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/filter.h>
18 #include <net/netlink.h>
19 #include <linux/file.h>
20 #include <linux/vmalloc.h>
22 /* bpf_check() is a static code analyzer that walks eBPF program
23 * instruction by instruction and updates register/stack state.
24 * All paths of conditional branches are analyzed until 'bpf_exit' insn.
26 * The first pass is depth-first-search to check that the program is a DAG.
27 * It rejects the following programs:
28 * - larger than BPF_MAXINSNS insns
29 * - if loop is present (detected via back-edge)
30 * - unreachable insns exist (shouldn't be a forest. program = one function)
31 * - out of bounds or malformed jumps
32 * The second pass is all possible path descent from the 1st insn.
33 * Since it's analyzing all pathes through the program, the length of the
34 * analysis is limited to 32k insn, which may be hit even if total number of
35 * insn is less then 4K, but there are too many branches that change stack/regs.
36 * Number of 'branches to be analyzed' is limited to 1k
38 * On entry to each instruction, each register has a type, and the instruction
39 * changes the types of the registers depending on instruction semantics.
40 * If instruction is BPF_MOV64_REG(BPF_REG_1, BPF_REG_5), then type of R5 is
43 * All registers are 64-bit.
44 * R0 - return register
45 * R1-R5 argument passing registers
46 * R6-R9 callee saved registers
47 * R10 - frame pointer read-only
49 * At the start of BPF program the register R1 contains a pointer to bpf_context
50 * and has type PTR_TO_CTX.
52 * Verifier tracks arithmetic operations on pointers in case:
53 * BPF_MOV64_REG(BPF_REG_1, BPF_REG_10),
54 * BPF_ALU64_IMM(BPF_ADD, BPF_REG_1, -20),
55 * 1st insn copies R10 (which has FRAME_PTR) type into R1
56 * and 2nd arithmetic instruction is pattern matched to recognize
57 * that it wants to construct a pointer to some element within stack.
58 * So after 2nd insn, the register R1 has type PTR_TO_STACK
59 * (and -20 constant is saved for further stack bounds checking).
60 * Meaning that this reg is a pointer to stack plus known immediate constant.
62 * Most of the time the registers have UNKNOWN_VALUE type, which
63 * means the register has some value, but it's not a valid pointer.
64 * (like pointer plus pointer becomes UNKNOWN_VALUE type)
66 * When verifier sees load or store instructions the type of base register
67 * can be: PTR_TO_MAP_VALUE, PTR_TO_CTX, FRAME_PTR. These are three pointer
68 * types recognized by check_mem_access() function.
70 * PTR_TO_MAP_VALUE means that this register is pointing to 'map element value'
71 * and the range of [ptr, ptr + map's value_size) is accessible.
73 * registers used to pass values to function calls are checked against
74 * function argument constraints.
76 * ARG_PTR_TO_MAP_KEY is one of such argument constraints.
77 * It means that the register type passed to this function must be
78 * PTR_TO_STACK and it will be used inside the function as
79 * 'pointer to map element key'
81 * For example the argument constraints for bpf_map_lookup_elem():
82 * .ret_type = RET_PTR_TO_MAP_VALUE_OR_NULL,
83 * .arg1_type = ARG_CONST_MAP_PTR,
84 * .arg2_type = ARG_PTR_TO_MAP_KEY,
86 * ret_type says that this function returns 'pointer to map elem value or null'
87 * function expects 1st argument to be a const pointer to 'struct bpf_map' and
88 * 2nd argument should be a pointer to stack, which will be used inside
89 * the helper function as a pointer to map element key.
91 * On the kernel side the helper function looks like:
92 * u64 bpf_map_lookup_elem(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5)
94 * struct bpf_map *map = (struct bpf_map *) (unsigned long) r1;
95 * void *key = (void *) (unsigned long) r2;
98 * here kernel can access 'key' and 'map' pointers safely, knowing that
99 * [key, key + map->key_size) bytes are valid and were initialized on
100 * the stack of eBPF program.
103 * Corresponding eBPF program may look like:
104 * BPF_MOV64_REG(BPF_REG_2, BPF_REG_10), // after this insn R2 type is FRAME_PTR
105 * BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, -4), // after this insn R2 type is PTR_TO_STACK
106 * BPF_LD_MAP_FD(BPF_REG_1, map_fd), // after this insn R1 type is CONST_PTR_TO_MAP
107 * BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0, BPF_FUNC_map_lookup_elem),
108 * here verifier looks at prototype of map_lookup_elem() and sees:
109 * .arg1_type == ARG_CONST_MAP_PTR and R1->type == CONST_PTR_TO_MAP, which is ok,
110 * Now verifier knows that this map has key of R1->map_ptr->key_size bytes
112 * Then .arg2_type == ARG_PTR_TO_MAP_KEY and R2->type == PTR_TO_STACK, ok so far,
113 * Now verifier checks that [R2, R2 + map's key_size) are within stack limits
114 * and were initialized prior to this call.
115 * If it's ok, then verifier allows this BPF_CALL insn and looks at
116 * .ret_type which is RET_PTR_TO_MAP_VALUE_OR_NULL, so it sets
117 * R0->type = PTR_TO_MAP_VALUE_OR_NULL which means bpf_map_lookup_elem() function
118 * returns ether pointer to map value or NULL.
120 * When type PTR_TO_MAP_VALUE_OR_NULL passes through 'if (reg != 0) goto +off'
121 * insn, the register holding that pointer in the true branch changes state to
122 * PTR_TO_MAP_VALUE and the same register changes state to CONST_IMM in the false
123 * branch. See check_cond_jmp_op().
125 * After the call R0 is set to return type of the function and registers R1-R5
126 * are set to NOT_INIT to indicate that they are no longer readable.
130 enum bpf_reg_type type
;
132 /* valid when type == CONST_IMM | PTR_TO_STACK | UNKNOWN_VALUE */
135 /* valid when type == PTR_TO_PACKET* */
142 /* valid when type == CONST_PTR_TO_MAP | PTR_TO_MAP_VALUE |
143 * PTR_TO_MAP_VALUE_OR_NULL
145 struct bpf_map
*map_ptr
;
149 enum bpf_stack_slot_type
{
150 STACK_INVALID
, /* nothing was stored in this stack slot */
151 STACK_SPILL
, /* register spilled into stack */
152 STACK_MISC
/* BPF program wrote some data into this slot */
155 #define BPF_REG_SIZE 8 /* size of eBPF register in bytes */
157 /* state of the program:
158 * type of all registers and stack info
160 struct verifier_state
{
161 struct reg_state regs
[MAX_BPF_REG
];
162 u8 stack_slot_type
[MAX_BPF_STACK
];
163 struct reg_state spilled_regs
[MAX_BPF_STACK
/ BPF_REG_SIZE
];
166 /* linked list of verifier states used to prune search */
167 struct verifier_state_list
{
168 struct verifier_state state
;
169 struct verifier_state_list
*next
;
172 /* verifier_state + insn_idx are pushed to stack when branch is encountered */
173 struct verifier_stack_elem
{
174 /* verifer state is 'st'
175 * before processing instruction 'insn_idx'
176 * and after processing instruction 'prev_insn_idx'
178 struct verifier_state st
;
181 struct verifier_stack_elem
*next
;
184 #define MAX_USED_MAPS 64 /* max number of maps accessed by one eBPF program */
186 /* single container for all structs
187 * one verifier_env per bpf_check() call
189 struct verifier_env
{
190 struct bpf_prog
*prog
; /* eBPF program being verified */
191 struct verifier_stack_elem
*head
; /* stack of verifier states to be processed */
192 int stack_size
; /* number of states to be processed */
193 struct verifier_state cur_state
; /* current verifier state */
194 struct verifier_state_list
**explored_states
; /* search pruning optimization */
195 struct bpf_map
*used_maps
[MAX_USED_MAPS
]; /* array of map's used by eBPF program */
196 u32 used_map_cnt
; /* number of used maps */
197 u32 id_gen
; /* used to generate unique reg IDs */
198 bool allow_ptr_leaks
;
201 #define BPF_COMPLEXITY_LIMIT_INSNS 65536
202 #define BPF_COMPLEXITY_LIMIT_STACK 1024
204 struct bpf_call_arg_meta
{
205 struct bpf_map
*map_ptr
;
211 /* verbose verifier prints what it's seeing
212 * bpf_check() is called under lock, so no race to access these global vars
214 static u32 log_level
, log_size
, log_len
;
215 static char *log_buf
;
217 static DEFINE_MUTEX(bpf_verifier_lock
);
219 /* log_level controls verbosity level of eBPF verifier.
220 * verbose() is used to dump the verification trace to the log, so the user
221 * can figure out what's wrong with the program
223 static __printf(1, 2) void verbose(const char *fmt
, ...)
