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>
26 static const struct bpf_verifier_ops
* const bpf_verifier_ops
[] = {
27 #define BPF_PROG_TYPE(_id, _name) \
28 [_id] = & _name ## _verifier_ops,
29 #define BPF_MAP_TYPE(_id, _ops)
30 #include <linux/bpf_types.h>
35 /* bpf_check() is a static code analyzer that walks eBPF program
36 * instruction by instruction and updates register/stack state.
37 * All paths of conditional branches are analyzed until 'bpf_exit' insn.
39 * The first pass is depth-first-search to check that the program is a DAG.
40 * It rejects the following programs:
41 * - larger than BPF_MAXINSNS insns
42 * - if loop is present (detected via back-edge)
43 * - unreachable insns exist (shouldn't be a forest. program = one function)
44 * - out of bounds or malformed jumps
45 * The second pass is all possible path descent from the 1st insn.
46 * Since it's analyzing all pathes through the program, the length of the
47 * analysis is limited to 64k insn, which may be hit even if total number of
48 * insn is less then 4K, but there are too many branches that change stack/regs.
49 * Number of 'branches to be analyzed' is limited to 1k
51 * On entry to each instruction, each register has a type, and the instruction
52 * changes the types of the registers depending on instruction semantics.
53 * If instruction is BPF_MOV64_REG(BPF_REG_1, BPF_REG_5), then type of R5 is
56 * All registers are 64-bit.
57 * R0 - return register
58 * R1-R5 argument passing registers
59 * R6-R9 callee saved registers
60 * R10 - frame pointer read-only
62 * At the start of BPF program the register R1 contains a pointer to bpf_context
63 * and has type PTR_TO_CTX.
65 * Verifier tracks arithmetic operations on pointers in case:
66 * BPF_MOV64_REG(BPF_REG_1, BPF_REG_10),
67 * BPF_ALU64_IMM(BPF_ADD, BPF_REG_1, -20),
68 * 1st insn copies R10 (which has FRAME_PTR) type into R1
69 * and 2nd arithmetic instruction is pattern matched to recognize
70 * that it wants to construct a pointer to some element within stack.
71 * So after 2nd insn, the register R1 has type PTR_TO_STACK
72 * (and -20 constant is saved for further stack bounds checking).
73 * Meaning that this reg is a pointer to stack plus known immediate constant.
75 * Most of the time the registers have SCALAR_VALUE type, which
76 * means the register has some value, but it's not a valid pointer.
77 * (like pointer plus pointer becomes SCALAR_VALUE type)
79 * When verifier sees load or store instructions the type of base register
80 * can be: PTR_TO_MAP_VALUE, PTR_TO_CTX, PTR_TO_STACK. These are three pointer
81 * types recognized by check_mem_access() function.
83 * PTR_TO_MAP_VALUE means that this register is pointing to 'map element value'
84 * and the range of [ptr, ptr + map's value_size) is accessible.
86 * registers used to pass values to function calls are checked against
87 * function argument constraints.
89 * ARG_PTR_TO_MAP_KEY is one of such argument constraints.
90 * It means that the register type passed to this function must be
91 * PTR_TO_STACK and it will be used inside the function as
92 * 'pointer to map element key'
94 * For example the argument constraints for bpf_map_lookup_elem():
95 * .ret_type = RET_PTR_TO_MAP_VALUE_OR_NULL,
96 * .arg1_type = ARG_CONST_MAP_PTR,
97 * .arg2_type = ARG_PTR_TO_MAP_KEY,
99 * ret_type says that this function returns 'pointer to map elem value or null'
100 * function expects 1st argument to be a const pointer to 'struct bpf_map' and
101 * 2nd argument should be a pointer to stack, which will be used inside
102 * the helper function as a pointer to map element key.
104 * On the kernel side the helper function looks like:
105 * u64 bpf_map_lookup_elem(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5)
107 * struct bpf_map *map = (struct bpf_map *) (unsigned long) r1;
108 * void *key = (void *) (unsigned long) r2;
111 * here kernel can access 'key' and 'map' pointers safely, knowing that
112 * [key, key + map->key_size) bytes are valid and were initialized on
113 * the stack of eBPF program.
116 * Corresponding eBPF program may look like:
117 * BPF_MOV64_REG(BPF_REG_2, BPF_REG_10), // after this insn R2 type is FRAME_PTR
118 * BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, -4), // after this insn R2 type is PTR_TO_STACK
119 * BPF_LD_MAP_FD(BPF_REG_1, map_fd), // after this insn R1 type is CONST_PTR_TO_MAP
120 * BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0, BPF_FUNC_map_lookup_elem),
121 * here verifier looks at prototype of map_lookup_elem() and sees:
122 * .arg1_type == ARG_CONST_MAP_PTR and R1->type == CONST_PTR_TO_MAP, which is ok,
123 * Now verifier knows that this map has key of R1->map_ptr->key_size bytes
125 * Then .arg2_type == ARG_PTR_TO_MAP_KEY and R2->type == PTR_TO_STACK, ok so far,
126 * Now verifier checks that [R2, R2 + map's key_size) are within stack limits
127 * and were initialized prior to this call.
128 * If it's ok, then verifier allows this BPF_CALL insn and looks at
129 * .ret_type which is RET_PTR_TO_MAP_VALUE_OR_NULL, so it sets
130 * R0->type = PTR_TO_MAP_VALUE_OR_NULL which means bpf_map_lookup_elem() function
131 * returns ether pointer to map value or NULL.
133 * When type PTR_TO_MAP_VALUE_OR_NULL passes through 'if (reg != 0) goto +off'
134 * insn, the register holding that pointer in the true branch changes state to
135 * PTR_TO_MAP_VALUE and the same register changes state to CONST_IMM in the false
136 * branch. See check_cond_jmp_op().
138 * After the call R0 is set to return type of the function and registers R1-R5
139 * are set to NOT_INIT to indicate that they are no longer readable.
142 /* verifier_state + insn_idx are pushed to stack when branch is encountered */
143 struct bpf_verifier_stack_elem
{
144 /* verifer state is 'st'
145 * before processing instruction 'insn_idx'
146 * and after processing instruction 'prev_insn_idx'
148 struct bpf_verifier_state st
;
151 struct bpf_verifier_stack_elem
*next
;
154 #define BPF_COMPLEXITY_LIMIT_INSNS 131072
155 #define BPF_COMPLEXITY_LIMIT_STACK 1024
157 #define BPF_MAP_PTR_POISON ((void *)0xeB9F + POISON_POINTER_DELTA)
159 struct bpf_call_arg_meta
{
160 struct bpf_map
*map_ptr
;
167 static DEFINE_MUTEX(bpf_verifier_lock
);
169 /* log_level controls verbosity level of eBPF verifier.
170 * verbose() is used to dump the verification trace to the log, so the user
171 * can figure out what's wrong with the program
173 static __printf(2, 3) void verbose(struct bpf_verifier_env
*env
,
174 const char *fmt
, ...)
176 struct bpf_verifer_log
*log
= &env
->log
;
180 if (!log
->level
|| !log
->ubuf
|| bpf_verifier_log_full(log
))
184 n
= vscnprintf(log
->kbuf
, BPF_VERIFIER_TMP_LOG_SIZE
, fmt
, args
);
187 WARN_ONCE(n
>= BPF_VERIFIER_TMP_LOG_SIZE
- 1,
188 "verifier log line truncated - local buffer too short\n");
190 n
= min(log
->len_total
- log
->len_used
- 1, n
);
193 if (!copy_to_user(log
->ubuf
+ log
->len_used
, log
->kbuf
, n
+ 1))
199 static bool type_is_pkt_pointer(enum bpf_reg_type type
)
201 return type
== PTR_TO_PACKET
||
202 type
== PTR_TO_PACKET_META
;
205 /* string representation of 'enum bpf_reg_type' */
206 static const char * const reg_type_str
[] = {
208 [SCALAR_VALUE
] = "inv",
209 [PTR_TO_CTX
] = "ctx",
210 [CONST_PTR_TO_MAP
] = "map_ptr",
211 [PTR_TO_MAP_VALUE
] = "map_value",
212 [PTR_TO_MAP_VALUE_OR_NULL
] = "map_value_or_null",
213 [PTR_TO_STACK
] = "fp",
214 [PTR_TO_PACKET
] = "pkt",
215 [PTR_TO_PACKET_META
] = "pkt_meta",
216 [PTR_TO_PACKET_END
] = "pkt_end",
219 static void print_verifier_state(struct bpf_verifier_env
*env
,
220 struct bpf_verifier_state
*state
)
222 struct bpf_reg_state
*reg
;
226 for (i
= 0; i
< MAX_BPF_REG
; i
++) {
227 reg
= &state
->regs
[i
];
231 verbose(env
, " R%d=%s", i
, reg_type_str
[t
]);
232 if ((t
== SCALAR_VALUE
|| t
== PTR_TO_STACK
) &&
233 tnum_is_const(reg
->var_off
)) {
234 /* reg->off should be 0 for SCALAR_VALUE */
235 verbose(env
, "%lld", reg
->var_off
.value
+ reg
->off
);
237 verbose(env
, "(id=%d", reg
->id
);
238 if (t
!= SCALAR_VALUE
)
239 verbose(env
, ",off=%d", reg
->off
);
240 if (type_is_pkt_pointer(t
))
241 verbose(env
, ",r=%d", reg
->range
);
242 else if (t
== CONST_PTR_TO_MAP
||
243 t
== PTR_TO_MAP_VALUE
||
244 t
== PTR_TO_MAP_VALUE_OR_NULL
)
245 verbose(env
, ",ks=%d,vs=%d",
246 reg
->map_ptr
->key_size
,
247 reg
->map_ptr
->value_size
);
248 if (tnum_is_const(reg
->var_off
)) {
249 /* Typically an immediate SCALAR_VALUE, but
250 * could be a pointer whose offset is too big
253 verbose(env
, ",imm=%llx", reg
->var_off
.value
);
255 if (reg
->smin_value
!= reg
->umin_value
&&
256 reg
->smin_value
!= S64_MIN
)
257 verbose(env
, ",smin_value=%lld",
258 (long long)reg
->smin_value
);
259 if (reg
->smax_value
!= reg
->umax_value
&&
260 reg
->smax_value
!= S64_MAX
)
261 verbose(env
, ",smax_value=%lld",
262 (long long)reg
->smax_value
);
263 if (reg
->umin_value
!= 0)
264 verbose(env
, ",umin_value=%llu",
265 (unsigned long long)reg
->umin_value
);
266 if (reg
->umax_value
!= U64_MAX
)
267 verbose(env
, ",umax_value=%llu",
268 (unsigned long long)reg
->umax_value
);
269 if (!tnum_is_unknown(reg
->var_off
)) {
272 tnum_strn(tn_buf
, sizeof(tn_buf
), reg
->var_off
);
273 verbose(env
, ",var_off=%s", tn_buf
);
279 for (i
= 0; i
< MAX_BPF_STACK
; i
+= BPF_REG_SIZE
) {
280 if (state
->stack_slot_type
[i
] == STACK_SPILL
)
281 verbose(env
, " fp%d=%s", -MAX_BPF_STACK
+ i
,
282 reg_type_str
[state
->spilled_regs
[i
/ BPF_REG_SIZE
].type
]);
287 static int pop_stack(struct bpf_verifier_env
*env
, int *prev_insn_idx
)
289 struct bpf_verifier_stack_elem
*elem
;
292 if (env
->head
== NULL
)
295 memcpy(&env
->cur_state
, &env
->head
->st
, sizeof(env
->cur_state
));
296 insn_idx
= env
->head
->insn_idx
;
298 *prev_insn_idx
= env
->head
->prev_insn_idx
;
299 elem
= env
->head
->next
;
306 static struct bpf_verifier_state
*push_stack(struct bpf_verifier_env
*env
,
307 int insn_idx
, int prev_insn_idx
)
309 struct bpf_verifier_stack_elem
*elem
;
311 elem
= kmalloc(sizeof(struct bpf_verifier_stack_elem
), GFP_KERNEL
);
315 memcpy(&elem
->st
, &env
->cur_state
, sizeof(env
->cur_state
));
316 elem
->insn_idx
= insn_idx
;
317 elem
->prev_insn_idx
= prev_insn_idx
;
318 elem
->next
= env
->head
;
321 if (env
->stack_size
> BPF_COMPLEXITY_LIMIT_STACK
) {
322 verbose(env
, "BPF program is too complex\n");
327 /* pop all elements and return */
328 while (pop_stack(env
, NULL
) >= 0);
332 #define CALLER_SAVED_REGS 6
333 static const int caller_saved
[CALLER_SAVED_REGS
] = {
334 BPF_REG_0
, BPF_REG_1
, BPF_REG_2
, BPF_REG_3
, BPF_REG_4
, BPF_REG_5
337 static void __mark_reg_not_init(struct bpf_reg_state
*reg
);
339 /* Mark the unknown part of a register (variable offset or scalar value) as
340 * known to have the value @imm.
342 static void __mark_reg_known(struct bpf_reg_state
*reg
, u64 imm
)
345 reg
->var_off
= tnum_const(imm
);
346 reg
->smin_value
= (s64
)imm
;
347 reg
->smax_value
= (s64
)imm
;
348 reg
->umin_value
= imm
;
349 reg
->umax_value
= imm
;
352 /* Mark the 'variable offset' part of a register as zero. This should be
353 * used only on registers holding a pointer type.
355 static void __mark_reg_known_zero(struct bpf_reg_state
*reg
)
357 __mark_reg_known(reg
, 0);
360 static void mark_reg_known_zero(struct bpf_verifier_env
*env
,
361 struct bpf_reg_state
*regs
, u32 regno
)
363 if (WARN_ON(regno
>= MAX_BPF_REG
)) {
364 verbose(env
, "mark_reg_known_zero(regs, %u)\n", regno
);
365 /* Something bad happened, let's kill all regs */
366 for (regno
= 0; regno
< MAX_BPF_REG
; regno
++)
367 __mark_reg_not_init(regs
+ regno
);
370 __mark_reg_known_zero(regs
+ regno
);
373 static bool reg_is_pkt_pointer(const struct bpf_reg_state
*reg
)
375 return type_is_pkt_pointer(reg
->type
);
378 static bool reg_is_pkt_pointer_any(const struct bpf_reg_state
*reg
)
380 return reg_is_pkt_pointer(reg
) ||
381 reg
->type
== PTR_TO_PACKET_END
;
384 /* Unmodified PTR_TO_PACKET[_META,_END] register from ctx access. */
385 static bool reg_is_init_pkt_pointer(const struct bpf_reg_state
*reg
,
386 enum bpf_reg_type which
)
388 /* The register can already have a range from prior markings.
389 * This is fine as long as it hasn't been advanced from its
392 return reg
->type
== which
&&
395 tnum_equals_const(reg
->var_off
, 0);
398 /* Attempts to improve min/max values based on var_off information */
399 static void __update_reg_bounds(struct bpf_reg_state
*reg
)
401 /* min signed is max(sign bit) | min(other bits) */
402 reg
->smin_value
= max_t(s64
, reg
->smin_value
,
403 reg
->var_off
.value
| (reg
->var_off
.mask
& S64_MIN
));
404 /* max signed is min(sign bit) | max(other bits) */
405 reg
->smax_value
= min_t(s64
, reg
->smax_value
,
406 reg
->var_off
.value
| (reg
->var_off
.mask
& S64_MAX
));
407 reg
->umin_value
= max(reg
->umin_value
, reg
->var_off
.value
);
408 reg
->umax_value
= min(reg
->umax_value
,
409 reg
->var_off
.value
| reg
->var_off
.mask
);
412 /* Uses signed min/max values to inform unsigned, and vice-versa */
413 static void __reg_deduce_bounds(struct bpf_reg_state
*reg
)
415 /* Learn sign from signed bounds.
416 * If we cannot cross the sign boundary, then signed and unsigned bounds
417 * are the same, so combine. This works even in the negative case, e.g.
418 * -3 s<= x s<= -1 implies 0xf...fd u<= x u<= 0xf...ff.
420 if (reg
->smin_value
>= 0 || reg
->smax_value
< 0) {
421 reg
->smin_value
= reg
->umin_value
= max_t(u64
, reg
->smin_value
,
423 reg
->smax_value
= reg
->umax_value
= min_t(u64
, reg
->smax_value
,
427 /* Learn sign from unsigned bounds. Signed bounds cross the sign
428 * boundary, so we must be careful.
430 if ((s64
)reg
->umax_value
>= 0) {
431 /* Positive. We can't learn anything from the smin, but smax
432 * is positive, hence safe.
434 reg
->smin_value
= reg
->umin_value
;
435 reg
->smax_value
= reg
->umax_value
= min_t(u64
, reg
->smax_value
,
437 } else if ((s64
)reg
->umin_value
< 0) {
438 /* Negative. We can't learn anything from the smax, but smin
439 * is negative, hence safe.
