1 // SPDX-License-Identifier: GPL-2.0-only
2 /* Copyright (c) 2011-2014 PLUMgrid, http://plumgrid.com
3 * Copyright (c) 2016 Facebook
4 * Copyright (c) 2018 Covalent IO, Inc. http://covalent.io
6 #include <uapi/linux/btf.h>
7 #include <linux/kernel.h>
8 #include <linux/types.h>
9 #include <linux/slab.h>
10 #include <linux/bpf.h>
11 #include <linux/btf.h>
12 #include <linux/bpf_verifier.h>
13 #include <linux/filter.h>
14 #include <net/netlink.h>
15 #include <linux/file.h>
16 #include <linux/vmalloc.h>
17 #include <linux/stringify.h>
18 #include <linux/bsearch.h>
19 #include <linux/sort.h>
20 #include <linux/perf_event.h>
21 #include <linux/ctype.h>
22 #include <linux/error-injection.h>
23 #include <linux/bpf_lsm.h>
24 #include <linux/btf_ids.h>
28 static const struct bpf_verifier_ops
* const bpf_verifier_ops
[] = {
29 #define BPF_PROG_TYPE(_id, _name, prog_ctx_type, kern_ctx_type) \
30 [_id] = & _name ## _verifier_ops,
31 #define BPF_MAP_TYPE(_id, _ops)
32 #define BPF_LINK_TYPE(_id, _name)
33 #include <linux/bpf_types.h>
39 /* bpf_check() is a static code analyzer that walks eBPF program
40 * instruction by instruction and updates register/stack state.
41 * All paths of conditional branches are analyzed until 'bpf_exit' insn.
43 * The first pass is depth-first-search to check that the program is a DAG.
44 * It rejects the following programs:
45 * - larger than BPF_MAXINSNS insns
46 * - if loop is present (detected via back-edge)
47 * - unreachable insns exist (shouldn't be a forest. program = one function)
48 * - out of bounds or malformed jumps
49 * The second pass is all possible path descent from the 1st insn.
50 * Since it's analyzing all paths through the program, the length of the
51 * analysis is limited to 64k insn, which may be hit even if total number of
52 * insn is less then 4K, but there are too many branches that change stack/regs.
53 * Number of 'branches to be analyzed' is limited to 1k
55 * On entry to each instruction, each register has a type, and the instruction
56 * changes the types of the registers depending on instruction semantics.
57 * If instruction is BPF_MOV64_REG(BPF_REG_1, BPF_REG_5), then type of R5 is
60 * All registers are 64-bit.
61 * R0 - return register
62 * R1-R5 argument passing registers
63 * R6-R9 callee saved registers
64 * R10 - frame pointer read-only
66 * At the start of BPF program the register R1 contains a pointer to bpf_context
67 * and has type PTR_TO_CTX.
69 * Verifier tracks arithmetic operations on pointers in case:
70 * BPF_MOV64_REG(BPF_REG_1, BPF_REG_10),
71 * BPF_ALU64_IMM(BPF_ADD, BPF_REG_1, -20),
72 * 1st insn copies R10 (which has FRAME_PTR) type into R1
73 * and 2nd arithmetic instruction is pattern matched to recognize
74 * that it wants to construct a pointer to some element within stack.
75 * So after 2nd insn, the register R1 has type PTR_TO_STACK
76 * (and -20 constant is saved for further stack bounds checking).
77 * Meaning that this reg is a pointer to stack plus known immediate constant.
79 * Most of the time the registers have SCALAR_VALUE type, which
80 * means the register has some value, but it's not a valid pointer.
81 * (like pointer plus pointer becomes SCALAR_VALUE type)
83 * When verifier sees load or store instructions the type of base register
84 * can be: PTR_TO_MAP_VALUE, PTR_TO_CTX, PTR_TO_STACK, PTR_TO_SOCKET. These are
85 * four pointer types recognized by check_mem_access() function.
87 * PTR_TO_MAP_VALUE means that this register is pointing to 'map element value'
88 * and the range of [ptr, ptr + map's value_size) is accessible.
90 * registers used to pass values to function calls are checked against
91 * function argument constraints.
93 * ARG_PTR_TO_MAP_KEY is one of such argument constraints.
94 * It means that the register type passed to this function must be
95 * PTR_TO_STACK and it will be used inside the function as
96 * 'pointer to map element key'
98 * For example the argument constraints for bpf_map_lookup_elem():
99 * .ret_type = RET_PTR_TO_MAP_VALUE_OR_NULL,
100 * .arg1_type = ARG_CONST_MAP_PTR,
101 * .arg2_type = ARG_PTR_TO_MAP_KEY,
103 * ret_type says that this function returns 'pointer to map elem value or null'
104 * function expects 1st argument to be a const pointer to 'struct bpf_map' and
105 * 2nd argument should be a pointer to stack, which will be used inside
106 * the helper function as a pointer to map element key.
108 * On the kernel side the helper function looks like:
109 * u64 bpf_map_lookup_elem(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5)
111 * struct bpf_map *map = (struct bpf_map *) (unsigned long) r1;
112 * void *key = (void *) (unsigned long) r2;
115 * here kernel can access 'key' and 'map' pointers safely, knowing that
116 * [key, key + map->key_size) bytes are valid and were initialized on
117 * the stack of eBPF program.
120 * Corresponding eBPF program may look like:
121 * BPF_MOV64_REG(BPF_REG_2, BPF_REG_10), // after this insn R2 type is FRAME_PTR
122 * BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, -4), // after this insn R2 type is PTR_TO_STACK
123 * BPF_LD_MAP_FD(BPF_REG_1, map_fd), // after this insn R1 type is CONST_PTR_TO_MAP
124 * BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0, BPF_FUNC_map_lookup_elem),
125 * here verifier looks at prototype of map_lookup_elem() and sees:
126 * .arg1_type == ARG_CONST_MAP_PTR and R1->type == CONST_PTR_TO_MAP, which is ok,
127 * Now verifier knows that this map has key of R1->map_ptr->key_size bytes
129 * Then .arg2_type == ARG_PTR_TO_MAP_KEY and R2->type == PTR_TO_STACK, ok so far,
130 * Now verifier checks that [R2, R2 + map's key_size) are within stack limits
131 * and were initialized prior to this call.
132 * If it's ok, then verifier allows this BPF_CALL insn and looks at
133 * .ret_type which is RET_PTR_TO_MAP_VALUE_OR_NULL, so it sets
134 * R0->type = PTR_TO_MAP_VALUE_OR_NULL which means bpf_map_lookup_elem() function
135 * returns either pointer to map value or NULL.
137 * When type PTR_TO_MAP_VALUE_OR_NULL passes through 'if (reg != 0) goto +off'
138 * insn, the register holding that pointer in the true branch changes state to
139 * PTR_TO_MAP_VALUE and the same register changes state to CONST_IMM in the false
140 * branch. See check_cond_jmp_op().
142 * After the call R0 is set to return type of the function and registers R1-R5
143 * are set to NOT_INIT to indicate that they are no longer readable.
145 * The following reference types represent a potential reference to a kernel
146 * resource which, after first being allocated, must be checked and freed by
148 * - PTR_TO_SOCKET_OR_NULL, PTR_TO_SOCKET
150 * When the verifier sees a helper call return a reference type, it allocates a
151 * pointer id for the reference and stores it in the current function state.
152 * Similar to the way that PTR_TO_MAP_VALUE_OR_NULL is converted into
153 * PTR_TO_MAP_VALUE, PTR_TO_SOCKET_OR_NULL becomes PTR_TO_SOCKET when the type
154 * passes through a NULL-check conditional. For the branch wherein the state is
155 * changed to CONST_IMM, the verifier releases the reference.
157 * For each helper function that allocates a reference, such as
158 * bpf_sk_lookup_tcp(), there is a corresponding release function, such as
159 * bpf_sk_release(). When a reference type passes into the release function,
160 * the verifier also releases the reference. If any unchecked or unreleased
161 * reference remains at the end of the program, the verifier rejects it.
164 /* verifier_state + insn_idx are pushed to stack when branch is encountered */
165 struct bpf_verifier_stack_elem
{
166 /* verifer state is 'st'
167 * before processing instruction 'insn_idx'
168 * and after processing instruction 'prev_insn_idx'
170 struct bpf_verifier_state st
;
173 struct bpf_verifier_stack_elem
*next
;
174 /* length of verifier log at the time this state was pushed on stack */
178 #define BPF_COMPLEXITY_LIMIT_JMP_SEQ 8192
179 #define BPF_COMPLEXITY_LIMIT_STATES 64
181 #define BPF_MAP_KEY_POISON (1ULL << 63)
182 #define BPF_MAP_KEY_SEEN (1ULL << 62)
184 #define BPF_MAP_PTR_UNPRIV 1UL
185 #define BPF_MAP_PTR_POISON ((void *)((0xeB9FUL << 1) + \
186 POISON_POINTER_DELTA))
187 #define BPF_MAP_PTR(X) ((struct bpf_map *)((X) & ~BPF_MAP_PTR_UNPRIV))
189 static bool bpf_map_ptr_poisoned(const struct bpf_insn_aux_data
*aux
)
191 return BPF_MAP_PTR(aux
->map_ptr_state
) == BPF_MAP_PTR_POISON
;
194 static bool bpf_map_ptr_unpriv(const struct bpf_insn_aux_data
*aux
)
196 return aux
->map_ptr_state
& BPF_MAP_PTR_UNPRIV
;
199 static void bpf_map_ptr_store(struct bpf_insn_aux_data
*aux
,
200 const struct bpf_map
*map
, bool unpriv
)
202 BUILD_BUG_ON((unsigned long)BPF_MAP_PTR_POISON
& BPF_MAP_PTR_UNPRIV
);
203 unpriv
|= bpf_map_ptr_unpriv(aux
);
204 aux
->map_ptr_state
= (unsigned long)map
|
205 (unpriv
? BPF_MAP_PTR_UNPRIV
: 0UL);
208 static bool bpf_map_key_poisoned(const struct bpf_insn_aux_data
*aux
)
210 return aux
->map_key_state
& BPF_MAP_KEY_POISON
;
213 static bool bpf_map_key_unseen(const struct bpf_insn_aux_data
*aux
)
215 return !(aux
->map_key_state
& BPF_MAP_KEY_SEEN
);
218 static u64
bpf_map_key_immediate(const struct bpf_insn_aux_data
*aux
)
220 return aux
->map_key_state
& ~(BPF_MAP_KEY_SEEN
| BPF_MAP_KEY_POISON
);
223 static void bpf_map_key_store(struct bpf_insn_aux_data
*aux
, u64 state
)
225 bool poisoned
= bpf_map_key_poisoned(aux
);
227 aux
->map_key_state
= state
| BPF_MAP_KEY_SEEN
|
228 (poisoned
? BPF_MAP_KEY_POISON
: 0ULL);
231 static bool bpf_pseudo_call(const struct bpf_insn
*insn
)
233 return insn
->code
== (BPF_JMP
| BPF_CALL
) &&
234 insn
->src_reg
== BPF_PSEUDO_CALL
;
237 static bool bpf_pseudo_kfunc_call(const struct bpf_insn
*insn
)
239 return insn
->code
== (BPF_JMP
| BPF_CALL
) &&
240 insn
->src_reg
== BPF_PSEUDO_KFUNC_CALL
;
243 static bool bpf_pseudo_func(const struct bpf_insn
*insn
)
245 return insn
->code
== (BPF_LD
| BPF_IMM
| BPF_DW
) &&
246 insn
->src_reg
== BPF_PSEUDO_FUNC
;
249 struct bpf_call_arg_meta
{
250 struct bpf_map
*map_ptr
;
267 struct btf
*btf_vmlinux
;
269 static DEFINE_MUTEX(bpf_verifier_lock
);
271 static const struct bpf_line_info
*
272 find_linfo(const struct bpf_verifier_env
*env
, u32 insn_off
)
274 const struct bpf_line_info
*linfo
;
275 const struct bpf_prog
*prog
;
279 nr_linfo
= prog
->aux
->nr_linfo
;
281 if (!nr_linfo
|| insn_off
>= prog
->len
)
284 linfo
= prog
->aux
->linfo
;
285 for (i
= 1; i
< nr_linfo
; i
++)
286 if (insn_off
< linfo
[i
].insn_off
)
289 return &linfo
[i
- 1];
292 void bpf_verifier_vlog(struct bpf_verifier_log
*log
, const char *fmt
,
297 n
= vscnprintf(log
->kbuf
, BPF_VERIFIER_TMP_LOG_SIZE
, fmt
, args
);
299 WARN_ONCE(n
>= BPF_VERIFIER_TMP_LOG_SIZE
- 1,
300 "verifier log line truncated - local buffer too short\n");
302 n
= min(log
->len_total
- log
->len_used
- 1, n
);
305 if (log
->level
== BPF_LOG_KERNEL
) {
306 pr_err("BPF:%s\n", log
->kbuf
);
309 if (!copy_to_user(log
->ubuf
+ log
->len_used
, log
->kbuf
, n
+ 1))
315 static void bpf_vlog_reset(struct bpf_verifier_log
*log
, u32 new_pos
)
319 if (!bpf_verifier_log_needed(log
))
322 log
->len_used
= new_pos
;
323 if (put_user(zero
, log
->ubuf
+ new_pos
))
327 /* log_level controls verbosity level of eBPF verifier.
328 * bpf_verifier_log_write() is used to dump the verification trace to the log,
329 * so the user can figure out what's wrong with the program
331 __printf(2, 3) void bpf_verifier_log_write(struct bpf_verifier_env
*env
,
332 const char *fmt
, ...)
336 if (!bpf_verifier_log_needed(&env
->log
))
340 bpf_verifier_vlog(&env
->log
, fmt
, args
);
343 EXPORT_SYMBOL_GPL(bpf_verifier_log_write
);
345 __printf(2, 3) static void verbose(void *private_data
, const char *fmt
, ...)
347 struct bpf_verifier_env
*env
= private_data
;
350 if (!bpf_verifier_log_needed(&env
->log
))
354 bpf_verifier_vlog(&env
->log
, fmt
, args
);
358 __printf(2, 3) void bpf_log(struct bpf_verifier_log
*log
,
359 const char *fmt
, ...)
363 if (!bpf_verifier_log_needed(log
))
367 bpf_verifier_vlog(log
, fmt
, args
);
371 static const char *ltrim(const char *s
)
379 __printf(3, 4) static void verbose_linfo(struct bpf_verifier_env
*env
,
381 const char *prefix_fmt
, ...)