227 if (log_level
== 0 || log_len
>= log_size
- 1)
231 log_len
+= vscnprintf(log_buf
+ log_len
, log_size
- log_len
, fmt
, args
);
235 /* string representation of 'enum bpf_reg_type' */
236 static const char * const reg_type_str
[] = {
238 [UNKNOWN_VALUE
] = "inv",
239 [PTR_TO_CTX
] = "ctx",
240 [CONST_PTR_TO_MAP
] = "map_ptr",
241 [PTR_TO_MAP_VALUE
] = "map_value",
242 [PTR_TO_MAP_VALUE_OR_NULL
] = "map_value_or_null",
244 [PTR_TO_STACK
] = "fp",
246 [PTR_TO_PACKET
] = "pkt",
247 [PTR_TO_PACKET_END
] = "pkt_end",
250 static void print_verifier_state(struct verifier_state
*state
)
252 struct reg_state
*reg
;
256 for (i
= 0; i
< MAX_BPF_REG
; i
++) {
257 reg
= &state
->regs
[i
];
261 verbose(" R%d=%s", i
, reg_type_str
[t
]);
262 if (t
== CONST_IMM
|| t
== PTR_TO_STACK
)
263 verbose("%lld", reg
->imm
);
264 else if (t
== PTR_TO_PACKET
)
265 verbose("(id=%d,off=%d,r=%d)",
266 reg
->id
, reg
->off
, reg
->range
);
267 else if (t
== UNKNOWN_VALUE
&& reg
->imm
)
268 verbose("%lld", reg
->imm
);
269 else if (t
== CONST_PTR_TO_MAP
|| t
== PTR_TO_MAP_VALUE
||
270 t
== PTR_TO_MAP_VALUE_OR_NULL
)
271 verbose("(ks=%d,vs=%d)",
272 reg
->map_ptr
->key_size
,
273 reg
->map_ptr
->value_size
);
275 for (i
= 0; i
< MAX_BPF_STACK
; i
+= BPF_REG_SIZE
) {
276 if (state
->stack_slot_type
[i
] == STACK_SPILL
)
277 verbose(" fp%d=%s", -MAX_BPF_STACK
+ i
,
278 reg_type_str
[state
->spilled_regs
[i
/ BPF_REG_SIZE
].type
]);
283 static const char *const bpf_class_string
[] = {
291 [BPF_ALU64
] = "alu64",
294 static const char *const bpf_alu_string
[16] = {
295 [BPF_ADD
>> 4] = "+=",
296 [BPF_SUB
>> 4] = "-=",
297 [BPF_MUL
>> 4] = "*=",
298 [BPF_DIV
>> 4] = "/=",
299 [BPF_OR
>> 4] = "|=",
300 [BPF_AND
>> 4] = "&=",
301 [BPF_LSH
>> 4] = "<<=",
302 [BPF_RSH
>> 4] = ">>=",
303 [BPF_NEG
>> 4] = "neg",
304 [BPF_MOD
>> 4] = "%=",
305 [BPF_XOR
>> 4] = "^=",
306 [BPF_MOV
>> 4] = "=",
307 [BPF_ARSH
>> 4] = "s>>=",
308 [BPF_END
>> 4] = "endian",
311 static const char *const bpf_ldst_string
[] = {
312 [BPF_W
>> 3] = "u32",
313 [BPF_H
>> 3] = "u16",
315 [BPF_DW
>> 3] = "u64",
318 static const char *const bpf_jmp_string
[16] = {
319 [BPF_JA
>> 4] = "jmp",
320 [BPF_JEQ
>> 4] = "==",
321 [BPF_JGT
>> 4] = ">",
322 [BPF_JGE
>> 4] = ">=",
323 [BPF_JSET
>> 4] = "&",
324 [BPF_JNE
>> 4] = "!=",
325 [BPF_JSGT
>> 4] = "s>",
326 [BPF_JSGE
>> 4] = "s>=",
327 [BPF_CALL
>> 4] = "call",
328 [BPF_EXIT
>> 4] = "exit",
331 static void print_bpf_insn(struct bpf_insn
*insn
)
333 u8
class = BPF_CLASS(insn
->code
);
335 if (class == BPF_ALU
|| class == BPF_ALU64
) {
336 if (BPF_SRC(insn
->code
) == BPF_X
)
337 verbose("(%02x) %sr%d %s %sr%d\n",
338 insn
->code
, class == BPF_ALU
? "(u32) " : "",
340 bpf_alu_string
[BPF_OP(insn
->code
) >> 4],
341 class == BPF_ALU
? "(u32) " : "",
344 verbose("(%02x) %sr%d %s %s%d\n",
345 insn
->code
, class == BPF_ALU
? "(u32) " : "",
347 bpf_alu_string
[BPF_OP(insn
->code
) >> 4],
348 class == BPF_ALU
? "(u32) " : "",
350 } else if (class == BPF_STX
) {
351 if (BPF_MODE(insn
->code
) == BPF_MEM
)
352 verbose("(%02x) *(%s *)(r%d %+d) = r%d\n",
354 bpf_ldst_string
[BPF_SIZE(insn
->code
) >> 3],
356 insn
->off
, insn
->src_reg
);
357 else if (BPF_MODE(insn
->code
) == BPF_XADD
)
358 verbose("(%02x) lock *(%s *)(r%d %+d) += r%d\n",
360 bpf_ldst_string
[BPF_SIZE(insn
->code
) >> 3],
361 insn
->dst_reg
, insn
->off
,
364 verbose("BUG_%02x\n", insn
->code
);
365 } else if (class == BPF_ST
) {
366 if (BPF_MODE(insn
->code
) != BPF_MEM
) {
367 verbose("BUG_st_%02x\n", insn
->code
);
370 verbose("(%02x) *(%s *)(r%d %+d) = %d\n",
372 bpf_ldst_string
[BPF_SIZE(insn
->code
) >> 3],
374 insn
->off
, insn
->imm
);
375 } else if (class == BPF_LDX
) {
376 if (BPF_MODE(insn
->code
) != BPF_MEM
) {
377 verbose("BUG_ldx_%02x\n", insn
->code
);
380 verbose("(%02x) r%d = *(%s *)(r%d %+d)\n",
381 insn
->code
, insn
->dst_reg
,
382 bpf_ldst_string
[BPF_SIZE(insn
->code
) >> 3],
383 insn
->src_reg
, insn
->off
);
384 } else if (class == BPF_LD
) {
385 if (BPF_MODE(insn
->code
) == BPF_ABS
) {
386 verbose("(%02x) r0 = *(%s *)skb[%d]\n",
388 bpf_ldst_string
[BPF_SIZE(insn
->code
) >> 3],
390 } else if (BPF_MODE(insn
->code
) == BPF_IND
) {
391 verbose("(%02x) r0 = *(%s *)skb[r%d + %d]\n",
393 bpf_ldst_string
[BPF_SIZE(insn
->code
) >> 3],
394 insn
->src_reg
, insn
->imm
);
395 } else if (BPF_MODE(insn
->code
) == BPF_IMM
) {
396 verbose("(%02x) r%d = 0x%x\n",
397 insn
->code
, insn
->dst_reg
, insn
->imm
);
399 verbose("BUG_ld_%02x\n", insn
->code
);
402 } else if (class == BPF_JMP
) {
403 u8 opcode
= BPF_OP(insn
->code
);
405 if (opcode
== BPF_CALL
) {
406 verbose("(%02x) call %d\n", insn
->code
, insn
->imm
);
407 } else if (insn
->code
== (BPF_JMP
| BPF_JA
)) {
408 verbose("(%02x) goto pc%+d\n",
409 insn
->code
, insn
->off
);
410 } else if (insn
->code
== (BPF_JMP
| BPF_EXIT
)) {
411 verbose("(%02x) exit\n", insn
->code
);
412 } else if (BPF_SRC(insn
->code
) == BPF_X
) {
413 verbose("(%02x) if r%d %s r%d goto pc%+d\n",
414 insn
->code
, insn
->dst_reg
,
415 bpf_jmp_string
[BPF_OP(insn
->code
) >> 4],
416 insn
->src_reg
, insn
->off
);
418 verbose("(%02x) if r%d %s 0x%x goto pc%+d\n",
419 insn
->code
, insn
->dst_reg
,
420 bpf_jmp_string
[BPF_OP(insn
->code
) >> 4],
421 insn
->imm
, insn
->off
);
424 verbose("(%02x) %s\n", insn
->code
, bpf_class_string
[class]);
428 static int pop_stack(struct verifier_env
*env
, int *prev_insn_idx
)
430 struct verifier_stack_elem
*elem
;
433 if (env
->head
== NULL
)
436 memcpy(&env
->cur_state
, &env
->head
->st
, sizeof(env
->cur_state
));
437 insn_idx
= env
->head
->insn_idx
;
439 *prev_insn_idx
= env
->head
->prev_insn_idx
;
440 elem
= env
->head
->next
;
447 static struct verifier_state
*push_stack(struct verifier_env
*env
, int insn_idx
,
450 struct verifier_stack_elem
*elem
;
452 elem
= kmalloc(sizeof(struct verifier_stack_elem
), GFP_KERNEL
);
456 memcpy(&elem
->st
, &env
->cur_state
, sizeof(env
->cur_state
));
457 elem
->insn_idx
= insn_idx
;
458 elem
->prev_insn_idx
= prev_insn_idx
;
459 elem
->next
= env
->head
;
462 if (env
->stack_size
> BPF_COMPLEXITY_LIMIT_STACK
) {
463 verbose("BPF program is too complex\n");
468 /* pop all elements and return */
469 while (pop_stack(env
, NULL
) >= 0);
473 #define CALLER_SAVED_REGS 6
474 static const int caller_saved
[CALLER_SAVED_REGS
] = {
475 BPF_REG_0
, BPF_REG_1
, BPF_REG_2
, BPF_REG_3
, BPF_REG_4
, BPF_REG_5
478 static void init_reg_state(struct reg_state
*regs
)
482 for (i
= 0; i
< MAX_BPF_REG
; i
++) {
483 regs
[i
].type
= NOT_INIT
;
488 regs
[BPF_REG_FP
].type
= FRAME_PTR
;
490 /* 1st arg to a function */
491 regs
[BPF_REG_1
].type
= PTR_TO_CTX
;
494 static void mark_reg_unknown_value(struct reg_state
*regs
, u32 regno
)
496 BUG_ON(regno
>= MAX_BPF_REG
);
497 regs
[regno
].type
= UNKNOWN_VALUE
;
502 SRC_OP
, /* register is used as source operand */
503 DST_OP
, /* register is used as destination operand */
504 DST_OP_NO_MARK
/* same as above, check only, don't mark */
507 static int check_reg_arg(struct reg_state
*regs
, u32 regno
,
510 if (regno
>= MAX_BPF_REG
) {
511 verbose("R%d is invalid\n", regno
);
516 /* check whether register used as source operand can be read */
517 if (regs
[regno
].type
== NOT_INIT
) {
518 verbose("R%d !read_ok\n", regno
);
522 /* check whether register used as dest operand can be written to */
523 if (regno
== BPF_REG_FP
) {
524 verbose("frame pointer is read only\n");
528 mark_reg_unknown_value(regs
, regno
);
533 static int bpf_size_to_bytes(int bpf_size
)
535 if (bpf_size
== BPF_W
)
537 else if (bpf_size
== BPF_H
)
539 else if (bpf_size
== BPF_B
)
541 else if (bpf_size
== BPF_DW
)
547 static bool is_spillable_regtype(enum bpf_reg_type type
)
550 case PTR_TO_MAP_VALUE
:
551 case PTR_TO_MAP_VALUE_OR_NULL
:
555 case PTR_TO_PACKET_END
:
557 case CONST_PTR_TO_MAP
:
564 /* check_stack_read/write functions track spill/fill of registers,
565 * stack boundary and alignment are checked in check_mem_access()
567 static int check_stack_write(struct verifier_state
*state
, int off
, int size
,
571 /* caller checked that off % size == 0 and -MAX_BPF_STACK <= off < 0,
572 * so it's aligned access and [off, off + size) are within stack limits
575 if (value_regno
>= 0 &&
576 is_spillable_regtype(state
->regs
[value_regno
].type
)) {
578 /* register containing pointer is being spilled into stack */
579 if (size
!= BPF_REG_SIZE
) {
580 verbose("invalid size of register spill\n");
584 /* save register state */
585 state
->spilled_regs
[(MAX_BPF_STACK
+ off
) / BPF_REG_SIZE
] =
586 state
->regs
[value_regno
];
588 for (i
= 0; i
< BPF_REG_SIZE
; i
++)
589 state
->stack_slot_type
[MAX_BPF_STACK
+ off
+ i
] = STACK_SPILL
;
591 /* regular write of data into stack */
592 state
->spilled_regs
[(MAX_BPF_STACK
+ off
) / BPF_REG_SIZE
] =
593 (struct reg_state
) {};
595 for (i
= 0; i
< size
; i
++)
596 state
->stack_slot_type
[MAX_BPF_STACK
+ off
+ i
] = STACK_MISC
;
601 static int check_stack_read(struct verifier_state
*state
, int off
, int size
,
607 slot_type
= &state
->stack_slot_type
[MAX_BPF_STACK
+ off
];
609 if (slot_type
[0] == STACK_SPILL
) {
610 if (size
!= BPF_REG_SIZE
) {
611 verbose("invalid size of register spill\n");
614 for (i
= 1; i
< BPF_REG_SIZE
; i
++) {
615 if (slot_type
[i
] != STACK_SPILL
) {
616 verbose("corrupted spill memory\n");
621 if (value_regno
>= 0)
622 /* restore register state from stack */
623 state
->regs
[value_regno
] =
624 state
