf_reg_state
*regs
= env
->cur_state
.regs
;
749 struct bpf_reg_state
*reg
= ®s
[regno
];
752 if (off
< 0 || size
<= 0 || off
+ size
> reg
->range
) {
753 verbose("invalid access to packet, off=%d size=%d, R%d(id=%d,off=%d,r=%d)\n",
754 off
, size
, regno
, reg
->id
, reg
->off
, reg
->range
);
760 /* check access to 'struct bpf_context' fields */
761 static int check_ctx_access(struct bpf_verifier_env
*env
, int off
, int size
,
762 enum bpf_access_type t
, enum bpf_reg_type
*reg_type
)
764 /* for analyzer ctx accesses are already validated and converted */
765 if (env
->analyzer_ops
)
768 if (env
->prog
->aux
->ops
->is_valid_access
&&
769 env
->prog
->aux
->ops
->is_valid_access(off
, size
, t
, reg_type
)) {
770 /* remember the offset of last byte accessed in ctx */
771 if (env
->prog
->aux
->max_ctx_offset
< off
+ size
)
772 env
->prog
->aux
->max_ctx_offset
= off
+ size
;
776 verbose("invalid bpf_context access off=%d size=%d\n", off
, size
);
780 static bool is_pointer_value(struct bpf_verifier_env
*env
, int regno
)
782 if (env
->allow_ptr_leaks
)
785 switch (env
->cur_state
.regs
[regno
].type
) {
794 static int check_pkt_ptr_alignment(const struct bpf_reg_state
*reg
,
795 int off
, int size
, bool strict
)
800 /* Byte size accesses are always allowed. */
801 if (!strict
|| size
== 1)
806 if (reg
->aux_off_align
% size
) {
807 verbose("Packet access is only %u byte aligned, %d byte access not allowed\n",
808 reg
->aux_off_align
, size
);
811 reg_off
+= reg
->aux_off
;
814 /* For platforms that do not have a Kconfig enabling
815 * CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS the value of
816 * NET_IP_ALIGN is universally set to '2'. And on platforms
817 * that do set CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS, we get
818 * to this code only in strict mode where we want to emulate
819 * the NET_IP_ALIGN==2 checking. Therefore use an
820 * unconditional IP align value of '2'.
823 if ((ip_align
+ reg_off
+ off
) % size
!= 0) {
824 verbose("misaligned packet access off %d+%d+%d size %d\n",
825 ip_align
, reg_off
, off
, size
);
832 static int check_val_ptr_alignment(const struct bpf_reg_state
*reg
,
833 int size
, bool strict
)
835 if (strict
&& size
!= 1) {
836 verbose("Unknown alignment. Only byte-sized access allowed in value access.\n");
843 static int check_ptr_alignment(struct bpf_verifier_env
*env
,
844 const struct bpf_reg_state
*reg
,
847 bool strict
= env
->strict_alignment
;
851 return check_pkt_ptr_alignment(reg
, off
, size
, strict
);
852 case PTR_TO_MAP_VALUE_ADJ
:
853 return check_val_ptr_alignment(reg
, size
, strict
);
855 if (off
% size
!= 0) {
856 verbose("misaligned access off %d size %d\n",
865 /* check whether memory at (regno + off) is accessible for t = (read | write)
866 * if t==write, value_regno is a register which value is stored into memory
867 * if t==read, value_regno is a register which will receive the value from memory
868 * if t==write && value_regno==-1, some unknown value is stored into memory
869 * if t==read && value_regno==-1, don't care what we read from memory
871 static int check_mem_access(struct bpf_verifier_env
*env
, u32 regno
, int off
,
872 int bpf_size
, enum bpf_access_type t
,
875 struct bpf_verifier_state
*state
= &env
->cur_state
;
876 struct bpf_reg_state
*reg
= &state
->regs
[regno
];
879 if (reg
->type
== PTR_TO_STACK
)
882 size
= bpf_size_to_bytes(bpf_size
);
886 err
= check_ptr_alignment(env
, reg
, off
, size
);
890 if (reg
->type
== PTR_TO_MAP_VALUE
||
891 reg
->type
== PTR_TO_MAP_VALUE_ADJ
) {
892 if (t
== BPF_WRITE
&& value_regno
>= 0 &&
893 is_pointer_value(env
, value_regno
)) {
894 verbose("R%d leaks addr into map\n", value_regno
);
898 if (reg
->type
== PTR_TO_MAP_VALUE_ADJ
)
899 err
= check_map_access_adj(env
, regno
, off
, size
);
901 err
= check_map_access(env
, regno
, off
, size
);
902 if (!err
&& t
== BPF_READ
&& value_regno
>= 0)
903 mark_reg_unknown_value_and_range(state
->regs
,
906 } else if (reg
->type
== PTR_TO_CTX
) {
907 enum bpf_reg_type reg_type
= UNKNOWN_VALUE
;
909 if (t
== BPF_WRITE
&& value_regno
>= 0 &&
910 is_pointer_value(env
, value_regno
)) {
911 verbose("R%d leaks addr into ctx\n", value_regno
);
914 err
= check_ctx_access(env
, off
, size
, t
, ®_type
);
915 if (!err
&& t
== BPF_READ
&& value_regno
>= 0) {
916 mark_reg_unknown_value_and_range(state
->regs
,
918 /* note that reg.[id|off|range] == 0 */
919 state
->regs
[value_regno
].type
= reg_type
;
920 state
->regs
[value_regno
].aux_off
= 0;
921 state
->regs
[value_regno
].aux_off_align
= 0;
924 } else if (reg
->type
== FRAME_PTR
|| reg
->type
== PTR_TO_STACK
) {
925 if (off
>= 0 || off
< -MAX_BPF_STACK
) {
926 verbose("invalid stack off=%d size=%d\n", off
, size
);
930 if (env
->prog
->aux
->stack_depth
< -off
)
931 env
->prog
->aux
->stack_depth
= -off
;
933 if (t
== BPF_WRITE
) {
934 if (!env
->allow_ptr_leaks
&&
935 state
->stack_slot_type
[MAX_BPF_STACK
+ off
] == STACK_SPILL
&&
936 size
!= BPF_REG_SIZE
) {
937 verbose("attempt to corrupt spilled pointer on stack\n");
940 err
= check_stack_write(state
, off
, size
, value_regno
);
942 err
= check_stack_read(state
, off
, size
, value_regno
);
944 } else if (state
->regs
[regno
].type
== PTR_TO_PACKET
) {
945 if (t
== BPF_WRITE
&& !may_access_direct_pkt_data(env
, NULL
, t
)) {
946 verbose("cannot write into packet\n");
949 if (t
== BPF_WRITE
&& value_regno
>= 0 &&
950 is_pointer_value(env
, value_regno
)) {
951 verbose("R%d leaks addr into packet\n", value_regno
);
954 err
= check_packet_access(env
, regno
, off
, size
);
955 if (!err
&& t
== BPF_READ
&& value_regno
>= 0)
956 mark_reg_unknown_value_and_range(state
->regs
,
959 verbose("R%d invalid mem access '%s'\n",
960 regno
, reg_type_str
[reg
->type
]);
964 if (!err
&& size
<= 2 && value_regno
>= 0 && env
->allow_ptr_leaks
&&
965 state
->regs
[value_regno
].type
== UNKNOWN_VALUE
) {
966 /* 1 or 2 byte load zero-extends, determine the number of
967 * zero upper bits. Not doing it fo 4 byte load, since
968 * such values cannot be added to ptr_to_packet anyway.
970 state
->regs
[value_regno
].imm
= 64 - size
* 8;
975 static int check_xadd(struct bpf_verifier_env
*env
, struct bpf_insn
*insn
)
977 struct bpf_reg_state
*regs
= env
->cur_state
.regs
;
980 if ((BPF_SIZE(insn
->code
) != BPF_W
&& BPF_SIZE(insn
->code
) != BPF_DW
) ||
982 verbose("BPF_XADD uses reserved fields\n");
986 /* check src1 operand */
987 err
= check_reg_arg(regs
, insn
->src_reg
, SRC_OP
);
991 /* check src2 operand */
992 err
= check_reg_arg(regs
, insn
->dst_reg
, SRC_OP
);
996 /* check whether atomic_add can read the memory */
997 err
= check_mem_access(env
, insn
->dst_reg
, insn
->off
,
998 BPF_SIZE(insn
->code
), BPF_READ
, -1);
1002 /* check whether atomic_add can write into the same memory */
1003 return check_mem_access(env
, insn
->dst_reg
, insn
->off
,
1004 BPF_SIZE(insn
->code
), BPF_WRITE
, -1);
1007 /* when register 'regno' is passed into function that will read 'access_size'
1008 * bytes from that pointer, make sure that it's within stack boundary
1009 * and all elements of stack are initialized
1011 static int check_stack_boundary(struct bpf_verifier_env
*env
, int regno
,
1012 int access_size
, bool zero_size_allowed
,
1013 struct bpf_call_arg_meta
*meta
)
1015 struct bpf_verifier_state
*state
= &env
->cur_state
;
1016 struct bpf_reg_state
*regs
= state
->regs
;
1019 if (regs
[regno
].type
!= PTR_TO_STACK
) {
1020 if (zero_size_allowed
&& access_size
== 0 &&
1021 regs
[regno
].type
== CONST_IMM
&&
1022 regs
[regno
].imm
== 0)
1025 verbose("R%d type=%s expected=%s\n", regno
,
1026 reg_type_str
[regs
[regno
].type
],
1027 reg_type_str
[PTR_TO_STACK
]);
1031 off
= regs
[regno
].imm
;
1032 if (off
>= 0 || off
< -MAX_BPF_STACK
|| off
+ access_size
> 0 ||
1034 verbose("invalid stack type R%d off=%d access_size=%d\n",
1035 regno
, off
, access_size
);
1039 if (env
->prog
->aux
->stack_depth
< -off
)
1040 env
->prog
->aux
->stack_depth
= -off
;
1042 if (meta
&& meta
->raw_mode
) {
1043 meta
->access_size
= access_size
;
1044 meta
->regno
= regno
;
1048 for (i
= 0; i
< access_size
; i
++) {
1049 if (state
->stack_slot_type
[MAX_BPF_STACK
+ off
+ i
] != STACK_MISC
) {
1050 verbose("invalid indirect read from stack off %d+%d size %d\n",
1051 off
, i
, access_size
);
1058 static int check_helper_mem_access(struct bpf_verifier_env
*env
, int regno
,
1059 int access_size
, bool zero_size_allowed
,
1060 struct bpf_call_arg_meta
*meta
)
1062 struct bpf_reg_state
*regs
= env
->cur_state
.regs
;
1064 switch (regs
[regno
].type
) {
1066 return check_packet_access(env
, regno
, 0, access_size
);
1067 case PTR_TO_MAP_VALUE
:
1068 return check_map_access(env
, regno
, 0, access_size
);
1069 case PTR_TO_MAP_VALUE_ADJ
:
1070 return check_map_access_adj(env
, regno
, 0, access_size
);
1071 default: /* const_imm|ptr_to_stack or invalid ptr */
1072 return check_stack_boundary(env
, regno
, access_size
,
1073 zero_size_allowed
, meta
);
1077 static int check_func_arg(struct bpf_verifier_env
*env
, u32 regno
,
1078 enum bpf_arg_type arg_type
,
1079 struct bpf_call_arg_meta
*meta
)
1081 struct bpf_reg_state
*regs
= env
->cur_state
.regs
, *reg
= ®s
[regno
];
1082 enum bpf_reg_type expected_type
, type
= reg
->type
;
1085 if (arg_type
== ARG_DONTCARE
)
1088 if (type
== NOT_INIT
) {
1089 verbose("R%d !read_ok\n", regno
);
1093 if (arg_type
== ARG_ANYTHING
) {
1094 if (is_pointer_value(env
, regno
)) {
1095 verbose("R%d leaks addr into helper function\n", regno
);
1101 if (type
== PTR_TO_PACKET
&&
1102 !may_access_direct_pkt_data(env
, meta
, BPF_READ
)) {
1103 verbose("helper access to the packet is not allowed\n");
1107 if (arg_type
== ARG_PTR_TO_MAP_KEY
||
1108 arg_type
== ARG_PTR_TO_MAP_VALUE
) {
1109 expected_type
= PTR_TO_STACK
;
1110 if (type
!= PTR_TO_PACKET
&& type
!= expected_type
)
1112 } else if (arg_type
== ARG_CONST_SIZE
||
1113 arg_type
== ARG_CONST_SIZE_OR_ZERO
) {
1114 expected_type
= CONST_IMM
;
1115 /* One exception. Allow UNKNOWN_VALUE registers when the
1116 * boundaries are known and don't cause unsafe memory accesses
1118 if (type
!= UNKNOWN_VALUE
&& type
!= expected_type
)
1120 } else if (arg_type
== ARG_CONST_MAP_PTR
) {
1121 expected_type
= CONST_PTR_TO_MAP
;
1122 if (type
!= expected_type
)
1124 } else if (arg_type
== ARG_PTR_TO_CTX
) {
1125 expected_type
= PTR_TO_CTX
;
1126 if (type
!= expected_type
)
1128 } else if (arg_type
== ARG_PTR_TO_MEM
||
1129 arg_type
== ARG_PTR_TO_UNINIT_MEM
) {
1130 expected_type
= PTR_TO_STACK
;
1131 /* One exception here. In case function allows for NULL to be
1132 * passed in as argument, it's a CONST_IMM type. Final test
1133 * happens during stack boundary checking.
