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>
23 #include <linux/bsearch.h>
24 #include <linux/sort.h>
28 static const struct bpf_verifier_ops
* const bpf_verifier_ops
[] = {
29 #define BPF_PROG_TYPE(_id, _name) \
30 [_id] = & _name ## _verifier_ops,
31 #define BPF_MAP_TYPE(_id, _ops)
32 #include <linux/bpf_types.h>
37 /* bpf_check() is a static code analyzer that walks eBPF program
38 * instruction by instruction and updates register/stack state.
39 * All paths of conditional branches are analyzed until 'bpf_exit' insn.
41 * The first pass is depth-first-search to check that the program is a DAG.
42 * It rejects the following programs:
43 * - larger than BPF_MAXINSNS insns
44 * - if loop is present (detected via back-edge)
45 * - unreachable insns exist (shouldn't be a forest. program = one function)
46 * - out of bounds or malformed jumps
47 * The second pass is all possible path descent from the 1st insn.
48 * Since it's analyzing all pathes through the program, the length of the
49 * analysis is limited to 64k insn, which may be hit even if total number of
50 * insn is less then 4K, but there are too many branches that change stack/regs.
51 * Number of 'branches to be analyzed' is limited to 1k
53 * On entry to each instruction, each register has a type, and the instruction
54 * changes the types of the registers depending on instruction semantics.
55 * If instruction is BPF_MOV64_REG(BPF_REG_1, BPF_REG_5), then type of R5 is
58 * All registers are 64-bit.
59 * R0 - return register
60 * R1-R5 argument passing registers
61 * R6-R9 callee saved registers
62 * R10 - frame pointer read-only
64 * At the start of BPF program the register R1 contains a pointer to bpf_context
65 * and has type PTR_TO_CTX.
67 * Verifier tracks arithmetic operations on pointers in case:
68 * BPF_MOV64_REG(BPF_REG_1, BPF_REG_10),
69 * BPF_ALU64_IMM(BPF_ADD, BPF_REG_1, -20),
70 * 1st insn copies R10 (which has FRAME_PTR) type into R1
71 * and 2nd arithmetic instruction is pattern matched to recognize
72 * that it wants to construct a pointer to some element within stack.
73 * So after 2nd insn, the register R1 has type PTR_TO_STACK
74 * (and -20 constant is saved for further stack bounds checking).
75 * Meaning that this reg is a pointer to stack plus known immediate constant.
77 * Most of the time the registers have SCALAR_VALUE type, which
78 * means the register has some value, but it's not a valid pointer.
79 * (like pointer plus pointer becomes SCALAR_VALUE type)
81 * When verifier sees load or store instructions the type of base register
82 * can be: PTR_TO_MAP_VALUE, PTR_TO_CTX, PTR_TO_STACK. These are three pointer
83 * types recognized by check_mem_access() function.
85 * PTR_TO_MAP_VALUE means that this register is pointing to 'map element value'
86 * and the range of [ptr, ptr + map's value_size) is accessible.
88 * registers used to pass values to function calls are checked against
89 * function argument constraints.
91 * ARG_PTR_TO_MAP_KEY is one of such argument constraints.
92 * It means that the register type passed to this function must be
93 * PTR_TO_STACK and it will be used inside the function as
94 * 'pointer to map element key'
96 * For example the argument constraints for bpf_map_lookup_elem():
97 * .ret_type = RET_PTR_TO_MAP_VALUE_OR_NULL,
98 * .arg1_type = ARG_CONST_MAP_PTR,
99 * .arg2_type = ARG_PTR_TO_MAP_KEY,
101 * ret_type says that this function returns 'pointer to map elem value or null'
102 * function expects 1st argument to be a const pointer to 'struct bpf_map' and
103 * 2nd argument should be a pointer to stack, which will be used inside
104 * the helper function as a pointer to map element key.
106 * On the kernel side the helper function looks like:
107 * u64 bpf_map_lookup_elem(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5)
109 * struct bpf_map *map = (struct bpf_map *) (unsigned long) r1;
110 * void *key = (void *) (unsigned long) r2;
113 * here kernel can access 'key' and 'map' pointers safely, knowing that
114 * [key, key + map->key_size) bytes are valid and were initialized on
115 * the stack of eBPF program.
118 * Corresponding eBPF program may look like:
119 * BPF_MOV64_REG(BPF_REG_2, BPF_REG_10), // after this insn R2 type is FRAME_PTR
120 * BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, -4), // after this insn R2 type is PTR_TO_STACK
121 * BPF_LD_MAP_FD(BPF_REG_1, map_fd), // after this insn R1 type is CONST_PTR_TO_MAP
122 * BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0, BPF_FUNC_map_lookup_elem),
123 * here verifier looks at prototype of map_lookup_elem() and sees:
124 * .arg1_type == ARG_CONST_MAP_PTR and R1->type == CONST_PTR_TO_MAP, which is ok,
125 * Now verifier knows that this map has key of R1->map_ptr->key_size bytes
127 * Then .arg2_type == ARG_PTR_TO_MAP_KEY and R2->type == PTR_TO_STACK, ok so far,
128 * Now verifier checks that [R2, R2 + map's key_size) are within stack limits
129 * and were initialized prior to this call.
130 * If it's ok, then verifier allows this BPF_CALL insn and looks at
131 * .ret_type which is RET_PTR_TO_MAP_VALUE_OR_NULL, so it sets
132 * R0->type = PTR_TO_MAP_VALUE_OR_NULL which means bpf_map_lookup_elem() function
133 * returns ether pointer to map value or NULL.
135 * When type PTR_TO_MAP_VALUE_OR_NULL passes through 'if (reg != 0) goto +off'
136 * insn, the register holding that pointer in the true branch changes state to
137 * PTR_TO_MAP_VALUE and the same register changes state to CONST_IMM in the false
138 * branch. See check_cond_jmp_op().
140 * After the call R0 is set to return type of the function and registers R1-R5
141 * are set to NOT_INIT to indicate that they are no longer readable.
144 /* verifier_state + insn_idx are pushed to stack when branch is encountered */
145 struct bpf_verifier_stack_elem
{
146 /* verifer state is 'st'
147 * before processing instruction 'insn_idx'
148 * and after processing instruction 'prev_insn_idx'
150 struct bpf_verifier_state st
;
153 struct bpf_verifier_stack_elem
*next
;
156 #define BPF_COMPLEXITY_LIMIT_INSNS 131072
157 #define BPF_COMPLEXITY_LIMIT_STACK 1024
159 #define BPF_MAP_PTR_POISON ((void *)0xeB9F + POISON_POINTER_DELTA)
161 struct bpf_call_arg_meta
{
162 struct bpf_map
*map_ptr
;
169 static DEFINE_MUTEX(bpf_verifier_lock
);
171 /* log_level controls verbosity level of eBPF verifier.
172 * bpf_verifier_log_write() is used to dump the verification trace to the log,
173 * so the user can figure out what's wrong with the program
175 __printf(2, 3) void bpf_verifier_log_write(struct bpf_verifier_env
*env
,
176 const char *fmt
, ...)
178 struct bpf_verifer_log
*log
= &env
->log
;
182 if (!log
->level
|| !log
->ubuf
|| bpf_verifier_log_full(log
))
186 n
= vscnprintf(log
->kbuf
, BPF_VERIFIER_TMP_LOG_SIZE
, fmt
, args
);
189 WARN_ONCE(n
>= BPF_VERIFIER_TMP_LOG_SIZE
- 1,
190 "verifier log line truncated - local buffer too short\n");
192 n
= min(log
->len_total
- log
->len_used
- 1, n
);
195 if (!copy_to_user(log
->ubuf
+ log
->len_used
, log
->kbuf
, n
+ 1))
200 EXPORT_SYMBOL_GPL(bpf_verifier_log_write
);
201 /* Historically bpf_verifier_log_write was called verbose, but the name was too
202 * generic for symbol export. The function was renamed, but not the calls in
203 * the verifier to avoid complicating backports. Hence the alias below.
205 static __printf(2, 3) void verbose(struct bpf_verifier_env
*env
,
206 const char *fmt
, ...)
207 __attribute__((alias("bpf_verifier_log_write")));
209 static bool type_is_pkt_pointer(enum bpf_reg_type type
)
211 return type
== PTR_TO_PACKET
||
212 type
== PTR_TO_PACKET_META
;
215 /* string representation of 'enum bpf_reg_type' */
216 static const char * const reg_type_str
[] = {
218 [SCALAR_VALUE
] = "inv",
219 [PTR_TO_CTX
] = "ctx",
220 [CONST_PTR_TO_MAP
] = "map_ptr",
221 [PTR_TO_MAP_VALUE
] = "map_value",
222 [PTR_TO_MAP_VALUE_OR_NULL
] = "map_value_or_null",
223 [PTR_TO_STACK
] = "fp",
224 [PTR_TO_PACKET
] = "pkt",
225 [PTR_TO_PACKET_META
] = "pkt_meta",
226 [PTR_TO_PACKET_END
] = "pkt_end",
229 static void print_liveness(struct bpf_verifier_env
*env
,
230 enum bpf_reg_liveness live
)
232 if (live
& (REG_LIVE_READ
| REG_LIVE_WRITTEN
))
234 if (live
& REG_LIVE_READ
)
236 if (live
& REG_LIVE_WRITTEN
)
240 static struct bpf_func_state
*func(struct bpf_verifier_env
*env
,
241 const struct bpf_reg_state
*reg
)
243 struct bpf_verifier_state
*cur
= env
->cur_state
;
245 return cur
->frame
[reg
->frameno
];
248 static void print_verifier_state(struct bpf_verifier_env
*env
,
249 const struct bpf_func_state
*state
)
251 const struct bpf_reg_state
*reg
;
256 verbose(env
, " frame%d:", state
->frameno
);
257 for (i
= 0; i
< MAX_BPF_REG
; i
++) {
258 reg
= &state
->regs
[i
];
262 verbose(env
, " R%d", i
);
263 print_liveness(env
, reg
->live
);
264 verbose(env
, "=%s", reg_type_str
[t
]);
265 if ((t
== SCALAR_VALUE
|| t
== PTR_TO_STACK
) &&
266 tnum_is_const(reg
->var_off
)) {
267 /* reg->off should be 0 for SCALAR_VALUE */
268 verbose(env
, "%lld", reg
->var_off
.value
+ reg
->off
);
269 if (t
== PTR_TO_STACK
)
270 verbose(env
, ",call_%d", func(env
, reg
)->callsite
);
272 verbose(env
, "(id=%d", reg
->id
);
273 if (t
!= SCALAR_VALUE
)
274 verbose(env
, ",off=%d", reg
->off
);
275 if (type_is_pkt_pointer(t
))
276 verbose(env
, ",r=%d", reg
->range
);
277 else if (t
== CONST_PTR_TO_MAP
||
278 t
== PTR_TO_MAP_VALUE
||
279 t
== PTR_TO_MAP_VALUE_OR_NULL
)
280 verbose(env
, ",ks=%d,vs=%d",
281 reg
->map_ptr
->key_size
,
282 reg
->map_ptr
->value_size
);
283 if (tnum_is_const(reg
->var_off
)) {
284 /* Typically an immediate SCALAR_VALUE, but
285 * could be a pointer whose offset is too big
288 verbose(env
, ",imm=%llx", reg
->var_off
.value
);
290 if (reg
->smin_value
!= reg
->umin_value
&&
291 reg
->smin_value
!= S64_MIN
)
292 verbose(env
, ",smin_value=%lld",
293 (long long)reg
->smin_value
);
294 if (reg
->smax_value
!= reg
->umax_value
&&
295 reg
->smax_value
!= S64_MAX
)
296 verbose(env
, ",smax_value=%lld",
297 (long long)reg
->smax_value
);
298 if (reg
->umin_value
!= 0)
299 verbose(env
, ",umin_value=%llu",
300 (unsigned long long)reg
->umin_value
);
301 if (reg
->umax_value
!= U64_MAX
)
302 verbose(env
, ",umax_value=%llu",
303 (unsigned long long)reg
->umax_value
);
304 if (!tnum_is_unknown(reg
->var_off
)) {
307 tnum_strn(tn_buf
, sizeof(tn_buf
), reg
->var_off
);
308 verbose(env
, ",var_off=%s", tn_buf
);
314 for (i
= 0; i
< state
->allocated_stack
/ BPF_REG_SIZE
; i
++) {
315 if (state
->stack
[i
].slot_type
[0] == STACK_SPILL
) {
316 verbose(env
, " fp%d",
317 (-i
- 1) * BPF_REG_SIZE
);
318 print_liveness(env
, state
->stack
[i
].spilled_ptr
.live
);
320 reg_type_str
[state
->stack
[i
].spilled_ptr
.type
]);
322 if (state
->stack
[i
].slot_type
[0] == STACK_ZERO
)
323 verbose(env
, " fp%d=0", (-i
- 1) * BPF_REG_SIZE
);
328 static int copy_stack_state(struct bpf_func_state
*dst
,
329 const struct bpf_func_state
*src
)
333 if (WARN_ON_ONCE(dst
->allocated_stack
< src
->allocated_stack
)) {
334 /* internal bug, make state invalid to reject the program */
335 memset(dst
, 0, sizeof(*dst
));
338 memcpy(dst
->stack
, src
->stack
,
339 sizeof(*src
->stack
) * (src
->allocated_stack
/ BPF_REG_SIZE
));
343 /* do_check() starts with zero-sized stack in struct bpf_verifier_state to
344 * make it consume minimal amount of memory. check_stack_write() access from
345 * the program calls into realloc_func_state() to grow the stack size.
346 * Note there is a non-zero 'parent' pointer inside bpf_verifier_state
347 * which this function copies over. It points to previous bpf_verifier_state
348 * which is never reallocated
350 static int realloc_func_state(struct bpf_func_state
*state
, int size
,
353 u32 old_size
= state
->allocated_stack
;
354 struct bpf_stack_state
*new_stack
;
355 int slot
= size
/ BPF_REG_SIZE
;
357 if (size
<= old_size
|| !size
) {
360 state
->allocated_stack
= slot
* BPF_REG_SIZE
;
361 if (!size
&& old_size
) {
367 new_stack
= kmalloc_array(slot
, sizeof(struct bpf_stack_state
),
373 memcpy(new_stack
, state
->stack
,
374 sizeof(*new_stack
) * (old_size
/ BPF_REG_SIZE
));
375 memset(new_stack
+ old_size
/ BPF_REG_SIZE
, 0,
376 sizeof(*new_stack
) * (size
- old_size
) / BPF_REG_SIZE
);
378 state
->allocated_stack
= slot
* BPF_REG_SIZE
;
380 state
->stack
= new_stack
;
384 static void free_func_state(struct bpf_func_state
*state
)
392 static void free_verifier_state(struct bpf_verifier_state
*state
,
397 for (i
= 0; i
<= state
->curframe
; i
++) {
398 free_func_state(state
->frame
[i
]);
399 state
->frame
[i
] = NULL
;
405 /* copy verifier state from src to dst growing dst stack space
406 * when necessary to accommodate larger src stack
408 static int copy_func_state(struct bpf_func_state
*dst
,
409 const struct bpf_func_state
*src
)
413 err
= realloc_func_state(dst
, src
->allocated_stack
, false);
416 memcpy(dst
, src
, offsetof(struct bpf_func_state
, allocated_stack
));
417 return copy_stack_state(dst
, src
);
420 static int copy_verifier_state(struct bpf_verifier_state
*dst_state
,
421 const struct bpf_verifier_state
*src
)
423 struct bpf_func_state
*dst
;
426 /* if dst has more stack frames then src frame, free them */
427 for (i
= src
->curframe
+ 1; i
<= dst_state
->curframe
; i
++) {
428 free_func_state(dst_state
->frame
[i
]);
429 dst_state
->frame
[i
] = NULL
;
431 dst_state
->curframe
= src
->curframe
;
432 dst_state
->parent
= src
->parent
;
433 for (i
= 0; i
<= src
->curframe
; i
++) {
434 dst
= dst_state
->frame
[i
];
436 dst
= kzalloc(sizeof(*dst
), GFP_KERNEL
);
439 dst_state
->frame
[i
] = dst
;
441 err
= copy_func_state(dst
, src
->frame
[i
]);
448 static int pop_stack(struct bpf_verifier_env
*env
, int *prev_insn_idx
,
451 struct bpf_verifier_state
*cur
= env
->cur_state
;
452 struct bpf_verifier_stack_elem
*elem
, *head
= env
->head
;
455 if (env
->head
== NULL
)
459 err
= copy_verifier_state(cur
, &head
->st
);
464 *insn_idx
= head
->insn_idx
;
466 *prev_insn_idx
= head
->prev_insn_idx
;
468 free_verifier_state(&head
->st
, false);
475 static struct bpf_verifier_state
*push_stack(struct bpf_verifier_env
*env
,
476 int insn_idx
, int prev_insn_idx
)
478 struct bpf_verifier_state
*cur
= env
->cur_state
;
479 struct bpf_verifier_stack_elem
*elem
;
482 elem
= kzalloc(sizeof(struct bpf_verifier_stack_elem
), GFP_KERNEL
);
486 elem
->insn_idx
= insn_idx
;
487 elem
->prev_insn_idx
= prev_insn_idx
;
488 elem
->next
= env
->head
;
491 err
= copy_verifier_state(&elem
->st
, cur
);
494 if (env
->stack_size
> BPF_COMPLEXITY_LIMIT_STACK
) {
495 verbose(env
, "BPF program is too complex\n");
500 free_verifier_state(env
->cur_state
, true);
501 env
->cur_state
= NULL
;
502 /* pop all elements and return */
503 while (!pop_stack(env
, NULL
, NULL
));
507 #define CALLER_SAVED_REGS 6
508 static const int caller_saved
[CALLER_SAVED_REGS
] = {
509 BPF_REG_0
, BPF_REG_1
, BPF_REG_2
, BPF_REG_3
, BPF_REG_4
, BPF_REG_5
511 #define CALLEE_SAVED_REGS 5
512 static const int callee_saved
[CALLEE_SAVED_REGS
] = {
513 BPF_REG_6
, BPF_REG_7
, BPF_REG_8
, BPF_REG_9
516 static void __mark_reg_not_init(struct bpf_reg_state
*reg
);
518 /* Mark the unknown part of a register (variable offset or scalar value) as
519 * known to have the value @imm.
521 static void __mark_reg_known(struct bpf_reg_state
*reg
, u64 imm
)
524 reg
->var_off
= tnum_const(imm
);
525 reg
->smin_value
= (s64
)imm
;
526 reg
->smax_value
= (s64
)imm
;
527 reg
->umin_value
= imm
;
528 reg
->umax_value
= imm
;
531 /* Mark the 'variable offset' part of a register as zero. This should be
532 * used only on registers holding a pointer type.
534 static void __mark_reg_known_zero(struct bpf_reg_state
*reg
)
536 __mark_reg_known(reg
, 0);
539 static void __mark_reg_const_zero(struct bpf_reg_state
*reg
)
541 __mark_reg_known(reg
, 0);
543 reg
->type
= SCALAR_VALUE
;
546 static void mark_reg_known_zero(struct bpf_verifier_env
*env
,
547 struct bpf_reg_state
*regs
, u32 regno
)
549 if (WARN_ON(regno
>= MAX_BPF_REG
)) {
550 verbose(env
, "mark_reg_known_zero(regs, %u)\n", regno
);
551 /* Something bad happened, let's kill all regs */
552 for (regno
= 0; regno
< MAX_BPF_REG
; regno
++)
553 __mark_reg_not_init(regs
+ regno
);
556 __mark_reg_known_zero(regs
+ regno
);
559 static bool reg_is_pkt_pointer(const struct bpf_reg_state
*reg
)
561 return type_is_pkt_pointer(reg
->type
);
564 static bool reg_is_pkt_pointer_any(const struct bpf_reg_state
*reg
)
566 return reg_is_pkt_pointer(reg
) ||
567 reg
->type
== PTR_TO_PACKET_END
;
570 /* Unmodified PTR_TO_PACKET[_META,_END] register from ctx access. */
571 static bool reg_is_init_pkt_pointer(const struct bpf_reg_state
*reg
,
572 enum bpf_reg_type which
)
574 /* The register can already have a range from prior markings.
575 * This is fine as long as it hasn't been advanced from its
578 return reg
->type
== which
&&
581 tnum_equals_const(reg
->var_off
, 0);
584 /* Attempts to improve min/max values based on var_off information */
585 static void __update_reg_bounds(struct bpf_reg_state
*reg
)
587 /* min signed is max(sign bit) | min(other bits) */
588 reg
->smin_value
= max_t(s64
, reg
->smin_value
,
589 reg
->var_off
.value
| (reg
->var_off
.mask
& S64_MIN
));
590 /* max signed is min(sign bit) | max(other bits) */
591 reg
->smax_value
= min_t(s64
, reg
->smax_value
,
592 reg
->var_off
.value
| (reg
->var_off
.mask
& S64_MAX
));
593 reg
->umin_value
= max(reg
->umin_value
, reg
->var_off
.value
);
594 reg
->umax_value
= min(reg
->umax_value
,
595 reg
->var_off
.value
| reg
->var_off
.mask
);
598 /* Uses signed min/max values to inform unsigned, and vice-versa */
599 static void __reg_deduce_bounds(struct bpf_reg_state
*reg
)
601 /* Learn sign from signed bounds.
602 * If we cannot cross the sign boundary, then signed and unsigned bounds
603 * are the same, so combine. This works even in the negative case, e.g.
604 * -3 s<= x s<= -1 implies 0xf...fd u<= x u<= 0xf...ff.
606 if (reg
->smin_value
>= 0 || reg
->smax_value
< 0) {
607 reg
->smin_value
= reg
->umin_value
= max_t(u64
, reg
->smin_value
,
609 reg
->smax_value
= reg
->umax_value
= min_t(u64
, reg
->smax_value
,
613 /* Learn sign from unsigned bounds. Signed bounds cross the sign
614 * boundary, so we must be careful.
616 if ((s64
)reg
->umax_value
>= 0) {
617 /* Positive. We can't learn anything from the smin, but smax
618 * is positive, hence safe.
620 reg
->smin_value
= reg
->umin_value
;
621 reg
->smax_value
= reg
->umax_value
= min_t(u64
, reg
->smax_value
,
623 } else if ((s64
)reg
->umin_value
< 0) {
624 /* Negative. We can't learn anything from the smax, but smin
625 * is negative, hence safe.