->spilled_regs
[(MAX_BPF_STACK
+ off
) / BPF_REG_SIZE
];
627 for (i
= 0; i
< size
; i
++) {
628 if (slot_type
[i
] != STACK_MISC
) {
629 verbose("invalid read from stack off %d+%d size %d\n",
634 if (value_regno
>= 0)
635 /* have read misc data from the stack */
636 mark_reg_unknown_value(state
->regs
, value_regno
);
641 /* check read/write into map element returned by bpf_map_lookup_elem() */
642 static int check_map_access(struct verifier_env
*env
, u32 regno
, int off
,
645 struct bpf_map
*map
= env
->cur_state
.regs
[regno
].map_ptr
;
647 if (off
< 0 || off
+ size
> map
->value_size
) {
648 verbose("invalid access to map value, value_size=%d off=%d size=%d\n",
649 map
->value_size
, off
, size
);
655 #define MAX_PACKET_OFF 0xffff
657 static bool may_write_pkt_data(enum bpf_prog_type type
)
660 case BPF_PROG_TYPE_XDP
:
667 static int check_packet_access(struct verifier_env
*env
, u32 regno
, int off
,
670 struct reg_state
*regs
= env
->cur_state
.regs
;
671 struct reg_state
*reg
= ®s
[regno
];
674 if (off
< 0 || off
+ size
> reg
->range
) {
675 verbose("invalid access to packet, off=%d size=%d, R%d(id=%d,off=%d,r=%d)\n",
676 off
, size
, regno
, reg
->id
, reg
->off
, reg
->range
);
682 /* check access to 'struct bpf_context' fields */
683 static int check_ctx_access(struct verifier_env
*env
, int off
, int size
,
684 enum bpf_access_type t
, enum bpf_reg_type
*reg_type
)
686 if (env
->prog
->aux
->ops
->is_valid_access
&&
687 env
->prog
->aux
->ops
->is_valid_access(off
, size
, t
, reg_type
)) {
688 /* remember the offset of last byte accessed in ctx */
689 if (env
->prog
->aux
->max_ctx_offset
< off
+ size
)
690 env
->prog
->aux
->max_ctx_offset
= off
+ size
;
694 verbose("invalid bpf_context access off=%d size=%d\n", off
, size
);
698 static bool is_pointer_value(struct verifier_env
*env
, int regno
)
700 if (env
->allow_ptr_leaks
)
703 switch (env
->cur_state
.regs
[regno
].type
) {
712 static int check_ptr_alignment(struct verifier_env
*env
, struct reg_state
*reg
,
715 if (reg
->type
!= PTR_TO_PACKET
) {
716 if (off
% size
!= 0) {
717 verbose("misaligned access off %d size %d\n", off
, size
);
724 switch (env
->prog
->type
) {
725 case BPF_PROG_TYPE_SCHED_CLS
:
726 case BPF_PROG_TYPE_SCHED_ACT
:
727 case BPF_PROG_TYPE_XDP
:
730 verbose("verifier is misconfigured\n");
734 if (IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS
))
735 /* misaligned access to packet is ok on x86,arm,arm64 */
738 if (reg
->id
&& size
!= 1) {
739 verbose("Unknown packet alignment. Only byte-sized access allowed\n");
743 /* skb->data is NET_IP_ALIGN-ed */
744 if ((NET_IP_ALIGN
+ reg
->off
+ off
) % size
!= 0) {
745 verbose("misaligned packet access off %d+%d+%d size %d\n",
746 NET_IP_ALIGN
, reg
->off
, off
, size
);
752 /* check whether memory at (regno + off) is accessible for t = (read | write)
753 * if t==write, value_regno is a register which value is stored into memory
754 * if t==read, value_regno is a register which will receive the value from memory
755 * if t==write && value_regno==-1, some unknown value is stored into memory
756 * if t==read && value_regno==-1, don't care what we read from memory
758 static int check_mem_access(struct verifier_env
*env
, u32 regno
, int off
,
759 int bpf_size
, enum bpf_access_type t
,
762 struct verifier_state
*state
= &env
->cur_state
;
763 struct reg_state
*reg
= &state
->regs
[regno
];
766 if (reg
->type
== PTR_TO_STACK
)
769 size
= bpf_size_to_bytes(bpf_size
);
773 err
= check_ptr_alignment(env
, reg
, off
, size
);
777 if (reg
->type
== PTR_TO_MAP_VALUE
) {
778 if (t
== BPF_WRITE
&& value_regno
>= 0 &&
779 is_pointer_value(env
, value_regno
)) {
780 verbose("R%d leaks addr into map\n", value_regno
);
783 err
= check_map_access(env
, regno
, off
, size
);
784 if (!err
&& t
== BPF_READ
&& value_regno
>= 0)
785 mark_reg_unknown_value(state
->regs
, value_regno
);
787 } else if (reg
->type
== PTR_TO_CTX
) {
788 enum bpf_reg_type reg_type
= UNKNOWN_VALUE
;
790 if (t
== BPF_WRITE
&& value_regno
>= 0 &&
791 is_pointer_value(env
, value_regno
)) {
792 verbose("R%d leaks addr into ctx\n", value_regno
);
795 err
= check_ctx_access(env
, off
, size
, t
, ®_type
);
796 if (!err
&& t
== BPF_READ
&& value_regno
>= 0) {
797 mark_reg_unknown_value(state
->regs
, value_regno
);
798 if (env
->allow_ptr_leaks
)
799 /* note that reg.[id|off|range] == 0 */
800 state
->regs
[value_regno
].type
= reg_type
;
803 } else if (reg
->type
== FRAME_PTR
|| reg
->type
== PTR_TO_STACK
) {
804 if (off
>= 0 || off
< -MAX_BPF_STACK
) {
805 verbose("invalid stack off=%d size=%d\n", off
, size
);
808 if (t
== BPF_WRITE
) {
809 if (!env
->allow_ptr_leaks
&&
810 state
->stack_slot_type
[MAX_BPF_STACK
+ off
] == STACK_SPILL
&&
811 size
!= BPF_REG_SIZE
) {
812 verbose("attempt to corrupt spilled pointer on stack\n");
815 err
= check_stack_write(state
, off
, size
, value_regno
);
817 err
= check_stack_read(state
, off
, size
, value_regno
);
819 } else if (state
->regs
[regno
].type
== PTR_TO_PACKET
) {
820 if (t
== BPF_WRITE
&& !may_write_pkt_data(env
->prog
->type
)) {
821 verbose("cannot write into packet\n");
824 if (t
== BPF_WRITE
&& value_regno
>= 0 &&
825 is_pointer_value(env
, value_regno
)) {
826 verbose("R%d leaks addr into packet\n", value_regno
);
829 err
= check_packet_access(env
, regno
, off
, size
);
830 if (!err
&& t
== BPF_READ
&& value_regno
>= 0)
831 mark_reg_unknown_value(state
->regs
, value_regno
);
833 verbose("R%d invalid mem access '%s'\n",
834 regno
, reg_type_str
[reg
->type
]);
838 if (!err
&& size
<= 2 && value_regno
>= 0 && env
->allow_ptr_leaks
&&
839 state
->regs
[value_regno
].type
== UNKNOWN_VALUE
) {
840 /* 1 or 2 byte load zero-extends, determine the number of
841 * zero upper bits. Not doing it fo 4 byte load, since
842 * such values cannot be added to ptr_to_packet anyway.
844 state
->regs
[value_regno
].imm
= 64 - size
* 8;
849 static int check_xadd(struct verifier_env
*env
, struct bpf_insn
*insn
)
851 struct reg_state
*regs
= env
->cur_state
.regs
;
854 if ((BPF_SIZE(insn
->code
) != BPF_W
&& BPF_SIZE(insn
->code
) != BPF_DW
) ||
856 verbose("BPF_XADD uses reserved fields\n");
860 /* check src1 operand */
861 err
= check_reg_arg(regs
, insn
->src_reg
, SRC_OP
);
865 /* check src2 operand */
866 err
= check_reg_arg(regs
, insn
->dst_reg
, SRC_OP
);
870 /* check whether atomic_add can read the memory */
871 err
= check_mem_access(env
, insn
->dst_reg
, insn
->off
,
872 BPF_SIZE(insn
->code
), BPF_READ
, -1);
876 /* check whether atomic_add can write into the same memory */
877 return check_mem_access(env
, insn
->dst_reg
, insn
->off
,
878 BPF_SIZE(insn
->code
), BPF_WRITE
, -1);
881 /* when register 'regno' is passed into function that will read 'access_size'
882 * bytes from that pointer, make sure that it's within stack boundary
883 * and all elements of stack are initialized
885 static int check_stack_boundary(struct verifier_env
*env
, int regno
,
886 int access_size
, bool zero_size_allowed
,
887 struct bpf_call_arg_meta
*meta
)
889 struct verifier_state
*state
= &env
->cur_state
;
890 struct reg_state
*regs
= state
->regs
;
893 if (regs
[regno
].type
!= PTR_TO_STACK
) {
894 if (zero_size_allowed
&& access_size
== 0 &&
895 regs
[regno
].type
== CONST_IMM
&&
896 regs
[regno
].imm
== 0)
899 verbose("R%d type=%s expected=%s\n", regno
,
900 reg_type_str
[regs
[regno
].type
],
901 reg_type_str
[PTR_TO_STACK
]);
905 off
= regs
[regno
].imm
;
906 if (off
>= 0 || off
< -MAX_BPF_STACK
|| off
+ access_size
> 0 ||
908 verbose("invalid stack type R%d off=%d access_size=%d\n",
909 regno
, off
, access_size
);
913 if (meta
&& meta
->raw_mode
) {
914 meta
->access_size
= access_size
;
919 for (i
= 0; i
< access_size
; i
++) {
920 if (state
->stack_slot_type
[MAX_BPF_STACK
+ off
+ i
] != STACK_MISC
) {
921 verbose("invalid indirect read from stack off %d+%d size %d\n",
922 off
, i
, access_size
);
929 static int check_func_arg(struct verifier_env
*env
, u32 regno
,
930 enum bpf_arg_type arg_type
,
931 struct bpf_call_arg_meta
*meta
)
933 struct reg_state
*regs
= env
->cur_state
.regs
, *reg
= ®s
[regno
];
934 enum bpf_reg_type expected_type
, type
= reg
->type
;
937 if (arg_type
== ARG_DONTCARE
)
940 if (type
== NOT_INIT
) {
941 verbose("R%d !read_ok\n", regno
);
945 if (arg_type
== ARG_ANYTHING
) {
946 if (is_pointer_value(env
, regno
)) {
947 verbose("R%d leaks addr into helper function\n", regno
);
953 if (type
== PTR_TO_PACKET
&& !may_write_pkt_data(env
->prog
->type
)) {
954 verbose("helper access to the packet is not allowed for clsact\n");
958 if (arg_type
== ARG_PTR_TO_MAP_KEY
||
959 arg_type
== ARG_PTR_TO_MAP_VALUE
) {
960 expected_type
= PTR_TO_STACK
;
961 if (type
!= PTR_TO_PACKET
&& type
!= expected_type
)
963 } else if (arg_type
== ARG_CONST_STACK_SIZE
||
964 arg_type
== ARG_CONST_STACK_SIZE_OR_ZERO
) {
965 expected_type
= CONST_IMM
;
966 if (type
!= expected_type
)
968 } else if (arg_type
== ARG_CONST_MAP_PTR
) {
969 expected_type
= CONST_PTR_TO_MAP
;
970 if (type
!= expected_type
)
972 } else if (arg_type
== ARG_PTR_TO_CTX
) {
973 expected_type
= PTR_TO_CTX
;
974 if (type
!= expected_type
)
976 } else if (arg_type
== ARG_PTR_TO_STACK
||
977 arg_type
== ARG_PTR_TO_RAW_STACK
) {
978 expected_type
= PTR_TO_STACK
;
979 /* One exception here. In case function allows for NULL to be
980 * passed in as argument, it's a CONST_IMM type. Final test
981 * happens during stack boundary checking.