1135 if (type
== CONST_IMM
&& reg
->imm
== 0)
1136 /* final test in check_stack_boundary() */;
1137 else if (type
!= PTR_TO_PACKET
&& type
!= PTR_TO_MAP_VALUE
&&
1138 type
!= PTR_TO_MAP_VALUE_ADJ
&& type
!= expected_type
)
1140 meta
->raw_mode
= arg_type
== ARG_PTR_TO_UNINIT_MEM
;
1142 verbose("unsupported arg_type %d\n", arg_type
);
1146 if (arg_type
== ARG_CONST_MAP_PTR
) {
1147 /* bpf_map_xxx(map_ptr) call: remember that map_ptr */
1148 meta
->map_ptr
= reg
->map_ptr
;
1149 } else if (arg_type
== ARG_PTR_TO_MAP_KEY
) {
1150 /* bpf_map_xxx(..., map_ptr, ..., key) call:
1151 * check that [key, key + map->key_size) are within
1152 * stack limits and initialized
1154 if (!meta
->map_ptr
) {
1155 /* in function declaration map_ptr must come before
1156 * map_key, so that it's verified and known before
1157 * we have to check map_key here. Otherwise it means
1158 * that kernel subsystem misconfigured verifier
1160 verbose("invalid map_ptr to access map->key\n");
1163 if (type
== PTR_TO_PACKET
)
1164 err
= check_packet_access(env
, regno
, 0,
1165 meta
->map_ptr
->key_size
);
1167 err
= check_stack_boundary(env
, regno
,
1168 meta
->map_ptr
->key_size
,
1170 } else if (arg_type
== ARG_PTR_TO_MAP_VALUE
) {
1171 /* bpf_map_xxx(..., map_ptr, ..., value) call:
1172 * check [value, value + map->value_size) validity
1174 if (!meta
->map_ptr
) {
1175 /* kernel subsystem misconfigured verifier */
1176 verbose("invalid map_ptr to access map->value\n");
1179 if (type
== PTR_TO_PACKET
)
1180 err
= check_packet_access(env
, regno
, 0,
1181 meta
->map_ptr
->value_size
);
1183 err
= check_stack_boundary(env
, regno
,
1184 meta
->map_ptr
->value_size
,
1186 } else if (arg_type
== ARG_CONST_SIZE
||
1187 arg_type
== ARG_CONST_SIZE_OR_ZERO
) {
1188 bool zero_size_allowed
= (arg_type
== ARG_CONST_SIZE_OR_ZERO
);
1190 /* bpf_xxx(..., buf, len) call will access 'len' bytes
1191 * from stack pointer 'buf'. Check it
1192 * note: regno == len, regno - 1 == buf
1195 /* kernel subsystem misconfigured verifier */
1196 verbose("ARG_CONST_SIZE cannot be first argument\n");
1200 /* If the register is UNKNOWN_VALUE, the access check happens
1201 * using its boundaries. Otherwise, just use its imm
1203 if (type
== UNKNOWN_VALUE
) {
1204 /* For unprivileged variable accesses, disable raw
1205 * mode so that the program is required to
1206 * initialize all the memory that the helper could
1207 * just partially fill up.
1211 if (reg
->min_value
< 0) {
1212 verbose("R%d min value is negative, either use unsigned or 'var &= const'\n",
1217 if (reg
->min_value
== 0) {
1218 err
= check_helper_mem_access(env
, regno
- 1, 0,
1225 if (reg
->max_value
== BPF_REGISTER_MAX_RANGE
) {
1226 verbose("R%d unbounded memory access, use 'var &= const' or 'if (var < const)'\n",
1230 err
= check_helper_mem_access(env
, regno
- 1,
1232 zero_size_allowed
, meta
);
1236 /* register is CONST_IMM */
1237 err
= check_helper_mem_access(env
, regno
- 1, reg
->imm
,
1238 zero_size_allowed
, meta
);
1244 verbose("R%d type=%s expected=%s\n", regno
,
1245 reg_type_str
[type
], reg_type_str
[expected_type
]);
1249 static int check_map_func_compatibility(struct bpf_map
*map
, int func_id
)
1254 /* We need a two way check, first is from map perspective ... */
1255 switch (map
->map_type
) {
1256 case BPF_MAP_TYPE_PROG_ARRAY
:
1257 if (func_id
!= BPF_FUNC_tail_call
)
1260 case BPF_MAP_TYPE_PERF_EVENT_ARRAY
:
1261 if (func_id
!= BPF_FUNC_perf_event_read
&&
1262 func_id
!= BPF_FUNC_perf_event_output
)
1265 case BPF_MAP_TYPE_STACK_TRACE
:
1266 if (func_id
!= BPF_FUNC_get_stackid
)
1269 case BPF_MAP_TYPE_CGROUP_ARRAY
:
1270 if (func_id
!= BPF_FUNC_skb_under_cgroup
&&
1271 func_id
!= BPF_FUNC_current_task_under_cgroup
)
1274 case BPF_MAP_TYPE_ARRAY_OF_MAPS
:
1275 case BPF_MAP_TYPE_HASH_OF_MAPS
:
1276 if (func_id
!= BPF_FUNC_map_lookup_elem
)
1282 /* ... and second from the function itself. */
1284 case BPF_FUNC_tail_call
:
1285 if (map
->map_type
!= BPF_MAP_TYPE_PROG_ARRAY
)
1288 case BPF_FUNC_perf_event_read
:
1289 case BPF_FUNC_perf_event_output
:
1290 if (map
->map_type
!= BPF_MAP_TYPE_PERF_EVENT_ARRAY
)
1293 case BPF_FUNC_get_stackid
:
1294 if (map
->map_type
!= BPF_MAP_TYPE_STACK_TRACE
)
1297 case BPF_FUNC_current_task_under_cgroup
:
1298 case BPF_FUNC_skb_under_cgroup
:
1299 if (map
->map_type
!= BPF_MAP_TYPE_CGROUP_ARRAY
)
1308 verbose("cannot pass map_type %d into func %s#%d\n",
1309 map
->map_type
, func_id_name(func_id
), func_id
);
1313 static int check_raw_mode(const struct bpf_func_proto
*fn
)
1317 if (fn
->arg1_type
== ARG_PTR_TO_UNINIT_MEM
)
1319 if (fn
->arg2_type
== ARG_PTR_TO_UNINIT_MEM
)
1321 if (fn
->arg3_type
== ARG_PTR_TO_UNINIT_MEM
)
1323 if (fn
->arg4_type
== ARG_PTR_TO_UNINIT_MEM
)
1325 if (fn
->arg5_type
== ARG_PTR_TO_UNINIT_MEM
)
1328 return count
> 1 ? -EINVAL
: 0;
1331 static void clear_all_pkt_pointers(struct bpf_verifier_env
*env
)
1333 struct bpf_verifier_state
*state
= &env
->cur_state
;
1334 struct bpf_reg_state
*regs
= state
->regs
, *reg
;
1337 for (i
= 0; i
< MAX_BPF_REG
; i
++)
1338 if (regs
[i
].type
== PTR_TO_PACKET
||
1339 regs
[i
].type
== PTR_TO_PACKET_END
)
1340 mark_reg_unknown_value(regs
, i
);
1342 for (i
= 0; i
< MAX_BPF_STACK
; i
+= BPF_REG_SIZE
) {
1343 if (state
->stack_slot_type
[i
] != STACK_SPILL
)
1345 reg
= &state
->spilled_regs
[i
/ BPF_REG_SIZE
];
1346 if (reg
->type
!= PTR_TO_PACKET
&&
1347 reg
->type
!= PTR_TO_PACKET_END
)
1349 reg
->type
= UNKNOWN_VALUE
;
1354 static int check_call(struct bpf_verifier_env
*env
, int func_id
, int insn_idx
)
1356 struct bpf_verifier_state
*state
= &env
->cur_state
;
1357 const struct bpf_func_proto
*fn
= NULL
;
1358 struct bpf_reg_state
*regs
= state
->regs
;
1359 struct bpf_call_arg_meta meta
;
1363 /* find function prototype */
1364 if (func_id
< 0 || func_id
>= __BPF_FUNC_MAX_ID
) {
1365 verbose("invalid func %s#%d\n", func_id_name(func_id
), func_id
);
1369 if (env
->prog
->aux
->ops
->get_func_proto
)
1370 fn
= env
->prog
->aux
->ops
->get_func_proto(func_id
);
1373 verbose("unknown func %s#%d\n", func_id_name(func_id
), func_id
);
1377 /* eBPF programs must be GPL compatible to use GPL-ed functions */
1378 if (!env
->prog
->gpl_compatible
&& fn
->gpl_only
) {
1379 verbose("cannot call GPL only function from proprietary program\n");
1383 changes_data
= bpf_helper_changes_pkt_data(fn
->func
);
1385 memset(&meta
, 0, sizeof(meta
));
1386 meta
.pkt_access
= fn
->pkt_access
;
1388 /* We only support one arg being in raw mode at the moment, which
1389 * is sufficient for the helper functions we have right now.
1391 err
= check_raw_mode(fn
);
1393 verbose("kernel subsystem misconfigured func %s#%d\n",
1394 func_id_name(func_id
), func_id
);
1399 err
= check_func_arg(env
, BPF_REG_1
, fn
->arg1_type
, &meta
);
1402 err
= check_func_arg(env
, BPF_REG_2
, fn
->arg2_type
, &meta
);
1405 err
= check_func_arg(env
, BPF_REG_3
, fn
->arg3_type
, &meta
);
1408 err
= check_func_arg(env
, BPF_REG_4
, fn
->arg4_type
, &meta
);
1411 err
= check_func_arg(env
, BPF_REG_5
, fn
->arg5_type
, &meta
);
1415 /* Mark slots with STACK_MISC in case of raw mode, stack offset
1416 * is inferred from register state.
1418 for (i
= 0; i
< meta
.access_size
; i
++) {
1419 err
= check_mem_access(env
, meta
.regno
, i
, BPF_B
, BPF_WRITE
, -1);
1424 /* reset caller saved regs */
1425 for (i
= 0; i
< CALLER_SAVED_REGS
; i
++)
1426 mark_reg_not_init(regs
, caller_saved
[i
]);
1428 /* update return register */
1429 if (fn
->ret_type
== RET_INTEGER
) {
1430 regs
[BPF_REG_0
].type
= UNKNOWN_VALUE
;
1431 } else if (fn
->ret_type
== RET_VOID
) {
1432 regs
[BPF_REG_0
].type
= NOT_INIT
;
1433 } else if (fn
->ret_type
== RET_PTR_TO_MAP_VALUE_OR_NULL
) {
1434 struct bpf_insn_aux_data
*insn_aux
;
1436 regs
[BPF_REG_0
].type
= PTR_TO_MAP_VALUE_OR_NULL
;
1437 regs
[BPF_REG_0
].max_value
= regs
[BPF_REG_0
].min_value
= 0;
1438 /* remember map_ptr, so that check_map_access()
1439 * can check 'value_size' boundary of memory access
1440 * to map element returned from bpf_map_lookup_elem()
1442 if (meta
.map_ptr
== NULL
) {
1443 verbose("kernel subsystem misconfigured verifier\n");
1446 regs
[BPF_REG_0
].map_ptr
= meta
.map_ptr
;
1447 regs
[BPF_REG_0
].id
= ++env
->id_gen
;
1448 insn_aux
= &env
->insn_aux_data
[insn_idx
];
1449 if (!insn_aux
->map_ptr
)
1450 insn_aux
->map_ptr
= meta
.map_ptr
;
1451 else if (insn_aux
->map_ptr
!= meta
.map_ptr
)
1452 insn_aux
->map_ptr
= BPF_MAP_PTR_POISON
;
1454 verbose("unknown return type %d of func %s#%d\n",
1455 fn
->ret_type
, func_id_name(func_id
), func_id
);
1459 err
= check_map_func_compatibility(meta
.map_ptr
, func_id
);
1464 clear_all_pkt_pointers(env
);
1468 static int check_packet_ptr_add(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 struct bpf_reg_state tmp_reg
;
1477 if (BPF_SRC(insn
->code
) == BPF_K
) {
1478 /* pkt_ptr += imm */
1483 verbose("addition of negative constant to packet pointer is not allowed\n");
1486 if (imm
>= MAX_PACKET_OFF
||
1487 imm
+ dst_reg
->off
>= MAX_PACKET_OFF
) {
1488 verbose("constant %d is too large to add to packet pointer\n",
1492 /* a constant was added to pkt_ptr.