441 reg
->smin_value
= reg
->umin_value
= max_t(u64
, reg
->smin_value
,
443 reg
->smax_value
= reg
->umax_value
;
447 /* Attempts to improve var_off based on unsigned min/max information */
448 static void __reg_bound_offset(struct bpf_reg_state
*reg
)
450 reg
->var_off
= tnum_intersect(reg
->var_off
,
451 tnum_range(reg
->umin_value
,
455 /* Reset the min/max bounds of a register */
456 static void __mark_reg_unbounded(struct bpf_reg_state
*reg
)
458 reg
->smin_value
= S64_MIN
;
459 reg
->smax_value
= S64_MAX
;
461 reg
->umax_value
= U64_MAX
;
464 /* Mark a register as having a completely unknown (scalar) value. */
465 static void __mark_reg_unknown(struct bpf_reg_state
*reg
)
467 reg
->type
= SCALAR_VALUE
;
470 reg
->var_off
= tnum_unknown
;
471 __mark_reg_unbounded(reg
);
474 static void mark_reg_unknown(struct bpf_verifier_env
*env
,
475 struct bpf_reg_state
*regs
, u32 regno
)
477 if (WARN_ON(regno
>= MAX_BPF_REG
)) {
478 verbose(env
, "mark_reg_unknown(regs, %u)\n", regno
);
479 /* Something bad happened, let's kill all regs */
480 for (regno
= 0; regno
< MAX_BPF_REG
; regno
++)
481 __mark_reg_not_init(regs
+ regno
);
484 __mark_reg_unknown(regs
+ regno
);
487 static void __mark_reg_not_init(struct bpf_reg_state
*reg
)
489 __mark_reg_unknown(reg
);
490 reg
->type
= NOT_INIT
;
493 static void mark_reg_not_init(struct bpf_verifier_env
*env
,
494 struct bpf_reg_state
*regs
, u32 regno
)
496 if (WARN_ON(regno
>= MAX_BPF_REG
)) {
497 verbose(env
, "mark_reg_not_init(regs, %u)\n", regno
);
498 /* Something bad happened, let's kill all regs */
499 for (regno
= 0; regno
< MAX_BPF_REG
; regno
++)
500 __mark_reg_not_init(regs
+ regno
);
503 __mark_reg_not_init(regs
+ regno
);
506 static void init_reg_state(struct bpf_verifier_env
*env
,
507 struct bpf_reg_state
*regs
)
511 for (i
= 0; i
< MAX_BPF_REG
; i
++) {
512 mark_reg_not_init(env
, regs
, i
);
513 regs
[i
].live
= REG_LIVE_NONE
;
517 regs
[BPF_REG_FP
].type
= PTR_TO_STACK
;
518 mark_reg_known_zero(env
, regs
, BPF_REG_FP
);
520 /* 1st arg to a function */
521 regs
[BPF_REG_1
].type
= PTR_TO_CTX
;
522 mark_reg_known_zero(env
, regs
, BPF_REG_1
);
526 SRC_OP
, /* register is used as source operand */
527 DST_OP
, /* register is used as destination operand */
528 DST_OP_NO_MARK
/* same as above, check only, don't mark */
531 static void mark_reg_read(const struct bpf_verifier_state
*state
, u32 regno
)
533 struct bpf_verifier_state
*parent
= state
->parent
;
535 if (regno
== BPF_REG_FP
)
536 /* We don't need to worry about FP liveness because it's read-only */
540 /* if read wasn't screened by an earlier write ... */
541 if (state
->regs
[regno
].live
& REG_LIVE_WRITTEN
)
543 /* ... then we depend on parent's value */
544 parent
->regs
[regno
].live
|= REG_LIVE_READ
;
546 parent
= state
->parent
;
550 static int check_reg_arg(struct bpf_verifier_env
*env
, u32 regno
,
553 struct bpf_reg_state
*regs
= env
->cur_state
.regs
;
555 if (regno
>= MAX_BPF_REG
) {
556 verbose(env
, "R%d is invalid\n", regno
);
561 /* check whether register used as source operand can be read */
562 if (regs
[regno
].type
== NOT_INIT
) {
563 verbose(env
, "R%d !read_ok\n", regno
);
566 mark_reg_read(&env
->cur_state
, regno
);
568 /* check whether register used as dest operand can be written to */
569 if (regno
== BPF_REG_FP
) {
570 verbose(env
, "frame pointer is read only\n");
573 regs
[regno
].live
|= REG_LIVE_WRITTEN
;
575 mark_reg_unknown(env
, regs
, regno
);
580 static bool is_spillable_regtype(enum bpf_reg_type type
)
583 case PTR_TO_MAP_VALUE
:
584 case PTR_TO_MAP_VALUE_OR_NULL
:
588 case PTR_TO_PACKET_META
:
589 case PTR_TO_PACKET_END
:
590 case CONST_PTR_TO_MAP
:
597 /* check_stack_read/write functions track spill/fill of registers,
598 * stack boundary and alignment are checked in check_mem_access()
600 static int check_stack_write(struct bpf_verifier_env
*env
,
601 struct bpf_verifier_state
*state
, int off
,
602 int size
, int value_regno
)
604 int i
, spi
= (MAX_BPF_STACK
+ off
) / BPF_REG_SIZE
;
605 /* caller checked that off % size == 0 and -MAX_BPF_STACK <= off < 0,
606 * so it's aligned access and [off, off + size) are within stack limits
609 if (value_regno
>= 0 &&
610 is_spillable_regtype(state
->regs
[value_regno
].type
)) {
612 /* register containing pointer is being spilled into stack */
613 if (size
!= BPF_REG_SIZE
) {
614 verbose(env
, "invalid size of register spill\n");
618 /* save register state */
619 state
->spilled_regs
[spi
] = state
->regs
[value_regno
];
620 state
->spilled_regs
[spi
].live
|= REG_LIVE_WRITTEN
;
622 for (i
= 0; i
< BPF_REG_SIZE
; i
++)
623 state
->stack_slot_type
[MAX_BPF_STACK
+ off
+ i
] = STACK_SPILL
;
625 /* regular write of data into stack */
626 state
->spilled_regs
[spi
] = (struct bpf_reg_state
) {};
628 for (i
= 0; i
< size
; i
++)
629 state
->stack_slot_type
[MAX_BPF_STACK
+ off
+ i
] = STACK_MISC
;
634 static void mark_stack_slot_read(const struct bpf_verifier_state
*state
, int slot
)
636 struct bpf_verifier_state
*parent
= state
->parent
;
639 /* if read wasn't screened by an earlier write ... */
640 if (state
->spilled_regs
[slot
].live
& REG_LIVE_WRITTEN
)
642 /* ... then we depend on parent's value */
643 parent
->spilled_regs
[slot
].live
|= REG_LIVE_READ
;
645 parent
= state
->parent
;
649 static int check_stack_read(struct bpf_verifier_env
*env
,
650 struct bpf_verifier_state
*state
, int off
, int size
,
656 slot_type
= &state
->stack_slot_type
[MAX_BPF_STACK
+ off
];
658 if (slot_type
[0] == STACK_SPILL
) {
659 if (size
!= BPF_REG_SIZE
) {
660 verbose(env
, "invalid size of register spill\n");
663 for (i
= 1; i
< BPF_REG_SIZE
; i
++) {
664 if (slot_type
[i
] != STACK_SPILL
) {
665 verbose(env
, "corrupted spill memory\n");
670 spi
= (MAX_BPF_STACK
+ off
) / BPF_REG_SIZE
;
672 if (value_regno
>= 0) {
673 /* restore register state from stack */
674 state
->regs
[value_regno
] = state
->spilled_regs
[spi
];
675 mark_stack_slot_read(state
, spi
);
679 for (i
= 0; i
< size
; i
++) {
680 if (slot_type
[i
] != STACK_MISC
) {
681 verbose(env
, "invalid read from stack off %d+%d size %d\n",
686 if (value_regno
>= 0)
687 /* have read misc data from the stack */
688 mark_reg_unknown(env
, state
->regs
, value_regno
);
693 /* check read/write into map element returned by bpf_map_lookup_elem() */
694 static int __check_map_access(struct bpf_verifier_env
*env
, u32 regno
, int off
,
697 struct bpf_map
*map
= env
->cur_state
.regs
[regno
].map_ptr
;
699 if (off
< 0 || size
<= 0 || off
+ size
> map
->value_size
) {
700 verbose(env
, "invalid access to map value, value_size=%d off=%d size=%d\n",
701 map
->value_size
, off
, size
);
707 /* check read/write into a map element with possible variable offset */
708 static int check_map_access(struct bpf_verifier_env
*env
, u32 regno
,
711 struct bpf_verifier_state
*state
= &env
->cur_state
;
712 struct bpf_reg_state
*reg
= &state
->regs
[regno
];
715 /* We may have adjusted the register to this map value, so we
716 * need to try adding each of min_value and max_value to off
717 * to make sure our theoretical access will be safe.
720 print_verifier_state(env
, state
);
721 /* The minimum value is only important with signed
722 * comparisons where we can't assume the floor of a
723 * value is 0. If we are using signed variables for our
724 * index'es we need to make sure that whatever we use
725 * will have a set floor within our range.
727 if (reg
->smin_value
< 0) {
728 verbose(env
, "R%d min value is negative, either use unsigned index or do a if (index >=0) check.\n",
732 err
= __check_map_access(env
, regno
, reg
->smin_value
+ off
, size
);
734 verbose(env
, "R%d min value is outside of the array range\n",
739 /* If we haven't set a max value then we need to bail since we can't be
740 * sure we won't do bad things.
741 * If reg->umax_value + off could overflow, treat that as unbounded too.
743 if (reg
->umax_value
>= BPF_MAX_VAR_OFF
) {
744 verbose(env
, "R%d unbounded memory access, make sure to bounds check any array access into a map\n",
748 err
= __check_map_access(env
, regno
, reg
->umax_value
+ off
, size
);
750 verbose(env
, "R%d max value is outside of the array range\n",
755 #define MAX_PACKET_OFF 0xffff
757 static bool may_access_direct_pkt_data(struct bpf_verifier_env
*env
,
758 const struct bpf_call_arg_meta
*meta
,
759 enum bpf_access_type t
)
761 switch (env
->prog
->type
) {
762 case BPF_PROG_TYPE_LWT_IN
:
763 case BPF_PROG_TYPE_LWT_OUT
:
764 /* dst_input() and dst_output() can't write for now */
768 case BPF_PROG_TYPE_SCHED_CLS
:
769 case BPF_PROG_TYPE_SCHED_ACT
:
770 case BPF_PROG_TYPE_XDP
:
771 case BPF_PROG_TYPE_LWT_XMIT
:
772 case BPF_PROG_TYPE_SK_SKB
:
774 return meta
->pkt_access
;
776 env
->seen_direct_write
= true;
783 static int __check_packet_access(struct bpf_verifier_env
*env
, u32 regno
,
786 struct bpf_reg_state
*regs
= env
->cur_state
.regs
;
787 struct bpf_reg_state
*reg
= ®s
[regno
];
789 if (off
< 0 || size
<= 0 || (u64
)off
+ size
> reg
->range
) {
790 verbose(env
, "invalid access to packet, off=%d size=%d, R%d(id=%d,off=%d,r=%d)\n",
791 off
, size
, regno
, reg
->id
, reg
->off
, reg
->range
);
797 static int check_packet_access(struct bpf_verifier_env
*env
, u32 regno
, int off
,
800 struct bpf_reg_state
*regs
= env
->cur_state
.regs
;
801 struct bpf_reg_state
*reg
= ®s
[regno
];
804 /* We may have added a variable offset to the packet pointer; but any
805 * reg->range we have comes after that. We are only checking the fixed
809 /* We don't allow negative numbers, because we aren't tracking enough
810 * detail to prove they're safe.
812 if (reg
->smin_value
< 0) {
813 verbose(env
, "R%d min value is negative, either use unsigned index or do a if (index >=0) check.\n",
817 err
= __check_packet_access(env
, regno
, off
, size
);
819 verbose(env
, "R%d offset is outside of the packet\n", regno
);
825 /* check access to 'struct bpf_context' fields. Supports fixed offsets only */
826 static int check_ctx_access(struct bpf_verifier_env
*env
, int insn_idx
, int off
, int size
,
827 enum bpf_access_type t
, enum bpf_reg_type
*reg_type
)
829 struct bpf_insn_access_aux info
= {
830 .reg_type
= *reg_type
,
833 if (env
->ops
->is_valid_access
&&
834 env
->ops
->is_valid_access(off
, size
, t
, &info
)) {
835 /* A non zero info.ctx_field_size indicates that this field is a
836 * candidate for later verifier transformation to load the whole
837 * field and then apply a mask when accessed with a narrower
838 * access than actual ctx access size. A zero info.ctx_field_size
839 * will only allow for whole field access and rejects any other
840 * type of narrower access.
842 *reg_type
= info
.reg_type
;
844 if (env
->analyzer_ops
)
847 env
->insn_aux_data
[insn_idx
].ctx_field_size
= info
.ctx_field_size
;
848 /* remember the offset of last byte accessed in ctx */
849 if (env
->prog
->aux
->max_ctx_offset
< off
+ size
)
850 env
->prog
->aux
->max_ctx_offset
= off
+ size
;
854 verbose(env
, "invalid bpf_context access off=%d size=%d\n", off
, size
);
858 static bool __is_pointer_value(bool allow_ptr_leaks
,
859 const struct bpf_reg_state
*reg
)
864 return reg
->type
!= SCALAR_VALUE
;
867 static bool is_pointer_value(struct bpf_verifier_env
*env
, int regno
)
869 return __is_pointer_value(env
->allow_ptr_leaks
, &env
->cur_state
.regs
[regno
]);
872 static int check_pkt_ptr_alignment(struct bpf_verifier_env
*env
,
873 const struct bpf_reg_state
*reg
,
874 int off
, int size
, bool strict
)
879 /* Byte size accesses are always allowed. */
880 if (!strict
|| size
== 1)
883 /* For platforms that do not have a Kconfig enabling
884 * CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS the value of
885 * NET_IP_ALIGN is universally set to '2'. And on platforms
886 * that do set CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS, we get
887 * to this code only in strict mode where we want to emulate
888 * the NET_IP_ALIGN==2 checking. Therefore use an
889 * unconditional IP align value of '2'.
893 reg_off
= tnum_add(reg
->var_off
, tnum_const(ip_align
+ reg
->off
+ off
));
894 if (!tnum_is_aligned(reg_off
, size
)) {
897 tnum_strn(tn_buf
, sizeof(tn_buf
), reg
->var_off
);
899 "misaligned packet access off %d+%s+%d+%d size %d\n",
900 ip_align
, tn_buf
, reg
->off
, off
, size
);
907 static int check_generic_ptr_alignment(struct bpf_verifier_env
*env
,
908 const struct bpf_reg_state
*reg
,
909 const char *pointer_desc
,
910 int off
, int size
, bool strict
)
914 /* Byte size accesses are always allowed. */
915 if (!strict
|| size
== 1)
918 reg_off
= tnum_add(reg
->var_off
, tnum_const(reg
->off
+ off
));
919 if (!tnum_is_aligned(reg_off
, size
)) {
922 tnum_strn(tn_buf
, sizeof(tn_buf
), reg
->var_off
);
923 verbose(env
, "misaligned %saccess off %s+%d+%d size %d\n",
924 pointer_desc
, tn_buf
, reg
->off
, off
, size
);
931 static int check_ptr_alignment(struct bpf_verifier_env
*env
,
932 const struct bpf_reg_state
*reg
,
935 bool strict
= env
->strict_alignment
;
936 const char *pointer_desc
= "";
940 case PTR_TO_PACKET_META
:
941 /* Special case, because of NET_IP_ALIGN. Given metadata sits
942 * right in front, treat it the very same way.
944 return check_pkt_ptr_alignment(env
, reg
, off
, size
, strict
);
945 case PTR_TO_MAP_VALUE
:
946 pointer_desc
= "value ";
949 pointer_desc
= "context ";
952 pointer_desc
= "stack ";
957 return check_generic_ptr_alignment(env
, reg
, pointer_desc
, off
, size
,
961 /* check whether memory at (regno + off) is accessible for t = (read | write)
962 * if t==write, value_regno is a register which value is stored into memory
963 * if t==read, value_regno is a register which will receive the value from memory
964 * if t==write && value_regno==-1, some unknown value is stored into memory
965 * if t==read && value_regno==-1, don't care what we read from memory
967 static int check_mem_access(struct bpf_verifier_env
*env
, int insn_idx
, u32 regno
, int off
,
968 int bpf_size
, enum bpf_access_type t
,
971 struct bpf_verifier_state
*state
= &env
->cur_state
;
972 struct bpf_reg_state
*reg
= &state
->regs
[regno
];
975 size
= bpf_size_to_bytes(bpf_size
);
979 /* alignment checks will add in reg->off themselves */
980 err
= check_ptr_alignment(env
, reg
, off
, size
);
984 /* for access checks, reg->off is just part of off */
987 if (reg
->type
== PTR_TO_MAP_VALUE
) {
988 if (t
== BPF_WRITE
&& value_regno
>= 0 &&
989 is_pointer_value(env
, value_regno
)) {
990 verbose(env
, "R%d leaks addr into map\n", value_regno
);
994 err
= check_map_access(env
, regno
, off
, size
);
995 if (!err
&& t
== BPF_READ
&& value_regno
>= 0)
996 mark_reg_unknown(env
, state
->regs
, value_regno
);
998 } else if (reg
->type
== PTR_TO_CTX
) {
999 enum bpf_reg_type reg_type
= SCALAR_VALUE
;
1001 if (t
== BPF_WRITE
&& value_regno
>= 0 &&
1002 is_pointer_value(env
, value_regno
)) {
1003 verbose(env
, "R%d leaks addr into ctx\n", value_regno
);
1006 /* ctx accesses must be at a fixed offset, so that we can
1007 * determine what type of data were returned.
1009 if (!tnum_is_const(reg
->var_off
)) {
1012 tnum_strn(tn_buf
, sizeof(tn_buf
), reg
->var_off
);
1014 "variable ctx access var_off=%s off=%d size=%d",
1018 off
+= reg
->var_off
.value
;
1019 err
= check_ctx_access(env
, insn_idx
, off
, size
, t
, ®_type
);
1020 if (!err
&& t
== BPF_READ
&& value_regno
>= 0) {
1021 /* ctx access returns either a scalar, or a
1022 * PTR_TO_PACKET[_META,_END]. In the latter
1023 * case, we know the offset is zero.
1025 if (reg_type
== SCALAR_VALUE
)
1026 mark_reg_unknown(env
, state
->regs
, value_regno
);
1028 mark_reg_known_zero(env
, state
->regs
,
1030 state
->regs
[value_regno
].id
= 0;
1031 state
->regs
[value_regno
].off
= 0;
1032 state
->regs
[value_regno
].range
= 0;
1033 state
->regs
[value_regno
].type
= reg_type
;
1036 } else if (reg
->type
== PTR_TO_STACK
) {
1037 /* stack accesses must be at a fixed offset, so that we can
1038 * determine what type of data were returned.
1039 * See check_stack_read().
1041 if (!tnum_is_const(reg
->var_off
)) {
1044 tnum_strn(tn_buf
, sizeof(tn_buf
), reg
->var_off
);
1045 verbose(env
, "variable stack access var_off=%s off=%d size=%d",
1049 off
+= reg
->var_off
.value
;
1050 if (off
>= 0 || off
< -MAX_BPF_STACK
) {
1051 verbose(env
, "invalid stack off=%d size=%d\n", off
,
1056 if (env
->prog
->aux
->stack_depth
< -off
)
1057 env
->prog
->aux
->stack_depth
= -off
;
1059 if (t
== BPF_WRITE
) {
1060 if (!env
->allow_ptr_leaks
&&
1061 state
->stack_slot_type
[MAX_BPF_STACK
+ off
] == STACK_SPILL
&&
1062 size
!= BPF_REG_SIZE
) {
1063 verbose(env
, "attempt to corrupt spilled pointer on stack\n");
1066 err
= check_stack_write(env
, state
, off
, size
,
1069 err
= check_stack_read(env
, state
, off
, size
,
1072 } else if (reg_is_pkt_pointer(reg
)) {
1073 if (t
== BPF_WRITE
&& !may_access_direct_pkt_data(env
, NULL
, t
)) {
1074 verbose(env
, "cannot write into packet\n");
1077 if (t
== BPF_WRITE
&& value_regno
>= 0 &&
1078 is_pointer_value(env
, value_regno
)) {
1079 verbose(env
, "R%d leaks addr into packet\n",
1083 err
= check_packet_access(env
, regno
, off
, size
);
1084 if (!err
&& t
== BPF_READ
&& value_regno
>= 0)
1085 mark_reg_unknown(env
, state
->regs
, value_regno
);
1087 verbose(env
, "R%d invalid mem access '%s'\n", regno
,
1088 reg_type_str
[reg
->type
]);
1092 if (!err
&& size
< BPF_REG_SIZE
&& value_regno
>= 0 && t
== BPF_READ
&&
1093 state
->regs
[value_regno
].type
== SCALAR_VALUE
) {
1094 /* b/h/w load zero-extends, mark upper bits as known 0 */
1095 state
->regs
[value_regno
].var_off
= tnum_cast(
1096 state
->regs
[value_regno
].var_off
, size
);
1097 __update_reg_bounds(&state
->regs
[value_regno
]);
1102 static int check_xadd(struct bpf_verifier_env
*env
, int insn_idx
, struct bpf_insn
*insn
)
1106 if ((BPF_SIZE(insn
->code
) != BPF_W
&& BPF_SIZE(insn
->code
) != BPF_DW
) ||
1108 verbose(env
, "BPF_XADD uses reserved fields\n");
1112 /* check src1 operand */
1113 err
= check_reg_arg(env
, insn
->src_reg
, SRC_OP
);
1117 /* check src2 operand */
1118 err
= check_reg_arg(env
, insn
->dst_reg
, SRC_OP
);
1122 if (is_pointer_value(env
, insn
->src_reg
)) {
1123 verbose(env
, "R%d leaks addr into mem\n", insn
->src_reg
);
1127 /* check whether atomic_add can read the memory */
1128 err
= check_mem_access(env
, insn_idx
, insn
->dst_reg
, insn
->off
,
1129 BPF_SIZE(insn
->code
), BPF_READ
, -1);
1133 /* check whether atomic_add can write into the same memory */
1134 return check_mem_access(env
, insn_idx
, insn
->dst_reg
, insn
->off
,
1135 BPF_SIZE(insn
->code
), BPF_WRITE
, -1);
1138 /* Does this register contain a constant zero? */
1139 static bool register_is_null(struct bpf_reg_state reg
)
1141 return reg
.type
== SCALAR_VALUE
&& tnum_equals_const(reg
.var_off
, 0);
1144 /* when register 'regno' is passed into function that will read 'access_size'
1145 * bytes from that pointer, make sure that it's within stack boundary
1146 * and all elements of stack are initialized.