627 reg
->smin_value
= reg
->umin_value
= max_t(u64
, reg
->smin_value
,
629 reg
->smax_value
= reg
->umax_value
;
633 /* Attempts to improve var_off based on unsigned min/max information */
634 static void __reg_bound_offset(struct bpf_reg_state
*reg
)
636 reg
->var_off
= tnum_intersect(reg
->var_off
,
637 tnum_range(reg
->umin_value
,
641 /* Reset the min/max bounds of a register */
642 static void __mark_reg_unbounded(struct bpf_reg_state
*reg
)
644 reg
->smin_value
= S64_MIN
;
645 reg
->smax_value
= S64_MAX
;
647 reg
->umax_value
= U64_MAX
;
650 /* Mark a register as having a completely unknown (scalar) value. */
651 static void __mark_reg_unknown(struct bpf_reg_state
*reg
)
653 reg
->type
= SCALAR_VALUE
;
656 reg
->var_off
= tnum_unknown
;
658 __mark_reg_unbounded(reg
);
661 static void mark_reg_unknown(struct bpf_verifier_env
*env
,
662 struct bpf_reg_state
*regs
, u32 regno
)
664 if (WARN_ON(regno
>= MAX_BPF_REG
)) {
665 verbose(env
, "mark_reg_unknown(regs, %u)\n", regno
);
666 /* Something bad happened, let's kill all regs except FP */
667 for (regno
= 0; regno
< BPF_REG_FP
; regno
++)
668 __mark_reg_not_init(regs
+ regno
);
671 __mark_reg_unknown(regs
+ regno
);
674 static void __mark_reg_not_init(struct bpf_reg_state
*reg
)
676 __mark_reg_unknown(reg
);
677 reg
->type
= NOT_INIT
;
680 static void mark_reg_not_init(struct bpf_verifier_env
*env
,
681 struct bpf_reg_state
*regs
, u32 regno
)
683 if (WARN_ON(regno
>= MAX_BPF_REG
)) {
684 verbose(env
, "mark_reg_not_init(regs, %u)\n", regno
);
685 /* Something bad happened, let's kill all regs except FP */
686 for (regno
= 0; regno
< BPF_REG_FP
; regno
++)
687 __mark_reg_not_init(regs
+ regno
);
690 __mark_reg_not_init(regs
+ regno
);
693 static void init_reg_state(struct bpf_verifier_env
*env
,
694 struct bpf_func_state
*state
)
696 struct bpf_reg_state
*regs
= state
->regs
;
699 for (i
= 0; i
< MAX_BPF_REG
; i
++) {
700 mark_reg_not_init(env
, regs
, i
);
701 regs
[i
].live
= REG_LIVE_NONE
;
705 regs
[BPF_REG_FP
].type
= PTR_TO_STACK
;
706 mark_reg_known_zero(env
, regs
, BPF_REG_FP
);
707 regs
[BPF_REG_FP
].frameno
= state
->frameno
;
709 /* 1st arg to a function */
710 regs
[BPF_REG_1
].type
= PTR_TO_CTX
;
711 mark_reg_known_zero(env
, regs
, BPF_REG_1
);
714 #define BPF_MAIN_FUNC (-1)
715 static void init_func_state(struct bpf_verifier_env
*env
,
716 struct bpf_func_state
*state
,
717 int callsite
, int frameno
, int subprogno
)
719 state
->callsite
= callsite
;
720 state
->frameno
= frameno
;
721 state
->subprogno
= subprogno
;
722 init_reg_state(env
, state
);
726 SRC_OP
, /* register is used as source operand */
727 DST_OP
, /* register is used as destination operand */
728 DST_OP_NO_MARK
/* same as above, check only, don't mark */
731 static int cmp_subprogs(const void *a
, const void *b
)
733 return *(int *)a
- *(int *)b
;
736 static int find_subprog(struct bpf_verifier_env
*env
, int off
)
740 p
= bsearch(&off
, env
->subprog_starts
, env
->subprog_cnt
,
741 sizeof(env
->subprog_starts
[0]), cmp_subprogs
);
744 return p
- env
->subprog_starts
;
748 static int add_subprog(struct bpf_verifier_env
*env
, int off
)
750 int insn_cnt
= env
->prog
->len
;
753 if (off
>= insn_cnt
|| off
< 0) {
754 verbose(env
, "call to invalid destination\n");
757 ret
= find_subprog(env
, off
);
760 if (env
->subprog_cnt
>= BPF_MAX_SUBPROGS
) {
761 verbose(env
, "too many subprograms\n");
764 env
->subprog_starts
[env
->subprog_cnt
++] = off
;
765 sort(env
->subprog_starts
, env
->subprog_cnt
,
766 sizeof(env
->subprog_starts
[0]), cmp_subprogs
, NULL
);
770 static int check_subprogs(struct bpf_verifier_env
*env
)
772 int i
, ret
, subprog_start
, subprog_end
, off
, cur_subprog
= 0;
773 struct bpf_insn
*insn
= env
->prog
->insnsi
;
774 int insn_cnt
= env
->prog
->len
;
776 /* determine subprog starts. The end is one before the next starts */
777 for (i
= 0; i
< insn_cnt
; i
++) {
778 if (insn
[i
].code
!= (BPF_JMP
| BPF_CALL
))
780 if (insn
[i
].src_reg
!= BPF_PSEUDO_CALL
)
782 if (!env
->allow_ptr_leaks
) {
783 verbose(env
, "function calls to other bpf functions are allowed for root only\n");
786 if (bpf_prog_is_dev_bound(env
->prog
->aux
)) {
787 verbose(env
, "function calls in offloaded programs are not supported yet\n");
790 ret
= add_subprog(env
, i
+ insn
[i
].imm
+ 1);
795 if (env
->log
.level
> 1)
796 for (i
= 0; i
< env
->subprog_cnt
; i
++)
797 verbose(env
, "func#%d @%d\n", i
, env
->subprog_starts
[i
]);
799 /* now check that all jumps are within the same subprog */
801 if (env
->subprog_cnt
== cur_subprog
)
802 subprog_end
= insn_cnt
;
804 subprog_end
= env
->subprog_starts
[cur_subprog
++];
805 for (i
= 0; i
< insn_cnt
; i
++) {
806 u8 code
= insn
[i
].code
;
808 if (BPF_CLASS(code
) != BPF_JMP
)
810 if (BPF_OP(code
) == BPF_EXIT
|| BPF_OP(code
) == BPF_CALL
)
812 off
= i
+ insn
[i
].off
+ 1;
813 if (off
< subprog_start
|| off
>= subprog_end
) {
814 verbose(env
, "jump out of range from insn %d to %d\n", i
, off
);
818 if (i
== subprog_end
- 1) {
819 /* to avoid fall-through from one subprog into another
820 * the last insn of the subprog should be either exit
821 * or unconditional jump back
823 if (code
!= (BPF_JMP
| BPF_EXIT
) &&
824 code
!= (BPF_JMP
| BPF_JA
)) {
825 verbose(env
, "last insn is not an exit or jmp\n");
828 subprog_start
= subprog_end
;
829 if (env
->subprog_cnt
== cur_subprog
)
830 subprog_end
= insn_cnt
;
832 subprog_end
= env
->subprog_starts
[cur_subprog
++];
839 struct bpf_verifier_state
*skip_callee(struct bpf_verifier_env
*env
,
840 const struct bpf_verifier_state
*state
,
841 struct bpf_verifier_state
*parent
,
844 struct bpf_verifier_state
*tmp
= NULL
;
846 /* 'parent' could be a state of caller and
847 * 'state' could be a state of callee. In such case
848 * parent->curframe < state->curframe
849 * and it's ok for r1 - r5 registers
851 * 'parent' could be a callee's state after it bpf_exit-ed.
852 * In such case parent->curframe > state->curframe
853 * and it's ok for r0 only
855 if (parent
->curframe
== state
->curframe
||
856 (parent
->curframe
< state
->curframe
&&
857 regno
>= BPF_REG_1
&& regno
<= BPF_REG_5
) ||
858 (parent
->curframe
> state
->curframe
&&
862 if (parent
->curframe
> state
->curframe
&&
863 regno
>= BPF_REG_6
) {
864 /* for callee saved regs we have to skip the whole chain
865 * of states that belong to callee and mark as LIVE_READ
866 * the registers before the call
869 while (tmp
&& tmp
->curframe
!= state
->curframe
) {
880 verbose(env
, "verifier bug regno %d tmp %p\n", regno
, tmp
);
881 verbose(env
, "regno %d parent frame %d current frame %d\n",
882 regno
, parent
->curframe
, state
->curframe
);
886 static int mark_reg_read(struct bpf_verifier_env
*env
,
887 const struct bpf_verifier_state
*state
,
888 struct bpf_verifier_state
*parent
,
891 bool writes
= parent
== state
->parent
; /* Observe write marks */
893 if (regno
== BPF_REG_FP
)
894 /* We don't need to worry about FP liveness because it's read-only */
898 /* if read wasn't screened by an earlier write ... */
899 if (writes
&& state
->frame
[state
->curframe
]->regs
[regno
].live
& REG_LIVE_WRITTEN
)
901 parent
= skip_callee(env
, state
, parent
, regno
);
904 /* ... then we depend on parent's value */
905 parent
->frame
[parent
->curframe
]->regs
[regno
].live
|= REG_LIVE_READ
;
907 parent
= state
->parent
;
913 static int check_reg_arg(struct bpf_verifier_env
*env
, u32 regno
,
916 struct bpf_verifier_state
*vstate
= env
->cur_state
;
917 struct bpf_func_state
*state
= vstate
->frame
[vstate
->curframe
];
918 struct bpf_reg_state
*regs
= state
->regs
;
920 if (regno
>= MAX_BPF_REG
) {
921 verbose(env
, "R%d is invalid\n", regno
);
926 /* check whether register used as source operand can be read */
927 if (regs
[regno
].type
== NOT_INIT
) {
928 verbose(env
, "R%d !read_ok\n", regno
);
931 return mark_reg_read(env
, vstate
, vstate
->parent
, regno
);
933 /* check whether register used as dest operand can be written to */
934 if (regno
== BPF_REG_FP
) {
935 verbose(env
, "frame pointer is read only\n");
938 regs
[regno
].live
|= REG_LIVE_WRITTEN
;
940 mark_reg_unknown(env
, regs
, regno
);
945 static bool is_spillable_regtype(enum bpf_reg_type type
)
948 case PTR_TO_MAP_VALUE
:
949 case PTR_TO_MAP_VALUE_OR_NULL
:
953 case PTR_TO_PACKET_META
:
954 case PTR_TO_PACKET_END
:
955 case CONST_PTR_TO_MAP
:
962 /* Does this register contain a constant zero? */
963 static bool register_is_null(struct bpf_reg_state
*reg
)
965 return reg
->type
== SCALAR_VALUE
&& tnum_equals_const(reg
->var_off
, 0);
968 /* check_stack_read/write functions track spill/fill of registers,
969 * stack boundary and alignment are checked in check_mem_access()
971 static int check_stack_write(struct bpf_verifier_env
*env
,
972 struct bpf_func_state
*state
, /* func where register points to */
973 int off
, int size
, int value_regno
)
975 struct bpf_func_state
*cur
; /* state of the current function */
976 int i
, slot
= -off
- 1, spi
= slot
/ BPF_REG_SIZE
, err
;
977 enum bpf_reg_type type
;
979 err
= realloc_func_state(state
, round_up(slot
+ 1, BPF_REG_SIZE
),
983 /* caller checked that off % size == 0 and -MAX_BPF_STACK <= off < 0,
984 * so it's aligned access and [off, off + size) are within stack limits
986 if (!env
->allow_ptr_leaks
&&
987 state
->stack
[spi
].slot_type
[0] == STACK_SPILL
&&
988 size
!= BPF_REG_SIZE
) {
989 verbose(env
, "attempt to corrupt spilled pointer on stack\n");
993 cur
= env
->cur_state
->frame
[env
->cur_state
->curframe
];
994 if (value_regno
>= 0 &&
995 is_spillable_regtype((type
= cur
->regs
[value_regno
].type
))) {
997 /* register containing pointer is being spilled into stack */
998 if (size
!= BPF_REG_SIZE
) {
999 verbose(env
, "invalid size of register spill\n");
1003 if (state
!= cur
&& type
== PTR_TO_STACK
) {
1004 verbose(env
, "cannot spill pointers to stack into stack frame of the caller\n");
1008 /* save register state */
1009 state
->stack
[spi
].spilled_ptr
= cur
->regs
[value_regno
];
1010 state
->stack
[spi
].spilled_ptr
.live
|= REG_LIVE_WRITTEN
;
1012 for (i
= 0; i
< BPF_REG_SIZE
; i
++)
1013 state
->stack
[spi
].slot_type
[i
] = STACK_SPILL
;
1015 u8 type
= STACK_MISC
;
1017 /* regular write of data into stack */
1018 state
->stack
[spi
].spilled_ptr
= (struct bpf_reg_state
) {};
1020 /* only mark the slot as written if all 8 bytes were written
1021 * otherwise read propagation may incorrectly stop too soon
1022 * when stack slots are partially written.
1023 * This heuristic means that read propagation will be
1024 * conservative, since it will add reg_live_read marks
1025 * to stack slots all the way to first state when programs
1026 * writes+reads less than 8 bytes
1028 if (size
== BPF_REG_SIZE
)
1029 state
->stack
[spi
].spilled_ptr
.live
|= REG_LIVE_WRITTEN
;
1031 /* when we zero initialize stack slots mark them as such */
1032 if (value_regno
>= 0 &&
1033 register_is_null(&cur
->regs
[value_regno
]))
1036 for (i
= 0; i
< size
; i
++)
1037 state
->stack
[spi
].slot_type
[(slot
- i
) % BPF_REG_SIZE
] =
1043 /* registers of every function are unique and mark_reg_read() propagates
1044 * the liveness in the following cases:
1045 * - from callee into caller for R1 - R5 that were used as arguments
1046 * - from caller into callee for R0 that used as result of the call
1047 * - from caller to the same caller skipping states of the callee for R6 - R9,
1048 * since R6 - R9 are callee saved by implicit function prologue and
1049 * caller's R6 != callee's R6, so when we propagate liveness up to
1050 * parent states we need to skip callee states for R6 - R9.
1052 * stack slot marking is different, since stacks of caller and callee are
1053 * accessible in both (since caller can pass a pointer to caller's stack to
1054 * callee which can pass it to another function), hence mark_stack_slot_read()
1055 * has to propagate the stack liveness to all parent states at given frame number.
1065 * First *ptr is reading from f1's stack and mark_stack_slot_read() has
1066 * to mark liveness at the f1's frame and not f2's frame.
1067 * Second *ptr is also reading from f1's stack and mark_stack_slot_read() has
1068 * to propagate liveness to f2 states at f1's frame level and further into
1069 * f1 states at f1's frame level until write into that stack slot
1071 static void mark_stack_slot_read(struct bpf_verifier_env
*env
,
1072 const struct bpf_verifier_state
*state
,
1073 struct bpf_verifier_state
*parent
,
1074 int slot
, int frameno
)
1076 bool writes
= parent
== state
->parent
; /* Observe write marks */
1079 if (parent
->frame
[frameno
]->allocated_stack
<= slot
* BPF_REG_SIZE
)
1080 /* since LIVE_WRITTEN mark is only done for full 8-byte
1081 * write the read marks are conservative and parent
1082 * state may not even have the stack allocated. In such case
1083 * end the propagation, since the loop reached beginning
1087 /* if read wasn't screened by an earlier write ... */
1088 if (writes
&& state
->frame
[frameno
]->stack
[slot
].spilled_ptr
.live
& REG_LIVE_WRITTEN
)
1090 /* ... then we depend on parent's value */
1091 parent
->frame
[frameno
]->stack
[slot
].spilled_ptr
.live
|= REG_LIVE_READ
;
1093 parent
= state
->parent
;
1098 static int check_stack_read(struct bpf_verifier_env
*env
,
1099 struct bpf_func_state
*reg_state
/* func where register points to */,
1100 int off
, int size
, int value_regno
)
1102 struct bpf_verifier_state
*vstate
= env
->cur_state
;
1103 struct bpf_func_state
*state
= vstate
->frame
[vstate
->curframe
];
1104 int i
, slot
= -off
- 1, spi
= slot
/ BPF_REG_SIZE
;
1107 if (reg_state
->allocated_stack
<= slot
) {
1108 verbose(env
, "invalid read from stack off %d+0 size %d\n",
1112 stype
= reg_state
->stack
[spi
].slot_type
;
1114 if (stype
[0] == STACK_SPILL
) {
1115 if (size
!= BPF_REG_SIZE
) {
1116 verbose(env
, "invalid size of register spill\n");
1119 for (i
= 1; i
< BPF_REG_SIZE
; i
++) {
1120 if (stype
[(slot
- i
) % BPF_REG_SIZE
] != STACK_SPILL
) {
1121 verbose(env
, "corrupted spill memory\n");
1126 if (value_regno
>= 0) {
1127 /* restore register state from stack */
1128 state
->regs
[value_regno
] = reg_state
->stack
[spi
].spilled_ptr
;
1129 /* mark reg as written since spilled pointer state likely
1130 * has its liveness marks cleared by is_state_visited()
1131 * which resets stack/reg liveness for state transitions
1133 state
->regs
[value_regno
].live
|= REG_LIVE_WRITTEN
;
1135 mark_stack_slot_read(env
, vstate
, vstate
->parent
, spi
,
1136 reg_state
->frameno
);
1141 for (i
= 0; i
< size
; i
++) {
1142 if (stype
[(slot
- i
) % BPF_REG_SIZE
] == STACK_MISC
)
1144 if (stype
[(slot
- i
) % BPF_REG_SIZE
] == STACK_ZERO
) {
1148 verbose(env
, "invalid read from stack off %d+%d size %d\n",
1152 mark_stack_slot_read(env
, vstate
, vstate
->parent
, spi
,
1153 reg_state
->frameno
);
1154 if (value_regno
>= 0) {
1155 if (zeros
== size
) {
1156 /* any size read into register is zero extended,
1157 * so the whole register == const_zero
1159 __mark_reg_const_zero(&state
->regs
[value_regno
]);
1161 /* have read misc data from the stack */
1162 mark_reg_unknown(env
, state
->regs
, value_regno
);
1164 state
->regs
[value_regno
].live
|= REG_LIVE_WRITTEN
;
1170 /* check read/write into map element returned by bpf_map_lookup_elem() */
1171 static int __check_map_access(struct bpf_verifier_env
*env
, u32 regno
, int off
,
1172 int size
, bool zero_size_allowed
)
1174 struct bpf_reg_state
*regs
= cur_regs(env
);
1175 struct bpf_map
*map
= regs
[regno
].map_ptr
;
1177 if (off
< 0 || size
< 0 || (size
== 0 && !zero_size_allowed
) ||
1178 off
+ size
> map
->value_size
) {
1179 verbose(env
, "invalid access to map value, value_size=%d off=%d size=%d\n",
1180 map
->value_size
, off
, size
);
1186 /* check read/write into a map element with possible variable offset */
1187 static int check_map_access(struct bpf_verifier_env
*env
, u32 regno
,
1188 int off
, int size
, bool zero_size_allowed
)
1190 struct bpf_verifier_state
*vstate
= env
->cur_state
;
1191 struct bpf_func_state
*state
= vstate
->frame
[vstate
->curframe
];
1192 struct bpf_reg_state
*reg
= &state
->regs
[regno
];
1195 /* We may have adjusted the register to this map value, so we
1196 * need to try adding each of min_value and max_value to off
1197 * to make sure our theoretical access will be safe.
1200 print_verifier_state(env
, state
);
1201 /* The minimum value is only important with signed
1202 * comparisons where we can't assume the floor of a
1203 * value is 0. If we are using signed variables for our
1204 * index'es we need to make sure that whatever we use
1205 * will have a set floor within our range.
1207 if (reg
->smin_value
< 0) {
1208 verbose(env
, "R%d min value is negative, either use unsigned index or do a if (index >=0) check.\n",
1212 err
= __check_map_access(env
, regno
, reg
->smin_value
+ off
, size
,
1215 verbose(env
, "R%d min value is outside of the array range\n",
1220 /* If we haven't set a max value then we need to bail since we can't be
1221 * sure we won't do bad things.
1222 * If reg->umax_value + off could overflow, treat that as unbounded too.
1224 if (reg
->umax_value
>= BPF_MAX_VAR_OFF
) {
1225 verbose(env
, "R%d unbounded memory access, make sure to bounds check any array access into a map\n",
1229 err
= __check_map_access(env
, regno
, reg
->umax_value
+ off
, size
,
1232 verbose(env
, "R%d max value is outside of the array range\n",
1237 #define MAX_PACKET_OFF 0xffff
1239 static bool may_access_direct_pkt_data(struct bpf_verifier_env
*env
,
1240 const struct bpf_call_arg_meta
*meta
,
1241 enum bpf_access_type t
)
1243 switch (env
->prog
->type
) {
1244 case BPF_PROG_TYPE_LWT_IN
:
1245 case BPF_PROG_TYPE_LWT_OUT
:
1246 /* dst_input() and dst_output() can't write for now */
1250 case BPF_PROG_TYPE_SCHED_CLS
:
1251 case BPF_PROG_TYPE_SCHED_ACT
:
1252 case BPF_PROG_TYPE_XDP
:
1253 case BPF_PROG_TYPE_LWT_XMIT
:
1254 case BPF_PROG_TYPE_SK_SKB
:
1256 return meta
->pkt_access
;
1258 env
->seen_direct_write
= true;
1265 static int __check_packet_access(struct bpf_verifier_env
*env
, u32 regno
,
1266 int off
, int size
, bool zero_size_allowed
)
1268 struct bpf_reg_state
*regs
= cur_regs(env
);
1269 struct bpf_reg_state
*reg
= ®s
[regno
];
1271 if (off
< 0 || size
< 0 || (size
== 0 && !zero_size_allowed
) ||
1272 (u64
)off
+ size
> reg
->range
) {
1273 verbose(env
, "invalid access to packet, off=%d size=%d, R%d(id=%d,off=%d,r=%d)\n",
1274 off
, size
, regno
, reg
->id
, reg
->off
, reg
->range
);
1280 static int check_packet_access(struct bpf_verifier_env
*env
, u32 regno
, int off
,
1281 int size
, bool zero_size_allowed
)
1283 struct bpf_reg_state
*regs
= cur_regs(env
);
1284 struct bpf_reg_state
*reg
= ®s
[regno
];
1287 /* We may have added a variable offset to the packet pointer; but any
1288 * reg->range we have comes after that. We are only checking the fixed
1292 /* We don't allow negative numbers, because we aren't tracking enough
1293 * detail to prove they're safe.
1295 if (reg
->smin_value
< 0) {
1296 verbose(env
, "R%d min value is negative, either use unsigned index or do a if (index >=0) check.\n",
1300 err
= __check_packet_access(env
, regno
, off
, size
, zero_size_allowed
);
1302 verbose(env
, "R%d offset is outside of the packet\n", regno
);
1308 /* check access to 'struct bpf_context' fields. Supports fixed offsets only */
1309 static int check_ctx_access(struct bpf_verifier_env
*env
, int insn_idx
, int off
, int size
,
1310 enum bpf_access_type t
, enum bpf_reg_type
*reg_type
)
1312 struct bpf_insn_access_aux info
= {
1313 .reg_type
= *reg_type
,
1316 if (env
->ops
->is_valid_access
&&
1317 env
->ops
->is_valid_access(off
, size
, t
, &info
)) {
1318 /* A non zero info.ctx_field_size indicates that this field is a
1319 * candidate for later verifier transformation to load the whole
1320 * field and then apply a mask when accessed with a narrower
1321 * access than actual ctx access size. A zero info.ctx_field_size
1322 * will only allow for whole field access and rejects any other
1323 * type of narrower access.
1325 *reg_type
= info
.reg_type
;
1327 env
->insn_aux_data
[insn_idx
].ctx_field_size
= info
.ctx_field_size
;
1328 /* remember the offset of last byte accessed in ctx */
1329 if (env
->prog
->aux
->max_ctx_offset
< off
+ size
)
1330 env
->prog
->aux
->max_ctx_offset
= off
+ size
;
1334 verbose(env
, "invalid bpf_context access off=%d size=%d\n", off
, size
);
1338 static bool __is_pointer_value(bool allow_ptr_leaks
,
1339 const struct bpf_reg_state
*reg
)
1341 if (allow_ptr_leaks
)
1344 return reg
->type
!= SCALAR_VALUE
;
1347 static bool is_pointer_value(struct bpf_verifier_env
*env
, int regno
)
1349 return __is_pointer_value(env
->allow_ptr_leaks
, cur_regs(env
) + regno
);
1352 static bool is_ctx_reg(struct bpf_verifier_env
*env
, int regno
)
1354 const struct bpf_reg_state
*reg
= cur_regs(env
) + regno
;
1356 return reg
->type
== PTR_TO_CTX
;
1359 static int check_pkt_ptr_alignment(struct bpf_verifier_env
*env
,
1360 const struct bpf_reg_state
*reg
,
1361 int off
, int size
, bool strict
)
1363 struct tnum reg_off
;
1366 /* Byte size accesses are always allowed. */
1367 if (!strict
|| size
== 1)
1370 /* For platforms that do not have a Kconfig enabling
1371 * CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS the value of
1372 * NET_IP_ALIGN is universally set to '2'. And on platforms
1373 * that do set CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS, we get
1374 * to this code only in strict mode where we want to emulate
1375 * the NET_IP_ALIGN==2 checking. Therefore use an
1376 * unconditional IP align value of '2'.
1380 reg_off
= tnum_add(reg
->var_off
, tnum_const(ip_align
+ reg
->off
+ off
));
1381 if (!tnum_is_aligned(reg_off
, size
)) {
1384 tnum_strn(tn_buf
, sizeof(tn_buf
), reg
->var_off
);
1386 "misaligned packet access off %d+%s+%d+%d size %d\n",
1387 ip_align
, tn_buf
, reg
->off
, off
, size
);
1394 static int check_generic_ptr_alignment(struct bpf_verifier_env
*env
,
1395 const struct bpf_reg_state
*reg
,
1396 const char *pointer_desc
,
1397 int off
, int size
, bool strict
)
1399 struct tnum reg_off
;
1401 /* Byte size accesses are always allowed. */
1402 if (!strict
|| size
== 1)
1405 reg_off
= tnum_add(reg
->var_off
, tnum_const(reg
->off
+ off
));
1406 if (!tnum_is_aligned(reg_off
, size
)) {
1409 tnum_strn(tn_buf
, sizeof(tn_buf
), reg
->var_off
);
1410 verbose(env
, "misaligned %saccess off %s+%d+%d size %d\n",
1411 pointer_desc
, tn_buf
, reg
->off
, off
, size
);
1418 static int check_ptr_alignment(struct bpf_verifier_env
*env
,
1419 const struct bpf_reg_state
*reg
,
1422 bool strict
= env
->strict_alignment
;
1423 const char *pointer_desc
= "";
1425 switch (reg
->type
) {
1427 case PTR_TO_PACKET_META
:
1428 /* Special case, because of NET_IP_ALIGN. Given metadata sits
1429 * right in front, treat it the very same way.
