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 65536
144 #define BPF_COMPLEXITY_LIMIT_STACK 1024
146 struct bpf_call_arg_meta
{
147 struct bpf_map
*map_ptr
;
154 /* verbose verifier prints what it's seeing
155 * bpf_check() is called under lock, so no race to access these global vars
157 static u32 log_level
, log_size
, log_len
;
158 static char *log_buf
;
160 static DEFINE_MUTEX(bpf_verifier_lock
);
162 /* log_level controls verbosity level of eBPF verifier.
163 * verbose() is used to dump the verification trace to the log, so the user
164 * can figure out what's wrong with the program
166 static __printf(1, 2) void verbose(const char *fmt
, ...)
170 if (log_level
== 0 || log_len
>= log_size
- 1)
174 log_len
+= vscnprintf(log_buf
+ log_len
, log_size
- log_len
, fmt
, args
);
178 /* string representation of 'enum bpf_reg_type' */
179 static const char * const reg_type_str
[] = {
181 [UNKNOWN_VALUE
] = "inv",
182 [PTR_TO_CTX
] = "ctx",
183 [CONST_PTR_TO_MAP
] = "map_ptr",
184 [PTR_TO_MAP_VALUE
] = "map_value",
185 [PTR_TO_MAP_VALUE_OR_NULL
] = "map_value_or_null",
186 [PTR_TO_MAP_VALUE_ADJ
] = "map_value_adj",
188 [PTR_TO_STACK
] = "fp",
190 [PTR_TO_PACKET
] = "pkt",
191 [PTR_TO_PACKET_END
] = "pkt_end",
194 #define __BPF_FUNC_STR_FN(x) [BPF_FUNC_ ## x] = __stringify(bpf_ ## x)
195 static const char * const func_id_str
[] = {
196 __BPF_FUNC_MAPPER(__BPF_FUNC_STR_FN
)
198 #undef __BPF_FUNC_STR_FN
200 static const char *func_id_name(int id
)
202 BUILD_BUG_ON(ARRAY_SIZE(func_id_str
) != __BPF_FUNC_MAX_ID
);
204 if (id
>= 0 && id
< __BPF_FUNC_MAX_ID
&& func_id_str
[id
])
205 return func_id_str
[id
];
210 static void print_verifier_state(struct bpf_verifier_state
*state
)
212 struct bpf_reg_state
*reg
;
216 for (i
= 0; i
< MAX_BPF_REG
; i
++) {
217 reg
= &state
->regs
[i
];
221 verbose(" R%d=%s", i
, reg_type_str
[t
]);
222 if (t
== CONST_IMM
|| t
== PTR_TO_STACK
)
223 verbose("%lld", reg
->imm
);
224 else if (t
== PTR_TO_PACKET
)
225 verbose("(id=%d,off=%d,r=%d)",
226 reg
->id
, reg
->off
, reg
->range
);
227 else if (t
== UNKNOWN_VALUE
&& reg
->imm
)
228 verbose("%lld", reg
->imm
);
229 else if (t
== CONST_PTR_TO_MAP
|| t
== PTR_TO_MAP_VALUE
||
230 t
== PTR_TO_MAP_VALUE_OR_NULL
||
231 t
== PTR_TO_MAP_VALUE_ADJ
)
232 verbose("(ks=%d,vs=%d,id=%u)",
233 reg
->map_ptr
->key_size
,
234 reg
->map_ptr
->value_size
,
236 if (reg
->min_value
!= BPF_REGISTER_MIN_RANGE
)
237 verbose(",min_value=%lld",
238 (long long)reg
->min_value
);
239 if (reg
->max_value
!= BPF_REGISTER_MAX_RANGE
)
240 verbose(",max_value=%llu",
241 (unsigned long long)reg
->max_value
);
243 for (i
= 0; i
< MAX_BPF_STACK
; i
+= BPF_REG_SIZE
) {
244 if (state
->stack_slot_type
[i
] == STACK_SPILL
)
245 verbose(" fp%d=%s", -MAX_BPF_STACK
+ i
,
246 reg_type_str
[state
->spilled_regs
[i
/ BPF_REG_SIZE
].type
]);
251 static const char *const bpf_class_string
[] = {
259 [BPF_ALU64
] = "alu64",
262 static const char *const bpf_alu_string
[16] = {
263 [BPF_ADD
>> 4] = "+=",
264 [BPF_SUB
>> 4] = "-=",
265 [BPF_MUL
>> 4] = "*=",
266 [BPF_DIV
>> 4] = "/=",
267 [BPF_OR
>> 4] = "|=",
268 [BPF_AND
>> 4] = "&=",
269 [BPF_LSH
>> 4] = "<<=",
270 [BPF_RSH
>> 4] = ">>=",
271 [BPF_NEG
>> 4] = "neg",
272 [BPF_MOD
>> 4] = "%=",
273 [BPF_XOR
>> 4] = "^=",
274 [BPF_MOV
>> 4] = "=",
275 [BPF_ARSH
>> 4] = "s>>=",
276 [BPF_END
>> 4] = "endian",
279 static const char *const bpf_ldst_string
[] = {
280 [BPF_W
>> 3] = "u32",
281 [BPF_H
>> 3] = "u16",
283 [BPF_DW
>> 3] = "u64",
286 static const char *const bpf_jmp_string
[16] = {
287 [BPF_JA
>> 4] = "jmp",
288 [BPF_JEQ
>> 4] = "==",
289 [BPF_JGT
>> 4] = ">",
290 [BPF_JGE
>> 4] = ">=",
291 [BPF_JSET
>> 4] = "&",
292 [BPF_JNE
>> 4] = "!=",
293 [BPF_JSGT
>> 4] = "s>",
294 [BPF_JSGE
>> 4] = "s>=",
295 [BPF_CALL
>> 4] = "call",
296 [BPF_EXIT
>> 4] = "exit",
299 static void print_bpf_insn(struct bpf_insn
*insn
)
301 u8
class = BPF_CLASS(insn
->code
);
303 if (class == BPF_ALU
|| class == BPF_ALU64
) {
304 if (BPF_SRC(insn
->code
) == BPF_X
)
305 verbose("(%02x) %sr%d %s %sr%d\n",
306 insn
->code
, class == BPF_ALU
? "(u32) " : "",
308 bpf_alu_string
[BPF_OP(insn
->code
) >> 4],
309 class == BPF_ALU
? "(u32) " : "",
312 verbose("(%02x) %sr%d %s %s%d\n",
313 insn
->code
, class == BPF_ALU
? "(u32) " : "",
315 bpf_alu_string
[BPF_OP(insn
->code
) >> 4],
316 class == BPF_ALU
? "(u32) " : "",
318 } else if (class == BPF_STX
) {
319 if (BPF_MODE(insn
->code
) == BPF_MEM
)
320 verbose("(%02x) *(%s *)(r%d %+d) = r%d\n",
322 bpf_ldst_string
[BPF_SIZE(insn
->code
) >> 3],
324 insn
->off
, insn
->src_reg
);
325 else if (BPF_MODE(insn
->code
) == BPF_XADD
)
326 verbose("(%02x) lock *(%s *)(r%d %+d) += r%d\n",
328 bpf_ldst_string
[BPF_SIZE(insn
->code
) >> 3],
329 insn
->dst_reg
, insn
->off
,
332 verbose("BUG_%02x\n", insn
->code
);
333 } else if (class == BPF_ST
) {
334 if (BPF_MODE(insn
->code
) != BPF_MEM
) {
335 verbose("BUG_st_%02x\n", insn
->code
);
338 verbose("(%02x) *(%s *)(r%d %+d) = %d\n",
340 bpf_ldst_string
[BPF_SIZE(insn
->code
) >> 3],
342 insn
->off
, insn
->imm
);
343 } else if (class == BPF_LDX
) {
344 if (BPF_MODE(insn
->code
) != BPF_MEM
) {
345 verbose("BUG_ldx_%02x\n", insn
->code
);
348 verbose("(%02x) r%d = *(%s *)(r%d %+d)\n",
349 insn
->code
, insn
->dst_reg
,
350 bpf_ldst_string
[BPF_SIZE(insn
->code
) >> 3],
351 insn
->src_reg
, insn
->off
);
352 } else if (class == BPF_LD
) {
353 if (BPF_MODE(insn
->code
) == BPF_ABS
) {
354 verbose("(%02x) r0 = *(%s *)skb[%d]\n",
356 bpf_ldst_string
[BPF_SIZE(insn
->code
) >> 3],
358 } else if (BPF_MODE(insn
->code
) == BPF_IND
) {
359 verbose("(%02x) r0 = *(%s *)skb[r%d + %d]\n",
361 bpf_ldst_string
[BPF_SIZE(insn
->code
) >> 3],
362 insn
->src_reg
, insn
->imm
);
363 } else if (BPF_MODE(insn
->code
) == BPF_IMM
) {
364 verbose("(%02x) r%d = 0x%x\n",
365 insn
->code
, insn
->dst_reg
, insn
->imm
);
367 verbose("BUG_ld_%02x\n", insn
->code
);
370 } else if (class == BPF_JMP
) {
371 u8 opcode
= BPF_OP(insn
->code
);
373 if (opcode
== BPF_CALL
) {
374 verbose("(%02x) call %s#%d\n", insn
->code
,
375 func_id_name(insn
->imm
), insn
->imm
);
376 } else if (insn
->code
== (BPF_JMP
| BPF_JA
)) {
377 verbose("(%02x) goto pc%+d\n",
378 insn
->code
, insn
->off
);
379 } else if (insn
->code
== (BPF_JMP
| BPF_EXIT
)) {
380 verbose("(%02x) exit\n", insn
->code
);
381 } else if (BPF_SRC(insn
->code
) == BPF_X
) {
382 verbose("(%02x) if r%d %s r%d goto pc%+d\n",
383 insn
->code
, insn
->dst_reg
,
384 bpf_jmp_string
[BPF_OP(insn
->code
) >> 4],
385 insn
->src_reg
, insn
->off
);
387 verbose("(%02x) if r%d %s 0x%x goto pc%+d\n",
388 insn
->code
, insn
->dst_reg
,
389 bpf_jmp_string
[BPF_OP(insn
->code
) >> 4],
390 insn
->imm
, insn
->off
);
393 verbose("(%02x) %s\n", insn
->code
, bpf_class_string
[class]);
397 static int pop_stack(struct bpf_verifier_env
*env
, int *prev_insn_idx
)
399 struct bpf_verifier_stack_elem
*elem
;
402 if (env
->head
== NULL
)
405 memcpy(&env
->cur_state
, &env
->head
->st
, sizeof(env
->cur_state
));
406 insn_idx
= env
->head
->insn_idx
;
408 *prev_insn_idx
= env
->head
->prev_insn_idx
;
409 elem
= env
->head
->next
;
416 static struct bpf_verifier_state
*push_stack(struct bpf_verifier_env
*env
,
417 int insn_idx
, int prev_insn_idx
)
419 struct bpf_verifier_stack_elem
*elem
;
421 elem
= kmalloc(sizeof(struct bpf_verifier_stack_elem
), GFP_KERNEL
);
425 memcpy(&elem
->st
, &env
->cur_state
, sizeof(env
->cur_state
));
426 elem
->insn_idx
= insn_idx
;
427 elem
->prev_insn_idx
= prev_insn_idx
;
428 elem
->next
= env
->head
;
431 if (env
->stack_size
> BPF_COMPLEXITY_LIMIT_STACK
) {
432 verbose("BPF program is too complex\n");
437 /* pop all elements and return */
438 while (pop_stack(env
, NULL
) >= 0);
442 #define CALLER_SAVED_REGS 6
443 static const int caller_saved
[CALLER_SAVED_REGS
] = {
444 BPF_REG_0
, BPF_REG_1
, BPF_REG_2
, BPF_REG_3
, BPF_REG_4
, BPF_REG_5
447 static void init_reg_state(struct bpf_reg_state
*regs
)
451 for (i
= 0; i
< MAX_BPF_REG
; i
++) {
452 regs
[i
].type
= NOT_INIT
;
454 regs
[i
].min_value
= BPF_REGISTER_MIN_RANGE
;
455 regs
[i
].max_value
= BPF_REGISTER_MAX_RANGE
;
459 regs
[BPF_REG_FP
].type
= FRAME_PTR
;
461 /* 1st arg to a function */
462 regs
[BPF_REG_1
].type
= PTR_TO_CTX
;
465 static void __mark_reg_unknown_value(struct bpf_reg_state
*regs
, u32 regno
)
467 regs
[regno
].type
= UNKNOWN_VALUE
;
472 static void mark_reg_unknown_value(struct bpf_reg_state
*regs
, u32 regno
)
474 BUG_ON(regno
>= MAX_BPF_REG
);
475 __mark_reg_unknown_value(regs
, regno
);
478 static void reset_reg_range_values(struct bpf_reg_state
*regs
, u32 regno
)
480 regs
[regno
].min_value
= BPF_REGISTER_MIN_RANGE
;
481 regs
[regno
].max_value
= BPF_REGISTER_MAX_RANGE
;
484 static void mark_reg_unknown_value_and_range(struct bpf_reg_state
*regs
,
487 mark_reg_unknown_value(regs
, regno
);
488 reset_reg_range_values(regs
, regno
);
492 SRC_OP
, /* register is used as source operand */
493 DST_OP
, /* register is used as destination operand */
494 DST_OP_NO_MARK
/* same as above, check only, don't mark */
497 static int check_reg_arg(struct bpf_reg_state
*regs
, u32 regno
,
500 if (regno
>= MAX_BPF_REG
) {
501 verbose("R%d is invalid\n", regno
);
506 /* check whether register used as source operand can be read */
507 if (regs
[regno
].type
== NOT_INIT
) {
508 verbose("R%d !read_ok\n", regno
);
512 /* check whether register used as dest operand can be written to */
513 if (regno
== BPF_REG_FP
) {
514 verbose("frame pointer is read only\n");
518 mark_reg_unknown_value(regs
, regno
);
523 static int bpf_size_to_bytes(int bpf_size
)
525 if (bpf_size
== BPF_W
)
527 else if (bpf_size
== BPF_H
)
529 else if (bpf_size
== BPF_B
)
531 else if (bpf_size
== BPF_DW
)
537 static bool is_spillable_regtype(enum bpf_reg_type type
)
540 case PTR_TO_MAP_VALUE
:
541 case PTR_TO_MAP_VALUE_OR_NULL
:
542 case PTR_TO_MAP_VALUE_ADJ
:
546 case PTR_TO_PACKET_END
:
548 case CONST_PTR_TO_MAP
:
555 /* check_stack_read/write functions track spill/fill of registers,
556 * stack boundary and alignment are checked in check_mem_access()
558 static int check_stack_write(struct bpf_verifier_state
*state
, int off
,
559 int size
, int value_regno
)
562 /* caller checked that off % size == 0 and -MAX_BPF_STACK <= off < 0,
563 * so it's aligned access and [off, off + size) are within stack limits
566 if (value_regno
>= 0 &&
567 is_spillable_regtype(state
->regs
[value_regno
].type
)) {
569 /* register containing pointer is being spilled into stack */
570 if (size
!= BPF_REG_SIZE
) {
571 verbose("invalid size of register spill\n");
575 /* save register state */
576 state
->spilled_regs
[(MAX_BPF_STACK
+ off
) / BPF_REG_SIZE
] =
577 state
->regs
[value_regno
];
579 for (i
= 0; i
< BPF_REG_SIZE
; i
++)
580 state
->stack_slot_type
[MAX_BPF_STACK
+ off
+ i
] = STACK_SPILL
;
582 /* regular write of data into stack */
583 state
->spilled_regs
[(MAX_BPF_STACK
+ off
) / BPF_REG_SIZE
] =
584 (struct bpf_reg_state
) {};
586 for (i
= 0; i
< size
; i
++)
587 state
->stack_slot_type
[MAX_BPF_STACK
+ off
+ i
] = STACK_MISC
;
592 static int check_stack_read(struct bpf_verifier_state
*state
, int off
, int size
,
598 slot_type
= &state
->stack_slot_type
[MAX_BPF_STACK
+ off
];
600 if (slot_type
[0] == STACK_SPILL
) {
601 if (size
!= BPF_REG_SIZE
) {
602 verbose("invalid size of register spill\n");
605 for (i
= 1; i
< BPF_REG_SIZE
; i
++) {
606 if (slot_type
[i
] != STACK_SPILL
) {
607 verbose("corrupted spill memory\n");
612 if (value_regno
>= 0)
613 /* restore register state from stack */
614 state
->regs
[value_regno
] =
615 state
->spilled_regs
[(MAX_BPF_STACK
+ off
) / BPF_REG_SIZE
];
618 for (i
= 0; i
< size
; i
++) {
619 if (slot_type
[i
] != STACK_MISC
) {
620 verbose("invalid read from stack off %d+%d size %d\n",
625 if (value_regno
>= 0)
626 /* have read misc data from the stack */
627 mark_reg_unknown_value_and_range(state
->regs
,
633 /* check read/write into map element returned by bpf_map_lookup_elem() */
634 static int check_map_access(struct bpf_verifier_env
*env
, u32 regno
, int off
,
637 struct bpf_map
*map
= env
->cur_state
.regs
[regno
].map_ptr
;
639 if (off
< 0 || size
<= 0 || off
+ size
> map
->value_size
) {
640 verbose("invalid access to map value, value_size=%d off=%d size=%d\n",
641 map
->value_size
, off
, size
);
647 /* check read/write into an adjusted map element */
648 static int check_map_access_adj(struct bpf_verifier_env
*env
, u32 regno
,
651 struct bpf_verifier_state
*state
= &env
->cur_state
;
652 struct bpf_reg_state
*reg
= &state
->regs
[regno
];
655 /* We adjusted the register to this map value, so we
656 * need to change off and size to min_value and max_value
657 * respectively to make sure our theoretical access will be
661 print_verifier_state(state
);
662 env
->varlen_map_value_access
= true;
663 /* The minimum value is only important with signed
664 * comparisons where we can't assume the floor of a
665 * value is 0. If we are using signed variables for our
666 * index'es we need to make sure that whatever we use
667 * will have a set floor within our range.
