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
< state
->allocated_stack
/ BPF_REG_SIZE
; i
++) {
280 if (state
->stack
[i
].slot_type
[0] == STACK_SPILL
)
281 verbose(env
, " fp%d=%s",
282 -MAX_BPF_STACK
+ i
* BPF_REG_SIZE
,
283 reg_type_str
[state
->stack
[i
].spilled_ptr
.type
]);
288 static int copy_stack_state(struct bpf_verifier_state
*dst
,
289 const struct bpf_verifier_state
*src
)
293 if (WARN_ON_ONCE(dst
->allocated_stack
< src
->allocated_stack
)) {
294 /* internal bug, make state invalid to reject the program */
295 memset(dst
, 0, sizeof(*dst
));
298 memcpy(dst
->stack
, src
->stack
,
299 sizeof(*src
->stack
) * (src
->allocated_stack
/ BPF_REG_SIZE
));
303 /* do_check() starts with zero-sized stack in struct bpf_verifier_state to
304 * make it consume minimal amount of memory. check_stack_write() access from
305 * the program calls into realloc_verifier_state() to grow the stack size.
306 * Note there is a non-zero 'parent' pointer inside bpf_verifier_state
307 * which this function copies over. It points to previous bpf_verifier_state
308 * which is never reallocated
310 static int realloc_verifier_state(struct bpf_verifier_state
*state
, int size
,
313 u32 old_size
= state
->allocated_stack
;
314 struct bpf_stack_state
*new_stack
;
315 int slot
= size
/ BPF_REG_SIZE
;
317 if (size
<= old_size
|| !size
) {
320 state
->allocated_stack
= slot
* BPF_REG_SIZE
;
321 if (!size
&& old_size
) {
327 new_stack
= kmalloc_array(slot
, sizeof(struct bpf_stack_state
),
333 memcpy(new_stack
, state
->stack
,
334 sizeof(*new_stack
) * (old_size
/ BPF_REG_SIZE
));
335 memset(new_stack
+ old_size
/ BPF_REG_SIZE
, 0,
336 sizeof(*new_stack
) * (size
- old_size
) / BPF_REG_SIZE
);
338 state
->allocated_stack
= slot
* BPF_REG_SIZE
;
340 state
->stack
= new_stack
;
344 static void free_verifier_state(struct bpf_verifier_state
*state
)
350 /* copy verifier state from src to dst growing dst stack space
351 * when necessary to accommodate larger src stack
353 static int copy_verifier_state(struct bpf_verifier_state
*dst
,
354 const struct bpf_verifier_state
*src
)
358 err
= realloc_verifier_state(dst
, src
->allocated_stack
, false);
361 memcpy(dst
, src
, offsetof(struct bpf_verifier_state
, allocated_stack
));
362 return copy_stack_state(dst
, src
);
365 static int pop_stack(struct bpf_verifier_env
*env
, int *prev_insn_idx
,
368 struct bpf_verifier_state
*cur
= env
->cur_state
;
369 struct bpf_verifier_stack_elem
*elem
, *head
= env
->head
;
372 if (env
->head
== NULL
)
376 err
= copy_verifier_state(cur
, &head
->st
);
381 *insn_idx
= head
->insn_idx
;
383 *prev_insn_idx
= head
->prev_insn_idx
;
391 static struct bpf_verifier_state
*push_stack(struct bpf_verifier_env
*env
,
392 int insn_idx
, int prev_insn_idx
)
394 struct bpf_verifier_state
*cur
= env
->cur_state
;
395 struct bpf_verifier_stack_elem
*elem
;
398 elem
= kzalloc(sizeof(struct bpf_verifier_stack_elem
), GFP_KERNEL
);
402 err
= copy_verifier_state(&elem
->st
, cur
);
405 elem
->insn_idx
= insn_idx
;
406 elem
->prev_insn_idx
= prev_insn_idx
;
407 elem
->next
= env
->head
;
410 if (env
->stack_size
> BPF_COMPLEXITY_LIMIT_STACK
) {
411 verbose(env
, "BPF program is too complex\n");
416 /* pop all elements and return */
417 while (!pop_stack(env
, NULL
, NULL
));
421 #define CALLER_SAVED_REGS 6
422 static const int caller_saved
[CALLER_SAVED_REGS
] = {
423 BPF_REG_0
, BPF_REG_1
, BPF_REG_2
, BPF_REG_3
, BPF_REG_4
, BPF_REG_5
426 static void __mark_reg_not_init(struct bpf_reg_state
*reg
);
428 /* Mark the unknown part of a register (variable offset or scalar value) as
429 * known to have the value @imm.
431 static void __mark_reg_known(struct bpf_reg_state
*reg
, u64 imm
)
434 reg
->var_off
= tnum_const(imm
);
435 reg
->smin_value
= (s64
)imm
;
436 reg
->smax_value
= (s64
)imm
;
437 reg
->umin_value
= imm
;
438 reg
->umax_value
= imm
;
441 /* Mark the 'variable offset' part of a register as zero. This should be
442 * used only on registers holding a pointer type.
444 static void __mark_reg_known_zero(struct bpf_reg_state
*reg
)
446 __mark_reg_known(reg
, 0);
449 static void mark_reg_known_zero(struct bpf_verifier_env
*env
,
450 struct bpf_reg_state
*regs
, u32 regno
)
452 if (WARN_ON(regno
>= MAX_BPF_REG
)) {
453 verbose(env
, "mark_reg_known_zero(regs, %u)\n", regno
);
454 /* Something bad happened, let's kill all regs */
455 for (regno
= 0; regno
< MAX_BPF_REG
; regno
++)
456 __mark_reg_not_init(regs
+ regno
);
459 __mark_reg_known_zero(regs
+ regno
);
462 static bool reg_is_pkt_pointer(const struct bpf_reg_state
*reg
)
464 return type_is_pkt_pointer(reg
->type
);
467 static bool reg_is_pkt_pointer_any(const struct bpf_reg_state
*reg
)
469 return reg_is_pkt_pointer(reg
) ||
470 reg
->type
== PTR_TO_PACKET_END
;
473 /* Unmodified PTR_TO_PACKET[_META,_END] register from ctx access. */
474 static bool reg_is_init_pkt_pointer(const struct bpf_reg_state
*reg
,
475 enum bpf_reg_type which
)
477 /* The register can already have a range from prior markings.
478 * This is fine as long as it hasn't been advanced from its
481 return reg
->type
== which
&&
484 tnum_equals_const(reg
->var_off
, 0);
487 /* Attempts to improve min/max values based on var_off information */
488 static void __update_reg_bounds(struct bpf_reg_state
*reg
)
490 /* min signed is max(sign bit) | min(other bits) */
491 reg
->smin_value
= max_t(s64
, reg
->smin_value
,
492 reg
->var_off
.value
| (reg
->var_off
.mask
& S64_MIN
));
493 /* max signed is min(sign bit) | max(other bits) */
494 reg
->smax_value
= min_t(s64
, reg
->smax_value
,
495 reg
->var_off
.value
| (reg
->var_off
.mask
& S64_MAX
));
496 reg
->umin_value
= max(reg
->umin_value
, reg
->var_off
.value
);
497 reg
->umax_value
= min(reg
->umax_value
,
498 reg
->var_off
.value
| reg
->var_off
.mask
);
501 /* Uses signed min/max values to inform unsigned, and vice-versa */
502 static void __reg_deduce_bounds(struct bpf_reg_state
*reg
)
504 /* Learn sign from signed bounds.
505 * If we cannot cross the sign boundary, then signed and unsigned bounds
506 * are the same, so combine. This works even in the negative case, e.g.
507 * -3 s<= x s<= -1 implies 0xf...fd u<= x u<= 0xf...ff.
509 if (reg
->smin_value
>= 0 || reg
->smax_value
< 0) {
510 reg
->smin_value
= reg
->umin_value
= max_t(u64
, reg
->smin_value
,
512 reg
->smax_value
= reg
->umax_value
= min_t(u64
, reg
->smax_value
,
516 /* Learn sign from unsigned bounds. Signed bounds cross the sign
517 * boundary, so we must be careful.
519 if ((s64
)reg
->umax_value
>= 0) {
520 /* Positive. We can't learn anything from the smin, but smax
521 * is positive, hence safe.
523 reg
->smin_value
= reg
->umin_value
;
524 reg
->smax_value
= reg
->umax_value
= min_t(u64
, reg
->smax_value
,
526 } else if ((s64
)reg
->umin_value
< 0) {
527 /* Negative. We can't learn anything from the smax, but smin
528 * is negative, hence safe.
530 reg
->smin_value
= reg
->umin_value
= max_t(u64
, reg
->smin_value
,
532 reg
->smax_value
= reg
->umax_value
;
536 /* Attempts to improve var_off based on unsigned min/max information */
537 static void __reg_bound_offset(struct bpf_reg_state
*reg
)
539 reg
->var_off
= tnum_intersect(reg
->var_off
,
540 tnum_range(reg
->umin_value
,
544 /* Reset the min/max bounds of a register */
545 static void __mark_reg_unbounded(struct bpf_reg_state
*reg
)
547 reg
->smin_value
= S64_MIN
;
548 reg
->smax_value
= S64_MAX
;
550 reg
->umax_value
= U64_MAX
;
553 /* Mark a register as having a completely unknown (scalar) value. */
554 static void __mark_reg_unknown(struct bpf_reg_state
*reg
)
556 reg
->type
= SCALAR_VALUE
;
559 reg
->var_off
= tnum_unknown
;
560 __mark_reg_unbounded(reg
);
563 static void mark_reg_unknown(struct bpf_verifier_env
*env
,
564 struct bpf_reg_state
*regs
, u32 regno
)
566 if (WARN_ON(regno
>= MAX_BPF_REG
)) {
567 verbose(env
, "mark_reg_unknown(regs, %u)\n", regno
);
568 /* Something bad happened, let's kill all regs */
569 for (regno
= 0; regno
< MAX_BPF_REG
; regno
++)
570 __mark_reg_not_init(regs
+ regno
);
573 __mark_reg_unknown(regs
+ regno
);
576 static void __mark_reg_not_init(struct bpf_reg_state
*reg
)
578 __mark_reg_unknown(reg
);
579 reg
->type
= NOT_INIT
;
582 static void mark_reg_not_init(struct bpf_verifier_env
*env
,
583 struct bpf_reg_state
*regs
, u32 regno
)
585 if (WARN_ON(regno
>= MAX_BPF_REG
)) {
586 verbose(env
, "mark_reg_not_init(regs, %u)\n", regno
);
587 /* Something bad happened, let's kill all regs */
588 for (regno
= 0; regno
< MAX_BPF_REG
; regno
++)
589 __mark_reg_not_init(regs
+ regno
);
592 __mark_reg_not_init(regs
+ regno
);
595 static void init_reg_state(struct bpf_verifier_env
*env
,
596 struct bpf_reg_state
*regs
)
600 for (i
= 0; i
< MAX_BPF_REG
; i
++) {
601 mark_reg_not_init(env
, regs
, i
);
602 regs
[i
].live
= REG_LIVE_NONE
;
606 regs
[BPF_REG_FP
].type
= PTR_TO_STACK
;
607 mark_reg_known_zero(env
, regs
, BPF_REG_FP
);
609 /* 1st arg to a function */
610 regs
[BPF_REG_1
].type
= PTR_TO_CTX
;
611 mark_reg_known_zero(env
, regs
, BPF_REG_1
);
615 SRC_OP
, /* register is used as source operand */
616 DST_OP
, /* register is used as destination operand */
617 DST_OP_NO_MARK
/* same as above, check only, don't mark */
620 static void mark_reg_read(const struct bpf_verifier_state
*state
, u32 regno
)
622 struct bpf_verifier_state
*parent
= state
->parent
;
624 if (regno
== BPF_REG_FP
)
625 /* We don't need to worry about FP liveness because it's read-only */
629 /* if read wasn't screened by an earlier write ... */
630 if (state
->regs
[regno
].live
& REG_LIVE_WRITTEN
)
632 /* ... then we depend on parent's value */
633 parent
->regs
[regno
].live
|= REG_LIVE_READ
;
635 parent
= state
->parent
;
639 static int check_reg_arg(struct bpf_verifier_env
*env
, u32 regno
,
642 struct bpf_reg_state
*regs
= env
->cur_state
->regs
;
644 if (regno
>= MAX_BPF_REG
) {
645 verbose(env
, "R%d is invalid\n", regno
);
650 /* check whether register used as source operand can be read */
651 if (regs
[regno
].type
== NOT_INIT
) {
652 verbose(env
, "R%d !read_ok\n", regno
);
655 mark_reg_read(env
->cur_state
, regno
);
657 /* check whether register used as dest operand can be written to */
658 if (regno
== BPF_REG_FP
) {
659 verbose(env
, "frame pointer is read only\n");
662 regs
[regno
].live
|= REG_LIVE_WRITTEN
;
664 mark_reg_unknown(env
, regs
, regno
);
669 static bool is_spillable_regtype(enum bpf_reg_type type
)
672 case PTR_TO_MAP_VALUE
:
673 case PTR_TO_MAP_VALUE_OR_NULL
:
677 case PTR_TO_PACKET_META
:
678 case PTR_TO_PACKET_END
:
679 case CONST_PTR_TO_MAP
:
686 /* check_stack_read/write functions track spill/fill of registers,
687 * stack boundary and alignment are checked in check_mem_access()
689 static int check_stack_write(struct bpf_verifier_env
*env
,
690 struct bpf_verifier_state
*state
, int off
,
691 int size
, int value_regno
)
693 int i
, slot
= -off
- 1, spi
= slot
/ BPF_REG_SIZE
, err
;
695 err
= realloc_verifier_state(state
, round_up(slot
+ 1, BPF_REG_SIZE
),
699 /* caller checked that off % size == 0 and -MAX_BPF_STACK <= off < 0,
700 * so it's aligned access and [off, off + size) are within stack limits
702 if (!env
->allow_ptr_leaks
&&
703 state
->stack
[spi
].slot_type
[0] == STACK_SPILL
&&
704 size
!= BPF_REG_SIZE
) {
705 verbose(env
, "attempt to corrupt spilled pointer on stack\n");
709 if (value_regno
>= 0 &&
710 is_spillable_regtype(state
->regs
[value_regno
].type
)) {
712 /* register containing pointer is being spilled into stack */
713 if (size
!= BPF_REG_SIZE
) {
714 verbose(env
, "invalid size of register spill\n");
718 /* save register state */
719 state
->stack
[spi
].spilled_ptr
= state
->regs
[value_regno
];
720 state
->stack
[spi
].spilled_ptr
.live
|= REG_LIVE_WRITTEN
;
722 for (i
= 0; i
< BPF_REG_SIZE
; i
++)
723 state
->stack
[spi
].slot_type
[i
] = STACK_SPILL
;
725 /* regular write of data into stack */
726 state
->stack
[spi
].spilled_ptr
= (struct bpf_reg_state
) {};
728 for (i
= 0; i
< size
; i
++)
729 state
->stack
[spi
].slot_type
[(slot
- i
) % BPF_REG_SIZE
] =
735 static void mark_stack_slot_read(const struct bpf_verifier_state
*state
, int slot
)
737 struct bpf_verifier_state
*parent
= state
->parent
;
740 /* if read wasn't screened by an earlier write ... */
741 if (state
->stack
[slot
].spilled_ptr
.live
& REG_LIVE_WRITTEN
)
743 /* ... then we depend on parent's value */
744 parent
->stack
[slot
].spilled_ptr
.live
|= REG_LIVE_READ
;
746 parent
= state
->parent
;
750 static int check_stack_read(struct bpf_verifier_env
*env
,
751 struct bpf_verifier_state
*state
, int off
, int size
,
754 int i
, slot
= -off
- 1, spi
= slot
/ BPF_REG_SIZE
;
757 if (state
->allocated_stack
<= slot
) {
758 verbose(env
, "invalid read from stack off %d+0 size %d\n",
762 stype
= state
->stack
[spi
].slot_type
;
764 if (stype
[0] == STACK_SPILL
) {
765 if (size
!= BPF_REG_SIZE
) {
766 verbose(env
, "invalid size of register spill\n");
769 for (i
= 1; i
< BPF_REG_SIZE
; i
++) {
770 if (stype
[(slot
- i
) % BPF_REG_SIZE
] != STACK_SPILL
) {
771 verbose(env
, "corrupted spill memory\n");
776 if (value_regno
>= 0) {
777 /* restore register state from stack */
778 state
->regs
[value_regno
] = state
->stack
[spi
].spilled_ptr
;
779 mark_stack_slot_read(state
, spi
);
783 for (i
= 0; i
< size
; i
++) {
784 if (stype
[(slot
- i
) % BPF_REG_SIZE
] != STACK_MISC
) {
785 verbose(env
, "invalid read from stack off %d+%d size %d\n",
790 if (value_regno
>= 0)
791 /* have read misc data from the stack */
792 mark_reg_unknown(env
, state
->regs
, value_regno
);
797 /* check read/write into map element returned by bpf_map_lookup_elem() */
798 static int __check_map_access(struct bpf_verifier_env
*env
, u32 regno
, int off
,
801 struct bpf_reg_state
*regs
= cur_regs(env
);
802 struct bpf_map
*map
= regs
[regno
].map_ptr
;
804 if (off
< 0 || size
<= 0 || off
+ size
> map
->value_size
) {
805 verbose(env
, "invalid access to map value, value_size=%d off=%d size=%d\n",
806 map
->value_size
, off
, size
);
812 /* check read/write into a map element with possible variable offset */
813 static int check_map_access(struct bpf_verifier_env
*env
, u32 regno
,
816 struct bpf_verifier_state
*state
= env
->cur_state
;
817 struct bpf_reg_state
*reg
= &state
->regs
[regno
];
820 /* We may have adjusted the register to this map value, so we
821 * need to try adding each of min_value and max_value to off
822 * to make sure our theoretical access will be safe.
825 print_verifier_state(env
, state
);
826 /* The minimum value is only important with signed
827 * comparisons where we can't assume the floor of a
828 * value is 0. If we are using signed variables for our
829 * index'es we need to make sure that whatever we use
830 * will have a set floor within our range.
832 if (reg
->smin_value
< 0) {
833 verbose(env
, "R%d min value is negative, either use unsigned index or do a if (index >=0) check.\n",
837 err
= __check_map_access(env
, regno
, reg
->smin_value
+ off
, size
);
839 verbose(env
, "R%d min value is outside of the array range\n",
844 /* If we haven't set a max value then we need to bail since we can't be
845 * sure we won't do bad things.
846 * If reg->umax_value + off could overflow, treat that as unbounded too.