1493 * Remember it while keeping the same 'id'
1495 dst_reg
->off
+= imm
;
1499 if (src_reg
->type
== PTR_TO_PACKET
) {
1500 /* R6=pkt(id=0,off=0,r=62) R7=imm22; r7 += r6 */
1501 tmp_reg
= *dst_reg
; /* save r7 state */
1502 *dst_reg
= *src_reg
; /* copy pkt_ptr state r6 into r7 */
1503 src_reg
= &tmp_reg
; /* pretend it's src_reg state */
1504 /* if the checks below reject it, the copy won't matter,
1505 * since we're rejecting the whole program. If all ok,
1506 * then imm22 state will be added to r7
1507 * and r7 will be pkt(id=0,off=22,r=62) while
1508 * r6 will stay as pkt(id=0,off=0,r=62)
1512 if (src_reg
->type
== CONST_IMM
) {
1513 /* pkt_ptr += reg where reg is known constant */
1517 /* disallow pkt_ptr += reg
1518 * if reg is not uknown_value with guaranteed zero upper bits
1519 * otherwise pkt_ptr may overflow and addition will become
1520 * subtraction which is not allowed
1522 if (src_reg
->type
!= UNKNOWN_VALUE
) {
1523 verbose("cannot add '%s' to ptr_to_packet\n",
1524 reg_type_str
[src_reg
->type
]);
1527 if (src_reg
->imm
< 48) {
1528 verbose("cannot add integer value with %lld upper zero bits to ptr_to_packet\n",
1533 had_id
= (dst_reg
->id
!= 0);
1535 /* dst_reg stays as pkt_ptr type and since some positive
1536 * integer value was added to the pointer, increment its 'id'
1538 dst_reg
->id
= ++env
->id_gen
;
1540 /* something was added to pkt_ptr, set range to zero */
1541 dst_reg
->aux_off
+= dst_reg
->off
;
1545 dst_reg
->aux_off_align
= min(dst_reg
->aux_off_align
,
1546 src_reg
->min_align
);
1548 dst_reg
->aux_off_align
= src_reg
->min_align
;
1553 static int evaluate_reg_alu(struct bpf_verifier_env
*env
, struct bpf_insn
*insn
)
1555 struct bpf_reg_state
*regs
= env
->cur_state
.regs
;
1556 struct bpf_reg_state
*dst_reg
= ®s
[insn
->dst_reg
];
1557 u8 opcode
= BPF_OP(insn
->code
);
1560 /* for type == UNKNOWN_VALUE:
1561 * imm > 0 -> number of zero upper bits
1562 * imm == 0 -> don't track which is the same as all bits can be non-zero
1565 if (BPF_SRC(insn
->code
) == BPF_X
) {
1566 struct bpf_reg_state
*src_reg
= ®s
[insn
->src_reg
];
1568 if (src_reg
->type
== UNKNOWN_VALUE
&& src_reg
->imm
> 0 &&
1569 dst_reg
->imm
&& opcode
== BPF_ADD
) {
1571 * where both have zero upper bits. Adding them
1572 * can only result making one more bit non-zero
1573 * in the larger value.
1574 * Ex. 0xffff (imm=48) + 1 (imm=63) = 0x10000 (imm=47)
1575 * 0xffff (imm=48) + 0xffff = 0x1fffe (imm=47)
1577 dst_reg
->imm
= min(dst_reg
->imm
, src_reg
->imm
);
1581 if (src_reg
->type
== CONST_IMM
&& src_reg
->imm
> 0 &&
1582 dst_reg
->imm
&& opcode
== BPF_ADD
) {
1584 * where dreg has zero upper bits and sreg is const.
1585 * Adding them can only result making one more bit
1586 * non-zero in the larger value.
1588 imm_log2
= __ilog2_u64((long long)src_reg
->imm
);
1589 dst_reg
->imm
= min(dst_reg
->imm
, 63 - imm_log2
);
1593 /* all other cases non supported yet, just mark dst_reg */
1598 /* sign extend 32-bit imm into 64-bit to make sure that
1599 * negative values occupy bit 63. Note ilog2() would have
1600 * been incorrect, since sizeof(insn->imm) == 4
1602 imm_log2
= __ilog2_u64((long long)insn
->imm
);
1604 if (dst_reg
->imm
&& opcode
== BPF_LSH
) {
1606 * if reg was a result of 2 byte load, then its imm == 48
1607 * which means that upper 48 bits are zero and shifting this reg
1608 * left by 4 would mean that upper 44 bits are still zero
1610 dst_reg
->imm
-= insn
->imm
;
1611 } else if (dst_reg
->imm
&& opcode
== BPF_MUL
) {
1613 * if multiplying by 14 subtract 4
1614 * This is conservative calculation of upper zero bits.
1615 * It's not trying to special case insn->imm == 1 or 0 cases
1617 dst_reg
->imm
-= imm_log2
+ 1;
1618 } else if (opcode
== BPF_AND
) {
1620 dst_reg
->imm
= 63 - imm_log2
;
1621 } else if (dst_reg
->imm
&& opcode
== BPF_ADD
) {
1623 dst_reg
->imm
= min(dst_reg
->imm
, 63 - imm_log2
);
1625 } else if (opcode
== BPF_RSH
) {
1627 * which means that after right shift, upper bits will be zero
1628 * note that verifier already checked that
1629 * 0 <= imm < 64 for shift insn
1631 dst_reg
->imm
+= insn
->imm
;
1632 if (unlikely(dst_reg
->imm
> 64))
1633 /* some dumb code did:
1636 * and all bits are zero now */
1639 /* all other alu ops, means that we don't know what will
1640 * happen to the value, mark it with unknown number of zero bits
1645 if (dst_reg
->imm
< 0) {
1646 /* all 64 bits of the register can contain non-zero bits
1647 * and such value cannot be added to ptr_to_packet, since it
1648 * may overflow, mark it as unknown to avoid further eval
1655 static int evaluate_reg_imm_alu(struct bpf_verifier_env
*env
,
1656 struct bpf_insn
*insn
)
1658 struct bpf_reg_state
*regs
= env
->cur_state
.regs
;
1659 struct bpf_reg_state
*dst_reg
= ®s
[insn
->dst_reg
];
1660 struct bpf_reg_state
*src_reg
= ®s
[insn
->src_reg
];
1661 u8 opcode
= BPF_OP(insn
->code
);
1662 u64 dst_imm
= dst_reg
->imm
;
1664 /* dst_reg->type == CONST_IMM here. Simulate execution of insns
1665 * containing ALU ops. Don't care about overflow or negative
1666 * values, just add/sub/... them; registers are in u64.
1668 if (opcode
== BPF_ADD
&& BPF_SRC(insn
->code
) == BPF_K
) {
1669 dst_imm
+= insn
->imm
;
1670 } else if (opcode
== BPF_ADD
&& BPF_SRC(insn
->code
) == BPF_X
&&
1671 src_reg
->type
== CONST_IMM
) {
1672 dst_imm
+= src_reg
->imm
;
1673 } else if (opcode
== BPF_SUB
&& BPF_SRC(insn
->code
) == BPF_K
) {
1674 dst_imm
-= insn
->imm
;
1675 } else if (opcode
== BPF_SUB
&& BPF_SRC(insn
->code
) == BPF_X
&&
1676 src_reg
->type
== CONST_IMM
) {
1677 dst_imm
-= src_reg
->imm
;
1678 } else if (opcode
== BPF_MUL
&& BPF_SRC(insn
->code
) == BPF_K
) {
1679 dst_imm
*= insn
->imm
;
1680 } else if (opcode
== BPF_MUL
&& BPF_SRC(insn
->code
) == BPF_X
&&
1681 src_reg
->type
== CONST_IMM
) {
1682 dst_imm
*= src_reg
->imm
;
1683 } else if (opcode
== BPF_OR
&& BPF_SRC(insn
->code
) == BPF_K
) {
1684 dst_imm
|= insn
->imm
;
1685 } else if (opcode
== BPF_OR
&& BPF_SRC(insn
->code
) == BPF_X
&&
1686 src_reg
->type
== CONST_IMM
) {
1687 dst_imm
|= src_reg
->imm
;
1688 } else if (opcode
== BPF_AND
&& BPF_SRC(insn
->code
) == BPF_K
) {
1689 dst_imm
&= insn
->imm
;
1690 } else if (opcode
== BPF_AND
&& BPF_SRC(insn
->code
) == BPF_X
&&
1691 src_reg
->type
== CONST_IMM
) {
1692 dst_imm
&= src_reg
->imm
;
1693 } else if (opcode
== BPF_RSH
&& BPF_SRC(insn
->code
) == BPF_K
) {
1694 dst_imm
>>= insn
->imm
;
1695 } else if (opcode
== BPF_RSH
&& BPF_SRC(insn
->code
) == BPF_X
&&
1696 src_reg
->type
== CONST_IMM
) {
1697 dst_imm
>>= src_reg
->imm
;
1698 } else if (opcode
== BPF_LSH
&& BPF_SRC(insn
->code
) == BPF_K
) {
1699 dst_imm
<<= insn
->imm
;
1700 } else if (opcode
== BPF_LSH
&& BPF_SRC(insn
->code
) == BPF_X
&&
1701 src_reg
->type
== CONST_IMM
) {
1702 dst_imm
<<= src_reg
->imm
;
1704 mark_reg_unknown_value(regs
, insn
->dst_reg
);
1708 dst_reg
->imm
= dst_imm
;
1713 static void check_reg_overflow(struct bpf_reg_state
*reg
)
1715 if (reg
->max_value
> BPF_REGISTER_MAX_RANGE
)
1716 reg
->max_value
= BPF_REGISTER_MAX_RANGE
;
1717 if (reg
->min_value
< BPF_REGISTER_MIN_RANGE
||
1718 reg
->min_value
> BPF_REGISTER_MAX_RANGE
)
1719 reg
->min_value
= BPF_REGISTER_MIN_RANGE
;
1722 static u32
calc_align(u32 imm
)
1726 return imm
- ((imm
- 1) & imm
);
1729 static void adjust_reg_min_max_vals(struct bpf_verifier_env
*env
,
1730 struct bpf_insn
*insn
)
1732 struct bpf_reg_state
*regs
= env
->cur_state
.regs
, *dst_reg
;
1733 s64 min_val
= BPF_REGISTER_MIN_RANGE
;
1734 u64 max_val
= BPF_REGISTER_MAX_RANGE
;
1735 u8 opcode
= BPF_OP(insn
->code
);
1736 u32 dst_align
, src_align
;
1738 dst_reg
= ®s
[insn
->dst_reg
];
1740 if (BPF_SRC(insn
->code
) == BPF_X
) {
1741 check_reg_overflow(®s
[insn
->src_reg
]);
1742 min_val
= regs
[insn
->src_reg
].min_value
;
1743 max_val
= regs
[insn
->src_reg
].max_value
;
1745 /* If the source register is a random pointer then the
1746 * min_value/max_value values represent the range of the known
1747 * accesses into that value, not the actual min/max value of the
1748 * register itself. In this case we have to reset the reg range
1749 * values so we know it is not safe to look at.
1751 if (regs
[insn
->src_reg
].type
!= CONST_IMM
&&
1752 regs
[insn
->src_reg
].type
!= UNKNOWN_VALUE
) {
1753 min_val
= BPF_REGISTER_MIN_RANGE
;
1754 max_val
= BPF_REGISTER_MAX_RANGE
;
1757 src_align
= regs
[insn
->src_reg
].min_align
;
1759 } else if (insn
->imm
< BPF_REGISTER_MAX_RANGE
&&
1760 (s64
)insn
->imm
> BPF_REGISTER_MIN_RANGE
) {
1761 min_val
= max_val
= insn
->imm
;
1762 src_align
= calc_align(insn
->imm
);
1765 dst_align
= dst_reg
->min_align
;
1767 /* We don't know anything about what was done to this register, mark it
1770 if (min_val
== BPF_REGISTER_MIN_RANGE
&&
1771 max_val
== BPF_REGISTER_MAX_RANGE
) {
1772 reset_reg_range_values(regs
, insn
->dst_reg
);
1776 /* If one of our values was at the end of our ranges then we can't just
1777 * do our normal operations to the register, we need to set the values
1778 * to the min/max since they are undefined.