983 if (type
== CONST_IMM
&& reg
->imm
== 0)
984 /* final test in check_stack_boundary() */;
985 else if (type
!= PTR_TO_PACKET
&& type
!= expected_type
)
987 meta
->raw_mode
= arg_type
== ARG_PTR_TO_RAW_STACK
;
989 verbose("unsupported arg_type %d\n", arg_type
);
993 if (arg_type
== ARG_CONST_MAP_PTR
) {
994 /* bpf_map_xxx(map_ptr) call: remember that map_ptr */
995 meta
->map_ptr
= reg
->map_ptr
;
996 } else if (arg_type
== ARG_PTR_TO_MAP_KEY
) {
997 /* bpf_map_xxx(..., map_ptr, ..., key) call:
998 * check that [key, key + map->key_size) are within
999 * stack limits and initialized
1001 if (!meta
->map_ptr
) {
1002 /* in function declaration map_ptr must come before
1003 * map_key, so that it's verified and known before
1004 * we have to check map_key here. Otherwise it means
1005 * that kernel subsystem misconfigured verifier
1007 verbose("invalid map_ptr to access map->key\n");
1010 if (type
== PTR_TO_PACKET
)
1011 err
= check_packet_access(env
, regno
, 0,
1012 meta
->map_ptr
->key_size
);
1014 err
= check_stack_boundary(env
, regno
,
1015 meta
->map_ptr
->key_size
,
1017 } else if (arg_type
== ARG_PTR_TO_MAP_VALUE
) {
1018 /* bpf_map_xxx(..., map_ptr, ..., value) call:
1019 * check [value, value + map->value_size) validity
1021 if (!meta
->map_ptr
) {
1022 /* kernel subsystem misconfigured verifier */
1023 verbose("invalid map_ptr to access map->value\n");
1026 if (type
== PTR_TO_PACKET
)
1027 err
= check_packet_access(env
, regno
, 0,
1028 meta
->map_ptr
->value_size
);
1030 err
= check_stack_boundary(env
, regno
,
1031 meta
->map_ptr
->value_size
,
1033 } else if (arg_type
== ARG_CONST_STACK_SIZE
||
1034 arg_type
== ARG_CONST_STACK_SIZE_OR_ZERO
) {
1035 bool zero_size_allowed
= (arg_type
== ARG_CONST_STACK_SIZE_OR_ZERO
);
1037 /* bpf_xxx(..., buf, len) call will access 'len' bytes
1038 * from stack pointer 'buf'. Check it
1039 * note: regno == len, regno - 1 == buf
1042 /* kernel subsystem misconfigured verifier */
1043 verbose("ARG_CONST_STACK_SIZE cannot be first argument\n");
1046 if (regs
[regno
- 1].type
== PTR_TO_PACKET
)
1047 err
= check_packet_access(env
, regno
- 1, 0, reg
->imm
);
1049 err
= check_stack_boundary(env
, regno
- 1, reg
->imm
,
1050 zero_size_allowed
, meta
);
1055 verbose("R%d type=%s expected=%s\n", regno
,
1056 reg_type_str
[type
], reg_type_str
[expected_type
]);
1060 static int check_map_func_compatibility(struct bpf_map
*map
, int func_id
)
1065 /* We need a two way check, first is from map perspective ... */
1066 switch (map
->map_type
) {
1067 case BPF_MAP_TYPE_PROG_ARRAY
:
1068 if (func_id
!= BPF_FUNC_tail_call
)
1071 case BPF_MAP_TYPE_PERF_EVENT_ARRAY
:
1072 if (func_id
!= BPF_FUNC_perf_event_read
&&
1073 func_id
!= BPF_FUNC_perf_event_output
)
1076 case BPF_MAP_TYPE_STACK_TRACE
:
1077 if (func_id
!= BPF_FUNC_get_stackid
)
1080 case BPF_MAP_TYPE_CGROUP_ARRAY
:
1081 if (func_id
!= BPF_FUNC_skb_under_cgroup
&&
1082 func_id
!= BPF_FUNC_current_task_under_cgroup
)
1089 /* ... and second from the function itself. */
1091 case BPF_FUNC_tail_call
:
1092 if (map
->map_type
!= BPF_MAP_TYPE_PROG_ARRAY
)
1095 case BPF_FUNC_perf_event_read
:
1096 case BPF_FUNC_perf_event_output
:
1097 if (map
->map_type
!= BPF_MAP_TYPE_PERF_EVENT_ARRAY
)
1100 case BPF_FUNC_get_stackid
:
1101 if (map
->map_type
!= BPF_MAP_TYPE_STACK_TRACE
)
1104 case BPF_FUNC_current_task_under_cgroup
:
1105 case BPF_FUNC_skb_under_cgroup
:
1106 if (map
->map_type
!= BPF_MAP_TYPE_CGROUP_ARRAY
)
1115 verbose("cannot pass map_type %d into func %d\n",
1116 map
->map_type
, func_id
);
1120 static int check_raw_mode(const struct bpf_func_proto
*fn
)
1124 if (fn
->arg1_type
== ARG_PTR_TO_RAW_STACK
)
1126 if (fn
->arg2_type
== ARG_PTR_TO_RAW_STACK
)
1128 if (fn
->arg3_type
== ARG_PTR_TO_RAW_STACK
)
1130 if (fn
->arg4_type
== ARG_PTR_TO_RAW_STACK
)
1132 if (fn
->arg5_type
== ARG_PTR_TO_RAW_STACK
)
1135 return count
> 1 ? -EINVAL
: 0;
1138 static void clear_all_pkt_pointers(struct verifier_env
*env
)
1140 struct verifier_state
*state
= &env
->cur_state
;
1141 struct reg_state
*regs
= state
->regs
, *reg
;
1144 for (i
= 0; i
< MAX_BPF_REG
; i
++)
1145 if (regs
[i
].type
== PTR_TO_PACKET
||
1146 regs
[i
].type
== PTR_TO_PACKET_END
)
1147 mark_reg_unknown_value(regs
, i
);
1149 for (i
= 0; i
< MAX_BPF_STACK
; i
+= BPF_REG_SIZE
) {
1150 if (state
->stack_slot_type
[i
] != STACK_SPILL
)
1152 reg
= &state
->spilled_regs
[i
/ BPF_REG_SIZE
];
1153 if (reg
->type
!= PTR_TO_PACKET
&&
1154 reg
->type
!= PTR_TO_PACKET_END
)
1156 reg
->type
= UNKNOWN_VALUE
;
1161 static int check_call(struct verifier_env
*env
, int func_id
)
1163 struct verifier_state
*state
= &env
->cur_state
;
1164 const struct bpf_func_proto
*fn
= NULL
;
1165 struct reg_state
*regs
= state
->regs
;
1166 struct reg_state
*reg
;
1167 struct bpf_call_arg_meta meta
;
1171 /* find function prototype */
1172 if (func_id
< 0 || func_id
>= __BPF_FUNC_MAX_ID
) {
1173 verbose("invalid func %d\n", func_id
);
1177 if (env
->prog
->aux
->ops
->get_func_proto
)
1178 fn
= env
->prog
->aux
->ops
->get_func_proto(func_id
);
1181 verbose("unknown func %d\n", func_id
);
1185 /* eBPF programs must be GPL compatible to use GPL-ed functions */
1186 if (!env
->prog
->gpl_compatible
&& fn
->gpl_only
) {
1187 verbose("cannot call GPL only function from proprietary program\n");
1191 changes_data
= bpf_helper_changes_skb_data(fn
->func
);
1193 memset(&meta
, 0, sizeof(meta
));
1195 /* We only support one arg being in raw mode at the moment, which
1196 * is sufficient for the helper functions we have right now.
1198 err
= check_raw_mode(fn
);
1200 verbose("kernel subsystem misconfigured func %d\n", func_id
);
1205 err
= check_func_arg(env
, BPF_REG_1
, fn
->arg1_type
, &meta
);
1208 err
= check_func_arg(env
, BPF_REG_2
, fn
->arg2_type
, &meta
);
1211 err
= check_func_arg(env
, BPF_REG_3
, fn
->arg3_type
, &meta
);
1214 err
= check_func_arg(env
, BPF_REG_4
, fn
->arg4_type
, &meta
);
1217 err
= check_func_arg(env
, BPF_REG_5
, fn
->arg5_type
, &meta
);
1221 /* Mark slots with STACK_MISC in case of raw mode, stack offset
1222 * is inferred from register state.
1224 for (i
= 0; i
< meta
.access_size
; i
++) {
1225 err
= check_mem_access(env
, meta
.regno
, i
, BPF_B
, BPF_WRITE
, -1);
1230 /* reset caller saved regs */
1231 for (i
= 0; i
< CALLER_SAVED_REGS
; i
++) {
1232 reg
= regs
+ caller_saved
[i
];
1233 reg
->type
= NOT_INIT
;
1237 /* update return register */
1238 if (fn
->ret_type
== RET_INTEGER
) {
1239 regs
[BPF_REG_0
].type
= UNKNOWN_VALUE
;
1240 } else if (fn
->ret_type
== RET_VOID
) {
1241 regs
[BPF_REG_0
].type
= NOT_INIT
;
1242 } else if (fn
->ret_type
== RET_PTR_TO_MAP_VALUE_OR_NULL
) {
1243 regs
[BPF_REG_0
].type
= PTR_TO_MAP_VALUE_OR_NULL
;
1244 /* remember map_ptr, so that check_map_access()
1245 * can check 'value_size' boundary of memory access
1246 * to map element returned from bpf_map_lookup_elem()
1248 if (meta
.map_ptr
== NULL
) {
1249 verbose("kernel subsystem misconfigured verifier\n");
1252 regs
[BPF_REG_0
].map_ptr
= meta
.map_ptr
;
1254 verbose("unknown return type %d of func %d\n",
1255 fn
->ret_type
, func_id
);
1259 err
= check_map_func_compatibility(meta
.map_ptr
, func_id
);
1264 clear_all_pkt_pointers(env
);
1268 static int check_packet_ptr_add(struct verifier_env
*env
, struct bpf_insn
*insn
)
1270 struct reg_state
*regs
= env
->cur_state
.regs
;
1271 struct reg_state
*dst_reg
= ®s
[insn
->dst_reg
];
1272 struct reg_state
*src_reg
= ®s
[insn
->src_reg
];
1273 struct reg_state tmp_reg
;
1276 if (BPF_SRC(insn
->code
) == BPF_K
) {
1277 /* pkt_ptr += imm */
1282 verbose("addition of negative constant to packet pointer is not allowed\n");
1285 if (imm
>= MAX_PACKET_OFF
||
1286 imm
+ dst_reg
->off
>= MAX_PACKET_OFF
) {
1287 verbose("constant %d is too large to add to packet pointer\n",
1291 /* a constant was added to pkt_ptr.
1292 * Remember it while keeping the same 'id'
1294 dst_reg
->off
+= imm
;
1296 if (src_reg
->type
== PTR_TO_PACKET
) {
1297 /* R6=pkt(id=0,off=0,r=62) R7=imm22; r7 += r6 */
1298 tmp_reg
= *dst_reg
; /* save r7 state */
1299 *dst_reg
= *src_reg
; /* copy pkt_ptr state r6 into r7 */
1300 src_reg
= &tmp_reg
; /* pretend it's src_reg state */
1301 /* if the checks below reject it, the copy won't matter,
1302 * since we're rejecting the whole program. If all ok,
1303 * then imm22 state will be added to r7
1304 * and r7 will be pkt(id=0,off=22,r=62) while
1305 * r6 will stay as pkt(id=0,off=0,r=62)
1309 if (src_reg
->type
== CONST_IMM
) {
1310 /* pkt_ptr += reg where reg is known constant */
1314 /* disallow pkt_ptr += reg
1315 * if reg is not uknown_value with guaranteed zero upper bits
1316 * otherwise pkt_ptr may overflow and addition will become
1317 * subtraction which is not allowed
1319 if (src_reg
->type
!= UNKNOWN_VALUE
) {
1320 verbose("cannot add '%s' to ptr_to_packet\n",
1321 reg_type_str
[src_reg
->type
]);
1324 if (src_reg
->imm
< 48) {
1325 verbose("cannot add integer value with %lld upper zero bits to ptr_to_packet\n",
1329 /* dst_reg stays as pkt_ptr type and since some positive
1330 * integer value was added to the pointer, increment its 'id'
1332 dst_reg
->id
= ++env
->id_gen
;
1334 /* something was added to pkt_ptr, set range and off to zero */
1341 static int evaluate_reg_alu(struct verifier_env
*env
, struct bpf_insn
*insn
)
1343 struct reg_state
*regs
= env
->cur_state
.regs
;
1344 struct reg_state
*dst_reg
= ®s
[insn
->dst_reg
];
1345 u8 opcode
= BPF_OP(insn
->code
);
1348 /* for type == UNKNOWN_VALUE:
1349 * imm > 0 -> number of zero upper bits
1350 * imm == 0 -> don't track which is the same as all bits can be non-zero
1353 if (BPF_SRC(insn
->code
) == BPF_X
) {
1354 struct reg_state
*src_reg
= ®s
[insn
->src_reg
];
1356 if (src_reg
->type
== UNKNOWN_VALUE
&& src_reg
->imm
> 0 &&
1357 dst_reg
->imm
&& opcode
== BPF_ADD
) {
1359 * where both have zero upper bits. Adding them
1360 * can only result making one more bit non-zero
1361 * in the larger value.