1147 * Unlike most pointer bounds-checking functions, this one doesn't take an
1148 * 'off' argument, so it has to add in reg->off itself.
1150 static int check_stack_boundary(struct bpf_verifier_env
*env
, int regno
,
1151 int access_size
, bool zero_size_allowed
,
1152 struct bpf_call_arg_meta
*meta
)
1154 struct bpf_verifier_state
*state
= &env
->cur_state
;
1155 struct bpf_reg_state
*regs
= state
->regs
;
1158 if (regs
[regno
].type
!= PTR_TO_STACK
) {
1159 /* Allow zero-byte read from NULL, regardless of pointer type */
1160 if (zero_size_allowed
&& access_size
== 0 &&
1161 register_is_null(regs
[regno
]))
1164 verbose(env
, "R%d type=%s expected=%s\n", regno
,
1165 reg_type_str
[regs
[regno
].type
],
1166 reg_type_str
[PTR_TO_STACK
]);
1170 /* Only allow fixed-offset stack reads */
1171 if (!tnum_is_const(regs
[regno
].var_off
)) {
1174 tnum_strn(tn_buf
, sizeof(tn_buf
), regs
[regno
].var_off
);
1175 verbose(env
, "invalid variable stack read R%d var_off=%s\n",
1178 off
= regs
[regno
].off
+ regs
[regno
].var_off
.value
;
1179 if (off
>= 0 || off
< -MAX_BPF_STACK
|| off
+ access_size
> 0 ||
1181 verbose(env
, "invalid stack type R%d off=%d access_size=%d\n",
1182 regno
, off
, access_size
);
1186 if (env
->prog
->aux
->stack_depth
< -off
)
1187 env
->prog
->aux
->stack_depth
= -off
;
1189 if (meta
&& meta
->raw_mode
) {
1190 meta
->access_size
= access_size
;
1191 meta
->regno
= regno
;
1195 for (i
= 0; i
< access_size
; i
++) {
1196 if (state
->stack_slot_type
[MAX_BPF_STACK
+ off
+ i
] != STACK_MISC
) {
1197 verbose(env
, "invalid indirect read from stack off %d+%d size %d\n",
1198 off
, i
, access_size
);
1205 static int check_helper_mem_access(struct bpf_verifier_env
*env
, int regno
,
1206 int access_size
, bool zero_size_allowed
,
1207 struct bpf_call_arg_meta
*meta
)
1209 struct bpf_reg_state
*regs
= env
->cur_state
.regs
, *reg
= ®s
[regno
];
1211 switch (reg
->type
) {
1213 case PTR_TO_PACKET_META
:
1214 return check_packet_access(env
, regno
, reg
->off
, access_size
);
1215 case PTR_TO_MAP_VALUE
:
1216 return check_map_access(env
, regno
, reg
->off
, access_size
);
1217 default: /* scalar_value|ptr_to_stack or invalid ptr */
1218 return check_stack_boundary(env
, regno
, access_size
,
1219 zero_size_allowed
, meta
);
1223 static int check_func_arg(struct bpf_verifier_env
*env
, u32 regno
,
1224 enum bpf_arg_type arg_type
,
1225 struct bpf_call_arg_meta
*meta
)
1227 struct bpf_reg_state
*regs
= env
->cur_state
.regs
, *reg
= ®s
[regno
];
1228 enum bpf_reg_type expected_type
, type
= reg
->type
;
1231 if (arg_type
== ARG_DONTCARE
)
1234 err
= check_reg_arg(env
, regno
, SRC_OP
);
1238 if (arg_type
== ARG_ANYTHING
) {
1239 if (is_pointer_value(env
, regno
)) {
1240 verbose(env
, "R%d leaks addr into helper function\n",
1247 if (type_is_pkt_pointer(type
) &&
1248 !may_access_direct_pkt_data(env
, meta
, BPF_READ
)) {
1249 verbose(env
, "helper access to the packet is not allowed\n");
1253 if (arg_type
== ARG_PTR_TO_MAP_KEY
||
1254 arg_type
== ARG_PTR_TO_MAP_VALUE
) {
1255 expected_type
= PTR_TO_STACK
;
1256 if (!type_is_pkt_pointer(type
) &&
1257 type
!= expected_type
)
1259 } else if (arg_type
== ARG_CONST_SIZE
||
1260 arg_type
== ARG_CONST_SIZE_OR_ZERO
) {
1261 expected_type
= SCALAR_VALUE
;
1262 if (type
!= expected_type
)
1264 } else if (arg_type
== ARG_CONST_MAP_PTR
) {
1265 expected_type
= CONST_PTR_TO_MAP
;
1266 if (type
!= expected_type
)
1268 } else if (arg_type
== ARG_PTR_TO_CTX
) {
1269 expected_type
= PTR_TO_CTX
;
1270 if (type
!= expected_type
)
1272 } else if (arg_type
== ARG_PTR_TO_MEM
||
1273 arg_type
== ARG_PTR_TO_UNINIT_MEM
) {
1274 expected_type
= PTR_TO_STACK
;
1275 /* One exception here. In case function allows for NULL to be
1276 * passed in as argument, it's a SCALAR_VALUE type. Final test
1277 * happens during stack boundary checking.
1279 if (register_is_null(*reg
))
1280 /* final test in check_stack_boundary() */;
1281 else if (!type_is_pkt_pointer(type
) &&
1282 type
!= PTR_TO_MAP_VALUE
&&
1283 type
!= expected_type
)
1285 meta
->raw_mode
= arg_type
== ARG_PTR_TO_UNINIT_MEM
;
1287 verbose(env
, "unsupported arg_type %d\n", arg_type
);
1291 if (arg_type
== ARG_CONST_MAP_PTR
) {
1292 /* bpf_map_xxx(map_ptr) call: remember that map_ptr */
1293 meta
->map_ptr
= reg
->map_ptr
;
1294 } else if (arg_type
== ARG_PTR_TO_MAP_KEY
) {
1295 /* bpf_map_xxx(..., map_ptr, ..., key) call:
1296 * check that [key, key + map->key_size) are within
1297 * stack limits and initialized
1299 if (!meta
->map_ptr
) {
1300 /* in function declaration map_ptr must come before
1301 * map_key, so that it's verified and known before
1302 * we have to check map_key here. Otherwise it means
1303 * that kernel subsystem misconfigured verifier
1305 verbose(env
, "invalid map_ptr to access map->key\n");
1308 if (type_is_pkt_pointer(type
))
1309 err
= check_packet_access(env
, regno
, reg
->off
,
1310 meta
->map_ptr
->key_size
);
1312 err
= check_stack_boundary(env
, regno
,
1313 meta
->map_ptr
->key_size
,
1315 } else if (arg_type
== ARG_PTR_TO_MAP_VALUE
) {
1316 /* bpf_map_xxx(..., map_ptr, ..., value) call:
1317 * check [value, value + map->value_size) validity
1319 if (!meta
->map_ptr
) {
1320 /* kernel subsystem misconfigured verifier */
1321 verbose(env
, "invalid map_ptr to access map->value\n");
1324 if (type_is_pkt_pointer(type
))
1325 err
= check_packet_access(env
, regno
, reg
->off
,
1326 meta
->map_ptr
->value_size
);
1328 err
= check_stack_boundary(env
, regno
,
1329 meta
->map_ptr
->value_size
,
1331 } else if (arg_type
== ARG_CONST_SIZE
||
1332 arg_type
== ARG_CONST_SIZE_OR_ZERO
) {
1333 bool zero_size_allowed
= (arg_type
== ARG_CONST_SIZE_OR_ZERO
);
1335 /* bpf_xxx(..., buf, len) call will access 'len' bytes
1336 * from stack pointer 'buf'. Check it
1337 * note: regno == len, regno - 1 == buf
1340 /* kernel subsystem misconfigured verifier */
1342 "ARG_CONST_SIZE cannot be first argument\n");
1346 /* The register is SCALAR_VALUE; the access check
1347 * happens using its boundaries.
1350 if (!tnum_is_const(reg
->var_off
))
1351 /* For unprivileged variable accesses, disable raw
1352 * mode so that the program is required to
1353 * initialize all the memory that the helper could
1354 * just partially fill up.
1358 if (reg
->smin_value
< 0) {
1359 verbose(env
, "R%d min value is negative, either use unsigned or 'var &= const'\n",
1364 if (reg
->umin_value
== 0) {
1365 err
= check_helper_mem_access(env
, regno
- 1, 0,
1372 if (reg
->umax_value
>= BPF_MAX_VAR_SIZ
) {
1373 verbose(env
, "R%d unbounded memory access, use 'var &= const' or 'if (var < const)'\n",
1377 err
= check_helper_mem_access(env
, regno
- 1,
1379 zero_size_allowed
, meta
);
1384 verbose(env
, "R%d type=%s expected=%s\n", regno
,
1385 reg_type_str
[type
], reg_type_str
[expected_type
]);
1389 static int check_map_func_compatibility(struct bpf_verifier_env
*env
,
1390 struct bpf_map
*map
, int func_id
)
1395 /* We need a two way check, first is from map perspective ... */
1396 switch (map
->map_type
) {
1397 case BPF_MAP_TYPE_PROG_ARRAY
:
1398 if (func_id
!= BPF_FUNC_tail_call
)
1401 case BPF_MAP_TYPE_PERF_EVENT_ARRAY
:
1402 if (func_id
!= BPF_FUNC_perf_event_read
&&
1403 func_id
!= BPF_FUNC_perf_event_output
&&
1404 func_id
!= BPF_FUNC_perf_event_read_value
)
1407 case BPF_MAP_TYPE_STACK_TRACE
:
1408 if (func_id
!= BPF_FUNC_get_stackid
)
1411 case BPF_MAP_TYPE_CGROUP_ARRAY
:
1412 if (func_id
!= BPF_FUNC_skb_under_cgroup
&&
1413 func_id
!= BPF_FUNC_current_task_under_cgroup
)
1416 /* devmap returns a pointer to a live net_device ifindex that we cannot
1417 * allow to be modified from bpf side. So do not allow lookup elements
1420 case BPF_MAP_TYPE_DEVMAP
:
1421 if (func_id
!= BPF_FUNC_redirect_map
)
1424 /* Restrict bpf side of cpumap, open when use-cases appear */
1425 case BPF_MAP_TYPE_CPUMAP
:
1426 if (func_id
!= BPF_FUNC_redirect_map
)
1429 case BPF_MAP_TYPE_ARRAY_OF_MAPS
:
1430 case BPF_MAP_TYPE_HASH_OF_MAPS
:
1431 if (func_id
!= BPF_FUNC_map_lookup_elem
)
1434 case BPF_MAP_TYPE_SOCKMAP
:
1435 if (func_id
!= BPF_FUNC_sk_redirect_map
&&
1436 func_id
!= BPF_FUNC_sock_map_update
&&
1437 func_id
!= BPF_FUNC_map_delete_elem
)
1444 /* ... and second from the function itself. */
1446 case BPF_FUNC_tail_call
:
1447 if (map
->map_type
!= BPF_MAP_TYPE_PROG_ARRAY
)
1450 case BPF_FUNC_perf_event_read
:
1451 case BPF_FUNC_perf_event_output
:
1452 case BPF_FUNC_perf_event_read_value
:
1453 if (map
->map_type
!= BPF_MAP_TYPE_PERF_EVENT_ARRAY
)
1456 case BPF_FUNC_get_stackid
:
1457 if (map
->map_type
!= BPF_MAP_TYPE_STACK_TRACE
)
1460 case BPF_FUNC_current_task_under_cgroup
:
1461 case BPF_FUNC_skb_under_cgroup
:
1462 if (map
->map_type
!= BPF_MAP_TYPE_CGROUP_ARRAY
)
1465 case BPF_FUNC_redirect_map
:
1466 if (map
->map_type
!= BPF_MAP_TYPE_DEVMAP
&&
1467 map
->map_type
!= BPF_MAP_TYPE_CPUMAP
)
1470 case BPF_FUNC_sk_redirect_map
:
1471 if (map
->map_type
!= BPF_MAP_TYPE_SOCKMAP
)
1474 case BPF_FUNC_sock_map_update
:
1475 if (map
->map_type
!= BPF_MAP_TYPE_SOCKMAP
)
1484 verbose(env
, "cannot pass map_type %d into func %s#%d\n",
1485 map
->map_type
, func_id_name(func_id
), func_id
);
1489 static int check_raw_mode(const struct bpf_func_proto
*fn
)
1493 if (fn
->arg1_type
== ARG_PTR_TO_UNINIT_MEM
)
1495 if (fn
->arg2_type
== ARG_PTR_TO_UNINIT_MEM
)
1497 if (fn
->arg3_type
== ARG_PTR_TO_UNINIT_MEM
)
1499 if (fn
->arg4_type
== ARG_PTR_TO_UNINIT_MEM
)
1501 if (fn
->arg5_type
== ARG_PTR_TO_UNINIT_MEM
)
1504 return count
> 1 ? -EINVAL
: 0;
1507 /* Packet data might have moved, any old PTR_TO_PACKET[_META,_END]
1508 * are now invalid, so turn them into unknown SCALAR_VALUE.
1510 static void clear_all_pkt_pointers(struct bpf_verifier_env
*env
)
1512 struct bpf_verifier_state
*state
= &env
->cur_state
;
1513 struct bpf_reg_state
*regs
= state
->regs
, *reg
;
1516 for (i
= 0; i
< MAX_BPF_REG
; i
++)
1517 if (reg_is_pkt_pointer_any(®s
[i
]))
1518 mark_reg_unknown(env
, regs
, i
);
1520 for (i
= 0; i
< MAX_BPF_STACK
; i
+= BPF_REG_SIZE
) {
1521 if (state
->stack_slot_type
[i
] != STACK_SPILL
)
1523 reg
= &state
->spilled_regs
[i
/ BPF_REG_SIZE
];
1524 if (reg_is_pkt_pointer_any(reg
))
1525 __mark_reg_unknown(reg
);
1529 static int check_call(struct bpf_verifier_env
*env
, int func_id
, int insn_idx
)
1531 struct bpf_verifier_state
*state
= &env
->cur_state
;
1532 const struct bpf_func_proto
*fn
= NULL
;
1533 struct bpf_reg_state
*regs
= state
->regs
;
1534 struct bpf_call_arg_meta meta
;
1538 /* find function prototype */
1539 if (func_id
< 0 || func_id
>= __BPF_FUNC_MAX_ID
) {
1540 verbose(env
, "invalid func %s#%d\n", func_id_name(func_id
),
1545 if (env
->ops
->get_func_proto
)
1546 fn
= env
->ops
->get_func_proto(func_id
);
1549 verbose(env
, "unknown func %s#%d\n", func_id_name(func_id
),
1554 /* eBPF programs must be GPL compatible to use GPL-ed functions */
1555 if (!env
->prog
->gpl_compatible
&& fn
->gpl_only
) {
1556 verbose(env
, "cannot call GPL only function from proprietary program\n");
1560 changes_data
= bpf_helper_changes_pkt_data(fn
->func
);
1562 memset(&meta
, 0, sizeof(meta
));
1563 meta
.pkt_access
= fn
->pkt_access
;
1565 /* We only support one arg being in raw mode at the moment, which
1566 * is sufficient for the helper functions we have right now.
1568 err
= check_raw_mode(fn
);
1570 verbose(env
, "kernel subsystem misconfigured func %s#%d\n",
1571 func_id_name(func_id
), func_id
);
1576 err
= check_func_arg(env
, BPF_REG_1
, fn
->arg1_type
, &meta
);
1579 err
= check_func_arg(env
, BPF_REG_2
, fn
->arg2_type
, &meta
);
1582 err
= check_func_arg(env
, BPF_REG_3
, fn
->arg3_type
, &meta
);
1585 err
= check_func_arg(env
, BPF_REG_4
, fn
->arg4_type
, &meta
);
1588 err
= check_func_arg(env
, BPF_REG_5
, fn
->arg5_type
, &meta
);
1592 /* Mark slots with STACK_MISC in case of raw mode, stack offset
1593 * is inferred from register state.
1595 for (i
= 0; i
< meta
.access_size
; i
++) {
1596 err
= check_mem_access(env
, insn_idx
, meta
.regno
, i
, BPF_B
, BPF_WRITE
, -1);
1601 /* reset caller saved regs */
1602 for (i
= 0; i
< CALLER_SAVED_REGS
; i
++) {
1603 mark_reg_not_init(env
, regs
, caller_saved
[i
]);
1604 check_reg_arg(env
, caller_saved
[i
], DST_OP_NO_MARK
);
1607 /* update return register (already marked as written above) */
1608 if (fn
->ret_type
== RET_INTEGER
) {
1609 /* sets type to SCALAR_VALUE */
1610 mark_reg_unknown(env
, regs
, BPF_REG_0
);
1611 } else if (fn
->ret_type
== RET_VOID
) {
1612 regs
[BPF_REG_0
].type
= NOT_INIT
;
1613 } else if (fn
->ret_type
== RET_PTR_TO_MAP_VALUE_OR_NULL
) {
1614 struct bpf_insn_aux_data
*insn_aux
;
1616 regs
[BPF_REG_0
].type
= PTR_TO_MAP_VALUE_OR_NULL
;
1617 /* There is no offset yet applied, variable or fixed */
1618 mark_reg_known_zero(env
, regs
, BPF_REG_0
);
1619 regs
[BPF_REG_0
].off
= 0;
1620 /* remember map_ptr, so that check_map_access()
1621 * can check 'value_size' boundary of memory access
1622 * to map element returned from bpf_map_lookup_elem()
1624 if (meta
.map_ptr
== NULL
) {
1626 "kernel subsystem misconfigured verifier\n");
1629 regs
[BPF_REG_0
].map_ptr
= meta
.map_ptr
;
1630 regs
[BPF_REG_0
].id
= ++env
->id_gen
;
1631 insn_aux
= &env
->insn_aux_data
[insn_idx
];
1632 if (!insn_aux
->map_ptr
)
1633 insn_aux
->map_ptr
= meta
.map_ptr
;
1634 else if (insn_aux
->map_ptr
!= meta
.map_ptr
)
1635 insn_aux
->map_ptr
= BPF_MAP_PTR_POISON
;
1637 verbose(env
, "unknown return type %d of func %s#%d\n",
1638 fn
->ret_type
, func_id_name(func_id
), func_id
);
1642 err
= check_map_func_compatibility(env
, meta
.map_ptr
, func_id
);
1647 clear_all_pkt_pointers(env
);
1651 static void coerce_reg_to_32(struct bpf_reg_state
*reg
)
1653 /* clear high 32 bits */
1654 reg
->var_off
= tnum_cast(reg
->var_off
, 4);
1656 __update_reg_bounds(reg
);
1659 static bool signed_add_overflows(s64 a
, s64 b
)
1661 /* Do the add in u64, where overflow is well-defined */
1662 s64 res
= (s64
)((u64
)a
+ (u64
)b
);
1669 static bool signed_sub_overflows(s64 a
, s64 b
)
1671 /* Do the sub in u64, where overflow is well-defined */
1672 s64 res
= (s64
)((u64
)a
- (u64
)b
);
1679 /* Handles arithmetic on a pointer and a scalar: computes new min/max and var_off.
1680 * Caller should also handle BPF_MOV case separately.
1681 * If we return -EACCES, caller may want to try again treating pointer as a
1682 * scalar. So we only emit a diagnostic if !env->allow_ptr_leaks.