848 if (reg
->umax_value
>= BPF_MAX_VAR_OFF
) {
849 verbose(env
, "R%d unbounded memory access, make sure to bounds check any array access into a map\n",
853 err
= __check_map_access(env
, regno
, reg
->umax_value
+ off
, size
);
855 verbose(env
, "R%d max value is outside of the array range\n",
860 #define MAX_PACKET_OFF 0xffff
862 static bool may_access_direct_pkt_data(struct bpf_verifier_env
*env
,
863 const struct bpf_call_arg_meta
*meta
,
864 enum bpf_access_type t
)
866 switch (env
->prog
->type
) {
867 case BPF_PROG_TYPE_LWT_IN
:
868 case BPF_PROG_TYPE_LWT_OUT
:
869 /* dst_input() and dst_output() can't write for now */
873 case BPF_PROG_TYPE_SCHED_CLS
:
874 case BPF_PROG_TYPE_SCHED_ACT
:
875 case BPF_PROG_TYPE_XDP
:
876 case BPF_PROG_TYPE_LWT_XMIT
:
877 case BPF_PROG_TYPE_SK_SKB
:
879 return meta
->pkt_access
;
881 env
->seen_direct_write
= true;
888 static int __check_packet_access(struct bpf_verifier_env
*env
, u32 regno
,
891 struct bpf_reg_state
*regs
= cur_regs(env
);
892 struct bpf_reg_state
*reg
= ®s
[regno
];
894 if (off
< 0 || size
<= 0 || (u64
)off
+ size
> reg
->range
) {
895 verbose(env
, "invalid access to packet, off=%d size=%d, R%d(id=%d,off=%d,r=%d)\n",
896 off
, size
, regno
, reg
->id
, reg
->off
, reg
->range
);
902 static int check_packet_access(struct bpf_verifier_env
*env
, u32 regno
, int off
,
905 struct bpf_reg_state
*regs
= cur_regs(env
);
906 struct bpf_reg_state
*reg
= ®s
[regno
];
909 /* We may have added a variable offset to the packet pointer; but any
910 * reg->range we have comes after that. We are only checking the fixed
914 /* We don't allow negative numbers, because we aren't tracking enough
915 * detail to prove they're safe.
917 if (reg
->smin_value
< 0) {
918 verbose(env
, "R%d min value is negative, either use unsigned index or do a if (index >=0) check.\n",
922 err
= __check_packet_access(env
, regno
, off
, size
);
924 verbose(env
, "R%d offset is outside of the packet\n", regno
);
930 /* check access to 'struct bpf_context' fields. Supports fixed offsets only */
931 static int check_ctx_access(struct bpf_verifier_env
*env
, int insn_idx
, int off
, int size
,
932 enum bpf_access_type t
, enum bpf_reg_type
*reg_type
)
934 struct bpf_insn_access_aux info
= {
935 .reg_type
= *reg_type
,
938 if (env
->ops
->is_valid_access
&&
939 env
->ops
->is_valid_access(off
, size
, t
, &info
)) {
940 /* A non zero info.ctx_field_size indicates that this field is a
941 * candidate for later verifier transformation to load the whole
942 * field and then apply a mask when accessed with a narrower
943 * access than actual ctx access size. A zero info.ctx_field_size
944 * will only allow for whole field access and rejects any other
945 * type of narrower access.
947 *reg_type
= info
.reg_type
;
949 if (env
->analyzer_ops
)
952 env
->insn_aux_data
[insn_idx
].ctx_field_size
= info
.ctx_field_size
;
953 /* remember the offset of last byte accessed in ctx */
954 if (env
->prog
->aux
->max_ctx_offset
< off
+ size
)
955 env
->prog
->aux
->max_ctx_offset
= off
+ size
;
959 verbose(env
, "invalid bpf_context access off=%d size=%d\n", off
, size
);
963 static bool __is_pointer_value(bool allow_ptr_leaks
,
964 const struct bpf_reg_state
*reg
)
969 return reg
->type
!= SCALAR_VALUE
;
972 static bool is_pointer_value(struct bpf_verifier_env
*env
, int regno
)
974 return __is_pointer_value(env
->allow_ptr_leaks
, cur_regs(env
) + regno
);
977 static int check_pkt_ptr_alignment(struct bpf_verifier_env
*env
,
978 const struct bpf_reg_state
*reg
,
979 int off
, int size
, bool strict
)
984 /* Byte size accesses are always allowed. */
985 if (!strict
|| size
== 1)
988 /* For platforms that do not have a Kconfig enabling
989 * CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS the value of
990 * NET_IP_ALIGN is universally set to '2'. And on platforms
991 * that do set CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS, we get
992 * to this code only in strict mode where we want to emulate
993 * the NET_IP_ALIGN==2 checking. Therefore use an
994 * unconditional IP align value of '2'.
998 reg_off
= tnum_add(reg
->var_off
, tnum_const(ip_align
+ reg
->off
+ off
));
999 if (!tnum_is_aligned(reg_off
, size
)) {
1002 tnum_strn(tn_buf
, sizeof(tn_buf
), reg
->var_off
);
1004 "misaligned packet access off %d+%s+%d+%d size %d\n",
1005 ip_align
, tn_buf
, reg
->off
, off
, size
);
1012 static int check_generic_ptr_alignment(struct bpf_verifier_env
*env
,
1013 const struct bpf_reg_state
*reg
,
1014 const char *pointer_desc
,
1015 int off
, int size
, bool strict
)
1017 struct tnum reg_off
;
1019 /* Byte size accesses are always allowed. */
1020 if (!strict
|| size
== 1)
1023 reg_off
= tnum_add(reg
->var_off
, tnum_const(reg
->off
+ off
));
1024 if (!tnum_is_aligned(reg_off
, size
)) {
1027 tnum_strn(tn_buf
, sizeof(tn_buf
), reg
->var_off
);
1028 verbose(env
, "misaligned %saccess off %s+%d+%d size %d\n",
1029 pointer_desc
, tn_buf
, reg
->off
, off
, size
);
1036 static int check_ptr_alignment(struct bpf_verifier_env
*env
,
1037 const struct bpf_reg_state
*reg
,
1040 bool strict
= env
->strict_alignment
;
1041 const char *pointer_desc
= "";
1043 switch (reg
->type
) {
1045 case PTR_TO_PACKET_META
:
1046 /* Special case, because of NET_IP_ALIGN. Given metadata sits
1047 * right in front, treat it the very same way.
1049 return check_pkt_ptr_alignment(env
, reg
, off
, size
, strict
);
1050 case PTR_TO_MAP_VALUE
:
1051 pointer_desc
= "value ";
1054 pointer_desc
= "context ";
1057 pointer_desc
= "stack ";
1062 return check_generic_ptr_alignment(env
, reg
, pointer_desc
, off
, size
,
1066 /* check whether memory at (regno + off) is accessible for t = (read | write)
1067 * if t==write, value_regno is a register which value is stored into memory
1068 * if t==read, value_regno is a register which will receive the value from memory
1069 * if t==write && value_regno==-1, some unknown value is stored into memory
1070 * if t==read && value_regno==-1, don't care what we read from memory
1072 static int check_mem_access(struct bpf_verifier_env
*env
, int insn_idx
, u32 regno
, int off
,
1073 int bpf_size
, enum bpf_access_type t
,
1076 struct bpf_verifier_state
*state
= env
->cur_state
;
1077 struct bpf_reg_state
*regs
= cur_regs(env
);
1078 struct bpf_reg_state
*reg
= regs
+ regno
;
1081 size
= bpf_size_to_bytes(bpf_size
);
1085 /* alignment checks will add in reg->off themselves */
1086 err
= check_ptr_alignment(env
, reg
, off
, size
);
1090 /* for access checks, reg->off is just part of off */
1093 if (reg
->type
== PTR_TO_MAP_VALUE
) {
1094 if (t
== BPF_WRITE
&& value_regno
>= 0 &&
1095 is_pointer_value(env
, value_regno
)) {
1096 verbose(env
, "R%d leaks addr into map\n", value_regno
);
1100 err
= check_map_access(env
, regno
, off
, size
);
1101 if (!err
&& t
== BPF_READ
&& value_regno
>= 0)
1102 mark_reg_unknown(env
, regs
, value_regno
);
1104 } else if (reg
->type
== PTR_TO_CTX
) {
1105 enum bpf_reg_type reg_type
= SCALAR_VALUE
;
1107 if (t
== BPF_WRITE
&& value_regno
>= 0 &&
1108 is_pointer_value(env
, value_regno
)) {
1109 verbose(env
, "R%d leaks addr into ctx\n", value_regno
);
1112 /* ctx accesses must be at a fixed offset, so that we can
1113 * determine what type of data were returned.
1117 "dereference of modified ctx ptr R%d off=%d+%d, ctx+const is allowed, ctx+const+const is not\n",
1118 regno
, reg
->off
, off
- reg
->off
);
1121 if (!tnum_is_const(reg
->var_off
) || reg
->var_off
.value
) {
1124 tnum_strn(tn_buf
, sizeof(tn_buf
), reg
->var_off
);
1126 "variable ctx access var_off=%s off=%d size=%d",
1130 err
= check_ctx_access(env
, insn_idx
, off
, size
, t
, ®_type
);
1131 if (!err
&& t
== BPF_READ
&& value_regno
>= 0) {
1132 /* ctx access returns either a scalar, or a
1133 * PTR_TO_PACKET[_META,_END]. In the latter
1134 * case, we know the offset is zero.
1136 if (reg_type
== SCALAR_VALUE
)
1137 mark_reg_unknown(env
, regs
, value_regno
);
1139 mark_reg_known_zero(env
, regs
,
1141 regs
[value_regno
].id
= 0;
1142 regs
[value_regno
].off
= 0;
1143 regs
[value_regno
].range
= 0;
1144 regs
[value_regno
].type
= reg_type
;
1147 } else if (reg
->type
== PTR_TO_STACK
) {
1148 /* stack accesses must be at a fixed offset, so that we can
1149 * determine what type of data were returned.
1150 * See check_stack_read().
1152 if (!tnum_is_const(reg
->var_off
)) {
1155 tnum_strn(tn_buf
, sizeof(tn_buf
), reg
->var_off
);
1156 verbose(env
, "variable stack access var_off=%s off=%d size=%d",
1160 off
+= reg
->var_off
.value
;
1161 if (off
>= 0 || off
< -MAX_BPF_STACK
) {
1162 verbose(env
, "invalid stack off=%d size=%d\n", off
,
1167 if (env
->prog
->aux
->stack_depth
< -off
)
1168 env
->prog
->aux
->stack_depth
= -off
;
1171 err
= check_stack_write(env
, state
, off
, size
,
1174 err
= check_stack_read(env
, state
, off
, size
,
1176 } else if (reg_is_pkt_pointer(reg
)) {
1177 if (t
== BPF_WRITE
&& !may_access_direct_pkt_data(env
, NULL
, t
)) {
1178 verbose(env
, "cannot write into packet\n");
1181 if (t
== BPF_WRITE
&& value_regno
>= 0 &&
1182 is_pointer_value(env
, value_regno
)) {
1183 verbose(env
, "R%d leaks addr into packet\n",
1187 err
= check_packet_access(env
, regno
, off
, size
);
1188 if (!err
&& t
== BPF_READ
&& value_regno
>= 0)
1189 mark_reg_unknown(env
, regs
, value_regno
);
1191 verbose(env
, "R%d invalid mem access '%s'\n", regno
,
1192 reg_type_str
[reg
->type
]);
1196 if (!err
&& size
< BPF_REG_SIZE
&& value_regno
>= 0 && t
== BPF_READ
&&
1197 regs
[value_regno
].type
== SCALAR_VALUE
) {
1198 /* b/h/w load zero-extends, mark upper bits as known 0 */
1199 regs
[value_regno
].var_off
=
1200 tnum_cast(regs
[value_regno
].var_off
, size
);
1201 __update_reg_bounds(®s
[value_regno
]);
1206 static int check_xadd(struct bpf_verifier_env
*env
, int insn_idx
, struct bpf_insn
*insn
)
1210 if ((BPF_SIZE(insn
->code
) != BPF_W
&& BPF_SIZE(insn
->code
) != BPF_DW
) ||
1212 verbose(env
, "BPF_XADD uses reserved fields\n");
1216 /* check src1 operand */
1217 err
= check_reg_arg(env
, insn
->src_reg
, SRC_OP
);
1221 /* check src2 operand */
1222 err
= check_reg_arg(env
, insn
->dst_reg
, SRC_OP
);
1226 if (is_pointer_value(env
, insn
->src_reg
)) {
1227 verbose(env
, "R%d leaks addr into mem\n", insn
->src_reg
);
1231 /* check whether atomic_add can read the memory */
1232 err
= check_mem_access(env
, insn_idx
, insn
->dst_reg
, insn
->off
,
1233 BPF_SIZE(insn
->code
), BPF_READ
, -1);
1237 /* check whether atomic_add can write into the same memory */
1238 return check_mem_access(env
, insn_idx
, insn
->dst_reg
, insn
->off
,
1239 BPF_SIZE(insn
->code
), BPF_WRITE
, -1);
1242 /* Does this register contain a constant zero? */
1243 static bool register_is_null(struct bpf_reg_state reg
)
1245 return reg
.type
== SCALAR_VALUE
&& tnum_equals_const(reg
.var_off
, 0);
1248 /* when register 'regno' is passed into function that will read 'access_size'
1249 * bytes from that pointer, make sure that it's within stack boundary
1250 * and all elements of stack are initialized.
1251 * Unlike most pointer bounds-checking functions, this one doesn't take an
1252 * 'off' argument, so it has to add in reg->off itself.
1254 static int check_stack_boundary(struct bpf_verifier_env
*env
, int regno
,
1255 int access_size
, bool zero_size_allowed
,
1256 struct bpf_call_arg_meta
*meta
)
1258 struct bpf_verifier_state
*state
= env
->cur_state
;
1259 struct bpf_reg_state
*regs
= state
->regs
;
1260 int off
, i
, slot
, spi
;
1262 if (regs
[regno
].type
!= PTR_TO_STACK
) {
1263 /* Allow zero-byte read from NULL, regardless of pointer type */
1264 if (zero_size_allowed
&& access_size
== 0 &&
1265 register_is_null(regs
[regno
]))
1268 verbose(env
, "R%d type=%s expected=%s\n", regno
,
1269 reg_type_str
[regs
[regno
].type
],
1270 reg_type_str
[PTR_TO_STACK
]);
1274 /* Only allow fixed-offset stack reads */
1275 if (!tnum_is_const(regs
[regno
].var_off
)) {
1278 tnum_strn(tn_buf
, sizeof(tn_buf
), regs
[regno
].var_off
);
1279 verbose(env
, "invalid variable stack read R%d var_off=%s\n",
1282 off
= regs
[regno
].off
+ regs
[regno
].var_off
.value
;
1283 if (off
>= 0 || off
< -MAX_BPF_STACK
|| off
+ access_size
> 0 ||
1285 verbose(env
, "invalid stack type R%d off=%d access_size=%d\n",
1286 regno
, off
, access_size
);
1290 if (env
->prog
->aux
->stack_depth
< -off
)
1291 env
->prog
->aux
->stack_depth
= -off
;
1293 if (meta
&& meta
->raw_mode
) {
1294 meta
->access_size
= access_size
;
1295 meta
->regno
= regno
;
1299 for (i
= 0; i
< access_size
; i
++) {
1300 slot
= -(off
+ i
) - 1;
1301 spi
= slot
/ BPF_REG_SIZE
;
1302 if (state
->allocated_stack
<= slot
||
1303 state
->stack
[spi
].slot_type
[slot
% BPF_REG_SIZE
] !=
1305 verbose(env
, "invalid indirect read from stack off %d+%d size %d\n",
1306 off
, i
, access_size
);
1313 static int check_helper_mem_access(struct bpf_verifier_env
*env
, int regno
,
1314 int access_size
, bool zero_size_allowed
,
1315 struct bpf_call_arg_meta
*meta
)
1317 struct bpf_reg_state
*regs
= cur_regs(env
), *reg
= ®s
[regno
];
1319 switch (reg
->type
) {
1321 case PTR_TO_PACKET_META
:
1322 return check_packet_access(env
, regno
, reg
->off
, access_size
);
1323 case PTR_TO_MAP_VALUE
:
1324 return check_map_access(env
, regno
, reg
->off
, access_size
);
1325 default: /* scalar_value|ptr_to_stack or invalid ptr */
1326 return check_stack_boundary(env
, regno
, access_size
,
1327 zero_size_allowed
, meta
);
1331 static int check_func_arg(struct bpf_verifier_env
*env
, u32 regno
,
1332 enum bpf_arg_type arg_type
,
1333 struct bpf_call_arg_meta
*meta
)
1335 struct bpf_reg_state
*regs
= cur_regs(env
), *reg
= ®s
[regno
];
1336 enum bpf_reg_type expected_type
, type
= reg
->type
;
1339 if (arg_type
== ARG_DONTCARE
)
1342 err
= check_reg_arg(env
, regno
, SRC_OP
);
1346 if (arg_type
== ARG_ANYTHING
) {
1347 if (is_pointer_value(env
, regno
)) {
1348 verbose(env
, "R%d leaks addr into helper function\n",
1355 if (type_is_pkt_pointer(type
) &&
1356 !may_access_direct_pkt_data(env
, meta
, BPF_READ
)) {
1357 verbose(env
, "helper access to the packet is not allowed\n");
1361 if (arg_type
== ARG_PTR_TO_MAP_KEY
||
1362 arg_type
== ARG_PTR_TO_MAP_VALUE
) {
1363 expected_type
= PTR_TO_STACK
;
1364 if (!type_is_pkt_pointer(type
) &&
1365 type
!= expected_type
)
1367 } else if (arg_type
== ARG_CONST_SIZE
||
1368 arg_type
== ARG_CONST_SIZE_OR_ZERO
) {
1369 expected_type
= SCALAR_VALUE
;
1370 if (type
!= expected_type
)
1372 } else if (arg_type
== ARG_CONST_MAP_PTR
) {
1373 expected_type
= CONST_PTR_TO_MAP
;
1374 if (type
!= expected_type
)
1376 } else if (arg_type
== ARG_PTR_TO_CTX
) {
1377 expected_type
= PTR_TO_CTX
;
1378 if (type
!= expected_type
)
1380 } else if (arg_type
== ARG_PTR_TO_MEM
||
1381 arg_type
== ARG_PTR_TO_UNINIT_MEM
) {
1382 expected_type
= PTR_TO_STACK
;
1383 /* One exception here. In case function allows for NULL to be
1384 * passed in as argument, it's a SCALAR_VALUE type. Final test
1385 * happens during stack boundary checking.