1362 * Ex. 0xffff (imm=48) + 1 (imm=63) = 0x10000 (imm=47)
1363 * 0xffff (imm=48) + 0xffff = 0x1fffe (imm=47)
1365 dst_reg
->imm
= min(dst_reg
->imm
, src_reg
->imm
);
1369 if (src_reg
->type
== CONST_IMM
&& src_reg
->imm
> 0 &&
1370 dst_reg
->imm
&& opcode
== BPF_ADD
) {
1372 * where dreg has zero upper bits and sreg is const.
1373 * Adding them can only result making one more bit
1374 * non-zero in the larger value.
1376 imm_log2
= __ilog2_u64((long long)src_reg
->imm
);
1377 dst_reg
->imm
= min(dst_reg
->imm
, 63 - imm_log2
);
1381 /* all other cases non supported yet, just mark dst_reg */
1386 /* sign extend 32-bit imm into 64-bit to make sure that
1387 * negative values occupy bit 63. Note ilog2() would have
1388 * been incorrect, since sizeof(insn->imm) == 4
1390 imm_log2
= __ilog2_u64((long long)insn
->imm
);
1392 if (dst_reg
->imm
&& opcode
== BPF_LSH
) {
1394 * if reg was a result of 2 byte load, then its imm == 48
1395 * which means that upper 48 bits are zero and shifting this reg
1396 * left by 4 would mean that upper 44 bits are still zero
1398 dst_reg
->imm
-= insn
->imm
;
1399 } else if (dst_reg
->imm
&& opcode
== BPF_MUL
) {
1401 * if multiplying by 14 subtract 4
1402 * This is conservative calculation of upper zero bits.
1403 * It's not trying to special case insn->imm == 1 or 0 cases
1405 dst_reg
->imm
-= imm_log2
+ 1;
1406 } else if (opcode
== BPF_AND
) {
1408 dst_reg
->imm
= 63 - imm_log2
;
1409 } else if (dst_reg
->imm
&& opcode
== BPF_ADD
) {
1411 dst_reg
->imm
= min(dst_reg
->imm
, 63 - imm_log2
);
1413 } else if (opcode
== BPF_RSH
) {
1415 * which means that after right shift, upper bits will be zero
1416 * note that verifier already checked that
1417 * 0 <= imm < 64 for shift insn
1419 dst_reg
->imm
+= insn
->imm
;
1420 if (unlikely(dst_reg
->imm
> 64))
1421 /* some dumb code did:
1424 * and all bits are zero now */
1427 /* all other alu ops, means that we don't know what will
1428 * happen to the value, mark it with unknown number of zero bits
1433 if (dst_reg
->imm
< 0) {
1434 /* all 64 bits of the register can contain non-zero bits
1435 * and such value cannot be added to ptr_to_packet, since it
1436 * may overflow, mark it as unknown to avoid further eval
1443 static int evaluate_reg_imm_alu(struct verifier_env
*env
, struct bpf_insn
*insn
)
1445 struct reg_state
*regs
= env
->cur_state
.regs
;
1446 struct reg_state
*dst_reg
= ®s
[insn
->dst_reg
];
1447 struct reg_state
*src_reg
= ®s
[insn
->src_reg
];
1448 u8 opcode
= BPF_OP(insn
->code
);
1450 /* dst_reg->type == CONST_IMM here, simulate execution of 'add' insn.
1451 * Don't care about overflow or negative values, just add them
1453 if (opcode
== BPF_ADD
&& BPF_SRC(insn
->code
) == BPF_K
)
1454 dst_reg
->imm
+= insn
->imm
;
1455 else if (opcode
== BPF_ADD
&& BPF_SRC(insn
->code
) == BPF_X
&&
1456 src_reg
->type
== CONST_IMM
)
1457 dst_reg
->imm
+= src_reg
->imm
;
1459 mark_reg_unknown_value(regs
, insn
->dst_reg
);
1463 /* check validity of 32-bit and 64-bit arithmetic operations */
1464 static int check_alu_op(struct verifier_env
*env
, struct bpf_insn
*insn
)
1466 struct reg_state
*regs
= env
->cur_state
.regs
, *dst_reg
;
1467 u8 opcode
= BPF_OP(insn
->code
);
1470 if (opcode
== BPF_END
|| opcode
== BPF_NEG
) {
1471 if (opcode
== BPF_NEG
) {
1472 if (BPF_SRC(insn
->code
) != 0 ||
1473 insn
->src_reg
!= BPF_REG_0
||
1474 insn
->off
!= 0 || insn
->imm
!= 0) {
1475 verbose("BPF_NEG uses reserved fields\n");
1479 if (insn
->src_reg
!= BPF_REG_0
|| insn
->off
!= 0 ||
1480 (insn
->imm
!= 16 && insn
->imm
!= 32 && insn
->imm
!= 64)) {
1481 verbose("BPF_END uses reserved fields\n");
1486 /* check src operand */
1487 err
= check_reg_arg(regs
, insn
->dst_reg
, SRC_OP
);
1491 if (is_pointer_value(env
, insn
->dst_reg
)) {
1492 verbose("R%d pointer arithmetic prohibited\n",
1497 /* check dest operand */
1498 err
= check_reg_arg(regs
, insn
->dst_reg
, DST_OP
);
1502 } else if (opcode
== BPF_MOV
) {
1504 if (BPF_SRC(insn
->code
) == BPF_X
) {
1505 if (insn
->imm
!= 0 || insn
->off
!= 0) {
1506 verbose("BPF_MOV uses reserved fields\n");
1510 /* check src operand */
1511 err
= check_reg_arg(regs
, insn
->src_reg
, SRC_OP
);
1515 if (insn
->src_reg
!= BPF_REG_0
|| insn
->off
!= 0) {
1516 verbose("BPF_MOV uses reserved fields\n");
1521 /* check dest operand */
1522 err
= check_reg_arg(regs
, insn
->dst_reg
, DST_OP
);
1526 if (BPF_SRC(insn
->code
) == BPF_X
) {
1527 if (BPF_CLASS(insn
->code
) == BPF_ALU64
) {
1529 * copy register state to dest reg
1531 regs
[insn
->dst_reg
] = regs
[insn
->src_reg
];
1533 if (is_pointer_value(env
, insn
->src_reg
)) {
1534 verbose("R%d partial copy of pointer\n",
1538 regs
[insn
->dst_reg
].type
= UNKNOWN_VALUE
;
1539 regs
[insn
->dst_reg
].map_ptr
= NULL
;
1543 * remember the value we stored into this reg
1545 regs
[insn
->dst_reg
].type
= CONST_IMM
;
1546 regs
[insn
->dst_reg
].imm
= insn
->imm
;
1549 } else if (opcode
> BPF_END
) {
1550 verbose("invalid BPF_ALU opcode %x\n", opcode
);
1553 } else { /* all other ALU ops: and, sub, xor, add, ... */
1555 if (BPF_SRC(insn
->code
) == BPF_X
) {
1556 if (insn
->imm
!= 0 || insn
->off
!= 0) {
1557 verbose("BPF_ALU uses reserved fields\n");
1560 /* check src1 operand */
1561 err
= check_reg_arg(regs
, insn
->src_reg
, SRC_OP
);
1565 if (insn
->src_reg
!= BPF_REG_0
|| insn
->off
!= 0) {
1566 verbose("BPF_ALU uses reserved fields\n");
1571 /* check src2 operand */
1572 err
= check_reg_arg(regs
, insn
->dst_reg
, SRC_OP
);
1576 if ((opcode
== BPF_MOD
|| opcode
== BPF_DIV
) &&
1577 BPF_SRC(insn
->code
) == BPF_K
&& insn
->imm
== 0) {
1578 verbose("div by zero\n");
1582 if ((opcode
== BPF_LSH
|| opcode
== BPF_RSH
||
1583 opcode
== BPF_ARSH
) && BPF_SRC(insn
->code
) == BPF_K
) {
1584 int size
= BPF_CLASS(insn
->code
) == BPF_ALU64
? 64 : 32;
1586 if (insn
->imm
< 0 || insn
->imm
>= size
) {
1587 verbose("invalid shift %d\n", insn
->imm
);
1592 /* check dest operand */
1593 err
= check_reg_arg(regs
, insn
->dst_reg
, DST_OP_NO_MARK
);
1597 dst_reg
= ®s
[insn
->dst_reg
];
1599 /* pattern match 'bpf_add Rx, imm' instruction */
1600 if (opcode
== BPF_ADD
&& BPF_CLASS(insn
->code
) == BPF_ALU64
&&
1601 dst_reg
->type
== FRAME_PTR
&& BPF_SRC(insn
->code
) == BPF_K
) {
1602 dst_reg
->type
= PTR_TO_STACK
;
1603 dst_reg
->imm
= insn
->imm
;
1605 } else if (opcode
== BPF_ADD
&&
1606 BPF_CLASS(insn
->code
) == BPF_ALU64
&&
1607 (dst_reg
->type
== PTR_TO_PACKET
||
1608 (BPF_SRC(insn
->code
) == BPF_X
&&
1609 regs
[insn
->src_reg
].type
== PTR_TO_PACKET
))) {
1610 /* ptr_to_packet += K|X */
1611 return check_packet_ptr_add(env
, insn
);
1612 } else if (BPF_CLASS(insn
->code
) == BPF_ALU64
&&
1613 dst_reg
->type
== UNKNOWN_VALUE
&&
1614 env
->allow_ptr_leaks
) {
1615 /* unknown += K|X */
1616 return evaluate_reg_alu(env
, insn
);
1617 } else if (BPF_CLASS(insn
->code
) == BPF_ALU64
&&
1618 dst_reg
->type
== CONST_IMM
&&
1619 env
->allow_ptr_leaks
) {
1620 /* reg_imm += K|X */
1621 return evaluate_reg_imm_alu(env
, insn
);
1622 } else if (is_pointer_value(env
, insn
->dst_reg
)) {
1623 verbose("R%d pointer arithmetic prohibited\n",
1626 } else if (BPF_SRC(insn
->code
) == BPF_X
&&
1627 is_pointer_value(env
, insn
->src_reg
)) {
1628 verbose("R%d pointer arithmetic prohibited\n",
1633 /* mark dest operand */
1634 mark_reg_unknown_value(regs
, insn
->dst_reg
);
1640 static void find_good_pkt_pointers(struct verifier_state
*state
,
1641 const struct reg_state
*dst_reg
)
1643 struct reg_state
*regs
= state
->regs
, *reg
;
1646 /* LLVM can generate two kind of checks:
1652 * if (r2 > pkt_end) goto <handle exception>
1656 * r2 == dst_reg, pkt_end == src_reg
1657 * r2=pkt(id=n,off=8,r=0)
1658 * r3=pkt(id=n,off=0,r=0)
1664 * if (pkt_end >= r2) goto <access okay>
1665 * <handle exception>
1668 * pkt_end == dst_reg, r2 == src_reg
1669 * r2=pkt(id=n,off=8,r=0)
1670 * r3=pkt(id=n,off=0,r=0)
1672 * Find register r3 and mark its range as r3=pkt(id=n,off=0,r=8)
1673 * so that range of bytes [r3, r3 + 8) is safe to access.