1684 static int adjust_ptr_min_max_vals(struct bpf_verifier_env
*env
,
1685 struct bpf_insn
*insn
,
1686 const struct bpf_reg_state
*ptr_reg
,
1687 const struct bpf_reg_state
*off_reg
)
1689 struct bpf_reg_state
*regs
= env
->cur_state
.regs
, *dst_reg
;
1690 bool known
= tnum_is_const(off_reg
->var_off
);
1691 s64 smin_val
= off_reg
->smin_value
, smax_val
= off_reg
->smax_value
,
1692 smin_ptr
= ptr_reg
->smin_value
, smax_ptr
= ptr_reg
->smax_value
;
1693 u64 umin_val
= off_reg
->umin_value
, umax_val
= off_reg
->umax_value
,
1694 umin_ptr
= ptr_reg
->umin_value
, umax_ptr
= ptr_reg
->umax_value
;
1695 u8 opcode
= BPF_OP(insn
->code
);
1696 u32 dst
= insn
->dst_reg
;
1698 dst_reg
= ®s
[dst
];
1700 if (WARN_ON_ONCE(known
&& (smin_val
!= smax_val
))) {
1701 print_verifier_state(env
, &env
->cur_state
);
1703 "verifier internal error: known but bad sbounds\n");
1706 if (WARN_ON_ONCE(known
&& (umin_val
!= umax_val
))) {
1707 print_verifier_state(env
, &env
->cur_state
);
1709 "verifier internal error: known but bad ubounds\n");
1713 if (BPF_CLASS(insn
->code
) != BPF_ALU64
) {
1714 /* 32-bit ALU ops on pointers produce (meaningless) scalars */
1715 if (!env
->allow_ptr_leaks
)
1717 "R%d 32-bit pointer arithmetic prohibited\n",
1722 if (ptr_reg
->type
== PTR_TO_MAP_VALUE_OR_NULL
) {
1723 if (!env
->allow_ptr_leaks
)
1724 verbose(env
, "R%d pointer arithmetic on PTR_TO_MAP_VALUE_OR_NULL prohibited, null-check it first\n",
1728 if (ptr_reg
->type
== CONST_PTR_TO_MAP
) {
1729 if (!env
->allow_ptr_leaks
)
1730 verbose(env
, "R%d pointer arithmetic on CONST_PTR_TO_MAP prohibited\n",
1734 if (ptr_reg
->type
== PTR_TO_PACKET_END
) {
1735 if (!env
->allow_ptr_leaks
)
1736 verbose(env
, "R%d pointer arithmetic on PTR_TO_PACKET_END prohibited\n",
1741 /* In case of 'scalar += pointer', dst_reg inherits pointer type and id.
1742 * The id may be overwritten later if we create a new variable offset.
1744 dst_reg
->type
= ptr_reg
->type
;
1745 dst_reg
->id
= ptr_reg
->id
;
1749 /* We can take a fixed offset as long as it doesn't overflow
1750 * the s32 'off' field
1752 if (known
&& (ptr_reg
->off
+ smin_val
==
1753 (s64
)(s32
)(ptr_reg
->off
+ smin_val
))) {
1754 /* pointer += K. Accumulate it into fixed offset */
1755 dst_reg
->smin_value
= smin_ptr
;
1756 dst_reg
->smax_value
= smax_ptr
;
1757 dst_reg
->umin_value
= umin_ptr
;
1758 dst_reg
->umax_value
= umax_ptr
;
1759 dst_reg
->var_off
= ptr_reg
->var_off
;
1760 dst_reg
->off
= ptr_reg
->off
+ smin_val
;
1761 dst_reg
->range
= ptr_reg
->range
;
1764 /* A new variable offset is created. Note that off_reg->off
1765 * == 0, since it's a scalar.
1766 * dst_reg gets the pointer type and since some positive
1767 * integer value was added to the pointer, give it a new 'id'
1768 * if it's a PTR_TO_PACKET.
1769 * this creates a new 'base' pointer, off_reg (variable) gets
1770 * added into the variable offset, and we copy the fixed offset
1773 if (signed_add_overflows(smin_ptr
, smin_val
) ||
1774 signed_add_overflows(smax_ptr
, smax_val
)) {
1775 dst_reg
->smin_value
= S64_MIN
;
1776 dst_reg
->smax_value
= S64_MAX
;
1778 dst_reg
->smin_value
= smin_ptr
+ smin_val
;
1779 dst_reg
->smax_value
= smax_ptr
+ smax_val
;
1781 if (umin_ptr
+ umin_val
< umin_ptr
||
1782 umax_ptr
+ umax_val
< umax_ptr
) {
1783 dst_reg
->umin_value
= 0;
1784 dst_reg
->umax_value
= U64_MAX
;
1786 dst_reg
->umin_value
= umin_ptr
+ umin_val
;
1787 dst_reg
->umax_value
= umax_ptr
+ umax_val
;
1789 dst_reg
->var_off
= tnum_add(ptr_reg
->var_off
, off_reg
->var_off
);
1790 dst_reg
->off
= ptr_reg
->off
;
1791 if (reg_is_pkt_pointer(ptr_reg
)) {
1792 dst_reg
->id
= ++env
->id_gen
;
1793 /* something was added to pkt_ptr, set range to zero */
1798 if (dst_reg
== off_reg
) {
1799 /* scalar -= pointer. Creates an unknown scalar */
1800 if (!env
->allow_ptr_leaks
)
1801 verbose(env
, "R%d tried to subtract pointer from scalar\n",
1805 /* We don't allow subtraction from FP, because (according to
1806 * test_verifier.c test "invalid fp arithmetic", JITs might not
1807 * be able to deal with it.
1809 if (ptr_reg
->type
== PTR_TO_STACK
) {
1810 if (!env
->allow_ptr_leaks
)
1811 verbose(env
, "R%d subtraction from stack pointer prohibited\n",
1815 if (known
&& (ptr_reg
->off
- smin_val
==
1816 (s64
)(s32
)(ptr_reg
->off
- smin_val
))) {
1817 /* pointer -= K. Subtract it from fixed offset */
1818 dst_reg
->smin_value
= smin_ptr
;
1819 dst_reg
->smax_value
= smax_ptr
;
1820 dst_reg
->umin_value
= umin_ptr
;
1821 dst_reg
->umax_value
= umax_ptr
;
1822 dst_reg
->var_off
= ptr_reg
->var_off
;
1823 dst_reg
->id
= ptr_reg
->id
;
1824 dst_reg
->off
= ptr_reg
->off
- smin_val
;
1825 dst_reg
->range
= ptr_reg
->range
;
1828 /* A new variable offset is created. If the subtrahend is known
1829 * nonnegative, then any reg->range we had before is still good.
1831 if (signed_sub_overflows(smin_ptr
, smax_val
) ||
1832 signed_sub_overflows(smax_ptr
, smin_val
)) {
1833 /* Overflow possible, we know nothing */
1834 dst_reg
->smin_value
= S64_MIN
;
1835 dst_reg
->smax_value
= S64_MAX
;
1837 dst_reg
->smin_value
= smin_ptr
- smax_val
;
1838 dst_reg
->smax_value
= smax_ptr
- smin_val
;
1840 if (umin_ptr
< umax_val
) {
1841 /* Overflow possible, we know nothing */
1842 dst_reg
->umin_value
= 0;
1843 dst_reg
->umax_value
= U64_MAX
;
1845 /* Cannot overflow (as long as bounds are consistent) */
1846 dst_reg
->umin_value
= umin_ptr
- umax_val
;
1847 dst_reg
->umax_value
= umax_ptr
- umin_val
;
1849 dst_reg
->var_off
= tnum_sub(ptr_reg
->var_off
, off_reg
->var_off
);
1850 dst_reg
->off
= ptr_reg
->off
;
1851 if (reg_is_pkt_pointer(ptr_reg
)) {
1852 dst_reg
->id
= ++env
->id_gen
;
1853 /* something was added to pkt_ptr, set range to zero */
1861 /* bitwise ops on pointers are troublesome, prohibit for now.
1862 * (However, in principle we could allow some cases, e.g.
1863 * ptr &= ~3 which would reduce min_value by 3.)
1865 if (!env
->allow_ptr_leaks
)
1866 verbose(env
, "R%d bitwise operator %s on pointer prohibited\n",
1867 dst
, bpf_alu_string
[opcode
>> 4]);
1870 /* other operators (e.g. MUL,LSH) produce non-pointer results */
1871 if (!env
->allow_ptr_leaks
)
1872 verbose(env
, "R%d pointer arithmetic with %s operator prohibited\n",
1873 dst
, bpf_alu_string
[opcode
>> 4]);
1877 __update_reg_bounds(dst_reg
);
1878 __reg_deduce_bounds(dst_reg
);
1879 __reg_bound_offset(dst_reg
);
1883 static int adjust_scalar_min_max_vals(struct bpf_verifier_env
*env
,
1884 struct bpf_insn
*insn
,
1885 struct bpf_reg_state
*dst_reg
,
1886 struct bpf_reg_state src_reg
)
1888 struct bpf_reg_state
*regs
= env
->cur_state
.regs
;
1889 u8 opcode
= BPF_OP(insn
->code
);
1890 bool src_known
, dst_known
;
1891 s64 smin_val
, smax_val
;
1892 u64 umin_val
, umax_val
;
1894 if (BPF_CLASS(insn
->code
) != BPF_ALU64
) {
1895 /* 32-bit ALU ops are (32,32)->64 */
1896 coerce_reg_to_32(dst_reg
);
1897 coerce_reg_to_32(&src_reg
);
1899 smin_val
= src_reg
.smin_value
;
1900 smax_val
= src_reg
.smax_value
;
1901 umin_val
= src_reg
.umin_value
;
1902 umax_val
= src_reg
.umax_value
;
1903 src_known
= tnum_is_const(src_reg
.var_off
);
1904 dst_known
= tnum_is_const(dst_reg
->var_off
);
1908 if (signed_add_overflows(dst_reg
->smin_value
, smin_val
) ||
1909 signed_add_overflows(dst_reg
->smax_value
, smax_val
)) {
1910 dst_reg
->smin_value
= S64_MIN
;
1911 dst_reg
->smax_value
= S64_MAX
;
1913 dst_reg
->smin_value
+= smin_val
;
1914 dst_reg
->smax_value
+= smax_val
;
1916 if (dst_reg
->umin_value
+ umin_val
< umin_val
||
1917 dst_reg
->umax_value
+ umax_val
< umax_val
) {
1918 dst_reg
->umin_value
= 0;
1919 dst_reg
->umax_value
= U64_MAX
;
1921 dst_reg
->umin_value
+= umin_val
;
1922 dst_reg
->umax_value
+= umax_val
;
1924 dst_reg
->var_off
= tnum_add(dst_reg
->var_off
, src_reg
.var_off
);
1927 if (signed_sub_overflows(dst_reg
->smin_value
, smax_val
) ||
1928 signed_sub_overflows(dst_reg
->smax_value
, smin_val
)) {
1929 /* Overflow possible, we know nothing */
1930 dst_reg
->smin_value
= S64_MIN
;
1931 dst_reg
->smax_value
= S64_MAX
;
1933 dst_reg
->smin_value
-= smax_val
;
1934 dst_reg
->smax_value
-= smin_val
;
1936 if (dst_reg
->umin_value
< umax_val
) {
1937 /* Overflow possible, we know nothing */
1938 dst_reg
->umin_value
= 0;
1939 dst_reg
->umax_value
= U64_MAX
;
1941 /* Cannot overflow (as long as bounds are consistent) */
1942 dst_reg
->umin_value
-= umax_val
;
1943 dst_reg
->umax_value
-= umin_val
;
1945 dst_reg
->var_off
= tnum_sub(dst_reg
->var_off
, src_reg
.var_off
);
1948 dst_reg
->var_off
= tnum_mul(dst_reg
->var_off
, src_reg
.var_off
);
1949 if (smin_val
< 0 || dst_reg
->smin_value
< 0) {
1950 /* Ain't nobody got time to multiply that sign */
1951 __mark_reg_unbounded(dst_reg
);
1952 __update_reg_bounds(dst_reg
);
1955 /* Both values are positive, so we can work with unsigned and
1956 * copy the result to signed (unless it exceeds S64_MAX).
1958 if (umax_val
> U32_MAX
|| dst_reg
->umax_value
> U32_MAX
) {
1959 /* Potential overflow, we know nothing */
1960 __mark_reg_unbounded(dst_reg
);
1961 /* (except what we can learn from the var_off) */
1962 __update_reg_bounds(dst_reg
);
1965 dst_reg
->umin_value
*= umin_val
;
1966 dst_reg
->umax_value
*= umax_val
;
1967 if (dst_reg
->umax_value
> S64_MAX
) {
1968 /* Overflow possible, we know nothing */
1969 dst_reg
->smin_value
= S64_MIN
;
1970 dst_reg
->smax_value
= S64_MAX
;
1972 dst_reg
->smin_value
= dst_reg
->umin_value
;
1973 dst_reg
->smax_value
= dst_reg
->umax_value
;
1977 if (src_known
&& dst_known
) {
1978 __mark_reg_known(dst_reg
, dst_reg
->var_off
.value
&
1979 src_reg
.var_off
.value
);
1982 /* We get our minimum from the var_off, since that's inherently
1983 * bitwise. Our maximum is the minimum of the operands' maxima.
1985 dst_reg
->var_off
= tnum_and(dst_reg
->var_off
, src_reg
.var_off
);
1986 dst_reg
->umin_value
= dst_reg
->var_off
.value
;
1987 dst_reg
->umax_value
= min(dst_reg
->umax_value
, umax_val
);
1988 if (dst_reg
->smin_value
< 0 || smin_val
< 0) {
1989 /* Lose signed bounds when ANDing negative numbers,
1990 * ain't nobody got time for that.
1992 dst_reg
->smin_value
= S64_MIN
;
1993 dst_reg
->smax_value
= S64_MAX
;
1995 /* ANDing two positives gives a positive, so safe to
1996 * cast result into s64.
1998 dst_reg
->smin_value
= dst_reg
->umin_value
;
1999 dst_reg
->smax_value
= dst_reg
->umax_value
;
2001 /* We may learn something more from the var_off */
2002 __update_reg_bounds(dst_reg
);
2005 if (src_known
&& dst_known
) {
2006 __mark_reg_known(dst_reg
, dst_reg
->var_off
.value
|
2007 src_reg
.var_off
.value
);
2010 /* We get our maximum from the var_off, and our minimum is the
2011 * maximum of the operands' minima
2013 dst_reg
->var_off
= tnum_or(dst_reg
->var_off
, src_reg
.var_off
);
2014 dst_reg
->umin_value
= max(dst_reg
->umin_value
, umin_val
);
2015 dst_reg
->umax_value
= dst_reg
->var_off
.value
|
2016 dst_reg
->var_off
.mask
;
2017 if (dst_reg
->smin_value
< 0 || smin_val
< 0) {
2018 /* Lose signed bounds when ORing negative numbers,
2019 * ain't nobody got time for that.
2021 dst_reg
->smin_value
= S64_MIN
;
2022 dst_reg
->smax_value
= S64_MAX
;
2024 /* ORing two positives gives a positive, so safe to
2025 * cast result into s64.
2027 dst_reg
->smin_value
= dst_reg
->umin_value
;
2028 dst_reg
->smax_value
= dst_reg
->umax_value
;
2030 /* We may learn something more from the var_off */
2031 __update_reg_bounds(dst_reg
);
2034 if (umax_val
> 63) {
2035 /* Shifts greater than 63 are undefined. This includes
2036 * shifts by a negative number.
2038 mark_reg_unknown(env
, regs
, insn
->dst_reg
);
2041 /* We lose all sign bit information (except what we can pick
2044 dst_reg
->smin_value
= S64_MIN
;
2045 dst_reg
->smax_value
= S64_MAX
;
2046 /* If we might shift our top bit out, then we know nothing */
2047 if (dst_reg
->umax_value
> 1ULL << (63 - umax_val
)) {
2048 dst_reg
->umin_value
= 0;
2049 dst_reg
->umax_value
= U64_MAX
;
2051 dst_reg
->umin_value
<<= umin_val
;
2052 dst_reg
->umax_value
<<= umax_val
;
2055 dst_reg
->var_off
= tnum_lshift(dst_reg
->var_off
, umin_val
);
2057 dst_reg
->var_off
= tnum_lshift(tnum_unknown
, umin_val
);
2058 /* We may learn something more from the var_off */
2059 __update_reg_bounds(dst_reg
);
2062 if (umax_val
> 63) {
2063 /* Shifts greater than 63 are undefined. This includes
2064 * shifts by a negative number.
2066 mark_reg_unknown(env
, regs
, insn
->dst_reg
);
2069 /* BPF_RSH is an unsigned shift, so make the appropriate casts */
2070 if (dst_reg
->smin_value
< 0) {
2072 /* Sign bit will be cleared */
2073 dst_reg
->smin_value
= 0;
2075 /* Lost sign bit information */
2076 dst_reg
->smin_value
= S64_MIN
;
2077 dst_reg
->smax_value
= S64_MAX
;
2080 dst_reg
->smin_value
=
2081 (u64
)(dst_reg
->smin_value
) >> umax_val
;
2084 dst_reg
->var_off
= tnum_rshift(dst_reg
->var_off
,
2087 dst_reg
->var_off
= tnum_rshift(tnum_unknown
, umin_val
);
2088 dst_reg
->umin_value
>>= umax_val
;
2089 dst_reg
->umax_value
>>= umin_val
;
2090 /* We may learn something more from the var_off */
2091 __update_reg_bounds(dst_reg
);
2094 mark_reg_unknown(env
, regs
, insn
->dst_reg
);
2098 __reg_deduce_bounds(dst_reg
);
2099 __reg_bound_offset(dst_reg
);
2103 /* Handles ALU ops other than BPF_END, BPF_NEG and BPF_MOV: computes new min/max
2106 static int adjust_reg_min_max_vals(struct bpf_verifier_env
*env
,
2107 struct bpf_insn
*insn
)
2109 struct bpf_reg_state
*regs
= env
->cur_state
.regs
, *dst_reg
, *src_reg
;
2110 struct bpf_reg_state
*ptr_reg
= NULL
, off_reg
= {0};
2111 u8 opcode
= BPF_OP(insn
->code
);
2114 dst_reg
= ®s
[insn
->dst_reg
];
2116 if (dst_reg
->type
!= SCALAR_VALUE
)
2118 if (BPF_SRC(insn
->code
) == BPF_X
) {
2119 src_reg
= ®s
[insn
->src_reg
];
2120 if (src_reg
->type
!= SCALAR_VALUE
) {
2121 if (dst_reg
->type
!= SCALAR_VALUE
) {
2122 /* Combining two pointers by any ALU op yields
2123 * an arbitrary scalar.
2125 if (!env
->allow_ptr_leaks
) {
2126 verbose(env
, "R%d pointer %s pointer prohibited\n",
2128 bpf_alu_string
[opcode
>> 4]);
2131 mark_reg_unknown(env
, regs
, insn
->dst_reg
);
2134 /* scalar += pointer
2135 * This is legal, but we have to reverse our
2136 * src/dest handling in computing the range
2138 rc
= adjust_ptr_min_max_vals(env
, insn
,
2140 if (rc
== -EACCES
&& env
->allow_ptr_leaks
) {
2141 /* scalar += unknown scalar */
2142 __mark_reg_unknown(&off_reg
);
2143 return adjust_scalar_min_max_vals(
2149 } else if (ptr_reg
) {
2150 /* pointer += scalar */
2151 rc
= adjust_ptr_min_max_vals(env
, insn
,
2153 if (rc
== -EACCES
&& env
->allow_ptr_leaks
) {
2154 /* unknown scalar += scalar */
2155 __mark_reg_unknown(dst_reg
);
2156 return adjust_scalar_min_max_vals(
2157 env
, insn
, dst_reg
, *src_reg
);
2162 /* Pretend the src is a reg with a known value, since we only
2163 * need to be able to read from this state.