383 const struct bpf_line_info
*linfo
;
385 if (!bpf_verifier_log_needed(&env
->log
))
388 linfo
= find_linfo(env
, insn_off
);
389 if (!linfo
|| linfo
== env
->prev_linfo
)
395 va_start(args
, prefix_fmt
);
396 bpf_verifier_vlog(&env
->log
, prefix_fmt
, args
);
401 ltrim(btf_name_by_offset(env
->prog
->aux
->btf
,
404 env
->prev_linfo
= linfo
;
407 static void verbose_invalid_scalar(struct bpf_verifier_env
*env
,
408 struct bpf_reg_state
*reg
,
409 struct tnum
*range
, const char *ctx
,
410 const char *reg_name
)
414 verbose(env
, "At %s the register %s ", ctx
, reg_name
);
415 if (!tnum_is_unknown(reg
->var_off
)) {
416 tnum_strn(tn_buf
, sizeof(tn_buf
), reg
->var_off
);
417 verbose(env
, "has value %s", tn_buf
);
419 verbose(env
, "has unknown scalar value");
421 tnum_strn(tn_buf
, sizeof(tn_buf
), *range
);
422 verbose(env
, " should have been in %s\n", tn_buf
);
425 static bool type_is_pkt_pointer(enum bpf_reg_type type
)
427 return type
== PTR_TO_PACKET
||
428 type
== PTR_TO_PACKET_META
;
431 static bool type_is_sk_pointer(enum bpf_reg_type type
)
433 return type
== PTR_TO_SOCKET
||
434 type
== PTR_TO_SOCK_COMMON
||
435 type
== PTR_TO_TCP_SOCK
||
436 type
== PTR_TO_XDP_SOCK
;
439 static bool reg_type_not_null(enum bpf_reg_type type
)
441 return type
== PTR_TO_SOCKET
||
442 type
== PTR_TO_TCP_SOCK
||
443 type
== PTR_TO_MAP_VALUE
||
444 type
== PTR_TO_MAP_KEY
||
445 type
== PTR_TO_SOCK_COMMON
;
448 static bool reg_type_may_be_null(enum bpf_reg_type type
)
450 return type
== PTR_TO_MAP_VALUE_OR_NULL
||
451 type
== PTR_TO_SOCKET_OR_NULL
||
452 type
== PTR_TO_SOCK_COMMON_OR_NULL
||
453 type
== PTR_TO_TCP_SOCK_OR_NULL
||
454 type
== PTR_TO_BTF_ID_OR_NULL
||
455 type
== PTR_TO_MEM_OR_NULL
||
456 type
== PTR_TO_RDONLY_BUF_OR_NULL
||
457 type
== PTR_TO_RDWR_BUF_OR_NULL
;
460 static bool reg_may_point_to_spin_lock(const struct bpf_reg_state
*reg
)
462 return reg
->type
== PTR_TO_MAP_VALUE
&&
463 map_value_has_spin_lock(reg
->map_ptr
);
466 static bool reg_type_may_be_refcounted_or_null(enum bpf_reg_type type
)
468 return type
== PTR_TO_SOCKET
||
469 type
== PTR_TO_SOCKET_OR_NULL
||
470 type
== PTR_TO_TCP_SOCK
||
471 type
== PTR_TO_TCP_SOCK_OR_NULL
||
472 type
== PTR_TO_MEM
||
473 type
== PTR_TO_MEM_OR_NULL
;
476 static bool arg_type_may_be_refcounted(enum bpf_arg_type type
)
478 return type
== ARG_PTR_TO_SOCK_COMMON
;
481 static bool arg_type_may_be_null(enum bpf_arg_type type
)
483 return type
== ARG_PTR_TO_MAP_VALUE_OR_NULL
||
484 type
== ARG_PTR_TO_MEM_OR_NULL
||
485 type
== ARG_PTR_TO_CTX_OR_NULL
||
486 type
== ARG_PTR_TO_SOCKET_OR_NULL
||
487 type
== ARG_PTR_TO_ALLOC_MEM_OR_NULL
||
488 type
== ARG_PTR_TO_STACK_OR_NULL
;
491 /* Determine whether the function releases some resources allocated by another
492 * function call. The first reference type argument will be assumed to be
493 * released by release_reference().
495 static bool is_release_function(enum bpf_func_id func_id
)
497 return func_id
== BPF_FUNC_sk_release
||
498 func_id
== BPF_FUNC_ringbuf_submit
||
499 func_id
== BPF_FUNC_ringbuf_discard
;
502 static bool may_be_acquire_function(enum bpf_func_id func_id
)
504 return func_id
== BPF_FUNC_sk_lookup_tcp
||
505 func_id
== BPF_FUNC_sk_lookup_udp
||
506 func_id
== BPF_FUNC_skc_lookup_tcp
||
507 func_id
== BPF_FUNC_map_lookup_elem
||
508 func_id
== BPF_FUNC_ringbuf_reserve
;
511 static bool is_acquire_function(enum bpf_func_id func_id
,
512 const struct bpf_map
*map
)
514 enum bpf_map_type map_type
= map
? map
->map_type
: BPF_MAP_TYPE_UNSPEC
;
516 if (func_id
== BPF_FUNC_sk_lookup_tcp
||
517 func_id
== BPF_FUNC_sk_lookup_udp
||
518 func_id
== BPF_FUNC_skc_lookup_tcp
||
519 func_id
== BPF_FUNC_ringbuf_reserve
)
522 if (func_id
== BPF_FUNC_map_lookup_elem
&&
523 (map_type
== BPF_MAP_TYPE_SOCKMAP
||
524 map_type
== BPF_MAP_TYPE_SOCKHASH
))
530 static bool is_ptr_cast_function(enum bpf_func_id func_id
)
532 return func_id
== BPF_FUNC_tcp_sock
||
533 func_id
== BPF_FUNC_sk_fullsock
||
534 func_id
== BPF_FUNC_skc_to_tcp_sock
||
535 func_id
== BPF_FUNC_skc_to_tcp6_sock
||
536 func_id
== BPF_FUNC_skc_to_udp6_sock
||
537 func_id
== BPF_FUNC_skc_to_tcp_timewait_sock
||
538 func_id
== BPF_FUNC_skc_to_tcp_request_sock
;
541 static bool is_cmpxchg_insn(const struct bpf_insn
*insn
)
543 return BPF_CLASS(insn
->code
) == BPF_STX
&&
544 BPF_MODE(insn
->code
) == BPF_ATOMIC
&&
545 insn
->imm
== BPF_CMPXCHG
;
548 /* string representation of 'enum bpf_reg_type' */
549 static const char * const reg_type_str
[] = {
551 [SCALAR_VALUE
] = "inv",
552 [PTR_TO_CTX
] = "ctx",
553 [CONST_PTR_TO_MAP
] = "map_ptr",
554 [PTR_TO_MAP_VALUE
] = "map_value",
555 [PTR_TO_MAP_VALUE_OR_NULL
] = "map_value_or_null",
556 [PTR_TO_STACK
] = "fp",
557 [PTR_TO_PACKET
] = "pkt",
558 [PTR_TO_PACKET_META
] = "pkt_meta",
559 [PTR_TO_PACKET_END
] = "pkt_end",
560 [PTR_TO_FLOW_KEYS
] = "flow_keys",
561 [PTR_TO_SOCKET
] = "sock",
562 [PTR_TO_SOCKET_OR_NULL
] = "sock_or_null",
563 [PTR_TO_SOCK_COMMON
] = "sock_common",
564 [PTR_TO_SOCK_COMMON_OR_NULL
] = "sock_common_or_null",
565 [PTR_TO_TCP_SOCK
] = "tcp_sock",
566 [PTR_TO_TCP_SOCK_OR_NULL
] = "tcp_sock_or_null",
567 [PTR_TO_TP_BUFFER
] = "tp_buffer",
568 [PTR_TO_XDP_SOCK
] = "xdp_sock",
569 [PTR_TO_BTF_ID
] = "ptr_",
570 [PTR_TO_BTF_ID_OR_NULL
] = "ptr_or_null_",
571 [PTR_TO_PERCPU_BTF_ID
] = "percpu_ptr_",
572 [PTR_TO_MEM
] = "mem",
573 [PTR_TO_MEM_OR_NULL
] = "mem_or_null",
574 [PTR_TO_RDONLY_BUF
] = "rdonly_buf",
575 [PTR_TO_RDONLY_BUF_OR_NULL
] = "rdonly_buf_or_null",
576 [PTR_TO_RDWR_BUF
] = "rdwr_buf",
577 [PTR_TO_RDWR_BUF_OR_NULL
] = "rdwr_buf_or_null",
578 [PTR_TO_FUNC
] = "func",
579 [PTR_TO_MAP_KEY
] = "map_key",
582 static char slot_type_char
[] = {
583 [STACK_INVALID
] = '?',
589 static void print_liveness(struct bpf_verifier_env
*env
,
590 enum bpf_reg_liveness live
)
592 if (live
& (REG_LIVE_READ
| REG_LIVE_WRITTEN
| REG_LIVE_DONE
))
594 if (live
& REG_LIVE_READ
)
596 if (live
& REG_LIVE_WRITTEN
)
598 if (live
& REG_LIVE_DONE
)
602 static struct bpf_func_state
*func(struct bpf_verifier_env
*env
,
603 const struct bpf_reg_state
*reg
)
605 struct bpf_verifier_state
*cur
= env
->cur_state
;
607 return cur
->frame
[reg
->frameno
];
610 static const char *kernel_type_name(const struct btf
* btf
, u32 id
)
612 return btf_name_by_offset(btf
, btf_type_by_id(btf
, id
)->name_off
);
615 static void print_verifier_state(struct bpf_verifier_env
*env
,
616 const struct bpf_func_state
*state
)
618 const struct bpf_reg_state
*reg
;
623 verbose(env
, " frame%d:", state
->frameno
);
624 for (i
= 0; i
< MAX_BPF_REG
; i
++) {
625 reg
= &state
->regs
[i
];
629 verbose(env
, " R%d", i
);
630 print_liveness(env
, reg
->live
);
631 verbose(env
, "=%s", reg_type_str
[t
]);
632 if (t
== SCALAR_VALUE
&& reg
->precise
)
634 if ((t
== SCALAR_VALUE
|| t
== PTR_TO_STACK
) &&
635 tnum_is_const(reg
->var_off
)) {
636 /* reg->off should be 0 for SCALAR_VALUE */
637 verbose(env
, "%lld", reg
->var_off
.value
+ reg
->off
);
639 if (t
== PTR_TO_BTF_ID
||
640 t
== PTR_TO_BTF_ID_OR_NULL
||
641 t
== PTR_TO_PERCPU_BTF_ID
)
642 verbose(env
, "%s", kernel_type_name(reg
->btf
, reg
->btf_id
));
643 verbose(env
, "(id=%d", reg
->id
);
644 if (reg_type_may_be_refcounted_or_null(t
))
645 verbose(env
, ",ref_obj_id=%d", reg
->ref_obj_id
);
646 if (t
!= SCALAR_VALUE
)
647 verbose(env
, ",off=%d", reg
->off
);
648 if (type_is_pkt_pointer(t
))
649 verbose(env
, ",r=%d", reg
->range
);
650 else if (t
== CONST_PTR_TO_MAP
||
651 t
== PTR_TO_MAP_KEY
||
652 t
== PTR_TO_MAP_VALUE
||
653 t
== PTR_TO_MAP_VALUE_OR_NULL
)
654 verbose(env
, ",ks=%d,vs=%d",
655 reg
->map_ptr
->key_size
,
656 reg
->map_ptr
->value_size
);
657 if (tnum_is_const(reg
->var_off
)) {
658 /* Typically an immediate SCALAR_VALUE, but
659 * could be a pointer whose offset is too big
662 verbose(env
, ",imm=%llx", reg
->var_off
.value
);
664 if (reg
->smin_value
!= reg
->umin_value
&&
665 reg
->smin_value
!= S64_MIN
)
666 verbose(env
, ",smin_value=%lld",
667 (long long)reg
->smin_value
);
668 if (reg
->smax_value
!= reg
->umax_value
&&
669 reg
->smax_value
!= S64_MAX
)
670 verbose(env
, ",smax_value=%lld",
671 (long long)reg
->smax_value
);
672 if (reg
->umin_value
!= 0)
673 verbose(env
, ",umin_value=%llu",
674 (unsigned long long)reg
->umin_value
);
675 if (reg
->umax_value
!= U64_MAX
)
676 verbose(env
, ",umax_value=%llu",
677 (unsigned long long)reg
->umax_value
);
678 if (!tnum_is_unknown(reg
->var_off
)) {
681 tnum_strn(tn_buf
, sizeof(tn_buf
), reg
->var_off
);
682 verbose(env
, ",var_off=%s", tn_buf
);
684 if (reg
->s32_min_value
!= reg
->smin_value
&&
685 reg
->s32_min_value
!= S32_MIN
)
686 verbose(env
, ",s32_min_value=%d",
687 (int)(reg
->s32_min_value
));
688 if (reg
->s32_max_value
!= reg
->smax_value
&&
689 reg
->s32_max_value
!= S32_MAX
)
690 verbose(env
, ",s32_max_value=%d",
691 (int)(reg
->s32_max_value
));
692 if (reg
->u32_min_value
!= reg
->umin_value
&&
693 reg
->u32_min_value
!= U32_MIN
)
694 verbose(env
, ",u32_min_value=%d",
695 (int)(reg
->u32_min_value
));
696 if (reg
->u32_max_value
!= reg
->umax_value
&&
697 reg
->u32_max_value
!= U32_MAX
)
698 verbose(env
, ",u32_max_value=%d",
699 (int)(reg
->u32_max_value
));
704 for (i
= 0; i
< state
->allocated_stack
/ BPF_REG_SIZE
; i
++) {
705 char types_buf
[BPF_REG_SIZE
+ 1];
709 for (j
= 0; j
< BPF_REG_SIZE
; j
++) {
710 if (state
->stack
[i
].slot_type
[j
] != STACK_INVALID
)
712 types_buf
[j
] = slot_type_char
[
713 state
->stack
[i
].slot_type
[j
]];
715 types_buf
[BPF_REG_SIZE
] = 0;
718 verbose(env
, " fp%d", (-i
- 1) * BPF_REG_SIZE
);
719 print_liveness(env
, state
->stack
[i
].spilled_ptr
.live
);
720 if (state
->stack
[i
].slot_type
[0] == STACK_SPILL
) {
721 reg
= &state
->stack
[i
].spilled_ptr
;
723 verbose(env
, "=%s", reg_type_str
[t
]);
724 if (t
== SCALAR_VALUE
&& reg
->precise
)
726 if (t
== SCALAR_VALUE
&& tnum_is_const(reg
->var_off
))
727 verbose(env
, "%lld", reg
->var_off
.value
+ reg
->off
);
729 verbose(env
, "=%s", types_buf
);
732 if (state
->acquired_refs
&& state
->refs
[0].id
) {
733 verbose(env
, " refs=%d", state
->refs
[0].id
);
734 for (i
= 1; i
< state
->acquired_refs
; i
++)
735 if (state
->refs
[i
].id
)
736 verbose(env
, ",%d", state
->refs
[i
].id
);
738 if (state
->in_callback_fn
)
740 if (state
->in_async_callback_fn
)
741 verbose(env
, " async_cb");
745 /* copy array src of length n * size bytes to dst. dst is reallocated if it's too
746 * small to hold src. This is different from krealloc since we don't want to preserve
747 * the contents of dst.
749 * Leaves dst untouched if src is NULL or length is zero. Returns NULL if memory could
752 static void *copy_array(void *dst
, const void *src
, size_t n
, size_t size
, gfp_t flags
)
756 if (ZERO_OR_NULL_PTR(src
))
759 if (unlikely(check_mul_overflow(n
, size
, &bytes
)))
762 if (ksize(dst
) < bytes
) {
764 dst
= kmalloc_track_caller(bytes
, flags
);
769 memcpy(dst
, src
, bytes
);
771 return dst
? dst
: ZERO_SIZE_PTR
;
774 /* resize an array from old_n items to new_n items. the array is reallocated if it's too
775 * small to hold new_n items. new items are zeroed out if the array grows.
777 * Contrary to krealloc_array, does not free arr if new_n is zero.
779 static void *realloc_array(void *arr
, size_t old_n
, size_t new_n
, size_t size
)
781 if (!new_n
|| old_n
== new_n
)
784 arr
= krealloc_array(arr
, new_n
, size
, GFP_KERNEL
);
789 memset(arr
+ old_n
* size
, 0, (new_n
- old_n
) * size
);
792 return arr
? arr
: ZERO_SIZE_PTR
;
795 static int copy_reference_state(struct bpf_func_state
*dst
, const struct bpf_func_state
*src
)
797 dst
->refs
= copy_array(dst
->refs
, src
->refs
, src
->acquired_refs
,
798 sizeof(struct bpf_reference_state
), GFP_KERNEL
);
802 dst
->acquired_refs
= src
->acquired_refs
;
806 static int copy_stack_state(struct bpf_func_state
*dst
, const struct bpf_func_state
*src
)
808 size_t n
= src
->allocated_stack
/ BPF_REG_SIZE
;
810 dst
->stack
= copy_array(dst
->stack
, src
->stack
, n
, sizeof(struct bpf_stack_state
),
815 dst
->allocated_stack
= src
->allocated_stack
;
819 static int resize_reference_state(struct bpf_func_state
*state
, size_t n
)
821 state
->refs
= realloc_array(state
->refs
, state
->acquired_refs
, n
,
822 sizeof(struct bpf_reference_state
));
826 state
->acquired_refs
= n
;
830 static int grow_stack_state(struct bpf_func_state
*state
, int size
)
832 size_t old_n
= state
->allocated_stack
/ BPF_REG_SIZE
, n
= size
/ BPF_REG_SIZE
;
837 state
->stack
= realloc_array(state
->stack
, old_n
, n
, sizeof(struct bpf_stack_state
));
841 state
->allocated_stack
= size
;
845 /* Acquire a pointer id from the env and update the state->refs to include
846 * this new pointer reference.
847 * On success, returns a valid pointer id to associate with the register
848 * On failure, returns a negative errno.