669 if (reg
->min_value
< 0) {
670 verbose("R%d min value is negative, either use unsigned index or do a if (index >=0) check.\n",
674 err
= check_map_access(env
, regno
, reg
->min_value
+ off
, size
);
676 verbose("R%d min value is outside of the array range\n",
681 /* If we haven't set a max value then we need to bail
682 * since we can't be sure we won't do bad things.
684 if (reg
->max_value
== BPF_REGISTER_MAX_RANGE
) {
685 verbose("R%d unbounded memory access, make sure to bounds check any array access into a map\n",
689 return check_map_access(env
, regno
, reg
->max_value
+ off
, size
);
692 #define MAX_PACKET_OFF 0xffff
694 static bool may_access_direct_pkt_data(struct bpf_verifier_env
*env
,
695 const struct bpf_call_arg_meta
*meta
,
696 enum bpf_access_type t
)
698 switch (env
->prog
->type
) {
699 case BPF_PROG_TYPE_LWT_IN
:
700 case BPF_PROG_TYPE_LWT_OUT
:
701 /* dst_input() and dst_output() can't write for now */
705 case BPF_PROG_TYPE_SCHED_CLS
:
706 case BPF_PROG_TYPE_SCHED_ACT
:
707 case BPF_PROG_TYPE_XDP
:
708 case BPF_PROG_TYPE_LWT_XMIT
:
710 return meta
->pkt_access
;
712 env
->seen_direct_write
= true;
719 static int check_packet_access(struct bpf_verifier_env
*env
, u32 regno
, int off
,
722 struct bpf_reg_state
*regs
= env
->cur_state
.regs
;
723 struct bpf_reg_state
*reg
= ®s
[regno
];
726 if (off
< 0 || size
<= 0 || off
+ size
> reg
->range
) {
727 verbose("invalid access to packet, off=%d size=%d, R%d(id=%d,off=%d,r=%d)\n",
728 off
, size
, regno
, reg
->id
, reg
->off
, reg
->range
);
734 /* check access to 'struct bpf_context' fields */
735 static int check_ctx_access(struct bpf_verifier_env
*env
, int off
, int size
,
736 enum bpf_access_type t
, enum bpf_reg_type
*reg_type
)
738 /* for analyzer ctx accesses are already validated and converted */
739 if (env
->analyzer_ops
)
742 if (env
->prog
->aux
->ops
->is_valid_access
&&
743 env
->prog
->aux
->ops
->is_valid_access(off
, size
, t
, reg_type
)) {
744 /* remember the offset of last byte accessed in ctx */
745 if (env
->prog
->aux
->max_ctx_offset
< off
+ size
)
746 env
->prog
->aux
->max_ctx_offset
= off
+ size
;
750 verbose("invalid bpf_context access off=%d size=%d\n", off
, size
);
754 static bool is_pointer_value(struct bpf_verifier_env
*env
, int regno
)
756 if (env
->allow_ptr_leaks
)
759 switch (env
->cur_state
.regs
[regno
].type
) {
768 static int check_ptr_alignment(struct bpf_verifier_env
*env
,
769 struct bpf_reg_state
*reg
, int off
, int size
)
771 if (reg
->type
!= PTR_TO_PACKET
&& reg
->type
!= PTR_TO_MAP_VALUE_ADJ
) {
772 if (off
% size
!= 0) {
773 verbose("misaligned access off %d size %d\n",
781 if (IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS
))
782 /* misaligned access to packet is ok on x86,arm,arm64 */
785 if (reg
->id
&& size
!= 1) {
786 verbose("Unknown packet alignment. Only byte-sized access allowed\n");
790 /* skb->data is NET_IP_ALIGN-ed */
791 if (reg
->type
== PTR_TO_PACKET
&&
792 (NET_IP_ALIGN
+ reg
->off
+ off
) % size
!= 0) {
793 verbose("misaligned packet access off %d+%d+%d size %d\n",
794 NET_IP_ALIGN
, reg
->off
, off
, size
);
800 /* check whether memory at (regno + off) is accessible for t = (read | write)
801 * if t==write, value_regno is a register which value is stored into memory
802 * if t==read, value_regno is a register which will receive the value from memory
803 * if t==write && value_regno==-1, some unknown value is stored into memory
804 * if t==read && value_regno==-1, don't care what we read from memory
806 static int check_mem_access(struct bpf_verifier_env
*env
, u32 regno
, int off
,
807 int bpf_size
, enum bpf_access_type t
,
810 struct bpf_verifier_state
*state
= &env
->cur_state
;
811 struct bpf_reg_state
*reg
= &state
->regs
[regno
];
814 if (reg
->type
== PTR_TO_STACK
)
817 size
= bpf_size_to_bytes(bpf_size
);
821 err
= check_ptr_alignment(env
, reg
, off
, size
);
825 if (reg
->type
== PTR_TO_MAP_VALUE
||
826 reg
->type
== PTR_TO_MAP_VALUE_ADJ
) {
827 if (t
== BPF_WRITE
&& value_regno
>= 0 &&
828 is_pointer_value(env
, value_regno
)) {
829 verbose("R%d leaks addr into map\n", value_regno
);
833 if (reg
->type
== PTR_TO_MAP_VALUE_ADJ
)
834 err
= check_map_access_adj(env
, regno
, off
, size
);
836 err
= check_map_access(env
, regno
, off
, size
);
837 if (!err
&& t
== BPF_READ
&& value_regno
>= 0)
838 mark_reg_unknown_value_and_range(state
->regs
,
841 } else if (reg
->type
== PTR_TO_CTX
) {
842 enum bpf_reg_type reg_type
= UNKNOWN_VALUE
;
844 if (t
== BPF_WRITE
&& value_regno
>= 0 &&
845 is_pointer_value(env
, value_regno
)) {
846 verbose("R%d leaks addr into ctx\n", value_regno
);
849 err
= check_ctx_access(env
, off
, size
, t
, ®_type
);
850 if (!err
&& t
== BPF_READ
&& value_regno
>= 0) {
851 mark_reg_unknown_value_and_range(state
->regs
,
853 /* note that reg.[id|off|range] == 0 */
854 state
->regs
[value_regno
].type
= reg_type
;
857 } else if (reg
->type
== FRAME_PTR
|| reg
->type
== PTR_TO_STACK
) {
858 if (off
>= 0 || off
< -MAX_BPF_STACK
) {
859 verbose("invalid stack off=%d size=%d\n", off
, size
);
862 if (t
== BPF_WRITE
) {
863 if (!env
->allow_ptr_leaks
&&
864 state
->stack_slot_type
[MAX_BPF_STACK
+ off
] == STACK_SPILL
&&
865 size
!= BPF_REG_SIZE
) {
866 verbose("attempt to corrupt spilled pointer on stack\n");
869 err
= check_stack_write(state
, off
, size
, value_regno
);
871 err
= check_stack_read(state
, off
, size
, value_regno
);
873 } else if (state
->regs
[regno
].type
== PTR_TO_PACKET
) {
874 if (t
== BPF_WRITE
&& !may_access_direct_pkt_data(env
, NULL
, t
)) {
875 verbose("cannot write into packet\n");
878 if (t
== BPF_WRITE
&& value_regno
>= 0 &&
879 is_pointer_value(env
, value_regno
)) {
880 verbose("R%d leaks addr into packet\n", value_regno
);
883 err
= check_packet_access(env
, regno
, off
, size
);
884 if (!err
&& t
== BPF_READ
&& value_regno
>= 0)
885 mark_reg_unknown_value_and_range(state
->regs
,
888 verbose("R%d invalid mem access '%s'\n",
889 regno
, reg_type_str
[reg
->type
]);
893 if (!err
&& size
<= 2 && value_regno
>= 0 && env
->allow_ptr_leaks
&&
894 state
->regs
[value_regno
].type
== UNKNOWN_VALUE
) {
895 /* 1 or 2 byte load zero-extends, determine the number of
896 * zero upper bits. Not doing it fo 4 byte load, since
897 * such values cannot be added to ptr_to_packet anyway.
899 state
->regs
[value_regno
].imm
= 64 - size
* 8;
904 static int check_xadd(struct bpf_verifier_env
*env
, struct bpf_insn
*insn
)
906 struct bpf_reg_state
*regs
= env
->cur_state
.regs
;
909 if ((BPF_SIZE(insn
->code
) != BPF_W
&& BPF_SIZE(insn
->code
) != BPF_DW
) ||
911 verbose("BPF_XADD uses reserved fields\n");
915 /* check src1 operand */
916 err
= check_reg_arg(regs
, insn
->src_reg
, SRC_OP
);
920 /* check src2 operand */
921 err
= check_reg_arg(regs
, insn
->dst_reg
, SRC_OP
);
925 /* check whether atomic_add can read the memory */
926 err
= check_mem_access(env
, insn
->dst_reg
, insn
->off
,
927 BPF_SIZE(insn
->code
), BPF_READ
, -1);
931 /* check whether atomic_add can write into the same memory */
932 return check_mem_access(env
, insn
->dst_reg
, insn
->off
,
933 BPF_SIZE(insn
->code
), BPF_WRITE
, -1);
936 /* when register 'regno' is passed into function that will read 'access_size'
937 * bytes from that pointer, make sure that it's within stack boundary
938 * and all elements of stack are initialized
940 static int check_stack_boundary(struct bpf_verifier_env
*env
, int regno
,
941 int access_size
, bool zero_size_allowed
,
942 struct bpf_call_arg_meta
*meta
)
944 struct bpf_verifier_state
*state
= &env
->cur_state
;
945 struct bpf_reg_state
*regs
= state
->regs
;
948 if (regs
[regno
].type
!= PTR_TO_STACK
) {
949 if (zero_size_allowed
&& access_size
== 0 &&
950 regs
[regno
].type
== CONST_IMM
&&
951 regs
[regno
].imm
== 0)
954 verbose("R%d type=%s expected=%s\n", regno
,
955 reg_type_str
[regs
[regno
].type
],
956 reg_type_str
[PTR_TO_STACK
]);
960 off
= regs
[regno
].imm
;
961 if (off
>= 0 || off
< -MAX_BPF_STACK
|| off
+ access_size
> 0 ||
963 verbose("invalid stack type R%d off=%d access_size=%d\n",
964 regno
, off
, access_size
);
968 if (meta
&& meta
->raw_mode
) {
969 meta
->access_size
= access_size
;
974 for (i
= 0; i
< access_size
; i
++) {
975 if (state
->stack_slot_type
[MAX_BPF_STACK
+ off
+ i
] != STACK_MISC
) {
976 verbose("invalid indirect read from stack off %d+%d size %d\n",
977 off
, i
, access_size
);
984 static int check_helper_mem_access(struct bpf_verifier_env
*env
, int regno
,
985 int access_size
, bool zero_size_allowed
,
986 struct bpf_call_arg_meta
*meta
)
988 struct bpf_reg_state
*regs
= env
->cur_state
.regs
;
990 switch (regs
[regno
].type
) {
992 return check_packet_access(env
, regno
, 0, access_size
);
993 case PTR_TO_MAP_VALUE
:
994 return check_map_access(env
, regno
, 0, access_size
);
995 case PTR_TO_MAP_VALUE_ADJ
:
996 return check_map_access_adj(env
, regno
, 0, access_size
);
997 default: /* const_imm|ptr_to_stack or invalid ptr */
998 return check_stack_boundary(env
, regno
, access_size
,
999 zero_size_allowed
, meta
);
1003 static int check_func_arg(struct bpf_verifier_env
*env
, u32 regno
,
1004 enum bpf_arg_type arg_type
,
1005 struct bpf_call_arg_meta
*meta
)
1007 struct bpf_reg_state
*regs
= env
->cur_state
.regs
, *reg
= ®s
[regno
];
1008 enum bpf_reg_type expected_type
, type
= reg
->type
;
1011 if (arg_type
== ARG_DONTCARE
)
1014 if (type
== NOT_INIT
) {
1015 verbose("R%d !read_ok\n", regno
);
1019 if (arg_type
== ARG_ANYTHING
) {
1020 if (is_pointer_value(env
, regno
)) {
1021 verbose("R%d leaks addr into helper function\n", regno
);
1027 if (type
== PTR_TO_PACKET
&&
1028 !may_access_direct_pkt_data(env
, meta
, BPF_READ
)) {
1029 verbose("helper access to the packet is not allowed\n");
1033 if (arg_type
== ARG_PTR_TO_MAP_KEY
||
1034 arg_type
== ARG_PTR_TO_MAP_VALUE
) {
1035 expected_type
= PTR_TO_STACK
;
1036 if (type
!= PTR_TO_PACKET
&& type
!= expected_type
)
1038 } else if (arg_type
== ARG_CONST_SIZE
||
1039 arg_type
== ARG_CONST_SIZE_OR_ZERO
) {
1040 expected_type
= CONST_IMM
;
1041 /* One exception. Allow UNKNOWN_VALUE registers when the
1042 * boundaries are known and don't cause unsafe memory accesses
1044 if (type
!= UNKNOWN_VALUE
&& type
!= expected_type
)
1046 } else if (arg_type
== ARG_CONST_MAP_PTR
) {
1047 expected_type
= CONST_PTR_TO_MAP
;
1048 if (type
!= expected_type
)
1050 } else if (arg_type
== ARG_PTR_TO_CTX
) {
1051 expected_type
= PTR_TO_CTX
;
1052 if (type
!= expected_type
)
1054 } else if (arg_type
== ARG_PTR_TO_MEM
||
1055 arg_type
== ARG_PTR_TO_UNINIT_MEM
) {
1056 expected_type
= PTR_TO_STACK
;
1057 /* One exception here. In case function allows for NULL to be
1058 * passed in as argument, it's a CONST_IMM type. Final test
1059 * happens during stack boundary checking.