1431 return check_pkt_ptr_alignment(env
, reg
, off
, size
, strict
);
1432 case PTR_TO_MAP_VALUE
:
1433 pointer_desc
= "value ";
1436 pointer_desc
= "context ";
1439 pointer_desc
= "stack ";
1440 /* The stack spill tracking logic in check_stack_write()
1441 * and check_stack_read() relies on stack accesses being
1449 return check_generic_ptr_alignment(env
, reg
, pointer_desc
, off
, size
,
1453 static int update_stack_depth(struct bpf_verifier_env
*env
,
1454 const struct bpf_func_state
*func
,
1457 u16 stack
= env
->subprog_stack_depth
[func
->subprogno
];
1462 /* update known max for given subprogram */
1463 env
->subprog_stack_depth
[func
->subprogno
] = -off
;
1467 /* starting from main bpf function walk all instructions of the function
1468 * and recursively walk all callees that given function can call.
1469 * Ignore jump and exit insns.
1470 * Since recursion is prevented by check_cfg() this algorithm
1471 * only needs a local stack of MAX_CALL_FRAMES to remember callsites
1473 static int check_max_stack_depth(struct bpf_verifier_env
*env
)
1475 int depth
= 0, frame
= 0, subprog
= 0, i
= 0, subprog_end
;
1476 struct bpf_insn
*insn
= env
->prog
->insnsi
;
1477 int insn_cnt
= env
->prog
->len
;
1478 int ret_insn
[MAX_CALL_FRAMES
];
1479 int ret_prog
[MAX_CALL_FRAMES
];
1482 /* round up to 32-bytes, since this is granularity
1483 * of interpreter stack size
1485 depth
+= round_up(max_t(u32
, env
->subprog_stack_depth
[subprog
], 1), 32);
1486 if (depth
> MAX_BPF_STACK
) {
1487 verbose(env
, "combined stack size of %d calls is %d. Too large\n",
1492 if (env
->subprog_cnt
== subprog
)
1493 subprog_end
= insn_cnt
;
1495 subprog_end
= env
->subprog_starts
[subprog
];
1496 for (; i
< subprog_end
; i
++) {
1497 if (insn
[i
].code
!= (BPF_JMP
| BPF_CALL
))
1499 if (insn
[i
].src_reg
!= BPF_PSEUDO_CALL
)
1501 /* remember insn and function to return to */
1502 ret_insn
[frame
] = i
+ 1;
1503 ret_prog
[frame
] = subprog
;
1505 /* find the callee */
1506 i
= i
+ insn
[i
].imm
+ 1;
1507 subprog
= find_subprog(env
, i
);
1509 WARN_ONCE(1, "verifier bug. No program starts at insn %d\n",
1515 if (frame
>= MAX_CALL_FRAMES
) {
1516 WARN_ONCE(1, "verifier bug. Call stack is too deep\n");
1521 /* end of for() loop means the last insn of the 'subprog'
1522 * was reached. Doesn't matter whether it was JA or EXIT
1526 depth
-= round_up(max_t(u32
, env
->subprog_stack_depth
[subprog
], 1), 32);
1528 i
= ret_insn
[frame
];
1529 subprog
= ret_prog
[frame
];
1533 #ifndef CONFIG_BPF_JIT_ALWAYS_ON
1534 static int get_callee_stack_depth(struct bpf_verifier_env
*env
,
1535 const struct bpf_insn
*insn
, int idx
)
1537 int start
= idx
+ insn
->imm
+ 1, subprog
;
1539 subprog
= find_subprog(env
, start
);
1541 WARN_ONCE(1, "verifier bug. No program starts at insn %d\n",
1546 return env
->subprog_stack_depth
[subprog
];
1550 /* truncate register to smaller size (in bytes)
1551 * must be called with size < BPF_REG_SIZE
1553 static void coerce_reg_to_size(struct bpf_reg_state
*reg
, int size
)
1557 /* clear high bits in bit representation */
1558 reg
->var_off
= tnum_cast(reg
->var_off
, size
);
1560 /* fix arithmetic bounds */
1561 mask
= ((u64
)1 << (size
* 8)) - 1;
1562 if ((reg
->umin_value
& ~mask
) == (reg
->umax_value
& ~mask
)) {
1563 reg
->umin_value
&= mask
;
1564 reg
->umax_value
&= mask
;
1566 reg
->umin_value
= 0;
1567 reg
->umax_value
= mask
;
1569 reg
->smin_value
= reg
->umin_value
;
1570 reg
->smax_value
= reg
->umax_value
;
1573 /* check whether memory at (regno + off) is accessible for t = (read | write)
1574 * if t==write, value_regno is a register which value is stored into memory
1575 * if t==read, value_regno is a register which will receive the value from memory
1576 * if t==write && value_regno==-1, some unknown value is stored into memory
1577 * if t==read && value_regno==-1, don't care what we read from memory
1579 static int check_mem_access(struct bpf_verifier_env
*env
, int insn_idx
, u32 regno
, int off
,
1580 int bpf_size
, enum bpf_access_type t
,
1583 struct bpf_reg_state
*regs
= cur_regs(env
);
1584 struct bpf_reg_state
*reg
= regs
+ regno
;
1585 struct bpf_func_state
*state
;
1588 size
= bpf_size_to_bytes(bpf_size
);
1592 /* alignment checks will add in reg->off themselves */
1593 err
= check_ptr_alignment(env
, reg
, off
, size
);
1597 /* for access checks, reg->off is just part of off */
1600 if (reg
->type
== PTR_TO_MAP_VALUE
) {
1601 if (t
== BPF_WRITE
&& value_regno
>= 0 &&
1602 is_pointer_value(env
, value_regno
)) {
1603 verbose(env
, "R%d leaks addr into map\n", value_regno
);
1607 err
= check_map_access(env
, regno
, off
, size
, false);
1608 if (!err
&& t
== BPF_READ
&& value_regno
>= 0)
1609 mark_reg_unknown(env
, regs
, value_regno
);
1611 } else if (reg
->type
== PTR_TO_CTX
) {
1612 enum bpf_reg_type reg_type
= SCALAR_VALUE
;
1614 if (t
== BPF_WRITE
&& value_regno
>= 0 &&
1615 is_pointer_value(env
, value_regno
)) {
1616 verbose(env
, "R%d leaks addr into ctx\n", value_regno
);
1619 /* ctx accesses must be at a fixed offset, so that we can
1620 * determine what type of data were returned.
1624 "dereference of modified ctx ptr R%d off=%d+%d, ctx+const is allowed, ctx+const+const is not\n",
1625 regno
, reg
->off
, off
- reg
->off
);
1628 if (!tnum_is_const(reg
->var_off
) || reg
->var_off
.value
) {
1631 tnum_strn(tn_buf
, sizeof(tn_buf
), reg
->var_off
);
1633 "variable ctx access var_off=%s off=%d size=%d",
1637 err
= check_ctx_access(env
, insn_idx
, off
, size
, t
, ®_type
);
1638 if (!err
&& t
== BPF_READ
&& value_regno
>= 0) {
1639 /* ctx access returns either a scalar, or a
1640 * PTR_TO_PACKET[_META,_END]. In the latter
1641 * case, we know the offset is zero.
1643 if (reg_type
== SCALAR_VALUE
)
1644 mark_reg_unknown(env
, regs
, value_regno
);
1646 mark_reg_known_zero(env
, regs
,
1648 regs
[value_regno
].id
= 0;
1649 regs
[value_regno
].off
= 0;
1650 regs
[value_regno
].range
= 0;
1651 regs
[value_regno
].type
= reg_type
;
1654 } else if (reg
->type
== PTR_TO_STACK
) {
1655 /* stack accesses must be at a fixed offset, so that we can
1656 * determine what type of data were returned.
1657 * See check_stack_read().
1659 if (!tnum_is_const(reg
->var_off
)) {
1662 tnum_strn(tn_buf
, sizeof(tn_buf
), reg
->var_off
);
1663 verbose(env
, "variable stack access var_off=%s off=%d size=%d",
1667 off
+= reg
->var_off
.value
;
1668 if (off
>= 0 || off
< -MAX_BPF_STACK
) {
1669 verbose(env
, "invalid stack off=%d size=%d\n", off
,
1674 state
= func(env
, reg
);
1675 err
= update_stack_depth(env
, state
, off
);
1680 err
= check_stack_write(env
, state
, off
, size
,
1683 err
= check_stack_read(env
, state
, off
, size
,
1685 } else if (reg_is_pkt_pointer(reg
)) {
1686 if (t
== BPF_WRITE
&& !may_access_direct_pkt_data(env
, NULL
, t
)) {
1687 verbose(env
, "cannot write into packet\n");
1690 if (t
== BPF_WRITE
&& value_regno
>= 0 &&
1691 is_pointer_value(env
, value_regno
)) {
1692 verbose(env
, "R%d leaks addr into packet\n",
1696 err
= check_packet_access(env
, regno
, off
, size
, false);
1697 if (!err
&& t
== BPF_READ
&& value_regno
>= 0)
1698 mark_reg_unknown(env
, regs
, value_regno
);
1700 verbose(env
, "R%d invalid mem access '%s'\n", regno
,
1701 reg_type_str
[reg
->type
]);
1705 if (!err
&& size
< BPF_REG_SIZE
&& value_regno
>= 0 && t
== BPF_READ
&&
1706 regs
[value_regno
].type
== SCALAR_VALUE
) {
1707 /* b/h/w load zero-extends, mark upper bits as known 0 */
1708 coerce_reg_to_size(®s
[value_regno
], size
);
1713 static int check_xadd(struct bpf_verifier_env
*env
, int insn_idx
, struct bpf_insn
*insn
)
1717 if ((BPF_SIZE(insn
->code
) != BPF_W
&& BPF_SIZE(insn
->code
) != BPF_DW
) ||
1719 verbose(env
, "BPF_XADD uses reserved fields\n");
1723 /* check src1 operand */
1724 err
= check_reg_arg(env
, insn
->src_reg
, SRC_OP
);
1728 /* check src2 operand */
1729 err
= check_reg_arg(env
, insn
->dst_reg
, SRC_OP
);
1733 if (is_pointer_value(env
, insn
->src_reg
)) {
1734 verbose(env
, "R%d leaks addr into mem\n", insn
->src_reg
);
1738 if (is_ctx_reg(env
, insn
->dst_reg
)) {
1739 verbose(env
, "BPF_XADD stores into R%d context is not allowed\n",
1744 /* check whether atomic_add can read the memory */
1745 err
= check_mem_access(env
, insn_idx
, insn
->dst_reg
, insn
->off
,
1746 BPF_SIZE(insn
->code
), BPF_READ
, -1);
1750 /* check whether atomic_add can write into the same memory */
1751 return check_mem_access(env
, insn_idx
, insn
->dst_reg
, insn
->off
,
1752 BPF_SIZE(insn
->code
), BPF_WRITE
, -1);
1755 /* when register 'regno' is passed into function that will read 'access_size'
1756 * bytes from that pointer, make sure that it's within stack boundary
1757 * and all elements of stack are initialized.
1758 * Unlike most pointer bounds-checking functions, this one doesn't take an
1759 * 'off' argument, so it has to add in reg->off itself.
1761 static int check_stack_boundary(struct bpf_verifier_env
*env
, int regno
,
1762 int access_size
, bool zero_size_allowed
,
1763 struct bpf_call_arg_meta
*meta
)
1765 struct bpf_reg_state
*reg
= cur_regs(env
) + regno
;
1766 struct bpf_func_state
*state
= func(env
, reg
);
1767 int off
, i
, slot
, spi
;
1769 if (reg
->type
!= PTR_TO_STACK
) {
1770 /* Allow zero-byte read from NULL, regardless of pointer type */
1771 if (zero_size_allowed
&& access_size
== 0 &&
1772 register_is_null(reg
))
1775 verbose(env
, "R%d type=%s expected=%s\n", regno
,
1776 reg_type_str
[reg
->type
],
1777 reg_type_str
[PTR_TO_STACK
]);
1781 /* Only allow fixed-offset stack reads */
1782 if (!tnum_is_const(reg
->var_off
)) {
1785 tnum_strn(tn_buf
, sizeof(tn_buf
), reg
->var_off
);
1786 verbose(env
, "invalid variable stack read R%d var_off=%s\n",
1790 off
= reg
->off
+ reg
->var_off
.value
;
1791 if (off
>= 0 || off
< -MAX_BPF_STACK
|| off
+ access_size
> 0 ||
1792 access_size
< 0 || (access_size
== 0 && !zero_size_allowed
)) {
1793 verbose(env
, "invalid stack type R%d off=%d access_size=%d\n",
1794 regno
, off
, access_size
);
1798 if (meta
&& meta
->raw_mode
) {
1799 meta
->access_size
= access_size
;
1800 meta
->regno
= regno
;
1804 for (i
= 0; i
< access_size
; i
++) {
1807 slot
= -(off
+ i
) - 1;
1808 spi
= slot
/ BPF_REG_SIZE
;
1809 if (state
->allocated_stack
<= slot
)
1811 stype
= &state
->stack
[spi
].slot_type
[slot
% BPF_REG_SIZE
];
1812 if (*stype
== STACK_MISC
)
1814 if (*stype
== STACK_ZERO
) {
1815 /* helper can write anything into the stack */
1816 *stype
= STACK_MISC
;
1820 verbose(env
, "invalid indirect read from stack off %d+%d size %d\n",
1821 off
, i
, access_size
);
1824 /* reading any byte out of 8-byte 'spill_slot' will cause
1825 * the whole slot to be marked as 'read'
1827 mark_stack_slot_read(env
, env
->cur_state
, env
->cur_state
->parent
,
1828 spi
, state
->frameno
);
1830 return update_stack_depth(env
, state
, off
);
1833 static int check_helper_mem_access(struct bpf_verifier_env
*env
, int regno
,
1834 int access_size
, bool zero_size_allowed
,
1835 struct bpf_call_arg_meta
*meta
)
1837 struct bpf_reg_state
*regs
= cur_regs(env
), *reg
= ®s
[regno
];
1839 switch (reg
->type
) {
1841 case PTR_TO_PACKET_META
:
1842 return check_packet_access(env
, regno
, reg
->off
, access_size
,
1844 case PTR_TO_MAP_VALUE
:
1845 return check_map_access(env
, regno
, reg
->off
, access_size
,
1847 default: /* scalar_value|ptr_to_stack or invalid ptr */
1848 return check_stack_boundary(env
, regno
, access_size
,
1849 zero_size_allowed
, meta
);
1853 static int check_func_arg(struct bpf_verifier_env
*env
, u32 regno
,
1854 enum bpf_arg_type arg_type
,
1855 struct bpf_call_arg_meta
*meta
)
1857 struct bpf_reg_state
*regs
= cur_regs(env
), *reg
= ®s
[regno
];
1858 enum bpf_reg_type expected_type
, type
= reg
->type
;
1861 if (arg_type
== ARG_DONTCARE
)
1864 err
= check_reg_arg(env
, regno
, SRC_OP
);
1868 if (arg_type
== ARG_ANYTHING
) {
1869 if (is_pointer_value(env
, regno
)) {
1870 verbose(env
, "R%d leaks addr into helper function\n",
1877 if (type_is_pkt_pointer(type
) &&
1878 !may_access_direct_pkt_data(env
, meta
, BPF_READ
)) {
1879 verbose(env
, "helper access to the packet is not allowed\n");
1883 if (arg_type
== ARG_PTR_TO_MAP_KEY
||
1884 arg_type
== ARG_PTR_TO_MAP_VALUE
) {
1885 expected_type
= PTR_TO_STACK
;
1886 if (!type_is_pkt_pointer(type
) &&
1887 type
!= expected_type
)
1889 } else if (arg_type
== ARG_CONST_SIZE
||
1890 arg_type
== ARG_CONST_SIZE_OR_ZERO
) {
1891 expected_type
= SCALAR_VALUE
;
1892 if (type
!= expected_type
)
1894 } else if (arg_type
== ARG_CONST_MAP_PTR
) {
1895 expected_type
= CONST_PTR_TO_MAP
;
1896 if (type
!= expected_type
)
1898 } else if (arg_type
== ARG_PTR_TO_CTX
) {
1899 expected_type
= PTR_TO_CTX
;
1900 if (type
!= expected_type
)
1902 } else if (arg_type
== ARG_PTR_TO_MEM
||
1903 arg_type
== ARG_PTR_TO_MEM_OR_NULL
||
1904 arg_type
== ARG_PTR_TO_UNINIT_MEM
) {
1905 expected_type
= PTR_TO_STACK
;
1906 /* One exception here. In case function allows for NULL to be
1907 * passed in as argument, it's a SCALAR_VALUE type. Final test
1908 * happens during stack boundary checking.
1910 if (register_is_null(reg
) &&
1911 arg_type
== ARG_PTR_TO_MEM_OR_NULL
)
1912 /* final test in check_stack_boundary() */;
1913 else if (!type_is_pkt_pointer(type
) &&
1914 type
!= PTR_TO_MAP_VALUE
&&
1915 type
!= expected_type
)
1917 meta
->raw_mode
= arg_type
== ARG_PTR_TO_UNINIT_MEM
;
1919 verbose(env
, "unsupported arg_type %d\n", arg_type
);
1923 if (arg_type
== ARG_CONST_MAP_PTR
) {
1924 /* bpf_map_xxx(map_ptr) call: remember that map_ptr */
1925 meta
->map_ptr
= reg
->map_ptr
;
1926 } else if (arg_type
== ARG_PTR_TO_MAP_KEY
) {
1927 /* bpf_map_xxx(..., map_ptr, ..., key) call:
1928 * check that [key, key + map->key_size) are within
1929 * stack limits and initialized
1931 if (!meta
->map_ptr
) {
1932 /* in function declaration map_ptr must come before
1933 * map_key, so that it's verified and known before
1934 * we have to check map_key here. Otherwise it means
1935 * that kernel subsystem misconfigured verifier
1937 verbose(env
, "invalid map_ptr to access map->key\n");
1940 if (type_is_pkt_pointer(type
))
1941 err
= check_packet_access(env
, regno
, reg
->off
,
1942 meta
->map_ptr
->key_size
,
1945 err
= check_stack_boundary(env
, regno
,
1946 meta
->map_ptr
->key_size
,
1948 } else if (arg_type
== ARG_PTR_TO_MAP_VALUE
) {
1949 /* bpf_map_xxx(..., map_ptr, ..., value) call:
1950 * check [value, value + map->value_size) validity
1952 if (!meta
->map_ptr
) {
1953 /* kernel subsystem misconfigured verifier */
1954 verbose(env
, "invalid map_ptr to access map->value\n");
1957 if (type_is_pkt_pointer(type
))
1958 err
= check_packet_access(env
, regno
, reg
->off
,
1959 meta
->map_ptr
->value_size
,
1962 err
= check_stack_boundary(env
, regno
,
1963 meta
->map_ptr
->value_size
,
1965 } else if (arg_type
== ARG_CONST_SIZE
||
1966 arg_type
== ARG_CONST_SIZE_OR_ZERO
) {
1967 bool zero_size_allowed
= (arg_type
== ARG_CONST_SIZE_OR_ZERO
);
1969 /* bpf_xxx(..., buf, len) call will access 'len' bytes
1970 * from stack pointer 'buf'. Check it
1971 * note: regno == len, regno - 1 == buf
1974 /* kernel subsystem misconfigured verifier */
1976 "ARG_CONST_SIZE cannot be first argument\n");
1980 /* The register is SCALAR_VALUE; the access check
1981 * happens using its boundaries.
1984 if (!tnum_is_const(reg
->var_off
))
1985 /* For unprivileged variable accesses, disable raw
1986 * mode so that the program is required to
1987 * initialize all the memory that the helper could
1988 * just partially fill up.
1992 if (reg
->smin_value
< 0) {
1993 verbose(env
, "R%d min value is negative, either use unsigned or 'var &= const'\n",
1998 if (reg
->umin_value
== 0) {
1999 err
= check_helper_mem_access(env
, regno
- 1, 0,
2006 if (reg
->umax_value
>= BPF_MAX_VAR_SIZ
) {
2007 verbose(env
, "R%d unbounded memory access, use 'var &= const' or 'if (var < const)'\n",
2011 err
= check_helper_mem_access(env
, regno
- 1,
2013 zero_size_allowed
, meta
);
2018 verbose(env
, "R%d type=%s expected=%s\n", regno
,
2019 reg_type_str
[type
], reg_type_str
[expected_type
]);
2023 static int check_map_func_compatibility(struct bpf_verifier_env
*env
,
2024 struct bpf_map
*map
, int func_id
)
2029 /* We need a two way check, first is from map perspective ... */
2030 switch (map
->map_type
) {
2031 case BPF_MAP_TYPE_PROG_ARRAY
:
2032 if (func_id
!= BPF_FUNC_tail_call
)
2035 case BPF_MAP_TYPE_PERF_EVENT_ARRAY
:
2036 if (func_id
!= BPF_FUNC_perf_event_read
&&
2037 func_id
!= BPF_FUNC_perf_event_output
&&
2038 func_id
!= BPF_FUNC_perf_event_read_value
)
2041 case BPF_MAP_TYPE_STACK_TRACE
:
2042 if (func_id
!= BPF_FUNC_get_stackid
)
2045 case BPF_MAP_TYPE_CGROUP_ARRAY
:
2046 if (func_id
!= BPF_FUNC_skb_under_cgroup
&&
2047 func_id
!= BPF_FUNC_current_task_under_cgroup
)
2050 /* devmap returns a pointer to a live net_device ifindex that we cannot
2051 * allow to be modified from bpf side. So do not allow lookup elements
2054 case BPF_MAP_TYPE_DEVMAP
:
2055 if (func_id
!= BPF_FUNC_redirect_map
)
2058 /* Restrict bpf side of cpumap, open when use-cases appear */
2059 case BPF_MAP_TYPE_CPUMAP
:
2060 if (func_id
!= BPF_FUNC_redirect_map
)
2063 case BPF_MAP_TYPE_ARRAY_OF_MAPS
:
2064 case BPF_MAP_TYPE_HASH_OF_MAPS
:
2065 if (func_id
!= BPF_FUNC_map_lookup_elem
)
2068 case BPF_MAP_TYPE_SOCKMAP
:
2069 if (func_id
!= BPF_FUNC_sk_redirect_map
&&
2070 func_id
!= BPF_FUNC_sock_map_update
&&
2071 func_id
!= BPF_FUNC_map_delete_elem
)
2078 /* ... and second from the function itself. */
2080 case BPF_FUNC_tail_call
:
2081 if (map
->map_type
!= BPF_MAP_TYPE_PROG_ARRAY
)
2083 if (env
->subprog_cnt
) {
2084 verbose(env
, "tail_calls are not allowed in programs with bpf-to-bpf calls\n");
2088 case BPF_FUNC_perf_event_read
:
2089 case BPF_FUNC_perf_event_output
:
2090 case BPF_FUNC_perf_event_read_value
:
2091 if (map
->map_type
!= BPF_MAP_TYPE_PERF_EVENT_ARRAY
)
2094 case BPF_FUNC_get_stackid
:
2095 if (map
->map_type
!= BPF_MAP_TYPE_STACK_TRACE
)
2098 case BPF_FUNC_current_task_under_cgroup
:
2099 case BPF_FUNC_skb_under_cgroup
:
2100 if (map
->map_type
!= BPF_MAP_TYPE_CGROUP_ARRAY
)
2103 case BPF_FUNC_redirect_map
:
2104 if (map
->map_type
!= BPF_MAP_TYPE_DEVMAP
&&
2105 map
->map_type
!= BPF_MAP_TYPE_CPUMAP
)
2108 case BPF_FUNC_sk_redirect_map
:
2109 if (map
->map_type
!= BPF_MAP_TYPE_SOCKMAP
)
2112 case BPF_FUNC_sock_map_update
:
2113 if (map
->map_type
!= BPF_MAP_TYPE_SOCKMAP
)
2122 verbose(env
, "cannot pass map_type %d into func %s#%d\n",
2123 map
->map_type
, func_id_name(func_id
), func_id
);
2127 static int check_raw_mode(const struct bpf_func_proto
*fn
)
2131 if (fn
->arg1_type
== ARG_PTR_TO_UNINIT_MEM
)
2133 if (fn
->arg2_type
== ARG_PTR_TO_UNINIT_MEM
)
2135 if (fn
->arg3_type
== ARG_PTR_TO_UNINIT_MEM
)
2137 if (fn
->arg4_type
== ARG_PTR_TO_UNINIT_MEM
)
2139 if (fn
->arg5_type
== ARG_PTR_TO_UNINIT_MEM
)
2142 return count
> 1 ? -EINVAL
: 0;
2145 /* Packet data might have moved, any old PTR_TO_PACKET[_META,_END]
2146 * are now invalid, so turn them into unknown SCALAR_VALUE.