850 static int acquire_reference_state(struct bpf_verifier_env
*env
, int insn_idx
)
852 struct bpf_func_state
*state
= cur_func(env
);
853 int new_ofs
= state
->acquired_refs
;
856 err
= resize_reference_state(state
, state
->acquired_refs
+ 1);
860 state
->refs
[new_ofs
].id
= id
;
861 state
->refs
[new_ofs
].insn_idx
= insn_idx
;
866 /* release function corresponding to acquire_reference_state(). Idempotent. */
867 static int release_reference_state(struct bpf_func_state
*state
, int ptr_id
)
871 last_idx
= state
->acquired_refs
- 1;
872 for (i
= 0; i
< state
->acquired_refs
; i
++) {
873 if (state
->refs
[i
].id
== ptr_id
) {
874 if (last_idx
&& i
!= last_idx
)
875 memcpy(&state
->refs
[i
], &state
->refs
[last_idx
],
876 sizeof(*state
->refs
));
877 memset(&state
->refs
[last_idx
], 0, sizeof(*state
->refs
));
878 state
->acquired_refs
--;
885 static void free_func_state(struct bpf_func_state
*state
)
894 static void clear_jmp_history(struct bpf_verifier_state
*state
)
896 kfree(state
->jmp_history
);
897 state
->jmp_history
= NULL
;
898 state
->jmp_history_cnt
= 0;
901 static void free_verifier_state(struct bpf_verifier_state
*state
,
906 for (i
= 0; i
<= state
->curframe
; i
++) {
907 free_func_state(state
->frame
[i
]);
908 state
->frame
[i
] = NULL
;
910 clear_jmp_history(state
);
915 /* copy verifier state from src to dst growing dst stack space
916 * when necessary to accommodate larger src stack
918 static int copy_func_state(struct bpf_func_state
*dst
,
919 const struct bpf_func_state
*src
)
923 memcpy(dst
, src
, offsetof(struct bpf_func_state
, acquired_refs
));
924 err
= copy_reference_state(dst
, src
);
927 return copy_stack_state(dst
, src
);
930 static int copy_verifier_state(struct bpf_verifier_state
*dst_state
,
931 const struct bpf_verifier_state
*src
)
933 struct bpf_func_state
*dst
;
936 dst_state
->jmp_history
= copy_array(dst_state
->jmp_history
, src
->jmp_history
,
937 src
->jmp_history_cnt
, sizeof(struct bpf_idx_pair
),
939 if (!dst_state
->jmp_history
)
941 dst_state
->jmp_history_cnt
= src
->jmp_history_cnt
;
943 /* if dst has more stack frames then src frame, free them */
944 for (i
= src
->curframe
+ 1; i
<= dst_state
->curframe
; i
++) {
945 free_func_state(dst_state
->frame
[i
]);
946 dst_state
->frame
[i
] = NULL
;
948 dst_state
->speculative
= src
->speculative
;
949 dst_state
->curframe
= src
->curframe
;
950 dst_state
->active_spin_lock
= src
->active_spin_lock
;
951 dst_state
->branches
= src
->branches
;
952 dst_state
->parent
= src
->parent
;
953 dst_state
->first_insn_idx
= src
->first_insn_idx
;
954 dst_state
->last_insn_idx
= src
->last_insn_idx
;
955 for (i
= 0; i
<= src
->curframe
; i
++) {
956 dst
= dst_state
->frame
[i
];
958 dst
= kzalloc(sizeof(*dst
), GFP_KERNEL
);
961 dst_state
->frame
[i
] = dst
;
963 err
= copy_func_state(dst
, src
->frame
[i
]);
970 static void update_branch_counts(struct bpf_verifier_env
*env
, struct bpf_verifier_state
*st
)
973 u32 br
= --st
->branches
;
975 /* WARN_ON(br > 1) technically makes sense here,
976 * but see comment in push_stack(), hence:
978 WARN_ONCE((int)br
< 0,
979 "BUG update_branch_counts:branches_to_explore=%d\n",
987 static int pop_stack(struct bpf_verifier_env
*env
, int *prev_insn_idx
,
988 int *insn_idx
, bool pop_log
)
990 struct bpf_verifier_state
*cur
= env
->cur_state
;
991 struct bpf_verifier_stack_elem
*elem
, *head
= env
->head
;
994 if (env
->head
== NULL
)
998 err
= copy_verifier_state(cur
, &head
->st
);
1003 bpf_vlog_reset(&env
->log
, head
->log_pos
);
1005 *insn_idx
= head
->insn_idx
;
1007 *prev_insn_idx
= head
->prev_insn_idx
;
1009 free_verifier_state(&head
->st
, false);
1016 static struct bpf_verifier_state
*push_stack(struct bpf_verifier_env
*env
,
1017 int insn_idx
, int prev_insn_idx
,
1020 struct bpf_verifier_state
*cur
= env
->cur_state
;
1021 struct bpf_verifier_stack_elem
*elem
;
1024 elem
= kzalloc(sizeof(struct bpf_verifier_stack_elem
), GFP_KERNEL
);
1028 elem
->insn_idx
= insn_idx
;
1029 elem
->prev_insn_idx
= prev_insn_idx
;
1030 elem
->next
= env
->head
;
1031 elem
->log_pos
= env
->log
.len_used
;
1034 err
= copy_verifier_state(&elem
->st
, cur
);
1037 elem
->st
.speculative
|= speculative
;
1038 if (env
->stack_size
> BPF_COMPLEXITY_LIMIT_JMP_SEQ
) {
1039 verbose(env
, "The sequence of %d jumps is too complex.\n",
1043 if (elem
->st
.parent
) {
1044 ++elem
->st
.parent
->branches
;
1045 /* WARN_ON(branches > 2) technically makes sense here,
1047 * 1. speculative states will bump 'branches' for non-branch
1049 * 2. is_state_visited() heuristics may decide not to create
1050 * a new state for a sequence of branches and all such current
1051 * and cloned states will be pointing to a single parent state
1052 * which might have large 'branches' count.
1057 free_verifier_state(env
->cur_state
, true);
1058 env
->cur_state
= NULL
;
1059 /* pop all elements and return */
1060 while (!pop_stack(env
, NULL
, NULL
, false));
1064 #define CALLER_SAVED_REGS 6
1065 static const int caller_saved
[CALLER_SAVED_REGS
] = {
1066 BPF_REG_0
, BPF_REG_1
, BPF_REG_2
, BPF_REG_3
, BPF_REG_4
, BPF_REG_5
1069 static void __mark_reg_not_init(const struct bpf_verifier_env
*env
,
1070 struct bpf_reg_state
*reg
);
1072 /* This helper doesn't clear reg->id */
1073 static void ___mark_reg_known(struct bpf_reg_state
*reg
, u64 imm
)
1075 reg
->var_off
= tnum_const(imm
);
1076 reg
->smin_value
= (s64
)imm
;
1077 reg
->smax_value
= (s64
)imm
;
1078 reg
->umin_value
= imm
;
1079 reg
->umax_value
= imm
;
1081 reg
->s32_min_value
= (s32
)imm
;
1082 reg
->s32_max_value
= (s32
)imm
;
1083 reg
->u32_min_value
= (u32
)imm
;
1084 reg
->u32_max_value
= (u32
)imm
;
1087 /* Mark the unknown part of a register (variable offset or scalar value) as
1088 * known to have the value @imm.
1090 static void __mark_reg_known(struct bpf_reg_state
*reg
, u64 imm
)
1092 /* Clear id, off, and union(map_ptr, range) */
1093 memset(((u8
*)reg
) + sizeof(reg
->type
), 0,
1094 offsetof(struct bpf_reg_state
, var_off
) - sizeof(reg
->type
));
1095 ___mark_reg_known(reg
, imm
);
1098 static void __mark_reg32_known(struct bpf_reg_state
*reg
, u64 imm
)
1100 reg
->var_off
= tnum_const_subreg(reg
->var_off
, imm
);
1101 reg
->s32_min_value
= (s32
)imm
;
1102 reg
->s32_max_value
= (s32
)imm
;
1103 reg
->u32_min_value
= (u32
)imm
;
1104 reg
->u32_max_value
= (u32
)imm
;
1107 /* Mark the 'variable offset' part of a register as zero. This should be
1108 * used only on registers holding a pointer type.
1110 static void __mark_reg_known_zero(struct bpf_reg_state
*reg
)
1112 __mark_reg_known(reg
, 0);
1115 static void __mark_reg_const_zero(struct bpf_reg_state
*reg
)
1117 __mark_reg_known(reg
, 0);
1118 reg
->type
= SCALAR_VALUE
;
1121 static void mark_reg_known_zero(struct bpf_verifier_env
*env
,
1122 struct bpf_reg_state
*regs
, u32 regno
)
1124 if (WARN_ON(regno
>= MAX_BPF_REG
)) {
1125 verbose(env
, "mark_reg_known_zero(regs, %u)\n", regno
);
1126 /* Something bad happened, let's kill all regs */
1127 for (regno
= 0; regno
< MAX_BPF_REG
; regno
++)
1128 __mark_reg_not_init(env
, regs
+ regno
);
1131 __mark_reg_known_zero(regs
+ regno
);
1134 static void mark_ptr_not_null_reg(struct bpf_reg_state
*reg
)
1136 switch (reg
->type
) {
1137 case PTR_TO_MAP_VALUE_OR_NULL
: {
1138 const struct bpf_map
*map
= reg
->map_ptr
;
1140 if (map
->inner_map_meta
) {
1141 reg
->type
= CONST_PTR_TO_MAP
;
1142 reg
->map_ptr
= map
->inner_map_meta
;
1143 /* transfer reg's id which is unique for every map_lookup_elem
1144 * as UID of the inner map.
1146 reg
->map_uid
= reg
->id
;
1147 } else if (map
->map_type
== BPF_MAP_TYPE_XSKMAP
) {
1148 reg
->type
= PTR_TO_XDP_SOCK
;
1149 } else if (map
->map_type
== BPF_MAP_TYPE_SOCKMAP
||
1150 map
->map_type
== BPF_MAP_TYPE_SOCKHASH
) {
1151 reg
->type
= PTR_TO_SOCKET
;
1153 reg
->type
= PTR_TO_MAP_VALUE
;
1157 case PTR_TO_SOCKET_OR_NULL
:
1158 reg
->type
= PTR_TO_SOCKET
;
1160 case PTR_TO_SOCK_COMMON_OR_NULL
:
1161 reg
->type
= PTR_TO_SOCK_COMMON
;
1163 case PTR_TO_TCP_SOCK_OR_NULL
:
1164 reg
->type
= PTR_TO_TCP_SOCK
;
1166 case PTR_TO_BTF_ID_OR_NULL
:
1167 reg
->type
= PTR_TO_BTF_ID
;
1169 case PTR_TO_MEM_OR_NULL
:
1170 reg
->type
= PTR_TO_MEM
;
1172 case PTR_TO_RDONLY_BUF_OR_NULL
:
1173 reg
->type
= PTR_TO_RDONLY_BUF
;
1175 case PTR_TO_RDWR_BUF_OR_NULL
:
1176 reg
->type
= PTR_TO_RDWR_BUF
;
1179 WARN_ONCE(1, "unknown nullable register type");
1183 static bool reg_is_pkt_pointer(const struct bpf_reg_state
*reg
)
1185 return type_is_pkt_pointer(reg
->type
);
1188 static bool reg_is_pkt_pointer_any(const struct bpf_reg_state
*reg
)
1190 return reg_is_pkt_pointer(reg
) ||
1191 reg
->type
== PTR_TO_PACKET_END
;
1194 /* Unmodified PTR_TO_PACKET[_META,_END] register from ctx access. */
1195 static bool reg_is_init_pkt_pointer(const struct bpf_reg_state
*reg
,
1196 enum bpf_reg_type which
)
1198 /* The register can already have a range from prior markings.
1199 * This is fine as long as it hasn't been advanced from its
1202 return reg
->type
== which
&&
1205 tnum_equals_const(reg
->var_off
, 0);
1208 /* Reset the min/max bounds of a register */
1209 static void __mark_reg_unbounded(struct bpf_reg_state
*reg
)
1211 reg
->smin_value
= S64_MIN
;
1212 reg
->smax_value
= S64_MAX
;
1213 reg
->umin_value
= 0;
1214 reg
->umax_value
= U64_MAX
;
1216 reg
->s32_min_value
= S32_MIN
;
1217 reg
->s32_max_value
= S32_MAX
;
1218 reg
->u32_min_value
= 0;
1219 reg
->u32_max_value
= U32_MAX
;
1222 static void __mark_reg64_unbounded(struct bpf_reg_state
*reg
)
1224 reg
->smin_value
= S64_MIN
;
1225 reg
->smax_value
= S64_MAX
;
1226 reg
->umin_value
= 0;
1227 reg
->umax_value
= U64_MAX
;
1230 static void __mark_reg32_unbounded(struct bpf_reg_state
*reg
)
1232 reg
->s32_min_value
= S32_MIN
;
1233 reg
->s32_max_value
= S32_MAX
;
1234 reg
->u32_min_value
= 0;
1235 reg
->u32_max_value
= U32_MAX
;
1238 static void __update_reg32_bounds(struct bpf_reg_state
*reg
)
1240 struct tnum var32_off
= tnum_subreg(reg
->var_off
);
1242 /* min signed is max(sign bit) | min(other bits) */
1243 reg
->s32_min_value
= max_t(s32
, reg
->s32_min_value
,
1244 var32_off
.value
| (var32_off
.mask
& S32_MIN
));
1245 /* max signed is min(sign bit) | max(other bits) */
1246 reg
->s32_max_value
= min_t(s32
, reg
->s32_max_value
,
1247 var32_off
.value
| (var32_off
.mask
& S32_MAX
));
1248 reg
->u32_min_value
= max_t(u32
, reg
->u32_min_value
, (u32
)var32_off
.value
);
1249 reg
->u32_max_value
= min(reg
->u32_max_value
,
1250 (u32
)(var32_off
.value
| var32_off
.mask
));
1253 static void __update_reg64_bounds(struct bpf_reg_state
*reg
)
1255 /* min signed is max(sign bit) | min(other bits) */
1256 reg
->smin_value
= max_t(s64
, reg
->smin_value
,
1257 reg
->var_off
.value
| (reg
->var_off
.mask
& S64_MIN
));
1258 /* max signed is min(sign bit) | max(other bits) */
1259 reg
->smax_value
= min_t(s64
, reg
->smax_value
,
1260 reg
->var_off
.value
| (reg
->var_off
.mask
& S64_MAX
));
1261 reg
->umin_value
= max(reg
->umin_value
, reg
->var_off
.value
);
1262 reg
->umax_value
= min(reg
->umax_value
,
1263 reg
->var_off
.value
| reg
->var_off
.mask
);
1266 static void __update_reg_bounds(struct bpf_reg_state
*reg
)
1268 __update_reg32_bounds(reg
);
1269 __update_reg64_bounds(reg
);
1272 /* Uses signed min/max values to inform unsigned, and vice-versa */
1273 static void __reg32_deduce_bounds(struct bpf_reg_state
*reg
)
1275 /* Learn sign from signed bounds.
1276 * If we cannot cross the sign boundary, then signed and unsigned bounds
1277 * are the same, so combine. This works even in the negative case, e.g.
1278 * -3 s<= x s<= -1 implies 0xf...fd u<= x u<= 0xf...ff.
1280 if (reg
->s32_min_value
>= 0 || reg
->s32_max_value
< 0) {
1281 reg
->s32_min_value
= reg
->u32_min_value
=
1282 max_t(u32
, reg
->s32_min_value
, reg
->u32_min_value
);
1283 reg
->s32_max_value
= reg
->u32_max_value
=
1284 min_t(u32
, reg
->s32_max_value
, reg
->u32_max_value
);
1287 /* Learn sign from unsigned bounds. Signed bounds cross the sign
1288 * boundary, so we must be careful.
1290 if ((s32
)reg
->u32_max_value
>= 0) {
1291 /* Positive. We can't learn anything from the smin, but smax
1292 * is positive, hence safe.
1294 reg
->s32_min_value
= reg
->u32_min_value
;
1295 reg
->s32_max_value
= reg
->u32_max_value
=
1296 min_t(u32
, reg
->s32_max_value
, reg
->u32_max_value
);
1297 } else if ((s32
)reg
->u32_min_value
< 0) {
1298 /* Negative. We can't learn anything from the smax, but smin
1299 * is negative, hence safe.
1301 reg
->s32_min_value
= reg
->u32_min_value
=
1302 max_t(u32
, reg
->s32_min_value
, reg
->u32_min_value
);
1303 reg
->s32_max_value
= reg
->u32_max_value
;
1307 static void __reg64_deduce_bounds(struct bpf_reg_state
*reg
)
1309 /* Learn sign from signed bounds.
1310 * If we cannot cross the sign boundary, then signed and unsigned bounds
1311 * are the same, so combine. This works even in the negative case, e.g.