2165 off_reg
.type
= SCALAR_VALUE
;
2166 __mark_reg_known(&off_reg
, insn
->imm
);
2168 if (ptr_reg
) { /* pointer += K */
2169 rc
= adjust_ptr_min_max_vals(env
, insn
,
2171 if (rc
== -EACCES
&& env
->allow_ptr_leaks
) {
2172 /* unknown scalar += K */
2173 __mark_reg_unknown(dst_reg
);
2174 return adjust_scalar_min_max_vals(
2175 env
, insn
, dst_reg
, off_reg
);
2181 /* Got here implies adding two SCALAR_VALUEs */
2182 if (WARN_ON_ONCE(ptr_reg
)) {
2183 print_verifier_state(env
, &env
->cur_state
);
2184 verbose(env
, "verifier internal error: unexpected ptr_reg\n");
2187 if (WARN_ON(!src_reg
)) {
2188 print_verifier_state(env
, &env
->cur_state
);
2189 verbose(env
, "verifier internal error: no src_reg\n");
2192 return adjust_scalar_min_max_vals(env
, insn
, dst_reg
, *src_reg
);
2195 /* check validity of 32-bit and 64-bit arithmetic operations */
2196 static int check_alu_op(struct bpf_verifier_env
*env
, struct bpf_insn
*insn
)
2198 struct bpf_reg_state
*regs
= env
->cur_state
.regs
;
2199 u8 opcode
= BPF_OP(insn
->code
);
2202 if (opcode
== BPF_END
|| opcode
== BPF_NEG
) {
2203 if (opcode
== BPF_NEG
) {
2204 if (BPF_SRC(insn
->code
) != 0 ||
2205 insn
->src_reg
!= BPF_REG_0
||
2206 insn
->off
!= 0 || insn
->imm
!= 0) {
2207 verbose(env
, "BPF_NEG uses reserved fields\n");
2211 if (insn
->src_reg
!= BPF_REG_0
|| insn
->off
!= 0 ||
2212 (insn
->imm
!= 16 && insn
->imm
!= 32 && insn
->imm
!= 64) ||
2213 BPF_CLASS(insn
->code
) == BPF_ALU64
) {
2214 verbose(env
, "BPF_END uses reserved fields\n");
2219 /* check src operand */
2220 err
= check_reg_arg(env
, insn
->dst_reg
, SRC_OP
);
2224 if (is_pointer_value(env
, insn
->dst_reg
)) {
2225 verbose(env
, "R%d pointer arithmetic prohibited\n",
2230 /* check dest operand */
2231 err
= check_reg_arg(env
, insn
->dst_reg
, DST_OP
);
2235 } else if (opcode
== BPF_MOV
) {
2237 if (BPF_SRC(insn
->code
) == BPF_X
) {
2238 if (insn
->imm
!= 0 || insn
->off
!= 0) {
2239 verbose(env
, "BPF_MOV uses reserved fields\n");
2243 /* check src operand */
2244 err
= check_reg_arg(env
, insn
->src_reg
, SRC_OP
);
2248 if (insn
->src_reg
!= BPF_REG_0
|| insn
->off
!= 0) {
2249 verbose(env
, "BPF_MOV uses reserved fields\n");
2254 /* check dest operand */
2255 err
= check_reg_arg(env
, insn
->dst_reg
, DST_OP
);
2259 if (BPF_SRC(insn
->code
) == BPF_X
) {
2260 if (BPF_CLASS(insn
->code
) == BPF_ALU64
) {
2262 * copy register state to dest reg
2264 regs
[insn
->dst_reg
] = regs
[insn
->src_reg
];
2265 regs
[insn
->dst_reg
].live
|= REG_LIVE_WRITTEN
;
2268 if (is_pointer_value(env
, insn
->src_reg
)) {
2270 "R%d partial copy of pointer\n",
2274 mark_reg_unknown(env
, regs
, insn
->dst_reg
);
2275 /* high 32 bits are known zero. */
2276 regs
[insn
->dst_reg
].var_off
= tnum_cast(
2277 regs
[insn
->dst_reg
].var_off
, 4);
2278 __update_reg_bounds(®s
[insn
->dst_reg
]);
2282 * remember the value we stored into this reg
2284 regs
[insn
->dst_reg
].type
= SCALAR_VALUE
;
2285 __mark_reg_known(regs
+ insn
->dst_reg
, insn
->imm
);
2288 } else if (opcode
> BPF_END
) {
2289 verbose(env
, "invalid BPF_ALU opcode %x\n", opcode
);
2292 } else { /* all other ALU ops: and, sub, xor, add, ... */
2294 if (BPF_SRC(insn
->code
) == BPF_X
) {
2295 if (insn
->imm
!= 0 || insn
->off
!= 0) {
2296 verbose(env
, "BPF_ALU uses reserved fields\n");
2299 /* check src1 operand */
2300 err
= check_reg_arg(env
, insn
->src_reg
, SRC_OP
);
2304 if (insn
->src_reg
!= BPF_REG_0
|| insn
->off
!= 0) {
2305 verbose(env
, "BPF_ALU uses reserved fields\n");
2310 /* check src2 operand */
2311 err
= check_reg_arg(env
, insn
->dst_reg
, SRC_OP
);
2315 if ((opcode
== BPF_MOD
|| opcode
== BPF_DIV
) &&
2316 BPF_SRC(insn
->code
) == BPF_K
&& insn
->imm
== 0) {
2317 verbose(env
, "div by zero\n");
2321 if ((opcode
== BPF_LSH
|| opcode
== BPF_RSH
||
2322 opcode
== BPF_ARSH
) && BPF_SRC(insn
->code
) == BPF_K
) {
2323 int size
= BPF_CLASS(insn
->code
) == BPF_ALU64
? 64 : 32;
2325 if (insn
->imm
< 0 || insn
->imm
>= size
) {
2326 verbose(env
, "invalid shift %d\n", insn
->imm
);
2331 /* check dest operand */
2332 err
= check_reg_arg(env
, insn
->dst_reg
, DST_OP_NO_MARK
);
2336 return adjust_reg_min_max_vals(env
, insn
);
2342 static void find_good_pkt_pointers(struct bpf_verifier_state
*state
,
2343 struct bpf_reg_state
*dst_reg
,
2344 enum bpf_reg_type type
)
2346 struct bpf_reg_state
*regs
= state
->regs
, *reg
;
2349 if (dst_reg
->off
< 0)
2350 /* This doesn't give us any range */
2353 if (dst_reg
->umax_value
> MAX_PACKET_OFF
||
2354 dst_reg
->umax_value
+ dst_reg
->off
> MAX_PACKET_OFF
)
2355 /* Risk of overflow. For instance, ptr + (1<<63) may be less
2356 * than pkt_end, but that's because it's also less than pkt.
2360 /* LLVM can generate four kind of checks:
2366 * if (r2 > pkt_end) goto <handle exception>
2371 * if (r2 < pkt_end) goto <access okay>
2372 * <handle exception>
2375 * r2 == dst_reg, pkt_end == src_reg
2376 * r2=pkt(id=n,off=8,r=0)
2377 * r3=pkt(id=n,off=0,r=0)
2383 * if (pkt_end >= r2) goto <access okay>
2384 * <handle exception>
2388 * if (pkt_end <= r2) goto <handle exception>
2392 * pkt_end == dst_reg, r2 == src_reg
2393 * r2=pkt(id=n,off=8,r=0)
2394 * r3=pkt(id=n,off=0,r=0)
2396 * Find register r3 and mark its range as r3=pkt(id=n,off=0,r=8)
2397 * so that range of bytes [r3, r3 + 8) is safe to access.
2400 /* If our ids match, then we must have the same max_value. And we
2401 * don't care about the other reg's fixed offset, since if it's too big
2402 * the range won't allow anything.
2403 * dst_reg->off is known < MAX_PACKET_OFF, therefore it fits in a u16.
2405 for (i
= 0; i
< MAX_BPF_REG
; i
++)
2406 if (regs
[i
].type
== type
&& regs
[i
].id
== dst_reg
->id
)
2407 /* keep the maximum range already checked */
2408 regs
[i
].range
= max_t(u16
, regs
[i
].range
, dst_reg
->off
);
2410 for (i
= 0; i
< MAX_BPF_STACK
; i
+= BPF_REG_SIZE
) {
2411 if (state
->stack_slot_type
[i
] != STACK_SPILL
)
2413 reg
= &state
->spilled_regs
[i
/ BPF_REG_SIZE
];
2414 if (reg
->type
== type
&& reg
->id
== dst_reg
->id
)
2415 reg
->range
= max_t(u16
, reg
->range
, dst_reg
->off
);
2419 /* Adjusts the register min/max values in the case that the dst_reg is the
2420 * variable register that we are working on, and src_reg is a constant or we're
2421 * simply doing a BPF_K check.
2422 * In JEQ/JNE cases we also adjust the var_off values.
2424 static void reg_set_min_max(struct bpf_reg_state
*true_reg
,
2425 struct bpf_reg_state
*false_reg
, u64 val
,
2428 /* If the dst_reg is a pointer, we can't learn anything about its
2429 * variable offset from the compare (unless src_reg were a pointer into
2430 * the same object, but we don't bother with that.
2431 * Since false_reg and true_reg have the same type by construction, we
2432 * only need to check one of them for pointerness.
2434 if (__is_pointer_value(false, false_reg
))
2439 /* If this is false then we know nothing Jon Snow, but if it is
2440 * true then we know for sure.
2442 __mark_reg_known(true_reg
, val
);
2445 /* If this is true we know nothing Jon Snow, but if it is false
2446 * we know the value for sure;
2448 __mark_reg_known(false_reg
, val
);
2451 false_reg
->umax_value
= min(false_reg
->umax_value
, val
);
2452 true_reg
->umin_value
= max(true_reg
->umin_value
, val
+ 1);
2455 false_reg
->smax_value
= min_t(s64
, false_reg
->smax_value
, val
);
2456 true_reg
->smin_value
= max_t(s64
, true_reg
->smin_value
, val
+ 1);
2459 false_reg
->umin_value
= max(false_reg
->umin_value
, val
);
2460 true_reg
->umax_value
= min(true_reg
->umax_value
, val
- 1);
2463 false_reg
->smin_value
= max_t(s64
, false_reg
->smin_value
, val
);
2464 true_reg
->smax_value
= min_t(s64
, true_reg
->smax_value
, val
- 1);
2467 false_reg
->umax_value
= min(false_reg
->umax_value
, val
- 1);
2468 true_reg
->umin_value
= max(true_reg
->umin_value
, val
);
2471 false_reg
->smax_value
= min_t(s64
, false_reg
->smax_value
, val
- 1);
2472 true_reg
->smin_value
= max_t(s64
, true_reg
->smin_value
, val
);
2475 false_reg
->umin_value
= max(false_reg
->umin_value
, val
+ 1);
2476 true_reg
->umax_value
= min(true_reg
->umax_value
, val
);
2479 false_reg
->smin_value
= max_t(s64
, false_reg
->smin_value
, val
+ 1);
2480 true_reg
->smax_value
= min_t(s64
, true_reg
->smax_value
, val
);
2486 __reg_deduce_bounds(false_reg
);
2487 __reg_deduce_bounds(true_reg
);
2488 /* We might have learned some bits from the bounds. */
2489 __reg_bound_offset(false_reg
);
2490 __reg_bound_offset(true_reg
);
2491 /* Intersecting with the old var_off might have improved our bounds
2492 * slightly. e.g. if umax was 0x7f...f and var_off was (0; 0xf...fc),
2493 * then new var_off is (0; 0x7f...fc) which improves our umax.
2495 __update_reg_bounds(false_reg
);
2496 __update_reg_bounds(true_reg
);
2499 /* Same as above, but for the case that dst_reg holds a constant and src_reg is
2502 static void reg_set_min_max_inv(struct bpf_reg_state
*true_reg
,
2503 struct bpf_reg_state
*false_reg
, u64 val
,
2506 if (__is_pointer_value(false, false_reg
))
2511 /* If this is false then we know nothing Jon Snow, but if it is
2512 * true then we know for sure.
2514 __mark_reg_known(true_reg
, val
);
2517 /* If this is true we know nothing Jon Snow, but if it is false
2518 * we know the value for sure;
2520 __mark_reg_known(false_reg
, val
);
2523 true_reg
->umax_value
= min(true_reg
->umax_value
, val
- 1);
2524 false_reg
->umin_value
= max(false_reg
->umin_value
, val
);
2527 true_reg
->smax_value
= min_t(s64
, true_reg
->smax_value
, val
- 1);
2528 false_reg
->smin_value
= max_t(s64
, false_reg
->smin_value
, val
);
2531 true_reg
->umin_value
= max(true_reg
->umin_value
, val
+ 1);
2532 false_reg
->umax_value
= min(false_reg
->umax_value
, val
);
2535 true_reg
->smin_value
= max_t(s64
, true_reg
->smin_value
, val
+ 1);
2536 false_reg
->smax_value
= min_t(s64
, false_reg
->smax_value
, val
);
2539 true_reg
->umax_value
= min(true_reg
->umax_value
, val
);
2540 false_reg
->umin_value
= max(false_reg
->umin_value
, val
+ 1);
2543 true_reg
->smax_value
= min_t(s64
, true_reg
->smax_value
, val
);
2544 false_reg
->smin_value
= max_t(s64
, false_reg
->smin_value
, val
+ 1);
2547 true_reg
->umin_value
= max(true_reg
->umin_value
, val
);
2548 false_reg
->umax_value
= min(false_reg
->umax_value
, val
- 1);
2551 true_reg
->smin_value
= max_t(s64
, true_reg
->smin_value
, val
);
2552 false_reg
->smax_value
= min_t(s64
, false_reg
->smax_value
, val
- 1);
2558 __reg_deduce_bounds(false_reg
);
2559 __reg_deduce_bounds(true_reg
);
2560 /* We might have learned some bits from the bounds. */
2561 __reg_bound_offset(false_reg
);
2562 __reg_bound_offset(true_reg
);
2563 /* Intersecting with the old var_off might have improved our bounds
2564 * slightly. e.g. if umax was 0x7f...f and var_off was (0; 0xf...fc),
2565 * then new var_off is (0; 0x7f...fc) which improves our umax.
2567 __update_reg_bounds(false_reg
);
2568 __update_reg_bounds(true_reg
);
2571 /* Regs are known to be equal, so intersect their min/max/var_off */
2572 static void __reg_combine_min_max(struct bpf_reg_state
*src_reg
,
2573 struct bpf_reg_state
*dst_reg
)
2575 src_reg
->umin_value
= dst_reg
->umin_value
= max(src_reg
->umin_value
,
2576 dst_reg
->umin_value
);
2577 src_reg
->umax_value
= dst_reg
->umax_value
= min(src_reg
->umax_value
,
2578 dst_reg
->umax_value
);
2579 src_reg
->smin_value
= dst_reg
->smin_value
= max(src_reg
->smin_value
,
2580 dst_reg
->smin_value
);
2581 src_reg
->smax_value
= dst_reg
->smax_value
= min(src_reg
->smax_value
,
2582 dst_reg
->smax_value
);
2583 src_reg
->var_off
= dst_reg
->var_off
= tnum_intersect(src_reg
->var_off
,
2585 /* We might have learned new bounds from the var_off. */
2586 __update_reg_bounds(src_reg
);
2587 __update_reg_bounds(dst_reg
);
2588 /* We might have learned something about the sign bit. */
2589 __reg_deduce_bounds(src_reg
);
2590 __reg_deduce_bounds(dst_reg
);
2591 /* We might have learned some bits from the bounds. */
2592 __reg_bound_offset(src_reg
);
2593 __reg_bound_offset(dst_reg
);
2594 /* Intersecting with the old var_off might have improved our bounds
2595 * slightly. e.g. if umax was 0x7f...f and var_off was (0; 0xf...fc),
2596 * then new var_off is (0; 0x7f...fc) which improves our umax.
2598 __update_reg_bounds(src_reg
);
2599 __update_reg_bounds(dst_reg
);
2602 static void reg_combine_min_max(struct bpf_reg_state
*true_src
,
2603 struct bpf_reg_state
*true_dst
,
2604 struct bpf_reg_state
*false_src
,
2605 struct bpf_reg_state
*false_dst
,
2610 __reg_combine_min_max(true_src
, true_dst
);
2613 __reg_combine_min_max(false_src
, false_dst
);
2618 static void mark_map_reg(struct bpf_reg_state
*regs
, u32 regno
, u32 id
,
2621 struct bpf_reg_state
*reg
= ®s
[regno
];
2623 if (reg
->type
== PTR_TO_MAP_VALUE_OR_NULL
&& reg
->id
== id
) {
2624 /* Old offset (both fixed and variable parts) should
2625 * have been known-zero, because we don't allow pointer
2626 * arithmetic on pointers that might be NULL.
2628 if (WARN_ON_ONCE(reg
->smin_value
|| reg
->smax_value
||
2629 !tnum_equals_const(reg
->var_off
, 0) ||
2631 __mark_reg_known_zero(reg
);
2635 reg
->type
= SCALAR_VALUE
;
2636 } else if (reg
->map_ptr
->inner_map_meta
) {
2637 reg
->type
= CONST_PTR_TO_MAP
;
2638 reg
->map_ptr
= reg
->map_ptr
->inner_map_meta
;
2640 reg
->type
= PTR_TO_MAP_VALUE
;
2642 /* We don't need id from this point onwards anymore, thus we
2643 * should better reset it, so that state pruning has chances
2650 /* The logic is similar to find_good_pkt_pointers(), both could eventually
2651 * be folded together at some point.
2653 static void mark_map_regs(struct bpf_verifier_state
*state
, u32 regno
,
2656 struct bpf_reg_state
*regs
= state
->regs
;
2657 u32 id
= regs
[regno
].id
;
2660 for (i
= 0; i
< MAX_BPF_REG
; i
++)
2661 mark_map_reg(regs
, i
, id
, is_null
);
2663 for (i
= 0; i
< MAX_BPF_STACK
; i
+= BPF_REG_SIZE
) {
2664 if (state
->stack_slot_type
[i
] != STACK_SPILL
)
2666 mark_map_reg(state
->spilled_regs
, i
/ BPF_REG_SIZE
, id
, is_null
);
2670 static int check_cond_jmp_op(struct bpf_verifier_env
*env
,
2671 struct bpf_insn
*insn
, int *insn_idx
)
2673 struct bpf_verifier_state
*other_branch
, *this_branch
= &env
->cur_state
;
2674 struct bpf_reg_state
*regs
= this_branch
->regs
, *dst_reg
;
2675 u8 opcode
= BPF_OP(insn
->code
);
2678 if (opcode
> BPF_JSLE
) {
2679 verbose(env
, "invalid BPF_JMP opcode %x\n", opcode
);
2683 if (BPF_SRC(insn
->code
) == BPF_X
) {
2684 if (insn
->imm
!= 0) {
2685 verbose(env
, "BPF_JMP uses reserved fields\n");
2689 /* check src1 operand */
2690 err
= check_reg_arg(env
, insn
->src_reg
, SRC_OP
);
2694 if (is_pointer_value(env
, insn
->src_reg
)) {
2695 verbose(env
, "R%d pointer comparison prohibited\n",
2700 if (insn
->src_reg
!= BPF_REG_0
) {
2701 verbose(env
, "BPF_JMP uses reserved fields\n");
2706 /* check src2 operand */
2707 err
= check_reg_arg(env
, insn
->dst_reg
, SRC_OP
);
2711 dst_reg
= ®s
[insn
->dst_reg
];
2713 /* detect if R == 0 where R was initialized to zero earlier */
2714 if (BPF_SRC(insn
->code
) == BPF_K
&&
2715 (opcode
== BPF_JEQ
|| opcode
== BPF_JNE
) &&
2716 dst_reg
->type
== SCALAR_VALUE
&&
2717 tnum_equals_const(dst_reg
->var_off
, insn
->imm
)) {
2718 if (opcode
== BPF_JEQ
) {
2719 /* if (imm == imm) goto pc+off;
2720 * only follow the goto, ignore fall-through
2722 *insn_idx
+= insn
->off
;
2725 /* if (imm != imm) goto pc+off;
2726 * only follow fall-through branch, since
2727 * that's where the program will go
2733 other_branch
= push_stack(env
, *insn_idx
+ insn
->off
+ 1, *insn_idx
);
2737 /* detect if we are comparing against a constant value so we can adjust
2738 * our min/max values for our dst register.