2148 static void __clear_all_pkt_pointers(struct bpf_verifier_env
*env
,
2149 struct bpf_func_state
*state
)
2151 struct bpf_reg_state
*regs
= state
->regs
, *reg
;
2154 for (i
= 0; i
< MAX_BPF_REG
; i
++)
2155 if (reg_is_pkt_pointer_any(®s
[i
]))
2156 mark_reg_unknown(env
, regs
, i
);
2158 for (i
= 0; i
< state
->allocated_stack
/ BPF_REG_SIZE
; i
++) {
2159 if (state
->stack
[i
].slot_type
[0] != STACK_SPILL
)
2161 reg
= &state
->stack
[i
].spilled_ptr
;
2162 if (reg_is_pkt_pointer_any(reg
))
2163 __mark_reg_unknown(reg
);
2167 static void clear_all_pkt_pointers(struct bpf_verifier_env
*env
)
2169 struct bpf_verifier_state
*vstate
= env
->cur_state
;
2172 for (i
= 0; i
<= vstate
->curframe
; i
++)
2173 __clear_all_pkt_pointers(env
, vstate
->frame
[i
]);
2176 static int check_func_call(struct bpf_verifier_env
*env
, struct bpf_insn
*insn
,
2179 struct bpf_verifier_state
*state
= env
->cur_state
;
2180 struct bpf_func_state
*caller
, *callee
;
2181 int i
, subprog
, target_insn
;
2183 if (state
->curframe
+ 1 >= MAX_CALL_FRAMES
) {
2184 verbose(env
, "the call stack of %d frames is too deep\n",
2185 state
->curframe
+ 2);
2189 target_insn
= *insn_idx
+ insn
->imm
;
2190 subprog
= find_subprog(env
, target_insn
+ 1);
2192 verbose(env
, "verifier bug. No program starts at insn %d\n",
2197 caller
= state
->frame
[state
->curframe
];
2198 if (state
->frame
[state
->curframe
+ 1]) {
2199 verbose(env
, "verifier bug. Frame %d already allocated\n",
2200 state
->curframe
+ 1);
2204 callee
= kzalloc(sizeof(*callee
), GFP_KERNEL
);
2207 state
->frame
[state
->curframe
+ 1] = callee
;
2209 /* callee cannot access r0, r6 - r9 for reading and has to write
2210 * into its own stack before reading from it.
2211 * callee can read/write into caller's stack
2213 init_func_state(env
, callee
,
2214 /* remember the callsite, it will be used by bpf_exit */
2215 *insn_idx
/* callsite */,
2216 state
->curframe
+ 1 /* frameno within this callchain */,
2217 subprog
+ 1 /* subprog number within this prog */);
2219 /* copy r1 - r5 args that callee can access */
2220 for (i
= BPF_REG_1
; i
<= BPF_REG_5
; i
++)
2221 callee
->regs
[i
] = caller
->regs
[i
];
2223 /* after the call regsiters r0 - r5 were scratched */
2224 for (i
= 0; i
< CALLER_SAVED_REGS
; i
++) {
2225 mark_reg_not_init(env
, caller
->regs
, caller_saved
[i
]);
2226 check_reg_arg(env
, caller_saved
[i
], DST_OP_NO_MARK
);
2229 /* only increment it after check_reg_arg() finished */
2232 /* and go analyze first insn of the callee */
2233 *insn_idx
= target_insn
;
2235 if (env
->log
.level
) {
2236 verbose(env
, "caller:\n");
2237 print_verifier_state(env
, caller
);
2238 verbose(env
, "callee:\n");
2239 print_verifier_state(env
, callee
);
2244 static int prepare_func_exit(struct bpf_verifier_env
*env
, int *insn_idx
)
2246 struct bpf_verifier_state
*state
= env
->cur_state
;
2247 struct bpf_func_state
*caller
, *callee
;
2248 struct bpf_reg_state
*r0
;
2250 callee
= state
->frame
[state
->curframe
];
2251 r0
= &callee
->regs
[BPF_REG_0
];
2252 if (r0
->type
== PTR_TO_STACK
) {
2253 /* technically it's ok to return caller's stack pointer
2254 * (or caller's caller's pointer) back to the caller,
2255 * since these pointers are valid. Only current stack
2256 * pointer will be invalid as soon as function exits,
2257 * but let's be conservative
2259 verbose(env
, "cannot return stack pointer to the caller\n");
2264 caller
= state
->frame
[state
->curframe
];
2265 /* return to the caller whatever r0 had in the callee */
2266 caller
->regs
[BPF_REG_0
] = *r0
;
2268 *insn_idx
= callee
->callsite
+ 1;
2269 if (env
->log
.level
) {
2270 verbose(env
, "returning from callee:\n");
2271 print_verifier_state(env
, callee
);
2272 verbose(env
, "to caller at %d:\n", *insn_idx
);
2273 print_verifier_state(env
, caller
);
2275 /* clear everything in the callee */
2276 free_func_state(callee
);
2277 state
->frame
[state
->curframe
+ 1] = NULL
;
2281 static int check_helper_call(struct bpf_verifier_env
*env
, int func_id
, int insn_idx
)
2283 const struct bpf_func_proto
*fn
= NULL
;
2284 struct bpf_reg_state
*regs
;
2285 struct bpf_call_arg_meta meta
;
2289 /* find function prototype */
2290 if (func_id
< 0 || func_id
>= __BPF_FUNC_MAX_ID
) {
2291 verbose(env
, "invalid func %s#%d\n", func_id_name(func_id
),
2296 if (env
->ops
->get_func_proto
)
2297 fn
= env
->ops
->get_func_proto(func_id
);
2300 verbose(env
, "unknown func %s#%d\n", func_id_name(func_id
),
2305 /* eBPF programs must be GPL compatible to use GPL-ed functions */
2306 if (!env
->prog
->gpl_compatible
&& fn
->gpl_only
) {
2307 verbose(env
, "cannot call GPL only function from proprietary program\n");
2311 /* With LD_ABS/IND some JITs save/restore skb from r1. */
2312 changes_data
= bpf_helper_changes_pkt_data(fn
->func
);
2313 if (changes_data
&& fn
->arg1_type
!= ARG_PTR_TO_CTX
) {
2314 verbose(env
, "kernel subsystem misconfigured func %s#%d: r1 != ctx\n",
2315 func_id_name(func_id
), func_id
);
2319 memset(&meta
, 0, sizeof(meta
));
2320 meta
.pkt_access
= fn
->pkt_access
;
2322 /* We only support one arg being in raw mode at the moment, which
2323 * is sufficient for the helper functions we have right now.
2325 err
= check_raw_mode(fn
);
2327 verbose(env
, "kernel subsystem misconfigured func %s#%d\n",
2328 func_id_name(func_id
), func_id
);
2333 err
= check_func_arg(env
, BPF_REG_1
, fn
->arg1_type
, &meta
);
2336 err
= check_func_arg(env
, BPF_REG_2
, fn
->arg2_type
, &meta
);
2339 if (func_id
== BPF_FUNC_tail_call
) {
2340 if (meta
.map_ptr
== NULL
) {
2341 verbose(env
, "verifier bug\n");
2344 env
->insn_aux_data
[insn_idx
].map_ptr
= meta
.map_ptr
;
2346 err
= check_func_arg(env
, BPF_REG_3
, fn
->arg3_type
, &meta
);
2349 err
= check_func_arg(env
, BPF_REG_4
, fn
->arg4_type
, &meta
);
2352 err
= check_func_arg(env
, BPF_REG_5
, fn
->arg5_type
, &meta
);
2356 /* Mark slots with STACK_MISC in case of raw mode, stack offset
2357 * is inferred from register state.
2359 for (i
= 0; i
< meta
.access_size
; i
++) {
2360 err
= check_mem_access(env
, insn_idx
, meta
.regno
, i
, BPF_B
, BPF_WRITE
, -1);
2365 regs
= cur_regs(env
);
2366 /* reset caller saved regs */
2367 for (i
= 0; i
< CALLER_SAVED_REGS
; i
++) {
2368 mark_reg_not_init(env
, regs
, caller_saved
[i
]);
2369 check_reg_arg(env
, caller_saved
[i
], DST_OP_NO_MARK
);
2372 /* update return register (already marked as written above) */
2373 if (fn
->ret_type
== RET_INTEGER
) {
2374 /* sets type to SCALAR_VALUE */
2375 mark_reg_unknown(env
, regs
, BPF_REG_0
);
2376 } else if (fn
->ret_type
== RET_VOID
) {
2377 regs
[BPF_REG_0
].type
= NOT_INIT
;
2378 } else if (fn
->ret_type
== RET_PTR_TO_MAP_VALUE_OR_NULL
) {
2379 struct bpf_insn_aux_data
*insn_aux
;
2381 regs
[BPF_REG_0
].type
= PTR_TO_MAP_VALUE_OR_NULL
;
2382 /* There is no offset yet applied, variable or fixed */
2383 mark_reg_known_zero(env
, regs
, BPF_REG_0
);
2384 regs
[BPF_REG_0
].off
= 0;
2385 /* remember map_ptr, so that check_map_access()
2386 * can check 'value_size' boundary of memory access
2387 * to map element returned from bpf_map_lookup_elem()
2389 if (meta
.map_ptr
== NULL
) {
2391 "kernel subsystem misconfigured verifier\n");
2394 regs
[BPF_REG_0
].map_ptr
= meta
.map_ptr
;
2395 regs
[BPF_REG_0
].id
= ++env
->id_gen
;
2396 insn_aux
= &env
->insn_aux_data
[insn_idx
];
2397 if (!insn_aux
->map_ptr
)
2398 insn_aux
->map_ptr
= meta
.map_ptr
;
2399 else if (insn_aux
->map_ptr
!= meta
.map_ptr
)
2400 insn_aux
->map_ptr
= BPF_MAP_PTR_POISON
;
2402 verbose(env
, "unknown return type %d of func %s#%d\n",
2403 fn
->ret_type
, func_id_name(func_id
), func_id
);
2407 err
= check_map_func_compatibility(env
, meta
.map_ptr
, func_id
);
2412 clear_all_pkt_pointers(env
);
2416 static bool signed_add_overflows(s64 a
, s64 b
)
2418 /* Do the add in u64, where overflow is well-defined */
2419 s64 res
= (s64
)((u64
)a
+ (u64
)b
);
2426 static bool signed_sub_overflows(s64 a
, s64 b
)
2428 /* Do the sub in u64, where overflow is well-defined */
2429 s64 res
= (s64
)((u64
)a
- (u64
)b
);
2436 static bool check_reg_sane_offset(struct bpf_verifier_env
*env
,
2437 const struct bpf_reg_state
*reg
,
2438 enum bpf_reg_type type
)
2440 bool known
= tnum_is_const(reg
->var_off
);
2441 s64 val
= reg
->var_off
.value
;
2442 s64 smin
= reg
->smin_value
;
2444 if (known
&& (val
>= BPF_MAX_VAR_OFF
|| val
<= -BPF_MAX_VAR_OFF
)) {
2445 verbose(env
, "math between %s pointer and %lld is not allowed\n",
2446 reg_type_str
[type
], val
);
2450 if (reg
->off
>= BPF_MAX_VAR_OFF
|| reg
->off
<= -BPF_MAX_VAR_OFF
) {
2451 verbose(env
, "%s pointer offset %d is not allowed\n",
2452 reg_type_str
[type
], reg
->off
);
2456 if (smin
== S64_MIN
) {
2457 verbose(env
, "math between %s pointer and register with unbounded min value is not allowed\n",
2458 reg_type_str
[type
]);
2462 if (smin
>= BPF_MAX_VAR_OFF
|| smin
<= -BPF_MAX_VAR_OFF
) {
2463 verbose(env
, "value %lld makes %s pointer be out of bounds\n",
2464 smin
, reg_type_str
[type
]);
2471 /* Handles arithmetic on a pointer and a scalar: computes new min/max and var_off.
2472 * Caller should also handle BPF_MOV case separately.
2473 * If we return -EACCES, caller may want to try again treating pointer as a
2474 * scalar. So we only emit a diagnostic if !env->allow_ptr_leaks.
2476 static int adjust_ptr_min_max_vals(struct bpf_verifier_env
*env
,
2477 struct bpf_insn
*insn
,
2478 const struct bpf_reg_state
*ptr_reg
,
2479 const struct bpf_reg_state
*off_reg
)
2481 struct bpf_verifier_state
*vstate
= env
->cur_state
;
2482 struct bpf_func_state
*state
= vstate
->frame
[vstate
->curframe
];
2483 struct bpf_reg_state
*regs
= state
->regs
, *dst_reg
;
2484 bool known
= tnum_is_const(off_reg
->var_off
);
2485 s64 smin_val
= off_reg
->smin_value
, smax_val
= off_reg
->smax_value
,
2486 smin_ptr
= ptr_reg
->smin_value
, smax_ptr
= ptr_reg
->smax_value
;
2487 u64 umin_val
= off_reg
->umin_value
, umax_val
= off_reg
->umax_value
,
2488 umin_ptr
= ptr_reg
->umin_value
, umax_ptr
= ptr_reg
->umax_value
;
2489 u8 opcode
= BPF_OP(insn
->code
);
2490 u32 dst
= insn
->dst_reg
;
2492 dst_reg
= ®s
[dst
];
2494 if ((known
&& (smin_val
!= smax_val
|| umin_val
!= umax_val
)) ||
2495 smin_val
> smax_val
|| umin_val
> umax_val
) {
2496 /* Taint dst register if offset had invalid bounds derived from
2497 * e.g. dead branches.
2499 __mark_reg_unknown(dst_reg
);
2503 if (BPF_CLASS(insn
->code
) != BPF_ALU64
) {
2504 /* 32-bit ALU ops on pointers produce (meaningless) scalars */
2506 "R%d 32-bit pointer arithmetic prohibited\n",
2511 if (ptr_reg
->type
== PTR_TO_MAP_VALUE_OR_NULL
) {
2512 verbose(env
, "R%d pointer arithmetic on PTR_TO_MAP_VALUE_OR_NULL prohibited, null-check it first\n",
2516 if (ptr_reg
->type
== CONST_PTR_TO_MAP
) {
2517 verbose(env
, "R%d pointer arithmetic on CONST_PTR_TO_MAP prohibited\n",
2521 if (ptr_reg
->type
== PTR_TO_PACKET_END
) {
2522 verbose(env
, "R%d pointer arithmetic on PTR_TO_PACKET_END prohibited\n",
2527 /* In case of 'scalar += pointer', dst_reg inherits pointer type and id.
2528 * The id may be overwritten later if we create a new variable offset.
2530 dst_reg
->type
= ptr_reg
->type
;
2531 dst_reg
->id
= ptr_reg
->id
;
2533 if (!check_reg_sane_offset(env
, off_reg
, ptr_reg
->type
) ||
2534 !check_reg_sane_offset(env
, ptr_reg
, ptr_reg
->type
))
2539 /* We can take a fixed offset as long as it doesn't overflow
2540 * the s32 'off' field
2542 if (known
&& (ptr_reg
->off
+ smin_val
==
2543 (s64
)(s32
)(ptr_reg
->off
+ smin_val
))) {
2544 /* pointer += K. Accumulate it into fixed offset */
2545 dst_reg
->smin_value
= smin_ptr
;
2546 dst_reg
->smax_value
= smax_ptr
;
2547 dst_reg
->umin_value
= umin_ptr
;
2548 dst_reg
->umax_value
= umax_ptr
;
2549 dst_reg
->var_off
= ptr_reg
->var_off
;
2550 dst_reg
->off
= ptr_reg
->off
+ smin_val
;
2551 dst_reg
->range
= ptr_reg
->range
;
2554 /* A new variable offset is created. Note that off_reg->off
2555 * == 0, since it's a scalar.
2556 * dst_reg gets the pointer type and since some positive
2557 * integer value was added to the pointer, give it a new 'id'
2558 * if it's a PTR_TO_PACKET.
2559 * this creates a new 'base' pointer, off_reg (variable) gets
2560 * added into the variable offset, and we copy the fixed offset
2563 if (signed_add_overflows(smin_ptr
, smin_val
) ||
2564 signed_add_overflows(smax_ptr
, smax_val
)) {
2565 dst_reg
->smin_value
= S64_MIN
;
2566 dst_reg
->smax_value
= S64_MAX
;
2568 dst_reg
->smin_value
= smin_ptr
+ smin_val
;
2569 dst_reg
->smax_value
= smax_ptr
+ smax_val
;
2571 if (umin_ptr
+ umin_val
< umin_ptr
||
2572 umax_ptr
+ umax_val
< umax_ptr
) {
2573 dst_reg
->umin_value
= 0;
2574 dst_reg
->umax_value
= U64_MAX
;
2576 dst_reg
->umin_value
= umin_ptr
+ umin_val
;
2577 dst_reg
->umax_value
= umax_ptr
+ umax_val
;
2579 dst_reg
->var_off
= tnum_add(ptr_reg
->var_off
, off_reg
->var_off
);
2580 dst_reg
->off
= ptr_reg
->off
;
2581 if (reg_is_pkt_pointer(ptr_reg
)) {
2582 dst_reg
->id
= ++env
->id_gen
;
2583 /* something was added to pkt_ptr, set range to zero */
2588 if (dst_reg
== off_reg
) {
2589 /* scalar -= pointer. Creates an unknown scalar */
2590 verbose(env
, "R%d tried to subtract pointer from scalar\n",
2594 /* We don't allow subtraction from FP, because (according to
2595 * test_verifier.c test "invalid fp arithmetic", JITs might not
2596 * be able to deal with it.
2598 if (ptr_reg
->type
== PTR_TO_STACK
) {
2599 verbose(env
, "R%d subtraction from stack pointer prohibited\n",
2603 if (known
&& (ptr_reg
->off
- smin_val
==
2604 (s64
)(s32
)(ptr_reg
->off
- smin_val
))) {
2605 /* pointer -= K. Subtract it from fixed offset */
2606 dst_reg
->smin_value
= smin_ptr
;
2607 dst_reg
->smax_value
= smax_ptr
;
2608 dst_reg
->umin_value
= umin_ptr
;
2609 dst_reg
->umax_value
= umax_ptr
;
2610 dst_reg
->var_off
= ptr_reg
->var_off
;
2611 dst_reg
->id
= ptr_reg
->id
;
2612 dst_reg
->off
= ptr_reg
->off
- smin_val
;
2613 dst_reg
->range
= ptr_reg
->range
;
2616 /* A new variable offset is created. If the subtrahend is known
2617 * nonnegative, then any reg->range we had before is still good.
2619 if (signed_sub_overflows(smin_ptr
, smax_val
) ||
2620 signed_sub_overflows(smax_ptr
, smin_val
)) {
2621 /* Overflow possible, we know nothing */
2622 dst_reg
->smin_value
= S64_MIN
;
2623 dst_reg
->smax_value
= S64_MAX
;
2625 dst_reg
->smin_value
= smin_ptr
- smax_val
;
2626 dst_reg
->smax_value
= smax_ptr
- smin_val
;
2628 if (umin_ptr
< umax_val
) {
2629 /* Overflow possible, we know nothing */
2630 dst_reg
->umin_value
= 0;
2631 dst_reg
->umax_value
= U64_MAX
;
2633 /* Cannot overflow (as long as bounds are consistent) */
2634 dst_reg
->umin_value
= umin_ptr
- umax_val
;
2635 dst_reg
->umax_value
= umax_ptr
- umin_val
;
2637 dst_reg
->var_off
= tnum_sub(ptr_reg
->var_off
, off_reg
->var_off
);
2638 dst_reg
->off
= ptr_reg
->off
;
2639 if (reg_is_pkt_pointer(ptr_reg
)) {
2640 dst_reg
->id
= ++env
->id_gen
;
2641 /* something was added to pkt_ptr, set range to zero */
2649 /* bitwise ops on pointers are troublesome, prohibit. */
2650 verbose(env
, "R%d bitwise operator %s on pointer prohibited\n",
2651 dst
, bpf_alu_string
[opcode
>> 4]);
2654 /* other operators (e.g. MUL,LSH) produce non-pointer results */
2655 verbose(env
, "R%d pointer arithmetic with %s operator prohibited\n",
2656 dst
, bpf_alu_string
[opcode
>> 4]);
2660 if (!check_reg_sane_offset(env
, dst_reg
, ptr_reg
->type
))
2663 __update_reg_bounds(dst_reg
);
2664 __reg_deduce_bounds(dst_reg
);
2665 __reg_bound_offset(dst_reg
);
2669 /* WARNING: This function does calculations on 64-bit values, but the actual
2670 * execution may occur on 32-bit values. Therefore, things like bitshifts
2671 * need extra checks in the 32-bit case.
2673 static int adjust_scalar_min_max_vals(struct bpf_verifier_env
*env
,
2674 struct bpf_insn
*insn
,
2675 struct bpf_reg_state
*dst_reg
,
2676 struct bpf_reg_state src_reg
)
2678 struct bpf_reg_state
*regs
= cur_regs(env
);
2679 u8 opcode
= BPF_OP(insn
->code
);
2680 bool src_known
, dst_known
;
2681 s64 smin_val
, smax_val
;
2682 u64 umin_val
, umax_val
;
2683 u64 insn_bitness
= (BPF_CLASS(insn
->code
) == BPF_ALU64
) ? 64 : 32;
2685 smin_val
= src_reg
.smin_value
;
2686 smax_val
= src_reg
.smax_value
;
2687 umin_val
= src_reg
.umin_value
;
2688 umax_val
= src_reg
.umax_value
;
2689 src_known
= tnum_is_const(src_reg
.var_off
);
2690 dst_known
= tnum_is_const(dst_reg
->var_off
);
2692 if ((src_known
&& (smin_val
!= smax_val
|| umin_val
!= umax_val
)) ||
2693 smin_val
> smax_val
|| umin_val
> umax_val
) {
2694 /* Taint dst register if offset had invalid bounds derived from
2695 * e.g. dead branches.
2697 __mark_reg_unknown(dst_reg
);
2702 opcode
!= BPF_ADD
&& opcode
!= BPF_SUB
&& opcode
!= BPF_AND
) {
2703 __mark_reg_unknown(dst_reg
);
2709 if (signed_add_overflows(dst_reg
->smin_value
, smin_val
) ||
2710 signed_add_overflows(dst_reg
->smax_value
, smax_val
)) {
2711 dst_reg
->smin_value
= S64_MIN
;
2712 dst_reg
->smax_value
= S64_MAX
;
2714 dst_reg
->smin_value
+= smin_val
;
2715 dst_reg
->smax_value
+= smax_val
;
2717 if (dst_reg
->umin_value
+ umin_val
< umin_val
||
2718 dst_reg
->umax_value
+ umax_val
< umax_val
) {
2719 dst_reg
->umin_value
= 0;
2720 dst_reg
->umax_value
= U64_MAX
;
2722 dst_reg
->umin_value
+= umin_val
;
2723 dst_reg
->umax_value
+= umax_val
;
2725 dst_reg
->var_off
= tnum_add(dst_reg
->var_off
, src_reg
.var_off
);
2728 if (signed_sub_overflows(dst_reg
->smin_value
, smax_val
) ||
2729 signed_sub_overflows(dst_reg
->smax_value
, smin_val
)) {
2730 /* Overflow possible, we know nothing */
2731 dst_reg
->smin_value
= S64_MIN
;
2732 dst_reg
->smax_value
= S64_MAX
;
2734 dst_reg
->smin_value
-= smax_val
;
2735 dst_reg
->smax_value
-= smin_val
;
2737 if (dst_reg
->umin_value
< umax_val
) {
2738 /* Overflow possible, we know nothing */
2739 dst_reg
->umin_value
= 0;
2740 dst_reg
->umax_value
= U64_MAX
;
2742 /* Cannot overflow (as long as bounds are consistent) */
2743 dst_reg
->umin_value
-= umax_val
;
2744 dst_reg
->umax_value
-= umin_val
;
2746 dst_reg
->var_off
= tnum_sub(dst_reg
->var_off
, src_reg
.var_off
);
2749 dst_reg
->var_off
= tnum_mul(dst_reg
->var_off
, src_reg
.var_off
);
2750 if (smin_val
< 0 || dst_reg
->smin_value
< 0) {
2751 /* Ain't nobody got time to multiply that sign */
2752 __mark_reg_unbounded(dst_reg
);
2753 __update_reg_bounds(dst_reg
);
2756 /* Both values are positive, so we can work with unsigned and
2757 * copy the result to signed (unless it exceeds S64_MAX).
2759 if (umax_val
> U32_MAX
|| dst_reg
->umax_value
> U32_MAX
) {
2760 /* Potential overflow, we know nothing */
2761 __mark_reg_unbounded(dst_reg
);
2762 /* (except what we can learn from the var_off) */
2763 __update_reg_bounds(dst_reg
);
2766 dst_reg
->umin_value
*= umin_val
;
2767 dst_reg
->umax_value
*= umax_val
;
2768 if (dst_reg
->umax_value
> S64_MAX
) {
2769 /* Overflow possible, we know nothing */
2770 dst_reg
->smin_value
= S64_MIN
;
2771 dst_reg
->smax_value
= S64_MAX
;
2773 dst_reg
->smin_value
= dst_reg
->umin_value
;
2774 dst_reg
->smax_value
= dst_reg
->umax_value
;
2778 if (src_known
&& dst_known
) {
2779 __mark_reg_known(dst_reg
, dst_reg
->var_off
.value
&
2780 src_reg
.var_off
.value
);
2783 /* We get our minimum from the var_off, since that's inherently
2784 * bitwise. Our maximum is the minimum of the operands' maxima.