1312 * -3 s<= x s<= -1 implies 0xf...fd u<= x u<= 0xf...ff.
1314 if (reg
->smin_value
>= 0 || reg
->smax_value
< 0) {
1315 reg
->smin_value
= reg
->umin_value
= max_t(u64
, reg
->smin_value
,
1317 reg
->smax_value
= reg
->umax_value
= min_t(u64
, reg
->smax_value
,
1321 /* Learn sign from unsigned bounds. Signed bounds cross the sign
1322 * boundary, so we must be careful.
1324 if ((s64
)reg
->umax_value
>= 0) {
1325 /* Positive. We can't learn anything from the smin, but smax
1326 * is positive, hence safe.
1328 reg
->smin_value
= reg
->umin_value
;
1329 reg
->smax_value
= reg
->umax_value
= min_t(u64
, reg
->smax_value
,
1331 } else if ((s64
)reg
->umin_value
< 0) {
1332 /* Negative. We can't learn anything from the smax, but smin
1333 * is negative, hence safe.
1335 reg
->smin_value
= reg
->umin_value
= max_t(u64
, reg
->smin_value
,
1337 reg
->smax_value
= reg
->umax_value
;
1341 static void __reg_deduce_bounds(struct bpf_reg_state
*reg
)
1343 __reg32_deduce_bounds(reg
);
1344 __reg64_deduce_bounds(reg
);
1347 /* Attempts to improve var_off based on unsigned min/max information */
1348 static void __reg_bound_offset(struct bpf_reg_state
*reg
)
1350 struct tnum var64_off
= tnum_intersect(reg
->var_off
,
1351 tnum_range(reg
->umin_value
,
1353 struct tnum var32_off
= tnum_intersect(tnum_subreg(reg
->var_off
),
1354 tnum_range(reg
->u32_min_value
,
1355 reg
->u32_max_value
));
1357 reg
->var_off
= tnum_or(tnum_clear_subreg(var64_off
), var32_off
);
1360 static void __reg_assign_32_into_64(struct bpf_reg_state
*reg
)
1362 reg
->umin_value
= reg
->u32_min_value
;
1363 reg
->umax_value
= reg
->u32_max_value
;
1364 /* Attempt to pull 32-bit signed bounds into 64-bit bounds
1365 * but must be positive otherwise set to worse case bounds
1366 * and refine later from tnum.
1368 if (reg
->s32_min_value
>= 0 && reg
->s32_max_value
>= 0)
1369 reg
->smax_value
= reg
->s32_max_value
;
1371 reg
->smax_value
= U32_MAX
;
1372 if (reg
->s32_min_value
>= 0)
1373 reg
->smin_value
= reg
->s32_min_value
;
1375 reg
->smin_value
= 0;
1378 static void __reg_combine_32_into_64(struct bpf_reg_state
*reg
)
1380 /* special case when 64-bit register has upper 32-bit register
1381 * zeroed. Typically happens after zext or <<32, >>32 sequence
1382 * allowing us to use 32-bit bounds directly,
1384 if (tnum_equals_const(tnum_clear_subreg(reg
->var_off
), 0)) {
1385 __reg_assign_32_into_64(reg
);
1387 /* Otherwise the best we can do is push lower 32bit known and
1388 * unknown bits into register (var_off set from jmp logic)
1389 * then learn as much as possible from the 64-bit tnum
1390 * known and unknown bits. The previous smin/smax bounds are
1391 * invalid here because of jmp32 compare so mark them unknown
1392 * so they do not impact tnum bounds calculation.
1394 __mark_reg64_unbounded(reg
);
1395 __update_reg_bounds(reg
);
1398 /* Intersecting with the old var_off might have improved our bounds
1399 * slightly. e.g. if umax was 0x7f...f and var_off was (0; 0xf...fc),
1400 * then new var_off is (0; 0x7f...fc) which improves our umax.
1402 __reg_deduce_bounds(reg
);
1403 __reg_bound_offset(reg
);
1404 __update_reg_bounds(reg
);
1407 static bool __reg64_bound_s32(s64 a
)
1409 return a
> S32_MIN
&& a
< S32_MAX
;
1412 static bool __reg64_bound_u32(u64 a
)
1414 return a
> U32_MIN
&& a
< U32_MAX
;
1417 static void __reg_combine_64_into_32(struct bpf_reg_state
*reg
)
1419 __mark_reg32_unbounded(reg
);
1421 if (__reg64_bound_s32(reg
->smin_value
) && __reg64_bound_s32(reg
->smax_value
)) {
1422 reg
->s32_min_value
= (s32
)reg
->smin_value
;
1423 reg
->s32_max_value
= (s32
)reg
->smax_value
;
1425 if (__reg64_bound_u32(reg
->umin_value
) && __reg64_bound_u32(reg
->umax_value
)) {
1426 reg
->u32_min_value
= (u32
)reg
->umin_value
;
1427 reg
->u32_max_value
= (u32
)reg
->umax_value
;
1430 /* Intersecting with the old var_off might have improved our bounds
1431 * slightly. e.g. if umax was 0x7f...f and var_off was (0; 0xf...fc),
1432 * then new var_off is (0; 0x7f...fc) which improves our umax.
1434 __reg_deduce_bounds(reg
);
1435 __reg_bound_offset(reg
);
1436 __update_reg_bounds(reg
);
1439 /* Mark a register as having a completely unknown (scalar) value. */
1440 static void __mark_reg_unknown(const struct bpf_verifier_env
*env
,
1441 struct bpf_reg_state
*reg
)
1444 * Clear type, id, off, and union(map_ptr, range) and
1445 * padding between 'type' and union
1447 memset(reg
, 0, offsetof(struct bpf_reg_state
, var_off
));
1448 reg
->type
= SCALAR_VALUE
;
1449 reg
->var_off
= tnum_unknown
;
1451 reg
->precise
= env
->subprog_cnt
> 1 || !env
->bpf_capable
;
1452 __mark_reg_unbounded(reg
);
1455 static void mark_reg_unknown(struct bpf_verifier_env
*env
,
1456 struct bpf_reg_state
*regs
, u32 regno
)
1458 if (WARN_ON(regno
>= MAX_BPF_REG
)) {
1459 verbose(env
, "mark_reg_unknown(regs, %u)\n", regno
);
1460 /* Something bad happened, let's kill all regs except FP */
1461 for (regno
= 0; regno
< BPF_REG_FP
; regno
++)
1462 __mark_reg_not_init(env
, regs
+ regno
);
1465 __mark_reg_unknown(env
, regs
+ regno
);
1468 static void __mark_reg_not_init(const struct bpf_verifier_env
*env
,
1469 struct bpf_reg_state
*reg
)
1471 __mark_reg_unknown(env
, reg
);
1472 reg
->type
= NOT_INIT
;
1475 static void mark_reg_not_init(struct bpf_verifier_env
*env
,
1476 struct bpf_reg_state
*regs
, u32 regno
)
1478 if (WARN_ON(regno
>= MAX_BPF_REG
)) {
1479 verbose(env
, "mark_reg_not_init(regs, %u)\n", regno
);
1480 /* Something bad happened, let's kill all regs except FP */
1481 for (regno
= 0; regno
< BPF_REG_FP
; regno
++)
1482 __mark_reg_not_init(env
, regs
+ regno
);
1485 __mark_reg_not_init(env
, regs
+ regno
);
1488 static void mark_btf_ld_reg(struct bpf_verifier_env
*env
,
1489 struct bpf_reg_state
*regs
, u32 regno
,
1490 enum bpf_reg_type reg_type
,
1491 struct btf
*btf
, u32 btf_id
)
1493 if (reg_type
== SCALAR_VALUE
) {
1494 mark_reg_unknown(env
, regs
, regno
);
1497 mark_reg_known_zero(env
, regs
, regno
);
1498 regs
[regno
].type
= PTR_TO_BTF_ID
;
1499 regs
[regno
].btf
= btf
;
1500 regs
[regno
].btf_id
= btf_id
;
1503 #define DEF_NOT_SUBREG (0)
1504 static void init_reg_state(struct bpf_verifier_env
*env
,
1505 struct bpf_func_state
*state
)
1507 struct bpf_reg_state
*regs
= state
->regs
;
1510 for (i
= 0; i
< MAX_BPF_REG
; i
++) {
1511 mark_reg_not_init(env
, regs
, i
);
1512 regs
[i
].live
= REG_LIVE_NONE
;
1513 regs
[i
].parent
= NULL
;
1514 regs
[i
].subreg_def
= DEF_NOT_SUBREG
;
1518 regs
[BPF_REG_FP
].type
= PTR_TO_STACK
;
1519 mark_reg_known_zero(env
, regs
, BPF_REG_FP
);
1520 regs
[BPF_REG_FP
].frameno
= state
->frameno
;
1523 #define BPF_MAIN_FUNC (-1)
1524 static void init_func_state(struct bpf_verifier_env
*env
,
1525 struct bpf_func_state
*state
,
1526 int callsite
, int frameno
, int subprogno
)
1528 state
->callsite
= callsite
;
1529 state
->frameno
= frameno
;
1530 state
->subprogno
= subprogno
;
1531 init_reg_state(env
, state
);
1534 /* Similar to push_stack(), but for async callbacks */
1535 static struct bpf_verifier_state
*push_async_cb(struct bpf_verifier_env
*env
,
1536 int insn_idx
, int prev_insn_idx
,
1539 struct bpf_verifier_stack_elem
*elem
;
1540 struct bpf_func_state
*frame
;
1542 elem
= kzalloc(sizeof(struct bpf_verifier_stack_elem
), GFP_KERNEL
);
1546 elem
->insn_idx
= insn_idx
;
1547 elem
->prev_insn_idx
= prev_insn_idx
;
1548 elem
->next
= env
->head
;
1549 elem
->log_pos
= env
->log
.len_used
;
1552 if (env
->stack_size
> BPF_COMPLEXITY_LIMIT_JMP_SEQ
) {
1554 "The sequence of %d jumps is too complex for async cb.\n",
1558 /* Unlike push_stack() do not copy_verifier_state().
1559 * The caller state doesn't matter.
1560 * This is async callback. It starts in a fresh stack.
1561 * Initialize it similar to do_check_common().
1563 elem
->st
.branches
= 1;
1564 frame
= kzalloc(sizeof(*frame
), GFP_KERNEL
);
1567 init_func_state(env
, frame
,
1568 BPF_MAIN_FUNC
/* callsite */,
1569 0 /* frameno within this callchain */,
1570 subprog
/* subprog number within this prog */);
1571 elem
->st
.frame
[0] = frame
;
1574 free_verifier_state(env
->cur_state
, true);
1575 env
->cur_state
= NULL
;
1576 /* pop all elements and return */
1577 while (!pop_stack(env
, NULL
, NULL
, false));
1583 SRC_OP
, /* register is used as source operand */
1584 DST_OP
, /* register is used as destination operand */
1585 DST_OP_NO_MARK
/* same as above, check only, don't mark */
1588 static int cmp_subprogs(const void *a
, const void *b
)
1590 return ((struct bpf_subprog_info
*)a
)->start
-
1591 ((struct bpf_subprog_info
*)b
)->start
;
1594 static int find_subprog(struct bpf_verifier_env
*env
, int off
)
1596 struct bpf_subprog_info
*p
;
1598 p
= bsearch(&off
, env
->subprog_info
, env
->subprog_cnt
,
1599 sizeof(env
->subprog_info
[0]), cmp_subprogs
);
1602 return p
- env
->subprog_info
;
1606 static int add_subprog(struct bpf_verifier_env
*env
, int off
)
1608 int insn_cnt
= env
->prog
->len
;
1611 if (off
>= insn_cnt
|| off
< 0) {
1612 verbose(env
, "call to invalid destination\n");
1615 ret
= find_subprog(env
, off
);
1618 if (env
->subprog_cnt
>= BPF_MAX_SUBPROGS
) {
1619 verbose(env
, "too many subprograms\n");
1622 /* determine subprog starts. The end is one before the next starts */
1623 env
->subprog_info
[env
->subprog_cnt
++].start
= off
;
1624 sort(env
->subprog_info
, env
->subprog_cnt
,
1625 sizeof(env
->subprog_info
[0]), cmp_subprogs
, NULL
);
1626 return env
->subprog_cnt
- 1;
1629 struct bpf_kfunc_desc
{
1630 struct btf_func_model func_model
;
1635 #define MAX_KFUNC_DESCS 256
1636 struct bpf_kfunc_desc_tab
{
1637 struct bpf_kfunc_desc descs
[MAX_KFUNC_DESCS
];
1641 static int kfunc_desc_cmp_by_id(const void *a
, const void *b
)
1643 const struct bpf_kfunc_desc
*d0
= a
;
1644 const struct bpf_kfunc_desc
*d1
= b
;
1646 /* func_id is not greater than BTF_MAX_TYPE */
1647 return d0
->func_id
- d1
->func_id
;
1650 static const struct bpf_kfunc_desc
*
1651 find_kfunc_desc(const struct bpf_prog
*prog
, u32 func_id
)
1653 struct bpf_kfunc_desc desc
= {
1656 struct bpf_kfunc_desc_tab
*tab
;
1658 tab
= prog
->aux
->kfunc_tab
;
1659 return bsearch(&desc
, tab
->descs
, tab
->nr_descs
,
1660 sizeof(tab
->descs
[0]), kfunc_desc_cmp_by_id
);
1663 static int add_kfunc_call(struct bpf_verifier_env
*env
, u32 func_id
)
1665 const struct btf_type
*func
, *func_proto
;
1666 struct bpf_kfunc_desc_tab
*tab
;
1667 struct bpf_prog_aux
*prog_aux
;
1668 struct bpf_kfunc_desc
*desc
;
1669 const char *func_name
;
1673 prog_aux
= env
->prog
->aux
;
1674 tab
= prog_aux
->kfunc_tab
;
1677 verbose(env
, "calling kernel function is not supported without CONFIG_DEBUG_INFO_BTF\n");
1681 if (!env
->prog
->jit_requested
) {
1682 verbose(env
, "JIT is required for calling kernel function\n");
1686 if (!bpf_jit_supports_kfunc_call()) {
1687 verbose(env
, "JIT does not support calling kernel function\n");
1691 if (!env
->prog
->gpl_compatible
) {
1692 verbose(env
, "cannot call kernel function from non-GPL compatible program\n");
1696 tab
= kzalloc(sizeof(*tab
), GFP_KERNEL
);
1699 prog_aux
->kfunc_tab
= tab
;
1702 if (find_kfunc_desc(env
->prog
, func_id
))
1705 if (tab
->nr_descs
== MAX_KFUNC_DESCS
) {
1706 verbose(env
, "too many different kernel function calls\n");
1710 func
= btf_type_by_id(btf_vmlinux
, func_id
);
1711 if (!func
|| !btf_type_is_func(func
)) {
1712 verbose(env
, "kernel btf_id %u is not a function\n",
1716 func_proto
= btf_type_by_id(btf_vmlinux
, func
->type
);
1717 if (!func_proto
|| !btf_type_is_func_proto(func_proto
)) {
1718 verbose(env
, "kernel function btf_id %u does not have a valid func_proto\n",
1723 func_name
= btf_name_by_offset(btf_vmlinux
, func
->name_off
);
1724 addr
= kallsyms_lookup_name(func_name
);
1726 verbose(env
, "cannot find address for kernel function %s\n",
1731 desc
= &tab
->descs
[tab
->nr_descs
++];
1732 desc
->func_id
= func_id
;
1733 desc
->imm
= BPF_CAST_CALL(addr
) - __bpf_call_base
;
1734 err
= btf_distill_func_proto(&env
->log
, btf_vmlinux
,
1735 func_proto
, func_name
,
1738 sort(tab
->descs
, tab
->nr_descs
, sizeof(tab
->descs
[0]),
1739 kfunc_desc_cmp_by_id
, NULL
);
1743 static int kfunc_desc_cmp_by_imm(const void *a
, const void *b
)
1745 const struct bpf_kfunc_desc
*d0
= a
;
1746 const struct bpf_kfunc_desc
*d1
= b
;
1748 if (d0
->imm
> d1
->imm
)
1750 else if (d0
->imm
< d1
->imm
)
1755 static void sort_kfunc_descs_by_imm(struct bpf_prog
*prog
)
1757 struct bpf_kfunc_desc_tab
*tab
;
1759 tab
= prog
->aux
->kfunc_tab
;
1763 sort(tab
->descs
, tab
->nr_descs
, sizeof(tab
->descs
[0]),
1764 kfunc_desc_cmp_by_imm
, NULL
);
1767 bool bpf_prog_has_kfunc_call(const struct bpf_prog
*prog
)
1769 return !!prog
->aux
->kfunc_tab
;
1772 const struct btf_func_model
*
1773 bpf_jit_find_kfunc_model(const struct bpf_prog
*prog
,
1774 const struct bpf_insn
*insn
)
1776 const struct bpf_kfunc_desc desc
= {
1779 const struct bpf_kfunc_desc
*res
;
1780 struct bpf_kfunc_desc_tab
*tab
;
1782 tab
= prog
->aux
->kfunc_tab
;
1783 res
= bsearch(&desc
, tab
->descs
, tab
->nr_descs
,
1784 sizeof(tab
->descs
[0]), kfunc_desc_cmp_by_imm
);
1786 return res
? &res
->func_model
: NULL
;
1789 static int add_subprog_and_kfunc(struct bpf_verifier_env
*env
)
1791 struct bpf_subprog_info
*subprog
= env
->subprog_info
;
1792 struct bpf_insn
*insn
= env
->prog
->insnsi
;
1793 int i
, ret
, insn_cnt
= env
->prog
->len
;
1795 /* Add entry function. */
1796 ret
= add_subprog(env
, 0);
1800 for (i
= 0; i
< insn_cnt
; i
++, insn
++) {
1801 if (!bpf_pseudo_func(insn
) && !bpf_pseudo_call(insn
) &&
1802 !bpf_pseudo_kfunc_call(insn
))
1805 if (!env
->bpf_capable
) {
1806 verbose(env
, "loading/calling other bpf or kernel functions are allowed for CAP_BPF and CAP_SYS_ADMIN\n");
1810 if (bpf_pseudo_func(insn
)) {
1811 ret
= add_subprog(env
, i
+ insn
->imm
+ 1);
1813 /* remember subprog */
1815 } else if (bpf_pseudo_call(insn
)) {
1816 ret
= add_subprog(env
, i
+ insn
->imm
+ 1);
1818 ret
= add_kfunc_call(env
, insn
->imm
);
1825 /* Add a fake 'exit' subprog which could simplify subprog iteration
1826 * logic. 'subprog_cnt' should not be increased.