2739 * this is only legit if both are scalars (or pointers to the same
2740 * object, I suppose, but we don't support that right now), because
2741 * otherwise the different base pointers mean the offsets aren't
2744 if (BPF_SRC(insn
->code
) == BPF_X
) {
2745 if (dst_reg
->type
== SCALAR_VALUE
&&
2746 regs
[insn
->src_reg
].type
== SCALAR_VALUE
) {
2747 if (tnum_is_const(regs
[insn
->src_reg
].var_off
))
2748 reg_set_min_max(&other_branch
->regs
[insn
->dst_reg
],
2749 dst_reg
, regs
[insn
->src_reg
].var_off
.value
,
2751 else if (tnum_is_const(dst_reg
->var_off
))
2752 reg_set_min_max_inv(&other_branch
->regs
[insn
->src_reg
],
2753 ®s
[insn
->src_reg
],
2754 dst_reg
->var_off
.value
, opcode
);
2755 else if (opcode
== BPF_JEQ
|| opcode
== BPF_JNE
)
2756 /* Comparing for equality, we can combine knowledge */
2757 reg_combine_min_max(&other_branch
->regs
[insn
->src_reg
],
2758 &other_branch
->regs
[insn
->dst_reg
],
2759 ®s
[insn
->src_reg
],
2760 ®s
[insn
->dst_reg
], opcode
);
2762 } else if (dst_reg
->type
== SCALAR_VALUE
) {
2763 reg_set_min_max(&other_branch
->regs
[insn
->dst_reg
],
2764 dst_reg
, insn
->imm
, opcode
);
2767 /* detect if R == 0 where R is returned from bpf_map_lookup_elem() */
2768 if (BPF_SRC(insn
->code
) == BPF_K
&&
2769 insn
->imm
== 0 && (opcode
== BPF_JEQ
|| opcode
== BPF_JNE
) &&
2770 dst_reg
->type
== PTR_TO_MAP_VALUE_OR_NULL
) {
2771 /* Mark all identical map registers in each branch as either
2772 * safe or unknown depending R == 0 or R != 0 conditional.
2774 mark_map_regs(this_branch
, insn
->dst_reg
, opcode
== BPF_JNE
);
2775 mark_map_regs(other_branch
, insn
->dst_reg
, opcode
== BPF_JEQ
);
2776 } else if (BPF_SRC(insn
->code
) == BPF_X
&& opcode
== BPF_JGT
&&
2777 dst_reg
->type
== PTR_TO_PACKET
&&
2778 regs
[insn
->src_reg
].type
== PTR_TO_PACKET_END
) {
2779 find_good_pkt_pointers(this_branch
, dst_reg
, PTR_TO_PACKET
);
2780 } else if (BPF_SRC(insn
->code
) == BPF_X
&& opcode
== BPF_JLT
&&
2781 dst_reg
->type
== PTR_TO_PACKET
&&
2782 regs
[insn
->src_reg
].type
== PTR_TO_PACKET_END
) {
2783 find_good_pkt_pointers(other_branch
, dst_reg
, PTR_TO_PACKET
);
2784 } else if (BPF_SRC(insn
->code
) == BPF_X
&& opcode
== BPF_JGE
&&
2785 dst_reg
->type
== PTR_TO_PACKET_END
&&
2786 regs
[insn
->src_reg
].type
== PTR_TO_PACKET
) {
2787 find_good_pkt_pointers(other_branch
, ®s
[insn
->src_reg
],
2789 } else if (BPF_SRC(insn
->code
) == BPF_X
&& opcode
== BPF_JLE
&&
2790 dst_reg
->type
== PTR_TO_PACKET_END
&&
2791 regs
[insn
->src_reg
].type
== PTR_TO_PACKET
) {
2792 find_good_pkt_pointers(this_branch
, ®s
[insn
->src_reg
],
2794 } else if (BPF_SRC(insn
->code
) == BPF_X
&& opcode
== BPF_JGT
&&
2795 dst_reg
->type
== PTR_TO_PACKET_META
&&
2796 reg_is_init_pkt_pointer(®s
[insn
->src_reg
], PTR_TO_PACKET
)) {
2797 find_good_pkt_pointers(this_branch
, dst_reg
, PTR_TO_PACKET_META
);
2798 } else if (BPF_SRC(insn
->code
) == BPF_X
&& opcode
== BPF_JLT
&&
2799 dst_reg
->type
== PTR_TO_PACKET_META
&&
2800 reg_is_init_pkt_pointer(®s
[insn
->src_reg
], PTR_TO_PACKET
)) {
2801 find_good_pkt_pointers(other_branch
, dst_reg
, PTR_TO_PACKET_META
);
2802 } else if (BPF_SRC(insn
->code
) == BPF_X
&& opcode
== BPF_JGE
&&
2803 reg_is_init_pkt_pointer(dst_reg
, PTR_TO_PACKET
) &&
2804 regs
[insn
->src_reg
].type
== PTR_TO_PACKET_META
) {
2805 find_good_pkt_pointers(other_branch
, ®s
[insn
->src_reg
],
2806 PTR_TO_PACKET_META
);
2807 } else if (BPF_SRC(insn
->code
) == BPF_X
&& opcode
== BPF_JLE
&&
2808 reg_is_init_pkt_pointer(dst_reg
, PTR_TO_PACKET
) &&
2809 regs
[insn
->src_reg
].type
== PTR_TO_PACKET_META
) {
2810 find_good_pkt_pointers(this_branch
, ®s
[insn
->src_reg
],
2811 PTR_TO_PACKET_META
);
2812 } else if (is_pointer_value(env
, insn
->dst_reg
)) {
2813 verbose(env
, "R%d pointer comparison prohibited\n",
2818 print_verifier_state(env
, this_branch
);
2822 /* return the map pointer stored inside BPF_LD_IMM64 instruction */
2823 static struct bpf_map
*ld_imm64_to_map_ptr(struct bpf_insn
*insn
)
2825 u64 imm64
= ((u64
) (u32
) insn
[0].imm
) | ((u64
) (u32
) insn
[1].imm
) << 32;
2827 return (struct bpf_map
*) (unsigned long) imm64
;
2830 /* verify BPF_LD_IMM64 instruction */
2831 static int check_ld_imm(struct bpf_verifier_env
*env
, struct bpf_insn
*insn
)
2833 struct bpf_reg_state
*regs
= env
->cur_state
.regs
;
2836 if (BPF_SIZE(insn
->code
) != BPF_DW
) {
2837 verbose(env
, "invalid BPF_LD_IMM insn\n");
2840 if (insn
->off
!= 0) {
2841 verbose(env
, "BPF_LD_IMM64 uses reserved fields\n");
2845 err
= check_reg_arg(env
, insn
->dst_reg
, DST_OP
);
2849 if (insn
->src_reg
== 0) {
2850 u64 imm
= ((u64
)(insn
+ 1)->imm
<< 32) | (u32
)insn
->imm
;
2852 regs
[insn
->dst_reg
].type
= SCALAR_VALUE
;
2853 __mark_reg_known(®s
[insn
->dst_reg
], imm
);
2857 /* replace_map_fd_with_map_ptr() should have caught bad ld_imm64 */
2858 BUG_ON(insn
->src_reg
!= BPF_PSEUDO_MAP_FD
);
2860 regs
[insn
->dst_reg
].type
= CONST_PTR_TO_MAP
;
2861 regs
[insn
->dst_reg
].map_ptr
= ld_imm64_to_map_ptr(insn
);
2865 static bool may_access_skb(enum bpf_prog_type type
)
2868 case BPF_PROG_TYPE_SOCKET_FILTER
:
2869 case BPF_PROG_TYPE_SCHED_CLS
:
2870 case BPF_PROG_TYPE_SCHED_ACT
:
2877 /* verify safety of LD_ABS|LD_IND instructions:
2878 * - they can only appear in the programs where ctx == skb
2879 * - since they are wrappers of function calls, they scratch R1-R5 registers,
2880 * preserve R6-R9, and store return value into R0
2883 * ctx == skb == R6 == CTX
2886 * SRC == any register
2887 * IMM == 32-bit immediate
2890 * R0 - 8/16/32-bit skb data converted to cpu endianness
2892 static int check_ld_abs(struct bpf_verifier_env
*env
, struct bpf_insn
*insn
)
2894 struct bpf_reg_state
*regs
= env
->cur_state
.regs
;
2895 u8 mode
= BPF_MODE(insn
->code
);
2898 if (!may_access_skb(env
->prog
->type
)) {
2899 verbose(env
, "BPF_LD_[ABS|IND] instructions not allowed for this program type\n");
2903 if (insn
->dst_reg
!= BPF_REG_0
|| insn
->off
!= 0 ||
2904 BPF_SIZE(insn
->code
) == BPF_DW
||
2905 (mode
== BPF_ABS
&& insn
->src_reg
!= BPF_REG_0
)) {
2906 verbose(env
, "BPF_LD_[ABS|IND] uses reserved fields\n");
2910 /* check whether implicit source operand (register R6) is readable */
2911 err
= check_reg_arg(env
, BPF_REG_6
, SRC_OP
);
2915 if (regs
[BPF_REG_6
].type
!= PTR_TO_CTX
) {
2917 "at the time of BPF_LD_ABS|IND R6 != pointer to skb\n");
2921 if (mode
== BPF_IND
) {
2922 /* check explicit source operand */
2923 err
= check_reg_arg(env
, insn
->src_reg
, SRC_OP
);
2928 /* reset caller saved regs to unreadable */
2929 for (i
= 0; i
< CALLER_SAVED_REGS
; i
++) {
2930 mark_reg_not_init(env
, regs
, caller_saved
[i
]);
2931 check_reg_arg(env
, caller_saved
[i
], DST_OP_NO_MARK
);
2934 /* mark destination R0 register as readable, since it contains
2935 * the value fetched from the packet.
2936 * Already marked as written above.
2938 mark_reg_unknown(env
, regs
, BPF_REG_0
);
2942 static int check_return_code(struct bpf_verifier_env
*env
)
2944 struct bpf_reg_state
*reg
;
2945 struct tnum range
= tnum_range(0, 1);
2947 switch (env
->prog
->type
) {
2948 case BPF_PROG_TYPE_CGROUP_SKB
:
2949 case BPF_PROG_TYPE_CGROUP_SOCK
:
2950 case BPF_PROG_TYPE_SOCK_OPS
:
2956 reg
= &env
->cur_state
.regs
[BPF_REG_0
];
2957 if (reg
->type
!= SCALAR_VALUE
) {
2958 verbose(env
, "At program exit the register R0 is not a known value (%s)\n",
2959 reg_type_str
[reg
->type
]);
2963 if (!tnum_in(range
, reg
->var_off
)) {
2964 verbose(env
, "At program exit the register R0 ");
2965 if (!tnum_is_unknown(reg
->var_off
)) {
2968 tnum_strn(tn_buf
, sizeof(tn_buf
), reg
->var_off
);
2969 verbose(env
, "has value %s", tn_buf
);
2971 verbose(env
, "has unknown scalar value");
2973 verbose(env
, " should have been 0 or 1\n");
2979 /* non-recursive DFS pseudo code
2980 * 1 procedure DFS-iterative(G,v):
2981 * 2 label v as discovered
2982 * 3 let S be a stack
2984 * 5 while S is not empty
2986 * 7 if t is what we're looking for:
2988 * 9 for all edges e in G.adjacentEdges(t) do
2989 * 10 if edge e is already labelled
2990 * 11 continue with the next edge
2991 * 12 w <- G.adjacentVertex(t,e)
2992 * 13 if vertex w is not discovered and not explored
2993 * 14 label e as tree-edge
2994 * 15 label w as discovered
2997 * 18 else if vertex w is discovered
2998 * 19 label e as back-edge
3000 * 21 // vertex w is explored
3001 * 22 label e as forward- or cross-edge
3002 * 23 label t as explored
3007 * 0x11 - discovered and fall-through edge labelled
3008 * 0x12 - discovered and fall-through and branch edges labelled
3019 #define STATE_LIST_MARK ((struct bpf_verifier_state_list *) -1L)
3021 static int *insn_stack
; /* stack of insns to process */
3022 static int cur_stack
; /* current stack index */
3023 static int *insn_state
;
3025 /* t, w, e - match pseudo-code above:
3026 * t - index of current instruction
3027 * w - next instruction
3030 static int push_insn(int t
, int w
, int e
, struct bpf_verifier_env
*env
)
3032 if (e
== FALLTHROUGH
&& insn_state
[t
] >= (DISCOVERED
| FALLTHROUGH
))
3035 if (e
== BRANCH
&& insn_state
[t
] >= (DISCOVERED
| BRANCH
))
3038 if (w
< 0 || w
>= env
->prog
->len
) {
3039 verbose(env
, "jump out of range from insn %d to %d\n", t
, w
);
3044 /* mark branch target for state pruning */
3045 env
->explored_states
[w
] = STATE_LIST_MARK
;
3047 if (insn_state
[w
] == 0) {
3049 insn_state
[t
] = DISCOVERED
| e
;
3050 insn_state
[w
] = DISCOVERED
;
3051 if (cur_stack
>= env
->prog
->len
)
3053 insn_stack
[cur_stack
++] = w
;
3055 } else if ((insn_state
[w
] & 0xF0) == DISCOVERED
) {
3056 verbose(env
, "back-edge from insn %d to %d\n", t
, w
);
3058 } else if (insn_state
[w
] == EXPLORED
) {
3059 /* forward- or cross-edge */
3060 insn_state
[t
] = DISCOVERED
| e
;
3062 verbose(env
, "insn state internal bug\n");
3068 /* non-recursive depth-first-search to detect loops in BPF program
3069 * loop == back-edge in directed graph
3071 static int check_cfg(struct bpf_verifier_env
*env
)
3073 struct bpf_insn
*insns
= env
->prog
->insnsi
;
3074 int insn_cnt
= env
->prog
->len
;
3078 insn_state
= kcalloc(insn_cnt
, sizeof(int), GFP_KERNEL
);
3082 insn_stack
= kcalloc(insn_cnt
, sizeof(int), GFP_KERNEL
);
3088 insn_state
[0] = DISCOVERED
; /* mark 1st insn as discovered */
3089 insn_stack
[0] = 0; /* 0 is the first instruction */
3095 t
= insn_stack
[cur_stack
- 1];
3097 if (BPF_CLASS(insns
[t
].code
) == BPF_JMP
) {
3098 u8 opcode
= BPF_OP(insns
[t
].code
);
3100 if (opcode
== BPF_EXIT
) {
3102 } else if (opcode
== BPF_CALL
) {
3103 ret
= push_insn(t
, t
+ 1, FALLTHROUGH
, env
);
3108 if (t
+ 1 < insn_cnt
)
3109 env
->explored_states
[t
+ 1] = STATE_LIST_MARK
;
3110 } else if (opcode
== BPF_JA
) {
3111 if (BPF_SRC(insns
[t
].code
) != BPF_K
) {
3115 /* unconditional jump with single edge */
3116 ret
= push_insn(t
, t
+ insns
[t
].off
+ 1,
3122 /* tell verifier to check for equivalent states
3123 * after every call and jump
3125 if (t
+ 1 < insn_cnt
)
3126 env
->explored_states
[t
+ 1] = STATE_LIST_MARK
;
3128 /* conditional jump with two edges */
3129 env
->explored_states
[t
] = STATE_LIST_MARK
;
3130 ret
= push_insn(t
, t
+ 1, FALLTHROUGH
, env
);
3136 ret
= push_insn(t
, t
+ insns
[t
].off
+ 1, BRANCH
, env
);
3143 /* all other non-branch instructions with single
3146 ret
= push_insn(t
, t
+ 1, FALLTHROUGH
, env
);
3154 insn_state
[t
] = EXPLORED
;
3155 if (cur_stack
-- <= 0) {
3156 verbose(env
, "pop stack internal bug\n");
3163 for (i
= 0; i
< insn_cnt
; i
++) {
3164 if (insn_state
[i
] != EXPLORED
) {
3165 verbose(env
, "unreachable insn %d\n", i
);
3170 ret
= 0; /* cfg looks good */
3178 /* check %cur's range satisfies %old's */
3179 static bool range_within(struct bpf_reg_state
*old
,
3180 struct bpf_reg_state
*cur
)
3182 return old
->umin_value
<= cur
->umin_value
&&
3183 old
->umax_value
>= cur
->umax_value
&&
3184 old
->smin_value
<= cur
->smin_value
&&
3185 old
->smax_value
>= cur
->smax_value
;
3188 /* Maximum number of register states that can exist at once */
3189 #define ID_MAP_SIZE (MAX_BPF_REG + MAX_BPF_STACK / BPF_REG_SIZE)
3195 /* If in the old state two registers had the same id, then they need to have
3196 * the same id in the new state as well. But that id could be different from
3197 * the old state, so we need to track the mapping from old to new ids.
3198 * Once we have seen that, say, a reg with old id 5 had new id 9, any subsequent
3199 * regs with old id 5 must also have new id 9 for the new state to be safe. But
3200 * regs with a different old id could still have new id 9, we don't care about
3202 * So we look through our idmap to see if this old id has been seen before. If
3203 * so, we require the new id to match; otherwise, we add the id pair to the map.
3205 static bool check_ids(u32 old_id
, u32 cur_id
, struct idpair
*idmap
)
3209 for (i
= 0; i
< ID_MAP_SIZE
; i
++) {
3210 if (!idmap
[i
].old
) {
3211 /* Reached an empty slot; haven't seen this id before */
3212 idmap
[i
].old
= old_id
;
3213 idmap
[i
].cur
= cur_id
;
3216 if (idmap
[i
].old
== old_id
)
3217 return idmap
[i
].cur
== cur_id
;
3219 /* We ran out of idmap slots, which should be impossible */
3224 /* Returns true if (rold safe implies rcur safe) */
3225 static bool regsafe(struct bpf_reg_state
*rold
, struct bpf_reg_state
*rcur
,
3226 struct idpair
*idmap
)
3228 if (!(rold
->live
& REG_LIVE_READ
))
3229 /* explored state didn't use this */
3232 if (memcmp(rold
, rcur
, offsetof(struct bpf_reg_state
, live
)) == 0)
3235 if (rold
->type
== NOT_INIT
)
3236 /* explored state can't have used this */
3238 if (rcur
->type
== NOT_INIT
)
3240 switch (rold
->type
) {
3242 if (rcur
->type
== SCALAR_VALUE
) {
3243 /* new val must satisfy old val knowledge */
3244 return range_within(rold
, rcur
) &&
3245 tnum_in(rold
->var_off
, rcur
->var_off
);
3247 /* if we knew anything about the old value, we're not
3248 * equal, because we can't know anything about the
3249 * scalar value of the pointer in the new value.