2786 dst_reg
->var_off
= tnum_and(dst_reg
->var_off
, src_reg
.var_off
);
2787 dst_reg
->umin_value
= dst_reg
->var_off
.value
;
2788 dst_reg
->umax_value
= min(dst_reg
->umax_value
, umax_val
);
2789 if (dst_reg
->smin_value
< 0 || smin_val
< 0) {
2790 /* Lose signed bounds when ANDing negative numbers,
2791 * ain't nobody got time for that.
2793 dst_reg
->smin_value
= S64_MIN
;
2794 dst_reg
->smax_value
= S64_MAX
;
2796 /* ANDing two positives gives a positive, so safe to
2797 * cast result into s64.
2799 dst_reg
->smin_value
= dst_reg
->umin_value
;
2800 dst_reg
->smax_value
= dst_reg
->umax_value
;
2802 /* We may learn something more from the var_off */
2803 __update_reg_bounds(dst_reg
);
2806 if (src_known
&& dst_known
) {
2807 __mark_reg_known(dst_reg
, dst_reg
->var_off
.value
|
2808 src_reg
.var_off
.value
);
2811 /* We get our maximum from the var_off, and our minimum is the
2812 * maximum of the operands' minima
2814 dst_reg
->var_off
= tnum_or(dst_reg
->var_off
, src_reg
.var_off
);
2815 dst_reg
->umin_value
= max(dst_reg
->umin_value
, umin_val
);
2816 dst_reg
->umax_value
= dst_reg
->var_off
.value
|
2817 dst_reg
->var_off
.mask
;
2818 if (dst_reg
->smin_value
< 0 || smin_val
< 0) {
2819 /* Lose signed bounds when ORing negative numbers,
2820 * ain't nobody got time for that.
2822 dst_reg
->smin_value
= S64_MIN
;
2823 dst_reg
->smax_value
= S64_MAX
;
2825 /* ORing two positives gives a positive, so safe to
2826 * cast result into s64.
2828 dst_reg
->smin_value
= dst_reg
->umin_value
;
2829 dst_reg
->smax_value
= dst_reg
->umax_value
;
2831 /* We may learn something more from the var_off */
2832 __update_reg_bounds(dst_reg
);
2835 if (umax_val
>= insn_bitness
) {
2836 /* Shifts greater than 31 or 63 are undefined.
2837 * This includes shifts by a negative number.
2839 mark_reg_unknown(env
, regs
, insn
->dst_reg
);
2842 /* We lose all sign bit information (except what we can pick
2845 dst_reg
->smin_value
= S64_MIN
;
2846 dst_reg
->smax_value
= S64_MAX
;
2847 /* If we might shift our top bit out, then we know nothing */
2848 if (dst_reg
->umax_value
> 1ULL << (63 - umax_val
)) {
2849 dst_reg
->umin_value
= 0;
2850 dst_reg
->umax_value
= U64_MAX
;
2852 dst_reg
->umin_value
<<= umin_val
;
2853 dst_reg
->umax_value
<<= umax_val
;
2856 dst_reg
->var_off
= tnum_lshift(dst_reg
->var_off
, umin_val
);
2858 dst_reg
->var_off
= tnum_lshift(tnum_unknown
, umin_val
);
2859 /* We may learn something more from the var_off */
2860 __update_reg_bounds(dst_reg
);
2863 if (umax_val
>= insn_bitness
) {
2864 /* Shifts greater than 31 or 63 are undefined.
2865 * This includes shifts by a negative number.
2867 mark_reg_unknown(env
, regs
, insn
->dst_reg
);
2870 /* BPF_RSH is an unsigned shift. If the value in dst_reg might
2871 * be negative, then either:
2872 * 1) src_reg might be zero, so the sign bit of the result is
2873 * unknown, so we lose our signed bounds
2874 * 2) it's known negative, thus the unsigned bounds capture the
2876 * 3) the signed bounds cross zero, so they tell us nothing
2878 * If the value in dst_reg is known nonnegative, then again the
2879 * unsigned bounts capture the signed bounds.
2880 * Thus, in all cases it suffices to blow away our signed bounds
2881 * and rely on inferring new ones from the unsigned bounds and
2882 * var_off of the result.
2884 dst_reg
->smin_value
= S64_MIN
;
2885 dst_reg
->smax_value
= S64_MAX
;
2887 dst_reg
->var_off
= tnum_rshift(dst_reg
->var_off
,
2890 dst_reg
->var_off
= tnum_rshift(tnum_unknown
, umin_val
);
2891 dst_reg
->umin_value
>>= umax_val
;
2892 dst_reg
->umax_value
>>= umin_val
;
2893 /* We may learn something more from the var_off */
2894 __update_reg_bounds(dst_reg
);
2897 mark_reg_unknown(env
, regs
, insn
->dst_reg
);
2901 if (BPF_CLASS(insn
->code
) != BPF_ALU64
) {
2902 /* 32-bit ALU ops are (32,32)->32 */
2903 coerce_reg_to_size(dst_reg
, 4);
2904 coerce_reg_to_size(&src_reg
, 4);
2907 __reg_deduce_bounds(dst_reg
);
2908 __reg_bound_offset(dst_reg
);
2912 /* Handles ALU ops other than BPF_END, BPF_NEG and BPF_MOV: computes new min/max
2915 static int adjust_reg_min_max_vals(struct bpf_verifier_env
*env
,
2916 struct bpf_insn
*insn
)
2918 struct bpf_verifier_state
*vstate
= env
->cur_state
;
2919 struct bpf_func_state
*state
= vstate
->frame
[vstate
->curframe
];
2920 struct bpf_reg_state
*regs
= state
->regs
, *dst_reg
, *src_reg
;
2921 struct bpf_reg_state
*ptr_reg
= NULL
, off_reg
= {0};
2922 u8 opcode
= BPF_OP(insn
->code
);
2924 dst_reg
= ®s
[insn
->dst_reg
];
2926 if (dst_reg
->type
!= SCALAR_VALUE
)
2928 if (BPF_SRC(insn
->code
) == BPF_X
) {
2929 src_reg
= ®s
[insn
->src_reg
];
2930 if (src_reg
->type
!= SCALAR_VALUE
) {
2931 if (dst_reg
->type
!= SCALAR_VALUE
) {
2932 /* Combining two pointers by any ALU op yields
2933 * an arbitrary scalar. Disallow all math except
2934 * pointer subtraction
2936 if (opcode
== BPF_SUB
){
2937 mark_reg_unknown(env
, regs
, insn
->dst_reg
);
2940 verbose(env
, "R%d pointer %s pointer prohibited\n",
2942 bpf_alu_string
[opcode
>> 4]);
2945 /* scalar += pointer
2946 * This is legal, but we have to reverse our
2947 * src/dest handling in computing the range
2949 return adjust_ptr_min_max_vals(env
, insn
,
2952 } else if (ptr_reg
) {
2953 /* pointer += scalar */
2954 return adjust_ptr_min_max_vals(env
, insn
,
2958 /* Pretend the src is a reg with a known value, since we only
2959 * need to be able to read from this state.
2961 off_reg
.type
= SCALAR_VALUE
;
2962 __mark_reg_known(&off_reg
, insn
->imm
);
2964 if (ptr_reg
) /* pointer += K */
2965 return adjust_ptr_min_max_vals(env
, insn
,
2969 /* Got here implies adding two SCALAR_VALUEs */
2970 if (WARN_ON_ONCE(ptr_reg
)) {
2971 print_verifier_state(env
, state
);
2972 verbose(env
, "verifier internal error: unexpected ptr_reg\n");
2975 if (WARN_ON(!src_reg
)) {
2976 print_verifier_state(env
, state
);
2977 verbose(env
, "verifier internal error: no src_reg\n");
2980 return adjust_scalar_min_max_vals(env
, insn
, dst_reg
, *src_reg
);
2983 /* check validity of 32-bit and 64-bit arithmetic operations */
2984 static int check_alu_op(struct bpf_verifier_env
*env
, struct bpf_insn
*insn
)
2986 struct bpf_reg_state
*regs
= cur_regs(env
);
2987 u8 opcode
= BPF_OP(insn
->code
);
2990 if (opcode
== BPF_END
|| opcode
== BPF_NEG
) {
2991 if (opcode
== BPF_NEG
) {
2992 if (BPF_SRC(insn
->code
) != 0 ||
2993 insn
->src_reg
!= BPF_REG_0
||
2994 insn
->off
!= 0 || insn
->imm
!= 0) {
2995 verbose(env
, "BPF_NEG uses reserved fields\n");
2999 if (insn
->src_reg
!= BPF_REG_0
|| insn
->off
!= 0 ||
3000 (insn
->imm
!= 16 && insn
->imm
!= 32 && insn
->imm
!= 64) ||
3001 BPF_CLASS(insn
->code
) == BPF_ALU64
) {
3002 verbose(env
, "BPF_END uses reserved fields\n");
3007 /* check src operand */
3008 err
= check_reg_arg(env
, insn
->dst_reg
, SRC_OP
);
3012 if (is_pointer_value(env
, insn
->dst_reg
)) {
3013 verbose(env
, "R%d pointer arithmetic prohibited\n",
3018 /* check dest operand */
3019 err
= check_reg_arg(env
, insn
->dst_reg
, DST_OP
);
3023 } else if (opcode
== BPF_MOV
) {
3025 if (BPF_SRC(insn
->code
) == BPF_X
) {
3026 if (insn
->imm
!= 0 || insn
->off
!= 0) {
3027 verbose(env
, "BPF_MOV uses reserved fields\n");
3031 /* check src operand */
3032 err
= check_reg_arg(env
, insn
->src_reg
, SRC_OP
);
3036 if (insn
->src_reg
!= BPF_REG_0
|| insn
->off
!= 0) {
3037 verbose(env
, "BPF_MOV uses reserved fields\n");
3042 /* check dest operand */
3043 err
= check_reg_arg(env
, insn
->dst_reg
, DST_OP
);
3047 if (BPF_SRC(insn
->code
) == BPF_X
) {
3048 if (BPF_CLASS(insn
->code
) == BPF_ALU64
) {
3050 * copy register state to dest reg
3052 regs
[insn
->dst_reg
] = regs
[insn
->src_reg
];
3053 regs
[insn
->dst_reg
].live
|= REG_LIVE_WRITTEN
;
3056 if (is_pointer_value(env
, insn
->src_reg
)) {
3058 "R%d partial copy of pointer\n",
3062 mark_reg_unknown(env
, regs
, insn
->dst_reg
);
3063 coerce_reg_to_size(®s
[insn
->dst_reg
], 4);
3067 * remember the value we stored into this reg
3069 regs
[insn
->dst_reg
].type
= SCALAR_VALUE
;
3070 if (BPF_CLASS(insn
->code
) == BPF_ALU64
) {
3071 __mark_reg_known(regs
+ insn
->dst_reg
,
3074 __mark_reg_known(regs
+ insn
->dst_reg
,
3079 } else if (opcode
> BPF_END
) {
3080 verbose(env
, "invalid BPF_ALU opcode %x\n", opcode
);
3083 } else { /* all other ALU ops: and, sub, xor, add, ... */
3085 if (BPF_SRC(insn
->code
) == BPF_X
) {
3086 if (insn
->imm
!= 0 || insn
->off
!= 0) {
3087 verbose(env
, "BPF_ALU uses reserved fields\n");
3090 /* check src1 operand */
3091 err
= check_reg_arg(env
, insn
->src_reg
, SRC_OP
);
3095 if (insn
->src_reg
!= BPF_REG_0
|| insn
->off
!= 0) {
3096 verbose(env
, "BPF_ALU uses reserved fields\n");
3101 /* check src2 operand */
3102 err
= check_reg_arg(env
, insn
->dst_reg
, SRC_OP
);
3106 if ((opcode
== BPF_MOD
|| opcode
== BPF_DIV
) &&
3107 BPF_SRC(insn
->code
) == BPF_K
&& insn
->imm
== 0) {
3108 verbose(env
, "div by zero\n");
3112 if (opcode
== BPF_ARSH
&& BPF_CLASS(insn
->code
) != BPF_ALU64
) {
3113 verbose(env
, "BPF_ARSH not supported for 32 bit ALU\n");
3117 if ((opcode
== BPF_LSH
|| opcode
== BPF_RSH
||
3118 opcode
== BPF_ARSH
) && BPF_SRC(insn
->code
) == BPF_K
) {
3119 int size
= BPF_CLASS(insn
->code
) == BPF_ALU64
? 64 : 32;
3121 if (insn
->imm
< 0 || insn
->imm
>= size
) {
3122 verbose(env
, "invalid shift %d\n", insn
->imm
);
3127 /* check dest operand */
3128 err
= check_reg_arg(env
, insn
->dst_reg
, DST_OP_NO_MARK
);
3132 return adjust_reg_min_max_vals(env
, insn
);
3138 static void find_good_pkt_pointers(struct bpf_verifier_state
*vstate
,
3139 struct bpf_reg_state
*dst_reg
,
3140 enum bpf_reg_type type
,
3141 bool range_right_open
)
3143 struct bpf_func_state
*state
= vstate
->frame
[vstate
->curframe
];
3144 struct bpf_reg_state
*regs
= state
->regs
, *reg
;
3148 if (dst_reg
->off
< 0 ||
3149 (dst_reg
->off
== 0 && range_right_open
))
3150 /* This doesn't give us any range */
3153 if (dst_reg
->umax_value
> MAX_PACKET_OFF
||
3154 dst_reg
->umax_value
+ dst_reg
->off
> MAX_PACKET_OFF
)
3155 /* Risk of overflow. For instance, ptr + (1<<63) may be less
3156 * than pkt_end, but that's because it's also less than pkt.
3160 new_range
= dst_reg
->off
;
3161 if (range_right_open
)
3164 /* Examples for register markings:
3166 * pkt_data in dst register:
3170 * if (r2 > pkt_end) goto <handle exception>
3175 * if (r2 < pkt_end) goto <access okay>
3176 * <handle exception>
3179 * r2 == dst_reg, pkt_end == src_reg
3180 * r2=pkt(id=n,off=8,r=0)
3181 * r3=pkt(id=n,off=0,r=0)
3183 * pkt_data in src register:
3187 * if (pkt_end >= r2) goto <access okay>
3188 * <handle exception>
3192 * if (pkt_end <= r2) goto <handle exception>
3196 * pkt_end == dst_reg, r2 == src_reg
3197 * r2=pkt(id=n,off=8,r=0)
3198 * r3=pkt(id=n,off=0,r=0)
3200 * Find register r3 and mark its range as r3=pkt(id=n,off=0,r=8)
3201 * or r3=pkt(id=n,off=0,r=8-1), so that range of bytes [r3, r3 + 8)
3202 * and [r3, r3 + 8-1) respectively is safe to access depending on
3206 /* If our ids match, then we must have the same max_value. And we
3207 * don't care about the other reg's fixed offset, since if it's too big
3208 * the range won't allow anything.
3209 * dst_reg->off is known < MAX_PACKET_OFF, therefore it fits in a u16.
3211 for (i
= 0; i
< MAX_BPF_REG
; i
++)
3212 if (regs
[i
].type
== type
&& regs
[i
].id
== dst_reg
->id
)
3213 /* keep the maximum range already checked */
3214 regs
[i
].range
= max(regs
[i
].range
, new_range
);
3216 for (j
= 0; j
<= vstate
->curframe
; j
++) {
3217 state
= vstate
->frame
[j
];
3218 for (i
= 0; i
< state
->allocated_stack
/ BPF_REG_SIZE
; i
++) {
3219 if (state
->stack
[i
].slot_type
[0] != STACK_SPILL
)
3221 reg
= &state
->stack
[i
].spilled_ptr
;
3222 if (reg
->type
== type
&& reg
->id
== dst_reg
->id
)
3223 reg
->range
= max(reg
->range
, new_range
);
3228 /* Adjusts the register min/max values in the case that the dst_reg is the
3229 * variable register that we are working on, and src_reg is a constant or we're
3230 * simply doing a BPF_K check.
3231 * In JEQ/JNE cases we also adjust the var_off values.
3233 static void reg_set_min_max(struct bpf_reg_state
*true_reg
,
3234 struct bpf_reg_state
*false_reg
, u64 val
,
3237 /* If the dst_reg is a pointer, we can't learn anything about its
3238 * variable offset from the compare (unless src_reg were a pointer into
3239 * the same object, but we don't bother with that.
3240 * Since false_reg and true_reg have the same type by construction, we
3241 * only need to check one of them for pointerness.
3243 if (__is_pointer_value(false, false_reg
))
3248 /* If this is false then we know nothing Jon Snow, but if it is
3249 * true then we know for sure.
3251 __mark_reg_known(true_reg
, val
);
3254 /* If this is true we know nothing Jon Snow, but if it is false
3255 * we know the value for sure;
3257 __mark_reg_known(false_reg
, val
);
3260 false_reg
->umax_value
= min(false_reg
->umax_value
, val
);
3261 true_reg
->umin_value
= max(true_reg
->umin_value
, val
+ 1);
3264 false_reg
->smax_value
= min_t(s64
, false_reg
->smax_value
, val
);
3265 true_reg
->smin_value
= max_t(s64
, true_reg
->smin_value
, val
+ 1);
3268 false_reg
->umin_value
= max(false_reg
->umin_value
, val
);
3269 true_reg
->umax_value
= min(true_reg
->umax_value
, val
- 1);
3272 false_reg
->smin_value
= max_t(s64
, false_reg
->smin_value
, val
);
3273 true_reg
->smax_value
= min_t(s64
, true_reg
->smax_value
, val
- 1);
3276 false_reg
->umax_value
= min(false_reg
->umax_value
, val
- 1);
3277 true_reg
->umin_value
= max(true_reg
->umin_value
, val
);
3280 false_reg
->smax_value
= min_t(s64
, false_reg
->smax_value
, val
- 1);
3281 true_reg
->smin_value
= max_t(s64
, true_reg
->smin_value
, val
);
3284 false_reg
->umin_value
= max(false_reg
->umin_value
, val
+ 1);
3285 true_reg
->umax_value
= min(true_reg
->umax_value
, val
);
3288 false_reg
->smin_value
= max_t(s64
, false_reg
->smin_value
, val
+ 1);
3289 true_reg
->smax_value
= min_t(s64
, true_reg
->smax_value
, val
);
3295 __reg_deduce_bounds(false_reg
);
3296 __reg_deduce_bounds(true_reg
);
3297 /* We might have learned some bits from the bounds. */
3298 __reg_bound_offset(false_reg
);
3299 __reg_bound_offset(true_reg
);
3300 /* Intersecting with the old var_off might have improved our bounds
3301 * slightly. e.g. if umax was 0x7f...f and var_off was (0; 0xf...fc),
3302 * then new var_off is (0; 0x7f...fc) which improves our umax.
3304 __update_reg_bounds(false_reg
);
3305 __update_reg_bounds(true_reg
);
3308 /* Same as above, but for the case that dst_reg holds a constant and src_reg is
3311 static void reg_set_min_max_inv(struct bpf_reg_state
*true_reg
,
3312 struct bpf_reg_state
*false_reg
, u64 val
,
3315 if (__is_pointer_value(false, false_reg
))
3320 /* If this is false then we know nothing Jon Snow, but if it is
3321 * true then we know for sure.
3323 __mark_reg_known(true_reg
, val
);
3326 /* If this is true we know nothing Jon Snow, but if it is false
3327 * we know the value for sure;
3329 __mark_reg_known(false_reg
, val
);
3332 true_reg
->umax_value
= min(true_reg
->umax_value
, val
- 1);
3333 false_reg
->umin_value
= max(false_reg
->umin_value
, val
);
3336 true_reg
->smax_value
= min_t(s64
, true_reg
->smax_value
, val
- 1);
3337 false_reg
->smin_value
= max_t(s64
, false_reg
->smin_value
, val
);
3340 true_reg
->umin_value
= max(true_reg
->umin_value
, val
+ 1);
3341 false_reg
->umax_value
= min(false_reg
->umax_value
, val
);
3344 true_reg
->smin_value
= max_t(s64
, true_reg
->smin_value
, val
+ 1);
3345 false_reg
->smax_value
= min_t(s64
, false_reg
->smax_value
, val
);
3348 true_reg
->umax_value
= min(true_reg
->umax_value
, val
);
3349 false_reg
->umin_value
= max(false_reg
->umin_value
, val
+ 1);
3352 true_reg
->smax_value
= min_t(s64
, true_reg
->smax_value
, val
);
3353 false_reg
->smin_value
= max_t(s64
, false_reg
->smin_value
, val
+ 1);
3356 true_reg
->umin_value
= max(true_reg
->umin_value
, val
);
3357 false_reg
->umax_value
= min(false_reg
->umax_value
, val
- 1);
3360 true_reg
->smin_value
= max_t(s64
, true_reg
->smin_value
, val
);
3361 false_reg
->smax_value
= min_t(s64
, false_reg
->smax_value
, val
- 1);
3367 __reg_deduce_bounds(false_reg
);
3368 __reg_deduce_bounds(true_reg
);
3369 /* We might have learned some bits from the bounds. */
3370 __reg_bound_offset(false_reg
);
3371 __reg_bound_offset(true_reg
);
3372 /* Intersecting with the old var_off might have improved our bounds
3373 * slightly. e.g. if umax was 0x7f...f and var_off was (0; 0xf...fc),
3374 * then new var_off is (0; 0x7f...fc) which improves our umax.
3376 __update_reg_bounds(false_reg
);
3377 __update_reg_bounds(true_reg
);
3380 /* Regs are known to be equal, so intersect their min/max/var_off */
3381 static void __reg_combine_min_max(struct bpf_reg_state
*src_reg
,
3382 struct bpf_reg_state
*dst_reg
)
3384 src_reg
->umin_value
= dst_reg
->umin_value
= max(src_reg
->umin_value
,
3385 dst_reg
->umin_value
);
3386 src_reg
->umax_value
= dst_reg
->umax_value
= min(src_reg
->umax_value
,
3387 dst_reg
->umax_value
);
3388 src_reg
->smin_value
= dst_reg
->smin_value
= max(src_reg
->smin_value
,
3389 dst_reg
->smin_value
);
3390 src_reg
->smax_value
= dst_reg
->smax_value
= min(src_reg
->smax_value
,
3391 dst_reg
->smax_value
);
3392 src_reg
->var_off
= dst_reg
->var_off
= tnum_intersect(src_reg
->var_off
,
3394 /* We might have learned new bounds from the var_off. */
3395 __update_reg_bounds(src_reg
);
3396 __update_reg_bounds(dst_reg
);
3397 /* We might have learned something about the sign bit. */
3398 __reg_deduce_bounds(src_reg
);
3399 __reg_deduce_bounds(dst_reg
);
3400 /* We might have learned some bits from the bounds. */
3401 __reg_bound_offset(src_reg
);
3402 __reg_bound_offset(dst_reg
);
3403 /* Intersecting with the old var_off might have improved our bounds
3404 * slightly. e.g. if umax was 0x7f...f and var_off was (0; 0xf...fc),
3405 * then new var_off is (0; 0x7f...fc) which improves our umax.
3407 __update_reg_bounds(src_reg
);
3408 __update_reg_bounds(dst_reg
);
3411 static void reg_combine_min_max(struct bpf_reg_state
*true_src
,
3412 struct bpf_reg_state
*true_dst
,
3413 struct bpf_reg_state
*false_src
,
3414 struct bpf_reg_state
*false_dst
,
3419 __reg_combine_min_max(true_src
, true_dst
);
3422 __reg_combine_min_max(false_src
, false_dst
);
3427 static void mark_map_reg(struct bpf_reg_state
*regs
, u32 regno
, u32 id
,
3430 struct bpf_reg_state
*reg
= ®s
[regno
];
3432 if (reg
->type
== PTR_TO_MAP_VALUE_OR_NULL
&& reg
->id
== id
) {
3433 /* Old offset (both fixed and variable parts) should
3434 * have been known-zero, because we don't allow pointer
3435 * arithmetic on pointers that might be NULL.
3437 if (WARN_ON_ONCE(reg
->smin_value
|| reg
->smax_value
||
3438 !tnum_equals_const(reg
->var_off
, 0) ||
3440 __mark_reg_known_zero(reg
);
3444 reg
->type
= SCALAR_VALUE
;
3445 } else if (reg
->map_ptr
->inner_map_meta
) {
3446 reg
->type
= CONST_PTR_TO_MAP
;
3447 reg
->map_ptr
= reg
->map_ptr
->inner_map_meta
;
3449 reg
->type
= PTR_TO_MAP_VALUE
;
3451 /* We don't need id from this point onwards anymore, thus we
3452 * should better reset it, so that state pruning has chances
3459 /* The logic is similar to find_good_pkt_pointers(), both could eventually
3460 * be folded together at some point.