1676 for (i
= 0; i
< MAX_BPF_REG
; i
++)
1677 if (regs
[i
].type
== PTR_TO_PACKET
&& regs
[i
].id
== dst_reg
->id
)
1678 regs
[i
].range
= dst_reg
->off
;
1680 for (i
= 0; i
< MAX_BPF_STACK
; i
+= BPF_REG_SIZE
) {
1681 if (state
->stack_slot_type
[i
] != STACK_SPILL
)
1683 reg
= &state
->spilled_regs
[i
/ BPF_REG_SIZE
];
1684 if (reg
->type
== PTR_TO_PACKET
&& reg
->id
== dst_reg
->id
)
1685 reg
->range
= dst_reg
->off
;
1689 static int check_cond_jmp_op(struct verifier_env
*env
,
1690 struct bpf_insn
*insn
, int *insn_idx
)
1692 struct verifier_state
*other_branch
, *this_branch
= &env
->cur_state
;
1693 struct reg_state
*regs
= this_branch
->regs
, *dst_reg
;
1694 u8 opcode
= BPF_OP(insn
->code
);
1697 if (opcode
> BPF_EXIT
) {
1698 verbose("invalid BPF_JMP opcode %x\n", opcode
);
1702 if (BPF_SRC(insn
->code
) == BPF_X
) {
1703 if (insn
->imm
!= 0) {
1704 verbose("BPF_JMP uses reserved fields\n");
1708 /* check src1 operand */
1709 err
= check_reg_arg(regs
, insn
->src_reg
, SRC_OP
);
1713 if (is_pointer_value(env
, insn
->src_reg
)) {
1714 verbose("R%d pointer comparison prohibited\n",
1719 if (insn
->src_reg
!= BPF_REG_0
) {
1720 verbose("BPF_JMP uses reserved fields\n");
1725 /* check src2 operand */
1726 err
= check_reg_arg(regs
, insn
->dst_reg
, SRC_OP
);
1730 dst_reg
= ®s
[insn
->dst_reg
];
1732 /* detect if R == 0 where R was initialized to zero earlier */
1733 if (BPF_SRC(insn
->code
) == BPF_K
&&
1734 (opcode
== BPF_JEQ
|| opcode
== BPF_JNE
) &&
1735 dst_reg
->type
== CONST_IMM
&& dst_reg
->imm
== insn
->imm
) {
1736 if (opcode
== BPF_JEQ
) {
1737 /* if (imm == imm) goto pc+off;
1738 * only follow the goto, ignore fall-through
1740 *insn_idx
+= insn
->off
;
1743 /* if (imm != imm) goto pc+off;
1744 * only follow fall-through branch, since
1745 * that's where the program will go
1751 other_branch
= push_stack(env
, *insn_idx
+ insn
->off
+ 1, *insn_idx
);
1755 /* detect if R == 0 where R is returned value from bpf_map_lookup_elem() */
1756 if (BPF_SRC(insn
->code
) == BPF_K
&&
1757 insn
->imm
== 0 && (opcode
== BPF_JEQ
|| opcode
== BPF_JNE
) &&
1758 dst_reg
->type
== PTR_TO_MAP_VALUE_OR_NULL
) {
1759 if (opcode
== BPF_JEQ
) {
1760 /* next fallthrough insn can access memory via
1763 regs
[insn
->dst_reg
].type
= PTR_TO_MAP_VALUE
;
1764 /* branch targer cannot access it, since reg == 0 */
1765 mark_reg_unknown_value(other_branch
->regs
,
1768 other_branch
->regs
[insn
->dst_reg
].type
= PTR_TO_MAP_VALUE
;
1769 mark_reg_unknown_value(regs
, insn
->dst_reg
);
1771 } else if (BPF_SRC(insn
->code
) == BPF_X
&& opcode
== BPF_JGT
&&
1772 dst_reg
->type
== PTR_TO_PACKET
&&
1773 regs
[insn
->src_reg
].type
== PTR_TO_PACKET_END
) {
1774 find_good_pkt_pointers(this_branch
, dst_reg
);
1775 } else if (BPF_SRC(insn
->code
) == BPF_X
&& opcode
== BPF_JGE
&&
1776 dst_reg
->type
== PTR_TO_PACKET_END
&&
1777 regs
[insn
->src_reg
].type
== PTR_TO_PACKET
) {
1778 find_good_pkt_pointers(other_branch
, ®s
[insn
->src_reg
]);
1779 } else if (is_pointer_value(env
, insn
->dst_reg
)) {
1780 verbose("R%d pointer comparison prohibited\n", insn
->dst_reg
);
1784 print_verifier_state(this_branch
);
1788 /* return the map pointer stored inside BPF_LD_IMM64 instruction */
1789 static struct bpf_map
*ld_imm64_to_map_ptr(struct bpf_insn
*insn
)
1791 u64 imm64
= ((u64
) (u32
) insn
[0].imm
) | ((u64
) (u32
) insn
[1].imm
) << 32;
1793 return (struct bpf_map
*) (unsigned long) imm64
;
1796 /* verify BPF_LD_IMM64 instruction */
1797 static int check_ld_imm(struct verifier_env
*env
, struct bpf_insn
*insn
)
1799 struct reg_state
*regs
= env
->cur_state
.regs
;
1802 if (BPF_SIZE(insn
->code
) != BPF_DW
) {
1803 verbose("invalid BPF_LD_IMM insn\n");
1806 if (insn
->off
!= 0) {
1807 verbose("BPF_LD_IMM64 uses reserved fields\n");
1811 err
= check_reg_arg(regs
, insn
->dst_reg
, DST_OP
);
1815 if (insn
->src_reg
== 0)
1816 /* generic move 64-bit immediate into a register */
1819 /* replace_map_fd_with_map_ptr() should have caught bad ld_imm64 */
1820 BUG_ON(insn
->src_reg
!= BPF_PSEUDO_MAP_FD
);
1822 regs
[insn
->dst_reg
].type
= CONST_PTR_TO_MAP
;
1823 regs
[insn
->dst_reg
].map_ptr
= ld_imm64_to_map_ptr(insn
);
1827 static bool may_access_skb(enum bpf_prog_type type
)
1830 case BPF_PROG_TYPE_SOCKET_FILTER
:
1831 case BPF_PROG_TYPE_SCHED_CLS
:
1832 case BPF_PROG_TYPE_SCHED_ACT
:
1839 /* verify safety of LD_ABS|LD_IND instructions:
1840 * - they can only appear in the programs where ctx == skb
1841 * - since they are wrappers of function calls, they scratch R1-R5 registers,
1842 * preserve R6-R9, and store return value into R0
1845 * ctx == skb == R6 == CTX
1848 * SRC == any register
1849 * IMM == 32-bit immediate
1852 * R0 - 8/16/32-bit skb data converted to cpu endianness
1854 static int check_ld_abs(struct verifier_env
*env
, struct bpf_insn
*insn
)
1856 struct reg_state
*regs
= env
->cur_state
.regs
;
1857 u8 mode
= BPF_MODE(insn
->code
);
1858 struct reg_state
*reg
;
1861 if (!may_access_skb(env
->prog
->type
)) {
1862 verbose("BPF_LD_[ABS|IND] instructions not allowed for this program type\n");
1866 if (insn
->dst_reg
!= BPF_REG_0
|| insn
->off
!= 0 ||
1867 BPF_SIZE(insn
->code
) == BPF_DW
||
1868 (mode
== BPF_ABS
&& insn
->src_reg
!= BPF_REG_0
)) {
1869 verbose("BPF_LD_[ABS|IND] uses reserved fields\n");
1873 /* check whether implicit source operand (register R6) is readable */
1874 err
= check_reg_arg(regs
, BPF_REG_6
, SRC_OP
);
1878 if (regs
[BPF_REG_6
].type
!= PTR_TO_CTX
) {
1879 verbose("at the time of BPF_LD_ABS|IND R6 != pointer to skb\n");
1883 if (mode
== BPF_IND
) {
1884 /* check explicit source operand */
1885 err
= check_reg_arg(regs
, insn
->src_reg
, SRC_OP
);
1890 /* reset caller saved regs to unreadable */
1891 for (i
= 0; i
< CALLER_SAVED_REGS
; i
++) {
1892 reg
= regs
+ caller_saved
[i
];
1893 reg
->type
= NOT_INIT
;
1897 /* mark destination R0 register as readable, since it contains
1898 * the value fetched from the packet
1900 regs
[BPF_REG_0
].type
= UNKNOWN_VALUE
;
1904 /* non-recursive DFS pseudo code
1905 * 1 procedure DFS-iterative(G,v):
1906 * 2 label v as discovered
1907 * 3 let S be a stack
1909 * 5 while S is not empty
1911 * 7 if t is what we're looking for:
1913 * 9 for all edges e in G.adjacentEdges(t) do
1914 * 10 if edge e is already labelled
1915 * 11 continue with the next edge
1916 * 12 w <- G.adjacentVertex(t,e)
1917 * 13 if vertex w is not discovered and not explored
1918 * 14 label e as tree-edge
1919 * 15 label w as discovered
1922 * 18 else if vertex w is discovered
1923 * 19 label e as back-edge
1925 * 21 // vertex w is explored
1926 * 22 label e as forward- or cross-edge
1927 * 23 label t as explored
1932 * 0x11 - discovered and fall-through edge labelled
1933 * 0x12 - discovered and fall-through and branch edges labelled
1944 #define STATE_LIST_MARK ((struct verifier_state_list *) -1L)
1946 static int *insn_stack
; /* stack of insns to process */
1947 static int cur_stack
; /* current stack index */
1948 static int *insn_state
;
1950 /* t, w, e - match pseudo-code above:
1951 * t - index of current instruction
1952 * w - next instruction
1955 static int push_insn(int t
, int w
, int e
, struct verifier_env
*env
)
1957 if (e
== FALLTHROUGH
&& insn_state
[t
] >= (DISCOVERED
| FALLTHROUGH
))
1960 if (e
== BRANCH
&& insn_state
[t
] >= (DISCOVERED
| BRANCH
))
1963 if (w
< 0 || w
>= env
->prog
->len
) {
1964 verbose("jump out of range from insn %d to %d\n", t
, w
);
1969 /* mark branch target for state pruning */
1970 env
->explored_states
[w
] = STATE_LIST_MARK
;
1972 if (insn_state
[w
] == 0) {
1974 insn_state
[t
] = DISCOVERED
| e
;
1975 insn_state
[w
] = DISCOVERED
;
1976 if (cur_stack
>= env
->prog
->len
)
1978 insn_stack
[cur_stack
++] = w
;
1980 } else if ((insn_state
[w
] & 0xF0) == DISCOVERED
) {
1981 verbose("back-edge from insn %d to %d\n", t
, w
);
1983 } else if (insn_state
[w
] == EXPLORED
) {
1984 /* forward- or cross-edge */
1985 insn_state
[t
] = DISCOVERED
| e
;
1987 verbose("insn state internal bug\n");
1993 /* non-recursive depth-first-search to detect loops in BPF program
1994 * loop == back-edge in directed graph
1996 static int check_cfg(struct verifier_env
*env
)
1998 struct bpf_insn
*insns
= env
->prog
->insnsi
;
1999 int insn_cnt
= env
->prog
->len
;
2003 insn_state
= kcalloc(insn_cnt
, sizeof(int), GFP_KERNEL
);
2007 insn_stack
= kcalloc(insn_cnt
, sizeof(int), GFP_KERNEL
);
2013 insn_state
[0] = DISCOVERED
; /* mark 1st insn as discovered */
2014 insn_stack
[0] = 0; /* 0 is the first instruction */
2020 t
= insn_stack
[cur_stack
- 1];
2022 if (BPF_CLASS(insns
[t
].code
) == BPF_JMP
) {
2023 u8 opcode
= BPF_OP(insns
[t
].code
);
2025 if (opcode
== BPF_EXIT
) {
2027 } else if (opcode
== BPF_CALL
) {
2028 ret
= push_insn(t
, t
+ 1, FALLTHROUGH
, env
);
2033 if (t
+ 1 < insn_cnt
)
2034 env
->explored_states
[t
+ 1] = STATE_LIST_MARK
;
2035 } else if (opcode
== BPF_JA
) {
2036 if (BPF_SRC(insns
[t
].code
) != BPF_K
) {
2040 /* unconditional jump with single edge */
2041 ret
= push_insn(t
, t
+ insns
[t
].off
+ 1,
2047 /* tell verifier to check for equivalent states
2048 * after every call and jump
2050 if (t
+ 1 < insn_cnt
)
2051 env
->explored_states
[t
+ 1] = STATE_LIST_MARK
;
2053 /* conditional jump with two edges */
2054 ret
= push_insn(t
, t
+ 1, FALLTHROUGH
, env
);
2060 ret
= push_insn(t
, t
+ insns
[t
].off
+ 1, BRANCH
, env
);
2067 /* all other non-branch instructions with single
2070 ret
= push_insn(t
, t
+ 1, FALLTHROUGH
, env
);
2078 insn_state
[t
] = EXPLORED
;
2079 if (cur_stack
-- <= 0) {
2080 verbose("pop stack internal bug\n");
2087 for (i
= 0; i
< insn_cnt
; i
++) {
2088 if (insn_state
[i
] != EXPLORED
) {
2089 verbose("unreachable insn %d\n", i
);
2094 ret
= 0; /* cfg looks good */
2102 /* the following conditions reduce the number of explored insns
2103 * from ~140k to ~80k for ultra large programs that use a lot of ptr_to_packet
2105 static bool compare_ptrs_to_packet(struct reg_state
*old
, struct reg_state
*cur
)
2107 if (old
->id
!= cur
->id
)
2110 /* old ptr_to_packet is more conservative, since it allows smaller
2112 * old(off=0,r=10) is equal to cur(off=0,r=20), because
2113 * old(off=0,r=10) means that with range=10 the verifier proceeded
2114 * further and found no issues with the program. Now we're in the same
2115 * spot with cur(off=0,r=20), so we're safe too, since anything further
2116 * will only be looking at most 10 bytes after this pointer.