1061 if (type
== CONST_IMM
&& reg
->imm
== 0)
1062 /* final test in check_stack_boundary() */;
1063 else if (type
!= PTR_TO_PACKET
&& type
!= PTR_TO_MAP_VALUE
&&
1064 type
!= PTR_TO_MAP_VALUE_ADJ
&& type
!= expected_type
)
1066 meta
->raw_mode
= arg_type
== ARG_PTR_TO_UNINIT_MEM
;
1068 verbose("unsupported arg_type %d\n", arg_type
);
1072 if (arg_type
== ARG_CONST_MAP_PTR
) {
1073 /* bpf_map_xxx(map_ptr) call: remember that map_ptr */
1074 meta
->map_ptr
= reg
->map_ptr
;
1075 } else if (arg_type
== ARG_PTR_TO_MAP_KEY
) {
1076 /* bpf_map_xxx(..., map_ptr, ..., key) call:
1077 * check that [key, key + map->key_size) are within
1078 * stack limits and initialized
1080 if (!meta
->map_ptr
) {
1081 /* in function declaration map_ptr must come before
1082 * map_key, so that it's verified and known before
1083 * we have to check map_key here. Otherwise it means
1084 * that kernel subsystem misconfigured verifier
1086 verbose("invalid map_ptr to access map->key\n");
1089 if (type
== PTR_TO_PACKET
)
1090 err
= check_packet_access(env
, regno
, 0,
1091 meta
->map_ptr
->key_size
);
1093 err
= check_stack_boundary(env
, regno
,
1094 meta
->map_ptr
->key_size
,
1096 } else if (arg_type
== ARG_PTR_TO_MAP_VALUE
) {
1097 /* bpf_map_xxx(..., map_ptr, ..., value) call:
1098 * check [value, value + map->value_size) validity
1100 if (!meta
->map_ptr
) {
1101 /* kernel subsystem misconfigured verifier */
1102 verbose("invalid map_ptr to access map->value\n");
1105 if (type
== PTR_TO_PACKET
)
1106 err
= check_packet_access(env
, regno
, 0,
1107 meta
->map_ptr
->value_size
);
1109 err
= check_stack_boundary(env
, regno
,
1110 meta
->map_ptr
->value_size
,
1112 } else if (arg_type
== ARG_CONST_SIZE
||
1113 arg_type
== ARG_CONST_SIZE_OR_ZERO
) {
1114 bool zero_size_allowed
= (arg_type
== ARG_CONST_SIZE_OR_ZERO
);
1116 /* bpf_xxx(..., buf, len) call will access 'len' bytes
1117 * from stack pointer 'buf'. Check it
1118 * note: regno == len, regno - 1 == buf
1121 /* kernel subsystem misconfigured verifier */
1122 verbose("ARG_CONST_SIZE cannot be first argument\n");
1126 /* If the register is UNKNOWN_VALUE, the access check happens
1127 * using its boundaries. Otherwise, just use its imm
1129 if (type
== UNKNOWN_VALUE
) {
1130 /* For unprivileged variable accesses, disable raw
1131 * mode so that the program is required to
1132 * initialize all the memory that the helper could
1133 * just partially fill up.
1137 if (reg
->min_value
< 0) {
1138 verbose("R%d min value is negative, either use unsigned or 'var &= const'\n",
1143 if (reg
->min_value
== 0) {
1144 err
= check_helper_mem_access(env
, regno
- 1, 0,
1151 if (reg
->max_value
== BPF_REGISTER_MAX_RANGE
) {
1152 verbose("R%d unbounded memory access, use 'var &= const' or 'if (var < const)'\n",
1156 err
= check_helper_mem_access(env
, regno
- 1,
1158 zero_size_allowed
, meta
);
1162 /* register is CONST_IMM */
1163 err
= check_helper_mem_access(env
, regno
- 1, reg
->imm
,
1164 zero_size_allowed
, meta
);
1170 verbose("R%d type=%s expected=%s\n", regno
,
1171 reg_type_str
[type
], reg_type_str
[expected_type
]);
1175 static int check_map_func_compatibility(struct bpf_map
*map
, int func_id
)
1180 /* We need a two way check, first is from map perspective ... */
1181 switch (map
->map_type
) {
1182 case BPF_MAP_TYPE_PROG_ARRAY
:
1183 if (func_id
!= BPF_FUNC_tail_call
)
1186 case BPF_MAP_TYPE_PERF_EVENT_ARRAY
:
1187 if (func_id
!= BPF_FUNC_perf_event_read
&&
1188 func_id
!= BPF_FUNC_perf_event_output
)
1191 case BPF_MAP_TYPE_STACK_TRACE
:
1192 if (func_id
!= BPF_FUNC_get_stackid
)
1195 case BPF_MAP_TYPE_CGROUP_ARRAY
:
1196 if (func_id
!= BPF_FUNC_skb_under_cgroup
&&
1197 func_id
!= BPF_FUNC_current_task_under_cgroup
)
1204 /* ... and second from the function itself. */
1206 case BPF_FUNC_tail_call
:
1207 if (map
->map_type
!= BPF_MAP_TYPE_PROG_ARRAY
)
1210 case BPF_FUNC_perf_event_read
:
1211 case BPF_FUNC_perf_event_output
:
1212 if (map
->map_type
!= BPF_MAP_TYPE_PERF_EVENT_ARRAY
)
1215 case BPF_FUNC_get_stackid
:
1216 if (map
->map_type
!= BPF_MAP_TYPE_STACK_TRACE
)
1219 case BPF_FUNC_current_task_under_cgroup
:
1220 case BPF_FUNC_skb_under_cgroup
:
1221 if (map
->map_type
!= BPF_MAP_TYPE_CGROUP_ARRAY
)
1230 verbose("cannot pass map_type %d into func %s#%d\n",
1231 map
->map_type
, func_id_name(func_id
), func_id
);
1235 static int check_raw_mode(const struct bpf_func_proto
*fn
)
1239 if (fn
->arg1_type
== ARG_PTR_TO_UNINIT_MEM
)
1241 if (fn
->arg2_type
== ARG_PTR_TO_UNINIT_MEM
)
1243 if (fn
->arg3_type
== ARG_PTR_TO_UNINIT_MEM
)
1245 if (fn
->arg4_type
== ARG_PTR_TO_UNINIT_MEM
)
1247 if (fn
->arg5_type
== ARG_PTR_TO_UNINIT_MEM
)
1250 return count
> 1 ? -EINVAL
: 0;
1253 static void clear_all_pkt_pointers(struct bpf_verifier_env
*env
)
1255 struct bpf_verifier_state
*state
= &env
->cur_state
;
1256 struct bpf_reg_state
*regs
= state
->regs
, *reg
;
1259 for (i
= 0; i
< MAX_BPF_REG
; i
++)
1260 if (regs
[i
].type
== PTR_TO_PACKET
||
1261 regs
[i
].type
== PTR_TO_PACKET_END
)
1262 mark_reg_unknown_value(regs
, i
);
1264 for (i
= 0; i
< MAX_BPF_STACK
; i
+= BPF_REG_SIZE
) {
1265 if (state
->stack_slot_type
[i
] != STACK_SPILL
)
1267 reg
= &state
->spilled_regs
[i
/ BPF_REG_SIZE
];
1268 if (reg
->type
!= PTR_TO_PACKET
&&
1269 reg
->type
!= PTR_TO_PACKET_END
)
1271 reg
->type
= UNKNOWN_VALUE
;
1276 static int check_call(struct bpf_verifier_env
*env
, int func_id
, int insn_idx
)
1278 struct bpf_verifier_state
*state
= &env
->cur_state
;
1279 const struct bpf_func_proto
*fn
= NULL
;
1280 struct bpf_reg_state
*regs
= state
->regs
;
1281 struct bpf_reg_state
*reg
;
1282 struct bpf_call_arg_meta meta
;
1286 /* find function prototype */
1287 if (func_id
< 0 || func_id
>= __BPF_FUNC_MAX_ID
) {
1288 verbose("invalid func %s#%d\n", func_id_name(func_id
), func_id
);
1292 if (env
->prog
->aux
->ops
->get_func_proto
)
1293 fn
= env
->prog
->aux
->ops
->get_func_proto(func_id
);
1296 verbose("unknown func %s#%d\n", func_id_name(func_id
), func_id
);
1300 /* eBPF programs must be GPL compatible to use GPL-ed functions */
1301 if (!env
->prog
->gpl_compatible
&& fn
->gpl_only
) {
1302 verbose("cannot call GPL only function from proprietary program\n");
1306 changes_data
= bpf_helper_changes_pkt_data(fn
->func
);
1308 memset(&meta
, 0, sizeof(meta
));
1309 meta
.pkt_access
= fn
->pkt_access
;
1311 /* We only support one arg being in raw mode at the moment, which
1312 * is sufficient for the helper functions we have right now.
1314 err
= check_raw_mode(fn
);
1316 verbose("kernel subsystem misconfigured func %s#%d\n",
1317 func_id_name(func_id
), func_id
);
1322 err
= check_func_arg(env
, BPF_REG_1
, fn
->arg1_type
, &meta
);
1325 err
= check_func_arg(env
, BPF_REG_2
, fn
->arg2_type
, &meta
);
1328 err
= check_func_arg(env
, BPF_REG_3
, fn
->arg3_type
, &meta
);
1331 err
= check_func_arg(env
, BPF_REG_4
, fn
->arg4_type
, &meta
);
1334 err
= check_func_arg(env
, BPF_REG_5
, fn
->arg5_type
, &meta
);
1338 /* Mark slots with STACK_MISC in case of raw mode, stack offset
1339 * is inferred from register state.
1341 for (i
= 0; i
< meta
.access_size
; i
++) {
1342 err
= check_mem_access(env
, meta
.regno
, i
, BPF_B
, BPF_WRITE
, -1);
1347 /* reset caller saved regs */
1348 for (i
= 0; i
< CALLER_SAVED_REGS
; i
++) {
1349 reg
= regs
+ caller_saved
[i
];
1350 reg
->type
= NOT_INIT
;
1354 /* update return register */
1355 if (fn
->ret_type
== RET_INTEGER
) {
1356 regs
[BPF_REG_0
].type
= UNKNOWN_VALUE
;
1357 } else if (fn
->ret_type
== RET_VOID
) {
1358 regs
[BPF_REG_0
].type
= NOT_INIT
;
1359 } else if (fn
->ret_type
== RET_PTR_TO_MAP_VALUE_OR_NULL
) {
1360 regs
[BPF_REG_0
].type
= PTR_TO_MAP_VALUE_OR_NULL
;
1361 regs
[BPF_REG_0
].max_value
= regs
[BPF_REG_0
].min_value
= 0;
1362 /* remember map_ptr, so that check_map_access()
1363 * can check 'value_size' boundary of memory access
1364 * to map element returned from bpf_map_lookup_elem()
1366 if (meta
.map_ptr
== NULL
) {
1367 verbose("kernel subsystem misconfigured verifier\n");
1370 regs
[BPF_REG_0
].map_ptr
= meta
.map_ptr
;
1371 regs
[BPF_REG_0
].id
= ++env
->id_gen
;
1372 env
->insn_aux_data
[insn_idx
].map_ptr
= meta
.map_ptr
;
1374 verbose("unknown return type %d of func %s#%d\n",
1375 fn
->ret_type
, func_id_name(func_id
), func_id
);
1379 err
= check_map_func_compatibility(meta
.map_ptr
, func_id
);
1384 clear_all_pkt_pointers(env
);
1388 static int check_packet_ptr_add(struct bpf_verifier_env
*env
,
1389 struct bpf_insn
*insn
)
1391 struct bpf_reg_state
*regs
= env
->cur_state
.regs
;
1392 struct bpf_reg_state
*dst_reg
= ®s
[insn
->dst_reg
];
1393 struct bpf_reg_state
*src_reg
= ®s
[insn
->src_reg
];
1394 struct bpf_reg_state tmp_reg
;
1397 if (BPF_SRC(insn
->code
) == BPF_K
) {
1398 /* pkt_ptr += imm */
1403 verbose("addition of negative constant to packet pointer is not allowed\n");
1406 if (imm
>= MAX_PACKET_OFF
||
1407 imm
+ dst_reg
->off
>= MAX_PACKET_OFF
) {
1408 verbose("constant %d is too large to add to packet pointer\n",
1412 /* a constant was added to pkt_ptr.