1387 if (register_is_null(*reg
))
1388 /* final test in check_stack_boundary() */;
1389 else if (!type_is_pkt_pointer(type
) &&
1390 type
!= PTR_TO_MAP_VALUE
&&
1391 type
!= expected_type
)
1393 meta
->raw_mode
= arg_type
== ARG_PTR_TO_UNINIT_MEM
;
1395 verbose(env
, "unsupported arg_type %d\n", arg_type
);
1399 if (arg_type
== ARG_CONST_MAP_PTR
) {
1400 /* bpf_map_xxx(map_ptr) call: remember that map_ptr */
1401 meta
->map_ptr
= reg
->map_ptr
;
1402 } else if (arg_type
== ARG_PTR_TO_MAP_KEY
) {
1403 /* bpf_map_xxx(..., map_ptr, ..., key) call:
1404 * check that [key, key + map->key_size) are within
1405 * stack limits and initialized
1407 if (!meta
->map_ptr
) {
1408 /* in function declaration map_ptr must come before
1409 * map_key, so that it's verified and known before
1410 * we have to check map_key here. Otherwise it means
1411 * that kernel subsystem misconfigured verifier
1413 verbose(env
, "invalid map_ptr to access map->key\n");
1416 if (type_is_pkt_pointer(type
))
1417 err
= check_packet_access(env
, regno
, reg
->off
,
1418 meta
->map_ptr
->key_size
);
1420 err
= check_stack_boundary(env
, regno
,
1421 meta
->map_ptr
->key_size
,
1423 } else if (arg_type
== ARG_PTR_TO_MAP_VALUE
) {
1424 /* bpf_map_xxx(..., map_ptr, ..., value) call:
1425 * check [value, value + map->value_size) validity
1427 if (!meta
->map_ptr
) {
1428 /* kernel subsystem misconfigured verifier */
1429 verbose(env
, "invalid map_ptr to access map->value\n");
1432 if (type_is_pkt_pointer(type
))
1433 err
= check_packet_access(env
, regno
, reg
->off
,
1434 meta
->map_ptr
->value_size
);
1436 err
= check_stack_boundary(env
, regno
,
1437 meta
->map_ptr
->value_size
,
1439 } else if (arg_type
== ARG_CONST_SIZE
||
1440 arg_type
== ARG_CONST_SIZE_OR_ZERO
) {
1441 bool zero_size_allowed
= (arg_type
== ARG_CONST_SIZE_OR_ZERO
);
1443 /* bpf_xxx(..., buf, len) call will access 'len' bytes
1444 * from stack pointer 'buf'. Check it
1445 * note: regno == len, regno - 1 == buf
1448 /* kernel subsystem misconfigured verifier */
1450 "ARG_CONST_SIZE cannot be first argument\n");
1454 /* The register is SCALAR_VALUE; the access check
1455 * happens using its boundaries.
1458 if (!tnum_is_const(reg
->var_off
))
1459 /* For unprivileged variable accesses, disable raw
1460 * mode so that the program is required to
1461 * initialize all the memory that the helper could
1462 * just partially fill up.
1466 if (reg
->smin_value
< 0) {
1467 verbose(env
, "R%d min value is negative, either use unsigned or 'var &= const'\n",
1472 if (reg
->umin_value
== 0) {
1473 err
= check_helper_mem_access(env
, regno
- 1, 0,
1480 if (reg
->umax_value
>= BPF_MAX_VAR_SIZ
) {
1481 verbose(env
, "R%d unbounded memory access, use 'var &= const' or 'if (var < const)'\n",
1485 err
= check_helper_mem_access(env
, regno
- 1,
1487 zero_size_allowed
, meta
);
1492 verbose(env
, "R%d type=%s expected=%s\n", regno
,
1493 reg_type_str
[type
], reg_type_str
[expected_type
]);
1497 static int check_map_func_compatibility(struct bpf_verifier_env
*env
,
1498 struct bpf_map
*map
, int func_id
)
1503 /* We need a two way check, first is from map perspective ... */
1504 switch (map
->map_type
) {
1505 case BPF_MAP_TYPE_PROG_ARRAY
:
1506 if (func_id
!= BPF_FUNC_tail_call
)
1509 case BPF_MAP_TYPE_PERF_EVENT_ARRAY
:
1510 if (func_id
!= BPF_FUNC_perf_event_read
&&
1511 func_id
!= BPF_FUNC_perf_event_output
&&
1512 func_id
!= BPF_FUNC_perf_event_read_value
)
1515 case BPF_MAP_TYPE_STACK_TRACE
:
1516 if (func_id
!= BPF_FUNC_get_stackid
)
1519 case BPF_MAP_TYPE_CGROUP_ARRAY
:
1520 if (func_id
!= BPF_FUNC_skb_under_cgroup
&&
1521 func_id
!= BPF_FUNC_current_task_under_cgroup
)
1524 /* devmap returns a pointer to a live net_device ifindex that we cannot
1525 * allow to be modified from bpf side. So do not allow lookup elements
1528 case BPF_MAP_TYPE_DEVMAP
:
1529 if (func_id
!= BPF_FUNC_redirect_map
)
1532 /* Restrict bpf side of cpumap, open when use-cases appear */
1533 case BPF_MAP_TYPE_CPUMAP
:
1534 if (func_id
!= BPF_FUNC_redirect_map
)
1537 case BPF_MAP_TYPE_ARRAY_OF_MAPS
:
1538 case BPF_MAP_TYPE_HASH_OF_MAPS
:
1539 if (func_id
!= BPF_FUNC_map_lookup_elem
)
1542 case BPF_MAP_TYPE_SOCKMAP
:
1543 if (func_id
!= BPF_FUNC_sk_redirect_map
&&
1544 func_id
!= BPF_FUNC_sock_map_update
&&
1545 func_id
!= BPF_FUNC_map_delete_elem
)
1552 /* ... and second from the function itself. */
1554 case BPF_FUNC_tail_call
:
1555 if (map
->map_type
!= BPF_MAP_TYPE_PROG_ARRAY
)
1558 case BPF_FUNC_perf_event_read
:
1559 case BPF_FUNC_perf_event_output
:
1560 case BPF_FUNC_perf_event_read_value
:
1561 if (map
->map_type
!= BPF_MAP_TYPE_PERF_EVENT_ARRAY
)
1564 case BPF_FUNC_get_stackid
:
1565 if (map
->map_type
!= BPF_MAP_TYPE_STACK_TRACE
)
1568 case BPF_FUNC_current_task_under_cgroup
:
1569 case BPF_FUNC_skb_under_cgroup
:
1570 if (map
->map_type
!= BPF_MAP_TYPE_CGROUP_ARRAY
)
1573 case BPF_FUNC_redirect_map
:
1574 if (map
->map_type
!= BPF_MAP_TYPE_DEVMAP
&&
1575 map
->map_type
!= BPF_MAP_TYPE_CPUMAP
)
1578 case BPF_FUNC_sk_redirect_map
:
1579 if (map
->map_type
!= BPF_MAP_TYPE_SOCKMAP
)
1582 case BPF_FUNC_sock_map_update
:
1583 if (map
->map_type
!= BPF_MAP_TYPE_SOCKMAP
)
1592 verbose(env
, "cannot pass map_type %d into func %s#%d\n",
1593 map
->map_type
, func_id_name(func_id
), func_id
);
1597 static int check_raw_mode(const struct bpf_func_proto
*fn
)
1601 if (fn
->arg1_type
== ARG_PTR_TO_UNINIT_MEM
)
1603 if (fn
->arg2_type
== ARG_PTR_TO_UNINIT_MEM
)
1605 if (fn
->arg3_type
== ARG_PTR_TO_UNINIT_MEM
)
1607 if (fn
->arg4_type
== ARG_PTR_TO_UNINIT_MEM
)
1609 if (fn
->arg5_type
== ARG_PTR_TO_UNINIT_MEM
)
1612 return count
> 1 ? -EINVAL
: 0;
1615 /* Packet data might have moved, any old PTR_TO_PACKET[_META,_END]
1616 * are now invalid, so turn them into unknown SCALAR_VALUE.
1618 static void clear_all_pkt_pointers(struct bpf_verifier_env
*env
)
1620 struct bpf_verifier_state
*state
= env
->cur_state
;
1621 struct bpf_reg_state
*regs
= state
->regs
, *reg
;
1624 for (i
= 0; i
< MAX_BPF_REG
; i
++)
1625 if (reg_is_pkt_pointer_any(®s
[i
]))
1626 mark_reg_unknown(env
, regs
, i
);
1628 for (i
= 0; i
< state
->allocated_stack
/ BPF_REG_SIZE
; i
++) {
1629 if (state
->stack
[i
].slot_type
[0] != STACK_SPILL
)
1631 reg
= &state
->stack
[i
].spilled_ptr
;
1632 if (reg_is_pkt_pointer_any(reg
))
1633 __mark_reg_unknown(reg
);
1637 static int check_call(struct bpf_verifier_env
*env
, int func_id
, int insn_idx
)
1639 const struct bpf_func_proto
*fn
= NULL
;
1640 struct bpf_reg_state
*regs
;
1641 struct bpf_call_arg_meta meta
;
1645 /* find function prototype */
1646 if (func_id
< 0 || func_id
>= __BPF_FUNC_MAX_ID
) {
1647 verbose(env
, "invalid func %s#%d\n", func_id_name(func_id
),
1652 if (env
->ops
->get_func_proto
)
1653 fn
= env
->ops
->get_func_proto(func_id
);
1656 verbose(env
, "unknown func %s#%d\n", func_id_name(func_id
),
1661 /* eBPF programs must be GPL compatible to use GPL-ed functions */
1662 if (!env
->prog
->gpl_compatible
&& fn
->gpl_only
) {
1663 verbose(env
, "cannot call GPL only function from proprietary program\n");
1667 changes_data
= bpf_helper_changes_pkt_data(fn
->func
);
1669 memset(&meta
, 0, sizeof(meta
));
1670 meta
.pkt_access
= fn
->pkt_access
;
1672 /* We only support one arg being in raw mode at the moment, which
1673 * is sufficient for the helper functions we have right now.
1675 err
= check_raw_mode(fn
);
1677 verbose(env
, "kernel subsystem misconfigured func %s#%d\n",
1678 func_id_name(func_id
), func_id
);
1683 err
= check_func_arg(env
, BPF_REG_1
, fn
->arg1_type
, &meta
);
1686 err
= check_func_arg(env
, BPF_REG_2
, fn
->arg2_type
, &meta
);
1689 err
= check_func_arg(env
, BPF_REG_3
, fn
->arg3_type
, &meta
);
1692 err
= check_func_arg(env
, BPF_REG_4
, fn
->arg4_type
, &meta
);
1695 err
= check_func_arg(env
, BPF_REG_5
, fn
->arg5_type
, &meta
);
1699 /* Mark slots with STACK_MISC in case of raw mode, stack offset
1700 * is inferred from register state.
1702 for (i
= 0; i
< meta
.access_size
; i
++) {
1703 err
= check_mem_access(env
, insn_idx
, meta
.regno
, i
, BPF_B
, BPF_WRITE
, -1);
1708 regs
= cur_regs(env
);
1709 /* reset caller saved regs */
1710 for (i
= 0; i
< CALLER_SAVED_REGS
; i
++) {
1711 mark_reg_not_init(env
, regs
, caller_saved
[i
]);
1712 check_reg_arg(env
, caller_saved
[i
], DST_OP_NO_MARK
);
1715 /* update return register (already marked as written above) */
1716 if (fn
->ret_type
== RET_INTEGER
) {
1717 /* sets type to SCALAR_VALUE */
1718 mark_reg_unknown(env
, regs
, BPF_REG_0
);
1719 } else if (fn
->ret_type
== RET_VOID
) {
1720 regs
[BPF_REG_0
].type
= NOT_INIT
;
1721 } else if (fn
->ret_type
== RET_PTR_TO_MAP_VALUE_OR_NULL
) {
1722 struct bpf_insn_aux_data
*insn_aux
;
1724 regs
[BPF_REG_0
].type
= PTR_TO_MAP_VALUE_OR_NULL
;
1725 /* There is no offset yet applied, variable or fixed */
1726 mark_reg_known_zero(env
, regs
, BPF_REG_0
);
1727 regs
[BPF_REG_0
].off
= 0;
1728 /* remember map_ptr, so that check_map_access()
1729 * can check 'value_size' boundary of memory access
1730 * to map element returned from bpf_map_lookup_elem()
1732 if (meta
.map_ptr
== NULL
) {
1734 "kernel subsystem misconfigured verifier\n");
1737 regs
[BPF_REG_0
].map_ptr
= meta
.map_ptr
;
1738 regs
[BPF_REG_0
].id
= ++env
->id_gen
;
1739 insn_aux
= &env
->insn_aux_data
[insn_idx
];
1740 if (!insn_aux
->map_ptr
)
1741 insn_aux
->map_ptr
= meta
.map_ptr
;
1742 else if (insn_aux
->map_ptr
!= meta
.map_ptr
)
1743 insn_aux
->map_ptr
= BPF_MAP_PTR_POISON
;
1745 verbose(env
, "unknown return type %d of func %s#%d\n",
1746 fn
->ret_type
, func_id_name(func_id
), func_id
);
1750 err
= check_map_func_compatibility(env
, meta
.map_ptr
, func_id
);
1755 clear_all_pkt_pointers(env
);
1759 static void coerce_reg_to_32(struct bpf_reg_state
*reg
)
1761 /* clear high 32 bits */
1762 reg
->var_off
= tnum_cast(reg
->var_off
, 4);
1764 __update_reg_bounds(reg
);
1767 static bool signed_add_overflows(s64 a
, s64 b
)
1769 /* Do the add in u64, where overflow is well-defined */
1770 s64 res
= (s64
)((u64
)a
+ (u64
)b
);
1777 static bool signed_sub_overflows(s64 a
, s64 b
)
1779 /* Do the sub in u64, where overflow is well-defined */
1780 s64 res
= (s64
)((u64
)a
- (u64
)b
);
1787 /* Handles arithmetic on a pointer and a scalar: computes new min/max and var_off.
1788 * Caller should also handle BPF_MOV case separately.
1789 * If we return -EACCES, caller may want to try again treating pointer as a
1790 * scalar. So we only emit a diagnostic if !env->allow_ptr_leaks.
1792 static int adjust_ptr_min_max_vals(struct bpf_verifier_env
*env
,
1793 struct bpf_insn
*insn
,
1794 const struct bpf_reg_state
*ptr_reg
,
1795 const struct bpf_reg_state
*off_reg
)
1797 struct bpf_reg_state
*regs
= cur_regs(env
), *dst_reg
;
1798 bool known
= tnum_is_const(off_reg
->var_off
);
1799 s64 smin_val
= off_reg
->smin_value
, smax_val
= off_reg
->smax_value
,
1800 smin_ptr
= ptr_reg
->smin_value
, smax_ptr
= ptr_reg
->smax_value
;
1801 u64 umin_val
= off_reg
->umin_value
, umax_val
= off_reg
->umax_value
,
1802 umin_ptr
= ptr_reg
->umin_value
, umax_ptr
= ptr_reg
->umax_value
;
1803 u8 opcode
= BPF_OP(insn
->code
);
1804 u32 dst
= insn
->dst_reg
;
1806 dst_reg
= ®s
[dst
];
1808 if (WARN_ON_ONCE(known
&& (smin_val
!= smax_val
))) {
1809 print_verifier_state(env
, env
->cur_state
);
1811 "verifier internal error: known but bad sbounds\n");
1814 if (WARN_ON_ONCE(known
&& (umin_val
!= umax_val
))) {
1815 print_verifier_state(env
, env
->cur_state
);
1817 "verifier internal error: known but bad ubounds\n");
1821 if (BPF_CLASS(insn
->code
) != BPF_ALU64
) {
1822 /* 32-bit ALU ops on pointers produce (meaningless) scalars */
1823 if (!env
->allow_ptr_leaks
)
1825 "R%d 32-bit pointer arithmetic prohibited\n",
1830 if (ptr_reg
->type
== PTR_TO_MAP_VALUE_OR_NULL
) {
1831 if (!env
->allow_ptr_leaks
)
1832 verbose(env
, "R%d pointer arithmetic on PTR_TO_MAP_VALUE_OR_NULL prohibited, null-check it first\n",
1836 if (ptr_reg
->type
== CONST_PTR_TO_MAP
) {
1837 if (!env
->allow_ptr_leaks
)
1838 verbose(env
, "R%d pointer arithmetic on CONST_PTR_TO_MAP prohibited\n",
1842 if (ptr_reg
->type
== PTR_TO_PACKET_END
) {
1843 if (!env
->allow_ptr_leaks
)
1844 verbose(env
, "R%d pointer arithmetic on PTR_TO_PACKET_END prohibited\n",
1849 /* In case of 'scalar += pointer', dst_reg inherits pointer type and id.
1850 * The id may be overwritten later if we create a new variable offset.
1852 dst_reg
->type
= ptr_reg
->type
;
1853 dst_reg
->id
= ptr_reg
->id
;
1857 /* We can take a fixed offset as long as it doesn't overflow
1858 * the s32 'off' field
1860 if (known
&& (ptr_reg
->off
+ smin_val
==
1861 (s64
)(s32
)(ptr_reg
->off
+ smin_val
))) {
1862 /* pointer += K. Accumulate it into fixed offset */
1863 dst_reg
->smin_value
= smin_ptr
;
1864 dst_reg
->smax_value
= smax_ptr
;
1865 dst_reg
->umin_value
= umin_ptr
;
1866 dst_reg
->umax_value
= umax_ptr
;
1867 dst_reg
->var_off
= ptr_reg
->var_off
;
1868 dst_reg
->off
= ptr_reg
->off
+ smin_val
;
1869 dst_reg
->range
= ptr_reg
->range
;
1872 /* A new variable offset is created. Note that off_reg->off
1873 * == 0, since it's a scalar.
1874 * dst_reg gets the pointer type and since some positive
1875 * integer value was added to the pointer, give it a new 'id'
1876 * if it's a PTR_TO_PACKET.
1877 * this creates a new 'base' pointer, off_reg (variable) gets
1878 * added into the variable offset, and we copy the fixed offset
1881 if (signed_add_overflows(smin_ptr
, smin_val
) ||
1882 signed_add_overflows(smax_ptr
, smax_val
)) {
1883 dst_reg
->smin_value
= S64_MIN
;
1884 dst_reg
->smax_value
= S64_MAX
;
1886 dst_reg
->smin_value
= smin_ptr
+ smin_val
;
1887 dst_reg
->smax_value
= smax_ptr
+ smax_val
;
1889 if (umin_ptr
+ umin_val
< umin_ptr
||
1890 umax_ptr
+ umax_val
< umax_ptr
) {
1891 dst_reg
->umin_value
= 0;
1892 dst_reg
->umax_value
= U64_MAX
;
1894 dst_reg
->umin_value
= umin_ptr
+ umin_val
;
1895 dst_reg
->umax_value
= umax_ptr
+ umax_val
;
1897 dst_reg
->var_off
= tnum_add(ptr_reg
->var_off
, off_reg
->var_off
);
1898 dst_reg
->off
= ptr_reg
->off
;
1899 if (reg_is_pkt_pointer(ptr_reg
)) {
1900 dst_reg
->id
= ++env
->id_gen
;
1901 /* something was added to pkt_ptr, set range to zero */
1906 if (dst_reg
== off_reg
) {
1907 /* scalar -= pointer. Creates an unknown scalar */
1908 if (!env
->allow_ptr_leaks
)
1909 verbose(env
, "R%d tried to subtract pointer from scalar\n",
1913 /* We don't allow subtraction from FP, because (according to
1914 * test_verifier.c test "invalid fp arithmetic", JITs might not
1915 * be able to deal with it.
1917 if (ptr_reg
->type
== PTR_TO_STACK
) {
1918 if (!env
->allow_ptr_leaks
)
1919 verbose(env
, "R%d subtraction from stack pointer prohibited\n",
1923 if (known
&& (ptr_reg
->off
- smin_val
==
1924 (s64
)(s32
)(ptr_reg
->off
- smin_val
))) {
1925 /* pointer -= K. Subtract it from fixed offset */
1926 dst_reg
->smin_value
= smin_ptr
;
1927 dst_reg
->smax_value
= smax_ptr
;
1928 dst_reg
->umin_value
= umin_ptr
;
1929 dst_reg
->umax_value
= umax_ptr
;
1930 dst_reg
->var_off
= ptr_reg
->var_off
;
1931 dst_reg
->id
= ptr_reg
->id
;
1932 dst_reg
->off
= ptr_reg
->off
- smin_val
;
1933 dst_reg
->range
= ptr_reg
->range
;
1936 /* A new variable offset is created. If the subtrahend is known
1937 * nonnegative, then any reg->range we had before is still good.