3462 static void mark_map_regs(struct bpf_verifier_state
*vstate
, u32 regno
,
3465 struct bpf_func_state
*state
= vstate
->frame
[vstate
->curframe
];
3466 struct bpf_reg_state
*regs
= state
->regs
;
3467 u32 id
= regs
[regno
].id
;
3470 for (i
= 0; i
< MAX_BPF_REG
; i
++)
3471 mark_map_reg(regs
, i
, id
, is_null
);
3473 for (j
= 0; j
<= vstate
->curframe
; j
++) {
3474 state
= vstate
->frame
[j
];
3475 for (i
= 0; i
< state
->allocated_stack
/ BPF_REG_SIZE
; i
++) {
3476 if (state
->stack
[i
].slot_type
[0] != STACK_SPILL
)
3478 mark_map_reg(&state
->stack
[i
].spilled_ptr
, 0, id
, is_null
);
3483 static bool try_match_pkt_pointers(const struct bpf_insn
*insn
,
3484 struct bpf_reg_state
*dst_reg
,
3485 struct bpf_reg_state
*src_reg
,
3486 struct bpf_verifier_state
*this_branch
,
3487 struct bpf_verifier_state
*other_branch
)
3489 if (BPF_SRC(insn
->code
) != BPF_X
)
3492 switch (BPF_OP(insn
->code
)) {
3494 if ((dst_reg
->type
== PTR_TO_PACKET
&&
3495 src_reg
->type
== PTR_TO_PACKET_END
) ||
3496 (dst_reg
->type
== PTR_TO_PACKET_META
&&
3497 reg_is_init_pkt_pointer(src_reg
, PTR_TO_PACKET
))) {
3498 /* pkt_data' > pkt_end, pkt_meta' > pkt_data */
3499 find_good_pkt_pointers(this_branch
, dst_reg
,
3500 dst_reg
->type
, false);
3501 } else if ((dst_reg
->type
== PTR_TO_PACKET_END
&&
3502 src_reg
->type
== PTR_TO_PACKET
) ||
3503 (reg_is_init_pkt_pointer(dst_reg
, PTR_TO_PACKET
) &&
3504 src_reg
->type
== PTR_TO_PACKET_META
)) {
3505 /* pkt_end > pkt_data', pkt_data > pkt_meta' */
3506 find_good_pkt_pointers(other_branch
, src_reg
,
3507 src_reg
->type
, true);
3513 if ((dst_reg
->type
== PTR_TO_PACKET
&&
3514 src_reg
->type
== PTR_TO_PACKET_END
) ||
3515 (dst_reg
->type
== PTR_TO_PACKET_META
&&
3516 reg_is_init_pkt_pointer(src_reg
, PTR_TO_PACKET
))) {
3517 /* pkt_data' < pkt_end, pkt_meta' < pkt_data */
3518 find_good_pkt_pointers(other_branch
, dst_reg
,
3519 dst_reg
->type
, true);
3520 } else if ((dst_reg
->type
== PTR_TO_PACKET_END
&&
3521 src_reg
->type
== PTR_TO_PACKET
) ||
3522 (reg_is_init_pkt_pointer(dst_reg
, PTR_TO_PACKET
) &&
3523 src_reg
->type
== PTR_TO_PACKET_META
)) {
3524 /* pkt_end < pkt_data', pkt_data > pkt_meta' */
3525 find_good_pkt_pointers(this_branch
, src_reg
,
3526 src_reg
->type
, false);
3532 if ((dst_reg
->type
== PTR_TO_PACKET
&&
3533 src_reg
->type
== PTR_TO_PACKET_END
) ||
3534 (dst_reg
->type
== PTR_TO_PACKET_META
&&
3535 reg_is_init_pkt_pointer(src_reg
, PTR_TO_PACKET
))) {
3536 /* pkt_data' >= pkt_end, pkt_meta' >= pkt_data */
3537 find_good_pkt_pointers(this_branch
, dst_reg
,
3538 dst_reg
->type
, true);
3539 } else if ((dst_reg
->type
== PTR_TO_PACKET_END
&&
3540 src_reg
->type
== PTR_TO_PACKET
) ||
3541 (reg_is_init_pkt_pointer(dst_reg
, PTR_TO_PACKET
) &&
3542 src_reg
->type
== PTR_TO_PACKET_META
)) {
3543 /* pkt_end >= pkt_data', pkt_data >= pkt_meta' */
3544 find_good_pkt_pointers(other_branch
, src_reg
,
3545 src_reg
->type
, false);
3551 if ((dst_reg
->type
== PTR_TO_PACKET
&&
3552 src_reg
->type
== PTR_TO_PACKET_END
) ||
3553 (dst_reg
->type
== PTR_TO_PACKET_META
&&
3554 reg_is_init_pkt_pointer(src_reg
, PTR_TO_PACKET
))) {
3555 /* pkt_data' <= pkt_end, pkt_meta' <= pkt_data */
3556 find_good_pkt_pointers(other_branch
, dst_reg
,
3557 dst_reg
->type
, false);
3558 } else if ((dst_reg
->type
== PTR_TO_PACKET_END
&&
3559 src_reg
->type
== PTR_TO_PACKET
) ||
3560 (reg_is_init_pkt_pointer(dst_reg
, PTR_TO_PACKET
) &&
3561 src_reg
->type
== PTR_TO_PACKET_META
)) {
3562 /* pkt_end <= pkt_data', pkt_data <= pkt_meta' */
3563 find_good_pkt_pointers(this_branch
, src_reg
,
3564 src_reg
->type
, true);
3576 static int check_cond_jmp_op(struct bpf_verifier_env
*env
,
3577 struct bpf_insn
*insn
, int *insn_idx
)
3579 struct bpf_verifier_state
*this_branch
= env
->cur_state
;
3580 struct bpf_verifier_state
*other_branch
;
3581 struct bpf_reg_state
*regs
= this_branch
->frame
[this_branch
->curframe
]->regs
;
3582 struct bpf_reg_state
*dst_reg
, *other_branch_regs
;
3583 u8 opcode
= BPF_OP(insn
->code
);
3586 if (opcode
> BPF_JSLE
) {
3587 verbose(env
, "invalid BPF_JMP opcode %x\n", opcode
);
3591 if (BPF_SRC(insn
->code
) == BPF_X
) {
3592 if (insn
->imm
!= 0) {
3593 verbose(env
, "BPF_JMP uses reserved fields\n");
3597 /* check src1 operand */
3598 err
= check_reg_arg(env
, insn
->src_reg
, SRC_OP
);
3602 if (is_pointer_value(env
, insn
->src_reg
)) {
3603 verbose(env
, "R%d pointer comparison prohibited\n",
3608 if (insn
->src_reg
!= BPF_REG_0
) {
3609 verbose(env
, "BPF_JMP uses reserved fields\n");
3614 /* check src2 operand */
3615 err
= check_reg_arg(env
, insn
->dst_reg
, SRC_OP
);
3619 dst_reg
= ®s
[insn
->dst_reg
];
3621 /* detect if R == 0 where R was initialized to zero earlier */
3622 if (BPF_SRC(insn
->code
) == BPF_K
&&
3623 (opcode
== BPF_JEQ
|| opcode
== BPF_JNE
) &&
3624 dst_reg
->type
== SCALAR_VALUE
&&
3625 tnum_is_const(dst_reg
->var_off
)) {
3626 if ((opcode
== BPF_JEQ
&& dst_reg
->var_off
.value
== insn
->imm
) ||
3627 (opcode
== BPF_JNE
&& dst_reg
->var_off
.value
!= insn
->imm
)) {
3628 /* if (imm == imm) goto pc+off;
3629 * only follow the goto, ignore fall-through
3631 *insn_idx
+= insn
->off
;
3634 /* if (imm != imm) goto pc+off;
3635 * only follow fall-through branch, since
3636 * that's where the program will go
3642 other_branch
= push_stack(env
, *insn_idx
+ insn
->off
+ 1, *insn_idx
);
3645 other_branch_regs
= other_branch
->frame
[other_branch
->curframe
]->regs
;
3647 /* detect if we are comparing against a constant value so we can adjust
3648 * our min/max values for our dst register.
3649 * this is only legit if both are scalars (or pointers to the same
3650 * object, I suppose, but we don't support that right now), because
3651 * otherwise the different base pointers mean the offsets aren't
3654 if (BPF_SRC(insn
->code
) == BPF_X
) {
3655 if (dst_reg
->type
== SCALAR_VALUE
&&
3656 regs
[insn
->src_reg
].type
== SCALAR_VALUE
) {
3657 if (tnum_is_const(regs
[insn
->src_reg
].var_off
))
3658 reg_set_min_max(&other_branch_regs
[insn
->dst_reg
],
3659 dst_reg
, regs
[insn
->src_reg
].var_off
.value
,
3661 else if (tnum_is_const(dst_reg
->var_off
))
3662 reg_set_min_max_inv(&other_branch_regs
[insn
->src_reg
],
3663 ®s
[insn
->src_reg
],
3664 dst_reg
->var_off
.value
, opcode
);
3665 else if (opcode
== BPF_JEQ
|| opcode
== BPF_JNE
)
3666 /* Comparing for equality, we can combine knowledge */
3667 reg_combine_min_max(&other_branch_regs
[insn
->src_reg
],
3668 &other_branch_regs
[insn
->dst_reg
],
3669 ®s
[insn
->src_reg
],
3670 ®s
[insn
->dst_reg
], opcode
);
3672 } else if (dst_reg
->type
== SCALAR_VALUE
) {
3673 reg_set_min_max(&other_branch_regs
[insn
->dst_reg
],
3674 dst_reg
, insn
->imm
, opcode
);
3677 /* detect if R == 0 where R is returned from bpf_map_lookup_elem() */
3678 if (BPF_SRC(insn
->code
) == BPF_K
&&
3679 insn
->imm
== 0 && (opcode
== BPF_JEQ
|| opcode
== BPF_JNE
) &&
3680 dst_reg
->type
== PTR_TO_MAP_VALUE_OR_NULL
) {
3681 /* Mark all identical map registers in each branch as either
3682 * safe or unknown depending R == 0 or R != 0 conditional.
3684 mark_map_regs(this_branch
, insn
->dst_reg
, opcode
== BPF_JNE
);
3685 mark_map_regs(other_branch
, insn
->dst_reg
, opcode
== BPF_JEQ
);
3686 } else if (!try_match_pkt_pointers(insn
, dst_reg
, ®s
[insn
->src_reg
],
3687 this_branch
, other_branch
) &&
3688 is_pointer_value(env
, insn
->dst_reg
)) {
3689 verbose(env
, "R%d pointer comparison prohibited\n",
3694 print_verifier_state(env
, this_branch
->frame
[this_branch
->curframe
]);
3698 /* return the map pointer stored inside BPF_LD_IMM64 instruction */
3699 static struct bpf_map
*ld_imm64_to_map_ptr(struct bpf_insn
*insn
)
3701 u64 imm64
= ((u64
) (u32
) insn
[0].imm
) | ((u64
) (u32
) insn
[1].imm
) << 32;
3703 return (struct bpf_map
*) (unsigned long) imm64
;
3706 /* verify BPF_LD_IMM64 instruction */
3707 static int check_ld_imm(struct bpf_verifier_env
*env
, struct bpf_insn
*insn
)
3709 struct bpf_reg_state
*regs
= cur_regs(env
);
3712 if (BPF_SIZE(insn
->code
) != BPF_DW
) {
3713 verbose(env
, "invalid BPF_LD_IMM insn\n");
3716 if (insn
->off
!= 0) {
3717 verbose(env
, "BPF_LD_IMM64 uses reserved fields\n");
3721 err
= check_reg_arg(env
, insn
->dst_reg
, DST_OP
);
3725 if (insn
->src_reg
== 0) {
3726 u64 imm
= ((u64
)(insn
+ 1)->imm
<< 32) | (u32
)insn
->imm
;
3728 regs
[insn
->dst_reg
].type
= SCALAR_VALUE
;
3729 __mark_reg_known(®s
[insn
->dst_reg
], imm
);
3733 /* replace_map_fd_with_map_ptr() should have caught bad ld_imm64 */
3734 BUG_ON(insn
->src_reg
!= BPF_PSEUDO_MAP_FD
);
3736 regs
[insn
->dst_reg
].type
= CONST_PTR_TO_MAP
;
3737 regs
[insn
->dst_reg
].map_ptr
= ld_imm64_to_map_ptr(insn
);
3741 static bool may_access_skb(enum bpf_prog_type type
)
3744 case BPF_PROG_TYPE_SOCKET_FILTER
:
3745 case BPF_PROG_TYPE_SCHED_CLS
:
3746 case BPF_PROG_TYPE_SCHED_ACT
:
3753 /* verify safety of LD_ABS|LD_IND instructions:
3754 * - they can only appear in the programs where ctx == skb
3755 * - since they are wrappers of function calls, they scratch R1-R5 registers,
3756 * preserve R6-R9, and store return value into R0
3759 * ctx == skb == R6 == CTX
3762 * SRC == any register
3763 * IMM == 32-bit immediate
3766 * R0 - 8/16/32-bit skb data converted to cpu endianness
3768 static int check_ld_abs(struct bpf_verifier_env
*env
, struct bpf_insn
*insn
)
3770 struct bpf_reg_state
*regs
= cur_regs(env
);
3771 u8 mode
= BPF_MODE(insn
->code
);
3774 if (!may_access_skb(env
->prog
->type
)) {
3775 verbose(env
, "BPF_LD_[ABS|IND] instructions not allowed for this program type\n");
3779 if (env
->subprog_cnt
) {
3780 /* when program has LD_ABS insn JITs and interpreter assume
3781 * that r1 == ctx == skb which is not the case for callees
3782 * that can have arbitrary arguments. It's problematic
3783 * for main prog as well since JITs would need to analyze
3784 * all functions in order to make proper register save/restore
3785 * decisions in the main prog. Hence disallow LD_ABS with calls
3787 verbose(env
, "BPF_LD_[ABS|IND] instructions cannot be mixed with bpf-to-bpf calls\n");
3791 if (insn
->dst_reg
!= BPF_REG_0
|| insn
->off
!= 0 ||
3792 BPF_SIZE(insn
->code
) == BPF_DW
||
3793 (mode
== BPF_ABS
&& insn
->src_reg
!= BPF_REG_0
)) {
3794 verbose(env
, "BPF_LD_[ABS|IND] uses reserved fields\n");
3798 /* check whether implicit source operand (register R6) is readable */
3799 err
= check_reg_arg(env
, BPF_REG_6
, SRC_OP
);
3803 if (regs
[BPF_REG_6
].type
!= PTR_TO_CTX
) {
3805 "at the time of BPF_LD_ABS|IND R6 != pointer to skb\n");
3809 if (mode
== BPF_IND
) {
3810 /* check explicit source operand */
3811 err
= check_reg_arg(env
, insn
->src_reg
, SRC_OP
);
3816 /* reset caller saved regs to unreadable */
3817 for (i
= 0; i
< CALLER_SAVED_REGS
; i
++) {
3818 mark_reg_not_init(env
, regs
, caller_saved
[i
]);
3819 check_reg_arg(env
, caller_saved
[i
], DST_OP_NO_MARK
);
3822 /* mark destination R0 register as readable, since it contains
3823 * the value fetched from the packet.
3824 * Already marked as written above.
3826 mark_reg_unknown(env
, regs
, BPF_REG_0
);
3830 static int check_return_code(struct bpf_verifier_env
*env
)
3832 struct bpf_reg_state
*reg
;
3833 struct tnum range
= tnum_range(0, 1);
3835 switch (env
->prog
->type
) {
3836 case BPF_PROG_TYPE_CGROUP_SKB
:
3837 case BPF_PROG_TYPE_CGROUP_SOCK
:
3838 case BPF_PROG_TYPE_SOCK_OPS
:
3839 case BPF_PROG_TYPE_CGROUP_DEVICE
:
3845 reg
= cur_regs(env
) + BPF_REG_0
;
3846 if (reg
->type
!= SCALAR_VALUE
) {
3847 verbose(env
, "At program exit the register R0 is not a known value (%s)\n",
3848 reg_type_str
[reg
->type
]);
3852 if (!tnum_in(range
, reg
->var_off
)) {
3853 verbose(env
, "At program exit the register R0 ");
3854 if (!tnum_is_unknown(reg
->var_off
)) {
3857 tnum_strn(tn_buf
, sizeof(tn_buf
), reg
->var_off
);
3858 verbose(env
, "has value %s", tn_buf
);
3860 verbose(env
, "has unknown scalar value");
3862 verbose(env
, " should have been 0 or 1\n");
3868 /* non-recursive DFS pseudo code
3869 * 1 procedure DFS-iterative(G,v):
3870 * 2 label v as discovered
3871 * 3 let S be a stack
3873 * 5 while S is not empty
3875 * 7 if t is what we're looking for:
3877 * 9 for all edges e in G.adjacentEdges(t) do
3878 * 10 if edge e is already labelled
3879 * 11 continue with the next edge
3880 * 12 w <- G.adjacentVertex(t,e)
3881 * 13 if vertex w is not discovered and not explored
3882 * 14 label e as tree-edge
3883 * 15 label w as discovered
3886 * 18 else if vertex w is discovered
3887 * 19 label e as back-edge
3889 * 21 // vertex w is explored
3890 * 22 label e as forward- or cross-edge
3891 * 23 label t as explored
3896 * 0x11 - discovered and fall-through edge labelled
3897 * 0x12 - discovered and fall-through and branch edges labelled
3908 #define STATE_LIST_MARK ((struct bpf_verifier_state_list *) -1L)
3910 static int *insn_stack
; /* stack of insns to process */
3911 static int cur_stack
; /* current stack index */
3912 static int *insn_state
;
3914 /* t, w, e - match pseudo-code above:
3915 * t - index of current instruction
3916 * w - next instruction
3919 static int push_insn(int t
, int w
, int e
, struct bpf_verifier_env
*env
)
3921 if (e
== FALLTHROUGH
&& insn_state
[t
] >= (DISCOVERED
| FALLTHROUGH
))
3924 if (e
== BRANCH
&& insn_state
[t
] >= (DISCOVERED
| BRANCH
))
3927 if (w
< 0 || w
>= env
->prog
->len
) {
3928 verbose(env
, "jump out of range from insn %d to %d\n", t
, w
);
3933 /* mark branch target for state pruning */
3934 env
->explored_states
[w
] = STATE_LIST_MARK
;
3936 if (insn_state
[w
] == 0) {
3938 insn_state
[t
] = DISCOVERED
| e
;
3939 insn_state
[w
] = DISCOVERED
;
3940 if (cur_stack
>= env
->prog
->len
)
3942 insn_stack
[cur_stack
++] = w
;
3944 } else if ((insn_state
[w
] & 0xF0) == DISCOVERED
) {
3945 verbose(env
, "back-edge from insn %d to %d\n", t
, w
);
3947 } else if (insn_state
[w
] == EXPLORED
) {
3948 /* forward- or cross-edge */
3949 insn_state
[t
] = DISCOVERED
| e
;
3951 verbose(env
, "insn state internal bug\n");
3957 /* non-recursive depth-first-search to detect loops in BPF program
3958 * loop == back-edge in directed graph
3960 static int check_cfg(struct bpf_verifier_env
*env
)
3962 struct bpf_insn
*insns
= env
->prog
->insnsi
;
3963 int insn_cnt
= env
->prog
->len
;
3967 ret
= check_subprogs(env
);
3971 insn_state
= kcalloc(insn_cnt
, sizeof(int), GFP_KERNEL
);
3975 insn_stack
= kcalloc(insn_cnt
, sizeof(int), GFP_KERNEL
);
3981 insn_state
[0] = DISCOVERED
; /* mark 1st insn as discovered */
3982 insn_stack
[0] = 0; /* 0 is the first instruction */
3988 t
= insn_stack
[cur_stack
- 1];
3990 if (BPF_CLASS(insns
[t
].code
) == BPF_JMP
) {
3991 u8 opcode
= BPF_OP(insns
[t
].code
);
3993 if (opcode
== BPF_EXIT
) {
3995 } else if (opcode
== BPF_CALL
) {
3996 ret
= push_insn(t
, t
+ 1, FALLTHROUGH
, env
);
4001 if (t
+ 1 < insn_cnt
)
4002 env
->explored_states
[t
+ 1] = STATE_LIST_MARK
;
4003 if (insns
[t
].src_reg
== BPF_PSEUDO_CALL
) {
4004 env
->explored_states
[t
] = STATE_LIST_MARK
;
4005 ret
= push_insn(t
, t
+ insns
[t
].imm
+ 1, BRANCH
, env
);
4011 } else if (opcode
== BPF_JA
) {
4012 if (BPF_SRC(insns
[t
].code
) != BPF_K
) {
4016 /* unconditional jump with single edge */
4017 ret
= push_insn(t
, t
+ insns
[t
].off
+ 1,
4023 /* tell verifier to check for equivalent states
4024 * after every call and jump
4026 if (t
+ 1 < insn_cnt
)
4027 env
->explored_states
[t
+ 1] = STATE_LIST_MARK
;
4029 /* conditional jump with two edges */
4030 env
->explored_states
[t
] = STATE_LIST_MARK
;
4031 ret
= push_insn(t
, t
+ 1, FALLTHROUGH
, env
);
4037 ret
= push_insn(t
, t
+ insns
[t
].off
+ 1, BRANCH
, env
);
4044 /* all other non-branch instructions with single
4047 ret
= push_insn(t
, t
+ 1, FALLTHROUGH
, env
);
4055 insn_state
[t
] = EXPLORED
;
4056 if (cur_stack
-- <= 0) {
4057 verbose(env
, "pop stack internal bug\n");
4064 for (i
= 0; i
< insn_cnt
; i
++) {
4065 if (insn_state
[i
] != EXPLORED
) {
4066 verbose(env
, "unreachable insn %d\n", i
);
4071 ret
= 0; /* cfg looks good */
4079 /* check %cur's range satisfies %old's */
4080 static bool range_within(struct bpf_reg_state
*old
,
4081 struct bpf_reg_state
*cur
)
4083 return old
->umin_value
<= cur
->umin_value
&&
4084 old
->umax_value
>= cur
->umax_value
&&
4085 old
->smin_value
<= cur
->smin_value
&&
4086 old
->smax_value
>= cur
->smax_value
;
4089 /* Maximum number of register states that can exist at once */
4090 #define ID_MAP_SIZE (MAX_BPF_REG + MAX_BPF_STACK / BPF_REG_SIZE)
4096 /* If in the old state two registers had the same id, then they need to have
4097 * the same id in the new state as well. But that id could be different from
4098 * the old state, so we need to track the mapping from old to new ids.
4099 * Once we have seen that, say, a reg with old id 5 had new id 9, any subsequent
4100 * regs with old id 5 must also have new id 9 for the new state to be safe. But
4101 * regs with a different old id could still have new id 9, we don't care about
4103 * So we look through our idmap to see if this old id has been seen before. If
4104 * so, we require the new id to match; otherwise, we add the id pair to the map.
4106 static bool check_ids(u32 old_id
, u32 cur_id
, struct idpair
*idmap
)
4110 for (i
= 0; i
< ID_MAP_SIZE
; i
++) {
4111 if (!idmap
[i
].old
) {
4112 /* Reached an empty slot; haven't seen this id before */
4113 idmap
[i
].old
= old_id
;
4114 idmap
[i
].cur
= cur_id
;
4117 if (idmap
[i
].old
== old_id
)
4118 return idmap
[i
].cur
== cur_id
;
4120 /* We ran out of idmap slots, which should be impossible */
4125 /* Returns true if (rold safe implies rcur safe) */
4126 static bool regsafe(struct bpf_reg_state
*rold
, struct bpf_reg_state
*rcur
,
4127 struct idpair
*idmap
)
4131 if (!(rold
->live
& REG_LIVE_READ
))
4132 /* explored state didn't use this */
4135 equal
= memcmp(rold
, rcur
, offsetof(struct bpf_reg_state
, frameno
)) == 0;
4137 if (rold
->type
== PTR_TO_STACK
)
4138 /* two stack pointers are equal only if they're pointing to
4139 * the same stack frame, since fp-8 in foo != fp-8 in bar
4141 return equal
&& rold
->frameno
== rcur
->frameno
;
4146 if (rold
->type
== NOT_INIT
)
4147 /* explored state can't have used this */
4149 if (rcur
->type
== NOT_INIT
)
4151 switch (rold
->type
) {
4153 if (rcur
->type
== SCALAR_VALUE
) {
4154 /* new val must satisfy old val knowledge */
4155 return range_within(rold
, rcur
) &&
4156 tnum_in(rold
->var_off
, rcur
->var_off
);
4158 /* We're trying to use a pointer in place of a scalar.
4159 * Even if the scalar was unbounded, this could lead to
4160 * pointer leaks because scalars are allowed to leak
4161 * while pointers are not. We could make this safe in
4162 * special cases if root is calling us, but it's
4163 * probably not worth the hassle.
4167 case PTR_TO_MAP_VALUE
:
4168 /* If the new min/max/var_off satisfy the old ones and
4169 * everything else matches, we are OK.