1780 if (min_val
== BPF_REGISTER_MIN_RANGE
)
1781 dst_reg
->min_value
= BPF_REGISTER_MIN_RANGE
;
1782 if (max_val
== BPF_REGISTER_MAX_RANGE
)
1783 dst_reg
->max_value
= BPF_REGISTER_MAX_RANGE
;
1787 if (dst_reg
->min_value
!= BPF_REGISTER_MIN_RANGE
)
1788 dst_reg
->min_value
+= min_val
;
1789 if (dst_reg
->max_value
!= BPF_REGISTER_MAX_RANGE
)
1790 dst_reg
->max_value
+= max_val
;
1791 dst_reg
->min_align
= min(src_align
, dst_align
);
1794 if (dst_reg
->min_value
!= BPF_REGISTER_MIN_RANGE
)
1795 dst_reg
->min_value
-= min_val
;
1796 if (dst_reg
->max_value
!= BPF_REGISTER_MAX_RANGE
)
1797 dst_reg
->max_value
-= max_val
;
1798 dst_reg
->min_align
= min(src_align
, dst_align
);
1801 if (dst_reg
->min_value
!= BPF_REGISTER_MIN_RANGE
)
1802 dst_reg
->min_value
*= min_val
;
1803 if (dst_reg
->max_value
!= BPF_REGISTER_MAX_RANGE
)
1804 dst_reg
->max_value
*= max_val
;
1805 dst_reg
->min_align
= max(src_align
, dst_align
);
1808 /* Disallow AND'ing of negative numbers, ain't nobody got time
1809 * for that. Otherwise the minimum is 0 and the max is the max
1810 * value we could AND against.
1813 dst_reg
->min_value
= BPF_REGISTER_MIN_RANGE
;
1815 dst_reg
->min_value
= 0;
1816 dst_reg
->max_value
= max_val
;
1817 dst_reg
->min_align
= max(src_align
, dst_align
);
1820 /* Gotta have special overflow logic here, if we're shifting
1821 * more than MAX_RANGE then just assume we have an invalid
1824 if (min_val
> ilog2(BPF_REGISTER_MAX_RANGE
)) {
1825 dst_reg
->min_value
= BPF_REGISTER_MIN_RANGE
;
1826 dst_reg
->min_align
= 1;
1828 if (dst_reg
->min_value
!= BPF_REGISTER_MIN_RANGE
)
1829 dst_reg
->min_value
<<= min_val
;
1830 if (!dst_reg
->min_align
)
1831 dst_reg
->min_align
= 1;
1832 dst_reg
->min_align
<<= min_val
;
1834 if (max_val
> ilog2(BPF_REGISTER_MAX_RANGE
))
1835 dst_reg
->max_value
= BPF_REGISTER_MAX_RANGE
;
1836 else if (dst_reg
->max_value
!= BPF_REGISTER_MAX_RANGE
)
1837 dst_reg
->max_value
<<= max_val
;
1840 /* RSH by a negative number is undefined, and the BPF_RSH is an
1841 * unsigned shift, so make the appropriate casts.
1843 if (min_val
< 0 || dst_reg
->min_value
< 0) {
1844 dst_reg
->min_value
= BPF_REGISTER_MIN_RANGE
;
1846 dst_reg
->min_value
=
1847 (u64
)(dst_reg
->min_value
) >> min_val
;
1850 dst_reg
->min_align
= 1;
1852 dst_reg
->min_align
>>= (u64
) min_val
;
1853 if (!dst_reg
->min_align
)
1854 dst_reg
->min_align
= 1;
1856 if (dst_reg
->max_value
!= BPF_REGISTER_MAX_RANGE
)
1857 dst_reg
->max_value
>>= max_val
;
1860 reset_reg_range_values(regs
, insn
->dst_reg
);
1864 check_reg_overflow(dst_reg
);
1867 /* check validity of 32-bit and 64-bit arithmetic operations */
1868 static int check_alu_op(struct bpf_verifier_env
*env
, struct bpf_insn
*insn
)
1870 struct bpf_reg_state
*regs
= env
->cur_state
.regs
, *dst_reg
;
1871 u8 opcode
= BPF_OP(insn
->code
);
1874 if (opcode
== BPF_END
|| opcode
== BPF_NEG
) {
1875 if (opcode
== BPF_NEG
) {
1876 if (BPF_SRC(insn
->code
) != 0 ||
1877 insn
->src_reg
!= BPF_REG_0
||
1878 insn
->off
!= 0 || insn
->imm
!= 0) {
1879 verbose("BPF_NEG uses reserved fields\n");
1883 if (insn
->src_reg
!= BPF_REG_0
|| insn
->off
!= 0 ||
1884 (insn
->imm
!= 16 && insn
->imm
!= 32 && insn
->imm
!= 64)) {
1885 verbose("BPF_END uses reserved fields\n");
1890 /* check src operand */
1891 err
= check_reg_arg(regs
, insn
->dst_reg
, SRC_OP
);
1895 if (is_pointer_value(env
, insn
->dst_reg
)) {
1896 verbose("R%d pointer arithmetic prohibited\n",
1901 /* check dest operand */
1902 err
= check_reg_arg(regs
, insn
->dst_reg
, DST_OP
);
1906 } else if (opcode
== BPF_MOV
) {
1908 if (BPF_SRC(insn
->code
) == BPF_X
) {
1909 if (insn
->imm
!= 0 || insn
->off
!= 0) {
1910 verbose("BPF_MOV uses reserved fields\n");
1914 /* check src operand */
1915 err
= check_reg_arg(regs
, insn
->src_reg
, SRC_OP
);
1919 if (insn
->src_reg
!= BPF_REG_0
|| insn
->off
!= 0) {
1920 verbose("BPF_MOV uses reserved fields\n");
1925 /* check dest operand */
1926 err
= check_reg_arg(regs
, insn
->dst_reg
, DST_OP
);
1930 /* we are setting our register to something new, we need to
1931 * reset its range values.
1933 reset_reg_range_values(regs
, insn
->dst_reg
);
1935 if (BPF_SRC(insn
->code
) == BPF_X
) {
1936 if (BPF_CLASS(insn
->code
) == BPF_ALU64
) {
1938 * copy register state to dest reg
1940 regs
[insn
->dst_reg
] = regs
[insn
->src_reg
];
1942 if (is_pointer_value(env
, insn
->src_reg
)) {
1943 verbose("R%d partial copy of pointer\n",
1947 mark_reg_unknown_value(regs
, insn
->dst_reg
);
1951 * remember the value we stored into this reg
1953 regs
[insn
->dst_reg
].type
= CONST_IMM
;
1954 regs
[insn
->dst_reg
].imm
= insn
->imm
;
1955 regs
[insn
->dst_reg
].max_value
= insn
->imm
;
1956 regs
[insn
->dst_reg
].min_value
= insn
->imm
;
1957 regs
[insn
->dst_reg
].min_align
= calc_align(insn
->imm
);
1960 } else if (opcode
> BPF_END
) {
1961 verbose("invalid BPF_ALU opcode %x\n", opcode
);
1964 } else { /* all other ALU ops: and, sub, xor, add, ... */
1966 if (BPF_SRC(insn
->code
) == BPF_X
) {
1967 if (insn
->imm
!= 0 || insn
->off
!= 0) {
1968 verbose("BPF_ALU uses reserved fields\n");
1971 /* check src1 operand */
1972 err
= check_reg_arg(regs
, insn
->src_reg
, SRC_OP
);
1976 if (insn
->src_reg
!= BPF_REG_0
|| insn
->off
!= 0) {
1977 verbose("BPF_ALU uses reserved fields\n");
1982 /* check src2 operand */
1983 err
= check_reg_arg(regs
, insn
->dst_reg
, SRC_OP
);
1987 if ((opcode
== BPF_MOD
|| opcode
== BPF_DIV
) &&
1988 BPF_SRC(insn
->code
) == BPF_K
&& insn
->imm
== 0) {
1989 verbose("div by zero\n");
1993 if ((opcode
== BPF_LSH
|| opcode
== BPF_RSH
||
1994 opcode
== BPF_ARSH
) && BPF_SRC(insn
->code
) == BPF_K
) {
1995 int size
= BPF_CLASS(insn
->code
) == BPF_ALU64
? 64 : 32;
1997 if (insn
->imm
< 0 || insn
->imm
>= size
) {
1998 verbose("invalid shift %d\n", insn
->imm
);
2003 /* check dest operand */
2004 err
= check_reg_arg(regs
, insn
->dst_reg
, DST_OP_NO_MARK
);
2008 dst_reg
= ®s
[insn
->dst_reg
];
2010 /* first we want to adjust our ranges. */
2011 adjust_reg_min_max_vals(env
, insn
);
2013 /* pattern match 'bpf_add Rx, imm' instruction */
2014 if (opcode
== BPF_ADD
&& BPF_CLASS(insn
->code
) == BPF_ALU64
&&
2015 dst_reg
->type
== FRAME_PTR
&& BPF_SRC(insn
->code
) == BPF_K
) {
2016 dst_reg
->type
= PTR_TO_STACK
;
2017 dst_reg
->imm
= insn
->imm
;
2019 } else if (opcode
== BPF_ADD
&&
2020 BPF_CLASS(insn
->code
) == BPF_ALU64
&&
2021 dst_reg
->type
== PTR_TO_STACK
&&
2022 ((BPF_SRC(insn
->code
) == BPF_X
&&
2023 regs
[insn
->src_reg
].type
== CONST_IMM
) ||
2024 BPF_SRC(insn
->code
) == BPF_K
)) {
2025 if (BPF_SRC(insn
->code
) == BPF_X
)
2026 dst_reg
->imm
+= regs
[insn
->src_reg
].imm
;
2028 dst_reg
->imm
+= insn
->imm
;
2030 } else if (opcode
== BPF_ADD
&&
2031 BPF_CLASS(insn
->code
) == BPF_ALU64
&&
2032 (dst_reg
->type
== PTR_TO_PACKET
||
2033 (BPF_SRC(insn
->code
) == BPF_X
&&
2034 regs
[insn
->src_reg
].type
== PTR_TO_PACKET
))) {
2035 /* ptr_to_packet += K|X */
2036 return check_packet_ptr_add(env
, insn
);
2037 } else if (BPF_CLASS(insn
->code
) == BPF_ALU64
&&
2038 dst_reg
->type
== UNKNOWN_VALUE
&&
2039 env
->allow_ptr_leaks
) {
2040 /* unknown += K|X */
2041 return evaluate_reg_alu(env
, insn
);
2042 } else if (BPF_CLASS(insn
->code
) == BPF_ALU64
&&
2043 dst_reg
->type
== CONST_IMM
&&
2044 env
->allow_ptr_leaks
) {
2045 /* reg_imm += K|X */
2046 return evaluate_reg_imm_alu(env
, insn
);
2047 } else if (is_pointer_value(env
, insn
->dst_reg
)) {
2048 verbose("R%d pointer arithmetic prohibited\n",
2051 } else if (BPF_SRC(insn
->code
) == BPF_X
&&
2052 is_pointer_value(env
, insn
->src_reg
)) {
2053 verbose("R%d pointer arithmetic prohibited\n",
2058 /* If we did pointer math on a map value then just set it to our
2059 * PTR_TO_MAP_VALUE_ADJ type so we can deal with any stores or
2060 * loads to this register appropriately, otherwise just mark the
2061 * register as unknown.
2063 if (env
->allow_ptr_leaks
&&
2064 BPF_CLASS(insn
->code
) == BPF_ALU64
&& opcode
== BPF_ADD
&&
2065 (dst_reg
->type
== PTR_TO_MAP_VALUE
||
2066 dst_reg
->type
== PTR_TO_MAP_VALUE_ADJ
))
2067 dst_reg
->type
= PTR_TO_MAP_VALUE_ADJ
;
2069 mark_reg_unknown_value(regs
, insn
->dst_reg
);
2075 static void find_good_pkt_pointers(struct bpf_verifier_state
*state
,
2076 struct bpf_reg_state
*dst_reg
)
2078 struct bpf_reg_state
*regs
= state
->regs
, *reg
;
2081 /* LLVM can generate two kind of checks:
2087 * if (r2 > pkt_end) goto <handle exception>
2091 * r2 == dst_reg, pkt_end == src_reg
2092 * r2=pkt(id=n,off=8,r=0)
2093 * r3=pkt(id=n,off=0,r=0)
2099 * if (pkt_end >= r2) goto <access okay>
2100 * <handle exception>
2103 * pkt_end == dst_reg, r2 == src_reg
2104 * r2=pkt(id=n,off=8,r=0)
2105 * r3=pkt(id=n,off=0,r=0)
2107 * Find register r3 and mark its range as r3=pkt(id=n,off=0,r=8)
2108 * so that range of bytes [r3, r3 + 8) is safe to access.
2111 for (i
= 0; i
< MAX_BPF_REG
; i
++)
2112 if (regs
[i
].type
== PTR_TO_PACKET
&& regs
[i
].id
== dst_reg
->id
)
2113 /* keep the maximum range already checked */
2114 regs
[i
].range
= max(regs
[i
].range
, dst_reg
->off
);
2116 for (i
= 0; i
< MAX_BPF_STACK
; i
+= BPF_REG_SIZE
) {
2117 if (state
->stack_slot_type
[i
] != STACK_SPILL
)
2119 reg
= &state
->spilled_regs
[i
/ BPF_REG_SIZE
];
2120 if (reg
->type
== PTR_TO_PACKET
&& reg
->id
== dst_reg
->id
)
2121 reg
->range
= max(reg
->range
, dst_reg
->off
);
2125 /* Adjusts the register min/max values in the case that the dst_reg is the
2126 * variable register that we are working on, and src_reg is a constant or we're
2127 * simply doing a BPF_K check.