1939 if (signed_sub_overflows(smin_ptr
, smax_val
) ||
1940 signed_sub_overflows(smax_ptr
, smin_val
)) {
1941 /* Overflow possible, we know nothing */
1942 dst_reg
->smin_value
= S64_MIN
;
1943 dst_reg
->smax_value
= S64_MAX
;
1945 dst_reg
->smin_value
= smin_ptr
- smax_val
;
1946 dst_reg
->smax_value
= smax_ptr
- smin_val
;
1948 if (umin_ptr
< umax_val
) {
1949 /* Overflow possible, we know nothing */
1950 dst_reg
->umin_value
= 0;
1951 dst_reg
->umax_value
= U64_MAX
;
1953 /* Cannot overflow (as long as bounds are consistent) */
1954 dst_reg
->umin_value
= umin_ptr
- umax_val
;
1955 dst_reg
->umax_value
= umax_ptr
- umin_val
;
1957 dst_reg
->var_off
= tnum_sub(ptr_reg
->var_off
, off_reg
->var_off
);
1958 dst_reg
->off
= ptr_reg
->off
;
1959 if (reg_is_pkt_pointer(ptr_reg
)) {
1960 dst_reg
->id
= ++env
->id_gen
;
1961 /* something was added to pkt_ptr, set range to zero */
1969 /* bitwise ops on pointers are troublesome, prohibit for now.
1970 * (However, in principle we could allow some cases, e.g.
1971 * ptr &= ~3 which would reduce min_value by 3.)
1973 if (!env
->allow_ptr_leaks
)
1974 verbose(env
, "R%d bitwise operator %s on pointer prohibited\n",
1975 dst
, bpf_alu_string
[opcode
>> 4]);
1978 /* other operators (e.g. MUL,LSH) produce non-pointer results */
1979 if (!env
->allow_ptr_leaks
)
1980 verbose(env
, "R%d pointer arithmetic with %s operator prohibited\n",
1981 dst
, bpf_alu_string
[opcode
>> 4]);
1985 __update_reg_bounds(dst_reg
);
1986 __reg_deduce_bounds(dst_reg
);
1987 __reg_bound_offset(dst_reg
);
1991 static int adjust_scalar_min_max_vals(struct bpf_verifier_env
*env
,
1992 struct bpf_insn
*insn
,
1993 struct bpf_reg_state
*dst_reg
,
1994 struct bpf_reg_state src_reg
)
1996 struct bpf_reg_state
*regs
= cur_regs(env
);
1997 u8 opcode
= BPF_OP(insn
->code
);
1998 bool src_known
, dst_known
;
1999 s64 smin_val
, smax_val
;
2000 u64 umin_val
, umax_val
;
2002 if (BPF_CLASS(insn
->code
) != BPF_ALU64
) {
2003 /* 32-bit ALU ops are (32,32)->64 */
2004 coerce_reg_to_32(dst_reg
);
2005 coerce_reg_to_32(&src_reg
);
2007 smin_val
= src_reg
.smin_value
;
2008 smax_val
= src_reg
.smax_value
;
2009 umin_val
= src_reg
.umin_value
;
2010 umax_val
= src_reg
.umax_value
;
2011 src_known
= tnum_is_const(src_reg
.var_off
);
2012 dst_known
= tnum_is_const(dst_reg
->var_off
);
2016 if (signed_add_overflows(dst_reg
->smin_value
, smin_val
) ||
2017 signed_add_overflows(dst_reg
->smax_value
, smax_val
)) {
2018 dst_reg
->smin_value
= S64_MIN
;
2019 dst_reg
->smax_value
= S64_MAX
;
2021 dst_reg
->smin_value
+= smin_val
;
2022 dst_reg
->smax_value
+= smax_val
;
2024 if (dst_reg
->umin_value
+ umin_val
< umin_val
||
2025 dst_reg
->umax_value
+ umax_val
< umax_val
) {
2026 dst_reg
->umin_value
= 0;
2027 dst_reg
->umax_value
= U64_MAX
;
2029 dst_reg
->umin_value
+= umin_val
;
2030 dst_reg
->umax_value
+= umax_val
;
2032 dst_reg
->var_off
= tnum_add(dst_reg
->var_off
, src_reg
.var_off
);
2035 if (signed_sub_overflows(dst_reg
->smin_value
, smax_val
) ||
2036 signed_sub_overflows(dst_reg
->smax_value
, smin_val
)) {
2037 /* Overflow possible, we know nothing */
2038 dst_reg
->smin_value
= S64_MIN
;
2039 dst_reg
->smax_value
= S64_MAX
;
2041 dst_reg
->smin_value
-= smax_val
;
2042 dst_reg
->smax_value
-= smin_val
;
2044 if (dst_reg
->umin_value
< umax_val
) {
2045 /* Overflow possible, we know nothing */
2046 dst_reg
->umin_value
= 0;
2047 dst_reg
->umax_value
= U64_MAX
;
2049 /* Cannot overflow (as long as bounds are consistent) */
2050 dst_reg
->umin_value
-= umax_val
;
2051 dst_reg
->umax_value
-= umin_val
;
2053 dst_reg
->var_off
= tnum_sub(dst_reg
->var_off
, src_reg
.var_off
);
2056 dst_reg
->var_off
= tnum_mul(dst_reg
->var_off
, src_reg
.var_off
);
2057 if (smin_val
< 0 || dst_reg
->smin_value
< 0) {
2058 /* Ain't nobody got time to multiply that sign */
2059 __mark_reg_unbounded(dst_reg
);
2060 __update_reg_bounds(dst_reg
);
2063 /* Both values are positive, so we can work with unsigned and
2064 * copy the result to signed (unless it exceeds S64_MAX).
2066 if (umax_val
> U32_MAX
|| dst_reg
->umax_value
> U32_MAX
) {
2067 /* Potential overflow, we know nothing */
2068 __mark_reg_unbounded(dst_reg
);
2069 /* (except what we can learn from the var_off) */
2070 __update_reg_bounds(dst_reg
);
2073 dst_reg
->umin_value
*= umin_val
;
2074 dst_reg
->umax_value
*= umax_val
;
2075 if (dst_reg
->umax_value
> S64_MAX
) {
2076 /* Overflow possible, we know nothing */
2077 dst_reg
->smin_value
= S64_MIN
;
2078 dst_reg
->smax_value
= S64_MAX
;
2080 dst_reg
->smin_value
= dst_reg
->umin_value
;
2081 dst_reg
->smax_value
= dst_reg
->umax_value
;
2085 if (src_known
&& dst_known
) {
2086 __mark_reg_known(dst_reg
, dst_reg
->var_off
.value
&
2087 src_reg
.var_off
.value
);
2090 /* We get our minimum from the var_off, since that's inherently
2091 * bitwise. Our maximum is the minimum of the operands' maxima.
2093 dst_reg
->var_off
= tnum_and(dst_reg
->var_off
, src_reg
.var_off
);
2094 dst_reg
->umin_value
= dst_reg
->var_off
.value
;
2095 dst_reg
->umax_value
= min(dst_reg
->umax_value
, umax_val
);
2096 if (dst_reg
->smin_value
< 0 || smin_val
< 0) {
2097 /* Lose signed bounds when ANDing negative numbers,
2098 * ain't nobody got time for that.
2100 dst_reg
->smin_value
= S64_MIN
;
2101 dst_reg
->smax_value
= S64_MAX
;
2103 /* ANDing two positives gives a positive, so safe to
2104 * cast result into s64.
2106 dst_reg
->smin_value
= dst_reg
->umin_value
;
2107 dst_reg
->smax_value
= dst_reg
->umax_value
;
2109 /* We may learn something more from the var_off */
2110 __update_reg_bounds(dst_reg
);
2113 if (src_known
&& dst_known
) {
2114 __mark_reg_known(dst_reg
, dst_reg
->var_off
.value
|
2115 src_reg
.var_off
.value
);
2118 /* We get our maximum from the var_off, and our minimum is the
2119 * maximum of the operands' minima
2121 dst_reg
->var_off
= tnum_or(dst_reg
->var_off
, src_reg
.var_off
);
2122 dst_reg
->umin_value
= max(dst_reg
->umin_value
, umin_val
);
2123 dst_reg
->umax_value
= dst_reg
->var_off
.value
|
2124 dst_reg
->var_off
.mask
;
2125 if (dst_reg
->smin_value
< 0 || smin_val
< 0) {
2126 /* Lose signed bounds when ORing negative numbers,
2127 * ain't nobody got time for that.
2129 dst_reg
->smin_value
= S64_MIN
;
2130 dst_reg
->smax_value
= S64_MAX
;
2132 /* ORing two positives gives a positive, so safe to
2133 * cast result into s64.
2135 dst_reg
->smin_value
= dst_reg
->umin_value
;
2136 dst_reg
->smax_value
= dst_reg
->umax_value
;
2138 /* We may learn something more from the var_off */
2139 __update_reg_bounds(dst_reg
);
2142 if (umax_val
> 63) {
2143 /* Shifts greater than 63 are undefined. This includes
2144 * shifts by a negative number.
2146 mark_reg_unknown(env
, regs
, insn
->dst_reg
);
2149 /* We lose all sign bit information (except what we can pick
2152 dst_reg
->smin_value
= S64_MIN
;
2153 dst_reg
->smax_value
= S64_MAX
;
2154 /* If we might shift our top bit out, then we know nothing */
2155 if (dst_reg
->umax_value
> 1ULL << (63 - umax_val
)) {
2156 dst_reg
->umin_value
= 0;
2157 dst_reg
->umax_value
= U64_MAX
;
2159 dst_reg
->umin_value
<<= umin_val
;
2160 dst_reg
->umax_value
<<= umax_val
;
2163 dst_reg
->var_off
= tnum_lshift(dst_reg
->var_off
, umin_val
);
2165 dst_reg
->var_off
= tnum_lshift(tnum_unknown
, umin_val
);
2166 /* We may learn something more from the var_off */
2167 __update_reg_bounds(dst_reg
);
2170 if (umax_val
> 63) {
2171 /* Shifts greater than 63 are undefined. This includes
2172 * shifts by a negative number.
2174 mark_reg_unknown(env
, regs
, insn
->dst_reg
);
2177 /* BPF_RSH is an unsigned shift, so make the appropriate casts */
2178 if (dst_reg
->smin_value
< 0) {
2180 /* Sign bit will be cleared */
2181 dst_reg
->smin_value
= 0;
2183 /* Lost sign bit information */
2184 dst_reg
->smin_value
= S64_MIN
;
2185 dst_reg
->smax_value
= S64_MAX
;
2188 dst_reg
->smin_value
=
2189 (u64
)(dst_reg
->smin_value
) >> umax_val
;
2192 dst_reg
->var_off
= tnum_rshift(dst_reg
->var_off
,
2195 dst_reg
->var_off
= tnum_rshift(tnum_unknown
, umin_val
);
2196 dst_reg
->umin_value
>>= umax_val
;
2197 dst_reg
->umax_value
>>= umin_val
;
2198 /* We may learn something more from the var_off */
2199 __update_reg_bounds(dst_reg
);
2202 mark_reg_unknown(env
, regs
, insn
->dst_reg
);
2206 __reg_deduce_bounds(dst_reg
);
2207 __reg_bound_offset(dst_reg
);
2211 /* Handles ALU ops other than BPF_END, BPF_NEG and BPF_MOV: computes new min/max
2214 static int adjust_reg_min_max_vals(struct bpf_verifier_env
*env
,
2215 struct bpf_insn
*insn
)
2217 struct bpf_reg_state
*regs
= cur_regs(env
), *dst_reg
, *src_reg
;
2218 struct bpf_reg_state
*ptr_reg
= NULL
, off_reg
= {0};
2219 u8 opcode
= BPF_OP(insn
->code
);
2222 dst_reg
= ®s
[insn
->dst_reg
];
2224 if (dst_reg
->type
!= SCALAR_VALUE
)
2226 if (BPF_SRC(insn
->code
) == BPF_X
) {
2227 src_reg
= ®s
[insn
->src_reg
];
2228 if (src_reg
->type
!= SCALAR_VALUE
) {
2229 if (dst_reg
->type
!= SCALAR_VALUE
) {
2230 /* Combining two pointers by any ALU op yields
2231 * an arbitrary scalar.
2233 if (!env
->allow_ptr_leaks
) {
2234 verbose(env
, "R%d pointer %s pointer prohibited\n",
2236 bpf_alu_string
[opcode
>> 4]);
2239 mark_reg_unknown(env
, regs
, insn
->dst_reg
);
2242 /* scalar += pointer
2243 * This is legal, but we have to reverse our
2244 * src/dest handling in computing the range
2246 rc
= adjust_ptr_min_max_vals(env
, insn
,
2248 if (rc
== -EACCES
&& env
->allow_ptr_leaks
) {
2249 /* scalar += unknown scalar */
2250 __mark_reg_unknown(&off_reg
);
2251 return adjust_scalar_min_max_vals(
2257 } else if (ptr_reg
) {
2258 /* pointer += scalar */
2259 rc
= adjust_ptr_min_max_vals(env
, insn
,
2261 if (rc
== -EACCES
&& env
->allow_ptr_leaks
) {
2262 /* unknown scalar += scalar */
2263 __mark_reg_unknown(dst_reg
);
2264 return adjust_scalar_min_max_vals(
2265 env
, insn
, dst_reg
, *src_reg
);
2270 /* Pretend the src is a reg with a known value, since we only
2271 * need to be able to read from this state.
2273 off_reg
.type
= SCALAR_VALUE
;
2274 __mark_reg_known(&off_reg
, insn
->imm
);
2276 if (ptr_reg
) { /* pointer += K */
2277 rc
= adjust_ptr_min_max_vals(env
, insn
,
2279 if (rc
== -EACCES
&& env
->allow_ptr_leaks
) {
2280 /* unknown scalar += K */
2281 __mark_reg_unknown(dst_reg
);
2282 return adjust_scalar_min_max_vals(
2283 env
, insn
, dst_reg
, off_reg
);
2289 /* Got here implies adding two SCALAR_VALUEs */
2290 if (WARN_ON_ONCE(ptr_reg
)) {
2291 print_verifier_state(env
, env
->cur_state
);
2292 verbose(env
, "verifier internal error: unexpected ptr_reg\n");
2295 if (WARN_ON(!src_reg
)) {
2296 print_verifier_state(env
, env
->cur_state
);
2297 verbose(env
, "verifier internal error: no src_reg\n");
2300 return adjust_scalar_min_max_vals(env
, insn
, dst_reg
, *src_reg
);
2303 /* check validity of 32-bit and 64-bit arithmetic operations */
2304 static int check_alu_op(struct bpf_verifier_env
*env
, struct bpf_insn
*insn
)
2306 struct bpf_reg_state
*regs
= cur_regs(env
);
2307 u8 opcode
= BPF_OP(insn
->code
);
2310 if (opcode
== BPF_END
|| opcode
== BPF_NEG
) {
2311 if (opcode
== BPF_NEG
) {
2312 if (BPF_SRC(insn
->code
) != 0 ||
2313 insn
->src_reg
!= BPF_REG_0
||
2314 insn
->off
!= 0 || insn
->imm
!= 0) {
2315 verbose(env
, "BPF_NEG uses reserved fields\n");
2319 if (insn
->src_reg
!= BPF_REG_0
|| insn
->off
!= 0 ||
2320 (insn
->imm
!= 16 && insn
->imm
!= 32 && insn
->imm
!= 64) ||
2321 BPF_CLASS(insn
->code
) == BPF_ALU64
) {
2322 verbose(env
, "BPF_END uses reserved fields\n");
2327 /* check src operand */
2328 err
= check_reg_arg(env
, insn
->dst_reg
, SRC_OP
);
2332 if (is_pointer_value(env
, insn
->dst_reg
)) {
2333 verbose(env
, "R%d pointer arithmetic prohibited\n",
2338 /* check dest operand */
2339 err
= check_reg_arg(env
, insn
->dst_reg
, DST_OP
);
2343 } else if (opcode
== BPF_MOV
) {
2345 if (BPF_SRC(insn
->code
) == BPF_X
) {
2346 if (insn
->imm
!= 0 || insn
->off
!= 0) {
2347 verbose(env
, "BPF_MOV uses reserved fields\n");
2351 /* check src operand */
2352 err
= check_reg_arg(env
, insn
->src_reg
, SRC_OP
);
2356 if (insn
->src_reg
!= BPF_REG_0
|| insn
->off
!= 0) {
2357 verbose(env
, "BPF_MOV uses reserved fields\n");
2362 /* check dest operand */
2363 err
= check_reg_arg(env
, insn
->dst_reg
, DST_OP
);
2367 if (BPF_SRC(insn
->code
) == BPF_X
) {
2368 if (BPF_CLASS(insn
->code
) == BPF_ALU64
) {
2370 * copy register state to dest reg
2372 regs
[insn
->dst_reg
] = regs
[insn
->src_reg
];
2373 regs
[insn
->dst_reg
].live
|= REG_LIVE_WRITTEN
;
2376 if (is_pointer_value(env
, insn
->src_reg
)) {
2378 "R%d partial copy of pointer\n",
2382 mark_reg_unknown(env
, regs
, insn
->dst_reg
);
2383 /* high 32 bits are known zero. */
2384 regs
[insn
->dst_reg
].var_off
= tnum_cast(
2385 regs
[insn
->dst_reg
].var_off
, 4);
2386 __update_reg_bounds(®s
[insn
->dst_reg
]);
2390 * remember the value we stored into this reg
2392 regs
[insn
->dst_reg
].type
= SCALAR_VALUE
;
2393 __mark_reg_known(regs
+ insn
->dst_reg
, insn
->imm
);
2396 } else if (opcode
> BPF_END
) {
2397 verbose(env
, "invalid BPF_ALU opcode %x\n", opcode
);
2400 } else { /* all other ALU ops: and, sub, xor, add, ... */
2402 if (BPF_SRC(insn
->code
) == BPF_X
) {
2403 if (insn
->imm
!= 0 || insn
->off
!= 0) {
2404 verbose(env
, "BPF_ALU uses reserved fields\n");
2407 /* check src1 operand */
2408 err
= check_reg_arg(env
, insn
->src_reg
, SRC_OP
);
2412 if (insn
->src_reg
!= BPF_REG_0
|| insn
->off
!= 0) {
2413 verbose(env
, "BPF_ALU uses reserved fields\n");
2418 /* check src2 operand */
2419 err
= check_reg_arg(env
, insn
->dst_reg
, SRC_OP
);
2423 if ((opcode
== BPF_MOD
|| opcode
== BPF_DIV
) &&
2424 BPF_SRC(insn
->code
) == BPF_K
&& insn
->imm
== 0) {
2425 verbose(env
, "div by zero\n");
2429 if ((opcode
== BPF_LSH
|| opcode
== BPF_RSH
||
2430 opcode
== BPF_ARSH
) && BPF_SRC(insn
->code
) == BPF_K
) {
2431 int size
= BPF_CLASS(insn
->code
) == BPF_ALU64
? 64 : 32;
2433 if (insn
->imm
< 0 || insn
->imm
>= size
) {
2434 verbose(env
, "invalid shift %d\n", insn
->imm
);
2439 /* check dest operand */
2440 err
= check_reg_arg(env
, insn
->dst_reg
, DST_OP_NO_MARK
);
2444 return adjust_reg_min_max_vals(env
, insn
);
2450 static void find_good_pkt_pointers(struct bpf_verifier_state
*state
,
2451 struct bpf_reg_state
*dst_reg
,
2452 enum bpf_reg_type type
,
2453 bool range_right_open
)
2455 struct bpf_reg_state
*regs
= state
->regs
, *reg
;
2459 if (dst_reg
->off
< 0 ||
2460 (dst_reg
->off
== 0 && range_right_open
))
2461 /* This doesn't give us any range */
2464 if (dst_reg
->umax_value
> MAX_PACKET_OFF
||
2465 dst_reg
->umax_value
+ dst_reg
->off
> MAX_PACKET_OFF
)
2466 /* Risk of overflow. For instance, ptr + (1<<63) may be less
2467 * than pkt_end, but that's because it's also less than pkt.