1413 * Remember it while keeping the same 'id'
1415 dst_reg
->off
+= imm
;
1417 if (src_reg
->type
== PTR_TO_PACKET
) {
1418 /* R6=pkt(id=0,off=0,r=62) R7=imm22; r7 += r6 */
1419 tmp_reg
= *dst_reg
; /* save r7 state */
1420 *dst_reg
= *src_reg
; /* copy pkt_ptr state r6 into r7 */
1421 src_reg
= &tmp_reg
; /* pretend it's src_reg state */
1422 /* if the checks below reject it, the copy won't matter,
1423 * since we're rejecting the whole program. If all ok,
1424 * then imm22 state will be added to r7
1425 * and r7 will be pkt(id=0,off=22,r=62) while
1426 * r6 will stay as pkt(id=0,off=0,r=62)
1430 if (src_reg
->type
== CONST_IMM
) {
1431 /* pkt_ptr += reg where reg is known constant */
1435 /* disallow pkt_ptr += reg
1436 * if reg is not uknown_value with guaranteed zero upper bits
1437 * otherwise pkt_ptr may overflow and addition will become
1438 * subtraction which is not allowed
1440 if (src_reg
->type
!= UNKNOWN_VALUE
) {
1441 verbose("cannot add '%s' to ptr_to_packet\n",
1442 reg_type_str
[src_reg
->type
]);
1445 if (src_reg
->imm
< 48) {
1446 verbose("cannot add integer value with %lld upper zero bits to ptr_to_packet\n",
1450 /* dst_reg stays as pkt_ptr type and since some positive
1451 * integer value was added to the pointer, increment its 'id'
1453 dst_reg
->id
= ++env
->id_gen
;
1455 /* something was added to pkt_ptr, set range and off to zero */
1462 static int evaluate_reg_alu(struct bpf_verifier_env
*env
, struct bpf_insn
*insn
)
1464 struct bpf_reg_state
*regs
= env
->cur_state
.regs
;
1465 struct bpf_reg_state
*dst_reg
= ®s
[insn
->dst_reg
];
1466 u8 opcode
= BPF_OP(insn
->code
);
1469 /* for type == UNKNOWN_VALUE:
1470 * imm > 0 -> number of zero upper bits
1471 * imm == 0 -> don't track which is the same as all bits can be non-zero
1474 if (BPF_SRC(insn
->code
) == BPF_X
) {
1475 struct bpf_reg_state
*src_reg
= ®s
[insn
->src_reg
];
1477 if (src_reg
->type
== UNKNOWN_VALUE
&& src_reg
->imm
> 0 &&
1478 dst_reg
->imm
&& opcode
== BPF_ADD
) {
1480 * where both have zero upper bits. Adding them
1481 * can only result making one more bit non-zero
1482 * in the larger value.
1483 * Ex. 0xffff (imm=48) + 1 (imm=63) = 0x10000 (imm=47)
1484 * 0xffff (imm=48) + 0xffff = 0x1fffe (imm=47)
1486 dst_reg
->imm
= min(dst_reg
->imm
, src_reg
->imm
);
1490 if (src_reg
->type
== CONST_IMM
&& src_reg
->imm
> 0 &&
1491 dst_reg
->imm
&& opcode
== BPF_ADD
) {
1493 * where dreg has zero upper bits and sreg is const.
1494 * Adding them can only result making one more bit
1495 * non-zero in the larger value.
1497 imm_log2
= __ilog2_u64((long long)src_reg
->imm
);
1498 dst_reg
->imm
= min(dst_reg
->imm
, 63 - imm_log2
);
1502 /* all other cases non supported yet, just mark dst_reg */
1507 /* sign extend 32-bit imm into 64-bit to make sure that
1508 * negative values occupy bit 63. Note ilog2() would have
1509 * been incorrect, since sizeof(insn->imm) == 4
1511 imm_log2
= __ilog2_u64((long long)insn
->imm
);
1513 if (dst_reg
->imm
&& opcode
== BPF_LSH
) {
1515 * if reg was a result of 2 byte load, then its imm == 48
1516 * which means that upper 48 bits are zero and shifting this reg
1517 * left by 4 would mean that upper 44 bits are still zero
1519 dst_reg
->imm
-= insn
->imm
;
1520 } else if (dst_reg
->imm
&& opcode
== BPF_MUL
) {
1522 * if multiplying by 14 subtract 4
1523 * This is conservative calculation of upper zero bits.
1524 * It's not trying to special case insn->imm == 1 or 0 cases
1526 dst_reg
->imm
-= imm_log2
+ 1;
1527 } else if (opcode
== BPF_AND
) {
1529 dst_reg
->imm
= 63 - imm_log2
;
1530 } else if (dst_reg
->imm
&& opcode
== BPF_ADD
) {
1532 dst_reg
->imm
= min(dst_reg
->imm
, 63 - imm_log2
);
1534 } else if (opcode
== BPF_RSH
) {
1536 * which means that after right shift, upper bits will be zero
1537 * note that verifier already checked that
1538 * 0 <= imm < 64 for shift insn
1540 dst_reg
->imm
+= insn
->imm
;
1541 if (unlikely(dst_reg
->imm
> 64))
1542 /* some dumb code did:
1545 * and all bits are zero now */
1548 /* all other alu ops, means that we don't know what will
1549 * happen to the value, mark it with unknown number of zero bits
1554 if (dst_reg
->imm
< 0) {
1555 /* all 64 bits of the register can contain non-zero bits
1556 * and such value cannot be added to ptr_to_packet, since it
1557 * may overflow, mark it as unknown to avoid further eval
1564 static int evaluate_reg_imm_alu(struct bpf_verifier_env
*env
,
1565 struct bpf_insn
*insn
)
1567 struct bpf_reg_state
*regs
= env
->cur_state
.regs
;
1568 struct bpf_reg_state
*dst_reg
= ®s
[insn
->dst_reg
];
1569 struct bpf_reg_state
*src_reg
= ®s
[insn
->src_reg
];
1570 u8 opcode
= BPF_OP(insn
->code
);
1571 u64 dst_imm
= dst_reg
->imm
;
1573 /* dst_reg->type == CONST_IMM here. Simulate execution of insns
1574 * containing ALU ops. Don't care about overflow or negative
1575 * values, just add/sub/... them; registers are in u64.
1577 if (opcode
== BPF_ADD
&& BPF_SRC(insn
->code
) == BPF_K
) {
1578 dst_imm
+= insn
->imm
;
1579 } else if (opcode
== BPF_ADD
&& BPF_SRC(insn
->code
) == BPF_X
&&
1580 src_reg
->type
== CONST_IMM
) {
1581 dst_imm
+= src_reg
->imm
;
1582 } else if (opcode
== BPF_SUB
&& BPF_SRC(insn
->code
) == BPF_K
) {
1583 dst_imm
-= insn
->imm
;
1584 } else if (opcode
== BPF_SUB
&& BPF_SRC(insn
->code
) == BPF_X
&&
1585 src_reg
->type
== CONST_IMM
) {
1586 dst_imm
-= src_reg
->imm
;
1587 } else if (opcode
== BPF_MUL
&& BPF_SRC(insn
->code
) == BPF_K
) {
1588 dst_imm
*= insn
->imm
;
1589 } else if (opcode
== BPF_MUL
&& BPF_SRC(insn
->code
) == BPF_X
&&
1590 src_reg
->type
== CONST_IMM
) {
1591 dst_imm
*= src_reg
->imm
;
1592 } else if (opcode
== BPF_OR
&& BPF_SRC(insn
->code
) == BPF_K
) {
1593 dst_imm
|= insn
->imm
;
1594 } else if (opcode
== BPF_OR
&& BPF_SRC(insn
->code
) == BPF_X
&&
1595 src_reg
->type
== CONST_IMM
) {
1596 dst_imm
|= src_reg
->imm
;
1597 } else if (opcode
== BPF_AND
&& BPF_SRC(insn
->code
) == BPF_K
) {
1598 dst_imm
&= insn
->imm
;
1599 } else if (opcode
== BPF_AND
&& BPF_SRC(insn
->code
) == BPF_X
&&
1600 src_reg
->type
== CONST_IMM
) {
1601 dst_imm
&= src_reg
->imm
;
1602 } else if (opcode
== BPF_RSH
&& BPF_SRC(insn
->code
) == BPF_K
) {
1603 dst_imm
>>= insn
->imm
;
1604 } else if (opcode
== BPF_RSH
&& BPF_SRC(insn
->code
) == BPF_X
&&
1605 src_reg
->type
== CONST_IMM
) {
1606 dst_imm
>>= src_reg
->imm
;
1607 } else if (opcode
== BPF_LSH
&& BPF_SRC(insn
->code
) == BPF_K
) {
1608 dst_imm
<<= insn
->imm
;
1609 } else if (opcode
== BPF_LSH
&& BPF_SRC(insn
->code
) == BPF_X
&&
1610 src_reg
->type
== CONST_IMM
) {
1611 dst_imm
<<= src_reg
->imm
;
1613 mark_reg_unknown_value(regs
, insn
->dst_reg
);
1617 dst_reg
->imm
= dst_imm
;
1622 static void check_reg_overflow(struct bpf_reg_state
*reg
)
1624 if (reg
->max_value
> BPF_REGISTER_MAX_RANGE
)
1625 reg
->max_value
= BPF_REGISTER_MAX_RANGE
;
1626 if (reg
->min_value
< BPF_REGISTER_MIN_RANGE
||
1627 reg
->min_value
> BPF_REGISTER_MAX_RANGE
)
1628 reg
->min_value
= BPF_REGISTER_MIN_RANGE
;
1631 static void adjust_reg_min_max_vals(struct bpf_verifier_env
*env
,
1632 struct bpf_insn
*insn
)
1634 struct bpf_reg_state
*regs
= env
->cur_state
.regs
, *dst_reg
;
1635 s64 min_val
= BPF_REGISTER_MIN_RANGE
;
1636 u64 max_val
= BPF_REGISTER_MAX_RANGE
;
1637 u8 opcode
= BPF_OP(insn
->code
);
1639 dst_reg
= ®s
[insn
->dst_reg
];
1640 if (BPF_SRC(insn
->code
) == BPF_X
) {
1641 check_reg_overflow(®s
[insn
->src_reg
]);
1642 min_val
= regs
[insn
->src_reg
].min_value
;
1643 max_val
= regs
[insn
->src_reg
].max_value
;
1645 /* If the source register is a random pointer then the
1646 * min_value/max_value values represent the range of the known
1647 * accesses into that value, not the actual min/max value of the
1648 * register itself. In this case we have to reset the reg range
1649 * values so we know it is not safe to look at.
1651 if (regs
[insn
->src_reg
].type
!= CONST_IMM
&&
1652 regs
[insn
->src_reg
].type
!= UNKNOWN_VALUE
) {
1653 min_val
= BPF_REGISTER_MIN_RANGE
;
1654 max_val
= BPF_REGISTER_MAX_RANGE
;
1656 } else if (insn
->imm
< BPF_REGISTER_MAX_RANGE
&&
1657 (s64
)insn
->imm
> BPF_REGISTER_MIN_RANGE
) {
1658 min_val
= max_val
= insn
->imm
;
1661 /* We don't know anything about what was done to this register, mark it
1664 if (min_val
== BPF_REGISTER_MIN_RANGE
&&
1665 max_val
== BPF_REGISTER_MAX_RANGE
) {
1666 reset_reg_range_values(regs
, insn
->dst_reg
);
1670 /* If one of our values was at the end of our ranges then we can't just
1671 * do our normal operations to the register, we need to set the values
1672 * to the min/max since they are undefined.
1674 if (min_val
== BPF_REGISTER_MIN_RANGE
)
1675 dst_reg
->min_value
= BPF_REGISTER_MIN_RANGE
;
1676 if (max_val
== BPF_REGISTER_MAX_RANGE
)
1677 dst_reg
->max_value
= BPF_REGISTER_MAX_RANGE
;
1681 if (dst_reg
->min_value
!= BPF_REGISTER_MIN_RANGE
)
1682 dst_reg
->min_value
+= min_val
;
1683 if (dst_reg
->max_value
!= BPF_REGISTER_MAX_RANGE
)
1684 dst_reg
->max_value
+= max_val
;
1687 if (dst_reg
->min_value
!= BPF_REGISTER_MIN_RANGE
)
1688 dst_reg
->min_value
-= min_val
;
1689 if (dst_reg
->max_value
!= BPF_REGISTER_MAX_RANGE
)
1690 dst_reg
->max_value
-= max_val
;
1693 if (dst_reg
->min_value
!= BPF_REGISTER_MIN_RANGE
)
1694 dst_reg
->min_value
*= min_val
;
1695 if (dst_reg
->max_value
!= BPF_REGISTER_MAX_RANGE
)
1696 dst_reg
->max_value
*= max_val
;
1699 /* Disallow AND'ing of negative numbers, ain't nobody got time
1700 * for that. Otherwise the minimum is 0 and the max is the max
1701 * value we could AND against.
1704 dst_reg
->min_value
= BPF_REGISTER_MIN_RANGE
;
1706 dst_reg
->min_value
= 0;
1707 dst_reg
->max_value
= max_val
;
1710 /* Gotta have special overflow logic here, if we're shifting
1711 * more than MAX_RANGE then just assume we have an invalid
1714 if (min_val
> ilog2(BPF_REGISTER_MAX_RANGE
))
1715 dst_reg
->min_value
= BPF_REGISTER_MIN_RANGE
;
1716 else if (dst_reg
->min_value
!= BPF_REGISTER_MIN_RANGE
)
1717 dst_reg
->min_value
<<= min_val
;
1719 if (max_val
> ilog2(BPF_REGISTER_MAX_RANGE
))
1720 dst_reg
->max_value
= BPF_REGISTER_MAX_RANGE
;
1721 else if (dst_reg
->max_value
!= BPF_REGISTER_MAX_RANGE
)
1722 dst_reg
->max_value
<<= max_val
;
1725 /* RSH by a negative number is undefined, and the BPF_RSH is an
1726 * unsigned shift, so make the appropriate casts.