2129 static void reg_set_min_max(struct bpf_reg_state
*true_reg
,
2130 struct bpf_reg_state
*false_reg
, u64 val
,
2135 /* If this is false then we know nothing Jon Snow, but if it is
2136 * true then we know for sure.
2138 true_reg
->max_value
= true_reg
->min_value
= val
;
2141 /* If this is true we know nothing Jon Snow, but if it is false
2142 * we know the value for sure;
2144 false_reg
->max_value
= false_reg
->min_value
= val
;
2147 /* Unsigned comparison, the minimum value is 0. */
2148 false_reg
->min_value
= 0;
2151 /* If this is false then we know the maximum val is val,
2152 * otherwise we know the min val is val+1.
2154 false_reg
->max_value
= val
;
2155 true_reg
->min_value
= val
+ 1;
2158 /* Unsigned comparison, the minimum value is 0. */
2159 false_reg
->min_value
= 0;
2162 /* If this is false then we know the maximum value is val - 1,
2163 * otherwise we know the mimimum value is val.
2165 false_reg
->max_value
= val
- 1;
2166 true_reg
->min_value
= val
;
2172 check_reg_overflow(false_reg
);
2173 check_reg_overflow(true_reg
);
2176 /* Same as above, but for the case that dst_reg is a CONST_IMM reg and src_reg
2177 * is the variable reg.
2179 static void reg_set_min_max_inv(struct bpf_reg_state
*true_reg
,
2180 struct bpf_reg_state
*false_reg
, u64 val
,
2185 /* If this is false then we know nothing Jon Snow, but if it is
2186 * true then we know for sure.
2188 true_reg
->max_value
= true_reg
->min_value
= val
;
2191 /* If this is true we know nothing Jon Snow, but if it is false
2192 * we know the value for sure;
2194 false_reg
->max_value
= false_reg
->min_value
= val
;
2197 /* Unsigned comparison, the minimum value is 0. */
2198 true_reg
->min_value
= 0;
2202 * If this is false, then the val is <= the register, if it is
2203 * true the register <= to the val.
2205 false_reg
->min_value
= val
;
2206 true_reg
->max_value
= val
- 1;
2209 /* Unsigned comparison, the minimum value is 0. */
2210 true_reg
->min_value
= 0;
2213 /* If this is false then constant < register, if it is true then
2214 * the register < constant.
2216 false_reg
->min_value
= val
+ 1;
2217 true_reg
->max_value
= val
;
2223 check_reg_overflow(false_reg
);
2224 check_reg_overflow(true_reg
);
2227 static void mark_map_reg(struct bpf_reg_state
*regs
, u32 regno
, u32 id
,
2228 enum bpf_reg_type type
)
2230 struct bpf_reg_state
*reg
= ®s
[regno
];
2232 if (reg
->type
== PTR_TO_MAP_VALUE_OR_NULL
&& reg
->id
== id
) {
2233 if (type
== UNKNOWN_VALUE
) {
2234 __mark_reg_unknown_value(regs
, regno
);
2235 } else if (reg
->map_ptr
->inner_map_meta
) {
2236 reg
->type
= CONST_PTR_TO_MAP
;
2237 reg
->map_ptr
= reg
->map_ptr
->inner_map_meta
;
2241 /* We don't need id from this point onwards anymore, thus we
2242 * should better reset it, so that state pruning has chances
2249 /* The logic is similar to find_good_pkt_pointers(), both could eventually
2250 * be folded together at some point.
2252 static void mark_map_regs(struct bpf_verifier_state
*state
, u32 regno
,
2253 enum bpf_reg_type type
)
2255 struct bpf_reg_state
*regs
= state
->regs
;
2256 u32 id
= regs
[regno
].id
;
2259 for (i
= 0; i
< MAX_BPF_REG
; i
++)
2260 mark_map_reg(regs
, i
, id
, type
);
2262 for (i
= 0; i
< MAX_BPF_STACK
; i
+= BPF_REG_SIZE
) {
2263 if (state
->stack_slot_type
[i
] != STACK_SPILL
)
2265 mark_map_reg(state
->spilled_regs
, i
/ BPF_REG_SIZE
, id
, type
);
2269 static int check_cond_jmp_op(struct bpf_verifier_env
*env
,
2270 struct bpf_insn
*insn
, int *insn_idx
)
2272 struct bpf_verifier_state
*other_branch
, *this_branch
= &env
->cur_state
;
2273 struct bpf_reg_state
*regs
= this_branch
->regs
, *dst_reg
;
2274 u8 opcode
= BPF_OP(insn
->code
);
2277 if (opcode
> BPF_EXIT
) {
2278 verbose("invalid BPF_JMP opcode %x\n", opcode
);
2282 if (BPF_SRC(insn
->code
) == BPF_X
) {
2283 if (insn
->imm
!= 0) {
2284 verbose("BPF_JMP uses reserved fields\n");
2288 /* check src1 operand */
2289 err
= check_reg_arg(regs
, insn
->src_reg
, SRC_OP
);
2293 if (is_pointer_value(env
, insn
->src_reg
)) {
2294 verbose("R%d pointer comparison prohibited\n",
2299 if (insn
->src_reg
!= BPF_REG_0
) {
2300 verbose("BPF_JMP uses reserved fields\n");
2305 /* check src2 operand */
2306 err
= check_reg_arg(regs
, insn
->dst_reg
, SRC_OP
);
2310 dst_reg
= ®s
[insn
->dst_reg
];
2312 /* detect if R == 0 where R was initialized to zero earlier */
2313 if (BPF_SRC(insn
->code
) == BPF_K
&&
2314 (opcode
== BPF_JEQ
|| opcode
== BPF_JNE
) &&
2315 dst_reg
->type
== CONST_IMM
&& dst_reg
->imm
== insn
->imm
) {
2316 if (opcode
== BPF_JEQ
) {
2317 /* if (imm == imm) goto pc+off;
2318 * only follow the goto, ignore fall-through
2320 *insn_idx
+= insn
->off
;
2323 /* if (imm != imm) goto pc+off;
2324 * only follow fall-through branch, since
2325 * that's where the program will go
2331 other_branch
= push_stack(env
, *insn_idx
+ insn
->off
+ 1, *insn_idx
);
2335 /* detect if we are comparing against a constant value so we can adjust
2336 * our min/max values for our dst register.
2338 if (BPF_SRC(insn
->code
) == BPF_X
) {
2339 if (regs
[insn
->src_reg
].type
== CONST_IMM
)
2340 reg_set_min_max(&other_branch
->regs
[insn
->dst_reg
],
2341 dst_reg
, regs
[insn
->src_reg
].imm
,
2343 else if (dst_reg
->type
== CONST_IMM
)
2344 reg_set_min_max_inv(&other_branch
->regs
[insn
->src_reg
],
2345 ®s
[insn
->src_reg
], dst_reg
->imm
,
2348 reg_set_min_max(&other_branch
->regs
[insn
->dst_reg
],
2349 dst_reg
, insn
->imm
, opcode
);
2352 /* detect if R == 0 where R is returned from bpf_map_lookup_elem() */
2353 if (BPF_SRC(insn
->code
) == BPF_K
&&
2354 insn
->imm
== 0 && (opcode
== BPF_JEQ
|| opcode
== BPF_JNE
) &&
2355 dst_reg
->type
== PTR_TO_MAP_VALUE_OR_NULL
) {
2356 /* Mark all identical map registers in each branch as either
2357 * safe or unknown depending R == 0 or R != 0 conditional.
2359 mark_map_regs(this_branch
, insn
->dst_reg
,
2360 opcode
== BPF_JEQ
? PTR_TO_MAP_VALUE
: UNKNOWN_VALUE
);
2361 mark_map_regs(other_branch
, insn
->dst_reg
,
2362 opcode
== BPF_JEQ
? UNKNOWN_VALUE
: PTR_TO_MAP_VALUE
);
2363 } else if (BPF_SRC(insn
->code
) == BPF_X
&& opcode
== BPF_JGT
&&
2364 dst_reg
->type
== PTR_TO_PACKET
&&
2365 regs
[insn
->src_reg
].type
== PTR_TO_PACKET_END
) {
2366 find_good_pkt_pointers(this_branch
, dst_reg
);
2367 } else if (BPF_SRC(insn
->code
) == BPF_X
&& opcode
== BPF_JGE
&&
2368 dst_reg
->type
== PTR_TO_PACKET_END
&&
2369 regs
[insn
->src_reg
].type
== PTR_TO_PACKET
) {
2370 find_good_pkt_pointers(other_branch
, ®s
[insn
->src_reg
]);
2371 } else if (is_pointer_value(env
, insn
->dst_reg
)) {
2372 verbose("R%d pointer comparison prohibited\n", insn
->dst_reg
);
2376 print_verifier_state(this_branch
);
2380 /* return the map pointer stored inside BPF_LD_IMM64 instruction */
2381 static struct bpf_map
*ld_imm64_to_map_ptr(struct bpf_insn
*insn
)
2383 u64 imm64
= ((u64
) (u32
) insn
[0].imm
) | ((u64
) (u32
) insn
[1].imm
) << 32;
2385 return (struct bpf_map
*) (unsigned long) imm64
;
2388 /* verify BPF_LD_IMM64 instruction */
2389 static int check_ld_imm(struct bpf_verifier_env
*env
, struct bpf_insn
*insn
)
2391 struct bpf_reg_state
*regs
= env
->cur_state
.regs
;
2394 if (BPF_SIZE(insn
->code
) != BPF_DW
) {
2395 verbose("invalid BPF_LD_IMM insn\n");
2398 if (insn
->off
!= 0) {
2399 verbose("BPF_LD_IMM64 uses reserved fields\n");
2403 err
= check_reg_arg(regs
, insn
->dst_reg
, DST_OP
);
2407 if (insn
->src_reg
== 0) {
2408 u64 imm
= ((u64
)(insn
+ 1)->imm
<< 32) | (u32
)insn
->imm
;
2410 regs
[insn
->dst_reg
].type
= CONST_IMM
;
2411 regs
[insn
->dst_reg
].imm
= imm
;
2415 /* replace_map_fd_with_map_ptr() should have caught bad ld_imm64 */
2416 BUG_ON(insn
->src_reg
!= BPF_PSEUDO_MAP_FD
);
2418 regs
[insn
->dst_reg
].type
= CONST_PTR_TO_MAP
;
2419 regs
[insn
->dst_reg
].map_ptr
= ld_imm64_to_map_ptr(insn
);
2423 static bool may_access_skb(enum bpf_prog_type type
)
2426 case BPF_PROG_TYPE_SOCKET_FILTER
:
2427 case BPF_PROG_TYPE_SCHED_CLS
:
2428 case BPF_PROG_TYPE_SCHED_ACT
:
2435 /* verify safety of LD_ABS|LD_IND instructions:
2436 * - they can only appear in the programs where ctx == skb
2437 * - since they are wrappers of function calls, they scratch R1-R5 registers,
2438 * preserve R6-R9, and store return value into R0
2441 * ctx == skb == R6 == CTX
2444 * SRC == any register
2445 * IMM == 32-bit immediate
2448 * R0 - 8/16/32-bit skb data converted to cpu endianness
2450 static int check_ld_abs(struct bpf_verifier_env
*env
, struct bpf_insn
*insn
)
2452 struct bpf_reg_state
*regs
= env
->cur_state
.regs
;
2453 u8 mode
= BPF_MODE(insn
->code
);
2456 if (!may_access_skb(env
->prog
->type
)) {
2457 verbose("BPF_LD_[ABS|IND] instructions not allowed for this program type\n");
2461 if (insn
->dst_reg
!= BPF_REG_0
|| insn
->off
!= 0 ||
2462 BPF_SIZE(insn
->code
) == BPF_DW
||
2463 (mode
== BPF_ABS
&& insn
->src_reg
!= BPF_REG_0
)) {
2464 verbose("BPF_LD_[ABS|IND] uses reserved fields\n");
2468 /* check whether implicit source operand (register R6) is readable */
2469 err
= check_reg_arg(regs
, BPF_REG_6
, SRC_OP
);
2473 if (regs
[BPF_REG_6
].type
!= PTR_TO_CTX
) {
2474 verbose("at the time of BPF_LD_ABS|IND R6 != pointer to skb\n");
2478 if (mode
== BPF_IND
) {
2479 /* check explicit source operand */
2480 err
= check_reg_arg(regs
, insn
->src_reg
, SRC_OP
);
2485 /* reset caller saved regs to unreadable */
2486 for (i
= 0; i
< CALLER_SAVED_REGS
; i
++)
2487 mark_reg_not_init(regs
, caller_saved
[i
]);
2489 /* mark destination R0 register as readable, since it contains
2490 * the value fetched from the packet
2492 regs
[BPF_REG_0
].type
= UNKNOWN_VALUE
;
2496 /* non-recursive DFS pseudo code
2497 * 1 procedure DFS-iterative(G,v):
2498 * 2 label v as discovered
2499 * 3 let S be a stack
2501 * 5 while S is not empty
2503 * 7 if t is what we're looking for:
2505 * 9 for all edges e in G.adjacentEdges(t) do
2506 * 10 if edge e is already labelled