2471 new_range
= dst_reg
->off
;
2472 if (range_right_open
)
2475 /* Examples for register markings:
2477 * pkt_data in dst register:
2481 * if (r2 > pkt_end) goto <handle exception>
2486 * if (r2 < pkt_end) goto <access okay>
2487 * <handle exception>
2490 * r2 == dst_reg, pkt_end == src_reg
2491 * r2=pkt(id=n,off=8,r=0)
2492 * r3=pkt(id=n,off=0,r=0)
2494 * pkt_data in src register:
2498 * if (pkt_end >= r2) goto <access okay>
2499 * <handle exception>
2503 * if (pkt_end <= r2) goto <handle exception>
2507 * pkt_end == dst_reg, r2 == src_reg
2508 * r2=pkt(id=n,off=8,r=0)
2509 * r3=pkt(id=n,off=0,r=0)
2511 * Find register r3 and mark its range as r3=pkt(id=n,off=0,r=8)
2512 * or r3=pkt(id=n,off=0,r=8-1), so that range of bytes [r3, r3 + 8)
2513 * and [r3, r3 + 8-1) respectively is safe to access depending on
2517 /* If our ids match, then we must have the same max_value. And we
2518 * don't care about the other reg's fixed offset, since if it's too big
2519 * the range won't allow anything.
2520 * dst_reg->off is known < MAX_PACKET_OFF, therefore it fits in a u16.
2522 for (i
= 0; i
< MAX_BPF_REG
; i
++)
2523 if (regs
[i
].type
== type
&& regs
[i
].id
== dst_reg
->id
)
2524 /* keep the maximum range already checked */
2525 regs
[i
].range
= max(regs
[i
].range
, new_range
);
2527 for (i
= 0; i
< state
->allocated_stack
/ BPF_REG_SIZE
; i
++) {
2528 if (state
->stack
[i
].slot_type
[0] != STACK_SPILL
)
2530 reg
= &state
->stack
[i
].spilled_ptr
;
2531 if (reg
->type
== type
&& reg
->id
== dst_reg
->id
)
2532 reg
->range
= max_t(u16
, reg
->range
, new_range
);
2536 /* Adjusts the register min/max values in the case that the dst_reg is the
2537 * variable register that we are working on, and src_reg is a constant or we're
2538 * simply doing a BPF_K check.
2539 * In JEQ/JNE cases we also adjust the var_off values.
2541 static void reg_set_min_max(struct bpf_reg_state
*true_reg
,
2542 struct bpf_reg_state
*false_reg
, u64 val
,
2545 /* If the dst_reg is a pointer, we can't learn anything about its
2546 * variable offset from the compare (unless src_reg were a pointer into
2547 * the same object, but we don't bother with that.
2548 * Since false_reg and true_reg have the same type by construction, we
2549 * only need to check one of them for pointerness.
2551 if (__is_pointer_value(false, false_reg
))
2556 /* If this is false then we know nothing Jon Snow, but if it is
2557 * true then we know for sure.
2559 __mark_reg_known(true_reg
, val
);
2562 /* If this is true we know nothing Jon Snow, but if it is false
2563 * we know the value for sure;
2565 __mark_reg_known(false_reg
, val
);
2568 false_reg
->umax_value
= min(false_reg
->umax_value
, val
);
2569 true_reg
->umin_value
= max(true_reg
->umin_value
, val
+ 1);
2572 false_reg
->smax_value
= min_t(s64
, false_reg
->smax_value
, val
);
2573 true_reg
->smin_value
= max_t(s64
, true_reg
->smin_value
, val
+ 1);
2576 false_reg
->umin_value
= max(false_reg
->umin_value
, val
);
2577 true_reg
->umax_value
= min(true_reg
->umax_value
, val
- 1);
2580 false_reg
->smin_value
= max_t(s64
, false_reg
->smin_value
, val
);
2581 true_reg
->smax_value
= min_t(s64
, true_reg
->smax_value
, val
- 1);
2584 false_reg
->umax_value
= min(false_reg
->umax_value
, val
- 1);
2585 true_reg
->umin_value
= max(true_reg
->umin_value
, val
);
2588 false_reg
->smax_value
= min_t(s64
, false_reg
->smax_value
, val
- 1);
2589 true_reg
->smin_value
= max_t(s64
, true_reg
->smin_value
, val
);
2592 false_reg
->umin_value
= max(false_reg
->umin_value
, val
+ 1);
2593 true_reg
->umax_value
= min(true_reg
->umax_value
, val
);
2596 false_reg
->smin_value
= max_t(s64
, false_reg
->smin_value
, val
+ 1);
2597 true_reg
->smax_value
= min_t(s64
, true_reg
->smax_value
, val
);
2603 __reg_deduce_bounds(false_reg
);
2604 __reg_deduce_bounds(true_reg
);
2605 /* We might have learned some bits from the bounds. */
2606 __reg_bound_offset(false_reg
);
2607 __reg_bound_offset(true_reg
);
2608 /* Intersecting with the old var_off might have improved our bounds
2609 * slightly. e.g. if umax was 0x7f...f and var_off was (0; 0xf...fc),
2610 * then new var_off is (0; 0x7f...fc) which improves our umax.
2612 __update_reg_bounds(false_reg
);
2613 __update_reg_bounds(true_reg
);
2616 /* Same as above, but for the case that dst_reg holds a constant and src_reg is
2619 static void reg_set_min_max_inv(struct bpf_reg_state
*true_reg
,
2620 struct bpf_reg_state
*false_reg
, u64 val
,
2623 if (__is_pointer_value(false, false_reg
))
2628 /* If this is false then we know nothing Jon Snow, but if it is
2629 * true then we know for sure.
2631 __mark_reg_known(true_reg
, val
);
2634 /* If this is true we know nothing Jon Snow, but if it is false
2635 * we know the value for sure;
2637 __mark_reg_known(false_reg
, val
);
2640 true_reg
->umax_value
= min(true_reg
->umax_value
, val
- 1);
2641 false_reg
->umin_value
= max(false_reg
->umin_value
, val
);
2644 true_reg
->smax_value
= min_t(s64
, true_reg
->smax_value
, val
- 1);
2645 false_reg
->smin_value
= max_t(s64
, false_reg
->smin_value
, val
);
2648 true_reg
->umin_value
= max(true_reg
->umin_value
, val
+ 1);
2649 false_reg
->umax_value
= min(false_reg
->umax_value
, val
);
2652 true_reg
->smin_value
= max_t(s64
, true_reg
->smin_value
, val
+ 1);
2653 false_reg
->smax_value
= min_t(s64
, false_reg
->smax_value
, val
);
2656 true_reg
->umax_value
= min(true_reg
->umax_value
, val
);
2657 false_reg
->umin_value
= max(false_reg
->umin_value
, val
+ 1);
2660 true_reg
->smax_value
= min_t(s64
, true_reg
->smax_value
, val
);
2661 false_reg
->smin_value
= max_t(s64
, false_reg
->smin_value
, val
+ 1);
2664 true_reg
->umin_value
= max(true_reg
->umin_value
, val
);
2665 false_reg
->umax_value
= min(false_reg
->umax_value
, val
- 1);
2668 true_reg
->smin_value
= max_t(s64
, true_reg
->smin_value
, val
);
2669 false_reg
->smax_value
= min_t(s64
, false_reg
->smax_value
, val
- 1);
2675 __reg_deduce_bounds(false_reg
);
2676 __reg_deduce_bounds(true_reg
);
2677 /* We might have learned some bits from the bounds. */
2678 __reg_bound_offset(false_reg
);
2679 __reg_bound_offset(true_reg
);
2680 /* Intersecting with the old var_off might have improved our bounds
2681 * slightly. e.g. if umax was 0x7f...f and var_off was (0; 0xf...fc),
2682 * then new var_off is (0; 0x7f...fc) which improves our umax.
2684 __update_reg_bounds(false_reg
);
2685 __update_reg_bounds(true_reg
);
2688 /* Regs are known to be equal, so intersect their min/max/var_off */
2689 static void __reg_combine_min_max(struct bpf_reg_state
*src_reg
,
2690 struct bpf_reg_state
*dst_reg
)
2692 src_reg
->umin_value
= dst_reg
->umin_value
= max(src_reg
->umin_value
,
2693 dst_reg
->umin_value
);
2694 src_reg
->umax_value
= dst_reg
->umax_value
= min(src_reg
->umax_value
,
2695 dst_reg
->umax_value
);
2696 src_reg
->smin_value
= dst_reg
->smin_value
= max(src_reg
->smin_value
,
2697 dst_reg
->smin_value
);
2698 src_reg
->smax_value
= dst_reg
->smax_value
= min(src_reg
->smax_value
,
2699 dst_reg
->smax_value
);
2700 src_reg
->var_off
= dst_reg
->var_off
= tnum_intersect(src_reg
->var_off
,
2702 /* We might have learned new bounds from the var_off. */
2703 __update_reg_bounds(src_reg
);
2704 __update_reg_bounds(dst_reg
);
2705 /* We might have learned something about the sign bit. */
2706 __reg_deduce_bounds(src_reg
);
2707 __reg_deduce_bounds(dst_reg
);
2708 /* We might have learned some bits from the bounds. */
2709 __reg_bound_offset(src_reg
);
2710 __reg_bound_offset(dst_reg
);
2711 /* Intersecting with the old var_off might have improved our bounds
2712 * slightly. e.g. if umax was 0x7f...f and var_off was (0; 0xf...fc),
2713 * then new var_off is (0; 0x7f...fc) which improves our umax.
2715 __update_reg_bounds(src_reg
);
2716 __update_reg_bounds(dst_reg
);
2719 static void reg_combine_min_max(struct bpf_reg_state
*true_src
,
2720 struct bpf_reg_state
*true_dst
,
2721 struct bpf_reg_state
*false_src
,
2722 struct bpf_reg_state
*false_dst
,
2727 __reg_combine_min_max(true_src
, true_dst
);
2730 __reg_combine_min_max(false_src
, false_dst
);
2735 static void mark_map_reg(struct bpf_reg_state
*regs
, u32 regno
, u32 id
,
2738 struct bpf_reg_state
*reg
= ®s
[regno
];
2740 if (reg
->type
== PTR_TO_MAP_VALUE_OR_NULL
&& reg
->id
== id
) {
2741 /* Old offset (both fixed and variable parts) should
2742 * have been known-zero, because we don't allow pointer
2743 * arithmetic on pointers that might be NULL.
2745 if (WARN_ON_ONCE(reg
->smin_value
|| reg
->smax_value
||
2746 !tnum_equals_const(reg
->var_off
, 0) ||
2748 __mark_reg_known_zero(reg
);
2752 reg
->type
= SCALAR_VALUE
;
2753 } else if (reg
->map_ptr
->inner_map_meta
) {
2754 reg
->type
= CONST_PTR_TO_MAP
;
2755 reg
->map_ptr
= reg
->map_ptr
->inner_map_meta
;
2757 reg
->type
= PTR_TO_MAP_VALUE
;
2759 /* We don't need id from this point onwards anymore, thus we
2760 * should better reset it, so that state pruning has chances
2767 /* The logic is similar to find_good_pkt_pointers(), both could eventually
2768 * be folded together at some point.
2770 static void mark_map_regs(struct bpf_verifier_state
*state
, u32 regno
,
2773 struct bpf_reg_state
*regs
= state
->regs
;
2774 u32 id
= regs
[regno
].id
;
2777 for (i
= 0; i
< MAX_BPF_REG
; i
++)
2778 mark_map_reg(regs
, i
, id
, is_null
);
2780 for (i
= 0; i
< state
->allocated_stack
/ BPF_REG_SIZE
; i
++) {
2781 if (state
->stack
[i
].slot_type
[0] != STACK_SPILL
)
2783 mark_map_reg(&state
->stack
[i
].spilled_ptr
, 0, id
, is_null
);
2787 static int check_cond_jmp_op(struct bpf_verifier_env
*env
,
2788 struct bpf_insn
*insn
, int *insn_idx
)
2790 struct bpf_verifier_state
*other_branch
, *this_branch
= env
->cur_state
;
2791 struct bpf_reg_state
*regs
= this_branch
->regs
, *dst_reg
;
2792 u8 opcode
= BPF_OP(insn
->code
);
2795 if (opcode
> BPF_JSLE
) {
2796 verbose(env
, "invalid BPF_JMP opcode %x\n", opcode
);
2800 if (BPF_SRC(insn
->code
) == BPF_X
) {
2801 if (insn
->imm
!= 0) {
2802 verbose(env
, "BPF_JMP uses reserved fields\n");
2806 /* check src1 operand */
2807 err
= check_reg_arg(env
, insn
->src_reg
, SRC_OP
);
2811 if (is_pointer_value(env
, insn
->src_reg
)) {
2812 verbose(env
, "R%d pointer comparison prohibited\n",
2817 if (insn
->src_reg
!= BPF_REG_0
) {
2818 verbose(env
, "BPF_JMP uses reserved fields\n");
2823 /* check src2 operand */
2824 err
= check_reg_arg(env
, insn
->dst_reg
, SRC_OP
);
2828 dst_reg
= ®s
[insn
->dst_reg
];
2830 /* detect if R == 0 where R was initialized to zero earlier */
2831 if (BPF_SRC(insn
->code
) == BPF_K
&&
2832 (opcode
== BPF_JEQ
|| opcode
== BPF_JNE
) &&
2833 dst_reg
->type
== SCALAR_VALUE
&&
2834 tnum_equals_const(dst_reg
->var_off
, insn
->imm
)) {
2835 if (opcode
== BPF_JEQ
) {
2836 /* if (imm == imm) goto pc+off;
2837 * only follow the goto, ignore fall-through
2839 *insn_idx
+= insn
->off
;
2842 /* if (imm != imm) goto pc+off;
2843 * only follow fall-through branch, since
2844 * that's where the program will go
2850 other_branch
= push_stack(env
, *insn_idx
+ insn
->off
+ 1, *insn_idx
);
2854 /* detect if we are comparing against a constant value so we can adjust
2855 * our min/max values for our dst register.
2856 * this is only legit if both are scalars (or pointers to the same
2857 * object, I suppose, but we don't support that right now), because
2858 * otherwise the different base pointers mean the offsets aren't
2861 if (BPF_SRC(insn
->code
) == BPF_X
) {
2862 if (dst_reg
->type
== SCALAR_VALUE
&&
2863 regs
[insn
->src_reg
].type
== SCALAR_VALUE
) {
2864 if (tnum_is_const(regs
[insn
->src_reg
].var_off
))
2865 reg_set_min_max(&other_branch
->regs
[insn
->dst_reg
],
2866 dst_reg
, regs
[insn
->src_reg
].var_off
.value
,
2868 else if (tnum_is_const(dst_reg
->var_off
))
2869 reg_set_min_max_inv(&other_branch
->regs
[insn
->src_reg
],
2870 ®s
[insn
->src_reg
],
2871 dst_reg
->var_off
.value
, opcode
);
2872 else if (opcode
== BPF_JEQ
|| opcode
== BPF_JNE
)
2873 /* Comparing for equality, we can combine knowledge */
2874 reg_combine_min_max(&other_branch
->regs
[insn
->src_reg
],
2875 &other_branch
->regs
[insn
->dst_reg
],
2876 ®s
[insn
->src_reg
],
2877 ®s
[insn
->dst_reg
], opcode
);
2879 } else if (dst_reg
->type
== SCALAR_VALUE
) {
2880 reg_set_min_max(&other_branch
->regs
[insn
->dst_reg
],
2881 dst_reg
, insn
->imm
, opcode
);
2884 /* detect if R == 0 where R is returned from bpf_map_lookup_elem() */
2885 if (BPF_SRC(insn
->code
) == BPF_K
&&
2886 insn
->imm
== 0 && (opcode
== BPF_JEQ
|| opcode
== BPF_JNE
) &&
2887 dst_reg
->type
== PTR_TO_MAP_VALUE_OR_NULL
) {
2888 /* Mark all identical map registers in each branch as either
2889 * safe or unknown depending R == 0 or R != 0 conditional.