1728 if (min_val
< 0 || dst_reg
->min_value
< 0)
1729 dst_reg
->min_value
= BPF_REGISTER_MIN_RANGE
;
1731 dst_reg
->min_value
=
1732 (u64
)(dst_reg
->min_value
) >> min_val
;
1733 if (dst_reg
->max_value
!= BPF_REGISTER_MAX_RANGE
)
1734 dst_reg
->max_value
>>= max_val
;
1737 reset_reg_range_values(regs
, insn
->dst_reg
);
1741 check_reg_overflow(dst_reg
);
1744 /* check validity of 32-bit and 64-bit arithmetic operations */
1745 static int check_alu_op(struct bpf_verifier_env
*env
, struct bpf_insn
*insn
)
1747 struct bpf_reg_state
*regs
= env
->cur_state
.regs
, *dst_reg
;
1748 u8 opcode
= BPF_OP(insn
->code
);
1751 if (opcode
== BPF_END
|| opcode
== BPF_NEG
) {
1752 if (opcode
== BPF_NEG
) {
1753 if (BPF_SRC(insn
->code
) != 0 ||
1754 insn
->src_reg
!= BPF_REG_0
||
1755 insn
->off
!= 0 || insn
->imm
!= 0) {
1756 verbose("BPF_NEG uses reserved fields\n");
1760 if (insn
->src_reg
!= BPF_REG_0
|| insn
->off
!= 0 ||
1761 (insn
->imm
!= 16 && insn
->imm
!= 32 && insn
->imm
!= 64)) {
1762 verbose("BPF_END uses reserved fields\n");
1767 /* check src operand */
1768 err
= check_reg_arg(regs
, insn
->dst_reg
, SRC_OP
);
1772 if (is_pointer_value(env
, insn
->dst_reg
)) {
1773 verbose("R%d pointer arithmetic prohibited\n",
1778 /* check dest operand */
1779 err
= check_reg_arg(regs
, insn
->dst_reg
, DST_OP
);
1783 } else if (opcode
== BPF_MOV
) {
1785 if (BPF_SRC(insn
->code
) == BPF_X
) {
1786 if (insn
->imm
!= 0 || insn
->off
!= 0) {
1787 verbose("BPF_MOV uses reserved fields\n");
1791 /* check src operand */
1792 err
= check_reg_arg(regs
, insn
->src_reg
, SRC_OP
);
1796 if (insn
->src_reg
!= BPF_REG_0
|| insn
->off
!= 0) {
1797 verbose("BPF_MOV uses reserved fields\n");
1802 /* check dest operand */
1803 err
= check_reg_arg(regs
, insn
->dst_reg
, DST_OP
);
1807 /* we are setting our register to something new, we need to
1808 * reset its range values.
1810 reset_reg_range_values(regs
, insn
->dst_reg
);
1812 if (BPF_SRC(insn
->code
) == BPF_X
) {
1813 if (BPF_CLASS(insn
->code
) == BPF_ALU64
) {
1815 * copy register state to dest reg
1817 regs
[insn
->dst_reg
] = regs
[insn
->src_reg
];
1819 if (is_pointer_value(env
, insn
->src_reg
)) {
1820 verbose("R%d partial copy of pointer\n",
1824 mark_reg_unknown_value(regs
, insn
->dst_reg
);
1828 * remember the value we stored into this reg
1830 regs
[insn
->dst_reg
].type
= CONST_IMM
;
1831 regs
[insn
->dst_reg
].imm
= insn
->imm
;
1832 regs
[insn
->dst_reg
].max_value
= insn
->imm
;
1833 regs
[insn
->dst_reg
].min_value
= insn
->imm
;
1836 } else if (opcode
> BPF_END
) {
1837 verbose("invalid BPF_ALU opcode %x\n", opcode
);
1840 } else { /* all other ALU ops: and, sub, xor, add, ... */
1842 if (BPF_SRC(insn
->code
) == BPF_X
) {
1843 if (insn
->imm
!= 0 || insn
->off
!= 0) {
1844 verbose("BPF_ALU uses reserved fields\n");
1847 /* check src1 operand */
1848 err
= check_reg_arg(regs
, insn
->src_reg
, SRC_OP
);
1852 if (insn
->src_reg
!= BPF_REG_0
|| insn
->off
!= 0) {
1853 verbose("BPF_ALU uses reserved fields\n");
1858 /* check src2 operand */
1859 err
= check_reg_arg(regs
, insn
->dst_reg
, SRC_OP
);
1863 if ((opcode
== BPF_MOD
|| opcode
== BPF_DIV
) &&
1864 BPF_SRC(insn
->code
) == BPF_K
&& insn
->imm
== 0) {
1865 verbose("div by zero\n");
1869 if ((opcode
== BPF_LSH
|| opcode
== BPF_RSH
||
1870 opcode
== BPF_ARSH
) && BPF_SRC(insn
->code
) == BPF_K
) {
1871 int size
= BPF_CLASS(insn
->code
) == BPF_ALU64
? 64 : 32;
1873 if (insn
->imm
< 0 || insn
->imm
>= size
) {
1874 verbose("invalid shift %d\n", insn
->imm
);
1879 /* check dest operand */
1880 err
= check_reg_arg(regs
, insn
->dst_reg
, DST_OP_NO_MARK
);
1884 dst_reg
= ®s
[insn
->dst_reg
];
1886 /* first we want to adjust our ranges. */
1887 adjust_reg_min_max_vals(env
, insn
);
1889 /* pattern match 'bpf_add Rx, imm' instruction */
1890 if (opcode
== BPF_ADD
&& BPF_CLASS(insn
->code
) == BPF_ALU64
&&
1891 dst_reg
->type
== FRAME_PTR
&& BPF_SRC(insn
->code
) == BPF_K
) {
1892 dst_reg
->type
= PTR_TO_STACK
;
1893 dst_reg
->imm
= insn
->imm
;
1895 } else if (opcode
== BPF_ADD
&&
1896 BPF_CLASS(insn
->code
) == BPF_ALU64
&&
1897 (dst_reg
->type
== PTR_TO_PACKET
||
1898 (BPF_SRC(insn
->code
) == BPF_X
&&
1899 regs
[insn
->src_reg
].type
== PTR_TO_PACKET
))) {
1900 /* ptr_to_packet += K|X */
1901 return check_packet_ptr_add(env
, insn
);
1902 } else if (BPF_CLASS(insn
->code
) == BPF_ALU64
&&
1903 dst_reg
->type
== UNKNOWN_VALUE
&&
1904 env
->allow_ptr_leaks
) {
1905 /* unknown += K|X */
1906 return evaluate_reg_alu(env
, insn
);
1907 } else if (BPF_CLASS(insn
->code
) == BPF_ALU64
&&
1908 dst_reg
->type
== CONST_IMM
&&
1909 env
->allow_ptr_leaks
) {
1910 /* reg_imm += K|X */
1911 return evaluate_reg_imm_alu(env
, insn
);
1912 } else if (is_pointer_value(env
, insn
->dst_reg
)) {
1913 verbose("R%d pointer arithmetic prohibited\n",
1916 } else if (BPF_SRC(insn
->code
) == BPF_X
&&
1917 is_pointer_value(env
, insn
->src_reg
)) {
1918 verbose("R%d pointer arithmetic prohibited\n",
1923 /* If we did pointer math on a map value then just set it to our
1924 * PTR_TO_MAP_VALUE_ADJ type so we can deal with any stores or
1925 * loads to this register appropriately, otherwise just mark the
1926 * register as unknown.
1928 if (env
->allow_ptr_leaks
&&
1929 (dst_reg
->type
== PTR_TO_MAP_VALUE
||
1930 dst_reg
->type
== PTR_TO_MAP_VALUE_ADJ
))
1931 dst_reg
->type
= PTR_TO_MAP_VALUE_ADJ
;
1933 mark_reg_unknown_value(regs
, insn
->dst_reg
);
1939 static void find_good_pkt_pointers(struct bpf_verifier_state
*state
,
1940 struct bpf_reg_state
*dst_reg
)
1942 struct bpf_reg_state
*regs
= state
->regs
, *reg
;
1945 /* LLVM can generate two kind of checks:
1951 * if (r2 > pkt_end) goto <handle exception>
1955 * r2 == dst_reg, pkt_end == src_reg
1956 * r2=pkt(id=n,off=8,r=0)
1957 * r3=pkt(id=n,off=0,r=0)
1963 * if (pkt_end >= r2) goto <access okay>
1964 * <handle exception>
1967 * pkt_end == dst_reg, r2 == src_reg
1968 * r2=pkt(id=n,off=8,r=0)
1969 * r3=pkt(id=n,off=0,r=0)
1971 * Find register r3 and mark its range as r3=pkt(id=n,off=0,r=8)
1972 * so that range of bytes [r3, r3 + 8) is safe to access.
1975 for (i
= 0; i
< MAX_BPF_REG
; i
++)
1976 if (regs
[i
].type
== PTR_TO_PACKET
&& regs
[i
].id
== dst_reg
->id
)
1977 regs
[i
].range
= dst_reg
->off
;
1979 for (i
= 0; i
< MAX_BPF_STACK
; i
+= BPF_REG_SIZE
) {
1980 if (state
->stack_slot_type
[i
] != STACK_SPILL
)
1982 reg
= &state
->spilled_regs
[i
/ BPF_REG_SIZE
];
1983 if (reg
->type
== PTR_TO_PACKET
&& reg
->id
== dst_reg
->id
)
1984 reg
->range
= dst_reg
->off
;
1988 /* Adjusts the register min/max values in the case that the dst_reg is the
1989 * variable register that we are working on, and src_reg is a constant or we're
1990 * simply doing a BPF_K check.
1992 static void reg_set_min_max(struct bpf_reg_state
*true_reg
,
1993 struct bpf_reg_state
*false_reg
, u64 val
,
1998 /* If this is false then we know nothing Jon Snow, but if it is
1999 * true then we know for sure.
2001 true_reg
->max_value
= true_reg
->min_value
= val
;
2004 /* If this is true we know nothing Jon Snow, but if it is false
2005 * we know the value for sure;
2007 false_reg
->max_value
= false_reg
->min_value
= val
;
2010 /* Unsigned comparison, the minimum value is 0. */
2011 false_reg
->min_value
= 0;
2014 /* If this is false then we know the maximum val is val,
2015 * otherwise we know the min val is val+1.
2017 false_reg
->max_value
= val
;
2018 true_reg
->min_value
= val
+ 1;
2021 /* Unsigned comparison, the minimum value is 0. */
2022 false_reg
->min_value
= 0;
2025 /* If this is false then we know the maximum value is val - 1,
2026 * otherwise we know the mimimum value is val.
2028 false_reg
->max_value
= val
- 1;
2029 true_reg
->min_value
= val
;
2035 check_reg_overflow(false_reg
);
2036 check_reg_overflow(true_reg
);
2039 /* Same as above, but for the case that dst_reg is a CONST_IMM reg and src_reg
2040 * is the variable reg.
2042 static void reg_set_min_max_inv(struct bpf_reg_state
*true_reg
,
2043 struct bpf_reg_state
*false_reg
, u64 val
,
2048 /* If this is false then we know nothing Jon Snow, but if it is
2049 * true then we know for sure.
2051 true_reg
->max_value
= true_reg
->min_value
= val
;
2054 /* If this is true we know nothing Jon Snow, but if it is false
2055 * we know the value for sure;
2057 false_reg
->max_value
= false_reg
->min_value
= val
;
2060 /* Unsigned comparison, the minimum value is 0. */
2061 true_reg
->min_value
= 0;
2065 * If this is false, then the val is <= the register, if it is
2066 * true the register <= to the val.
2068 false_reg
->min_value
= val
;
2069 true_reg
->max_value
= val
- 1;
2072 /* Unsigned comparison, the minimum value is 0. */
2073 true_reg
->min_value
= 0;
2076 /* If this is false then constant < register, if it is true then
2077 * the register < constant.
2079 false_reg
->min_value
= val
+ 1;
2080 true_reg
->max_value
= val
;
2086 check_reg_overflow(false_reg
);
2087 check_reg_overflow(true_reg
);
2090 static void mark_map_reg(struct bpf_reg_state
*regs
, u32 regno
, u32 id
,
2091 enum bpf_reg_type type
)
2093 struct bpf_reg_state
*reg
= ®s
[regno
];
2095 if (reg
->type
== PTR_TO_MAP_VALUE_OR_NULL
&& reg
->id
== id
) {
2097 /* We don't need id from this point onwards anymore, thus we
2098 * should better reset it, so that state pruning has chances
2102 if (type
== UNKNOWN_VALUE
)
2103 __mark_reg_unknown_value(regs
, regno
);
2107 /* The logic is similar to find_good_pkt_pointers(), both could eventually
2108 * be folded together at some point.
2110 static void mark_map_regs(struct bpf_verifier_state
*state
, u32 regno
,
2111 enum bpf_reg_type type
)
2113 struct bpf_reg_state
*regs
= state
->regs
;
2114 u32 id
= regs
[regno
].id
;
2117 for (i
= 0; i
< MAX_BPF_REG
; i
++)
2118 mark_map_reg(regs
, i
, id
, type
);
2120 for (i
= 0; i
< MAX_BPF_STACK
; i
+= BPF_REG_SIZE
) {
2121 if (state
->stack_slot_type
[i
] != STACK_SPILL
)
2123 mark_map_reg(state
->spilled_regs
, i
/ BPF_REG_SIZE
, id
, type
);
2127 static int check_cond_jmp_op(struct bpf_verifier_env
*env
,
2128 struct bpf_insn
*insn
, int *insn_idx
)
2130 struct bpf_verifier_state
*other_branch
, *this_branch
= &env
->cur_state
;
2131 struct bpf_reg_state
*regs
= this_branch
->regs
, *dst_reg
;
2132 u8 opcode
= BPF_OP(insn
->code
);
2135 if (opcode
> BPF_EXIT
) {
2136 verbose("invalid BPF_JMP opcode %x\n", opcode
);
2140 if (BPF_SRC(insn
->code
) == BPF_X
) {
2141 if (insn
->imm
!= 0) {
2142 verbose("BPF_JMP uses reserved fields\n");
2146 /* check src1 operand */
2147 err
= check_reg_arg(regs
, insn
->src_reg
, SRC_OP
);
2151 if (is_pointer_value(env
, insn
->src_reg
)) {
2152 verbose("R%d pointer comparison prohibited\n",
2157 if (insn
->src_reg
!= BPF_REG_0
) {
2158 verbose("BPF_JMP uses reserved fields\n");
2163 /* check src2 operand */
2164 err
= check_reg_arg(regs
, insn
->dst_reg
, SRC_OP
);
2168 dst_reg
= ®s
[insn
->dst_reg
];
2170 /* detect if R == 0 where R was initialized to zero earlier */
2171 if (BPF_SRC(insn
->code
) == BPF_K
&&
2172 (opcode
== BPF_JEQ
|| opcode
== BPF_JNE
) &&
2173 dst_reg
->type
== CONST_IMM
&& dst_reg
->imm
== insn
->imm
) {
2174 if (opcode
== BPF_JEQ
) {
2175 /* if (imm == imm) goto pc+off;
2176 * only follow the goto, ignore fall-through
2178 *insn_idx
+= insn
->off
;
2181 /* if (imm != imm) goto pc+off;
2182 * only follow fall-through branch, since
2183 * that's where the program will go
2189 other_branch
= push_stack(env
, *insn_idx
+ insn
->off
+ 1, *insn_idx
);
2193 /* detect if we are comparing against a constant value so we can adjust
2194 * our min/max values for our dst register.