2507 * 11 continue with the next edge
2508 * 12 w <- G.adjacentVertex(t,e)
2509 * 13 if vertex w is not discovered and not explored
2510 * 14 label e as tree-edge
2511 * 15 label w as discovered
2514 * 18 else if vertex w is discovered
2515 * 19 label e as back-edge
2517 * 21 // vertex w is explored
2518 * 22 label e as forward- or cross-edge
2519 * 23 label t as explored
2524 * 0x11 - discovered and fall-through edge labelled
2525 * 0x12 - discovered and fall-through and branch edges labelled
2536 #define STATE_LIST_MARK ((struct bpf_verifier_state_list *) -1L)
2538 static int *insn_stack
; /* stack of insns to process */
2539 static int cur_stack
; /* current stack index */
2540 static int *insn_state
;
2542 /* t, w, e - match pseudo-code above:
2543 * t - index of current instruction
2544 * w - next instruction
2547 static int push_insn(int t
, int w
, int e
, struct bpf_verifier_env
*env
)
2549 if (e
== FALLTHROUGH
&& insn_state
[t
] >= (DISCOVERED
| FALLTHROUGH
))
2552 if (e
== BRANCH
&& insn_state
[t
] >= (DISCOVERED
| BRANCH
))
2555 if (w
< 0 || w
>= env
->prog
->len
) {
2556 verbose("jump out of range from insn %d to %d\n", t
, w
);
2561 /* mark branch target for state pruning */
2562 env
->explored_states
[w
] = STATE_LIST_MARK
;
2564 if (insn_state
[w
] == 0) {
2566 insn_state
[t
] = DISCOVERED
| e
;
2567 insn_state
[w
] = DISCOVERED
;
2568 if (cur_stack
>= env
->prog
->len
)
2570 insn_stack
[cur_stack
++] = w
;
2572 } else if ((insn_state
[w
] & 0xF0) == DISCOVERED
) {
2573 verbose("back-edge from insn %d to %d\n", t
, w
);
2575 } else if (insn_state
[w
] == EXPLORED
) {
2576 /* forward- or cross-edge */
2577 insn_state
[t
] = DISCOVERED
| e
;
2579 verbose("insn state internal bug\n");
2585 /* non-recursive depth-first-search to detect loops in BPF program
2586 * loop == back-edge in directed graph
2588 static int check_cfg(struct bpf_verifier_env
*env
)
2590 struct bpf_insn
*insns
= env
->prog
->insnsi
;
2591 int insn_cnt
= env
->prog
->len
;
2595 insn_state
= kcalloc(insn_cnt
, sizeof(int), GFP_KERNEL
);
2599 insn_stack
= kcalloc(insn_cnt
, sizeof(int), GFP_KERNEL
);
2605 insn_state
[0] = DISCOVERED
; /* mark 1st insn as discovered */
2606 insn_stack
[0] = 0; /* 0 is the first instruction */
2612 t
= insn_stack
[cur_stack
- 1];
2614 if (BPF_CLASS(insns
[t
].code
) == BPF_JMP
) {
2615 u8 opcode
= BPF_OP(insns
[t
].code
);
2617 if (opcode
== BPF_EXIT
) {
2619 } else if (opcode
== BPF_CALL
) {
2620 ret
= push_insn(t
, t
+ 1, FALLTHROUGH
, env
);
2625 if (t
+ 1 < insn_cnt
)
2626 env
->explored_states
[t
+ 1] = STATE_LIST_MARK
;
2627 } else if (opcode
== BPF_JA
) {
2628 if (BPF_SRC(insns
[t
].code
) != BPF_K
) {
2632 /* unconditional jump with single edge */
2633 ret
= push_insn(t
, t
+ insns
[t
].off
+ 1,
2639 /* tell verifier to check for equivalent states
2640 * after every call and jump
2642 if (t
+ 1 < insn_cnt
)
2643 env
->explored_states
[t
+ 1] = STATE_LIST_MARK
;
2645 /* conditional jump with two edges */
2646 env
->explored_states
[t
] = STATE_LIST_MARK
;
2647 ret
= push_insn(t
, t
+ 1, FALLTHROUGH
, env
);
2653 ret
= push_insn(t
, t
+ insns
[t
].off
+ 1, BRANCH
, env
);
2660 /* all other non-branch instructions with single
2663 ret
= push_insn(t
, t
+ 1, FALLTHROUGH
, env
);
2671 insn_state
[t
] = EXPLORED
;
2672 if (cur_stack
-- <= 0) {
2673 verbose("pop stack internal bug\n");
2680 for (i
= 0; i
< insn_cnt
; i
++) {
2681 if (insn_state
[i
] != EXPLORED
) {
2682 verbose("unreachable insn %d\n", i
);
2687 ret
= 0; /* cfg looks good */
2695 /* the following conditions reduce the number of explored insns
2696 * from ~140k to ~80k for ultra large programs that use a lot of ptr_to_packet
2698 static bool compare_ptrs_to_packet(struct bpf_verifier_env
*env
,
2699 struct bpf_reg_state
*old
,
2700 struct bpf_reg_state
*cur
)
2702 if (old
->id
!= cur
->id
)
2705 /* old ptr_to_packet is more conservative, since it allows smaller
2707 * old(off=0,r=10) is equal to cur(off=0,r=20), because
2708 * old(off=0,r=10) means that with range=10 the verifier proceeded
2709 * further and found no issues with the program. Now we're in the same
2710 * spot with cur(off=0,r=20), so we're safe too, since anything further
2711 * will only be looking at most 10 bytes after this pointer.
2713 if (old
->off
== cur
->off
&& old
->range
< cur
->range
)
2716 /* old(off=20,r=10) is equal to cur(off=22,re=22 or 5 or 0)
2717 * since both cannot be used for packet access and safe(old)
2718 * pointer has smaller off that could be used for further
2719 * 'if (ptr > data_end)' check
2721 * old(off=20,r=10) and cur(off=22,r=22) and cur(off=22,r=0) mean
2722 * that we cannot access the packet.
2723 * The safe range is:
2724 * [ptr, ptr + range - off)
2725 * so whenever off >=range, it means no safe bytes from this pointer.
2726 * When comparing old->off <= cur->off, it means that older code
2727 * went with smaller offset and that offset was later
2728 * used to figure out the safe range after 'if (ptr > data_end)' check
2729 * Say, 'old' state was explored like:
2730 * ... R3(off=0, r=0)
2732 * ... now R4(off=20,r=0) <-- here
2733 * if (R4 > data_end)
2734 * ... R4(off=20,r=20), R3(off=0,r=20) and R3 can be used to access.
2735 * ... the code further went all the way to bpf_exit.
2736 * Now the 'cur' state at the mark 'here' has R4(off=30,r=0).
2737 * old_R4(off=20,r=0) equal to cur_R4(off=30,r=0), since if the verifier
2738 * goes further, such cur_R4 will give larger safe packet range after
2739 * 'if (R4 > data_end)' and all further insn were already good with r=20,
2740 * so they will be good with r=30 and we can prune the search.
2742 if (!env
->strict_alignment
&& old
->off
<= cur
->off
&&
2743 old
->off
>= old
->range
&& cur
->off
>= cur
->range
)
2749 /* compare two verifier states
2751 * all states stored in state_list are known to be valid, since
2752 * verifier reached 'bpf_exit' instruction through them
2754 * this function is called when verifier exploring different branches of
2755 * execution popped from the state stack. If it sees an old state that has
2756 * more strict register state and more strict stack state then this execution
2757 * branch doesn't need to be explored further, since verifier already
2758 * concluded that more strict state leads to valid finish.
2760 * Therefore two states are equivalent if register state is more conservative
2761 * and explored stack state is more conservative than the current one.
2764 * (slot1=INV slot2=MISC) == (slot1=MISC slot2=MISC)
2765 * (slot1=MISC slot2=MISC) != (slot1=INV slot2=MISC)
2767 * In other words if current stack state (one being explored) has more
2768 * valid slots than old one that already passed validation, it means
2769 * the verifier can stop exploring and conclude that current state is valid too
2771 * Similarly with registers. If explored state has register type as invalid
2772 * whereas register type in current state is meaningful, it means that
2773 * the current state will reach 'bpf_exit' instruction safely
2775 static bool states_equal(struct bpf_verifier_env
*env
,
2776 struct bpf_verifier_state
*old
,
2777 struct bpf_verifier_state
*cur
)
2779 bool varlen_map_access
= env
->varlen_map_value_access
;
2780 struct bpf_reg_state
*rold
, *rcur
;
2783 for (i
= 0; i
< MAX_BPF_REG
; i
++) {
2784 rold
= &old
->regs
[i
];
2785 rcur
= &cur
->regs
[i
];
2787 if (memcmp(rold
, rcur
, sizeof(*rold
)) == 0)
2790 /* If the ranges were not the same, but everything else was and
2791 * we didn't do a variable access into a map then we are a-ok.
2793 if (!varlen_map_access
&&
2794 memcmp(rold
, rcur
, offsetofend(struct bpf_reg_state
, id
)) == 0)
2797 /* If we didn't map access then again we don't care about the
2798 * mismatched range values and it's ok if our old type was
2799 * UNKNOWN and we didn't go to a NOT_INIT'ed reg.
2801 if (rold
->type
== NOT_INIT
||
2802 (!varlen_map_access
&& rold
->type
== UNKNOWN_VALUE
&&
2803 rcur
->type
!= NOT_INIT
))
2806 /* Don't care about the reg->id in this case. */
2807 if (rold
->type
== PTR_TO_MAP_VALUE_OR_NULL
&&
2808 rcur
->type
== PTR_TO_MAP_VALUE_OR_NULL
&&
2809 rold
->map_ptr
== rcur
->map_ptr
)
2812 if (rold
->type
== PTR_TO_PACKET
&& rcur
->type
== PTR_TO_PACKET
&&
2813 compare_ptrs_to_packet(env
, rold
, rcur
))
2819 for (i
= 0; i
< MAX_BPF_STACK
; i
++) {
2820 if (old
->stack_slot_type
[i
] == STACK_INVALID
)
2822 if (old
->stack_slot_type
[i
] != cur
->stack_slot_type
[i
])
2823 /* Ex: old explored (safe) state has STACK_SPILL in
2824 * this stack slot, but current has has STACK_MISC ->
2825 * this verifier states are not equivalent,
2826 * return false to continue verification of this path
2829 if (i
% BPF_REG_SIZE
)
2831 if (old
->stack_slot_type
[i
] != STACK_SPILL
)
2833 if (memcmp(&old
->spilled_regs
[i
/ BPF_REG_SIZE
],
2834 &cur
->spilled_regs
[i
/ BPF_REG_SIZE
],
2835 sizeof(old
->spilled_regs
[0])))
2836 /* when explored and current stack slot types are
2837 * the same, check that stored pointers types
2838 * are the same as well.
2839 * Ex: explored safe path could have stored
2840 * (bpf_reg_state) {.type = PTR_TO_STACK, .imm = -8}
2841 * but current path has stored:
2842 * (bpf_reg_state) {.type = PTR_TO_STACK, .imm = -16}
2843 * such verifier states are not equivalent.
2844 * return false to continue verification of this path
2853 static int is_state_visited(struct bpf_verifier_env
*env
, int insn_idx
)
2855 struct bpf_verifier_state_list
*new_sl
;
2856 struct bpf_verifier_state_list
*sl
;
2858 sl
= env
->explored_states
[insn_idx
];
2860 /* this 'insn_idx' instruction wasn't marked, so we will not
2861 * be doing state search here
2865 while (sl
!= STATE_LIST_MARK
) {
2866 if (states_equal(env
, &sl
->state
, &env
->cur_state
))
2867 /* reached equivalent register/stack state,
2874 /* there were no equivalent states, remember current one.