2891 mark_map_regs(this_branch
, insn
->dst_reg
, opcode
== BPF_JNE
);
2892 mark_map_regs(other_branch
, insn
->dst_reg
, opcode
== BPF_JEQ
);
2893 } else if (BPF_SRC(insn
->code
) == BPF_X
&& opcode
== BPF_JGT
&&
2894 dst_reg
->type
== PTR_TO_PACKET
&&
2895 regs
[insn
->src_reg
].type
== PTR_TO_PACKET_END
) {
2896 /* pkt_data' > pkt_end */
2897 find_good_pkt_pointers(this_branch
, dst_reg
,
2898 PTR_TO_PACKET
, false);
2899 } else if (BPF_SRC(insn
->code
) == BPF_X
&& opcode
== BPF_JGT
&&
2900 dst_reg
->type
== PTR_TO_PACKET_END
&&
2901 regs
[insn
->src_reg
].type
== PTR_TO_PACKET
) {
2902 /* pkt_end > pkt_data' */
2903 find_good_pkt_pointers(other_branch
, ®s
[insn
->src_reg
],
2904 PTR_TO_PACKET
, true);
2905 } else if (BPF_SRC(insn
->code
) == BPF_X
&& opcode
== BPF_JLT
&&
2906 dst_reg
->type
== PTR_TO_PACKET
&&
2907 regs
[insn
->src_reg
].type
== PTR_TO_PACKET_END
) {
2908 /* pkt_data' < pkt_end */
2909 find_good_pkt_pointers(other_branch
, dst_reg
, PTR_TO_PACKET
,
2911 } else if (BPF_SRC(insn
->code
) == BPF_X
&& opcode
== BPF_JLT
&&
2912 dst_reg
->type
== PTR_TO_PACKET_END
&&
2913 regs
[insn
->src_reg
].type
== PTR_TO_PACKET
) {
2914 /* pkt_end < pkt_data' */
2915 find_good_pkt_pointers(this_branch
, ®s
[insn
->src_reg
],
2916 PTR_TO_PACKET
, false);
2917 } else if (BPF_SRC(insn
->code
) == BPF_X
&& opcode
== BPF_JGE
&&
2918 dst_reg
->type
== PTR_TO_PACKET
&&
2919 regs
[insn
->src_reg
].type
== PTR_TO_PACKET_END
) {
2920 /* pkt_data' >= pkt_end */
2921 find_good_pkt_pointers(this_branch
, dst_reg
,
2922 PTR_TO_PACKET
, true);
2923 } else if (BPF_SRC(insn
->code
) == BPF_X
&& opcode
== BPF_JGE
&&
2924 dst_reg
->type
== PTR_TO_PACKET_END
&&
2925 regs
[insn
->src_reg
].type
== PTR_TO_PACKET
) {
2926 /* pkt_end >= pkt_data' */
2927 find_good_pkt_pointers(other_branch
, ®s
[insn
->src_reg
],
2928 PTR_TO_PACKET
, false);
2929 } else if (BPF_SRC(insn
->code
) == BPF_X
&& opcode
== BPF_JLE
&&
2930 dst_reg
->type
== PTR_TO_PACKET
&&
2931 regs
[insn
->src_reg
].type
== PTR_TO_PACKET_END
) {
2932 /* pkt_data' <= pkt_end */
2933 find_good_pkt_pointers(other_branch
, dst_reg
,
2934 PTR_TO_PACKET
, false);
2935 } else if (BPF_SRC(insn
->code
) == BPF_X
&& opcode
== BPF_JLE
&&
2936 dst_reg
->type
== PTR_TO_PACKET_END
&&
2937 regs
[insn
->src_reg
].type
== PTR_TO_PACKET
) {
2938 /* pkt_end <= pkt_data' */
2939 find_good_pkt_pointers(this_branch
, ®s
[insn
->src_reg
],
2940 PTR_TO_PACKET
, true);
2941 } else if (BPF_SRC(insn
->code
) == BPF_X
&& opcode
== BPF_JGT
&&
2942 dst_reg
->type
== PTR_TO_PACKET_META
&&
2943 reg_is_init_pkt_pointer(®s
[insn
->src_reg
], PTR_TO_PACKET
)) {
2944 find_good_pkt_pointers(this_branch
, dst_reg
,
2945 PTR_TO_PACKET_META
, false);
2946 } else if (BPF_SRC(insn
->code
) == BPF_X
&& opcode
== BPF_JLT
&&
2947 dst_reg
->type
== PTR_TO_PACKET_META
&&
2948 reg_is_init_pkt_pointer(®s
[insn
->src_reg
], PTR_TO_PACKET
)) {
2949 find_good_pkt_pointers(other_branch
, dst_reg
,
2950 PTR_TO_PACKET_META
, false);
2951 } else if (BPF_SRC(insn
->code
) == BPF_X
&& opcode
== BPF_JGE
&&
2952 reg_is_init_pkt_pointer(dst_reg
, PTR_TO_PACKET
) &&
2953 regs
[insn
->src_reg
].type
== PTR_TO_PACKET_META
) {
2954 find_good_pkt_pointers(other_branch
, ®s
[insn
->src_reg
],
2955 PTR_TO_PACKET_META
, false);
2956 } else if (BPF_SRC(insn
->code
) == BPF_X
&& opcode
== BPF_JLE
&&
2957 reg_is_init_pkt_pointer(dst_reg
, PTR_TO_PACKET
) &&
2958 regs
[insn
->src_reg
].type
== PTR_TO_PACKET_META
) {
2959 find_good_pkt_pointers(this_branch
, ®s
[insn
->src_reg
],
2960 PTR_TO_PACKET_META
, false);
2961 } else if (is_pointer_value(env
, insn
->dst_reg
)) {
2962 verbose(env
, "R%d pointer comparison prohibited\n",
2967 print_verifier_state(env
, this_branch
);
2971 /* return the map pointer stored inside BPF_LD_IMM64 instruction */
2972 static struct bpf_map
*ld_imm64_to_map_ptr(struct bpf_insn
*insn
)
2974 u64 imm64
= ((u64
) (u32
) insn
[0].imm
) | ((u64
) (u32
) insn
[1].imm
) << 32;
2976 return (struct bpf_map
*) (unsigned long) imm64
;
2979 /* verify BPF_LD_IMM64 instruction */
2980 static int check_ld_imm(struct bpf_verifier_env
*env
, struct bpf_insn
*insn
)
2982 struct bpf_reg_state
*regs
= cur_regs(env
);
2985 if (BPF_SIZE(insn
->code
) != BPF_DW
) {
2986 verbose(env
, "invalid BPF_LD_IMM insn\n");
2989 if (insn
->off
!= 0) {
2990 verbose(env
, "BPF_LD_IMM64 uses reserved fields\n");
2994 err
= check_reg_arg(env
, insn
->dst_reg
, DST_OP
);
2998 if (insn
->src_reg
== 0) {
2999 u64 imm
= ((u64
)(insn
+ 1)->imm
<< 32) | (u32
)insn
->imm
;
3001 regs
[insn
->dst_reg
].type
= SCALAR_VALUE
;
3002 __mark_reg_known(®s
[insn
->dst_reg
], imm
);
3006 /* replace_map_fd_with_map_ptr() should have caught bad ld_imm64 */
3007 BUG_ON(insn
->src_reg
!= BPF_PSEUDO_MAP_FD
);
3009 regs
[insn
->dst_reg
].type
= CONST_PTR_TO_MAP
;
3010 regs
[insn
->dst_reg
].map_ptr
= ld_imm64_to_map_ptr(insn
);
3014 static bool may_access_skb(enum bpf_prog_type type
)
3017 case BPF_PROG_TYPE_SOCKET_FILTER
:
3018 case BPF_PROG_TYPE_SCHED_CLS
:
3019 case BPF_PROG_TYPE_SCHED_ACT
:
3026 /* verify safety of LD_ABS|LD_IND instructions:
3027 * - they can only appear in the programs where ctx == skb
3028 * - since they are wrappers of function calls, they scratch R1-R5 registers,
3029 * preserve R6-R9, and store return value into R0
3032 * ctx == skb == R6 == CTX
3035 * SRC == any register
3036 * IMM == 32-bit immediate
3039 * R0 - 8/16/32-bit skb data converted to cpu endianness
3041 static int check_ld_abs(struct bpf_verifier_env
*env
, struct bpf_insn
*insn
)
3043 struct bpf_reg_state
*regs
= cur_regs(env
);
3044 u8 mode
= BPF_MODE(insn
->code
);
3047 if (!may_access_skb(env
->prog
->type
)) {
3048 verbose(env
, "BPF_LD_[ABS|IND] instructions not allowed for this program type\n");
3052 if (insn
->dst_reg
!= BPF_REG_0
|| insn
->off
!= 0 ||
3053 BPF_SIZE(insn
->code
) == BPF_DW
||
3054 (mode
== BPF_ABS
&& insn
->src_reg
!= BPF_REG_0
)) {
3055 verbose(env
, "BPF_LD_[ABS|IND] uses reserved fields\n");
3059 /* check whether implicit source operand (register R6) is readable */
3060 err
= check_reg_arg(env
, BPF_REG_6
, SRC_OP
);
3064 if (regs
[BPF_REG_6
].type
!= PTR_TO_CTX
) {
3066 "at the time of BPF_LD_ABS|IND R6 != pointer to skb\n");
3070 if (mode
== BPF_IND
) {
3071 /* check explicit source operand */
3072 err
= check_reg_arg(env
, insn
->src_reg
, SRC_OP
);
3077 /* reset caller saved regs to unreadable */
3078 for (i
= 0; i
< CALLER_SAVED_REGS
; i
++) {
3079 mark_reg_not_init(env
, regs
, caller_saved
[i
]);
3080 check_reg_arg(env
, caller_saved
[i
], DST_OP_NO_MARK
);
3083 /* mark destination R0 register as readable, since it contains
3084 * the value fetched from the packet.
3085 * Already marked as written above.
3087 mark_reg_unknown(env
, regs
, BPF_REG_0
);
3091 static int check_return_code(struct bpf_verifier_env
*env
)
3093 struct bpf_reg_state
*reg
;
3094 struct tnum range
= tnum_range(0, 1);
3096 switch (env
->prog
->type
) {
3097 case BPF_PROG_TYPE_CGROUP_SKB
:
3098 case BPF_PROG_TYPE_CGROUP_SOCK
:
3099 case BPF_PROG_TYPE_SOCK_OPS
:
3105 reg
= cur_regs(env
) + BPF_REG_0
;
3106 if (reg
->type
!= SCALAR_VALUE
) {
3107 verbose(env
, "At program exit the register R0 is not a known value (%s)\n",
3108 reg_type_str
[reg
->type
]);
3112 if (!tnum_in(range
, reg
->var_off
)) {
3113 verbose(env
, "At program exit the register R0 ");
3114 if (!tnum_is_unknown(reg
->var_off
)) {
3117 tnum_strn(tn_buf
, sizeof(tn_buf
), reg
->var_off
);
3118 verbose(env
, "has value %s", tn_buf
);
3120 verbose(env
, "has unknown scalar value");
3122 verbose(env
, " should have been 0 or 1\n");
3128 /* non-recursive DFS pseudo code
3129 * 1 procedure DFS-iterative(G,v):
3130 * 2 label v as discovered
3131 * 3 let S be a stack
3133 * 5 while S is not empty
3135 * 7 if t is what we're looking for:
3137 * 9 for all edges e in G.adjacentEdges(t) do
3138 * 10 if edge e is already labelled
3139 * 11 continue with the next edge
3140 * 12 w <- G.adjacentVertex(t,e)
3141 * 13 if vertex w is not discovered and not explored
3142 * 14 label e as tree-edge
3143 * 15 label w as discovered
3146 * 18 else if vertex w is discovered
3147 * 19 label e as back-edge
3149 * 21 // vertex w is explored
3150 * 22 label e as forward- or cross-edge
3151 * 23 label t as explored
3156 * 0x11 - discovered and fall-through edge labelled
3157 * 0x12 - discovered and fall-through and branch edges labelled
3168 #define STATE_LIST_MARK ((struct bpf_verifier_state_list *) -1L)
3170 static int *insn_stack
; /* stack of insns to process */
3171 static int cur_stack
; /* current stack index */
3172 static int *insn_state
;
3174 /* t, w, e - match pseudo-code above:
3175 * t - index of current instruction
3176 * w - next instruction
3179 static int push_insn(int t
, int w
, int e
, struct bpf_verifier_env
*env
)
3181 if (e
== FALLTHROUGH
&& insn_state
[t
] >= (DISCOVERED
| FALLTHROUGH
))
3184 if (e
== BRANCH
&& insn_state
[t
] >= (DISCOVERED
| BRANCH
))
3187 if (w
< 0 || w
>= env
->prog
->len
) {
3188 verbose(env
, "jump out of range from insn %d to %d\n", t
, w
);
3193 /* mark branch target for state pruning */
3194 env
->explored_states
[w
] = STATE_LIST_MARK
;
3196 if (insn_state
[w
] == 0) {
3198 insn_state
[t
] = DISCOVERED
| e
;
3199 insn_state
[w
] = DISCOVERED
;
3200 if (cur_stack
>= env
->prog
->len
)
3202 insn_stack
[cur_stack
++] = w
;
3204 } else if ((insn_state
[w
] & 0xF0) == DISCOVERED
) {
3205 verbose(env
, "back-edge from insn %d to %d\n", t
, w
);
3207 } else if (insn_state
[w
] == EXPLORED
) {
3208 /* forward- or cross-edge */
3209 insn_state
[t
] = DISCOVERED
| e
;
3211 verbose(env
, "insn state internal bug\n");
3217 /* non-recursive depth-first-search to detect loops in BPF program
3218 * loop == back-edge in directed graph
3220 static int check_cfg(struct bpf_verifier_env
*env
)
3222 struct bpf_insn
*insns
= env
->prog
->insnsi
;
3223 int insn_cnt
= env
->prog
->len
;
3227 insn_state
= kcalloc(insn_cnt
, sizeof(int), GFP_KERNEL
);
3231 insn_stack
= kcalloc(insn_cnt
, sizeof(int), GFP_KERNEL
);
3237 insn_state
[0] = DISCOVERED
; /* mark 1st insn as discovered */
3238 insn_stack
[0] = 0; /* 0 is the first instruction */
3244 t
= insn_stack
[cur_stack
- 1];
3246 if (BPF_CLASS(insns
[t
].code
) == BPF_JMP
) {
3247 u8 opcode
= BPF_OP(insns
[t
].code
);
3249 if (opcode
== BPF_EXIT
) {
3251 } else if (opcode
== BPF_CALL
) {
3252 ret
= push_insn(t
, t
+ 1, FALLTHROUGH
, env
);
3257 if (t
+ 1 < insn_cnt
)
3258 env
->explored_states
[t
+ 1] = STATE_LIST_MARK
;
3259 } else if (opcode
== BPF_JA
) {
3260 if (BPF_SRC(insns
[t
].code
) != BPF_K
) {
3264 /* unconditional jump with single edge */
3265 ret
= push_insn(t
, t
+ insns
[t
].off
+ 1,
3271 /* tell verifier to check for equivalent states
3272 * after every call and jump
3274 if (t
+ 1 < insn_cnt
)
3275 env
->explored_states
[t
+ 1] = STATE_LIST_MARK
;
3277 /* conditional jump with two edges */
3278 env
->explored_states
[t
] = STATE_LIST_MARK
;
3279 ret
= push_insn(t
, t
+ 1, FALLTHROUGH
, env
);
3285 ret
= push_insn(t
, t
+ insns
[t
].off
+ 1, BRANCH
, env
);
3292 /* all other non-branch instructions with single
3295 ret
= push_insn(t
, t
+ 1, FALLTHROUGH
, env
);
3303 insn_state
[t
] = EXPLORED
;
3304 if (cur_stack
-- <= 0) {
3305 verbose(env
, "pop stack internal bug\n");
3312 for (i
= 0; i
< insn_cnt
; i
++) {
3313 if (insn_state
[i
] != EXPLORED
) {
3314 verbose(env
, "unreachable insn %d\n", i
);
3319 ret
= 0; /* cfg looks good */
3327 /* check %cur's range satisfies %old's */
3328 static bool range_within(struct bpf_reg_state
*old
,
3329 struct bpf_reg_state
*cur
)
3331 return old
->umin_value
<= cur
->umin_value
&&
3332 old
->umax_value
>= cur
->umax_value
&&
3333 old
->smin_value
<= cur
->smin_value
&&
3334 old
->smax_value
>= cur
->smax_value
;
3337 /* Maximum number of register states that can exist at once */
3338 #define ID_MAP_SIZE (MAX_BPF_REG + MAX_BPF_STACK / BPF_REG_SIZE)
3344 /* If in the old state two registers had the same id, then they need to have
3345 * the same id in the new state as well. But that id could be different from
3346 * the old state, so we need to track the mapping from old to new ids.
3347 * Once we have seen that, say, a reg with old id 5 had new id 9, any subsequent
3348 * regs with old id 5 must also have new id 9 for the new state to be safe. But
3349 * regs with a different old id could still have new id 9, we don't care about
3351 * So we look through our idmap to see if this old id has been seen before. If
3352 * so, we require the new id to match; otherwise, we add the id pair to the map.
3354 static bool check_ids(u32 old_id
, u32 cur_id
, struct idpair
*idmap
)
3358 for (i
= 0; i
< ID_MAP_SIZE
; i
++) {
3359 if (!idmap
[i
].old
) {
3360 /* Reached an empty slot; haven't seen this id before */
3361 idmap
[i
].old
= old_id
;
3362 idmap
[i
].cur
= cur_id
;
3365 if (idmap
[i
].old
== old_id
)
3366 return idmap
[i
].cur
== cur_id
;
3368 /* We ran out of idmap slots, which should be impossible */
3373 /* Returns true if (rold safe implies rcur safe) */
3374 static bool regsafe(struct bpf_reg_state
*rold
, struct bpf_reg_state
*rcur
,
3375 struct idpair
*idmap
)
3377 if (!(rold
->live
& REG_LIVE_READ
))
3378 /* explored state didn't use this */
3381 if (memcmp(rold
, rcur
, offsetof(struct bpf_reg_state
, live
)) == 0)
3384 if (rold
->type
== NOT_INIT
)
3385 /* explored state can't have used this */
3387 if (rcur
->type
== NOT_INIT
)
3389 switch (rold
->type
) {
3391 if (rcur
->type
== SCALAR_VALUE
) {
3392 /* new val must satisfy old val knowledge */
3393 return range_within(rold
, rcur
) &&
3394 tnum_in(rold
->var_off
, rcur
->var_off
);
3396 /* if we knew anything about the old value, we're not
3397 * equal, because we can't know anything about the
3398 * scalar value of the pointer in the new value.
3400 return rold
->umin_value
== 0 &&
3401 rold
->umax_value
== U64_MAX
&&
3402 rold
->smin_value
== S64_MIN
&&
3403 rold
->smax_value
== S64_MAX
&&
3404 tnum_is_unknown(rold
->var_off
);
3406 case PTR_TO_MAP_VALUE
:
3407 /* If the new min/max/var_off satisfy the old ones and
3408 * everything else matches, we are OK.
3409 * We don't care about the 'id' value, because nothing
3410 * uses it for PTR_TO_MAP_VALUE (only for ..._OR_NULL)
3412 return memcmp(rold
, rcur
, offsetof(struct bpf_reg_state
, id
)) == 0 &&
3413 range_within(rold
, rcur
) &&
3414 tnum_in(rold
->var_off
, rcur
->var_off
);
3415 case PTR_TO_MAP_VALUE_OR_NULL
:
3416 /* a PTR_TO_MAP_VALUE could be safe to use as a
3417 * PTR_TO_MAP_VALUE_OR_NULL into the same map.
3418 * However, if the old PTR_TO_MAP_VALUE_OR_NULL then got NULL-
3419 * checked, doing so could have affected others with the same
3420 * id, and we can't check for that because we lost the id when
3421 * we converted to a PTR_TO_MAP_VALUE.
3423 if (rcur
->type
!= PTR_TO_MAP_VALUE_OR_NULL
)
3425 if (memcmp(rold
, rcur
, offsetof(struct bpf_reg_state
, id
)))
3427 /* Check our ids match any regs they're supposed to */
3428 return check_ids(rold
->id
, rcur
->id
, idmap
);
3429 case PTR_TO_PACKET_META
:
3431 if (rcur
->type
!= rold
->type
)
3433 /* We must have at least as much range as the old ptr
3434 * did, so that any accesses which were safe before are
3435 * still safe. This is true even if old range < old off,
3436 * since someone could have accessed through (ptr - k), or
3437 * even done ptr -= k in a register, to get a safe access.
3439 if (rold
->range
> rcur
->range
)
3441 /* If the offsets don't match, we can't trust our alignment;
3442 * nor can we be sure that we won't fall out of range.