4170 * We don't care about the 'id' value, because nothing
4171 * uses it for PTR_TO_MAP_VALUE (only for ..._OR_NULL)
4173 return memcmp(rold
, rcur
, offsetof(struct bpf_reg_state
, id
)) == 0 &&
4174 range_within(rold
, rcur
) &&
4175 tnum_in(rold
->var_off
, rcur
->var_off
);
4176 case PTR_TO_MAP_VALUE_OR_NULL
:
4177 /* a PTR_TO_MAP_VALUE could be safe to use as a
4178 * PTR_TO_MAP_VALUE_OR_NULL into the same map.
4179 * However, if the old PTR_TO_MAP_VALUE_OR_NULL then got NULL-
4180 * checked, doing so could have affected others with the same
4181 * id, and we can't check for that because we lost the id when
4182 * we converted to a PTR_TO_MAP_VALUE.
4184 if (rcur
->type
!= PTR_TO_MAP_VALUE_OR_NULL
)
4186 if (memcmp(rold
, rcur
, offsetof(struct bpf_reg_state
, id
)))
4188 /* Check our ids match any regs they're supposed to */
4189 return check_ids(rold
->id
, rcur
->id
, idmap
);
4190 case PTR_TO_PACKET_META
:
4192 if (rcur
->type
!= rold
->type
)
4194 /* We must have at least as much range as the old ptr
4195 * did, so that any accesses which were safe before are
4196 * still safe. This is true even if old range < old off,
4197 * since someone could have accessed through (ptr - k), or
4198 * even done ptr -= k in a register, to get a safe access.
4200 if (rold
->range
> rcur
->range
)
4202 /* If the offsets don't match, we can't trust our alignment;
4203 * nor can we be sure that we won't fall out of range.
4205 if (rold
->off
!= rcur
->off
)
4207 /* id relations must be preserved */
4208 if (rold
->id
&& !check_ids(rold
->id
, rcur
->id
, idmap
))
4210 /* new val must satisfy old val knowledge */
4211 return range_within(rold
, rcur
) &&
4212 tnum_in(rold
->var_off
, rcur
->var_off
);
4214 case CONST_PTR_TO_MAP
:
4215 case PTR_TO_PACKET_END
:
4216 /* Only valid matches are exact, which memcmp() above
4217 * would have accepted
4220 /* Don't know what's going on, just say it's not safe */
4224 /* Shouldn't get here; if we do, say it's not safe */
4229 static bool stacksafe(struct bpf_func_state
*old
,
4230 struct bpf_func_state
*cur
,
4231 struct idpair
*idmap
)
4235 /* if explored stack has more populated slots than current stack
4236 * such stacks are not equivalent
4238 if (old
->allocated_stack
> cur
->allocated_stack
)
4241 /* walk slots of the explored stack and ignore any additional
4242 * slots in the current stack, since explored(safe) state
4245 for (i
= 0; i
< old
->allocated_stack
; i
++) {
4246 spi
= i
/ BPF_REG_SIZE
;
4248 if (!(old
->stack
[spi
].spilled_ptr
.live
& REG_LIVE_READ
))
4249 /* explored state didn't use this */
4252 if (old
->stack
[spi
].slot_type
[i
% BPF_REG_SIZE
] == STACK_INVALID
)
4254 /* if old state was safe with misc data in the stack
4255 * it will be safe with zero-initialized stack.
4256 * The opposite is not true
4258 if (old
->stack
[spi
].slot_type
[i
% BPF_REG_SIZE
] == STACK_MISC
&&
4259 cur
->stack
[spi
].slot_type
[i
% BPF_REG_SIZE
] == STACK_ZERO
)
4261 if (old
->stack
[spi
].slot_type
[i
% BPF_REG_SIZE
] !=
4262 cur
->stack
[spi
].slot_type
[i
% BPF_REG_SIZE
])
4263 /* Ex: old explored (safe) state has STACK_SPILL in
4264 * this stack slot, but current has has STACK_MISC ->
4265 * this verifier states are not equivalent,
4266 * return false to continue verification of this path
4269 if (i
% BPF_REG_SIZE
)
4271 if (old
->stack
[spi
].slot_type
[0] != STACK_SPILL
)
4273 if (!regsafe(&old
->stack
[spi
].spilled_ptr
,
4274 &cur
->stack
[spi
].spilled_ptr
,
4276 /* when explored and current stack slot are both storing
4277 * spilled registers, check that stored pointers types
4278 * are the same as well.
4279 * Ex: explored safe path could have stored
4280 * (bpf_reg_state) {.type = PTR_TO_STACK, .off = -8}
4281 * but current path has stored:
4282 * (bpf_reg_state) {.type = PTR_TO_STACK, .off = -16}
4283 * such verifier states are not equivalent.
4284 * return false to continue verification of this path
4291 /* compare two verifier states
4293 * all states stored in state_list are known to be valid, since
4294 * verifier reached 'bpf_exit' instruction through them
4296 * this function is called when verifier exploring different branches of
4297 * execution popped from the state stack. If it sees an old state that has
4298 * more strict register state and more strict stack state then this execution
4299 * branch doesn't need to be explored further, since verifier already
4300 * concluded that more strict state leads to valid finish.
4302 * Therefore two states are equivalent if register state is more conservative
4303 * and explored stack state is more conservative than the current one.
4306 * (slot1=INV slot2=MISC) == (slot1=MISC slot2=MISC)
4307 * (slot1=MISC slot2=MISC) != (slot1=INV slot2=MISC)
4309 * In other words if current stack state (one being explored) has more
4310 * valid slots than old one that already passed validation, it means
4311 * the verifier can stop exploring and conclude that current state is valid too
4313 * Similarly with registers. If explored state has register type as invalid
4314 * whereas register type in current state is meaningful, it means that
4315 * the current state will reach 'bpf_exit' instruction safely
4317 static bool func_states_equal(struct bpf_func_state
*old
,
4318 struct bpf_func_state
*cur
)
4320 struct idpair
*idmap
;
4324 idmap
= kcalloc(ID_MAP_SIZE
, sizeof(struct idpair
), GFP_KERNEL
);
4325 /* If we failed to allocate the idmap, just say it's not safe */
4329 for (i
= 0; i
< MAX_BPF_REG
; i
++) {
4330 if (!regsafe(&old
->regs
[i
], &cur
->regs
[i
], idmap
))
4334 if (!stacksafe(old
, cur
, idmap
))
4342 static bool states_equal(struct bpf_verifier_env
*env
,
4343 struct bpf_verifier_state
*old
,
4344 struct bpf_verifier_state
*cur
)
4348 if (old
->curframe
!= cur
->curframe
)
4351 /* for states to be equal callsites have to be the same
4352 * and all frame states need to be equivalent
4354 for (i
= 0; i
<= old
->curframe
; i
++) {
4355 if (old
->frame
[i
]->callsite
!= cur
->frame
[i
]->callsite
)
4357 if (!func_states_equal(old
->frame
[i
], cur
->frame
[i
]))
4363 /* A write screens off any subsequent reads; but write marks come from the
4364 * straight-line code between a state and its parent. When we arrive at an
4365 * equivalent state (jump target or such) we didn't arrive by the straight-line
4366 * code, so read marks in the state must propagate to the parent regardless
4367 * of the state's write marks. That's what 'parent == state->parent' comparison
4368 * in mark_reg_read() and mark_stack_slot_read() is for.
4370 static int propagate_liveness(struct bpf_verifier_env
*env
,
4371 const struct bpf_verifier_state
*vstate
,
4372 struct bpf_verifier_state
*vparent
)
4374 int i
, frame
, err
= 0;
4375 struct bpf_func_state
*state
, *parent
;
4377 if (vparent
->curframe
!= vstate
->curframe
) {
4378 WARN(1, "propagate_live: parent frame %d current frame %d\n",
4379 vparent
->curframe
, vstate
->curframe
);
4382 /* Propagate read liveness of registers... */
4383 BUILD_BUG_ON(BPF_REG_FP
+ 1 != MAX_BPF_REG
);
4384 /* We don't need to worry about FP liveness because it's read-only */
4385 for (i
= 0; i
< BPF_REG_FP
; i
++) {
4386 if (vparent
->frame
[vparent
->curframe
]->regs
[i
].live
& REG_LIVE_READ
)
4388 if (vstate
->frame
[vstate
->curframe
]->regs
[i
].live
& REG_LIVE_READ
) {
4389 err
= mark_reg_read(env
, vstate
, vparent
, i
);
4395 /* ... and stack slots */
4396 for (frame
= 0; frame
<= vstate
->curframe
; frame
++) {
4397 state
= vstate
->frame
[frame
];
4398 parent
= vparent
->frame
[frame
];
4399 for (i
= 0; i
< state
->allocated_stack
/ BPF_REG_SIZE
&&
4400 i
< parent
->allocated_stack
/ BPF_REG_SIZE
; i
++) {
4401 if (parent
->stack
[i
].spilled_ptr
.live
& REG_LIVE_READ
)
4403 if (state
->stack
[i
].spilled_ptr
.live
& REG_LIVE_READ
)
4404 mark_stack_slot_read(env
, vstate
, vparent
, i
, frame
);
4410 static int is_state_visited(struct bpf_verifier_env
*env
, int insn_idx
)
4412 struct bpf_verifier_state_list
*new_sl
;
4413 struct bpf_verifier_state_list
*sl
;
4414 struct bpf_verifier_state
*cur
= env
->cur_state
;
4417 sl
= env
->explored_states
[insn_idx
];
4419 /* this 'insn_idx' instruction wasn't marked, so we will not
4420 * be doing state search here
4424 while (sl
!= STATE_LIST_MARK
) {
4425 if (states_equal(env
, &sl
->state
, cur
)) {
4426 /* reached equivalent register/stack state,
4428 * Registers read by the continuation are read by us.
4429 * If we have any write marks in env->cur_state, they
4430 * will prevent corresponding reads in the continuation
4431 * from reaching our parent (an explored_state). Our
4432 * own state will get the read marks recorded, but
4433 * they'll be immediately forgotten as we're pruning
4434 * this state and will pop a new one.
4436 err
= propagate_liveness(env
, &sl
->state
, cur
);
4444 /* there were no equivalent states, remember current one.
4445 * technically the current state is not proven to be safe yet,
4446 * but it will either reach outer most bpf_exit (which means it's safe)
4447 * or it will be rejected. Since there are no loops, we won't be
4448 * seeing this tuple (frame[0].callsite, frame[1].callsite, .. insn_idx)
4449 * again on the way to bpf_exit
4451 new_sl
= kzalloc(sizeof(struct bpf_verifier_state_list
), GFP_KERNEL
);
4455 /* add new state to the head of linked list */
4456 err
= copy_verifier_state(&new_sl
->state
, cur
);
4458 free_verifier_state(&new_sl
->state
, false);
4462 new_sl
->next
= env
->explored_states
[insn_idx
];
4463 env
->explored_states
[insn_idx
] = new_sl
;
4464 /* connect new state to parentage chain */
4465 cur
->parent
= &new_sl
->state
;
4466 /* clear write marks in current state: the writes we did are not writes
4467 * our child did, so they don't screen off its reads from us.
4468 * (There are no read marks in current state, because reads always mark
4469 * their parent and current state never has children yet. Only
4470 * explored_states can get read marks.)
4472 for (i
= 0; i
< BPF_REG_FP
; i
++)
4473 cur
->frame
[cur
->curframe
]->regs
[i
].live
= REG_LIVE_NONE
;
4475 /* all stack frames are accessible from callee, clear them all */
4476 for (j
= 0; j
<= cur
->curframe
; j
++) {
4477 struct bpf_func_state
*frame
= cur
->frame
[j
];
4479 for (i
= 0; i
< frame
->allocated_stack
/ BPF_REG_SIZE
; i
++)
4480 frame
->stack
[i
].spilled_ptr
.live
= REG_LIVE_NONE
;
4485 static int do_check(struct bpf_verifier_env
*env
)
4487 struct bpf_verifier_state
*state
;
4488 struct bpf_insn
*insns
= env
->prog
->insnsi
;
4489 struct bpf_reg_state
*regs
;
4490 int insn_cnt
= env
->prog
->len
, i
;
4491 int insn_idx
, prev_insn_idx
= 0;
4492 int insn_processed
= 0;
4493 bool do_print_state
= false;
4495 state
= kzalloc(sizeof(struct bpf_verifier_state
), GFP_KERNEL
);
4498 state
->curframe
= 0;
4499 state
->parent
= NULL
;
4500 state
->frame
[0] = kzalloc(sizeof(struct bpf_func_state
), GFP_KERNEL
);
4501 if (!state
->frame
[0]) {
4505 env
->cur_state
= state
;
4506 init_func_state(env
, state
->frame
[0],
4507 BPF_MAIN_FUNC
/* callsite */,
4509 0 /* subprogno, zero == main subprog */);
4512 struct bpf_insn
*insn
;
4516 if (insn_idx
>= insn_cnt
) {
4517 verbose(env
, "invalid insn idx %d insn_cnt %d\n",
4518 insn_idx
, insn_cnt
);
4522 insn
= &insns
[insn_idx
];
4523 class = BPF_CLASS(insn
->code
);
4525 if (++insn_processed
> BPF_COMPLEXITY_LIMIT_INSNS
) {
4527 "BPF program is too large. Processed %d insn\n",
4532 err
= is_state_visited(env
, insn_idx
);
4536 /* found equivalent state, can prune the search */
4537 if (env
->log
.level
) {
4539 verbose(env
, "\nfrom %d to %d: safe\n",
4540 prev_insn_idx
, insn_idx
);
4542 verbose(env
, "%d: safe\n", insn_idx
);
4544 goto process_bpf_exit
;
4550 if (env
->log
.level
> 1 || (env
->log
.level
&& do_print_state
)) {
4551 if (env
->log
.level
> 1)
4552 verbose(env
, "%d:", insn_idx
);
4554 verbose(env
, "\nfrom %d to %d:",
4555 prev_insn_idx
, insn_idx
);
4556 print_verifier_state(env
, state
->frame
[state
->curframe
]);
4557 do_print_state
= false;
4560 if (env
->log
.level
) {
4561 const struct bpf_insn_cbs cbs
= {
4562 .cb_print
= verbose
,
4565 verbose(env
, "%d: ", insn_idx
);
4566 print_bpf_insn(&cbs
, env
, insn
, env
->allow_ptr_leaks
);
4569 if (bpf_prog_is_dev_bound(env
->prog
->aux
)) {
4570 err
= bpf_prog_offload_verify_insn(env
, insn_idx
,
4576 regs
= cur_regs(env
);
4577 env
->insn_aux_data
[insn_idx
].seen
= true;
4578 if (class == BPF_ALU
|| class == BPF_ALU64
) {
4579 err
= check_alu_op(env
, insn
);
4583 } else if (class == BPF_LDX
) {
4584 enum bpf_reg_type
*prev_src_type
, src_reg_type
;
4586 /* check for reserved fields is already done */
4588 /* check src operand */
4589 err
= check_reg_arg(env
, insn
->src_reg
, SRC_OP
);
4593 err
= check_reg_arg(env
, insn
->dst_reg
, DST_OP_NO_MARK
);
4597 src_reg_type
= regs
[insn
->src_reg
].type
;
4599 /* check that memory (src_reg + off) is readable,
4600 * the state of dst_reg will be updated by this func
4602 err
= check_mem_access(env
, insn_idx
, insn
->src_reg
, insn
->off
,
4603 BPF_SIZE(insn
->code
), BPF_READ
,
4608 prev_src_type
= &env
->insn_aux_data
[insn_idx
].ptr_type
;
4610 if (*prev_src_type
== NOT_INIT
) {
4612 * dst_reg = *(u32 *)(src_reg + off)
4613 * save type to validate intersecting paths
4615 *prev_src_type
= src_reg_type
;
4617 } else if (src_reg_type
!= *prev_src_type
&&
4618 (src_reg_type
== PTR_TO_CTX
||
4619 *prev_src_type
== PTR_TO_CTX
)) {
4620 /* ABuser program is trying to use the same insn
4621 * dst_reg = *(u32*) (src_reg + off)
4622 * with different pointer types:
4623 * src_reg == ctx in one branch and
4624 * src_reg == stack|map in some other branch.
4627 verbose(env
, "same insn cannot be used with different pointers\n");
4631 } else if (class == BPF_STX
) {
4632 enum bpf_reg_type
*prev_dst_type
, dst_reg_type
;
4634 if (BPF_MODE(insn
->code
) == BPF_XADD
) {
4635 err
= check_xadd(env
, insn_idx
, insn
);
4642 /* check src1 operand */
4643 err
= check_reg_arg(env
, insn
->src_reg
, SRC_OP
);
4646 /* check src2 operand */
4647 err
= check_reg_arg(env
, insn
->dst_reg
, SRC_OP
);
4651 dst_reg_type
= regs
[insn
->dst_reg
].type
;
4653 /* check that memory (dst_reg + off) is writeable */
4654 err
= check_mem_access(env
, insn_idx
, insn
->dst_reg
, insn
->off
,
4655 BPF_SIZE(insn
->code
), BPF_WRITE
,
4660 prev_dst_type
= &env
->insn_aux_data
[insn_idx
].ptr_type
;
4662 if (*prev_dst_type
== NOT_INIT
) {
4663 *prev_dst_type
= dst_reg_type
;
4664 } else if (dst_reg_type
!= *prev_dst_type
&&
4665 (dst_reg_type
== PTR_TO_CTX
||
4666 *prev_dst_type
== PTR_TO_CTX
)) {
4667 verbose(env
, "same insn cannot be used with different pointers\n");
4671 } else if (class == BPF_ST
) {
4672 if (BPF_MODE(insn
->code
) != BPF_MEM
||
4673 insn
->src_reg
!= BPF_REG_0
) {
4674 verbose(env
, "BPF_ST uses reserved fields\n");
4677 /* check src operand */
4678 err
= check_reg_arg(env
, insn
->dst_reg
, SRC_OP
);
4682 if (is_ctx_reg(env
, insn
->dst_reg
)) {
4683 verbose(env
, "BPF_ST stores into R%d context is not allowed\n",
4688 /* check that memory (dst_reg + off) is writeable */
4689 err
= check_mem_access(env
, insn_idx
, insn
->dst_reg
, insn
->off
,
4690 BPF_SIZE(insn
->code
), BPF_WRITE
,
4695 } else if (class == BPF_JMP
) {
4696 u8 opcode
= BPF_OP(insn
->code
);
4698 if (opcode
== BPF_CALL
) {
4699 if (BPF_SRC(insn
->code
) != BPF_K
||
4701 (insn
->src_reg
!= BPF_REG_0
&&
4702 insn
->src_reg
!= BPF_PSEUDO_CALL
) ||
4703 insn
->dst_reg
!= BPF_REG_0
) {
4704 verbose(env
, "BPF_CALL uses reserved fields\n");
4708 if (insn
->src_reg
== BPF_PSEUDO_CALL
)
4709 err
= check_func_call(env
, insn
, &insn_idx
);
4711 err
= check_helper_call(env
, insn
->imm
, insn_idx
);
4715 } else if (opcode
== BPF_JA
) {
4716 if (BPF_SRC(insn
->code
) != BPF_K
||
4718 insn
->src_reg
!= BPF_REG_0
||
4719 insn
->dst_reg
!= BPF_REG_0
) {
4720 verbose(env
, "BPF_JA uses reserved fields\n");
4724 insn_idx
+= insn
->off
+ 1;
4727 } else if (opcode
== BPF_EXIT
) {
4728 if (BPF_SRC(insn
->code
) != BPF_K
||
4730 insn
->src_reg
!= BPF_REG_0
||
4731 insn
->dst_reg
!= BPF_REG_0
) {
4732 verbose(env
, "BPF_EXIT uses reserved fields\n");
4736 if (state
->curframe
) {
4737 /* exit from nested function */
4738 prev_insn_idx
= insn_idx
;
4739 err
= prepare_func_exit(env
, &insn_idx
);
4742 do_print_state
= true;
4746 /* eBPF calling convetion is such that R0 is used
4747 * to return the value from eBPF program.
4748 * Make sure that it's readable at this time
4749 * of bpf_exit, which means that program wrote
4750 * something into it earlier
4752 err
= check_reg_arg(env
, BPF_REG_0
, SRC_OP
);
4756 if (is_pointer_value(env
, BPF_REG_0
)) {
4757 verbose(env
, "R0 leaks addr as return value\n");
4761 err
= check_return_code(env
);
4765 err
= pop_stack(env
, &prev_insn_idx
, &insn_idx
);
4771 do_print_state
= true;
4775 err
= check_cond_jmp_op(env
, insn
, &insn_idx
);
4779 } else if (class == BPF_LD
) {
4780 u8 mode
= BPF_MODE(insn
->code
);
4782 if (mode
== BPF_ABS
|| mode
== BPF_IND
) {
4783 err
= check_ld_abs(env
, insn
);
4787 } else if (mode
== BPF_IMM
) {
4788 err
= check_ld_imm(env
, insn
);
4793 env
->insn_aux_data
[insn_idx
].seen
= true;
4795 verbose(env
, "invalid BPF_LD mode\n");
4799 verbose(env
, "unknown insn class %d\n", class);
4806 verbose(env
, "processed %d insns, stack depth ", insn_processed
);
4807 for (i
= 0; i
< env
->subprog_cnt
+ 1; i
++) {
4808 u32 depth
= env
->subprog_stack_depth
[i
];
4810 verbose(env
, "%d", depth
);
4811 if (i
+ 1 < env
->subprog_cnt
+ 1)
4815 env
->prog
->aux
->stack_depth
= env
->subprog_stack_depth
[0];
4819 static int check_map_prealloc(struct bpf_map
*map
)
4821 return (map
->map_type
!= BPF_MAP_TYPE_HASH
&&
4822 map
->map_type
!= BPF_MAP_TYPE_PERCPU_HASH
&&
4823 map
->map_type
!= BPF_MAP_TYPE_HASH_OF_MAPS
) ||
4824 !(map
->map_flags
& BPF_F_NO_PREALLOC
);
4827 static int check_map_prog_compatibility(struct bpf_verifier_env
*env
,
4828 struct bpf_map
*map
,
4829 struct bpf_prog
*prog
)
4832 /* Make sure that BPF_PROG_TYPE_PERF_EVENT programs only use
4833 * preallocated hash maps, since doing memory allocation
4834 * in overflow_handler can crash depending on where nmi got
4837 if (prog
->type
== BPF_PROG_TYPE_PERF_EVENT
) {
4838 if (!check_map_prealloc(map
)) {
4839 verbose(env
, "perf_event programs can only use preallocated hash map\n");
4842 if (map
->inner_map_meta
&&
4843 !check_map_prealloc(map
->inner_map_meta
)) {
4844 verbose(env
, "perf_event programs can only use preallocated inner hash map\n");
4849 if ((bpf_prog_is_dev_bound(prog
->aux
) || bpf_map_is_dev_bound(map
)) &&
4850 !bpf_offload_dev_match(prog
, map
)) {
4851 verbose(env
, "offload device mismatch between prog and map\n");
4858 /* look for pseudo eBPF instructions that access map FDs and
4859 * replace them with actual map pointers
4861 static int replace_map_fd_with_map_ptr(struct bpf_verifier_env
*env
)
4863 struct bpf_insn
*insn
= env
->prog
->insnsi
;
4864 int insn_cnt
= env
->prog
->len
;
4867 err
= bpf_prog_calc_tag(env
->prog
);
4871 for (i
= 0; i
< insn_cnt
; i
++, insn
++) {
4872 if (BPF_CLASS(insn
->code
) == BPF_LDX
&&
4873 (BPF_MODE(insn
->code
) != BPF_MEM
|| insn
->imm
!= 0)) {
4874 verbose(env
, "BPF_LDX uses reserved fields\n");
4878 if (BPF_CLASS(insn
->code
) == BPF_STX
&&
4879 ((BPF_MODE(insn
->code
) != BPF_MEM
&&
4880 BPF_MODE(insn
->code
) != BPF_XADD
) || insn
->imm
!= 0)) {
4881 verbose(env
, "BPF_STX uses reserved fields\n");
4885 if (insn
[0].code
== (BPF_LD
| BPF_IMM
| BPF_DW
)) {
4886 struct bpf_map
*map
;
4889 if (i
== insn_cnt
- 1 || insn
[1].code
!= 0 ||
4890 insn
[1].dst_reg
!= 0 || insn
[1].src_reg
!= 0 ||
4892 verbose(env
, "invalid bpf_ld_imm64 insn\n");
4896 if (insn
->src_reg
== 0)
4897 /* valid generic load 64-bit imm */
4900 if (insn
->src_reg
!= BPF_PSEUDO_MAP_FD
) {
4902 "unrecognized bpf_ld_imm64 insn\n");
4906 f
= fdget(insn
->imm
);
4907 map
= __bpf_map_get(f
);
4909 verbose(env
, "fd %d is not pointing to valid bpf_map\n",
4911 return PTR_ERR(map
);
4914 err
= check_map_prog_compatibility(env
, map
, env
->prog
);
4920 /* store map pointer inside BPF_LD_IMM64 instruction */
4921 insn
[0].imm
= (u32
) (unsigned long) map
;
4922 insn
[1].imm
= ((u64
) (unsigned long) map
) >> 32;
4924 /* check whether we recorded this map already */
4925 for (j
= 0; j
< env
->used_map_cnt
; j
++)
4926 if (env
->used_maps
[j
] == map
) {
4931 if (env
->used_map_cnt
>= MAX_USED_MAPS
) {
4936 /* hold the map. If the program is rejected by verifier,
4937 * the map will be released by release_maps() or it
4938 * will be used by the valid program until it's unloaded
4939 * and all maps are released in free_bpf_prog_info()
4941 map
= bpf_map_inc(map
, false);
4944 return PTR_ERR(map
);
4946 env
->used_maps
[env
->used_map_cnt
++] = map
;
4955 /* now all pseudo BPF_LD_IMM64 instructions load valid
4956 * 'struct bpf_map *' into a register instead of user map_fd.