2875 * technically the current state is not proven to be safe yet,
2876 * but it will either reach bpf_exit (which means it's safe) or
2877 * it will be rejected. Since there are no loops, we won't be
2878 * seeing this 'insn_idx' instruction again on the way to bpf_exit
2880 new_sl
= kmalloc(sizeof(struct bpf_verifier_state_list
), GFP_USER
);
2884 /* add new state to the head of linked list */
2885 memcpy(&new_sl
->state
, &env
->cur_state
, sizeof(env
->cur_state
));
2886 new_sl
->next
= env
->explored_states
[insn_idx
];
2887 env
->explored_states
[insn_idx
] = new_sl
;
2891 static int ext_analyzer_insn_hook(struct bpf_verifier_env
*env
,
2892 int insn_idx
, int prev_insn_idx
)
2894 if (!env
->analyzer_ops
|| !env
->analyzer_ops
->insn_hook
)
2897 return env
->analyzer_ops
->insn_hook(env
, insn_idx
, prev_insn_idx
);
2900 static int do_check(struct bpf_verifier_env
*env
)
2902 struct bpf_verifier_state
*state
= &env
->cur_state
;
2903 struct bpf_insn
*insns
= env
->prog
->insnsi
;
2904 struct bpf_reg_state
*regs
= state
->regs
;
2905 int insn_cnt
= env
->prog
->len
;
2906 int insn_idx
, prev_insn_idx
= 0;
2907 int insn_processed
= 0;
2908 bool do_print_state
= false;
2910 init_reg_state(regs
);
2912 env
->varlen_map_value_access
= false;
2914 struct bpf_insn
*insn
;
2918 if (insn_idx
>= insn_cnt
) {
2919 verbose("invalid insn idx %d insn_cnt %d\n",
2920 insn_idx
, insn_cnt
);
2924 insn
= &insns
[insn_idx
];
2925 class = BPF_CLASS(insn
->code
);
2927 if (++insn_processed
> BPF_COMPLEXITY_LIMIT_INSNS
) {
2928 verbose("BPF program is too large. Processed %d insn\n",
2933 err
= is_state_visited(env
, insn_idx
);
2937 /* found equivalent state, can prune the search */
2940 verbose("\nfrom %d to %d: safe\n",
2941 prev_insn_idx
, insn_idx
);
2943 verbose("%d: safe\n", insn_idx
);
2945 goto process_bpf_exit
;
2951 if (log_level
> 1 || (log_level
&& do_print_state
)) {
2953 verbose("%d:", insn_idx
);
2955 verbose("\nfrom %d to %d:",
2956 prev_insn_idx
, insn_idx
);
2957 print_verifier_state(&env
->cur_state
);
2958 do_print_state
= false;
2962 verbose("%d: ", insn_idx
);
2963 print_bpf_insn(env
, insn
);
2966 err
= ext_analyzer_insn_hook(env
, insn_idx
, prev_insn_idx
);
2970 if (class == BPF_ALU
|| class == BPF_ALU64
) {
2971 err
= check_alu_op(env
, insn
);
2975 } else if (class == BPF_LDX
) {
2976 enum bpf_reg_type
*prev_src_type
, src_reg_type
;
2978 /* check for reserved fields is already done */
2980 /* check src operand */
2981 err
= check_reg_arg(regs
, insn
->src_reg
, SRC_OP
);
2985 err
= check_reg_arg(regs
, insn
->dst_reg
, DST_OP_NO_MARK
);
2989 src_reg_type
= regs
[insn
->src_reg
].type
;
2991 /* check that memory (src_reg + off) is readable,
2992 * the state of dst_reg will be updated by this func
2994 err
= check_mem_access(env
, insn
->src_reg
, insn
->off
,
2995 BPF_SIZE(insn
->code
), BPF_READ
,
3000 if (BPF_SIZE(insn
->code
) != BPF_W
&&
3001 BPF_SIZE(insn
->code
) != BPF_DW
) {
3006 prev_src_type
= &env
->insn_aux_data
[insn_idx
].ptr_type
;
3008 if (*prev_src_type
== NOT_INIT
) {
3010 * dst_reg = *(u32 *)(src_reg + off)
3011 * save type to validate intersecting paths
3013 *prev_src_type
= src_reg_type
;
3015 } else if (src_reg_type
!= *prev_src_type
&&
3016 (src_reg_type
== PTR_TO_CTX
||
3017 *prev_src_type
== PTR_TO_CTX
)) {
3018 /* ABuser program is trying to use the same insn
3019 * dst_reg = *(u32*) (src_reg + off)
3020 * with different pointer types:
3021 * src_reg == ctx in one branch and
3022 * src_reg == stack|map in some other branch.
3025 verbose("same insn cannot be used with different pointers\n");
3029 } else if (class == BPF_STX
) {
3030 enum bpf_reg_type
*prev_dst_type
, dst_reg_type
;
3032 if (BPF_MODE(insn
->code
) == BPF_XADD
) {
3033 err
= check_xadd(env
, insn
);
3040 /* check src1 operand */
3041 err
= check_reg_arg(regs
, insn
->src_reg
, SRC_OP
);
3044 /* check src2 operand */
3045 err
= check_reg_arg(regs
, insn
->dst_reg
, SRC_OP
);
3049 dst_reg_type
= regs
[insn
->dst_reg
].type
;
3051 /* check that memory (dst_reg + off) is writeable */
3052 err
= check_mem_access(env
, insn
->dst_reg
, insn
->off
,
3053 BPF_SIZE(insn
->code
), BPF_WRITE
,
3058 prev_dst_type
= &env
->insn_aux_data
[insn_idx
].ptr_type
;
3060 if (*prev_dst_type
== NOT_INIT
) {
3061 *prev_dst_type
= dst_reg_type
;
3062 } else if (dst_reg_type
!= *prev_dst_type
&&
3063 (dst_reg_type
== PTR_TO_CTX
||
3064 *prev_dst_type
== PTR_TO_CTX
)) {
3065 verbose("same insn cannot be used with different pointers\n");
3069 } else if (class == BPF_ST
) {
3070 if (BPF_MODE(insn
->code
) != BPF_MEM
||
3071 insn
->src_reg
!= BPF_REG_0
) {
3072 verbose("BPF_ST uses reserved fields\n");
3075 /* check src operand */
3076 err
= check_reg_arg(regs
, insn
->dst_reg
, SRC_OP
);
3080 /* check that memory (dst_reg + off) is writeable */
3081 err
= check_mem_access(env
, insn
->dst_reg
, insn
->off
,
3082 BPF_SIZE(insn
->code
), BPF_WRITE
,
3087 } else if (class == BPF_JMP
) {
3088 u8 opcode
= BPF_OP(insn
->code
);
3090 if (opcode
== BPF_CALL
) {
3091 if (BPF_SRC(insn
->code
) != BPF_K
||
3093 insn
->src_reg
!= BPF_REG_0
||
3094 insn
->dst_reg
!= BPF_REG_0
) {
3095 verbose("BPF_CALL uses reserved fields\n");
3099 err
= check_call(env
, insn
->imm
, insn_idx
);
3103 } else if (opcode
== BPF_JA
) {
3104 if (BPF_SRC(insn
->code
) != BPF_K
||
3106 insn
->src_reg
!= BPF_REG_0
||
3107 insn
->dst_reg
!= BPF_REG_0
) {
3108 verbose("BPF_JA uses reserved fields\n");
3112 insn_idx
+= insn
->off
+ 1;
3115 } else if (opcode
== BPF_EXIT
) {
3116 if (BPF_SRC(insn
->code
) != BPF_K
||
3118 insn
->src_reg
!= BPF_REG_0
||
3119 insn
->dst_reg
!= BPF_REG_0
) {
3120 verbose("BPF_EXIT uses reserved fields\n");
3124 /* eBPF calling convetion is such that R0 is used
3125 * to return the value from eBPF program.
3126 * Make sure that it's readable at this time
3127 * of bpf_exit, which means that program wrote
3128 * something into it earlier
3130 err
= check_reg_arg(regs
, BPF_REG_0
, SRC_OP
);
3134 if (is_pointer_value(env
, BPF_REG_0
)) {
3135 verbose("R0 leaks addr as return value\n");
3140 insn_idx
= pop_stack(env
, &prev_insn_idx
);
3144 do_print_state
= true;
3148 err
= check_cond_jmp_op(env
, insn
, &insn_idx
);
3152 } else if (class == BPF_LD
) {
3153 u8 mode
= BPF_MODE(insn
->code
);
3155 if (mode
== BPF_ABS
|| mode
== BPF_IND
) {
3156 err
= check_ld_abs(env
, insn
);
3160 } else if (mode
== BPF_IMM
) {
3161 err
= check_ld_imm(env
, insn
);
3167 verbose("invalid BPF_LD mode\n");
3170 reset_reg_range_values(regs
, insn
->dst_reg
);
3172 verbose("unknown insn class %d\n", class);
3179 verbose("processed %d insns, stack depth %d\n",
3180 insn_processed
, env
->prog
->aux
->stack_depth
);
3184 static int check_map_prealloc(struct bpf_map
*map
)
3186 return (map
->map_type
!= BPF_MAP_TYPE_HASH
&&
3187 map
->map_type
!= BPF_MAP_TYPE_PERCPU_HASH
&&
3188 map
->map_type
!= BPF_MAP_TYPE_HASH_OF_MAPS
) ||
3189 !(map
->map_flags
& BPF_F_NO_PREALLOC
);
3192 static int check_map_prog_compatibility(struct bpf_map
*map
,
3193 struct bpf_prog
*prog
)
3196 /* Make sure that BPF_PROG_TYPE_PERF_EVENT programs only use
3197 * preallocated hash maps, since doing memory allocation
3198 * in overflow_handler can crash depending on where nmi got
3201 if (prog
->type
== BPF_PROG_TYPE_PERF_EVENT
) {
3202 if (!check_map_prealloc(map
)) {
3203 verbose("perf_event programs can only use preallocated hash map\n");
3206 if (map
->inner_map_meta
&&
3207 !check_map_prealloc(map
->inner_map_meta
)) {
3208 verbose("perf_event programs can only use preallocated inner hash map\n");
3215 /* look for pseudo eBPF instructions that access map FDs and
3216 * replace them with actual map pointers
3218 static int replace_map_fd_with_map_ptr(struct bpf_verifier_env
*env
)
3220 struct bpf_insn
*insn
= env
->prog
->insnsi
;
3221 int insn_cnt
= env
->prog
->len
;
3224 err
= bpf_prog_calc_tag(env
->prog
);
3228 for (i
= 0; i
< insn_cnt
; i
++, insn
++) {
3229 if (BPF_CLASS(insn
->code
) == BPF_LDX
&&
3230 (BPF_MODE(insn
->code
) != BPF_MEM
|| insn
->imm
!= 0)) {
3231 verbose("BPF_LDX uses reserved fields\n");
3235 if (BPF_CLASS(insn
->code
) == BPF_STX
&&
3236 ((BPF_MODE(insn
->code
) != BPF_MEM
&&
3237 BPF_MODE(insn
->code
) != BPF_XADD
) || insn
->imm
!= 0)) {
3238 verbose("BPF_STX uses reserved fields\n");
3242 if (insn
[0].code
== (BPF_LD
| BPF_IMM
| BPF_DW
)) {
3243 struct bpf_map
*map
;
3246 if (i
== insn_cnt
- 1 || insn
[1].code
!= 0 ||
3247 insn
[1].dst_reg
!= 0 || insn
[1].src_reg
!= 0 ||
3249 verbose("invalid bpf_ld_imm64 insn\n");
3253 if (insn
->src_reg
== 0)
3254 /* valid generic load 64-bit imm */
3257 if (insn
->src_reg
!= BPF_PSEUDO_MAP_FD
) {
3258 verbose("unrecognized bpf_ld_imm64 insn\n");
3262 f
= fdget(insn
->imm
);
3263 map
= __bpf_map_get(f
);
3265 verbose("fd %d is not pointing to valid bpf_map\n",
3267 return PTR_ERR(map
);
3270 err
= check_map_prog_compatibility(map
, env
->prog
);
3276 /* store map pointer inside BPF_LD_IMM64 instruction */
3277 insn
[0].imm
= (u32
) (unsigned long) map
;
3278 insn
[1].imm
= ((u64
) (unsigned long) map
) >> 32;
3280 /* check whether we recorded this map already */
3281 for (j
= 0; j
< env
->used_map_cnt
; j
++)
3282 if (env
->used_maps
[j
] == map
) {
3287 if (env
->used_map_cnt
>= MAX_USED_MAPS
) {
3292 /* hold the map. If the program is rejected by verifier,
3293 * the map will be released by release_maps() or it
3294 * will be used by the valid program until it's unloaded
3295 * and all maps are released in free_bpf_prog_info()
3297 map
= bpf_map_inc(map
, false);
3300 return PTR_ERR(map
);
3302 env
->used_maps
[env
->used_map_cnt
++] = map
;
3311 /* now all pseudo BPF_LD_IMM64 instructions load valid
3312 * 'struct bpf_map *' into a register instead of user map_fd.
3313 * These pointers will be used later by verifier to validate map access.