3444 if (rold
->off
!= rcur
->off
)
3446 /* id relations must be preserved */
3447 if (rold
->id
&& !check_ids(rold
->id
, rcur
->id
, idmap
))
3449 /* new val must satisfy old val knowledge */
3450 return range_within(rold
, rcur
) &&
3451 tnum_in(rold
->var_off
, rcur
->var_off
);
3453 case CONST_PTR_TO_MAP
:
3455 case PTR_TO_PACKET_END
:
3456 /* Only valid matches are exact, which memcmp() above
3457 * would have accepted
3460 /* Don't know what's going on, just say it's not safe */
3464 /* Shouldn't get here; if we do, say it's not safe */
3469 static bool stacksafe(struct bpf_verifier_state
*old
,
3470 struct bpf_verifier_state
*cur
,
3471 struct idpair
*idmap
)
3475 /* if explored stack has more populated slots than current stack
3476 * such stacks are not equivalent
3478 if (old
->allocated_stack
> cur
->allocated_stack
)
3481 /* walk slots of the explored stack and ignore any additional
3482 * slots in the current stack, since explored(safe) state
3485 for (i
= 0; i
< old
->allocated_stack
; i
++) {
3486 spi
= i
/ BPF_REG_SIZE
;
3488 if (old
->stack
[spi
].slot_type
[i
% BPF_REG_SIZE
] == STACK_INVALID
)
3490 if (old
->stack
[spi
].slot_type
[i
% BPF_REG_SIZE
] !=
3491 cur
->stack
[spi
].slot_type
[i
% BPF_REG_SIZE
])
3492 /* Ex: old explored (safe) state has STACK_SPILL in
3493 * this stack slot, but current has has STACK_MISC ->
3494 * this verifier states are not equivalent,
3495 * return false to continue verification of this path
3498 if (i
% BPF_REG_SIZE
)
3500 if (old
->stack
[spi
].slot_type
[0] != STACK_SPILL
)
3502 if (!regsafe(&old
->stack
[spi
].spilled_ptr
,
3503 &cur
->stack
[spi
].spilled_ptr
,
3505 /* when explored and current stack slot are both storing
3506 * spilled registers, check that stored pointers types
3507 * are the same as well.
3508 * Ex: explored safe path could have stored
3509 * (bpf_reg_state) {.type = PTR_TO_STACK, .off = -8}
3510 * but current path has stored:
3511 * (bpf_reg_state) {.type = PTR_TO_STACK, .off = -16}
3512 * such verifier states are not equivalent.
3513 * return false to continue verification of this path
3520 /* compare two verifier states
3522 * all states stored in state_list are known to be valid, since
3523 * verifier reached 'bpf_exit' instruction through them
3525 * this function is called when verifier exploring different branches of
3526 * execution popped from the state stack. If it sees an old state that has
3527 * more strict register state and more strict stack state then this execution
3528 * branch doesn't need to be explored further, since verifier already
3529 * concluded that more strict state leads to valid finish.
3531 * Therefore two states are equivalent if register state is more conservative
3532 * and explored stack state is more conservative than the current one.
3535 * (slot1=INV slot2=MISC) == (slot1=MISC slot2=MISC)
3536 * (slot1=MISC slot2=MISC) != (slot1=INV slot2=MISC)
3538 * In other words if current stack state (one being explored) has more
3539 * valid slots than old one that already passed validation, it means
3540 * the verifier can stop exploring and conclude that current state is valid too
3542 * Similarly with registers. If explored state has register type as invalid
3543 * whereas register type in current state is meaningful, it means that
3544 * the current state will reach 'bpf_exit' instruction safely
3546 static bool states_equal(struct bpf_verifier_env
*env
,
3547 struct bpf_verifier_state
*old
,
3548 struct bpf_verifier_state
*cur
)
3550 struct idpair
*idmap
;
3554 idmap
= kcalloc(ID_MAP_SIZE
, sizeof(struct idpair
), GFP_KERNEL
);
3555 /* If we failed to allocate the idmap, just say it's not safe */
3559 for (i
= 0; i
< MAX_BPF_REG
; i
++) {
3560 if (!regsafe(&old
->regs
[i
], &cur
->regs
[i
], idmap
))
3564 if (!stacksafe(old
, cur
, idmap
))
3572 /* A write screens off any subsequent reads; but write marks come from the
3573 * straight-line code between a state and its parent. When we arrive at a
3574 * jump target (in the first iteration of the propagate_liveness() loop),
3575 * we didn't arrive by the straight-line code, so read marks in state must
3576 * propagate to parent regardless of state's write marks.
3578 static bool do_propagate_liveness(const struct bpf_verifier_state
*state
,
3579 struct bpf_verifier_state
*parent
)
3581 bool writes
= parent
== state
->parent
; /* Observe write marks */
3582 bool touched
= false; /* any changes made? */
3587 /* Propagate read liveness of registers... */
3588 BUILD_BUG_ON(BPF_REG_FP
+ 1 != MAX_BPF_REG
);
3589 /* We don't need to worry about FP liveness because it's read-only */
3590 for (i
= 0; i
< BPF_REG_FP
; i
++) {
3591 if (parent
->regs
[i
].live
& REG_LIVE_READ
)
3593 if (writes
&& (state
->regs
[i
].live
& REG_LIVE_WRITTEN
))
3595 if (state
->regs
[i
].live
& REG_LIVE_READ
) {
3596 parent
->regs
[i
].live
|= REG_LIVE_READ
;
3600 /* ... and stack slots */
3601 for (i
= 0; i
< state
->allocated_stack
/ BPF_REG_SIZE
&&
3602 i
< parent
->allocated_stack
/ BPF_REG_SIZE
; i
++) {
3603 if (parent
->stack
[i
].slot_type
[0] != STACK_SPILL
)
3605 if (state
->stack
[i
].slot_type
[0] != STACK_SPILL
)
3607 if (parent
->stack
[i
].spilled_ptr
.live
& REG_LIVE_READ
)
3610 (state
->stack
[i
].spilled_ptr
.live
& REG_LIVE_WRITTEN
))
3612 if (state
->stack
[i
].spilled_ptr
.live
& REG_LIVE_READ
) {
3613 parent
->stack
[i
].spilled_ptr
.live
|= REG_LIVE_READ
;
3620 /* "parent" is "a state from which we reach the current state", but initially
3621 * it is not the state->parent (i.e. "the state whose straight-line code leads
3622 * to the current state"), instead it is the state that happened to arrive at
3623 * a (prunable) equivalent of the current state. See comment above
3624 * do_propagate_liveness() for consequences of this.
3625 * This function is just a more efficient way of calling mark_reg_read() or
3626 * mark_stack_slot_read() on each reg in "parent" that is read in "state",
3627 * though it requires that parent != state->parent in the call arguments.
3629 static void propagate_liveness(const struct bpf_verifier_state
*state
,
3630 struct bpf_verifier_state
*parent
)
3632 while (do_propagate_liveness(state
, parent
)) {
3633 /* Something changed, so we need to feed those changes onward */
3635 parent
= state
->parent
;
3639 static int is_state_visited(struct bpf_verifier_env
*env
, int insn_idx
)
3641 struct bpf_verifier_state_list
*new_sl
;
3642 struct bpf_verifier_state_list
*sl
;
3643 struct bpf_verifier_state
*cur
= env
->cur_state
;
3646 sl
= env
->explored_states
[insn_idx
];
3648 /* this 'insn_idx' instruction wasn't marked, so we will not
3649 * be doing state search here
3653 while (sl
!= STATE_LIST_MARK
) {
3654 if (states_equal(env
, &sl
->state
, cur
)) {
3655 /* reached equivalent register/stack state,
3657 * Registers read by the continuation are read by us.
3658 * If we have any write marks in env->cur_state, they
3659 * will prevent corresponding reads in the continuation
3660 * from reaching our parent (an explored_state). Our
3661 * own state will get the read marks recorded, but
3662 * they'll be immediately forgotten as we're pruning
3663 * this state and will pop a new one.
3665 propagate_liveness(&sl
->state
, cur
);
3671 /* there were no equivalent states, remember current one.
3672 * technically the current state is not proven to be safe yet,
3673 * but it will either reach bpf_exit (which means it's safe) or
3674 * it will be rejected. Since there are no loops, we won't be
3675 * seeing this 'insn_idx' instruction again on the way to bpf_exit
3677 new_sl
= kzalloc(sizeof(struct bpf_verifier_state_list
), GFP_KERNEL
);
3681 /* add new state to the head of linked list */
3682 copy_verifier_state(&new_sl
->state
, cur
);
3683 new_sl
->next
= env
->explored_states
[insn_idx
];
3684 env
->explored_states
[insn_idx
] = new_sl
;
3685 /* connect new state to parentage chain */
3686 cur
->parent
= &new_sl
->state
;
3687 /* clear write marks in current state: the writes we did are not writes
3688 * our child did, so they don't screen off its reads from us.
3689 * (There are no read marks in current state, because reads always mark
3690 * their parent and current state never has children yet. Only
3691 * explored_states can get read marks.)
3693 for (i
= 0; i
< BPF_REG_FP
; i
++)
3694 cur
->regs
[i
].live
= REG_LIVE_NONE
;
3695 for (i
= 0; i
< cur
->allocated_stack
/ BPF_REG_SIZE
; i
++)
3696 if (cur
->stack
[i
].slot_type
[0] == STACK_SPILL
)
3697 cur
->stack
[i
].spilled_ptr
.live
= REG_LIVE_NONE
;
3701 static int ext_analyzer_insn_hook(struct bpf_verifier_env
*env
,
3702 int insn_idx
, int prev_insn_idx
)
3704 if (!env
->analyzer_ops
|| !env
->analyzer_ops
->insn_hook
)
3707 return env
->analyzer_ops
->insn_hook(env
, insn_idx
, prev_insn_idx
);
3710 static int do_check(struct bpf_verifier_env
*env
)
3712 struct bpf_verifier_state
*state
;
3713 struct bpf_insn
*insns
= env
->prog
->insnsi
;
3714 struct bpf_reg_state
*regs
;
3715 int insn_cnt
= env
->prog
->len
;
3716 int insn_idx
, prev_insn_idx
= 0;
3717 int insn_processed
= 0;
3718 bool do_print_state
= false;
3720 state
= kzalloc(sizeof(struct bpf_verifier_state
), GFP_KERNEL
);
3723 env
->cur_state
= state
;
3724 init_reg_state(env
, state
->regs
);
3725 state
->parent
= NULL
;
3728 struct bpf_insn
*insn
;
3732 if (insn_idx
>= insn_cnt
) {
3733 verbose(env
, "invalid insn idx %d insn_cnt %d\n",
3734 insn_idx
, insn_cnt
);
3738 insn
= &insns
[insn_idx
];
3739 class = BPF_CLASS(insn
->code
);
3741 if (++insn_processed
> BPF_COMPLEXITY_LIMIT_INSNS
) {
3743 "BPF program is too large. Processed %d insn\n",
3748 err
= is_state_visited(env
, insn_idx
);
3752 /* found equivalent state, can prune the search */
3753 if (env
->log
.level
) {
3755 verbose(env
, "\nfrom %d to %d: safe\n",
3756 prev_insn_idx
, insn_idx
);
3758 verbose(env
, "%d: safe\n", insn_idx
);
3760 goto process_bpf_exit
;
3766 if (env
->log
.level
> 1 || (env
->log
.level
&& do_print_state
)) {
3767 if (env
->log
.level
> 1)
3768 verbose(env
, "%d:", insn_idx
);
3770 verbose(env
, "\nfrom %d to %d:",
3771 prev_insn_idx
, insn_idx
);
3772 print_verifier_state(env
, state
);
3773 do_print_state
= false;
3776 if (env
->log
.level
) {
3777 verbose(env
, "%d: ", insn_idx
);
3778 print_bpf_insn(verbose
, env
, insn
,
3779 env
->allow_ptr_leaks
);
3782 err
= ext_analyzer_insn_hook(env
, insn_idx
, prev_insn_idx
);
3786 regs
= cur_regs(env
);
3787 if (class == BPF_ALU
|| class == BPF_ALU64
) {
3788 err
= check_alu_op(env
, insn
);
3792 } else if (class == BPF_LDX
) {
3793 enum bpf_reg_type
*prev_src_type
, src_reg_type
;
3795 /* check for reserved fields is already done */
3797 /* check src operand */
3798 err
= check_reg_arg(env
, insn
->src_reg
, SRC_OP
);
3802 err
= check_reg_arg(env
, insn
->dst_reg
, DST_OP_NO_MARK
);
3806 src_reg_type
= regs
[insn
->src_reg
].type
;
3808 /* check that memory (src_reg + off) is readable,
3809 * the state of dst_reg will be updated by this func
3811 err
= check_mem_access(env
, insn_idx
, insn
->src_reg
, insn
->off
,
3812 BPF_SIZE(insn
->code
), BPF_READ
,
3817 prev_src_type
= &env
->insn_aux_data
[insn_idx
].ptr_type
;
3819 if (*prev_src_type
== NOT_INIT
) {
3821 * dst_reg = *(u32 *)(src_reg + off)
3822 * save type to validate intersecting paths
3824 *prev_src_type
= src_reg_type
;
3826 } else if (src_reg_type
!= *prev_src_type
&&
3827 (src_reg_type
== PTR_TO_CTX
||
3828 *prev_src_type
== PTR_TO_CTX
)) {
3829 /* ABuser program is trying to use the same insn
3830 * dst_reg = *(u32*) (src_reg + off)
3831 * with different pointer types:
3832 * src_reg == ctx in one branch and
3833 * src_reg == stack|map in some other branch.
3836 verbose(env
, "same insn cannot be used with different pointers\n");
3840 } else if (class == BPF_STX
) {
3841 enum bpf_reg_type
*prev_dst_type
, dst_reg_type
;
3843 if (BPF_MODE(insn
->code
) == BPF_XADD
) {
3844 err
= check_xadd(env
, insn_idx
, insn
);
3851 /* check src1 operand */
3852 err
= check_reg_arg(env
, insn
->src_reg
, SRC_OP
);
3855 /* check src2 operand */
3856 err
= check_reg_arg(env
, insn
->dst_reg
, SRC_OP
);
3860 dst_reg_type
= regs
[insn
->dst_reg
].type
;
3862 /* check that memory (dst_reg + off) is writeable */
3863 err
= check_mem_access(env
, insn_idx
, insn
->dst_reg
, insn
->off
,
3864 BPF_SIZE(insn
->code
), BPF_WRITE
,
3869 prev_dst_type
= &env
->insn_aux_data
[insn_idx
].ptr_type
;
3871 if (*prev_dst_type
== NOT_INIT
) {
3872 *prev_dst_type
= dst_reg_type
;
3873 } else if (dst_reg_type
!= *prev_dst_type
&&
3874 (dst_reg_type
== PTR_TO_CTX
||
3875 *prev_dst_type
== PTR_TO_CTX
)) {
3876 verbose(env
, "same insn cannot be used with different pointers\n");
3880 } else if (class == BPF_ST
) {
3881 if (BPF_MODE(insn
->code
) != BPF_MEM
||
3882 insn
->src_reg
!= BPF_REG_0
) {
3883 verbose(env
, "BPF_ST uses reserved fields\n");
3886 /* check src operand */
3887 err
= check_reg_arg(env
, insn
->dst_reg
, SRC_OP
);
3891 /* check that memory (dst_reg + off) is writeable */
3892 err
= check_mem_access(env
, insn_idx
, insn
->dst_reg
, insn
->off
,
3893 BPF_SIZE(insn
->code
), BPF_WRITE
,
3898 } else if (class == BPF_JMP
) {
3899 u8 opcode
= BPF_OP(insn
->code
);
3901 if (opcode
== BPF_CALL
) {
3902 if (BPF_SRC(insn
->code
) != BPF_K
||
3904 insn
->src_reg
!= BPF_REG_0
||
3905 insn
->dst_reg
!= BPF_REG_0
) {
3906 verbose(env
, "BPF_CALL uses reserved fields\n");
3910 err
= check_call(env
, insn
->imm
, insn_idx
);
3914 } else if (opcode
== BPF_JA
) {
3915 if (BPF_SRC(insn
->code
) != BPF_K
||
3917 insn
->src_reg
!= BPF_REG_0
||
3918 insn
->dst_reg
!= BPF_REG_0
) {
3919 verbose(env
, "BPF_JA uses reserved fields\n");
3923 insn_idx
+= insn
->off
+ 1;
3926 } else if (opcode
== BPF_EXIT
) {
3927 if (BPF_SRC(insn
->code
) != BPF_K
||
3929 insn
->src_reg
!= BPF_REG_0
||
3930 insn
->dst_reg
!= BPF_REG_0
) {
3931 verbose(env
, "BPF_EXIT uses reserved fields\n");
3935 /* eBPF calling convetion is such that R0 is used
3936 * to return the value from eBPF program.