4957 * These pointers will be used later by verifier to validate map access.
4962 /* drop refcnt of maps used by the rejected program */
4963 static void release_maps(struct bpf_verifier_env
*env
)
4967 for (i
= 0; i
< env
->used_map_cnt
; i
++)
4968 bpf_map_put(env
->used_maps
[i
]);
4971 /* convert pseudo BPF_LD_IMM64 into generic BPF_LD_IMM64 */
4972 static void convert_pseudo_ld_imm64(struct bpf_verifier_env
*env
)
4974 struct bpf_insn
*insn
= env
->prog
->insnsi
;
4975 int insn_cnt
= env
->prog
->len
;
4978 for (i
= 0; i
< insn_cnt
; i
++, insn
++)
4979 if (insn
->code
== (BPF_LD
| BPF_IMM
| BPF_DW
))
4983 /* single env->prog->insni[off] instruction was replaced with the range
4984 * insni[off, off + cnt). Adjust corresponding insn_aux_data by copying
4985 * [0, off) and [off, end) to new locations, so the patched range stays zero
4987 static int adjust_insn_aux_data(struct bpf_verifier_env
*env
, u32 prog_len
,
4990 struct bpf_insn_aux_data
*new_data
, *old_data
= env
->insn_aux_data
;
4995 new_data
= vzalloc(sizeof(struct bpf_insn_aux_data
) * prog_len
);
4998 memcpy(new_data
, old_data
, sizeof(struct bpf_insn_aux_data
) * off
);
4999 memcpy(new_data
+ off
+ cnt
- 1, old_data
+ off
,
5000 sizeof(struct bpf_insn_aux_data
) * (prog_len
- off
- cnt
+ 1));
5001 for (i
= off
; i
< off
+ cnt
- 1; i
++)
5002 new_data
[i
].seen
= true;
5003 env
->insn_aux_data
= new_data
;
5008 static void adjust_subprog_starts(struct bpf_verifier_env
*env
, u32 off
, u32 len
)
5014 for (i
= 0; i
< env
->subprog_cnt
; i
++) {
5015 if (env
->subprog_starts
[i
] < off
)
5017 env
->subprog_starts
[i
] += len
- 1;
5021 static struct bpf_prog
*bpf_patch_insn_data(struct bpf_verifier_env
*env
, u32 off
,
5022 const struct bpf_insn
*patch
, u32 len
)
5024 struct bpf_prog
*new_prog
;
5026 new_prog
= bpf_patch_insn_single(env
->prog
, off
, patch
, len
);
5029 if (adjust_insn_aux_data(env
, new_prog
->len
, off
, len
))
5031 adjust_subprog_starts(env
, off
, len
);
5035 /* The verifier does more data flow analysis than llvm and will not explore
5036 * branches that are dead at run time. Malicious programs can have dead code
5037 * too. Therefore replace all dead at-run-time code with nops.
5039 static void sanitize_dead_code(struct bpf_verifier_env
*env
)
5041 struct bpf_insn_aux_data
*aux_data
= env
->insn_aux_data
;
5042 struct bpf_insn nop
= BPF_MOV64_REG(BPF_REG_0
, BPF_REG_0
);
5043 struct bpf_insn
*insn
= env
->prog
->insnsi
;
5044 const int insn_cnt
= env
->prog
->len
;
5047 for (i
= 0; i
< insn_cnt
; i
++) {
5048 if (aux_data
[i
].seen
)
5050 memcpy(insn
+ i
, &nop
, sizeof(nop
));
5054 /* convert load instructions that access fields of 'struct __sk_buff'
5055 * into sequence of instructions that access fields of 'struct sk_buff'
5057 static int convert_ctx_accesses(struct bpf_verifier_env
*env
)
5059 const struct bpf_verifier_ops
*ops
= env
->ops
;
5060 int i
, cnt
, size
, ctx_field_size
, delta
= 0;
5061 const int insn_cnt
= env
->prog
->len
;
5062 struct bpf_insn insn_buf
[16], *insn
;
5063 struct bpf_prog
*new_prog
;
5064 enum bpf_access_type type
;
5065 bool is_narrower_load
;
5068 if (ops
->gen_prologue
) {
5069 cnt
= ops
->gen_prologue(insn_buf
, env
->seen_direct_write
,
5071 if (cnt
>= ARRAY_SIZE(insn_buf
)) {
5072 verbose(env
, "bpf verifier is misconfigured\n");
5075 new_prog
= bpf_patch_insn_data(env
, 0, insn_buf
, cnt
);
5079 env
->prog
= new_prog
;
5084 if (!ops
->convert_ctx_access
)
5087 insn
= env
->prog
->insnsi
+ delta
;
5089 for (i
= 0; i
< insn_cnt
; i
++, insn
++) {
5090 if (insn
->code
== (BPF_LDX
| BPF_MEM
| BPF_B
) ||
5091 insn
->code
== (BPF_LDX
| BPF_MEM
| BPF_H
) ||
5092 insn
->code
== (BPF_LDX
| BPF_MEM
| BPF_W
) ||
5093 insn
->code
== (BPF_LDX
| BPF_MEM
| BPF_DW
))
5095 else if (insn
->code
== (BPF_STX
| BPF_MEM
| BPF_B
) ||
5096 insn
->code
== (BPF_STX
| BPF_MEM
| BPF_H
) ||
5097 insn
->code
== (BPF_STX
| BPF_MEM
| BPF_W
) ||
5098 insn
->code
== (BPF_STX
| BPF_MEM
| BPF_DW
))
5103 if (env
->insn_aux_data
[i
+ delta
].ptr_type
!= PTR_TO_CTX
)
5106 ctx_field_size
= env
->insn_aux_data
[i
+ delta
].ctx_field_size
;
5107 size
= BPF_LDST_BYTES(insn
);
5109 /* If the read access is a narrower load of the field,
5110 * convert to a 4/8-byte load, to minimum program type specific
5111 * convert_ctx_access changes. If conversion is successful,
5112 * we will apply proper mask to the result.
5114 is_narrower_load
= size
< ctx_field_size
;
5115 if (is_narrower_load
) {
5116 u32 off
= insn
->off
;
5119 if (type
== BPF_WRITE
) {
5120 verbose(env
, "bpf verifier narrow ctx access misconfigured\n");
5125 if (ctx_field_size
== 4)
5127 else if (ctx_field_size
== 8)
5130 insn
->off
= off
& ~(ctx_field_size
- 1);
5131 insn
->code
= BPF_LDX
| BPF_MEM
| size_code
;
5135 cnt
= ops
->convert_ctx_access(type
, insn
, insn_buf
, env
->prog
,
5137 if (cnt
== 0 || cnt
>= ARRAY_SIZE(insn_buf
) ||
5138 (ctx_field_size
&& !target_size
)) {
5139 verbose(env
, "bpf verifier is misconfigured\n");
5143 if (is_narrower_load
&& size
< target_size
) {
5144 if (ctx_field_size
<= 4)
5145 insn_buf
[cnt
++] = BPF_ALU32_IMM(BPF_AND
, insn
->dst_reg
,
5146 (1 << size
* 8) - 1);
5148 insn_buf
[cnt
++] = BPF_ALU64_IMM(BPF_AND
, insn
->dst_reg
,
5149 (1 << size
* 8) - 1);
5152 new_prog
= bpf_patch_insn_data(env
, i
+ delta
, insn_buf
, cnt
);
5158 /* keep walking new program and skip insns we just inserted */
5159 env
->prog
= new_prog
;
5160 insn
= new_prog
->insnsi
+ i
+ delta
;
5166 static int jit_subprogs(struct bpf_verifier_env
*env
)
5168 struct bpf_prog
*prog
= env
->prog
, **func
, *tmp
;
5169 int i
, j
, subprog_start
, subprog_end
= 0, len
, subprog
;
5170 struct bpf_insn
*insn
;
5174 if (env
->subprog_cnt
== 0)
5177 for (i
= 0, insn
= prog
->insnsi
; i
< prog
->len
; i
++, insn
++) {
5178 if (insn
->code
!= (BPF_JMP
| BPF_CALL
) ||
5179 insn
->src_reg
!= BPF_PSEUDO_CALL
)
5181 subprog
= find_subprog(env
, i
+ insn
->imm
+ 1);
5183 WARN_ONCE(1, "verifier bug. No program starts at insn %d\n",
5187 /* temporarily remember subprog id inside insn instead of
5188 * aux_data, since next loop will split up all insns into funcs
5190 insn
->off
= subprog
+ 1;
5191 /* remember original imm in case JIT fails and fallback
5192 * to interpreter will be needed
5194 env
->insn_aux_data
[i
].call_imm
= insn
->imm
;
5195 /* point imm to __bpf_call_base+1 from JITs point of view */
5199 func
= kzalloc(sizeof(prog
) * (env
->subprog_cnt
+ 1), GFP_KERNEL
);
5203 for (i
= 0; i
<= env
->subprog_cnt
; i
++) {
5204 subprog_start
= subprog_end
;
5205 if (env
->subprog_cnt
== i
)
5206 subprog_end
= prog
->len
;
5208 subprog_end
= env
->subprog_starts
[i
];
5210 len
= subprog_end
- subprog_start
;
5211 func
[i
] = bpf_prog_alloc(bpf_prog_size(len
), GFP_USER
);
5214 memcpy(func
[i
]->insnsi
, &prog
->insnsi
[subprog_start
],
5215 len
* sizeof(struct bpf_insn
));
5216 func
[i
]->type
= prog
->type
;
5218 if (bpf_prog_calc_tag(func
[i
]))
5220 func
[i
]->is_func
= 1;
5221 /* Use bpf_prog_F_tag to indicate functions in stack traces.
5222 * Long term would need debug info to populate names
5224 func
[i
]->aux
->name
[0] = 'F';
5225 func
[i
]->aux
->stack_depth
= env
->subprog_stack_depth
[i
];
5226 func
[i
]->jit_requested
= 1;
5227 func
[i
] = bpf_int_jit_compile(func
[i
]);
5228 if (!func
[i
]->jited
) {
5234 /* at this point all bpf functions were successfully JITed
5235 * now populate all bpf_calls with correct addresses and
5236 * run last pass of JIT
5238 for (i
= 0; i
<= env
->subprog_cnt
; i
++) {
5239 insn
= func
[i
]->insnsi
;
5240 for (j
= 0; j
< func
[i
]->len
; j
++, insn
++) {
5241 if (insn
->code
!= (BPF_JMP
| BPF_CALL
) ||
5242 insn
->src_reg
!= BPF_PSEUDO_CALL
)
5244 subprog
= insn
->off
;
5246 insn
->imm
= (u64 (*)(u64
, u64
, u64
, u64
, u64
))
5247 func
[subprog
]->bpf_func
-
5251 for (i
= 0; i
<= env
->subprog_cnt
; i
++) {
5252 old_bpf_func
= func
[i
]->bpf_func
;
5253 tmp
= bpf_int_jit_compile(func
[i
]);
5254 if (tmp
!= func
[i
] || func
[i
]->bpf_func
!= old_bpf_func
) {
5255 verbose(env
, "JIT doesn't support bpf-to-bpf calls\n");
5262 /* finally lock prog and jit images for all functions and
5265 for (i
= 0; i
<= env
->subprog_cnt
; i
++) {
5266 bpf_prog_lock_ro(func
[i
]);
5267 bpf_prog_kallsyms_add(func
[i
]);
5270 /* Last step: make now unused interpreter insns from main
5271 * prog consistent for later dump requests, so they can
5272 * later look the same as if they were interpreted only.
5274 for (i
= 0, insn
= prog
->insnsi
; i
< prog
->len
; i
++, insn
++) {
5277 if (insn
->code
!= (BPF_JMP
| BPF_CALL
) ||
5278 insn
->src_reg
!= BPF_PSEUDO_CALL
)
5280 insn
->off
= env
->insn_aux_data
[i
].call_imm
;
5281 subprog
= find_subprog(env
, i
+ insn
->off
+ 1);
5282 addr
= (unsigned long)func
[subprog
+ 1]->bpf_func
;
5284 insn
->imm
= (u64 (*)(u64
, u64
, u64
, u64
, u64
))
5285 addr
- __bpf_call_base
;
5289 prog
->bpf_func
= func
[0]->bpf_func
;
5290 prog
->aux
->func
= func
;
5291 prog
->aux
->func_cnt
= env
->subprog_cnt
+ 1;
5294 for (i
= 0; i
<= env
->subprog_cnt
; i
++)
5296 bpf_jit_free(func
[i
]);
5298 /* cleanup main prog to be interpreted */
5299 prog
->jit_requested
= 0;
5300 for (i
= 0, insn
= prog
->insnsi
; i
< prog
->len
; i
++, insn
++) {
5301 if (insn
->code
!= (BPF_JMP
| BPF_CALL
) ||
5302 insn
->src_reg
!= BPF_PSEUDO_CALL
)
5305 insn
->imm
= env
->insn_aux_data
[i
].call_imm
;
5310 static int fixup_call_args(struct bpf_verifier_env
*env
)
5312 #ifndef CONFIG_BPF_JIT_ALWAYS_ON
5313 struct bpf_prog
*prog
= env
->prog
;
5314 struct bpf_insn
*insn
= prog
->insnsi
;
5320 if (env
->prog
->jit_requested
) {
5321 err
= jit_subprogs(env
);
5325 #ifndef CONFIG_BPF_JIT_ALWAYS_ON
5326 for (i
= 0; i
< prog
->len
; i
++, insn
++) {
5327 if (insn
->code
!= (BPF_JMP
| BPF_CALL
) ||
5328 insn
->src_reg
!= BPF_PSEUDO_CALL
)
5330 depth
= get_callee_stack_depth(env
, insn
, i
);
5333 bpf_patch_call_args(insn
, depth
);
5340 /* fixup insn->imm field of bpf_call instructions
5341 * and inline eligible helpers as explicit sequence of BPF instructions
5343 * this function is called after eBPF program passed verification
5345 static int fixup_bpf_calls(struct bpf_verifier_env
*env
)
5347 struct bpf_prog
*prog
= env
->prog
;
5348 struct bpf_insn
*insn
= prog
->insnsi
;
5349 const struct bpf_func_proto
*fn
;
5350 const int insn_cnt
= prog
->len
;
5351 struct bpf_insn insn_buf
[16];
5352 struct bpf_prog
*new_prog
;
5353 struct bpf_map
*map_ptr
;
5354 int i
, cnt
, delta
= 0;
5356 for (i
= 0; i
< insn_cnt
; i
++, insn
++) {
5357 if (insn
->code
== (BPF_ALU
| BPF_MOD
| BPF_X
) ||
5358 insn
->code
== (BPF_ALU
| BPF_DIV
| BPF_X
)) {
5359 /* due to JIT bugs clear upper 32-bits of src register
5360 * before div/mod operation
5362 insn_buf
[0] = BPF_MOV32_REG(insn
->src_reg
, insn
->src_reg
);
5363 insn_buf
[1] = *insn
;
5365 new_prog
= bpf_patch_insn_data(env
, i
+ delta
, insn_buf
, cnt
);
5370 env
->prog
= prog
= new_prog
;
5371 insn
= new_prog
->insnsi
+ i
+ delta
;
5375 if (insn
->code
!= (BPF_JMP
| BPF_CALL
))
5377 if (insn
->src_reg
== BPF_PSEUDO_CALL
)
5380 if (insn
->imm
== BPF_FUNC_get_route_realm
)
5381 prog
->dst_needed
= 1;
5382 if (insn
->imm
== BPF_FUNC_get_prandom_u32
)
5383 bpf_user_rnd_init_once();
5384 if (insn
->imm
== BPF_FUNC_override_return
)
5385 prog
->kprobe_override
= 1;
5386 if (insn
->imm
== BPF_FUNC_tail_call
) {
5387 /* If we tail call into other programs, we
5388 * cannot make any assumptions since they can
5389 * be replaced dynamically during runtime in
5390 * the program array.
5392 prog
->cb_access
= 1;
5393 env
->prog
->aux
->stack_depth
= MAX_BPF_STACK
;
5395 /* mark bpf_tail_call as different opcode to avoid
5396 * conditional branch in the interpeter for every normal
5397 * call and to prevent accidental JITing by JIT compiler
5398 * that doesn't support bpf_tail_call yet
5401 insn
->code
= BPF_JMP
| BPF_TAIL_CALL
;
5403 /* instead of changing every JIT dealing with tail_call
5404 * emit two extra insns:
5405 * if (index >= max_entries) goto out;
5406 * index &= array->index_mask;
5407 * to avoid out-of-bounds cpu speculation
5409 map_ptr
= env
->insn_aux_data
[i
+ delta
].map_ptr
;
5410 if (map_ptr
== BPF_MAP_PTR_POISON
) {
5411 verbose(env
, "tail_call abusing map_ptr\n");
5414 if (!map_ptr
->unpriv_array
)
5416 insn_buf
[0] = BPF_JMP_IMM(BPF_JGE
, BPF_REG_3
,
5417 map_ptr
->max_entries
, 2);
5418 insn_buf
[1] = BPF_ALU32_IMM(BPF_AND
, BPF_REG_3
,
5419 container_of(map_ptr
,
5422 insn_buf
[2] = *insn
;
5424 new_prog
= bpf_patch_insn_data(env
, i
+ delta
, insn_buf
, cnt
);
5429 env
->prog
= prog
= new_prog
;
5430 insn
= new_prog
->insnsi
+ i
+ delta
;
5434 /* BPF_EMIT_CALL() assumptions in some of the map_gen_lookup
5435 * handlers are currently limited to 64 bit only.
5437 if (prog
->jit_requested
&& BITS_PER_LONG
== 64 &&
5438 insn
->imm
== BPF_FUNC_map_lookup_elem
) {
5439 map_ptr
= env
->insn_aux_data
[i
+ delta
].map_ptr
;
5440 if (map_ptr
== BPF_MAP_PTR_POISON
||
5441 !map_ptr
->ops
->map_gen_lookup
)
5442 goto patch_call_imm
;
5444 cnt
= map_ptr
->ops
->map_gen_lookup(map_ptr
, insn_buf
);
5445 if (cnt
== 0 || cnt
>= ARRAY_SIZE(insn_buf
)) {
5446 verbose(env
, "bpf verifier is misconfigured\n");
5450 new_prog
= bpf_patch_insn_data(env
, i
+ delta
, insn_buf
,
5457 /* keep walking new program and skip insns we just inserted */
5458 env
->prog
= prog
= new_prog
;
5459 insn
= new_prog
->insnsi
+ i
+ delta
;
5463 if (insn
->imm
== BPF_FUNC_redirect_map
) {
5464 /* Note, we cannot use prog directly as imm as subsequent
5465 * rewrites would still change the prog pointer. The only
5466 * stable address we can use is aux, which also works with
5467 * prog clones during blinding.
5469 u64 addr
= (unsigned long)prog
->aux
;
5470 struct bpf_insn r4_ld
[] = {
5471 BPF_LD_IMM64(BPF_REG_4
, addr
),
5474 cnt
= ARRAY_SIZE(r4_ld
);
5476 new_prog
= bpf_patch_insn_data(env
, i
+ delta
, r4_ld
, cnt
);
5481 env
->prog
= prog
= new_prog
;
5482 insn
= new_prog
->insnsi
+ i
+ delta
;
5485 fn
= env
->ops
->get_func_proto(insn
->imm
);
5486 /* all functions that have prototype and verifier allowed
5487 * programs to call them, must be real in-kernel functions
5491 "kernel subsystem misconfigured func %s#%d\n",
5492 func_id_name(insn
->imm
), insn
->imm
);
5495 insn
->imm
= fn
->func
- __bpf_call_base
;
5501 static void free_states(struct bpf_verifier_env
*env
)
5503 struct bpf_verifier_state_list
*sl
, *sln
;
5506 if (!env
->explored_states
)
5509 for (i
= 0; i
< env
->prog
->len
; i
++) {
5510 sl
= env
->explored_states
[i
];
5513 while (sl
!= STATE_LIST_MARK
) {
5515 free_verifier_state(&sl
->state
, false);
5521 kfree(env
->explored_states
);
5524 int bpf_check(struct bpf_prog
**prog
, union bpf_attr
*attr
)
5526 struct bpf_verifier_env
*env
;
5527 struct bpf_verifer_log
*log
;
5530 /* no program is valid */
5531 if (ARRAY_SIZE(bpf_verifier_ops
) == 0)
5534 /* 'struct bpf_verifier_env' can be global, but since it's not small,
5535 * allocate/free it every time bpf_check() is called
5537 env
= kzalloc(sizeof(struct bpf_verifier_env
), GFP_KERNEL
);
5542 env
->insn_aux_data
= vzalloc(sizeof(struct bpf_insn_aux_data
) *
5545 if (!env
->insn_aux_data
)
5548 env
->ops
= bpf_verifier_ops
[env
->prog
->type
];
5550 /* grab the mutex to protect few globals used by verifier */
5551 mutex_lock(&bpf_verifier_lock
);
5553 if (attr
->log_level
|| attr
->log_buf
|| attr
->log_size
) {
5554 /* user requested verbose verifier output
5555 * and supplied buffer to store the verification trace
5557 log
->level
= attr
->log_level
;
5558 log
->ubuf
= (char __user
*) (unsigned long) attr
->log_buf
;
5559 log
->len_total
= attr
->log_size
;
5562 /* log attributes have to be sane */
5563 if (log
->len_total
< 128 || log
->len_total
> UINT_MAX
>> 8 ||
5564 !log
->level
|| !log
->ubuf
)
5568 env
->strict_alignment
= !!(attr
->prog_flags
& BPF_F_STRICT_ALIGNMENT
);
5569 if (!IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS
))
5570 env
->strict_alignment
= true;
5572 if (bpf_prog_is_dev_bound(env
->prog
->aux
)) {
5573 ret
= bpf_prog_offload_verifier_prep(env
);
5578 ret
= replace_map_fd_with_map_ptr(env
);
5580 goto skip_full_check
;
5582 env
->explored_states
= kcalloc(env
->prog
->len
,
5583 sizeof(struct bpf_verifier_state_list
*),
5586 if (!env
->explored_states
)
5587 goto skip_full_check
;
5589 env
->allow_ptr_leaks
= capable(CAP_SYS_ADMIN
);
5591 ret
= check_cfg(env
);
5593 goto skip_full_check
;
5595 ret
= do_check(env
);
5596 if (env
->cur_state
) {
5597 free_verifier_state(env
->cur_state
, true);
5598 env
->cur_state
= NULL
;
5602 while (!pop_stack(env
, NULL
, NULL
));
5606 sanitize_dead_code(env
);
5609 ret
= check_max_stack_depth(env
);
5612 /* program is valid, convert *(u32*)(ctx + off) accesses */
5613 ret
= convert_ctx_accesses(env
);
5616 ret
= fixup_bpf_calls(env
);
5619 ret
= fixup_call_args(env
);
5621 if (log
->level
&& bpf_verifier_log_full(log
))
5623 if (log
->level
&& !log
->ubuf
) {
5625 goto err_release_maps
;
5628 if (ret
== 0 && env
->used_map_cnt
) {
5629 /* if program passed verifier, update used_maps in bpf_prog_info */
5630 env
->prog
->aux
->used_maps
= kmalloc_array(env
->used_map_cnt
,
5631 sizeof(env
->used_maps
[0]),
5634 if (!env
->prog
->aux
->used_maps
) {
5636 goto err_release_maps
;
5639 memcpy(env
->prog
->aux
->used_maps
, env
->used_maps
,
5640 sizeof(env
->used_maps
[0]) * env
->used_map_cnt
);
5641 env
->prog
->aux
->used_map_cnt
= env
->used_map_cnt
;
5643 /* program is valid. Convert pseudo bpf_ld_imm64 into generic
5644 * bpf_ld_imm64 instructions
5646 convert_pseudo_ld_imm64(env
);
5650 if (!env
->prog
->aux
->used_maps
)
5651 /* if we didn't copy map pointers into bpf_prog_info, release
5652 * them now. Otherwise free_bpf_prog_info() will release them.
5657 mutex_unlock(&bpf_verifier_lock
);
5658 vfree(env
->insn_aux_data
);