3318 /* drop refcnt of maps used by the rejected program */
3319 static void release_maps(struct bpf_verifier_env
*env
)
3323 for (i
= 0; i
< env
->used_map_cnt
; i
++)
3324 bpf_map_put(env
->used_maps
[i
]);
3327 /* convert pseudo BPF_LD_IMM64 into generic BPF_LD_IMM64 */
3328 static void convert_pseudo_ld_imm64(struct bpf_verifier_env
*env
)
3330 struct bpf_insn
*insn
= env
->prog
->insnsi
;
3331 int insn_cnt
= env
->prog
->len
;
3334 for (i
= 0; i
< insn_cnt
; i
++, insn
++)
3335 if (insn
->code
== (BPF_LD
| BPF_IMM
| BPF_DW
))
3339 /* single env->prog->insni[off] instruction was replaced with the range
3340 * insni[off, off + cnt). Adjust corresponding insn_aux_data by copying
3341 * [0, off) and [off, end) to new locations, so the patched range stays zero
3343 static int adjust_insn_aux_data(struct bpf_verifier_env
*env
, u32 prog_len
,
3346 struct bpf_insn_aux_data
*new_data
, *old_data
= env
->insn_aux_data
;
3350 new_data
= vzalloc(sizeof(struct bpf_insn_aux_data
) * prog_len
);
3353 memcpy(new_data
, old_data
, sizeof(struct bpf_insn_aux_data
) * off
);
3354 memcpy(new_data
+ off
+ cnt
- 1, old_data
+ off
,
3355 sizeof(struct bpf_insn_aux_data
) * (prog_len
- off
- cnt
+ 1));
3356 env
->insn_aux_data
= new_data
;
3361 static struct bpf_prog
*bpf_patch_insn_data(struct bpf_verifier_env
*env
, u32 off
,
3362 const struct bpf_insn
*patch
, u32 len
)
3364 struct bpf_prog
*new_prog
;
3366 new_prog
= bpf_patch_insn_single(env
->prog
, off
, patch
, len
);
3369 if (adjust_insn_aux_data(env
, new_prog
->len
, off
, len
))
3374 /* convert load instructions that access fields of 'struct __sk_buff'
3375 * into sequence of instructions that access fields of 'struct sk_buff'
3377 static int convert_ctx_accesses(struct bpf_verifier_env
*env
)
3379 const struct bpf_verifier_ops
*ops
= env
->prog
->aux
->ops
;
3380 const int insn_cnt
= env
->prog
->len
;
3381 struct bpf_insn insn_buf
[16], *insn
;
3382 struct bpf_prog
*new_prog
;
3383 enum bpf_access_type type
;
3384 int i
, cnt
, delta
= 0;
3386 if (ops
->gen_prologue
) {
3387 cnt
= ops
->gen_prologue(insn_buf
, env
->seen_direct_write
,
3389 if (cnt
>= ARRAY_SIZE(insn_buf
)) {
3390 verbose("bpf verifier is misconfigured\n");
3393 new_prog
= bpf_patch_insn_data(env
, 0, insn_buf
, cnt
);
3397 env
->prog
= new_prog
;
3402 if (!ops
->convert_ctx_access
)
3405 insn
= env
->prog
->insnsi
+ delta
;
3407 for (i
= 0; i
< insn_cnt
; i
++, insn
++) {
3408 if (insn
->code
== (BPF_LDX
| BPF_MEM
| BPF_B
) ||
3409 insn
->code
== (BPF_LDX
| BPF_MEM
| BPF_H
) ||
3410 insn
->code
== (BPF_LDX
| BPF_MEM
| BPF_W
) ||
3411 insn
->code
== (BPF_LDX
| BPF_MEM
| BPF_DW
))
3413 else if (insn
->code
== (BPF_STX
| BPF_MEM
| BPF_B
) ||
3414 insn
->code
== (BPF_STX
| BPF_MEM
| BPF_H
) ||
3415 insn
->code
== (BPF_STX
| BPF_MEM
| BPF_W
) ||
3416 insn
->code
== (BPF_STX
| BPF_MEM
| BPF_DW
))
3421 if (env
->insn_aux_data
[i
+ delta
].ptr_type
!= PTR_TO_CTX
)
3424 cnt
= ops
->convert_ctx_access(type
, insn
, insn_buf
, env
->prog
);
3425 if (cnt
== 0 || cnt
>= ARRAY_SIZE(insn_buf
)) {
3426 verbose("bpf verifier is misconfigured\n");
3430 new_prog
= bpf_patch_insn_data(env
, i
+ delta
, insn_buf
, cnt
);
3436 /* keep walking new program and skip insns we just inserted */
3437 env
->prog
= new_prog
;
3438 insn
= new_prog
->insnsi
+ i
+ delta
;
3444 /* fixup insn->imm field of bpf_call instructions
3445 * and inline eligible helpers as explicit sequence of BPF instructions
3447 * this function is called after eBPF program passed verification
3449 static int fixup_bpf_calls(struct bpf_verifier_env
*env
)
3451 struct bpf_prog
*prog
= env
->prog
;
3452 struct bpf_insn
*insn
= prog
->insnsi
;
3453 const struct bpf_func_proto
*fn
;
3454 const int insn_cnt
= prog
->len
;
3455 struct bpf_insn insn_buf
[16];
3456 struct bpf_prog
*new_prog
;
3457 struct bpf_map
*map_ptr
;
3458 int i
, cnt
, delta
= 0;
3460 for (i
= 0; i
< insn_cnt
; i
++, insn
++) {
3461 if (insn
->code
!= (BPF_JMP
| BPF_CALL
))
3464 if (insn
->imm
== BPF_FUNC_get_route_realm
)
3465 prog
->dst_needed
= 1;
3466 if (insn
->imm
== BPF_FUNC_get_prandom_u32
)
3467 bpf_user_rnd_init_once();
3468 if (insn
->imm
== BPF_FUNC_tail_call
) {
3469 /* If we tail call into other programs, we
3470 * cannot make any assumptions since they can
3471 * be replaced dynamically during runtime in
3472 * the program array.
3474 prog
->cb_access
= 1;
3475 env
->prog
->aux
->stack_depth
= MAX_BPF_STACK
;
3477 /* mark bpf_tail_call as different opcode to avoid
3478 * conditional branch in the interpeter for every normal
3479 * call and to prevent accidental JITing by JIT compiler
3480 * that doesn't support bpf_tail_call yet
3483 insn
->code
= BPF_JMP
| BPF_TAIL_CALL
;
3487 if (ebpf_jit_enabled() && insn
->imm
== BPF_FUNC_map_lookup_elem
) {
3488 map_ptr
= env
->insn_aux_data
[i
+ delta
].map_ptr
;
3489 if (map_ptr
== BPF_MAP_PTR_POISON
||
3490 !map_ptr
->ops
->map_gen_lookup
)
3491 goto patch_call_imm
;
3493 cnt
= map_ptr
->ops
->map_gen_lookup(map_ptr
, insn_buf
);
3494 if (cnt
== 0 || cnt
>= ARRAY_SIZE(insn_buf
)) {
3495 verbose("bpf verifier is misconfigured\n");
3499 new_prog
= bpf_patch_insn_data(env
, i
+ delta
, insn_buf
,
3506 /* keep walking new program and skip insns we just inserted */
3507 env
->prog
= prog
= new_prog
;
3508 insn
= new_prog
->insnsi
+ i
+ delta
;
3513 fn
= prog
->aux
->ops
->get_func_proto(insn
->imm
);
3514 /* all functions that have prototype and verifier allowed
3515 * programs to call them, must be real in-kernel functions
3518 verbose("kernel subsystem misconfigured func %s#%d\n",
3519 func_id_name(insn
->imm
), insn
->imm
);
3522 insn
->imm
= fn
->func
- __bpf_call_base
;
3528 static void free_states(struct bpf_verifier_env
*env
)
3530 struct bpf_verifier_state_list
*sl
, *sln
;
3533 if (!env
->explored_states
)
3536 for (i
= 0; i
< env
->prog
->len
; i
++) {
3537 sl
= env
->explored_states
[i
];
3540 while (sl
!= STATE_LIST_MARK
) {
3547 kfree(env
->explored_states
);
3550 int bpf_check(struct bpf_prog
**prog
, union bpf_attr
*attr
)
3552 char __user
*log_ubuf
= NULL
;
3553 struct bpf_verifier_env
*env
;
3556 /* 'struct bpf_verifier_env' can be global, but since it's not small,
3557 * allocate/free it every time bpf_check() is called
3559 env
= kzalloc(sizeof(struct bpf_verifier_env
), GFP_KERNEL
);
3563 env
->insn_aux_data
= vzalloc(sizeof(struct bpf_insn_aux_data
) *
3566 if (!env
->insn_aux_data
)
3570 /* grab the mutex to protect few globals used by verifier */
3571 mutex_lock(&bpf_verifier_lock
);
3573 if (attr
->log_level
|| attr
->log_buf
|| attr
->log_size
) {
3574 /* user requested verbose verifier output
3575 * and supplied buffer to store the verification trace
3577 log_level
= attr
->log_level
;
3578 log_ubuf
= (char __user
*) (unsigned long) attr
->log_buf
;
3579 log_size
= attr
->log_size
;
3583 /* log_* values have to be sane */
3584 if (log_size
< 128 || log_size
> UINT_MAX
>> 8 ||
3585 log_level
== 0 || log_ubuf
== NULL
)
3589 log_buf
= vmalloc(log_size
);
3596 env
->strict_alignment
= !!(attr
->prog_flags
& BPF_F_STRICT_ALIGNMENT
);
3597 if (!IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS
))
3598 env
->strict_alignment
= true;
3600 ret
= replace_map_fd_with_map_ptr(env
);
3602 goto skip_full_check
;
3604 env
->explored_states
= kcalloc(env
->prog
->len
,
3605 sizeof(struct bpf_verifier_state_list
*),
3608 if (!env
->explored_states
)
3609 goto skip_full_check
;
3611 ret
= check_cfg(env
);
3613 goto skip_full_check
;
3615 env
->allow_ptr_leaks
= capable(CAP_SYS_ADMIN
);
3617 ret
= do_check(env
);
3620 while (pop_stack(env
, NULL
) >= 0);
3624 /* program is valid, convert *(u32*)(ctx + off) accesses */
3625 ret
= convert_ctx_accesses(env
);
3628 ret
= fixup_bpf_calls(env
);
3630 if (log_level
&& log_len
>= log_size
- 1) {
3631 BUG_ON(log_len
>= log_size
);
3632 /* verifier log exceeded user supplied buffer */
3634 /* fall through to return what was recorded */
3637 /* copy verifier log back to user space including trailing zero */
3638 if (log_level
&& copy_to_user(log_ubuf
, log_buf
, log_len
+ 1) != 0) {
3643 if (ret
== 0 && env
->used_map_cnt
) {
3644 /* if program passed verifier, update used_maps in bpf_prog_info */
3645 env
->prog
->aux
->used_maps
= kmalloc_array(env
->used_map_cnt
,
3646 sizeof(env
->used_maps
[0]),
3649 if (!env
->prog
->aux
->used_maps
) {
3654 memcpy(env
->prog
->aux
->used_maps
, env
->used_maps
,
3655 sizeof(env
->used_maps
[0]) * env
->used_map_cnt
);
3656 env
->prog
->aux
->used_map_cnt
= env
->used_map_cnt
;
3658 /* program is valid. Convert pseudo bpf_ld_imm64 into generic
3659 * bpf_ld_imm64 instructions
3661 convert_pseudo_ld_imm64(env
);
3667 if (!env
->prog
->aux
->used_maps
)
3668 /* if we didn't copy map pointers into bpf_prog_info, release
3669 * them now. Otherwise free_bpf_prog_info() will release them.
3674 mutex_unlock(&bpf_verifier_lock
);
3675 vfree(env
->insn_aux_data
);
3681 int bpf_analyzer(struct bpf_prog
*prog
, const struct bpf_ext_analyzer_ops
*ops
,
3684 struct bpf_verifier_env
*env
;
3687 env
= kzalloc(sizeof(struct bpf_verifier_env
), GFP_KERNEL
);
3691 env
->insn_aux_data
= vzalloc(sizeof(struct bpf_insn_aux_data
) *
3694 if (!env
->insn_aux_data
)
3697 env
->analyzer_ops
= ops
;
3698 env
->analyzer_priv
= priv
;
3700 /* grab the mutex to protect few globals used by verifier */
3701 mutex_lock(&bpf_verifier_lock
);
3705 env
->strict_alignment
= false;
3706 if (!IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS
))
3707 env
->strict_alignment
= true;
3709 env
->explored_states
= kcalloc(env
->prog
->len
,
3710 sizeof(struct bpf_verifier_state_list
*),
3713 if (!env
->explored_states
)
3714 goto skip_full_check
;
3716 ret
= check_cfg(env
);
3718 goto skip_full_check
;
3720 env
->allow_ptr_leaks
= capable(CAP_SYS_ADMIN
);
3722 ret
= do_check(env
);
3725 while (pop_stack(env
, NULL
) >= 0);
3728 mutex_unlock(&bpf_verifier_lock
);
3729 vfree(env
->insn_aux_data
);
3734 EXPORT_SYMBOL_GPL(bpf_analyzer
);