2196 if (BPF_SRC(insn
->code
) == BPF_X
) {
2197 if (regs
[insn
->src_reg
].type
== CONST_IMM
)
2198 reg_set_min_max(&other_branch
->regs
[insn
->dst_reg
],
2199 dst_reg
, regs
[insn
->src_reg
].imm
,
2201 else if (dst_reg
->type
== CONST_IMM
)
2202 reg_set_min_max_inv(&other_branch
->regs
[insn
->src_reg
],
2203 ®s
[insn
->src_reg
], dst_reg
->imm
,
2206 reg_set_min_max(&other_branch
->regs
[insn
->dst_reg
],
2207 dst_reg
, insn
->imm
, opcode
);
2210 /* detect if R == 0 where R is returned from bpf_map_lookup_elem() */
2211 if (BPF_SRC(insn
->code
) == BPF_K
&&
2212 insn
->imm
== 0 && (opcode
== BPF_JEQ
|| opcode
== BPF_JNE
) &&
2213 dst_reg
->type
== PTR_TO_MAP_VALUE_OR_NULL
) {
2214 /* Mark all identical map registers in each branch as either
2215 * safe or unknown depending R == 0 or R != 0 conditional.
2217 mark_map_regs(this_branch
, insn
->dst_reg
,
2218 opcode
== BPF_JEQ
? PTR_TO_MAP_VALUE
: UNKNOWN_VALUE
);
2219 mark_map_regs(other_branch
, insn
->dst_reg
,
2220 opcode
== BPF_JEQ
? UNKNOWN_VALUE
: PTR_TO_MAP_VALUE
);
2221 } else if (BPF_SRC(insn
->code
) == BPF_X
&& opcode
== BPF_JGT
&&
2222 dst_reg
->type
== PTR_TO_PACKET
&&
2223 regs
[insn
->src_reg
].type
== PTR_TO_PACKET_END
) {
2224 find_good_pkt_pointers(this_branch
, dst_reg
);
2225 } else if (BPF_SRC(insn
->code
) == BPF_X
&& opcode
== BPF_JGE
&&
2226 dst_reg
->type
== PTR_TO_PACKET_END
&&
2227 regs
[insn
->src_reg
].type
== PTR_TO_PACKET
) {
2228 find_good_pkt_pointers(other_branch
, ®s
[insn
->src_reg
]);
2229 } else if (is_pointer_value(env
, insn
->dst_reg
)) {
2230 verbose("R%d pointer comparison prohibited\n", insn
->dst_reg
);
2234 print_verifier_state(this_branch
);
2238 /* return the map pointer stored inside BPF_LD_IMM64 instruction */
2239 static struct bpf_map
*ld_imm64_to_map_ptr(struct bpf_insn
*insn
)
2241 u64 imm64
= ((u64
) (u32
) insn
[0].imm
) | ((u64
) (u32
) insn
[1].imm
) << 32;
2243 return (struct bpf_map
*) (unsigned long) imm64
;
2246 /* verify BPF_LD_IMM64 instruction */
2247 static int check_ld_imm(struct bpf_verifier_env
*env
, struct bpf_insn
*insn
)
2249 struct bpf_reg_state
*regs
= env
->cur_state
.regs
;
2252 if (BPF_SIZE(insn
->code
) != BPF_DW
) {
2253 verbose("invalid BPF_LD_IMM insn\n");
2256 if (insn
->off
!= 0) {
2257 verbose("BPF_LD_IMM64 uses reserved fields\n");
2261 err
= check_reg_arg(regs
, insn
->dst_reg
, DST_OP
);
2265 if (insn
->src_reg
== 0) {
2266 u64 imm
= ((u64
)(insn
+ 1)->imm
<< 32) | (u32
)insn
->imm
;
2268 regs
[insn
->dst_reg
].type
= CONST_IMM
;
2269 regs
[insn
->dst_reg
].imm
= imm
;
2273 /* replace_map_fd_with_map_ptr() should have caught bad ld_imm64 */
2274 BUG_ON(insn
->src_reg
!= BPF_PSEUDO_MAP_FD
);
2276 regs
[insn
->dst_reg
].type
= CONST_PTR_TO_MAP
;
2277 regs
[insn
->dst_reg
].map_ptr
= ld_imm64_to_map_ptr(insn
);
2281 static bool may_access_skb(enum bpf_prog_type type
)
2284 case BPF_PROG_TYPE_SOCKET_FILTER
:
2285 case BPF_PROG_TYPE_SCHED_CLS
:
2286 case BPF_PROG_TYPE_SCHED_ACT
:
2293 /* verify safety of LD_ABS|LD_IND instructions:
2294 * - they can only appear in the programs where ctx == skb
2295 * - since they are wrappers of function calls, they scratch R1-R5 registers,
2296 * preserve R6-R9, and store return value into R0
2299 * ctx == skb == R6 == CTX
2302 * SRC == any register
2303 * IMM == 32-bit immediate
2306 * R0 - 8/16/32-bit skb data converted to cpu endianness
2308 static int check_ld_abs(struct bpf_verifier_env
*env
, struct bpf_insn
*insn
)
2310 struct bpf_reg_state
*regs
= env
->cur_state
.regs
;
2311 u8 mode
= BPF_MODE(insn
->code
);
2312 struct bpf_reg_state
*reg
;
2315 if (!may_access_skb(env
->prog
->type
)) {
2316 verbose("BPF_LD_[ABS|IND] instructions not allowed for this program type\n");
2320 if (insn
->dst_reg
!= BPF_REG_0
|| insn
->off
!= 0 ||
2321 BPF_SIZE(insn
->code
) == BPF_DW
||
2322 (mode
== BPF_ABS
&& insn
->src_reg
!= BPF_REG_0
)) {
2323 verbose("BPF_LD_[ABS|IND] uses reserved fields\n");
2327 /* check whether implicit source operand (register R6) is readable */
2328 err
= check_reg_arg(regs
, BPF_REG_6
, SRC_OP
);
2332 if (regs
[BPF_REG_6
].type
!= PTR_TO_CTX
) {
2333 verbose("at the time of BPF_LD_ABS|IND R6 != pointer to skb\n");
2337 if (mode
== BPF_IND
) {
2338 /* check explicit source operand */
2339 err
= check_reg_arg(regs
, insn
->src_reg
, SRC_OP
);
2344 /* reset caller saved regs to unreadable */
2345 for (i
= 0; i
< CALLER_SAVED_REGS
; i
++) {
2346 reg
= regs
+ caller_saved
[i
];
2347 reg
->type
= NOT_INIT
;
2351 /* mark destination R0 register as readable, since it contains
2352 * the value fetched from the packet
2354 regs
[BPF_REG_0
].type
= UNKNOWN_VALUE
;
2358 /* non-recursive DFS pseudo code
2359 * 1 procedure DFS-iterative(G,v):
2360 * 2 label v as discovered
2361 * 3 let S be a stack
2363 * 5 while S is not empty
2365 * 7 if t is what we're looking for:
2367 * 9 for all edges e in G.adjacentEdges(t) do
2368 * 10 if edge e is already labelled
2369 * 11 continue with the next edge
2370 * 12 w <- G.adjacentVertex(t,e)
2371 * 13 if vertex w is not discovered and not explored
2372 * 14 label e as tree-edge
2373 * 15 label w as discovered
2376 * 18 else if vertex w is discovered
2377 * 19 label e as back-edge
2379 * 21 // vertex w is explored
2380 * 22 label e as forward- or cross-edge
2381 * 23 label t as explored
2386 * 0x11 - discovered and fall-through edge labelled
2387 * 0x12 - discovered and fall-through and branch edges labelled
2398 #define STATE_LIST_MARK ((struct bpf_verifier_state_list *) -1L)
2400 static int *insn_stack
; /* stack of insns to process */
2401 static int cur_stack
; /* current stack index */
2402 static int *insn_state
;
2404 /* t, w, e - match pseudo-code above:
2405 * t - index of current instruction
2406 * w - next instruction
2409 static int push_insn(int t
, int w
, int e
, struct bpf_verifier_env
*env
)
2411 if (e
== FALLTHROUGH
&& insn_state
[t
] >= (DISCOVERED
| FALLTHROUGH
))
2414 if (e
== BRANCH
&& insn_state
[t
] >= (DISCOVERED
| BRANCH
))
2417 if (w
< 0 || w
>= env
->prog
->len
) {
2418 verbose("jump out of range from insn %d to %d\n", t
, w
);
2423 /* mark branch target for state pruning */
2424 env
->explored_states
[w
] = STATE_LIST_MARK
;
2426 if (insn_state
[w
] == 0) {
2428 insn_state
[t
] = DISCOVERED
| e
;
2429 insn_state
[w
] = DISCOVERED
;
2430 if (cur_stack
>= env
->prog
->len
)
2432 insn_stack
[cur_stack
++] = w
;
2434 } else if ((insn_state
[w
] & 0xF0) == DISCOVERED
) {
2435 verbose("back-edge from insn %d to %d\n", t
, w
);
2437 } else if (insn_state
[w
] == EXPLORED
) {
2438 /* forward- or cross-edge */
2439 insn_state
[t
] = DISCOVERED
| e
;
2441 verbose("insn state internal bug\n");
2447 /* non-recursive depth-first-search to detect loops in BPF program
2448 * loop == back-edge in directed graph
2450 static int check_cfg(struct bpf_verifier_env
*env
)
2452 struct bpf_insn
*insns
= env
->prog
->insnsi
;
2453 int insn_cnt
= env
->prog
->len
;
2457 insn_state
= kcalloc(insn_cnt
, sizeof(int), GFP_KERNEL
);
2461 insn_stack
= kcalloc(insn_cnt
, sizeof(int), GFP_KERNEL
);
2467 insn_state
[0] = DISCOVERED
; /* mark 1st insn as discovered */
2468 insn_stack
[0] = 0; /* 0 is the first instruction */
2474 t
= insn_stack
[cur_stack
- 1];
2476 if (BPF_CLASS(insns
[t
].code
) == BPF_JMP
) {
2477 u8 opcode
= BPF_OP(insns
[t
].code
);
2479 if (opcode
== BPF_EXIT
) {
2481 } else if (opcode
== BPF_CALL
) {
2482 ret
= push_insn(t
, t
+ 1, FALLTHROUGH
, env
);
2487 if (t
+ 1 < insn_cnt
)
2488 env
->explored_states
[t
+ 1] = STATE_LIST_MARK
;
2489 } else if (opcode
== BPF_JA
) {
2490 if (BPF_SRC(insns
[t
].code
) != BPF_K
) {
2494 /* unconditional jump with single edge */
2495 ret
= push_insn(t
, t
+ insns
[t
].off
+ 1,
2501 /* tell verifier to check for equivalent states
2502 * after every call and jump
2504 if (t
+ 1 < insn_cnt
)
2505 env
->explored_states
[t
+ 1] = STATE_LIST_MARK
;
2507 /* conditional jump with two edges */
2508 ret
= push_insn(t
, t
+ 1, FALLTHROUGH
, env
);
2514 ret
= push_insn(t
, t
+ insns
[t
].off
+ 1, BRANCH
, env
);
2521 /* all other non-branch instructions with single
2524 ret
= push_insn(t
, t
+ 1, FALLTHROUGH
, env
);
2532 insn_state
[t
] = EXPLORED
;
2533 if (cur_stack
-- <= 0) {
2534 verbose("pop stack internal bug\n");
2541 for (i
= 0; i
< insn_cnt
; i
++) {
2542 if (insn_state
[i
] != EXPLORED
) {
2543 verbose("unreachable insn %d\n", i
);
2548 ret
= 0; /* cfg looks good */
2556 /* the following conditions reduce the number of explored insns
2557 * from ~140k to ~80k for ultra large programs that use a lot of ptr_to_packet
2559 static bool compare_ptrs_to_packet(struct bpf_reg_state
*old
,
2560 struct bpf_reg_state
*cur
)
2562 if (old
->id
!= cur
->id
)
2565 /* old ptr_to_packet is more conservative, since it allows smaller
2567 * old(off=0,r=10) is equal to cur(off=0,r=20), because
2568 * old(off=0,r=10) means that with range=10 the verifier proceeded
2569 * further and found no issues with the program. Now we're in the same
2570 * spot with cur(off=0,r=20), so we're safe too, since anything further
2571 * will only be looking at most 10 bytes after this pointer.
2573 if (old
->off
== cur
->off
&& old
->range
< cur
->range
)
2576 /* old(off=20,r=10) is equal to cur(off=22,re=22 or 5 or 0)
2577 * since both cannot be used for packet access and safe(old)
2578 * pointer has smaller off that could be used for further
2579 * 'if (ptr > data_end)' check
2581 * old(off=20,r=10) and cur(off=22,r=22) and cur(off=22,r=0) mean
2582 * that we cannot access the packet.
2583 * The safe range is:
2584 * [ptr, ptr + range - off)
2585 * so whenever off >=range, it means no safe bytes from this pointer.
2586 * When comparing old->off <= cur->off, it means that older code
2587 * went with smaller offset and that offset was later
2588 * used to figure out the safe range after 'if (ptr > data_end)' check
2589 * Say, 'old' state was explored like:
2590 * ... R3(off=0, r=0)
2592 * ... now R4(off=20,r=0) <-- here
2593 * if (R4 > data_end)
2594 * ... R4(off=20,r=20), R3(off=0,r=20) and R3 can be used to access.
2595 * ... the code further went all the way to bpf_exit.
2596 * Now the 'cur' state at the mark 'here' has R4(off=30,r=0).
2597 * old_R4(off=20,r=0) equal to cur_R4(off=30,r=0), since if the verifier
2598 * goes further, such cur_R4 will give larger safe packet range after
2599 * 'if (R4 > data_end)' and all further insn were already good with r=20,
2600 * so they will be good with r=30 and we can prune the search.