3251 return rold
->umin_value
== 0 &&
3252 rold
->umax_value
== U64_MAX
&&
3253 rold
->smin_value
== S64_MIN
&&
3254 rold
->smax_value
== S64_MAX
&&
3255 tnum_is_unknown(rold
->var_off
);
3257 case PTR_TO_MAP_VALUE
:
3258 /* If the new min/max/var_off satisfy the old ones and
3259 * everything else matches, we are OK.
3260 * We don't care about the 'id' value, because nothing
3261 * uses it for PTR_TO_MAP_VALUE (only for ..._OR_NULL)
3263 return memcmp(rold
, rcur
, offsetof(struct bpf_reg_state
, id
)) == 0 &&
3264 range_within(rold
, rcur
) &&
3265 tnum_in(rold
->var_off
, rcur
->var_off
);
3266 case PTR_TO_MAP_VALUE_OR_NULL
:
3267 /* a PTR_TO_MAP_VALUE could be safe to use as a
3268 * PTR_TO_MAP_VALUE_OR_NULL into the same map.
3269 * However, if the old PTR_TO_MAP_VALUE_OR_NULL then got NULL-
3270 * checked, doing so could have affected others with the same
3271 * id, and we can't check for that because we lost the id when
3272 * we converted to a PTR_TO_MAP_VALUE.
3274 if (rcur
->type
!= PTR_TO_MAP_VALUE_OR_NULL
)
3276 if (memcmp(rold
, rcur
, offsetof(struct bpf_reg_state
, id
)))
3278 /* Check our ids match any regs they're supposed to */
3279 return check_ids(rold
->id
, rcur
->id
, idmap
);
3280 case PTR_TO_PACKET_META
:
3282 if (rcur
->type
!= rold
->type
)
3284 /* We must have at least as much range as the old ptr
3285 * did, so that any accesses which were safe before are
3286 * still safe. This is true even if old range < old off,
3287 * since someone could have accessed through (ptr - k), or
3288 * even done ptr -= k in a register, to get a safe access.
3290 if (rold
->range
> rcur
->range
)
3292 /* If the offsets don't match, we can't trust our alignment;
3293 * nor can we be sure that we won't fall out of range.
3295 if (rold
->off
!= rcur
->off
)
3297 /* id relations must be preserved */
3298 if (rold
->id
&& !check_ids(rold
->id
, rcur
->id
, idmap
))
3300 /* new val must satisfy old val knowledge */
3301 return range_within(rold
, rcur
) &&
3302 tnum_in(rold
->var_off
, rcur
->var_off
);
3304 case CONST_PTR_TO_MAP
:
3306 case PTR_TO_PACKET_END
:
3307 /* Only valid matches are exact, which memcmp() above
3308 * would have accepted
3311 /* Don't know what's going on, just say it's not safe */
3315 /* Shouldn't get here; if we do, say it's not safe */
3320 /* compare two verifier states
3322 * all states stored in state_list are known to be valid, since
3323 * verifier reached 'bpf_exit' instruction through them
3325 * this function is called when verifier exploring different branches of
3326 * execution popped from the state stack. If it sees an old state that has
3327 * more strict register state and more strict stack state then this execution
3328 * branch doesn't need to be explored further, since verifier already
3329 * concluded that more strict state leads to valid finish.
3331 * Therefore two states are equivalent if register state is more conservative
3332 * and explored stack state is more conservative than the current one.
3335 * (slot1=INV slot2=MISC) == (slot1=MISC slot2=MISC)
3336 * (slot1=MISC slot2=MISC) != (slot1=INV slot2=MISC)
3338 * In other words if current stack state (one being explored) has more
3339 * valid slots than old one that already passed validation, it means
3340 * the verifier can stop exploring and conclude that current state is valid too
3342 * Similarly with registers. If explored state has register type as invalid
3343 * whereas register type in current state is meaningful, it means that
3344 * the current state will reach 'bpf_exit' instruction safely
3346 static bool states_equal(struct bpf_verifier_env
*env
,
3347 struct bpf_verifier_state
*old
,
3348 struct bpf_verifier_state
*cur
)
3350 struct idpair
*idmap
;
3354 idmap
= kcalloc(ID_MAP_SIZE
, sizeof(struct idpair
), GFP_KERNEL
);
3355 /* If we failed to allocate the idmap, just say it's not safe */
3359 for (i
= 0; i
< MAX_BPF_REG
; i
++) {
3360 if (!regsafe(&old
->regs
[i
], &cur
->regs
[i
], idmap
))
3364 for (i
= 0; i
< MAX_BPF_STACK
; i
++) {
3365 if (old
->stack_slot_type
[i
] == STACK_INVALID
)
3367 if (old
->stack_slot_type
[i
] != cur
->stack_slot_type
[i
])
3368 /* Ex: old explored (safe) state has STACK_SPILL in
3369 * this stack slot, but current has has STACK_MISC ->
3370 * this verifier states are not equivalent,
3371 * return false to continue verification of this path
3374 if (i
% BPF_REG_SIZE
)
3376 if (old
->stack_slot_type
[i
] != STACK_SPILL
)
3378 if (!regsafe(&old
->spilled_regs
[i
/ BPF_REG_SIZE
],
3379 &cur
->spilled_regs
[i
/ BPF_REG_SIZE
],
3381 /* when explored and current stack slot are both storing
3382 * spilled registers, check that stored pointers types
3383 * are the same as well.
3384 * Ex: explored safe path could have stored
3385 * (bpf_reg_state) {.type = PTR_TO_STACK, .off = -8}
3386 * but current path has stored:
3387 * (bpf_reg_state) {.type = PTR_TO_STACK, .off = -16}
3388 * such verifier states are not equivalent.
3389 * return false to continue verification of this path
3401 /* A write screens off any subsequent reads; but write marks come from the
3402 * straight-line code between a state and its parent. When we arrive at a
3403 * jump target (in the first iteration of the propagate_liveness() loop),
3404 * we didn't arrive by the straight-line code, so read marks in state must
3405 * propagate to parent regardless of state's write marks.
3407 static bool do_propagate_liveness(const struct bpf_verifier_state
*state
,
3408 struct bpf_verifier_state
*parent
)
3410 bool writes
= parent
== state
->parent
; /* Observe write marks */
3411 bool touched
= false; /* any changes made? */
3416 /* Propagate read liveness of registers... */
3417 BUILD_BUG_ON(BPF_REG_FP
+ 1 != MAX_BPF_REG
);
3418 /* We don't need to worry about FP liveness because it's read-only */
3419 for (i
= 0; i
< BPF_REG_FP
; i
++) {
3420 if (parent
->regs
[i
].live
& REG_LIVE_READ
)
3422 if (writes
&& (state
->regs
[i
].live
& REG_LIVE_WRITTEN
))
3424 if (state
->regs
[i
].live
& REG_LIVE_READ
) {
3425 parent
->regs
[i
].live
|= REG_LIVE_READ
;
3429 /* ... and stack slots */
3430 for (i
= 0; i
< MAX_BPF_STACK
/ BPF_REG_SIZE
; i
++) {
3431 if (parent
->stack_slot_type
[i
* BPF_REG_SIZE
] != STACK_SPILL
)
3433 if (state
->stack_slot_type
[i
* BPF_REG_SIZE
] != STACK_SPILL
)
3435 if (parent
->spilled_regs
[i
].live
& REG_LIVE_READ
)
3437 if (writes
&& (state
->spilled_regs
[i
].live
& REG_LIVE_WRITTEN
))
3439 if (state
->spilled_regs
[i
].live
& REG_LIVE_READ
) {
3440 parent
->spilled_regs
[i
].live
|= REG_LIVE_READ
;
3447 /* "parent" is "a state from which we reach the current state", but initially
3448 * it is not the state->parent (i.e. "the state whose straight-line code leads
3449 * to the current state"), instead it is the state that happened to arrive at
3450 * a (prunable) equivalent of the current state. See comment above
3451 * do_propagate_liveness() for consequences of this.
3452 * This function is just a more efficient way of calling mark_reg_read() or
3453 * mark_stack_slot_read() on each reg in "parent" that is read in "state",
3454 * though it requires that parent != state->parent in the call arguments.
3456 static void propagate_liveness(const struct bpf_verifier_state
*state
,
3457 struct bpf_verifier_state
*parent
)
3459 while (do_propagate_liveness(state
, parent
)) {
3460 /* Something changed, so we need to feed those changes onward */
3462 parent
= state
->parent
;
3466 static int is_state_visited(struct bpf_verifier_env
*env
, int insn_idx
)
3468 struct bpf_verifier_state_list
*new_sl
;
3469 struct bpf_verifier_state_list
*sl
;
3472 sl
= env
->explored_states
[insn_idx
];
3474 /* this 'insn_idx' instruction wasn't marked, so we will not
3475 * be doing state search here
3479 while (sl
!= STATE_LIST_MARK
) {
3480 if (states_equal(env
, &sl
->state
, &env
->cur_state
)) {
3481 /* reached equivalent register/stack state,
3483 * Registers read by the continuation are read by us.
3484 * If we have any write marks in env->cur_state, they
3485 * will prevent corresponding reads in the continuation
3486 * from reaching our parent (an explored_state). Our
3487 * own state will get the read marks recorded, but
3488 * they'll be immediately forgotten as we're pruning
3489 * this state and will pop a new one.
3491 propagate_liveness(&sl
->state
, &env
->cur_state
);
3497 /* there were no equivalent states, remember current one.
3498 * technically the current state is not proven to be safe yet,
3499 * but it will either reach bpf_exit (which means it's safe) or
3500 * it will be rejected. Since there are no loops, we won't be
3501 * seeing this 'insn_idx' instruction again on the way to bpf_exit
3503 new_sl
= kmalloc(sizeof(struct bpf_verifier_state_list
), GFP_USER
);
3507 /* add new state to the head of linked list */
3508 memcpy(&new_sl
->state
, &env
->cur_state
, sizeof(env
->cur_state
));
3509 new_sl
->next
= env
->explored_states
[insn_idx
];
3510 env
->explored_states
[insn_idx
] = new_sl
;
3511 /* connect new state to parentage chain */
3512 env
->cur_state
.parent
= &new_sl
->state
;
3513 /* clear write marks in current state: the writes we did are not writes
3514 * our child did, so they don't screen off its reads from us.
3515 * (There are no read marks in current state, because reads always mark
3516 * their parent and current state never has children yet. Only
3517 * explored_states can get read marks.)
3519 for (i
= 0; i
< BPF_REG_FP
; i
++)
3520 env
->cur_state
.regs
[i
].live
= REG_LIVE_NONE
;
3521 for (i
= 0; i
< MAX_BPF_STACK
/ BPF_REG_SIZE
; i
++)
3522 if (env
->cur_state
.stack_slot_type
[i
* BPF_REG_SIZE
] == STACK_SPILL
)
3523 env
->cur_state
.spilled_regs
[i
].live
= REG_LIVE_NONE
;
3527 static int ext_analyzer_insn_hook(struct bpf_verifier_env
*env
,
3528 int insn_idx
, int prev_insn_idx
)
3530 if (!env
->analyzer_ops
|| !env
->analyzer_ops
->insn_hook
)
3533 return env
->analyzer_ops
->insn_hook(env
, insn_idx
, prev_insn_idx
);
3536 static int do_check(struct bpf_verifier_env
*env
)
3538 struct bpf_verifier_state
*state
= &env
->cur_state
;
3539 struct bpf_insn
*insns
= env
->prog
->insnsi
;
3540 struct bpf_reg_state
*regs
= state
->regs
;
3541 int insn_cnt
= env
->prog
->len
;
3542 int insn_idx
, prev_insn_idx
= 0;
3543 int insn_processed
= 0;
3544 bool do_print_state
= false;
3546 init_reg_state(env
, regs
);
3547 state
->parent
= NULL
;
3550 struct bpf_insn
*insn
;
3554 if (insn_idx
>= insn_cnt
) {
3555 verbose(env
, "invalid insn idx %d insn_cnt %d\n",
3556 insn_idx
, insn_cnt
);
3560 insn
= &insns
[insn_idx
];
3561 class = BPF_CLASS(insn
->code
);
3563 if (++insn_processed
> BPF_COMPLEXITY_LIMIT_INSNS
) {
3565 "BPF program is too large. Processed %d insn\n",
3570 err
= is_state_visited(env
, insn_idx
);
3574 /* found equivalent state, can prune the search */
3575 if (env
->log
.level
) {
3577 verbose(env
, "\nfrom %d to %d: safe\n",
3578 prev_insn_idx
, insn_idx
);
3580 verbose(env
, "%d: safe\n", insn_idx
);
3582 goto process_bpf_exit
;
3588 if (env
->log
.level
> 1 || (env
->log
.level
&& do_print_state
)) {
3589 if (env
->log
.level
> 1)
3590 verbose(env
, "%d:", insn_idx
);
3592 verbose(env
, "\nfrom %d to %d:",
3593 prev_insn_idx
, insn_idx
);
3594 print_verifier_state(env
, &env
->cur_state
);
3595 do_print_state
= false;
3598 if (env
->log
.level
) {
3599 verbose(env
, "%d: ", insn_idx
);
3600 print_bpf_insn(verbose
, env
, insn
,
3601 env
->allow_ptr_leaks
);
3604 err
= ext_analyzer_insn_hook(env
, insn_idx
, prev_insn_idx
);
3608 if (class == BPF_ALU
|| class == BPF_ALU64
) {
3609 err
= check_alu_op(env
, insn
);
3613 } else if (class == BPF_LDX
) {
3614 enum bpf_reg_type
*prev_src_type
, src_reg_type
;
3616 /* check for reserved fields is already done */
3618 /* check src operand */
3619 err
= check_reg_arg(env
, insn
->src_reg
, SRC_OP
);
3623 err
= check_reg_arg(env
, insn
->dst_reg
, DST_OP_NO_MARK
);
3627 src_reg_type
= regs
[insn
->src_reg
].type
;
3629 /* check that memory (src_reg + off) is readable,
3630 * the state of dst_reg will be updated by this func
3632 err
= check_mem_access(env
, insn_idx
, insn
->src_reg
, insn
->off
,
3633 BPF_SIZE(insn
->code
), BPF_READ
,
3638 prev_src_type
= &env
->insn_aux_data
[insn_idx
].ptr_type
;
3640 if (*prev_src_type
== NOT_INIT
) {
3642 * dst_reg = *(u32 *)(src_reg + off)
3643 * save type to validate intersecting paths
3645 *prev_src_type
= src_reg_type
;
3647 } else if (src_reg_type
!= *prev_src_type
&&
3648 (src_reg_type
== PTR_TO_CTX
||
3649 *prev_src_type
== PTR_TO_CTX
)) {
3650 /* ABuser program is trying to use the same insn
3651 * dst_reg = *(u32*) (src_reg + off)
3652 * with different pointer types:
3653 * src_reg == ctx in one branch and
3654 * src_reg == stack|map in some other branch.
3657 verbose(env
, "same insn cannot be used with different pointers\n");
3661 } else if (class == BPF_STX
) {
3662 enum bpf_reg_type
*prev_dst_type
, dst_reg_type
;
3664 if (BPF_MODE(insn
->code
) == BPF_XADD
) {
3665 err
= check_xadd(env
, insn_idx
, insn
);
3672 /* check src1 operand */
3673 err
= check_reg_arg(env
, insn
->src_reg
, SRC_OP
);
3676 /* check src2 operand */
3677 err
= check_reg_arg(env
, insn
->dst_reg
, SRC_OP
);
3681 dst_reg_type
= regs
[insn
->dst_reg
].type
;
3683 /* check that memory (dst_reg + off) is writeable */
3684 err
= check_mem_access(env
, insn_idx
, insn
->dst_reg
, insn
->off
,
3685 BPF_SIZE(insn
->code
), BPF_WRITE
,
3690 prev_dst_type
= &env
->insn_aux_data
[insn_idx
].ptr_type
;
3692 if (*prev_dst_type
== NOT_INIT
) {
3693 *prev_dst_type
= dst_reg_type
;
3694 } else if (dst_reg_type
!= *prev_dst_type
&&
3695 (dst_reg_type
== PTR_TO_CTX
||
3696 *prev_dst_type
== PTR_TO_CTX
)) {
3697 verbose(env
, "same insn cannot be used with different pointers\n");
3701 } else if (class == BPF_ST
) {
3702 if (BPF_MODE(insn
->code
) != BPF_MEM
||
3703 insn
->src_reg
!= BPF_REG_0
) {
3704 verbose(env
, "BPF_ST uses reserved fields\n");
3707 /* check src operand */
3708 err
= check_reg_arg(env
, insn
->dst_reg
, SRC_OP
);
3712 /* check that memory (dst_reg + off) is writeable */
3713 err
= check_mem_access(env
, insn_idx
, insn
->dst_reg
, insn
->off
,
3714 BPF_SIZE(insn
->code
), BPF_WRITE
,
3719 } else if (class == BPF_JMP
) {
3720 u8 opcode
= BPF_OP(insn
->code
);
3722 if (opcode
== BPF_CALL
) {
3723 if (BPF_SRC(insn
->code
) != BPF_K
||
3725 insn
->src_reg
!= BPF_REG_0
||
3726 insn
->dst_reg
!= BPF_REG_0
) {
3727 verbose(env
, "BPF_CALL uses reserved fields\n");
3731 err
= check_call(env
, insn
->imm
, insn_idx
);
3735 } else if (opcode
== BPF_JA
) {
3736 if (BPF_SRC(insn
->code
) != BPF_K
||
3738 insn
->src_reg
!= BPF_REG_0
||
3739 insn
->dst_reg
!= BPF_REG_0
) {
3740 verbose(env
, "BPF_JA uses reserved fields\n");
3744 insn_idx
+= insn
->off
+ 1;
3747 } else if (opcode
== BPF_EXIT
) {
3748 if (BPF_SRC(insn
->code
) != BPF_K
||
3750 insn
->src_reg
!= BPF_REG_0
||
3751 insn
->dst_reg
!= BPF_REG_0
) {
3752 verbose(env
, "BPF_EXIT uses reserved fields\n");
3756 /* eBPF calling convetion is such that R0 is used
3757 * to return the value from eBPF program.