1828 subprog
[env
->subprog_cnt
].start
= insn_cnt
;
1830 if (env
->log
.level
& BPF_LOG_LEVEL2
)
1831 for (i
= 0; i
< env
->subprog_cnt
; i
++)
1832 verbose(env
, "func#%d @%d\n", i
, subprog
[i
].start
);
1837 static int check_subprogs(struct bpf_verifier_env
*env
)
1839 int i
, subprog_start
, subprog_end
, off
, cur_subprog
= 0;
1840 struct bpf_subprog_info
*subprog
= env
->subprog_info
;
1841 struct bpf_insn
*insn
= env
->prog
->insnsi
;
1842 int insn_cnt
= env
->prog
->len
;
1844 /* now check that all jumps are within the same subprog */
1845 subprog_start
= subprog
[cur_subprog
].start
;
1846 subprog_end
= subprog
[cur_subprog
+ 1].start
;
1847 for (i
= 0; i
< insn_cnt
; i
++) {
1848 u8 code
= insn
[i
].code
;
1850 if (code
== (BPF_JMP
| BPF_CALL
) &&
1851 insn
[i
].imm
== BPF_FUNC_tail_call
&&
1852 insn
[i
].src_reg
!= BPF_PSEUDO_CALL
)
1853 subprog
[cur_subprog
].has_tail_call
= true;
1854 if (BPF_CLASS(code
) == BPF_LD
&&
1855 (BPF_MODE(code
) == BPF_ABS
|| BPF_MODE(code
) == BPF_IND
))
1856 subprog
[cur_subprog
].has_ld_abs
= true;
1857 if (BPF_CLASS(code
) != BPF_JMP
&& BPF_CLASS(code
) != BPF_JMP32
)
1859 if (BPF_OP(code
) == BPF_EXIT
|| BPF_OP(code
) == BPF_CALL
)
1861 off
= i
+ insn
[i
].off
+ 1;
1862 if (off
< subprog_start
|| off
>= subprog_end
) {
1863 verbose(env
, "jump out of range from insn %d to %d\n", i
, off
);
1867 if (i
== subprog_end
- 1) {
1868 /* to avoid fall-through from one subprog into another
1869 * the last insn of the subprog should be either exit
1870 * or unconditional jump back
1872 if (code
!= (BPF_JMP
| BPF_EXIT
) &&
1873 code
!= (BPF_JMP
| BPF_JA
)) {
1874 verbose(env
, "last insn is not an exit or jmp\n");
1877 subprog_start
= subprog_end
;
1879 if (cur_subprog
< env
->subprog_cnt
)
1880 subprog_end
= subprog
[cur_subprog
+ 1].start
;
1886 /* Parentage chain of this register (or stack slot) should take care of all
1887 * issues like callee-saved registers, stack slot allocation time, etc.
1889 static int mark_reg_read(struct bpf_verifier_env
*env
,
1890 const struct bpf_reg_state
*state
,
1891 struct bpf_reg_state
*parent
, u8 flag
)
1893 bool writes
= parent
== state
->parent
; /* Observe write marks */
1897 /* if read wasn't screened by an earlier write ... */
1898 if (writes
&& state
->live
& REG_LIVE_WRITTEN
)
1900 if (parent
->live
& REG_LIVE_DONE
) {
1901 verbose(env
, "verifier BUG type %s var_off %lld off %d\n",
1902 reg_type_str
[parent
->type
],
1903 parent
->var_off
.value
, parent
->off
);
1906 /* The first condition is more likely to be true than the
1907 * second, checked it first.
1909 if ((parent
->live
& REG_LIVE_READ
) == flag
||
1910 parent
->live
& REG_LIVE_READ64
)
1911 /* The parentage chain never changes and
1912 * this parent was already marked as LIVE_READ.
1913 * There is no need to keep walking the chain again and
1914 * keep re-marking all parents as LIVE_READ.
1915 * This case happens when the same register is read
1916 * multiple times without writes into it in-between.
1917 * Also, if parent has the stronger REG_LIVE_READ64 set,
1918 * then no need to set the weak REG_LIVE_READ32.
1921 /* ... then we depend on parent's value */
1922 parent
->live
|= flag
;
1923 /* REG_LIVE_READ64 overrides REG_LIVE_READ32. */
1924 if (flag
== REG_LIVE_READ64
)
1925 parent
->live
&= ~REG_LIVE_READ32
;
1927 parent
= state
->parent
;
1932 if (env
->longest_mark_read_walk
< cnt
)
1933 env
->longest_mark_read_walk
= cnt
;
1937 /* This function is supposed to be used by the following 32-bit optimization
1938 * code only. It returns TRUE if the source or destination register operates
1939 * on 64-bit, otherwise return FALSE.
1941 static bool is_reg64(struct bpf_verifier_env
*env
, struct bpf_insn
*insn
,
1942 u32 regno
, struct bpf_reg_state
*reg
, enum reg_arg_type t
)
1947 class = BPF_CLASS(code
);
1949 if (class == BPF_JMP
) {
1950 /* BPF_EXIT for "main" will reach here. Return TRUE
1955 if (op
== BPF_CALL
) {
1956 /* BPF to BPF call will reach here because of marking
1957 * caller saved clobber with DST_OP_NO_MARK for which we
1958 * don't care the register def because they are anyway
1959 * marked as NOT_INIT already.
1961 if (insn
->src_reg
== BPF_PSEUDO_CALL
)
1963 /* Helper call will reach here because of arg type
1964 * check, conservatively return TRUE.
1973 if (class == BPF_ALU64
|| class == BPF_JMP
||
1974 /* BPF_END always use BPF_ALU class. */
1975 (class == BPF_ALU
&& op
== BPF_END
&& insn
->imm
== 64))
1978 if (class == BPF_ALU
|| class == BPF_JMP32
)
1981 if (class == BPF_LDX
) {
1983 return BPF_SIZE(code
) == BPF_DW
;
1984 /* LDX source must be ptr. */
1988 if (class == BPF_STX
) {
1989 /* BPF_STX (including atomic variants) has multiple source
1990 * operands, one of which is a ptr. Check whether the caller is
1993 if (t
== SRC_OP
&& reg
->type
!= SCALAR_VALUE
)
1995 return BPF_SIZE(code
) == BPF_DW
;
1998 if (class == BPF_LD
) {
1999 u8 mode
= BPF_MODE(code
);
2002 if (mode
== BPF_IMM
)
2005 /* Both LD_IND and LD_ABS return 32-bit data. */
2009 /* Implicit ctx ptr. */
2010 if (regno
== BPF_REG_6
)
2013 /* Explicit source could be any width. */
2017 if (class == BPF_ST
)
2018 /* The only source register for BPF_ST is a ptr. */
2021 /* Conservatively return true at default. */
2025 /* Return the regno defined by the insn, or -1. */
2026 static int insn_def_regno(const struct bpf_insn
*insn
)
2028 switch (BPF_CLASS(insn
->code
)) {
2034 if (BPF_MODE(insn
->code
) == BPF_ATOMIC
&&
2035 (insn
->imm
& BPF_FETCH
)) {
2036 if (insn
->imm
== BPF_CMPXCHG
)
2039 return insn
->src_reg
;
2044 return insn
->dst_reg
;
2048 /* Return TRUE if INSN has defined any 32-bit value explicitly. */
2049 static bool insn_has_def32(struct bpf_verifier_env
*env
, struct bpf_insn
*insn
)
2051 int dst_reg
= insn_def_regno(insn
);
2056 return !is_reg64(env
, insn
, dst_reg
, NULL
, DST_OP
);
2059 static void mark_insn_zext(struct bpf_verifier_env
*env
,
2060 struct bpf_reg_state
*reg
)
2062 s32 def_idx
= reg
->subreg_def
;
2064 if (def_idx
== DEF_NOT_SUBREG
)
2067 env
->insn_aux_data
[def_idx
- 1].zext_dst
= true;
2068 /* The dst will be zero extended, so won't be sub-register anymore. */
2069 reg
->subreg_def
= DEF_NOT_SUBREG
;
2072 static int check_reg_arg(struct bpf_verifier_env
*env
, u32 regno
,
2073 enum reg_arg_type t
)
2075 struct bpf_verifier_state
*vstate
= env
->cur_state
;
2076 struct bpf_func_state
*state
= vstate
->frame
[vstate
->curframe
];
2077 struct bpf_insn
*insn
= env
->prog
->insnsi
+ env
->insn_idx
;
2078 struct bpf_reg_state
*reg
, *regs
= state
->regs
;
2081 if (regno
>= MAX_BPF_REG
) {
2082 verbose(env
, "R%d is invalid\n", regno
);
2087 rw64
= is_reg64(env
, insn
, regno
, reg
, t
);
2089 /* check whether register used as source operand can be read */
2090 if (reg
->type
== NOT_INIT
) {
2091 verbose(env
, "R%d !read_ok\n", regno
);
2094 /* We don't need to worry about FP liveness because it's read-only */
2095 if (regno
== BPF_REG_FP
)
2099 mark_insn_zext(env
, reg
);
2101 return mark_reg_read(env
, reg
, reg
->parent
,
2102 rw64
? REG_LIVE_READ64
: REG_LIVE_READ32
);
2104 /* check whether register used as dest operand can be written to */
2105 if (regno
== BPF_REG_FP
) {
2106 verbose(env
, "frame pointer is read only\n");
2109 reg
->live
|= REG_LIVE_WRITTEN
;
2110 reg
->subreg_def
= rw64
? DEF_NOT_SUBREG
: env
->insn_idx
+ 1;
2112 mark_reg_unknown(env
, regs
, regno
);
2117 /* for any branch, call, exit record the history of jmps in the given state */
2118 static int push_jmp_history(struct bpf_verifier_env
*env
,
2119 struct bpf_verifier_state
*cur
)
2121 u32 cnt
= cur
->jmp_history_cnt
;
2122 struct bpf_idx_pair
*p
;
2125 p
= krealloc(cur
->jmp_history
, cnt
* sizeof(*p
), GFP_USER
);
2128 p
[cnt
- 1].idx
= env
->insn_idx
;
2129 p
[cnt
- 1].prev_idx
= env
->prev_insn_idx
;
2130 cur
->jmp_history
= p
;
2131 cur
->jmp_history_cnt
= cnt
;
2135 /* Backtrack one insn at a time. If idx is not at the top of recorded
2136 * history then previous instruction came from straight line execution.
2138 static int get_prev_insn_idx(struct bpf_verifier_state
*st
, int i
,
2143 if (cnt
&& st
->jmp_history
[cnt
- 1].idx
== i
) {
2144 i
= st
->jmp_history
[cnt
- 1].prev_idx
;
2152 static const char *disasm_kfunc_name(void *data
, const struct bpf_insn
*insn
)
2154 const struct btf_type
*func
;
2156 if (insn
->src_reg
!= BPF_PSEUDO_KFUNC_CALL
)
2159 func
= btf_type_by_id(btf_vmlinux
, insn
->imm
);
2160 return btf_name_by_offset(btf_vmlinux
, func
->name_off
);
2163 /* For given verifier state backtrack_insn() is called from the last insn to
2164 * the first insn. Its purpose is to compute a bitmask of registers and
2165 * stack slots that needs precision in the parent verifier state.
2167 static int backtrack_insn(struct bpf_verifier_env
*env
, int idx
,
2168 u32
*reg_mask
, u64
*stack_mask
)
2170 const struct bpf_insn_cbs cbs
= {
2171 .cb_call
= disasm_kfunc_name
,
2172 .cb_print
= verbose
,
2173 .private_data
= env
,
2175 struct bpf_insn
*insn
= env
->prog
->insnsi
+ idx
;
2176 u8
class = BPF_CLASS(insn
->code
);
2177 u8 opcode
= BPF_OP(insn
->code
);
2178 u8 mode
= BPF_MODE(insn
->code
);
2179 u32 dreg
= 1u << insn
->dst_reg
;
2180 u32 sreg
= 1u << insn
->src_reg
;
2183 if (insn
->code
== 0)
2185 if (env
->log
.level
& BPF_LOG_LEVEL
) {
2186 verbose(env
, "regs=%x stack=%llx before ", *reg_mask
, *stack_mask
);
2187 verbose(env
, "%d: ", idx
);
2188 print_bpf_insn(&cbs
, insn
, env
->allow_ptr_leaks
);
2191 if (class == BPF_ALU
|| class == BPF_ALU64
) {
2192 if (!(*reg_mask
& dreg
))
2194 if (opcode
== BPF_MOV
) {
2195 if (BPF_SRC(insn
->code
) == BPF_X
) {
2197 * dreg needs precision after this insn
2198 * sreg needs precision before this insn
2204 * dreg needs precision after this insn.
2205 * Corresponding register is already marked
2206 * as precise=true in this verifier state.
2207 * No further markings in parent are necessary
2212 if (BPF_SRC(insn
->code
) == BPF_X
) {
2214 * both dreg and sreg need precision
2219 * dreg still needs precision before this insn
2222 } else if (class == BPF_LDX
) {
2223 if (!(*reg_mask
& dreg
))
2227 /* scalars can only be spilled into stack w/o losing precision.
2228 * Load from any other memory can be zero extended.
2229 * The desire to keep that precision is already indicated
2230 * by 'precise' mark in corresponding register of this state.
2231 * No further tracking necessary.
2233 if (insn
->src_reg
!= BPF_REG_FP
)
2235 if (BPF_SIZE(insn
->code
) != BPF_DW
)
2238 /* dreg = *(u64 *)[fp - off] was a fill from the stack.
2239 * that [fp - off] slot contains scalar that needs to be
2240 * tracked with precision
2242 spi
= (-insn
->off
- 1) / BPF_REG_SIZE
;
2244 verbose(env
, "BUG spi %d\n", spi
);
2245 WARN_ONCE(1, "verifier backtracking bug");
2248 *stack_mask
|= 1ull << spi
;
2249 } else if (class == BPF_STX
|| class == BPF_ST
) {
2250 if (*reg_mask
& dreg
)
2251 /* stx & st shouldn't be using _scalar_ dst_reg
2252 * to access memory. It means backtracking
2253 * encountered a case of pointer subtraction.