2118 if (old
->off
== cur
->off
&& old
->range
< cur
->range
)
2121 /* old(off=20,r=10) is equal to cur(off=22,re=22 or 5 or 0)
2122 * since both cannot be used for packet access and safe(old)
2123 * pointer has smaller off that could be used for further
2124 * 'if (ptr > data_end)' check
2126 * old(off=20,r=10) and cur(off=22,r=22) and cur(off=22,r=0) mean
2127 * that we cannot access the packet.
2128 * The safe range is:
2129 * [ptr, ptr + range - off)
2130 * so whenever off >=range, it means no safe bytes from this pointer.
2131 * When comparing old->off <= cur->off, it means that older code
2132 * went with smaller offset and that offset was later
2133 * used to figure out the safe range after 'if (ptr > data_end)' check
2134 * Say, 'old' state was explored like:
2135 * ... R3(off=0, r=0)
2137 * ... now R4(off=20,r=0) <-- here
2138 * if (R4 > data_end)
2139 * ... R4(off=20,r=20), R3(off=0,r=20) and R3 can be used to access.
2140 * ... the code further went all the way to bpf_exit.
2141 * Now the 'cur' state at the mark 'here' has R4(off=30,r=0).
2142 * old_R4(off=20,r=0) equal to cur_R4(off=30,r=0), since if the verifier
2143 * goes further, such cur_R4 will give larger safe packet range after
2144 * 'if (R4 > data_end)' and all further insn were already good with r=20,
2145 * so they will be good with r=30 and we can prune the search.
2147 if (old
->off
<= cur
->off
&&
2148 old
->off
>= old
->range
&& cur
->off
>= cur
->range
)
2154 /* compare two verifier states
2156 * all states stored in state_list are known to be valid, since
2157 * verifier reached 'bpf_exit' instruction through them
2159 * this function is called when verifier exploring different branches of
2160 * execution popped from the state stack. If it sees an old state that has
2161 * more strict register state and more strict stack state then this execution
2162 * branch doesn't need to be explored further, since verifier already
2163 * concluded that more strict state leads to valid finish.
2165 * Therefore two states are equivalent if register state is more conservative
2166 * and explored stack state is more conservative than the current one.
2169 * (slot1=INV slot2=MISC) == (slot1=MISC slot2=MISC)
2170 * (slot1=MISC slot2=MISC) != (slot1=INV slot2=MISC)
2172 * In other words if current stack state (one being explored) has more
2173 * valid slots than old one that already passed validation, it means
2174 * the verifier can stop exploring and conclude that current state is valid too
2176 * Similarly with registers. If explored state has register type as invalid
2177 * whereas register type in current state is meaningful, it means that
2178 * the current state will reach 'bpf_exit' instruction safely
2180 static bool states_equal(struct verifier_state
*old
, struct verifier_state
*cur
)
2182 struct reg_state
*rold
, *rcur
;
2185 for (i
= 0; i
< MAX_BPF_REG
; i
++) {
2186 rold
= &old
->regs
[i
];
2187 rcur
= &cur
->regs
[i
];
2189 if (memcmp(rold
, rcur
, sizeof(*rold
)) == 0)
2192 if (rold
->type
== NOT_INIT
||
2193 (rold
->type
== UNKNOWN_VALUE
&& rcur
->type
!= NOT_INIT
))
2196 if (rold
->type
== PTR_TO_PACKET
&& rcur
->type
== PTR_TO_PACKET
&&
2197 compare_ptrs_to_packet(rold
, rcur
))
2203 for (i
= 0; i
< MAX_BPF_STACK
; i
++) {
2204 if (old
->stack_slot_type
[i
] == STACK_INVALID
)
2206 if (old
->stack_slot_type
[i
] != cur
->stack_slot_type
[i
])
2207 /* Ex: old explored (safe) state has STACK_SPILL in
2208 * this stack slot, but current has has STACK_MISC ->
2209 * this verifier states are not equivalent,
2210 * return false to continue verification of this path
2213 if (i
% BPF_REG_SIZE
)
2215 if (memcmp(&old
->spilled_regs
[i
/ BPF_REG_SIZE
],
2216 &cur
->spilled_regs
[i
/ BPF_REG_SIZE
],
2217 sizeof(old
->spilled_regs
[0])))
2218 /* when explored and current stack slot types are
2219 * the same, check that stored pointers types
2220 * are the same as well.
2221 * Ex: explored safe path could have stored
2222 * (struct reg_state) {.type = PTR_TO_STACK, .imm = -8}
2223 * but current path has stored:
2224 * (struct reg_state) {.type = PTR_TO_STACK, .imm = -16}
2225 * such verifier states are not equivalent.
2226 * return false to continue verification of this path
2235 static int is_state_visited(struct verifier_env
*env
, int insn_idx
)
2237 struct verifier_state_list
*new_sl
;
2238 struct verifier_state_list
*sl
;
2240 sl
= env
->explored_states
[insn_idx
];
2242 /* this 'insn_idx' instruction wasn't marked, so we will not
2243 * be doing state search here
2247 while (sl
!= STATE_LIST_MARK
) {
2248 if (states_equal(&sl
->state
, &env
->cur_state
))
2249 /* reached equivalent register/stack state,
2256 /* there were no equivalent states, remember current one.
2257 * technically the current state is not proven to be safe yet,
2258 * but it will either reach bpf_exit (which means it's safe) or
2259 * it will be rejected. Since there are no loops, we won't be
2260 * seeing this 'insn_idx' instruction again on the way to bpf_exit
2262 new_sl
= kmalloc(sizeof(struct verifier_state_list
), GFP_USER
);
2266 /* add new state to the head of linked list */
2267 memcpy(&new_sl
->state
, &env
->cur_state
, sizeof(env
->cur_state
));
2268 new_sl
->next
= env
->explored_states
[insn_idx
];
2269 env
->explored_states
[insn_idx
] = new_sl
;
2273 static int do_check(struct verifier_env
*env
)
2275 struct verifier_state
*state
= &env
->cur_state
;
2276 struct bpf_insn
*insns
= env
->prog
->insnsi
;
2277 struct reg_state
*regs
= state
->regs
;
2278 int insn_cnt
= env
->prog
->len
;
2279 int insn_idx
, prev_insn_idx
= 0;
2280 int insn_processed
= 0;
2281 bool do_print_state
= false;
2283 init_reg_state(regs
);
2286 struct bpf_insn
*insn
;
2290 if (insn_idx
>= insn_cnt
) {
2291 verbose("invalid insn idx %d insn_cnt %d\n",
2292 insn_idx
, insn_cnt
);
2296 insn
= &insns
[insn_idx
];
2297 class = BPF_CLASS(insn
->code
);
2299 if (++insn_processed
> BPF_COMPLEXITY_LIMIT_INSNS
) {
2300 verbose("BPF program is too large. Proccessed %d insn\n",
2305 err
= is_state_visited(env
, insn_idx
);
2309 /* found equivalent state, can prune the search */
2312 verbose("\nfrom %d to %d: safe\n",
2313 prev_insn_idx
, insn_idx
);
2315 verbose("%d: safe\n", insn_idx
);
2317 goto process_bpf_exit
;
2320 if (log_level
&& do_print_state
) {
2321 verbose("\nfrom %d to %d:", prev_insn_idx
, insn_idx
);
2322 print_verifier_state(&env
->cur_state
);
2323 do_print_state
= false;
2327 verbose("%d: ", insn_idx
);
2328 print_bpf_insn(insn
);
2331 if (class == BPF_ALU
|| class == BPF_ALU64
) {
2332 err
= check_alu_op(env
, insn
);
2336 } else if (class == BPF_LDX
) {
2337 enum bpf_reg_type src_reg_type
;
2339 /* check for reserved fields is already done */
2341 /* check src operand */
2342 err
= check_reg_arg(regs
, insn
->src_reg
, SRC_OP
);
2346 err
= check_reg_arg(regs
, insn
->dst_reg
, DST_OP_NO_MARK
);
2350 src_reg_type
= regs
[insn
->src_reg
].type
;
2352 /* check that memory (src_reg + off) is readable,
2353 * the state of dst_reg will be updated by this func
2355 err
= check_mem_access(env
, insn
->src_reg
, insn
->off
,
2356 BPF_SIZE(insn
->code
), BPF_READ
,
2361 if (BPF_SIZE(insn
->code
) != BPF_W
&&
2362 BPF_SIZE(insn
->code
) != BPF_DW
) {
2367 if (insn
->imm
== 0) {
2369 * dst_reg = *(u32 *)(src_reg + off)
2370 * use reserved 'imm' field to mark this insn
2372 insn
->imm
= src_reg_type
;
2374 } else if (src_reg_type
!= insn
->imm
&&
2375 (src_reg_type
== PTR_TO_CTX
||
2376 insn
->imm
== PTR_TO_CTX
)) {
2377 /* ABuser program is trying to use the same insn
2378 * dst_reg = *(u32*) (src_reg + off)
2379 * with different pointer types:
2380 * src_reg == ctx in one branch and
2381 * src_reg == stack|map in some other branch.
2384 verbose("same insn cannot be used with different pointers\n");
2388 } else if (class == BPF_STX
) {
2389 enum bpf_reg_type dst_reg_type
;
2391 if (BPF_MODE(insn
->code
) == BPF_XADD
) {
2392 err
= check_xadd(env
, insn
);
2399 /* check src1 operand */
2400 err
= check_reg_arg(regs
, insn
->src_reg
, SRC_OP
);
2403 /* check src2 operand */
2404 err
= check_reg_arg(regs
, insn
->dst_reg
, SRC_OP
);
2408 dst_reg_type
= regs
[insn
->dst_reg
].type
;
2410 /* check that memory (dst_reg + off) is writeable */
2411 err
= check_mem_access(env
, insn
->dst_reg
, insn
->off
,
2412 BPF_SIZE(insn
->code
), BPF_WRITE
,
2417 if (insn
->imm
== 0) {
2418 insn
->imm
= dst_reg_type
;
2419 } else if (dst_reg_type
!= insn
->imm
&&
2420 (dst_reg_type
== PTR_TO_CTX
||
2421 insn
->imm
== PTR_TO_CTX
)) {
2422 verbose("same insn cannot be used with different pointers\n");
2426 } else if (class == BPF_ST
) {
2427 if (BPF_MODE(insn
->code
) != BPF_MEM
||
2428 insn
->src_reg
!= BPF_REG_0
) {
2429 verbose("BPF_ST uses reserved fields\n");
2432 /* check src operand */
2433 err
= check_reg_arg(regs
, insn
->dst_reg
, SRC_OP
);
2437 /* check that memory (dst_reg + off) is writeable */
2438 err
= check_mem_access(env
, insn
->dst_reg
, insn
->off
,
2439 BPF_SIZE(insn
->code
), BPF_WRITE
,
2444 } else if (class == BPF_JMP
) {
2445 u8 opcode
= BPF_OP(insn
->code
);
2447 if (opcode
== BPF_CALL
) {
2448 if (BPF_SRC(insn
->code
) != BPF_K
||
2450 insn
->src_reg
!= BPF_REG_0
||
2451 insn
->dst_reg
!= BPF_REG_0
) {
2452 verbose("BPF_CALL uses reserved fields\n");
2456 err
= check_call(env
, insn
->imm
);
2460 } else if (opcode
== BPF_JA
) {
2461 if (BPF_SRC(insn
->code
) != BPF_K
||
2463 insn
->src_reg
!= BPF_REG_0
||
2464 insn
->dst_reg
!= BPF_REG_0
) {
2465 verbose("BPF_JA uses reserved fields\n");
2469 insn_idx
+= insn
->off
+ 1;
2472 } else if (opcode
== BPF_EXIT
) {
2473 if (BPF_SRC(insn
->code
) != BPF_K
||
2475 insn
->src_reg
!= BPF_REG_0
||
2476 insn
->dst_reg
!= BPF_REG_0
) {
2477 verbose("BPF_EXIT uses reserved fields\n");
2481 /* eBPF calling convetion is such that R0 is used
2482 * to return the value from eBPF program.