3758 * Make sure that it's readable at this time
3759 * of bpf_exit, which means that program wrote
3760 * something into it earlier
3762 err
= check_reg_arg(env
, BPF_REG_0
, SRC_OP
);
3766 if (is_pointer_value(env
, BPF_REG_0
)) {
3767 verbose(env
, "R0 leaks addr as return value\n");
3771 err
= check_return_code(env
);
3775 insn_idx
= pop_stack(env
, &prev_insn_idx
);
3779 do_print_state
= true;
3783 err
= check_cond_jmp_op(env
, insn
, &insn_idx
);
3787 } else if (class == BPF_LD
) {
3788 u8 mode
= BPF_MODE(insn
->code
);
3790 if (mode
== BPF_ABS
|| mode
== BPF_IND
) {
3791 err
= check_ld_abs(env
, insn
);
3795 } else if (mode
== BPF_IMM
) {
3796 err
= check_ld_imm(env
, insn
);
3802 verbose(env
, "invalid BPF_LD mode\n");
3806 verbose(env
, "unknown insn class %d\n", class);
3813 verbose(env
, "processed %d insns, stack depth %d\n", insn_processed
,
3814 env
->prog
->aux
->stack_depth
);
3818 static int check_map_prealloc(struct bpf_map
*map
)
3820 return (map
->map_type
!= BPF_MAP_TYPE_HASH
&&
3821 map
->map_type
!= BPF_MAP_TYPE_PERCPU_HASH
&&
3822 map
->map_type
!= BPF_MAP_TYPE_HASH_OF_MAPS
) ||
3823 !(map
->map_flags
& BPF_F_NO_PREALLOC
);
3826 static int check_map_prog_compatibility(struct bpf_verifier_env
*env
,
3827 struct bpf_map
*map
,
3828 struct bpf_prog
*prog
)
3831 /* Make sure that BPF_PROG_TYPE_PERF_EVENT programs only use
3832 * preallocated hash maps, since doing memory allocation
3833 * in overflow_handler can crash depending on where nmi got
3836 if (prog
->type
== BPF_PROG_TYPE_PERF_EVENT
) {
3837 if (!check_map_prealloc(map
)) {
3838 verbose(env
, "perf_event programs can only use preallocated hash map\n");
3841 if (map
->inner_map_meta
&&
3842 !check_map_prealloc(map
->inner_map_meta
)) {
3843 verbose(env
, "perf_event programs can only use preallocated inner hash map\n");
3850 /* look for pseudo eBPF instructions that access map FDs and
3851 * replace them with actual map pointers
3853 static int replace_map_fd_with_map_ptr(struct bpf_verifier_env
*env
)
3855 struct bpf_insn
*insn
= env
->prog
->insnsi
;
3856 int insn_cnt
= env
->prog
->len
;
3859 err
= bpf_prog_calc_tag(env
->prog
);
3863 for (i
= 0; i
< insn_cnt
; i
++, insn
++) {
3864 if (BPF_CLASS(insn
->code
) == BPF_LDX
&&
3865 (BPF_MODE(insn
->code
) != BPF_MEM
|| insn
->imm
!= 0)) {
3866 verbose(env
, "BPF_LDX uses reserved fields\n");
3870 if (BPF_CLASS(insn
->code
) == BPF_STX
&&
3871 ((BPF_MODE(insn
->code
) != BPF_MEM
&&
3872 BPF_MODE(insn
->code
) != BPF_XADD
) || insn
->imm
!= 0)) {
3873 verbose(env
, "BPF_STX uses reserved fields\n");
3877 if (insn
[0].code
== (BPF_LD
| BPF_IMM
| BPF_DW
)) {
3878 struct bpf_map
*map
;
3881 if (i
== insn_cnt
- 1 || insn
[1].code
!= 0 ||
3882 insn
[1].dst_reg
!= 0 || insn
[1].src_reg
!= 0 ||
3884 verbose(env
, "invalid bpf_ld_imm64 insn\n");
3888 if (insn
->src_reg
== 0)
3889 /* valid generic load 64-bit imm */
3892 if (insn
->src_reg
!= BPF_PSEUDO_MAP_FD
) {
3894 "unrecognized bpf_ld_imm64 insn\n");
3898 f
= fdget(insn
->imm
);
3899 map
= __bpf_map_get(f
);
3901 verbose(env
, "fd %d is not pointing to valid bpf_map\n",
3903 return PTR_ERR(map
);
3906 err
= check_map_prog_compatibility(env
, map
, env
->prog
);
3912 /* store map pointer inside BPF_LD_IMM64 instruction */
3913 insn
[0].imm
= (u32
) (unsigned long) map
;
3914 insn
[1].imm
= ((u64
) (unsigned long) map
) >> 32;
3916 /* check whether we recorded this map already */
3917 for (j
= 0; j
< env
->used_map_cnt
; j
++)
3918 if (env
->used_maps
[j
] == map
) {
3923 if (env
->used_map_cnt
>= MAX_USED_MAPS
) {
3928 /* hold the map. If the program is rejected by verifier,
3929 * the map will be released by release_maps() or it
3930 * will be used by the valid program until it's unloaded
3931 * and all maps are released in free_bpf_prog_info()
3933 map
= bpf_map_inc(map
, false);
3936 return PTR_ERR(map
);
3938 env
->used_maps
[env
->used_map_cnt
++] = map
;
3947 /* now all pseudo BPF_LD_IMM64 instructions load valid
3948 * 'struct bpf_map *' into a register instead of user map_fd.
3949 * These pointers will be used later by verifier to validate map access.
3954 /* drop refcnt of maps used by the rejected program */
3955 static void release_maps(struct bpf_verifier_env
*env
)
3959 for (i
= 0; i
< env
->used_map_cnt
; i
++)
3960 bpf_map_put(env
->used_maps
[i
]);
3963 /* convert pseudo BPF_LD_IMM64 into generic BPF_LD_IMM64 */
3964 static void convert_pseudo_ld_imm64(struct bpf_verifier_env
*env
)
3966 struct bpf_insn
*insn
= env
->prog
->insnsi
;
3967 int insn_cnt
= env
->prog
->len
;
3970 for (i
= 0; i
< insn_cnt
; i
++, insn
++)
3971 if (insn
->code
== (BPF_LD
| BPF_IMM
| BPF_DW
))
3975 /* single env->prog->insni[off] instruction was replaced with the range
3976 * insni[off, off + cnt). Adjust corresponding insn_aux_data by copying
3977 * [0, off) and [off, end) to new locations, so the patched range stays zero
3979 static int adjust_insn_aux_data(struct bpf_verifier_env
*env
, u32 prog_len
,
3982 struct bpf_insn_aux_data
*new_data
, *old_data
= env
->insn_aux_data
;
3986 new_data
= vzalloc(sizeof(struct bpf_insn_aux_data
) * prog_len
);
3989 memcpy(new_data
, old_data
, sizeof(struct bpf_insn_aux_data
) * off
);
3990 memcpy(new_data
+ off
+ cnt
- 1, old_data
+ off
,
3991 sizeof(struct bpf_insn_aux_data
) * (prog_len
- off
- cnt
+ 1));
3992 env
->insn_aux_data
= new_data
;
3997 static struct bpf_prog
*bpf_patch_insn_data(struct bpf_verifier_env
*env
, u32 off
,
3998 const struct bpf_insn
*patch
, u32 len
)
4000 struct bpf_prog
*new_prog
;
4002 new_prog
= bpf_patch_insn_single(env
->prog
, off
, patch
, len
);
4005 if (adjust_insn_aux_data(env
, new_prog
->len
, off
, len
))
4010 /* convert load instructions that access fields of 'struct __sk_buff'
4011 * into sequence of instructions that access fields of 'struct sk_buff'
4013 static int convert_ctx_accesses(struct bpf_verifier_env
*env
)
4015 const struct bpf_verifier_ops
*ops
= env
->ops
;
4016 int i
, cnt
, size
, ctx_field_size
, delta
= 0;
4017 const int insn_cnt
= env
->prog
->len
;
4018 struct bpf_insn insn_buf
[16], *insn
;
4019 struct bpf_prog
*new_prog
;
4020 enum bpf_access_type type
;
4021 bool is_narrower_load
;
4024 if (ops
->gen_prologue
) {
4025 cnt
= ops
->gen_prologue(insn_buf
, env
->seen_direct_write
,
4027 if (cnt
>= ARRAY_SIZE(insn_buf
)) {
4028 verbose(env
, "bpf verifier is misconfigured\n");
4031 new_prog
= bpf_patch_insn_data(env
, 0, insn_buf
, cnt
);
4035 env
->prog
= new_prog
;
4040 if (!ops
->convert_ctx_access
)
4043 insn
= env
->prog
->insnsi
+ delta
;
4045 for (i
= 0; i
< insn_cnt
; i
++, insn
++) {
4046 if (insn
->code
== (BPF_LDX
| BPF_MEM
| BPF_B
) ||
4047 insn
->code
== (BPF_LDX
| BPF_MEM
| BPF_H
) ||
4048 insn
->code
== (BPF_LDX
| BPF_MEM
| BPF_W
) ||
4049 insn
->code
== (BPF_LDX
| BPF_MEM
| BPF_DW
))
4051 else if (insn
->code
== (BPF_STX
| BPF_MEM
| BPF_B
) ||
4052 insn
->code
== (BPF_STX
| BPF_MEM
| BPF_H
) ||
4053 insn
->code
== (BPF_STX
| BPF_MEM
| BPF_W
) ||
4054 insn
->code
== (BPF_STX
| BPF_MEM
| BPF_DW
))
4059 if (env
->insn_aux_data
[i
+ delta
].ptr_type
!= PTR_TO_CTX
)
4062 ctx_field_size
= env
->insn_aux_data
[i
+ delta
].ctx_field_size
;
4063 size
= BPF_LDST_BYTES(insn
);
4065 /* If the read access is a narrower load of the field,
4066 * convert to a 4/8-byte load, to minimum program type specific
4067 * convert_ctx_access changes. If conversion is successful,
4068 * we will apply proper mask to the result.
4070 is_narrower_load
= size
< ctx_field_size
;
4071 if (is_narrower_load
) {
4072 u32 off
= insn
->off
;
4075 if (type
== BPF_WRITE
) {
4076 verbose(env
, "bpf verifier narrow ctx access misconfigured\n");
4081 if (ctx_field_size
== 4)
4083 else if (ctx_field_size
== 8)
4086 insn
->off
= off
& ~(ctx_field_size
- 1);
4087 insn
->code
= BPF_LDX
| BPF_MEM
| size_code
;
4091 cnt
= ops
->convert_ctx_access(type
, insn
, insn_buf
, env
->prog
,
4093 if (cnt
== 0 || cnt
>= ARRAY_SIZE(insn_buf
) ||
4094 (ctx_field_size
&& !target_size
)) {
4095 verbose(env
, "bpf verifier is misconfigured\n");
4099 if (is_narrower_load
&& size
< target_size
) {
4100 if (ctx_field_size
<= 4)
4101 insn_buf
[cnt
++] = BPF_ALU32_IMM(BPF_AND
, insn
->dst_reg
,
4102 (1 << size
* 8) - 1);
4104 insn_buf
[cnt
++] = BPF_ALU64_IMM(BPF_AND
, insn
->dst_reg
,
4105 (1 << size
* 8) - 1);
4108 new_prog
= bpf_patch_insn_data(env
, i
+ delta
, insn_buf
, cnt
);
4114 /* keep walking new program and skip insns we just inserted */
4115 env
->prog
= new_prog
;
4116 insn
= new_prog
->insnsi
+ i
+ delta
;
4122 /* fixup insn->imm field of bpf_call instructions
4123 * and inline eligible helpers as explicit sequence of BPF instructions
4125 * this function is called after eBPF program passed verification
4127 static int fixup_bpf_calls(struct bpf_verifier_env
*env
)
4129 struct bpf_prog
*prog
= env
->prog
;
4130 struct bpf_insn
*insn
= prog
->insnsi
;
4131 const struct bpf_func_proto
*fn
;
4132 const int insn_cnt
= prog
->len
;
4133 struct bpf_insn insn_buf
[16];
4134 struct bpf_prog
*new_prog
;
4135 struct bpf_map
*map_ptr
;
4136 int i
, cnt
, delta
= 0;
4138 for (i
= 0; i
< insn_cnt
; i
++, insn
++) {
4139 if (insn
->code
!= (BPF_JMP
| BPF_CALL
))
4142 if (insn
->imm
== BPF_FUNC_get_route_realm
)
4143 prog
->dst_needed
= 1;
4144 if (insn
->imm
== BPF_FUNC_get_prandom_u32
)
4145 bpf_user_rnd_init_once();
4146 if (insn
->imm
== BPF_FUNC_tail_call
) {
4147 /* If we tail call into other programs, we
4148 * cannot make any assumptions since they can
4149 * be replaced dynamically during runtime in
4150 * the program array.
4152 prog
->cb_access
= 1;
4153 env
->prog
->aux
->stack_depth
= MAX_BPF_STACK
;
4155 /* mark bpf_tail_call as different opcode to avoid
4156 * conditional branch in the interpeter for every normal
4157 * call and to prevent accidental JITing by JIT compiler
4158 * that doesn't support bpf_tail_call yet
4161 insn
->code
= BPF_JMP
| BPF_TAIL_CALL
;
4165 /* BPF_EMIT_CALL() assumptions in some of the map_gen_lookup
4166 * handlers are currently limited to 64 bit only.
4168 if (ebpf_jit_enabled() && BITS_PER_LONG
== 64 &&
4169 insn
->imm
== BPF_FUNC_map_lookup_elem
) {
4170 map_ptr
= env
->insn_aux_data
[i
+ delta
].map_ptr
;
4171 if (map_ptr
== BPF_MAP_PTR_POISON
||
4172 !map_ptr
->ops
->map_gen_lookup
)
4173 goto patch_call_imm
;
4175 cnt
= map_ptr
->ops
->map_gen_lookup(map_ptr
, insn_buf
);
4176 if (cnt
== 0 || cnt
>= ARRAY_SIZE(insn_buf
)) {
4177 verbose(env
, "bpf verifier is misconfigured\n");
4181 new_prog
= bpf_patch_insn_data(env
, i
+ delta
, insn_buf
,
4188 /* keep walking new program and skip insns we just inserted */
4189 env
->prog
= prog
= new_prog
;
4190 insn
= new_prog
->insnsi
+ i
+ delta
;
4194 if (insn
->imm
== BPF_FUNC_redirect_map
) {
4195 /* Note, we cannot use prog directly as imm as subsequent
4196 * rewrites would still change the prog pointer. The only
4197 * stable address we can use is aux, which also works with
4198 * prog clones during blinding.
4200 u64 addr
= (unsigned long)prog
->aux
;
4201 struct bpf_insn r4_ld
[] = {
4202 BPF_LD_IMM64(BPF_REG_4
, addr
),
4205 cnt
= ARRAY_SIZE(r4_ld
);
4207 new_prog
= bpf_patch_insn_data(env
, i
+ delta
, r4_ld
, cnt
);
4212 env
->prog
= prog
= new_prog
;
4213 insn
= new_prog
->insnsi
+ i
+ delta
;
4216 fn
= env
->ops
->get_func_proto(insn
->imm
);
4217 /* all functions that have prototype and verifier allowed
4218 * programs to call them, must be real in-kernel functions
4222 "kernel subsystem misconfigured func %s#%d\n",
4223 func_id_name(insn
->imm
), insn
->imm
);
4226 insn
->imm
= fn
->func
- __bpf_call_base
;
4232 static void free_states(struct bpf_verifier_env
*env
)
4234 struct bpf_verifier_state_list
*sl
, *sln
;
4237 if (!env
->explored_states
)
4240 for (i
= 0; i
< env
->prog
->len
; i
++) {
4241 sl
= env
->explored_states
[i
];
4244 while (sl
!= STATE_LIST_MARK
) {
4251 kfree(env
->explored_states
);
4254 int bpf_check(struct bpf_prog
**prog
, union bpf_attr
*attr
)
4256 struct bpf_verifier_env
*env
;
4257 struct bpf_verifer_log
*log
;
4260 /* 'struct bpf_verifier_env' can be global, but since it's not small,
4261 * allocate/free it every time bpf_check() is called
4263 env
= kzalloc(sizeof(struct bpf_verifier_env
), GFP_KERNEL
);
4268 env
->insn_aux_data
= vzalloc(sizeof(struct bpf_insn_aux_data
) *
4271 if (!env
->insn_aux_data
)
4274 env
->ops
= bpf_verifier_ops
[env
->prog
->type
];
4276 /* grab the mutex to protect few globals used by verifier */
4277 mutex_lock(&bpf_verifier_lock
);
4279 if (attr
->log_level
|| attr
->log_buf
|| attr
->log_size
) {
4280 /* user requested verbose verifier output
4281 * and supplied buffer to store the verification trace
4283 log
->level
= attr
->log_level
;
4284 log
->ubuf
= (char __user
*) (unsigned long) attr
->log_buf
;
4285 log
->len_total
= attr
->log_size
;
4288 /* log attributes have to be sane */
4289 if (log
->len_total
< 128 || log
->len_total
> UINT_MAX
>> 8 ||
4290 !log
->level
|| !log
->ubuf
)
4294 env
->strict_alignment
= !!(attr
->prog_flags
& BPF_F_STRICT_ALIGNMENT
);
4295 if (!IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS
))
4296 env
->strict_alignment
= true;
4298 ret
= replace_map_fd_with_map_ptr(env
);
4300 goto skip_full_check
;
4302 env
->explored_states
= kcalloc(env
->prog
->len
,
4303 sizeof(struct bpf_verifier_state_list
*),
4306 if (!env
->explored_states
)
4307 goto skip_full_check
;
4309 ret
= check_cfg(env
);
4311 goto skip_full_check
;
4313 env
->allow_ptr_leaks
= capable(CAP_SYS_ADMIN
);
4315 ret
= do_check(env
);
4318 while (pop_stack(env
, NULL
) >= 0);
4322 /* program is valid, convert *(u32*)(ctx + off) accesses */
4323 ret
= convert_ctx_accesses(env
);
4326 ret
= fixup_bpf_calls(env
);
4328 if (log
->level
&& bpf_verifier_log_full(log
))
4330 if (log
->level
&& !log
->ubuf
) {
4332 goto err_release_maps
;
4335 if (ret
== 0 && env
->used_map_cnt
) {
4336 /* if program passed verifier, update used_maps in bpf_prog_info */
4337 env
->prog
->aux
->used_maps
= kmalloc_array(env
->used_map_cnt
,
4338 sizeof(env
->used_maps
[0]),
4341 if (!env
->prog
->aux
->used_maps
) {
4343 goto err_release_maps
;
4346 memcpy(env
->prog
->aux
->used_maps
, env
->used_maps
,
4347 sizeof(env
->used_maps
[0]) * env
->used_map_cnt
);
4348 env
->prog
->aux
->used_map_cnt
= env
->used_map_cnt
;
4350 /* program is valid. Convert pseudo bpf_ld_imm64 into generic
4351 * bpf_ld_imm64 instructions
4353 convert_pseudo_ld_imm64(env
);
4357 if (!env
->prog
->aux
->used_maps
)
4358 /* if we didn't copy map pointers into bpf_prog_info, release
4359 * them now. Otherwise free_bpf_prog_info() will release them.
4364 mutex_unlock(&bpf_verifier_lock
);
4365 vfree(env
->insn_aux_data
);
4371 static const struct bpf_verifier_ops
* const bpf_analyzer_ops
[] = {
4372 [BPF_PROG_TYPE_XDP
] = &xdp_analyzer_ops
,
4373 [BPF_PROG_TYPE_SCHED_CLS
] = &tc_cls_act_analyzer_ops
,
4376 int bpf_analyzer(struct bpf_prog
*prog
, const struct bpf_ext_analyzer_ops
*ops
,
4379 struct bpf_verifier_env
*env
;
4382 if (prog
->type
>= ARRAY_SIZE(bpf_analyzer_ops
) ||
4383 !bpf_analyzer_ops
[prog
->type
])
4386 env
= kzalloc(sizeof(struct bpf_verifier_env
), GFP_KERNEL
);
4390 env
->insn_aux_data
= vzalloc(sizeof(struct bpf_insn_aux_data
) *
4393 if (!env
->insn_aux_data
)
4396 env
->ops
= bpf_analyzer_ops
[env
->prog
->type
];
4397 env
->analyzer_ops
= ops
;
4398 env
->analyzer_priv
= priv
;
4400 /* grab the mutex to protect few globals used by verifier */
4401 mutex_lock(&bpf_verifier_lock
);
4403 env
->strict_alignment
= false;
4404 if (!IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS
))
4405 env
->strict_alignment
= true;
4407 env
->explored_states
= kcalloc(env
->prog
->len
,
4408 sizeof(struct bpf_verifier_state_list
*),
4411 if (!env
->explored_states
)
4412 goto skip_full_check
;
4414 ret
= check_cfg(env
);
4416 goto skip_full_check
;
4418 env
->allow_ptr_leaks
= capable(CAP_SYS_ADMIN
);
4420 ret
= do_check(env
);
4423 while (pop_stack(env
, NULL
) >= 0);
4426 mutex_unlock(&bpf_verifier_lock
);
4427 vfree(env
->insn_aux_data
);
4432 EXPORT_SYMBOL_GPL(bpf_analyzer
);