2256 /* scalars can only be spilled into stack */
2257 if (insn
->dst_reg
!= BPF_REG_FP
)
2259 if (BPF_SIZE(insn
->code
) != BPF_DW
)
2261 spi
= (-insn
->off
- 1) / BPF_REG_SIZE
;
2263 verbose(env
, "BUG spi %d\n", spi
);
2264 WARN_ONCE(1, "verifier backtracking bug");
2267 if (!(*stack_mask
& (1ull << spi
)))
2269 *stack_mask
&= ~(1ull << spi
);
2270 if (class == BPF_STX
)
2272 } else if (class == BPF_JMP
|| class == BPF_JMP32
) {
2273 if (opcode
== BPF_CALL
) {
2274 if (insn
->src_reg
== BPF_PSEUDO_CALL
)
2276 /* regular helper call sets R0 */
2278 if (*reg_mask
& 0x3f) {
2279 /* if backtracing was looking for registers R1-R5
2280 * they should have been found already.
2282 verbose(env
, "BUG regs %x\n", *reg_mask
);
2283 WARN_ONCE(1, "verifier backtracking bug");
2286 } else if (opcode
== BPF_EXIT
) {
2289 } else if (class == BPF_LD
) {
2290 if (!(*reg_mask
& dreg
))
2293 /* It's ld_imm64 or ld_abs or ld_ind.
2294 * For ld_imm64 no further tracking of precision
2295 * into parent is necessary
2297 if (mode
== BPF_IND
|| mode
== BPF_ABS
)
2298 /* to be analyzed */
2304 /* the scalar precision tracking algorithm:
2305 * . at the start all registers have precise=false.
2306 * . scalar ranges are tracked as normal through alu and jmp insns.
2307 * . once precise value of the scalar register is used in:
2308 * . ptr + scalar alu
2309 * . if (scalar cond K|scalar)
2310 * . helper_call(.., scalar, ...) where ARG_CONST is expected
2311 * backtrack through the verifier states and mark all registers and
2312 * stack slots with spilled constants that these scalar regisers
2313 * should be precise.
2314 * . during state pruning two registers (or spilled stack slots)
2315 * are equivalent if both are not precise.
2317 * Note the verifier cannot simply walk register parentage chain,
2318 * since many different registers and stack slots could have been
2319 * used to compute single precise scalar.
2321 * The approach of starting with precise=true for all registers and then
2322 * backtrack to mark a register as not precise when the verifier detects
2323 * that program doesn't care about specific value (e.g., when helper
2324 * takes register as ARG_ANYTHING parameter) is not safe.
2326 * It's ok to walk single parentage chain of the verifier states.
2327 * It's possible that this backtracking will go all the way till 1st insn.
2328 * All other branches will be explored for needing precision later.
2330 * The backtracking needs to deal with cases like:
2331 * R8=map_value(id=0,off=0,ks=4,vs=1952,imm=0) R9_w=map_value(id=0,off=40,ks=4,vs=1952,imm=0)
2334 * if r5 > 0x79f goto pc+7
2335 * R5_w=inv(id=0,umax_value=1951,var_off=(0x0; 0x7ff))
2338 * call bpf_perf_event_output#25
2339 * where .arg5_type = ARG_CONST_SIZE_OR_ZERO
2343 * call foo // uses callee's r6 inside to compute r0
2347 * to track above reg_mask/stack_mask needs to be independent for each frame.
2349 * Also if parent's curframe > frame where backtracking started,
2350 * the verifier need to mark registers in both frames, otherwise callees
2351 * may incorrectly prune callers. This is similar to
2352 * commit 7640ead93924 ("bpf: verifier: make sure callees don't prune with caller differences")
2354 * For now backtracking falls back into conservative marking.
2356 static void mark_all_scalars_precise(struct bpf_verifier_env
*env
,
2357 struct bpf_verifier_state
*st
)
2359 struct bpf_func_state
*func
;
2360 struct bpf_reg_state
*reg
;
2363 /* big hammer: mark all scalars precise in this path.
2364 * pop_stack may still get !precise scalars.
2366 for (; st
; st
= st
->parent
)
2367 for (i
= 0; i
<= st
->curframe
; i
++) {
2368 func
= st
->frame
[i
];
2369 for (j
= 0; j
< BPF_REG_FP
; j
++) {
2370 reg
= &func
->regs
[j
];
2371 if (reg
->type
!= SCALAR_VALUE
)
2373 reg
->precise
= true;
2375 for (j
= 0; j
< func
->allocated_stack
/ BPF_REG_SIZE
; j
++) {
2376 if (func
->stack
[j
].slot_type
[0] != STACK_SPILL
)
2378 reg
= &func
->stack
[j
].spilled_ptr
;
2379 if (reg
->type
!= SCALAR_VALUE
)
2381 reg
->precise
= true;
2386 static int __mark_chain_precision(struct bpf_verifier_env
*env
, int regno
,
2389 struct bpf_verifier_state
*st
= env
->cur_state
;
2390 int first_idx
= st
->first_insn_idx
;
2391 int last_idx
= env
->insn_idx
;
2392 struct bpf_func_state
*func
;
2393 struct bpf_reg_state
*reg
;
2394 u32 reg_mask
= regno
>= 0 ? 1u << regno
: 0;
2395 u64 stack_mask
= spi
>= 0 ? 1ull << spi
: 0;
2396 bool skip_first
= true;
2397 bool new_marks
= false;
2400 if (!env
->bpf_capable
)
2403 func
= st
->frame
[st
->curframe
];
2405 reg
= &func
->regs
[regno
];
2406 if (reg
->type
!= SCALAR_VALUE
) {
2407 WARN_ONCE(1, "backtracing misuse");
2414 reg
->precise
= true;
2418 if (func
->stack
[spi
].slot_type
[0] != STACK_SPILL
) {
2422 reg
= &func
->stack
[spi
].spilled_ptr
;
2423 if (reg
->type
!= SCALAR_VALUE
) {
2431 reg
->precise
= true;
2437 if (!reg_mask
&& !stack_mask
)
2440 DECLARE_BITMAP(mask
, 64);
2441 u32 history
= st
->jmp_history_cnt
;
2443 if (env
->log
.level
& BPF_LOG_LEVEL
)
2444 verbose(env
, "last_idx %d first_idx %d\n", last_idx
, first_idx
);
2445 for (i
= last_idx
;;) {
2450 err
= backtrack_insn(env
, i
, ®_mask
, &stack_mask
);
2452 if (err
== -ENOTSUPP
) {
2453 mark_all_scalars_precise(env
, st
);
2458 if (!reg_mask
&& !stack_mask
)
2459 /* Found assignment(s) into tracked register in this state.
2460 * Since this state is already marked, just return.
2461 * Nothing to be tracked further in the parent state.
2466 i
= get_prev_insn_idx(st
, i
, &history
);
2467 if (i
>= env
->prog
->len
) {
2468 /* This can happen if backtracking reached insn 0
2469 * and there are still reg_mask or stack_mask
2471 * It means the backtracking missed the spot where
2472 * particular register was initialized with a constant.
2474 verbose(env
, "BUG backtracking idx %d\n", i
);
2475 WARN_ONCE(1, "verifier backtracking bug");
2484 func
= st
->frame
[st
->curframe
];
2485 bitmap_from_u64(mask
, reg_mask
);
2486 for_each_set_bit(i
, mask
, 32) {
2487 reg
= &func
->regs
[i
];
2488 if (reg
->type
!= SCALAR_VALUE
) {
2489 reg_mask
&= ~(1u << i
);
2494 reg
->precise
= true;
2497 bitmap_from_u64(mask
, stack_mask
);
2498 for_each_set_bit(i
, mask
, 64) {
2499 if (i
>= func
->allocated_stack
/ BPF_REG_SIZE
) {
2500 /* the sequence of instructions:
2502 * 3: (7b) *(u64 *)(r3 -8) = r0
2503 * 4: (79) r4 = *(u64 *)(r10 -8)
2504 * doesn't contain jmps. It's backtracked
2505 * as a single block.
2506 * During backtracking insn 3 is not recognized as
2507 * stack access, so at the end of backtracking
2508 * stack slot fp-8 is still marked in stack_mask.
2509 * However the parent state may not have accessed
2510 * fp-8 and it's "unallocated" stack space.
2511 * In such case fallback to conservative.
2513 mark_all_scalars_precise(env
, st
);
2517 if (func
->stack
[i
].slot_type
[0] != STACK_SPILL
) {
2518 stack_mask
&= ~(1ull << i
);
2521 reg
= &func
->stack
[i
].spilled_ptr
;
2522 if (reg
->type
!= SCALAR_VALUE
) {
2523 stack_mask
&= ~(1ull << i
);
2528 reg
->precise
= true;
2530 if (env
->log
.level
& BPF_LOG_LEVEL
) {
2531 print_verifier_state(env
, func
);
2532 verbose(env
, "parent %s regs=%x stack=%llx marks\n",
2533 new_marks
? "didn't have" : "already had",
2534 reg_mask
, stack_mask
);
2537 if (!reg_mask
&& !stack_mask
)
2542 last_idx
= st
->last_insn_idx
;
2543 first_idx
= st
->first_insn_idx
;
2548 static int mark_chain_precision(struct bpf_verifier_env
*env
, int regno
)
2550 return __mark_chain_precision(env
, regno
, -1);
2553 static int mark_chain_precision_stack(struct bpf_verifier_env
*env
, int spi
)
2555 return __mark_chain_precision(env
, -1, spi
);
2558 static bool is_spillable_regtype(enum bpf_reg_type type
)
2561 case PTR_TO_MAP_VALUE
:
2562 case PTR_TO_MAP_VALUE_OR_NULL
:
2566 case PTR_TO_PACKET_META
:
2567 case PTR_TO_PACKET_END
:
2568 case PTR_TO_FLOW_KEYS
:
2569 case CONST_PTR_TO_MAP
:
2571 case PTR_TO_SOCKET_OR_NULL
:
2572 case PTR_TO_SOCK_COMMON
:
2573 case PTR_TO_SOCK_COMMON_OR_NULL
:
2574 case PTR_TO_TCP_SOCK
:
2575 case PTR_TO_TCP_SOCK_OR_NULL
:
2576 case PTR_TO_XDP_SOCK
:
2578 case PTR_TO_BTF_ID_OR_NULL
:
2579 case PTR_TO_RDONLY_BUF
:
2580 case PTR_TO_RDONLY_BUF_OR_NULL
:
2581 case PTR_TO_RDWR_BUF
:
2582 case PTR_TO_RDWR_BUF_OR_NULL
:
2583 case PTR_TO_PERCPU_BTF_ID
:
2585 case PTR_TO_MEM_OR_NULL
:
2587 case PTR_TO_MAP_KEY
:
2594 /* Does this register contain a constant zero? */
2595 static bool register_is_null(struct bpf_reg_state
*reg
)
2597 return reg
->type
== SCALAR_VALUE
&& tnum_equals_const(reg
->var_off
, 0);
2600 static bool register_is_const(struct bpf_reg_state
*reg
)
2602 return reg
->type
== SCALAR_VALUE
&& tnum_is_const(reg
->var_off
);
2605 static bool __is_scalar_unbounded(struct bpf_reg_state
*reg
)
2607 return tnum_is_unknown(reg
->var_off
) &&
2608 reg
->smin_value
== S64_MIN
&& reg
->smax_value
== S64_MAX
&&
2609 reg
->umin_value
== 0 && reg
->umax_value
== U64_MAX
&&
2610 reg
->s32_min_value
== S32_MIN
&& reg
->s32_max_value
== S32_MAX
&&
2611 reg
->u32_min_value
== 0 && reg
->u32_max_value
== U32_MAX
;
2614 static bool register_is_bounded(struct bpf_reg_state
*reg
)
2616 return reg
->type
== SCALAR_VALUE
&& !__is_scalar_unbounded(reg
);
2619 static bool __is_pointer_value(bool allow_ptr_leaks
,
2620 const struct bpf_reg_state
*reg
)
2622 if (allow_ptr_leaks
)
2625 return reg
->type
!= SCALAR_VALUE
;
2628 static void save_register_state(struct bpf_func_state
*state
,
2629 int spi
, struct bpf_reg_state
*reg
)
2633 state
->stack
[spi
].spilled_ptr
= *reg
;
2634 state
->stack
[spi
].spilled_ptr
.live
|= REG_LIVE_WRITTEN
;
2636 for (i
= 0; i
< BPF_REG_SIZE
; i
++)
2637 state
->stack
[spi
].slot_type
[i
] = STACK_SPILL
;
2640 /* check_stack_{read,write}_fixed_off functions track spill/fill of registers,
2641 * stack boundary and alignment are checked in check_mem_access()
2643 static int check_stack_write_fixed_off(struct bpf_verifier_env
*env
,
2644 /* stack frame we're writing to */
2645 struct bpf_func_state
*state
,
2646 int off
, int size
, int value_regno
,
2649 struct bpf_func_state
*cur
; /* state of the current function */
2650 int i
, slot
= -off
- 1, spi
= slot
/ BPF_REG_SIZE
, err
;
2651 u32 dst_reg
= env
->prog
->insnsi
[insn_idx
].dst_reg
;
2652 struct bpf_reg_state
*reg
= NULL
;
2654 err
= grow_stack_state(state
, round_up(slot
+ 1, BPF_REG_SIZE
));
2657 /* caller checked that off % size == 0 and -MAX_BPF_STACK <= off < 0,
2658 * so it's aligned access and [off, off + size) are within stack limits
2660 if (!env
->allow_ptr_leaks
&&
2661 state
->stack
[spi
].slot_type
[0] == STACK_SPILL
&&
2662 size
!= BPF_REG_SIZE
) {
2663 verbose(env
, "attempt to corrupt spilled pointer on stack\n");
2667 cur
= env
->cur_state
->frame
[env
->cur_state
->curframe
];
2668 if (value_regno
>= 0)
2669 reg
= &cur
->regs
[value_regno
];
2670 if (!env
->bypass_spec_v4
) {
2671 bool sanitize
= reg
&& is_spillable_regtype(reg
->type
);
2673 for (i
= 0; i
< size
; i
++) {
2674 if (state
->stack
[spi
].slot_type
[i
] == STACK_INVALID
) {
2681 env
->insn_aux_data
[insn_idx
].sanitize_stack_spill
= true;
2684 if (reg
&& size
== BPF_REG_SIZE
&& register_is_bounded(reg
) &&
2685 !register_is_null(reg
) && env
->bpf_capable
) {
2686 if (dst_reg
!= BPF_REG_FP
) {
2687 /* The backtracking logic can only recognize explicit
2688 * stack slot address like [fp - 8]. Other spill of
2689 * scalar via different register has to be conservative.
2690 * Backtrack from here and mark all registers as precise
2691 * that contributed into 'reg' being a constant.
2693 err
= mark_chain_precision(env
, value_regno
);
2697 save_register_state(state
, spi
, reg
);
2698 } else if (reg
&& is_spillable_regtype(reg
->type
)) {
2699 /* register containing pointer is being spilled into stack */
2700 if (size
!= BPF_REG_SIZE
) {
2701 verbose_linfo(env
, insn_idx
, "; ");
2702 verbose(env
, "invalid size of register spill\n");
2705 if (state
!= cur
&& reg
->type
== PTR_TO_STACK
) {
2706 verbose(env
, "cannot spill pointers to stack into stack frame of the caller\n");
2709 save_register_state(state
, spi
, reg
);
2711 u8 type
= STACK_MISC
;
2713 /* regular write of data into stack destroys any spilled ptr */
2714 state
->stack
[spi
].spilled_ptr
.type
= NOT_INIT
;
2715 /* Mark slots as STACK_MISC if they belonged to spilled ptr. */
2716 if (state
->stack
[spi
].slot_type
[0] == STACK_SPILL
)
2717 for (i
= 0; i
< BPF_REG_SIZE
; i
++)
2718 state
->stack
[spi
].slot_type
[i
] = STACK_MISC
;
2720 /* only mark the slot as written if all 8 bytes were written
2721 * otherwise read propagation may incorrectly stop too soon
2722 * when stack slots are partially written.
2723 * This heuristic means that read propagation will be
2724 * conservative, since it will add reg_live_read marks
2725 * to stack slots all the way to first state when programs
2726 * writes+reads less than 8 bytes
2728 if (size
== BPF_REG_SIZE
)
2729 state
->stack
[spi
].spilled_ptr
.live
|= REG_LIVE_WRITTEN
;
2731 /* when we zero initialize stack slots mark them as such */
2732 if (reg
&& register_is_null(reg
)) {
2733 /* backtracking doesn't work for STACK_ZERO yet. */
2734 err
= mark_chain_precision(env
, value_regno
);
2740 /* Mark slots affected by this stack write. */
2741 for (i
= 0; i
< size
; i
++)
2742 state
->stack
[spi
].slot_type
[(slot
- i
) % BPF_REG_SIZE
] =
2748 /* Write the stack: 'stack[ptr_regno + off] = value_regno'. 'ptr_regno' is
2749 * known to contain a variable offset.