2602 if (old
->off
<= cur
->off
&&
2603 old
->off
>= old
->range
&& cur
->off
>= cur
->range
)
2609 /* compare two verifier states
2611 * all states stored in state_list are known to be valid, since
2612 * verifier reached 'bpf_exit' instruction through them
2614 * this function is called when verifier exploring different branches of
2615 * execution popped from the state stack. If it sees an old state that has
2616 * more strict register state and more strict stack state then this execution
2617 * branch doesn't need to be explored further, since verifier already
2618 * concluded that more strict state leads to valid finish.
2620 * Therefore two states are equivalent if register state is more conservative
2621 * and explored stack state is more conservative than the current one.
2624 * (slot1=INV slot2=MISC) == (slot1=MISC slot2=MISC)
2625 * (slot1=MISC slot2=MISC) != (slot1=INV slot2=MISC)
2627 * In other words if current stack state (one being explored) has more
2628 * valid slots than old one that already passed validation, it means
2629 * the verifier can stop exploring and conclude that current state is valid too
2631 * Similarly with registers. If explored state has register type as invalid
2632 * whereas register type in current state is meaningful, it means that
2633 * the current state will reach 'bpf_exit' instruction safely
2635 static bool states_equal(struct bpf_verifier_env
*env
,
2636 struct bpf_verifier_state
*old
,
2637 struct bpf_verifier_state
*cur
)
2639 bool varlen_map_access
= env
->varlen_map_value_access
;
2640 struct bpf_reg_state
*rold
, *rcur
;
2643 for (i
= 0; i
< MAX_BPF_REG
; i
++) {
2644 rold
= &old
->regs
[i
];
2645 rcur
= &cur
->regs
[i
];
2647 if (memcmp(rold
, rcur
, sizeof(*rold
)) == 0)
2650 /* If the ranges were not the same, but everything else was and
2651 * we didn't do a variable access into a map then we are a-ok.
2653 if (!varlen_map_access
&&
2654 memcmp(rold
, rcur
, offsetofend(struct bpf_reg_state
, id
)) == 0)
2657 /* If we didn't map access then again we don't care about the
2658 * mismatched range values and it's ok if our old type was
2659 * UNKNOWN and we didn't go to a NOT_INIT'ed reg.
2661 if (rold
->type
== NOT_INIT
||
2662 (!varlen_map_access
&& rold
->type
== UNKNOWN_VALUE
&&
2663 rcur
->type
!= NOT_INIT
))
2666 if (rold
->type
== PTR_TO_PACKET
&& rcur
->type
== PTR_TO_PACKET
&&
2667 compare_ptrs_to_packet(rold
, rcur
))
2673 for (i
= 0; i
< MAX_BPF_STACK
; i
++) {
2674 if (old
->stack_slot_type
[i
] == STACK_INVALID
)
2676 if (old
->stack_slot_type
[i
] != cur
->stack_slot_type
[i
])
2677 /* Ex: old explored (safe) state has STACK_SPILL in
2678 * this stack slot, but current has has STACK_MISC ->
2679 * this verifier states are not equivalent,
2680 * return false to continue verification of this path
2683 if (i
% BPF_REG_SIZE
)
2685 if (memcmp(&old
->spilled_regs
[i
/ BPF_REG_SIZE
],
2686 &cur
->spilled_regs
[i
/ BPF_REG_SIZE
],
2687 sizeof(old
->spilled_regs
[0])))
2688 /* when explored and current stack slot types are
2689 * the same, check that stored pointers types
2690 * are the same as well.
2691 * Ex: explored safe path could have stored
2692 * (bpf_reg_state) {.type = PTR_TO_STACK, .imm = -8}
2693 * but current path has stored:
2694 * (bpf_reg_state) {.type = PTR_TO_STACK, .imm = -16}
2695 * such verifier states are not equivalent.
2696 * return false to continue verification of this path
2705 static int is_state_visited(struct bpf_verifier_env
*env
, int insn_idx
)
2707 struct bpf_verifier_state_list
*new_sl
;
2708 struct bpf_verifier_state_list
*sl
;
2710 sl
= env
->explored_states
[insn_idx
];
2712 /* this 'insn_idx' instruction wasn't marked, so we will not
2713 * be doing state search here
2717 while (sl
!= STATE_LIST_MARK
) {
2718 if (states_equal(env
, &sl
->state
, &env
->cur_state
))
2719 /* reached equivalent register/stack state,
2726 /* there were no equivalent states, remember current one.
2727 * technically the current state is not proven to be safe yet,
2728 * but it will either reach bpf_exit (which means it's safe) or
2729 * it will be rejected. Since there are no loops, we won't be
2730 * seeing this 'insn_idx' instruction again on the way to bpf_exit
2732 new_sl
= kmalloc(sizeof(struct bpf_verifier_state_list
), GFP_USER
);
2736 /* add new state to the head of linked list */
2737 memcpy(&new_sl
->state
, &env
->cur_state
, sizeof(env
->cur_state
));
2738 new_sl
->next
= env
->explored_states
[insn_idx
];
2739 env
->explored_states
[insn_idx
] = new_sl
;
2743 static int ext_analyzer_insn_hook(struct bpf_verifier_env
*env
,
2744 int insn_idx
, int prev_insn_idx
)
2746 if (!env
->analyzer_ops
|| !env
->analyzer_ops
->insn_hook
)
2749 return env
->analyzer_ops
->insn_hook(env
, insn_idx
, prev_insn_idx
);
2752 static int do_check(struct bpf_verifier_env
*env
)
2754 struct bpf_verifier_state
*state
= &env
->cur_state
;
2755 struct bpf_insn
*insns
= env
->prog
->insnsi
;
2756 struct bpf_reg_state
*regs
= state
->regs
;
2757 int insn_cnt
= env
->prog
->len
;
2758 int insn_idx
, prev_insn_idx
= 0;
2759 int insn_processed
= 0;
2760 bool do_print_state
= false;
2762 init_reg_state(regs
);
2764 env
->varlen_map_value_access
= false;
2766 struct bpf_insn
*insn
;
2770 if (insn_idx
>= insn_cnt
) {
2771 verbose("invalid insn idx %d insn_cnt %d\n",
2772 insn_idx
, insn_cnt
);
2776 insn
= &insns
[insn_idx
];
2777 class = BPF_CLASS(insn
->code
);
2779 if (++insn_processed
> BPF_COMPLEXITY_LIMIT_INSNS
) {
2780 verbose("BPF program is too large. Processed %d insn\n",
2785 err
= is_state_visited(env
, insn_idx
);
2789 /* found equivalent state, can prune the search */
2792 verbose("\nfrom %d to %d: safe\n",
2793 prev_insn_idx
, insn_idx
);
2795 verbose("%d: safe\n", insn_idx
);
2797 goto process_bpf_exit
;
2800 if (log_level
&& do_print_state
) {
2801 verbose("\nfrom %d to %d:", prev_insn_idx
, insn_idx
);
2802 print_verifier_state(&env
->cur_state
);
2803 do_print_state
= false;
2807 verbose("%d: ", insn_idx
);
2808 print_bpf_insn(insn
);
2811 err
= ext_analyzer_insn_hook(env
, insn_idx
, prev_insn_idx
);
2815 if (class == BPF_ALU
|| class == BPF_ALU64
) {
2816 err
= check_alu_op(env
, insn
);
2820 } else if (class == BPF_LDX
) {
2821 enum bpf_reg_type
*prev_src_type
, src_reg_type
;
2823 /* check for reserved fields is already done */
2825 /* check src operand */
2826 err
= check_reg_arg(regs
, insn
->src_reg
, SRC_OP
);
2830 err
= check_reg_arg(regs
, insn
->dst_reg
, DST_OP_NO_MARK
);
2834 src_reg_type
= regs
[insn
->src_reg
].type
;
2836 /* check that memory (src_reg + off) is readable,
2837 * the state of dst_reg will be updated by this func
2839 err
= check_mem_access(env
, insn
->src_reg
, insn
->off
,
2840 BPF_SIZE(insn
->code
), BPF_READ
,
2845 if (BPF_SIZE(insn
->code
) != BPF_W
&&
2846 BPF_SIZE(insn
->code
) != BPF_DW
) {
2851 prev_src_type
= &env
->insn_aux_data
[insn_idx
].ptr_type
;
2853 if (*prev_src_type
== NOT_INIT
) {
2855 * dst_reg = *(u32 *)(src_reg + off)
2856 * save type to validate intersecting paths
2858 *prev_src_type
= src_reg_type
;
2860 } else if (src_reg_type
!= *prev_src_type
&&
2861 (src_reg_type
== PTR_TO_CTX
||
2862 *prev_src_type
== PTR_TO_CTX
)) {
2863 /* ABuser program is trying to use the same insn
2864 * dst_reg = *(u32*) (src_reg + off)
2865 * with different pointer types:
2866 * src_reg == ctx in one branch and
2867 * src_reg == stack|map in some other branch.
2870 verbose("same insn cannot be used with different pointers\n");
2874 } else if (class == BPF_STX
) {
2875 enum bpf_reg_type
*prev_dst_type
, dst_reg_type
;
2877 if (BPF_MODE(insn
->code
) == BPF_XADD
) {
2878 err
= check_xadd(env
, insn
);
2885 /* check src1 operand */
2886 err
= check_reg_arg(regs
, insn
->src_reg
, SRC_OP
);
2889 /* check src2 operand */
2890 err
= check_reg_arg(regs
, insn
->dst_reg
, SRC_OP
);
2894 dst_reg_type
= regs
[insn
->dst_reg
].type
;
2896 /* check that memory (dst_reg + off) is writeable */
2897 err
= check_mem_access(env
, insn
->dst_reg
, insn
->off
,
2898 BPF_SIZE(insn
->code
), BPF_WRITE
,
2903 prev_dst_type
= &env
->insn_aux_data
[insn_idx
].ptr_type
;
2905 if (*prev_dst_type
== NOT_INIT
) {
2906 *prev_dst_type
= dst_reg_type
;
2907 } else if (dst_reg_type
!= *prev_dst_type
&&
2908 (dst_reg_type
== PTR_TO_CTX
||
2909 *prev_dst_type
== PTR_TO_CTX
)) {
2910 verbose("same insn cannot be used with different pointers\n");
2914 } else if (class == BPF_ST
) {
2915 if (BPF_MODE(insn
->code
) != BPF_MEM
||
2916 insn
->src_reg
!= BPF_REG_0
) {
2917 verbose("BPF_ST uses reserved fields\n");
2920 /* check src operand */
2921 err
= check_reg_arg(regs
, insn
->dst_reg
, SRC_OP
);
2925 /* check that memory (dst_reg + off) is writeable */
2926 err
= check_mem_access(env
, insn
->dst_reg
, insn
->off
,
2927 BPF_SIZE(insn
->code
), BPF_WRITE
,
2932 } else if (class == BPF_JMP
) {
2933 u8 opcode
= BPF_OP(insn
->code
);
2935 if (opcode
== BPF_CALL
) {
2936 if (BPF_SRC(insn
->code
) != BPF_K
||
2938 insn
->src_reg
!= BPF_REG_0
||
2939 insn
->dst_reg
!= BPF_REG_0
) {
2940 verbose("BPF_CALL uses reserved fields\n");
2944 err
= check_call(env
, insn
->imm
, insn_idx
);
2948 } else if (opcode
== BPF_JA
) {
2949 if (BPF_SRC(insn
->code
) != BPF_K
||
2951 insn
->src_reg
!= BPF_REG_0
||
2952 insn
->dst_reg
!= BPF_REG_0
) {
2953 verbose("BPF_JA uses reserved fields\n");
2957 insn_idx
+= insn
->off
+ 1;
2960 } else if (opcode
== BPF_EXIT
) {
2961 if (BPF_SRC(insn
->code
) != BPF_K
||
2963 insn
->src_reg
!= BPF_REG_0
||
2964 insn
->dst_reg
!= BPF_REG_0
) {
2965 verbose("BPF_EXIT uses reserved fields\n");
2969 /* eBPF calling convetion is such that R0 is used
2970 * to return the value from eBPF program.
2971 * Make sure that it's readable at this time
2972 * of bpf_exit, which means that program wrote
2973 * something into it earlier
2975 err
= check_reg_arg(regs
, BPF_REG_0
, SRC_OP
);
2979 if (is_pointer_value(env
, BPF_REG_0
)) {
2980 verbose("R0 leaks addr as return value\n");
2985 insn_idx
= pop_stack(env
, &prev_insn_idx
);
2989 do_print_state
= true;
2993 err
= check_cond_jmp_op(env
, insn
, &insn_idx
);
2997 } else if (class == BPF_LD
) {
2998 u8 mode
= BPF_MODE(insn
->code
);
3000 if (mode
== BPF_ABS
|| mode
== BPF_IND
) {
3001 err
= check_ld_abs(env
, insn
);
3005 } else if (mode
== BPF_IMM
) {
3006 err
= check_ld_imm(env
, insn
);
3012 verbose("invalid BPF_LD mode\n");
3015 reset_reg_range_values(regs
, insn
->dst_reg
);
3017 verbose("unknown insn class %d\n", class);
3024 verbose("processed %d insns\n", insn_processed
);
3028 static int check_map_prog_compatibility(struct bpf_map
*map
,
3029 struct bpf_prog
*prog
)
3032 if (prog
->type
== BPF_PROG_TYPE_PERF_EVENT
&&
3033 (map
->map_type
== BPF_MAP_TYPE_HASH
||
3034 map
->map_type
== BPF_MAP_TYPE_PERCPU_HASH
) &&
3035 (map
->map_flags
& BPF_F_NO_PREALLOC
)) {
3036 verbose("perf_event programs can only use preallocated hash map\n");
3042 /* look for pseudo eBPF instructions that access map FDs and
3043 * replace them with actual map pointers
3045 static int replace_map_fd_with_map_ptr(struct bpf_verifier_env
*env
)
3047 struct bpf_insn
*insn
= env
->prog
->insnsi
;
3048 int insn_cnt
= env
->prog
->len
;
3051 err
= bpf_prog_calc_tag(env
->prog
);
3055 for (i
= 0; i
< insn_cnt
; i
++, insn
++) {
3056 if (BPF_CLASS(insn
->code
) == BPF_LDX
&&
3057 (BPF_MODE(insn
->code
) != BPF_MEM
|| insn
->imm
!= 0)) {
3058 verbose("BPF_LDX uses reserved fields\n");
3062 if (BPF_CLASS(insn
->code
) == BPF_STX
&&
3063 ((BPF_MODE(insn
->code
) != BPF_MEM
&&
3064 BPF_MODE(insn
->code
) != BPF_XADD
) || insn
->imm
!= 0)) {
3065 verbose("BPF_STX uses reserved fields\n");
3069 if (insn
[0].code
== (BPF_LD
| BPF_IMM
| BPF_DW
)) {
3070 struct bpf_map
*map
;
3073 if (i
== insn_cnt
- 1 || insn
[1].code
!= 0 ||
3074 insn
[1].dst_reg
!= 0 || insn
[1].src_reg
!= 0 ||
3076 verbose("invalid bpf_ld_imm64 insn\n");
3080 if (insn
->src_reg
== 0)
3081 /* valid generic load 64-bit imm */
3084 if (insn
->src_reg
!= BPF_PSEUDO_MAP_FD
) {
3085 verbose("unrecognized bpf_ld_imm64 insn\n");
3089 f
= fdget(insn
->imm
);
3090 map
= __bpf_map_get(f
);
3092 verbose("fd %d is not pointing to valid bpf_map\n",
3094 return PTR_ERR(map
);
3097 err
= check_map_prog_compatibility(map
, env
->prog
);
3103 /* store map pointer inside BPF_LD_IMM64 instruction */
3104 insn
[0].imm
= (u32
) (unsigned long) map
;
3105 insn
[1].imm
= ((u64
) (unsigned long) map
) >> 32;
3107 /* check whether we recorded this map already */
3108 for (j
= 0; j
< env
->used_map_cnt
; j
++)
3109 if (env
->used_maps
[j
] == map
) {
3114 if (env
->used_map_cnt
>= MAX_USED_MAPS
) {
3119 /* hold the map. If the program is rejected by verifier,
3120 * the map will be released by release_maps() or it
3121 * will be used by the valid program until it's unloaded
3122 * and all maps are released in free_bpf_prog_info()
3124 map
= bpf_map_inc(map
, false);
3127 return PTR_ERR(map
);
3129 env
->used_maps
[env
->used_map_cnt
++] = map
;
3138 /* now all pseudo BPF_LD_IMM64 instructions load valid
3139 * 'struct bpf_map *' into a register instead of user map_fd.