2483 * Make sure that it's readable at this time
2484 * of bpf_exit, which means that program wrote
2485 * something into it earlier
2487 err
= check_reg_arg(regs
, BPF_REG_0
, SRC_OP
);
2491 if (is_pointer_value(env
, BPF_REG_0
)) {
2492 verbose("R0 leaks addr as return value\n");
2497 insn_idx
= pop_stack(env
, &prev_insn_idx
);
2501 do_print_state
= true;
2505 err
= check_cond_jmp_op(env
, insn
, &insn_idx
);
2509 } else if (class == BPF_LD
) {
2510 u8 mode
= BPF_MODE(insn
->code
);
2512 if (mode
== BPF_ABS
|| mode
== BPF_IND
) {
2513 err
= check_ld_abs(env
, insn
);
2517 } else if (mode
== BPF_IMM
) {
2518 err
= check_ld_imm(env
, insn
);
2524 verbose("invalid BPF_LD mode\n");
2528 verbose("unknown insn class %d\n", class);
2535 verbose("processed %d insns\n", insn_processed
);
2539 static int check_map_prog_compatibility(struct bpf_map
*map
,
2540 struct bpf_prog
*prog
)
2543 if (prog
->type
== BPF_PROG_TYPE_PERF_EVENT
&&
2544 (map
->map_type
== BPF_MAP_TYPE_HASH
||
2545 map
->map_type
== BPF_MAP_TYPE_PERCPU_HASH
) &&
2546 (map
->map_flags
& BPF_F_NO_PREALLOC
)) {
2547 verbose("perf_event programs can only use preallocated hash map\n");
2553 /* look for pseudo eBPF instructions that access map FDs and
2554 * replace them with actual map pointers
2556 static int replace_map_fd_with_map_ptr(struct verifier_env
*env
)
2558 struct bpf_insn
*insn
= env
->prog
->insnsi
;
2559 int insn_cnt
= env
->prog
->len
;
2562 for (i
= 0; i
< insn_cnt
; i
++, insn
++) {
2563 if (BPF_CLASS(insn
->code
) == BPF_LDX
&&
2564 (BPF_MODE(insn
->code
) != BPF_MEM
|| insn
->imm
!= 0)) {
2565 verbose("BPF_LDX uses reserved fields\n");
2569 if (BPF_CLASS(insn
->code
) == BPF_STX
&&
2570 ((BPF_MODE(insn
->code
) != BPF_MEM
&&
2571 BPF_MODE(insn
->code
) != BPF_XADD
) || insn
->imm
!= 0)) {
2572 verbose("BPF_STX uses reserved fields\n");
2576 if (insn
[0].code
== (BPF_LD
| BPF_IMM
| BPF_DW
)) {
2577 struct bpf_map
*map
;
2580 if (i
== insn_cnt
- 1 || insn
[1].code
!= 0 ||
2581 insn
[1].dst_reg
!= 0 || insn
[1].src_reg
!= 0 ||
2583 verbose("invalid bpf_ld_imm64 insn\n");
2587 if (insn
->src_reg
== 0)
2588 /* valid generic load 64-bit imm */
2591 if (insn
->src_reg
!= BPF_PSEUDO_MAP_FD
) {
2592 verbose("unrecognized bpf_ld_imm64 insn\n");
2596 f
= fdget(insn
->imm
);
2597 map
= __bpf_map_get(f
);
2599 verbose("fd %d is not pointing to valid bpf_map\n",
2601 return PTR_ERR(map
);
2604 err
= check_map_prog_compatibility(map
, env
->prog
);
2610 /* store map pointer inside BPF_LD_IMM64 instruction */
2611 insn
[0].imm
= (u32
) (unsigned long) map
;
2612 insn
[1].imm
= ((u64
) (unsigned long) map
) >> 32;
2614 /* check whether we recorded this map already */
2615 for (j
= 0; j
< env
->used_map_cnt
; j
++)
2616 if (env
->used_maps
[j
] == map
) {
2621 if (env
->used_map_cnt
>= MAX_USED_MAPS
) {
2626 /* hold the map. If the program is rejected by verifier,
2627 * the map will be released by release_maps() or it
2628 * will be used by the valid program until it's unloaded
2629 * and all maps are released in free_bpf_prog_info()
2631 map
= bpf_map_inc(map
, false);
2634 return PTR_ERR(map
);
2636 env
->used_maps
[env
->used_map_cnt
++] = map
;
2645 /* now all pseudo BPF_LD_IMM64 instructions load valid
2646 * 'struct bpf_map *' into a register instead of user map_fd.
2647 * These pointers will be used later by verifier to validate map access.
2652 /* drop refcnt of maps used by the rejected program */
2653 static void release_maps(struct verifier_env
*env
)
2657 for (i
= 0; i
< env
->used_map_cnt
; i
++)
2658 bpf_map_put(env
->used_maps
[i
]);
2661 /* convert pseudo BPF_LD_IMM64 into generic BPF_LD_IMM64 */
2662 static void convert_pseudo_ld_imm64(struct verifier_env
*env
)
2664 struct bpf_insn
*insn
= env
->prog
->insnsi
;
2665 int insn_cnt
= env
->prog
->len
;
2668 for (i
= 0; i
< insn_cnt
; i
++, insn
++)
2669 if (insn
->code
== (BPF_LD
| BPF_IMM
| BPF_DW
))
2673 /* convert load instructions that access fields of 'struct __sk_buff'
2674 * into sequence of instructions that access fields of 'struct sk_buff'
2676 static int convert_ctx_accesses(struct verifier_env
*env
)
2678 struct bpf_insn
*insn
= env
->prog
->insnsi
;
2679 int insn_cnt
= env
->prog
->len
;
2680 struct bpf_insn insn_buf
[16];
2681 struct bpf_prog
*new_prog
;
2682 enum bpf_access_type type
;
2685 if (!env
->prog
->aux
->ops
->convert_ctx_access
)
2688 for (i
= 0; i
< insn_cnt
; i
++, insn
++) {
2689 u32 insn_delta
, cnt
;
2691 if (insn
->code
== (BPF_LDX
| BPF_MEM
| BPF_W
) ||
2692 insn
->code
== (BPF_LDX
| BPF_MEM
| BPF_DW
))
2694 else if (insn
->code
== (BPF_STX
| BPF_MEM
| BPF_W
) ||
2695 insn
->code
== (BPF_STX
| BPF_MEM
| BPF_DW
))
2700 if (insn
->imm
!= PTR_TO_CTX
) {
2701 /* clear internal mark */
2706 cnt
= env
->prog
->aux
->ops
->
2707 convert_ctx_access(type
, insn
->dst_reg
, insn
->src_reg
,
2708 insn
->off
, insn_buf
, env
->prog
);
2709 if (cnt
== 0 || cnt
>= ARRAY_SIZE(insn_buf
)) {
2710 verbose("bpf verifier is misconfigured\n");
2714 new_prog
= bpf_patch_insn_single(env
->prog
, i
, insn_buf
, cnt
);
2718 insn_delta
= cnt
- 1;
2720 /* keep walking new program and skip insns we just inserted */
2721 env
->prog
= new_prog
;
2722 insn
= new_prog
->insnsi
+ i
+ insn_delta
;
2724 insn_cnt
+= insn_delta
;
2731 static void free_states(struct verifier_env
*env
)
2733 struct verifier_state_list
*sl
, *sln
;
2736 if (!env
->explored_states
)
2739 for (i
= 0; i
< env
->prog
->len
; i
++) {
2740 sl
= env
->explored_states
[i
];
2743 while (sl
!= STATE_LIST_MARK
) {
2750 kfree(env
->explored_states
);
2753 int bpf_check(struct bpf_prog
**prog
, union bpf_attr
*attr
)
2755 char __user
*log_ubuf
= NULL
;
2756 struct verifier_env
*env
;
2759 if ((*prog
)->len
<= 0 || (*prog
)->len
> BPF_MAXINSNS
)
2762 /* 'struct verifier_env' can be global, but since it's not small,
2763 * allocate/free it every time bpf_check() is called
2765 env
= kzalloc(sizeof(struct verifier_env
), GFP_KERNEL
);
2771 /* grab the mutex to protect few globals used by verifier */
2772 mutex_lock(&bpf_verifier_lock
);
2774 if (attr
->log_level
|| attr
->log_buf
|| attr
->log_size
) {
2775 /* user requested verbose verifier output
2776 * and supplied buffer to store the verification trace
2778 log_level
= attr
->log_level
;
2779 log_ubuf
= (char __user
*) (unsigned long) attr
->log_buf
;
2780 log_size
= attr
->log_size
;
2784 /* log_* values have to be sane */
2785 if (log_size
< 128 || log_size
> UINT_MAX
>> 8 ||
2786 log_level
== 0 || log_ubuf
== NULL
)
2790 log_buf
= vmalloc(log_size
);
2797 ret
= replace_map_fd_with_map_ptr(env
);
2799 goto skip_full_check
;
2801 env
->explored_states
= kcalloc(env
->prog
->len
,
2802 sizeof(struct verifier_state_list
*),
2805 if (!env
->explored_states
)
2806 goto skip_full_check
;
2808 ret
= check_cfg(env
);
2810 goto skip_full_check
;
2812 env
->allow_ptr_leaks
= capable(CAP_SYS_ADMIN
);
2814 ret
= do_check(env
);
2817 while (pop_stack(env
, NULL
) >= 0);
2821 /* program is valid, convert *(u32*)(ctx + off) accesses */
2822 ret
= convert_ctx_accesses(env
);
2824 if (log_level
&& log_len
>= log_size
- 1) {
2825 BUG_ON(log_len
>= log_size
);
2826 /* verifier log exceeded user supplied buffer */
2828 /* fall through to return what was recorded */
2831 /* copy verifier log back to user space including trailing zero */
2832 if (log_level
&& copy_to_user(log_ubuf
, log_buf
, log_len
+ 1) != 0) {
2837 if (ret
== 0 && env
->used_map_cnt
) {
2838 /* if program passed verifier, update used_maps in bpf_prog_info */
2839 env
->prog
->aux
->used_maps
= kmalloc_array(env
->used_map_cnt
,
2840 sizeof(env
->used_maps
[0]),
2843 if (!env
->prog
->aux
->used_maps
) {
2848 memcpy(env
->prog
->aux
->used_maps
, env
->used_maps
,
2849 sizeof(env
->used_maps
[0]) * env
->used_map_cnt
);
2850 env
->prog
->aux
->used_map_cnt
= env
->used_map_cnt
;
2852 /* program is valid. Convert pseudo bpf_ld_imm64 into generic
2853 * bpf_ld_imm64 instructions
2855 convert_pseudo_ld_imm64(env
);
2862 if (!env
->prog
->aux
->used_maps
)
2863 /* if we didn't copy map pointers into bpf_prog_info, release
2864 * them now. Otherwise free_bpf_prog_info() will release them.
2869 mutex_unlock(&bpf_verifier_lock
);