2750 * This function checks whether the write is permitted and conservatively
2751 * tracks the effects of the write, considering that each stack slot in the
2752 * dynamic range is potentially written to.
2754 * 'off' includes 'regno->off'.
2755 * 'value_regno' can be -1, meaning that an unknown value is being written to
2758 * Spilled pointers in range are not marked as written because we don't know
2759 * what's going to be actually written. This means that read propagation for
2760 * future reads cannot be terminated by this write.
2762 * For privileged programs, uninitialized stack slots are considered
2763 * initialized by this write (even though we don't know exactly what offsets
2764 * are going to be written to). The idea is that we don't want the verifier to
2765 * reject future reads that access slots written to through variable offsets.
2767 static int check_stack_write_var_off(struct bpf_verifier_env
*env
,
2768 /* func where register points to */
2769 struct bpf_func_state
*state
,
2770 int ptr_regno
, int off
, int size
,
2771 int value_regno
, int insn_idx
)
2773 struct bpf_func_state
*cur
; /* state of the current function */
2774 int min_off
, max_off
;
2776 struct bpf_reg_state
*ptr_reg
= NULL
, *value_reg
= NULL
;
2777 bool writing_zero
= false;
2778 /* set if the fact that we're writing a zero is used to let any
2779 * stack slots remain STACK_ZERO
2781 bool zero_used
= false;
2783 cur
= env
->cur_state
->frame
[env
->cur_state
->curframe
];
2784 ptr_reg
= &cur
->regs
[ptr_regno
];
2785 min_off
= ptr_reg
->smin_value
+ off
;
2786 max_off
= ptr_reg
->smax_value
+ off
+ size
;
2787 if (value_regno
>= 0)
2788 value_reg
= &cur
->regs
[value_regno
];
2789 if (value_reg
&& register_is_null(value_reg
))
2790 writing_zero
= true;
2792 err
= grow_stack_state(state
, round_up(-min_off
, BPF_REG_SIZE
));
2797 /* Variable offset writes destroy any spilled pointers in range. */
2798 for (i
= min_off
; i
< max_off
; i
++) {
2799 u8 new_type
, *stype
;
2803 spi
= slot
/ BPF_REG_SIZE
;
2804 stype
= &state
->stack
[spi
].slot_type
[slot
% BPF_REG_SIZE
];
2806 if (!env
->allow_ptr_leaks
2807 && *stype
!= NOT_INIT
2808 && *stype
!= SCALAR_VALUE
) {
2809 /* Reject the write if there's are spilled pointers in
2810 * range. If we didn't reject here, the ptr status
2811 * would be erased below (even though not all slots are
2812 * actually overwritten), possibly opening the door to
2815 verbose(env
, "spilled ptr in range of var-offset stack write; insn %d, ptr off: %d",
2820 /* Erase all spilled pointers. */
2821 state
->stack
[spi
].spilled_ptr
.type
= NOT_INIT
;
2823 /* Update the slot type. */
2824 new_type
= STACK_MISC
;
2825 if (writing_zero
&& *stype
== STACK_ZERO
) {
2826 new_type
= STACK_ZERO
;
2829 /* If the slot is STACK_INVALID, we check whether it's OK to
2830 * pretend that it will be initialized by this write. The slot
2831 * might not actually be written to, and so if we mark it as
2832 * initialized future reads might leak uninitialized memory.
2833 * For privileged programs, we will accept such reads to slots
2834 * that may or may not be written because, if we're reject
2835 * them, the error would be too confusing.
2837 if (*stype
== STACK_INVALID
&& !env
->allow_uninit_stack
) {
2838 verbose(env
, "uninit stack in range of var-offset write prohibited for !root; insn %d, off: %d",
2845 /* backtracking doesn't work for STACK_ZERO yet. */
2846 err
= mark_chain_precision(env
, value_regno
);
2853 /* When register 'dst_regno' is assigned some values from stack[min_off,
2854 * max_off), we set the register's type according to the types of the
2855 * respective stack slots. If all the stack values are known to be zeros, then
2856 * so is the destination reg. Otherwise, the register is considered to be
2857 * SCALAR. This function does not deal with register filling; the caller must
2858 * ensure that all spilled registers in the stack range have been marked as
2861 static void mark_reg_stack_read(struct bpf_verifier_env
*env
,
2862 /* func where src register points to */
2863 struct bpf_func_state
*ptr_state
,
2864 int min_off
, int max_off
, int dst_regno
)
2866 struct bpf_verifier_state
*vstate
= env
->cur_state
;
2867 struct bpf_func_state
*state
= vstate
->frame
[vstate
->curframe
];
2872 for (i
= min_off
; i
< max_off
; i
++) {
2874 spi
= slot
/ BPF_REG_SIZE
;
2875 stype
= ptr_state
->stack
[spi
].slot_type
;
2876 if (stype
[slot
% BPF_REG_SIZE
] != STACK_ZERO
)
2880 if (zeros
== max_off
- min_off
) {
2881 /* any access_size read into register is zero extended,
2882 * so the whole register == const_zero
2884 __mark_reg_const_zero(&state
->regs
[dst_regno
]);
2885 /* backtracking doesn't support STACK_ZERO yet,
2886 * so mark it precise here, so that later
2887 * backtracking can stop here.
2888 * Backtracking may not need this if this register
2889 * doesn't participate in pointer adjustment.
2890 * Forward propagation of precise flag is not
2891 * necessary either. This mark is only to stop
2892 * backtracking. Any register that contributed
2893 * to const 0 was marked precise before spill.
2895 state
->regs
[dst_regno
].precise
= true;
2897 /* have read misc data from the stack */
2898 mark_reg_unknown(env
, state
->regs
, dst_regno
);
2900 state
->regs
[dst_regno
].live
|= REG_LIVE_WRITTEN
;
2903 /* Read the stack at 'off' and put the results into the register indicated by
2904 * 'dst_regno'. It handles reg filling if the addressed stack slot is a
2907 * 'dst_regno' can be -1, meaning that the read value is not going to a
2910 * The access is assumed to be within the current stack bounds.
2912 static int check_stack_read_fixed_off(struct bpf_verifier_env
*env
,
2913 /* func where src register points to */
2914 struct bpf_func_state
*reg_state
,
2915 int off
, int size
, int dst_regno
)
2917 struct bpf_verifier_state
*vstate
= env
->cur_state
;
2918 struct bpf_func_state
*state
= vstate
->frame
[vstate
->curframe
];
2919 int i
, slot
= -off
- 1, spi
= slot
/ BPF_REG_SIZE
;
2920 struct bpf_reg_state
*reg
;
2923 stype
= reg_state
->stack
[spi
].slot_type
;
2924 reg
= ®_state
->stack
[spi
].spilled_ptr
;
2926 if (stype
[0] == STACK_SPILL
) {
2927 if (size
!= BPF_REG_SIZE
) {
2928 if (reg
->type
!= SCALAR_VALUE
) {
2929 verbose_linfo(env
, env
->insn_idx
, "; ");
2930 verbose(env
, "invalid size of register fill\n");
2933 if (dst_regno
>= 0) {
2934 mark_reg_unknown(env
, state
->regs
, dst_regno
);
2935 state
->regs
[dst_regno
].live
|= REG_LIVE_WRITTEN
;
2937 mark_reg_read(env
, reg
, reg
->parent
, REG_LIVE_READ64
);
2940 for (i
= 1; i
< BPF_REG_SIZE
; i
++) {
2941 if (stype
[(slot
- i
) % BPF_REG_SIZE
] != STACK_SPILL
) {
2942 verbose(env
, "corrupted spill memory\n");
2947 if (dst_regno
>= 0) {
2948 /* restore register state from stack */
2949 state
->regs
[dst_regno
] = *reg
;
2950 /* mark reg as written since spilled pointer state likely
2951 * has its liveness marks cleared by is_state_visited()
2952 * which resets stack/reg liveness for state transitions
2954 state
->regs
[dst_regno
].live
|= REG_LIVE_WRITTEN
;
2955 } else if (__is_pointer_value(env
->allow_ptr_leaks
, reg
)) {
2956 /* If dst_regno==-1, the caller is asking us whether
2957 * it is acceptable to use this value as a SCALAR_VALUE
2959 * We must not allow unprivileged callers to do that
2960 * with spilled pointers.
2962 verbose(env
, "leaking pointer from stack off %d\n",
2966 mark_reg_read(env
, reg
, reg
->parent
, REG_LIVE_READ64
);
2970 for (i
= 0; i
< size
; i
++) {
2971 type
= stype
[(slot
- i
) % BPF_REG_SIZE
];
2972 if (type
== STACK_MISC
)
2974 if (type
== STACK_ZERO
)
2976 verbose(env
, "invalid read from stack off %d+%d size %d\n",
2980 mark_reg_read(env
, reg
, reg
->parent
, REG_LIVE_READ64
);
2982 mark_reg_stack_read(env
, reg_state
, off
, off
+ size
, dst_regno
);
2987 enum stack_access_src
{
2988 ACCESS_DIRECT
= 1, /* the access is performed by an instruction */
2989 ACCESS_HELPER
= 2, /* the access is performed by a helper */
2992 static int check_stack_range_initialized(struct bpf_verifier_env
*env
,
2993 int regno
, int off
, int access_size
,
2994 bool zero_size_allowed
,
2995 enum stack_access_src type
,
2996 struct bpf_call_arg_meta
*meta
);
2998 static struct bpf_reg_state
*reg_state(struct bpf_verifier_env
*env
, int regno
)
3000 return cur_regs(env
) + regno
;
3003 /* Read the stack at 'ptr_regno + off' and put the result into the register
3005 * 'off' includes the pointer register's fixed offset(i.e. 'ptr_regno.off'),
3006 * but not its variable offset.
3007 * 'size' is assumed to be <= reg size and the access is assumed to be aligned.
3009 * As opposed to check_stack_read_fixed_off, this function doesn't deal with
3010 * filling registers (i.e. reads of spilled register cannot be detected when
3011 * the offset is not fixed). We conservatively mark 'dst_regno' as containing
3012 * SCALAR_VALUE. That's why we assert that the 'ptr_regno' has a variable
3013 * offset; for a fixed offset check_stack_read_fixed_off should be used
3016 static int check_stack_read_var_off(struct bpf_verifier_env
*env
,
3017 int ptr_regno
, int off
, int size
, int dst_regno
)
3019 /* The state of the source register. */
3020 struct bpf_reg_state
*reg
= reg_state(env
, ptr_regno
);
3021 struct bpf_func_state
*ptr_state
= func(env
, reg
);
3023 int min_off
, max_off
;
3025 /* Note that we pass a NULL meta, so raw access will not be permitted.
3027 err
= check_stack_range_initialized(env
, ptr_regno
, off
, size
,
3028 false, ACCESS_DIRECT
, NULL
);
3032 min_off
= reg
->smin_value
+ off
;
3033 max_off
= reg
->smax_value
+ off
;
3034 mark_reg_stack_read(env
, ptr_state
, min_off
, max_off
+ size
, dst_regno
);
3038 /* check_stack_read dispatches to check_stack_read_fixed_off or
3039 * check_stack_read_var_off.
3041 * The caller must ensure that the offset falls within the allocated stack
3044 * 'dst_regno' is a register which will receive the value from the stack. It
3045 * can be -1, meaning that the read value is not going to a register.
3047 static int check_stack_read(struct bpf_verifier_env
*env
,
3048 int ptr_regno
, int off
, int size
,
3051 struct bpf_reg_state
*reg
= reg_state(env
, ptr_regno
);
3052 struct bpf_func_state
*state
= func(env
, reg
);
3054 /* Some accesses are only permitted with a static offset. */
3055 bool var_off
= !tnum_is_const(reg
->var_off
);
3057 /* The offset is required to be static when reads don't go to a
3058 * register, in order to not leak pointers (see
3059 * check_stack_read_fixed_off).
3061 if (dst_regno
< 0 && var_off
) {
3064 tnum_strn(tn_buf
, sizeof(tn_buf
), reg
->var_off
);
3065 verbose(env
, "variable offset stack pointer cannot be passed into helper function; var_off=%s off=%d size=%d\n",
3069 /* Variable offset is prohibited for unprivileged mode for simplicity
3070 * since it requires corresponding support in Spectre masking for stack
3071 * ALU. See also retrieve_ptr_limit().
3073 if (!env
->bypass_spec_v1
&& var_off
) {
3076 tnum_strn(tn_buf
, sizeof(tn_buf
), reg
->var_off
);
3077 verbose(env
, "R%d variable offset stack access prohibited for !root, var_off=%s\n",
3083 off
+= reg
->var_off
.value
;
3084 err
= check_stack_read_fixed_off(env
, state
, off
, size
,
3087 /* Variable offset stack reads need more conservative handling
3088 * than fixed offset ones. Note that dst_regno >= 0 on this
3091 err
= check_stack_read_var_off(env
, ptr_regno
, off
, size
,
3098 /* check_stack_write dispatches to check_stack_write_fixed_off or
3099 * check_stack_write_var_off.
3101 * 'ptr_regno' is the register used as a pointer into the stack.
3102 * 'off' includes 'ptr_regno->off', but not its variable offset (if any).
3103 * 'value_regno' is the register whose value we're writing to the stack. It can
3104 * be -1, meaning that we're not writing from a register.
3106 * The caller must ensure that the offset falls within the maximum stack size.
3108 static int check_stack_write(struct bpf_verifier_env
*env
,
3109 int ptr_regno
, int off
, int size
,
3110 int value_regno
, int insn_idx
)
3112 struct bpf_reg_state
*reg
= reg_state(env
, ptr_regno
);
3113 struct bpf_func_state
*state
= func(env
, reg
);
3116 if (tnum_is_const(reg
->var_off
)) {
3117 off
+= reg
->var_off
.value
;
3118 err
= check_stack_write_fixed_off(env
, state
, off
, size
,
3119 value_regno
, insn_idx
);
3121 /* Variable offset stack reads need more conservative handling
3122 * than fixed offset ones.
3124 err
= check_stack_write_var_off(env
, state
,
3125 ptr_regno
, off
, size
,
3126 value_regno
, insn_idx
);
3131 static int check_map_access_type(struct bpf_verifier_env
*env
, u32 regno
,
3132 int off
, int size
, enum bpf_access_type type
)
3134 struct bpf_reg_state
*regs
= cur_regs(env
);
3135 struct bpf_map
*map
= regs
[regno
].map_ptr
;
3136 u32 cap
= bpf_map_flags_to_cap(map
);
3138 if (type
== BPF_WRITE
&& !(cap
& BPF_MAP_CAN_WRITE
)) {
3139 verbose(env
, "write into map forbidden, value_size=%d off=%d size=%d\n",
3140 map
->value_size
, off
, size
);
3144 if (type
== BPF_READ
&& !(cap
& BPF_MAP_CAN_READ
)) {
3145 verbose(env
, "read from map forbidden, value_size=%d off=%d size=%d\n",
3146 map
->value_size
, off
, size
);
3153 /* check read/write into memory region (e.g., map value, ringbuf sample, etc) */
3154 static int __check_mem_access(struct bpf_verifier_env
*env
, int regno
,
3155 int off
, int size
, u32 mem_size
,
3156 bool zero_size_allowed
)
3158 bool size_ok
= size
> 0 || (size
== 0 && zero_size_allowed
);
3159 struct bpf_reg_state
*reg
;
3161 if (off
>= 0 && size_ok
&& (u64
)off
+ size
<= mem_size
)
3164 reg
= &cur_regs(env
)[regno
];
3165 switch (reg
->type
) {
3166 case PTR_TO_MAP_KEY
:
3167 verbose(env
, "invalid access to map key, key_size=%d off=%d size=%d\n",
3168 mem_size
, off
, size
);
3170 case PTR_TO_MAP_VALUE
:
3171 verbose(env
, "invalid access to map value, value_size=%d off=%d size=%d\n",
3172 mem_size
, off
, size
);
3175 case PTR_TO_PACKET_META
:
3176 case PTR_TO_PACKET_END
:
3177 verbose(env
, "invalid access to packet, off=%d size=%d, R%d(id=%d,off=%d,r=%d)\n",
3178 off
, size
, regno
, reg
->id
, off
, mem_size
);
3182 verbose(env
, "invalid access to memory, mem_size=%u off=%d size=%d\n",
3183 mem_size
, off
, size
);
3189 /* check read/write into a memory region with possible variable offset */
3190 static int check_mem_region_access(struct bpf_verifier_env
*env
, u32 regno
,
3191 int off
, int size
, u32 mem_size
,
3192 bool zero_size_allowed
)
3194 struct bpf_verifier_state
*vstate
= env
->cur_state
;
3195 struct bpf_func_state
*state
= vstate
->frame
[vstate
->curframe
];
3196 struct bpf_reg_state
*reg
= &state
->regs
[regno
];
3199 /* We may have adjusted the register pointing to memory region, so we
3200 * need to try adding each of min_value and max_value to off
3201 * to make sure our theoretical access will be safe.