3140 * These pointers will be used later by verifier to validate map access.
3145 /* drop refcnt of maps used by the rejected program */
3146 static void release_maps(struct bpf_verifier_env
*env
)
3150 for (i
= 0; i
< env
->used_map_cnt
; i
++)
3151 bpf_map_put(env
->used_maps
[i
]);
3154 /* convert pseudo BPF_LD_IMM64 into generic BPF_LD_IMM64 */
3155 static void convert_pseudo_ld_imm64(struct bpf_verifier_env
*env
)
3157 struct bpf_insn
*insn
= env
->prog
->insnsi
;
3158 int insn_cnt
= env
->prog
->len
;
3161 for (i
= 0; i
< insn_cnt
; i
++, insn
++)
3162 if (insn
->code
== (BPF_LD
| BPF_IMM
| BPF_DW
))
3166 /* single env->prog->insni[off] instruction was replaced with the range
3167 * insni[off, off + cnt). Adjust corresponding insn_aux_data by copying
3168 * [0, off) and [off, end) to new locations, so the patched range stays zero
3170 static int adjust_insn_aux_data(struct bpf_verifier_env
*env
, u32 prog_len
,
3173 struct bpf_insn_aux_data
*new_data
, *old_data
= env
->insn_aux_data
;
3177 new_data
= vzalloc(sizeof(struct bpf_insn_aux_data
) * prog_len
);
3180 memcpy(new_data
, old_data
, sizeof(struct bpf_insn_aux_data
) * off
);
3181 memcpy(new_data
+ off
+ cnt
- 1, old_data
+ off
,
3182 sizeof(struct bpf_insn_aux_data
) * (prog_len
- off
- cnt
+ 1));
3183 env
->insn_aux_data
= new_data
;
3188 static struct bpf_prog
*bpf_patch_insn_data(struct bpf_verifier_env
*env
, u32 off
,
3189 const struct bpf_insn
*patch
, u32 len
)
3191 struct bpf_prog
*new_prog
;
3193 new_prog
= bpf_patch_insn_single(env
->prog
, off
, patch
, len
);
3196 if (adjust_insn_aux_data(env
, new_prog
->len
, off
, len
))
3201 /* convert load instructions that access fields of 'struct __sk_buff'
3202 * into sequence of instructions that access fields of 'struct sk_buff'
3204 static int convert_ctx_accesses(struct bpf_verifier_env
*env
)
3206 const struct bpf_verifier_ops
*ops
= env
->prog
->aux
->ops
;
3207 const int insn_cnt
= env
->prog
->len
;
3208 struct bpf_insn insn_buf
[16], *insn
;
3209 struct bpf_prog
*new_prog
;
3210 enum bpf_access_type type
;
3211 int i
, cnt
, delta
= 0;
3213 if (ops
->gen_prologue
) {
3214 cnt
= ops
->gen_prologue(insn_buf
, env
->seen_direct_write
,
3216 if (cnt
>= ARRAY_SIZE(insn_buf
)) {
3217 verbose("bpf verifier is misconfigured\n");
3220 new_prog
= bpf_patch_insn_data(env
, 0, insn_buf
, cnt
);
3224 env
->prog
= new_prog
;
3229 if (!ops
->convert_ctx_access
)
3232 insn
= env
->prog
->insnsi
+ delta
;
3234 for (i
= 0; i
< insn_cnt
; i
++, insn
++) {
3235 if (insn
->code
== (BPF_LDX
| BPF_MEM
| BPF_B
) ||
3236 insn
->code
== (BPF_LDX
| BPF_MEM
| BPF_H
) ||
3237 insn
->code
== (BPF_LDX
| BPF_MEM
| BPF_W
) ||
3238 insn
->code
== (BPF_LDX
| BPF_MEM
| BPF_DW
))
3240 else if (insn
->code
== (BPF_STX
| BPF_MEM
| BPF_B
) ||
3241 insn
->code
== (BPF_STX
| BPF_MEM
| BPF_H
) ||
3242 insn
->code
== (BPF_STX
| BPF_MEM
| BPF_W
) ||
3243 insn
->code
== (BPF_STX
| BPF_MEM
| BPF_DW
))
3248 if (env
->insn_aux_data
[i
+ delta
].ptr_type
!= PTR_TO_CTX
)
3251 cnt
= ops
->convert_ctx_access(type
, insn
, insn_buf
, env
->prog
);
3252 if (cnt
== 0 || cnt
>= ARRAY_SIZE(insn_buf
)) {
3253 verbose("bpf verifier is misconfigured\n");
3257 new_prog
= bpf_patch_insn_data(env
, i
+ delta
, insn_buf
, cnt
);
3263 /* keep walking new program and skip insns we just inserted */
3264 env
->prog
= new_prog
;
3265 insn
= new_prog
->insnsi
+ i
+ delta
;
3271 /* fixup insn->imm field of bpf_call instructions
3272 * and inline eligible helpers as explicit sequence of BPF instructions
3274 * this function is called after eBPF program passed verification
3276 static int fixup_bpf_calls(struct bpf_verifier_env
*env
)
3278 struct bpf_prog
*prog
= env
->prog
;
3279 struct bpf_insn
*insn
= prog
->insnsi
;
3280 const struct bpf_func_proto
*fn
;
3281 const int insn_cnt
= prog
->len
;
3282 struct bpf_insn insn_buf
[16];
3283 struct bpf_prog
*new_prog
;
3284 struct bpf_map
*map_ptr
;
3285 int i
, cnt
, delta
= 0;
3287 for (i
= 0; i
< insn_cnt
; i
++, insn
++) {
3288 if (insn
->code
!= (BPF_JMP
| BPF_CALL
))
3291 if (insn
->imm
== BPF_FUNC_get_route_realm
)
3292 prog
->dst_needed
= 1;
3293 if (insn
->imm
== BPF_FUNC_get_prandom_u32
)
3294 bpf_user_rnd_init_once();
3295 if (insn
->imm
== BPF_FUNC_xdp_adjust_head
)
3296 prog
->xdp_adjust_head
= 1;
3297 if (insn
->imm
== BPF_FUNC_tail_call
) {
3298 /* mark bpf_tail_call as different opcode to avoid
3299 * conditional branch in the interpeter for every normal
3300 * call and to prevent accidental JITing by JIT compiler
3301 * that doesn't support bpf_tail_call yet
3304 insn
->code
|= BPF_X
;
3308 if (ebpf_jit_enabled() && insn
->imm
== BPF_FUNC_map_lookup_elem
) {
3309 map_ptr
= env
->insn_aux_data
[i
+ delta
].map_ptr
;
3310 if (!map_ptr
->ops
->map_gen_lookup
)
3311 goto patch_call_imm
;
3313 cnt
= map_ptr
->ops
->map_gen_lookup(map_ptr
, insn_buf
);
3314 if (cnt
== 0 || cnt
>= ARRAY_SIZE(insn_buf
)) {
3315 verbose("bpf verifier is misconfigured\n");
3319 new_prog
= bpf_patch_insn_data(env
, i
+ delta
, insn_buf
,
3326 /* keep walking new program and skip insns we just inserted */
3327 env
->prog
= prog
= new_prog
;
3328 insn
= new_prog
->insnsi
+ i
+ delta
;
3333 fn
= prog
->aux
->ops
->get_func_proto(insn
->imm
);
3334 /* all functions that have prototype and verifier allowed
3335 * programs to call them, must be real in-kernel functions
3338 verbose("kernel subsystem misconfigured func %s#%d\n",
3339 func_id_name(insn
->imm
), insn
->imm
);
3342 insn
->imm
= fn
->func
- __bpf_call_base
;
3348 static void free_states(struct bpf_verifier_env
*env
)
3350 struct bpf_verifier_state_list
*sl
, *sln
;
3353 if (!env
->explored_states
)
3356 for (i
= 0; i
< env
->prog
->len
; i
++) {
3357 sl
= env
->explored_states
[i
];
3360 while (sl
!= STATE_LIST_MARK
) {
3367 kfree(env
->explored_states
);
3370 int bpf_check(struct bpf_prog
**prog
, union bpf_attr
*attr
)
3372 char __user
*log_ubuf
= NULL
;
3373 struct bpf_verifier_env
*env
;
3376 /* 'struct bpf_verifier_env' can be global, but since it's not small,
3377 * allocate/free it every time bpf_check() is called
3379 env
= kzalloc(sizeof(struct bpf_verifier_env
), GFP_KERNEL
);
3383 env
->insn_aux_data
= vzalloc(sizeof(struct bpf_insn_aux_data
) *
3386 if (!env
->insn_aux_data
)
3390 /* grab the mutex to protect few globals used by verifier */
3391 mutex_lock(&bpf_verifier_lock
);
3393 if (attr
->log_level
|| attr
->log_buf
|| attr
->log_size
) {
3394 /* user requested verbose verifier output
3395 * and supplied buffer to store the verification trace
3397 log_level
= attr
->log_level
;
3398 log_ubuf
= (char __user
*) (unsigned long) attr
->log_buf
;
3399 log_size
= attr
->log_size
;
3403 /* log_* values have to be sane */
3404 if (log_size
< 128 || log_size
> UINT_MAX
>> 8 ||
3405 log_level
== 0 || log_ubuf
== NULL
)
3409 log_buf
= vmalloc(log_size
);
3416 ret
= replace_map_fd_with_map_ptr(env
);
3418 goto skip_full_check
;
3420 env
->explored_states
= kcalloc(env
->prog
->len
,
3421 sizeof(struct bpf_verifier_state_list
*),
3424 if (!env
->explored_states
)
3425 goto skip_full_check
;
3427 ret
= check_cfg(env
);
3429 goto skip_full_check
;
3431 env
->allow_ptr_leaks
= capable(CAP_SYS_ADMIN
);
3433 ret
= do_check(env
);
3436 while (pop_stack(env
, NULL
) >= 0);
3440 /* program is valid, convert *(u32*)(ctx + off) accesses */
3441 ret
= convert_ctx_accesses(env
);
3444 ret
= fixup_bpf_calls(env
);
3446 if (log_level
&& log_len
>= log_size
- 1) {
3447 BUG_ON(log_len
>= log_size
);
3448 /* verifier log exceeded user supplied buffer */
3450 /* fall through to return what was recorded */
3453 /* copy verifier log back to user space including trailing zero */
3454 if (log_level
&& copy_to_user(log_ubuf
, log_buf
, log_len
+ 1) != 0) {
3459 if (ret
== 0 && env
->used_map_cnt
) {
3460 /* if program passed verifier, update used_maps in bpf_prog_info */
3461 env
->prog
->aux
->used_maps
= kmalloc_array(env
->used_map_cnt
,
3462 sizeof(env
->used_maps
[0]),
3465 if (!env
->prog
->aux
->used_maps
) {
3470 memcpy(env
->prog
->aux
->used_maps
, env
->used_maps
,
3471 sizeof(env
->used_maps
[0]) * env
->used_map_cnt
);
3472 env
->prog
->aux
->used_map_cnt
= env
->used_map_cnt
;
3474 /* program is valid. Convert pseudo bpf_ld_imm64 into generic
3475 * bpf_ld_imm64 instructions
3477 convert_pseudo_ld_imm64(env
);
3483 if (!env
->prog
->aux
->used_maps
)
3484 /* if we didn't copy map pointers into bpf_prog_info, release
3485 * them now. Otherwise free_bpf_prog_info() will release them.
3490 mutex_unlock(&bpf_verifier_lock
);
3491 vfree(env
->insn_aux_data
);
3497 int bpf_analyzer(struct bpf_prog
*prog
, const struct bpf_ext_analyzer_ops
*ops
,
3500 struct bpf_verifier_env
*env
;
3503 env
= kzalloc(sizeof(struct bpf_verifier_env
), GFP_KERNEL
);
3507 env
->insn_aux_data
= vzalloc(sizeof(struct bpf_insn_aux_data
) *
3510 if (!env
->insn_aux_data
)
3513 env
->analyzer_ops
= ops
;
3514 env
->analyzer_priv
= priv
;
3516 /* grab the mutex to protect few globals used by verifier */
3517 mutex_lock(&bpf_verifier_lock
);
3521 env
->explored_states
= kcalloc(env
->prog
->len
,
3522 sizeof(struct bpf_verifier_state_list
*),
3525 if (!env
->explored_states
)
3526 goto skip_full_check
;
3528 ret
= check_cfg(env
);
3530 goto skip_full_check
;
3532 env
->allow_ptr_leaks
= capable(CAP_SYS_ADMIN
);
3534 ret
= do_check(env
);
3537 while (pop_stack(env
, NULL
) >= 0);
3540 mutex_unlock(&bpf_verifier_lock
);
3541 vfree(env
->insn_aux_data
);
3546 EXPORT_SYMBOL_GPL(bpf_analyzer
);