3937 * Make sure that it's readable at this time
3938 * of bpf_exit, which means that program wrote
3939 * something into it earlier
3941 err
= check_reg_arg(env
, BPF_REG_0
, SRC_OP
);
3945 if (is_pointer_value(env
, BPF_REG_0
)) {
3946 verbose(env
, "R0 leaks addr as return value\n");
3950 err
= check_return_code(env
);
3954 err
= pop_stack(env
, &prev_insn_idx
, &insn_idx
);
3960 do_print_state
= true;
3964 err
= check_cond_jmp_op(env
, insn
, &insn_idx
);
3968 } else if (class == BPF_LD
) {
3969 u8 mode
= BPF_MODE(insn
->code
);
3971 if (mode
== BPF_ABS
|| mode
== BPF_IND
) {
3972 err
= check_ld_abs(env
, insn
);
3976 } else if (mode
== BPF_IMM
) {
3977 err
= check_ld_imm(env
, insn
);
3983 verbose(env
, "invalid BPF_LD mode\n");
3987 verbose(env
, "unknown insn class %d\n", class);
3994 verbose(env
, "processed %d insns, stack depth %d\n", insn_processed
,
3995 env
->prog
->aux
->stack_depth
);
3999 static int check_map_prealloc(struct bpf_map
*map
)
4001 return (map
->map_type
!= BPF_MAP_TYPE_HASH
&&
4002 map
->map_type
!= BPF_MAP_TYPE_PERCPU_HASH
&&
4003 map
->map_type
!= BPF_MAP_TYPE_HASH_OF_MAPS
) ||
4004 !(map
->map_flags
& BPF_F_NO_PREALLOC
);
4007 static int check_map_prog_compatibility(struct bpf_verifier_env
*env
,
4008 struct bpf_map
*map
,
4009 struct bpf_prog
*prog
)
4012 /* Make sure that BPF_PROG_TYPE_PERF_EVENT programs only use
4013 * preallocated hash maps, since doing memory allocation
4014 * in overflow_handler can crash depending on where nmi got
4017 if (prog
->type
== BPF_PROG_TYPE_PERF_EVENT
) {
4018 if (!check_map_prealloc(map
)) {
4019 verbose(env
, "perf_event programs can only use preallocated hash map\n");
4022 if (map
->inner_map_meta
&&
4023 !check_map_prealloc(map
->inner_map_meta
)) {
4024 verbose(env
, "perf_event programs can only use preallocated inner hash map\n");
4031 /* look for pseudo eBPF instructions that access map FDs and
4032 * replace them with actual map pointers
4034 static int replace_map_fd_with_map_ptr(struct bpf_verifier_env
*env
)
4036 struct bpf_insn
*insn
= env
->prog
->insnsi
;
4037 int insn_cnt
= env
->prog
->len
;
4040 err
= bpf_prog_calc_tag(env
->prog
);
4044 for (i
= 0; i
< insn_cnt
; i
++, insn
++) {
4045 if (BPF_CLASS(insn
->code
) == BPF_LDX
&&
4046 (BPF_MODE(insn
->code
) != BPF_MEM
|| insn
->imm
!= 0)) {
4047 verbose(env
, "BPF_LDX uses reserved fields\n");
4051 if (BPF_CLASS(insn
->code
) == BPF_STX
&&
4052 ((BPF_MODE(insn
->code
) != BPF_MEM
&&
4053 BPF_MODE(insn
->code
) != BPF_XADD
) || insn
->imm
!= 0)) {
4054 verbose(env
, "BPF_STX uses reserved fields\n");
4058 if (insn
[0].code
== (BPF_LD
| BPF_IMM
| BPF_DW
)) {
4059 struct bpf_map
*map
;
4062 if (i
== insn_cnt
- 1 || insn
[1].code
!= 0 ||
4063 insn
[1].dst_reg
!= 0 || insn
[1].src_reg
!= 0 ||
4065 verbose(env
, "invalid bpf_ld_imm64 insn\n");
4069 if (insn
->src_reg
== 0)
4070 /* valid generic load 64-bit imm */
4073 if (insn
->src_reg
!= BPF_PSEUDO_MAP_FD
) {
4075 "unrecognized bpf_ld_imm64 insn\n");
4079 f
= fdget(insn
->imm
);
4080 map
= __bpf_map_get(f
);
4082 verbose(env
, "fd %d is not pointing to valid bpf_map\n",
4084 return PTR_ERR(map
);
4087 err
= check_map_prog_compatibility(env
, map
, env
->prog
);
4093 /* store map pointer inside BPF_LD_IMM64 instruction */
4094 insn
[0].imm
= (u32
) (unsigned long) map
;
4095 insn
[1].imm
= ((u64
) (unsigned long) map
) >> 32;
4097 /* check whether we recorded this map already */
4098 for (j
= 0; j
< env
->used_map_cnt
; j
++)
4099 if (env
->used_maps
[j
] == map
) {
4104 if (env
->used_map_cnt
>= MAX_USED_MAPS
) {
4109 /* hold the map. If the program is rejected by verifier,
4110 * the map will be released by release_maps() or it
4111 * will be used by the valid program until it's unloaded
4112 * and all maps are released in free_bpf_prog_info()
4114 map
= bpf_map_inc(map
, false);
4117 return PTR_ERR(map
);
4119 env
->used_maps
[env
->used_map_cnt
++] = map
;
4128 /* now all pseudo BPF_LD_IMM64 instructions load valid
4129 * 'struct bpf_map *' into a register instead of user map_fd.
4130 * These pointers will be used later by verifier to validate map access.
4135 /* drop refcnt of maps used by the rejected program */
4136 static void release_maps(struct bpf_verifier_env
*env
)
4140 for (i
= 0; i
< env
->used_map_cnt
; i
++)
4141 bpf_map_put(env
->used_maps
[i
]);
4144 /* convert pseudo BPF_LD_IMM64 into generic BPF_LD_IMM64 */
4145 static void convert_pseudo_ld_imm64(struct bpf_verifier_env
*env
)
4147 struct bpf_insn
*insn
= env
->prog
->insnsi
;
4148 int insn_cnt
= env
->prog
->len
;
4151 for (i
= 0; i
< insn_cnt
; i
++, insn
++)
4152 if (insn
->code
== (BPF_LD
| BPF_IMM
| BPF_DW
))
4156 /* single env->prog->insni[off] instruction was replaced with the range
4157 * insni[off, off + cnt). Adjust corresponding insn_aux_data by copying
4158 * [0, off) and [off, end) to new locations, so the patched range stays zero
4160 static int adjust_insn_aux_data(struct bpf_verifier_env
*env
, u32 prog_len
,
4163 struct bpf_insn_aux_data
*new_data
, *old_data
= env
->insn_aux_data
;
4167 new_data
= vzalloc(sizeof(struct bpf_insn_aux_data
) * prog_len
);
4170 memcpy(new_data
, old_data
, sizeof(struct bpf_insn_aux_data
) * off
);
4171 memcpy(new_data
+ off
+ cnt
- 1, old_data
+ off
,
4172 sizeof(struct bpf_insn_aux_data
) * (prog_len
- off
- cnt
+ 1));
4173 env
->insn_aux_data
= new_data
;
4178 static struct bpf_prog
*bpf_patch_insn_data(struct bpf_verifier_env
*env
, u32 off
,
4179 const struct bpf_insn
*patch
, u32 len
)
4181 struct bpf_prog
*new_prog
;
4183 new_prog
= bpf_patch_insn_single(env
->prog
, off
, patch
, len
);
4186 if (adjust_insn_aux_data(env
, new_prog
->len
, off
, len
))
4191 /* convert load instructions that access fields of 'struct __sk_buff'
4192 * into sequence of instructions that access fields of 'struct sk_buff'
4194 static int convert_ctx_accesses(struct bpf_verifier_env
*env
)
4196 const struct bpf_verifier_ops
*ops
= env
->ops
;
4197 int i
, cnt
, size
, ctx_field_size
, delta
= 0;
4198 const int insn_cnt
= env
->prog
->len
;
4199 struct bpf_insn insn_buf
[16], *insn
;
4200 struct bpf_prog
*new_prog
;
4201 enum bpf_access_type type
;
4202 bool is_narrower_load
;
4205 if (ops
->gen_prologue
) {
4206 cnt
= ops
->gen_prologue(insn_buf
, env
->seen_direct_write
,
4208 if (cnt
>= ARRAY_SIZE(insn_buf
)) {
4209 verbose(env
, "bpf verifier is misconfigured\n");
4212 new_prog
= bpf_patch_insn_data(env
, 0, insn_buf
, cnt
);
4216 env
->prog
= new_prog
;
4221 if (!ops
->convert_ctx_access
)
4224 insn
= env
->prog
->insnsi
+ delta
;
4226 for (i
= 0; i
< insn_cnt
; i
++, insn
++) {
4227 if (insn
->code
== (BPF_LDX
| BPF_MEM
| BPF_B
) ||
4228 insn
->code
== (BPF_LDX
| BPF_MEM
| BPF_H
) ||
4229 insn
->code
== (BPF_LDX
| BPF_MEM
| BPF_W
) ||
4230 insn
->code
== (BPF_LDX
| BPF_MEM
| BPF_DW
))
4232 else if (insn
->code
== (BPF_STX
| BPF_MEM
| BPF_B
) ||
4233 insn
->code
== (BPF_STX
| BPF_MEM
| BPF_H
) ||
4234 insn
->code
== (BPF_STX
| BPF_MEM
| BPF_W
) ||
4235 insn
->code
== (BPF_STX
| BPF_MEM
| BPF_DW
))
4240 if (env
->insn_aux_data
[i
+ delta
].ptr_type
!= PTR_TO_CTX
)
4243 ctx_field_size
= env
->insn_aux_data
[i
+ delta
].ctx_field_size
;
4244 size
= BPF_LDST_BYTES(insn
);
4246 /* If the read access is a narrower load of the field,
4247 * convert to a 4/8-byte load, to minimum program type specific
4248 * convert_ctx_access changes. If conversion is successful,
4249 * we will apply proper mask to the result.
4251 is_narrower_load
= size
< ctx_field_size
;
4252 if (is_narrower_load
) {
4253 u32 off
= insn
->off
;
4256 if (type
== BPF_WRITE
) {
4257 verbose(env
, "bpf verifier narrow ctx access misconfigured\n");
4262 if (ctx_field_size
== 4)
4264 else if (ctx_field_size
== 8)
4267 insn
->off
= off
& ~(ctx_field_size
- 1);
4268 insn
->code
= BPF_LDX
| BPF_MEM
| size_code
;
4272 cnt
= ops
->convert_ctx_access(type
, insn
, insn_buf
, env
->prog
,
4274 if (cnt
== 0 || cnt
>= ARRAY_SIZE(insn_buf
) ||
4275 (ctx_field_size
&& !target_size
)) {
4276 verbose(env
, "bpf verifier is misconfigured\n");
4280 if (is_narrower_load
&& size
< target_size
) {
4281 if (ctx_field_size
<= 4)
4282 insn_buf
[cnt
++] = BPF_ALU32_IMM(BPF_AND
, insn
->dst_reg
,
4283 (1 << size
* 8) - 1);
4285 insn_buf
[cnt
++] = BPF_ALU64_IMM(BPF_AND
, insn
->dst_reg
,
4286 (1 << size
* 8) - 1);
4289 new_prog
= bpf_patch_insn_data(env
, i
+ delta
, insn_buf
, cnt
);
4295 /* keep walking new program and skip insns we just inserted */
4296 env
->prog
= new_prog
;
4297 insn
= new_prog
->insnsi
+ i
+ delta
;
4303 /* fixup insn->imm field of bpf_call instructions
4304 * and inline eligible helpers as explicit sequence of BPF instructions
4306 * this function is called after eBPF program passed verification
4308 static int fixup_bpf_calls(struct bpf_verifier_env
*env
)
4310 struct bpf_prog
*prog
= env
->prog
;
4311 struct bpf_insn
*insn
= prog
->insnsi
;
4312 const struct bpf_func_proto
*fn
;
4313 const int insn_cnt
= prog
->len
;
4314 struct bpf_insn insn_buf
[16];
4315 struct bpf_prog
*new_prog
;
4316 struct bpf_map
*map_ptr
;
4317 int i
, cnt
, delta
= 0;
4319 for (i
= 0; i
< insn_cnt
; i
++, insn
++) {
4320 if (insn
->code
!= (BPF_JMP
| BPF_CALL
))
4323 if (insn
->imm
== BPF_FUNC_get_route_realm
)
4324 prog
->dst_needed
= 1;
4325 if (insn
->imm
== BPF_FUNC_get_prandom_u32
)
4326 bpf_user_rnd_init_once();
4327 if (insn
->imm
== BPF_FUNC_tail_call
) {
4328 /* If we tail call into other programs, we
4329 * cannot make any assumptions since they can
4330 * be replaced dynamically during runtime in
4331 * the program array.
4333 prog
->cb_access
= 1;
4334 env
->prog
->aux
->stack_depth
= MAX_BPF_STACK
;
4336 /* mark bpf_tail_call as different opcode to avoid
4337 * conditional branch in the interpeter for every normal
4338 * call and to prevent accidental JITing by JIT compiler
4339 * that doesn't support bpf_tail_call yet
4342 insn
->code
= BPF_JMP
| BPF_TAIL_CALL
;
4346 /* BPF_EMIT_CALL() assumptions in some of the map_gen_lookup
4347 * handlers are currently limited to 64 bit only.
4349 if (ebpf_jit_enabled() && BITS_PER_LONG
== 64 &&
4350 insn
->imm
== BPF_FUNC_map_lookup_elem
) {
4351 map_ptr
= env
->insn_aux_data
[i
+ delta
].map_ptr
;
4352 if (map_ptr
== BPF_MAP_PTR_POISON
||
4353 !map_ptr
->ops
->map_gen_lookup
)
4354 goto patch_call_imm
;
4356 cnt
= map_ptr
->ops
->map_gen_lookup(map_ptr
, insn_buf
);
4357 if (cnt
== 0 || cnt
>= ARRAY_SIZE(insn_buf
)) {
4358 verbose(env
, "bpf verifier is misconfigured\n");
4362 new_prog
= bpf_patch_insn_data(env
, i
+ delta
, insn_buf
,
4369 /* keep walking new program and skip insns we just inserted */
4370 env
->prog
= prog
= new_prog
;
4371 insn
= new_prog
->insnsi
+ i
+ delta
;
4375 if (insn
->imm
== BPF_FUNC_redirect_map
) {
4376 /* Note, we cannot use prog directly as imm as subsequent
4377 * rewrites would still change the prog pointer. The only
4378 * stable address we can use is aux, which also works with
4379 * prog clones during blinding.
4381 u64 addr
= (unsigned long)prog
->aux
;
4382 struct bpf_insn r4_ld
[] = {
4383 BPF_LD_IMM64(BPF_REG_4
, addr
),
4386 cnt
= ARRAY_SIZE(r4_ld
);
4388 new_prog
= bpf_patch_insn_data(env
, i
+ delta
, r4_ld
, cnt
);
4393 env
->prog
= prog
= new_prog
;
4394 insn
= new_prog
->insnsi
+ i
+ delta
;
4397 fn
= env
->ops
->get_func_proto(insn
->imm
);
4398 /* all functions that have prototype and verifier allowed
4399 * programs to call them, must be real in-kernel functions
4403 "kernel subsystem misconfigured func %s#%d\n",
4404 func_id_name(insn
->imm
), insn
->imm
);
4407 insn
->imm
= fn
->func
- __bpf_call_base
;
4413 static void free_states(struct bpf_verifier_env
*env
)
4415 struct bpf_verifier_state_list
*sl
, *sln
;
4418 if (!env
->explored_states
)
4421 for (i
= 0; i
< env
->prog
->len
; i
++) {
4422 sl
= env
->explored_states
[i
];
4425 while (sl
!= STATE_LIST_MARK
) {
4432 kfree(env
->explored_states
);
4435 int bpf_check(struct bpf_prog
**prog
, union bpf_attr
*attr
)
4437 struct bpf_verifier_env
*env
;
4438 struct bpf_verifer_log
*log
;
4441 /* 'struct bpf_verifier_env' can be global, but since it's not small,
4442 * allocate/free it every time bpf_check() is called
4444 env
= kzalloc(sizeof(struct bpf_verifier_env
), GFP_KERNEL
);
4449 env
->insn_aux_data
= vzalloc(sizeof(struct bpf_insn_aux_data
) *
4452 if (!env
->insn_aux_data
)
4455 env
->ops
= bpf_verifier_ops
[env
->prog
->type
];
4457 /* grab the mutex to protect few globals used by verifier */
4458 mutex_lock(&bpf_verifier_lock
);
4460 if (attr
->log_level
|| attr
->log_buf
|| attr
->log_size
) {
4461 /* user requested verbose verifier output
4462 * and supplied buffer to store the verification trace
4464 log
->level
= attr
->log_level
;
4465 log
->ubuf
= (char __user
*) (unsigned long) attr
->log_buf
;
4466 log
->len_total
= attr
->log_size
;
4469 /* log attributes have to be sane */
4470 if (log
->len_total
< 128 || log
->len_total
> UINT_MAX
>> 8 ||
4471 !log
->level
|| !log
->ubuf
)
4475 env
->strict_alignment
= !!(attr
->prog_flags
& BPF_F_STRICT_ALIGNMENT
);
4476 if (!IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS
))
4477 env
->strict_alignment
= true;
4479 ret
= replace_map_fd_with_map_ptr(env
);
4481 goto skip_full_check
;
4483 env
->explored_states
= kcalloc(env
->prog
->len
,
4484 sizeof(struct bpf_verifier_state_list
*),
4487 if (!env
->explored_states
)
4488 goto skip_full_check
;
4490 ret
= check_cfg(env
);
4492 goto skip_full_check
;
4494 env
->allow_ptr_leaks
= capable(CAP_SYS_ADMIN
);
4496 ret
= do_check(env
);
4497 free_verifier_state(env
->cur_state
);
4498 env
->cur_state
= NULL
;
4501 while (!pop_stack(env
, NULL
, NULL
));
4505 /* program is valid, convert *(u32*)(ctx + off) accesses */
4506 ret
= convert_ctx_accesses(env
);
4509 ret
= fixup_bpf_calls(env
);
4511 if (log
->level
&& bpf_verifier_log_full(log
))
4513 if (log
->level
&& !log
->ubuf
) {
4515 goto err_release_maps
;
4518 if (ret
== 0 && env
->used_map_cnt
) {
4519 /* if program passed verifier, update used_maps in bpf_prog_info */
4520 env
->prog
->aux
->used_maps
= kmalloc_array(env
->used_map_cnt
,
4521 sizeof(env
->used_maps
[0]),
4524 if (!env
->prog
->aux
->used_maps
) {
4526 goto err_release_maps
;
4529 memcpy(env
->prog
->aux
->used_maps
, env
->used_maps
,
4530 sizeof(env
->used_maps
[0]) * env
->used_map_cnt
);
4531 env
->prog
->aux
->used_map_cnt
= env
->used_map_cnt
;
4533 /* program is valid. Convert pseudo bpf_ld_imm64 into generic
4534 * bpf_ld_imm64 instructions
4536 convert_pseudo_ld_imm64(env
);
4540 if (!env
->prog
->aux
->used_maps
)
4541 /* if we didn't copy map pointers into bpf_prog_info, release
4542 * them now. Otherwise free_bpf_prog_info() will release them.
4547 mutex_unlock(&bpf_verifier_lock
);
4548 vfree(env
->insn_aux_data
);
4554 static const struct bpf_verifier_ops
* const bpf_analyzer_ops
[] = {
4555 [BPF_PROG_TYPE_XDP
] = &xdp_analyzer_ops
,
4556 [BPF_PROG_TYPE_SCHED_CLS
] = &tc_cls_act_analyzer_ops
,
4559 int bpf_analyzer(struct bpf_prog
*prog
, const struct bpf_ext_analyzer_ops
*ops
,
4562 struct bpf_verifier_env
*env
;
4565 if (prog
->type
>= ARRAY_SIZE(bpf_analyzer_ops
) ||
4566 !bpf_analyzer_ops
[prog
->type
])
4569 env
= kzalloc(sizeof(struct bpf_verifier_env
), GFP_KERNEL
);
4573 env
->insn_aux_data
= vzalloc(sizeof(struct bpf_insn_aux_data
) *
4576 if (!env
->insn_aux_data
)
4579 env
->ops
= bpf_analyzer_ops
[env
->prog
->type
];
4580 env
->analyzer_ops
= ops
;
4581 env
->analyzer_priv
= priv
;
4583 /* grab the mutex to protect few globals used by verifier */
4584 mutex_lock(&bpf_verifier_lock
);
4586 env
->strict_alignment
= false;
4587 if (!IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS
))
4588 env
->strict_alignment
= true;
4590 env
->explored_states
= kcalloc(env
->prog
->len
,
4591 sizeof(struct bpf_verifier_state_list
*),
4594 if (!env
->explored_states
)
4595 goto skip_full_check
;
4597 ret
= check_cfg(env
);
4599 goto skip_full_check
;
4601 env
->allow_ptr_leaks
= capable(CAP_SYS_ADMIN
);
4603 ret
= do_check(env
);
4604 free_verifier_state(env
->cur_state
);
4605 env
->cur_state
= NULL
;
4608 while (!pop_stack(env
, NULL
, NULL
));
4611 mutex_unlock(&bpf_verifier_lock
);
4612 vfree(env
->insn_aux_data
);
4617 EXPORT_SYMBOL_GPL(bpf_analyzer
);