3203 if (env
->log
.level
& BPF_LOG_LEVEL
)
3204 print_verifier_state(env
, state
);
3206 /* The minimum value is only important with signed
3207 * comparisons where we can't assume the floor of a
3208 * value is 0. If we are using signed variables for our
3209 * index'es we need to make sure that whatever we use
3210 * will have a set floor within our range.
3212 if (reg
->smin_value
< 0 &&
3213 (reg
->smin_value
== S64_MIN
||
3214 (off
+ reg
->smin_value
!= (s64
)(s32
)(off
+ reg
->smin_value
)) ||
3215 reg
->smin_value
+ off
< 0)) {
3216 verbose(env
, "R%d min value is negative, either use unsigned index or do a if (index >=0) check.\n",
3220 err
= __check_mem_access(env
, regno
, reg
->smin_value
+ off
, size
,
3221 mem_size
, zero_size_allowed
);
3223 verbose(env
, "R%d min value is outside of the allowed memory range\n",
3228 /* If we haven't set a max value then we need to bail since we can't be
3229 * sure we won't do bad things.
3230 * If reg->umax_value + off could overflow, treat that as unbounded too.
3232 if (reg
->umax_value
>= BPF_MAX_VAR_OFF
) {
3233 verbose(env
, "R%d unbounded memory access, make sure to bounds check any such access\n",
3237 err
= __check_mem_access(env
, regno
, reg
->umax_value
+ off
, size
,
3238 mem_size
, zero_size_allowed
);
3240 verbose(env
, "R%d max value is outside of the allowed memory range\n",
3248 /* check read/write into a map element with possible variable offset */
3249 static int check_map_access(struct bpf_verifier_env
*env
, u32 regno
,
3250 int off
, int size
, bool zero_size_allowed
)
3252 struct bpf_verifier_state
*vstate
= env
->cur_state
;
3253 struct bpf_func_state
*state
= vstate
->frame
[vstate
->curframe
];
3254 struct bpf_reg_state
*reg
= &state
->regs
[regno
];
3255 struct bpf_map
*map
= reg
->map_ptr
;
3258 err
= check_mem_region_access(env
, regno
, off
, size
, map
->value_size
,
3263 if (map_value_has_spin_lock(map
)) {
3264 u32 lock
= map
->spin_lock_off
;
3266 /* if any part of struct bpf_spin_lock can be touched by
3267 * load/store reject this program.
3268 * To check that [x1, x2) overlaps with [y1, y2)
3269 * it is sufficient to check x1 < y2 && y1 < x2.
3271 if (reg
->smin_value
+ off
< lock
+ sizeof(struct bpf_spin_lock
) &&
3272 lock
< reg
->umax_value
+ off
+ size
) {
3273 verbose(env
, "bpf_spin_lock cannot be accessed directly by load/store\n");
3277 if (map_value_has_timer(map
)) {
3278 u32 t
= map
->timer_off
;
3280 if (reg
->smin_value
+ off
< t
+ sizeof(struct bpf_timer
) &&
3281 t
< reg
->umax_value
+ off
+ size
) {
3282 verbose(env
, "bpf_timer cannot be accessed directly by load/store\n");
3289 #define MAX_PACKET_OFF 0xffff
3291 static enum bpf_prog_type
resolve_prog_type(struct bpf_prog
*prog
)
3293 return prog
->aux
->dst_prog
? prog
->aux
->dst_prog
->type
: prog
->type
;
3296 static bool may_access_direct_pkt_data(struct bpf_verifier_env
*env
,
3297 const struct bpf_call_arg_meta
*meta
,
3298 enum bpf_access_type t
)
3300 enum bpf_prog_type prog_type
= resolve_prog_type(env
->prog
);
3302 switch (prog_type
) {
3303 /* Program types only with direct read access go here! */
3304 case BPF_PROG_TYPE_LWT_IN
:
3305 case BPF_PROG_TYPE_LWT_OUT
:
3306 case BPF_PROG_TYPE_LWT_SEG6LOCAL
:
3307 case BPF_PROG_TYPE_SK_REUSEPORT
:
3308 case BPF_PROG_TYPE_FLOW_DISSECTOR
:
3309 case BPF_PROG_TYPE_CGROUP_SKB
:
3314 /* Program types with direct read + write access go here! */
3315 case BPF_PROG_TYPE_SCHED_CLS
:
3316 case BPF_PROG_TYPE_SCHED_ACT
:
3317 case BPF_PROG_TYPE_XDP
:
3318 case BPF_PROG_TYPE_LWT_XMIT
:
3319 case BPF_PROG_TYPE_SK_SKB
:
3320 case BPF_PROG_TYPE_SK_MSG
:
3322 return meta
->pkt_access
;
3324 env
->seen_direct_write
= true;
3327 case BPF_PROG_TYPE_CGROUP_SOCKOPT
:
3329 env
->seen_direct_write
= true;
3338 static int check_packet_access(struct bpf_verifier_env
*env
, u32 regno
, int off
,
3339 int size
, bool zero_size_allowed
)
3341 struct bpf_reg_state
*regs
= cur_regs(env
);
3342 struct bpf_reg_state
*reg
= ®s
[regno
];
3345 /* We may have added a variable offset to the packet pointer; but any
3346 * reg->range we have comes after that. We are only checking the fixed
3350 /* We don't allow negative numbers, because we aren't tracking enough
3351 * detail to prove they're safe.
3353 if (reg
->smin_value
< 0) {
3354 verbose(env
, "R%d min value is negative, either use unsigned index or do a if (index >=0) check.\n",
3359 err
= reg
->range
< 0 ? -EINVAL
:
3360 __check_mem_access(env
, regno
, off
, size
, reg
->range
,
3363 verbose(env
, "R%d offset is outside of the packet\n", regno
);
3367 /* __check_mem_access has made sure "off + size - 1" is within u16.
3368 * reg->umax_value can't be bigger than MAX_PACKET_OFF which is 0xffff,
3369 * otherwise find_good_pkt_pointers would have refused to set range info
3370 * that __check_mem_access would have rejected this pkt access.
3371 * Therefore, "off + reg->umax_value + size - 1" won't overflow u32.
3373 env
->prog
->aux
->max_pkt_offset
=
3374 max_t(u32
, env
->prog
->aux
->max_pkt_offset
,
3375 off
+ reg
->umax_value
+ size
- 1);
3380 /* check access to 'struct bpf_context' fields. Supports fixed offsets only */
3381 static int check_ctx_access(struct bpf_verifier_env
*env
, int insn_idx
, int off
, int size
,
3382 enum bpf_access_type t
, enum bpf_reg_type
*reg_type
,
3383 struct btf
**btf
, u32
*btf_id
)
3385 struct bpf_insn_access_aux info
= {
3386 .reg_type
= *reg_type
,
3390 if (env
->ops
->is_valid_access
&&
3391 env
->ops
->is_valid_access(off
, size
, t
, env
->prog
, &info
)) {
3392 /* A non zero info.ctx_field_size indicates that this field is a
3393 * candidate for later verifier transformation to load the whole
3394 * field and then apply a mask when accessed with a narrower
3395 * access than actual ctx access size. A zero info.ctx_field_size
3396 * will only allow for whole field access and rejects any other
3397 * type of narrower access.
3399 *reg_type
= info
.reg_type
;
3401 if (*reg_type
== PTR_TO_BTF_ID
|| *reg_type
== PTR_TO_BTF_ID_OR_NULL
) {
3403 *btf_id
= info
.btf_id
;
3405 env
->insn_aux_data
[insn_idx
].ctx_field_size
= info
.ctx_field_size
;
3407 /* remember the offset of last byte accessed in ctx */
3408 if (env
->prog
->aux
->max_ctx_offset
< off
+ size
)
3409 env
->prog
->aux
->max_ctx_offset
= off
+ size
;
3413 verbose(env
, "invalid bpf_context access off=%d size=%d\n", off
, size
);
3417 static int check_flow_keys_access(struct bpf_verifier_env
*env
, int off
,
3420 if (size
< 0 || off
< 0 ||
3421 (u64
)off
+ size
> sizeof(struct bpf_flow_keys
)) {
3422 verbose(env
, "invalid access to flow keys off=%d size=%d\n",
3429 static int check_sock_access(struct bpf_verifier_env
*env
, int insn_idx
,
3430 u32 regno
, int off
, int size
,
3431 enum bpf_access_type t
)
3433 struct bpf_reg_state
*regs
= cur_regs(env
);
3434 struct bpf_reg_state
*reg
= ®s
[regno
];
3435 struct bpf_insn_access_aux info
= {};
3438 if (reg
->smin_value
< 0) {
3439 verbose(env
, "R%d min value is negative, either use unsigned index or do a if (index >=0) check.\n",
3444 switch (reg
->type
) {
3445 case PTR_TO_SOCK_COMMON
:
3446 valid
= bpf_sock_common_is_valid_access(off
, size
, t
, &info
);
3449 valid
= bpf_sock_is_valid_access(off
, size
, t
, &info
);
3451 case PTR_TO_TCP_SOCK
:
3452 valid
= bpf_tcp_sock_is_valid_access(off
, size
, t
, &info
);
3454 case PTR_TO_XDP_SOCK
:
3455 valid
= bpf_xdp_sock_is_valid_access(off
, size
, t
, &info
);
3463 env
->insn_aux_data
[insn_idx
].ctx_field_size
=
3464 info
.ctx_field_size
;
3468 verbose(env
, "R%d invalid %s access off=%d size=%d\n",
3469 regno
, reg_type_str
[reg
->type
], off
, size
);
3474 static bool is_pointer_value(struct bpf_verifier_env
*env
, int regno
)
3476 return __is_pointer_value(env
->allow_ptr_leaks
, reg_state(env
, regno
));
3479 static bool is_ctx_reg(struct bpf_verifier_env
*env
, int regno
)
3481 const struct bpf_reg_state
*reg
= reg_state(env
, regno
);
3483 return reg
->type
== PTR_TO_CTX
;
3486 static bool is_sk_reg(struct bpf_verifier_env
*env
, int regno
)
3488 const struct bpf_reg_state
*reg
= reg_state(env
, regno
);
3490 return type_is_sk_pointer(reg
->type
);
3493 static bool is_pkt_reg(struct bpf_verifier_env
*env
, int regno
)
3495 const struct bpf_reg_state
*reg
= reg_state(env
, regno
);
3497 return type_is_pkt_pointer(reg
->type
);
3500 static bool is_flow_key_reg(struct bpf_verifier_env
*env
, int regno
)
3502 const struct bpf_reg_state
*reg
= reg_state(env
, regno
);
3504 /* Separate to is_ctx_reg() since we still want to allow BPF_ST here. */
3505 return reg
->type
== PTR_TO_FLOW_KEYS
;
3508 static int check_pkt_ptr_alignment(struct bpf_verifier_env
*env
,
3509 const struct bpf_reg_state
*reg
,
3510 int off
, int size
, bool strict
)
3512 struct tnum reg_off
;
3515 /* Byte size accesses are always allowed. */
3516 if (!strict
|| size
== 1)
3519 /* For platforms that do not have a Kconfig enabling
3520 * CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS the value of
3521 * NET_IP_ALIGN is universally set to '2'. And on platforms
3522 * that do set CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS, we get
3523 * to this code only in strict mode where we want to emulate
3524 * the NET_IP_ALIGN==2 checking. Therefore use an
3525 * unconditional IP align value of '2'.
3529 reg_off
= tnum_add(reg
->var_off
, tnum_const(ip_align
+ reg
->off
+ off
));
3530 if (!tnum_is_aligned(reg_off
, size
)) {
3533 tnum_strn(tn_buf
, sizeof(tn_buf
), reg
->var_off
);
3535 "misaligned packet access off %d+%s+%d+%d size %d\n",
3536 ip_align
, tn_buf
, reg
->off
, off
, size
);
3543 static int check_generic_ptr_alignment(struct bpf_verifier_env
*env
,
3544 const struct bpf_reg_state
*reg
,
3545 const char *pointer_desc
,
3546 int off
, int size
, bool strict
)
3548 struct tnum reg_off
;
3550 /* Byte size accesses are always allowed. */
3551 if (!strict
|| size
== 1)
3554 reg_off
= tnum_add(reg
->var_off
, tnum_const(reg
->off
+ off
));
3555 if (!tnum_is_aligned(reg_off
, size
)) {
3558 tnum_strn(tn_buf
, sizeof(tn_buf
), reg
->var_off
);
3559 verbose(env
, "misaligned %saccess off %s+%d+%d size %d\n",
3560 pointer_desc
, tn_buf
, reg
->off
, off
, size
);
3567 static int check_ptr_alignment(struct bpf_verifier_env
*env
,
3568 const struct bpf_reg_state
*reg
, int off
,
3569 int size
, bool strict_alignment_once
)
3571 bool strict
= env
->strict_alignment
|| strict_alignment_once
;
3572 const char *pointer_desc
= "";
3574 switch (reg
->type
) {
3576 case PTR_TO_PACKET_META
:
3577 /* Special case, because of NET_IP_ALIGN. Given metadata sits
3578 * right in front, treat it the very same way.
3580 return check_pkt_ptr_alignment(env
, reg
, off
, size
, strict
);
3581 case PTR_TO_FLOW_KEYS
:
3582 pointer_desc
= "flow keys ";
3584 case PTR_TO_MAP_KEY
:
3585 pointer_desc
= "key ";
3587 case PTR_TO_MAP_VALUE
:
3588 pointer_desc
= "value ";
3591 pointer_desc
= "context ";
3594 pointer_desc
= "stack ";
3595 /* The stack spill tracking logic in check_stack_write_fixed_off()
3596 * and check_stack_read_fixed_off() relies on stack accesses being
3602 pointer_desc
= "sock ";
3604 case PTR_TO_SOCK_COMMON
:
3605 pointer_desc
= "sock_common ";
3607 case PTR_TO_TCP_SOCK
:
3608 pointer_desc
= "tcp_sock ";
3610 case PTR_TO_XDP_SOCK
:
3611 pointer_desc
= "xdp_sock ";
3616 return check_generic_ptr_alignment(env
, reg
, pointer_desc
, off
, size
,
3620 static int update_stack_depth(struct bpf_verifier_env
*env
,
3621 const struct bpf_func_state
*func
,
3624 u16 stack
= env
->subprog_info
[func
->subprogno
].stack_depth
;
3629 /* update known max for given subprogram */
3630 env
->subprog_info
[func
->subprogno
].stack_depth
= -off
;
3634 /* starting from main bpf function walk all instructions of the function
3635 * and recursively walk all callees that given function can call.
3636 * Ignore jump and exit insns.
3637 * Since recursion is prevented by check_cfg() this algorithm
3638 * only needs a local stack of MAX_CALL_FRAMES to remember callsites
3640 static int check_max_stack_depth(struct bpf_verifier_env
*env
)
3642 int depth
= 0, frame
= 0, idx
= 0, i
= 0, subprog_end
;
3643 struct bpf_subprog_info
*subprog
= env
->subprog_info
;
3644 struct bpf_insn
*insn
= env
->prog
->insnsi
;
3645 bool tail_call_reachable
= false;
3646 int ret_insn
[MAX_CALL_FRAMES
];
3647 int ret_prog
[MAX_CALL_FRAMES
];
3651 /* protect against potential stack overflow that might happen when
3652 * bpf2bpf calls get combined with tailcalls. Limit the caller's stack
3653 * depth for such case down to 256 so that the worst case scenario
3654 * would result in 8k stack size (32 which is tailcall limit * 256 =
3657 * To get the idea what might happen, see an example:
3658 * func1 -> sub rsp, 128
3659 * subfunc1 -> sub rsp, 256
3660 * tailcall1 -> add rsp, 256
3661 * func2 -> sub rsp, 192 (total stack size = 128 + 192 = 320)
3662 * subfunc2 -> sub rsp, 64
3663 * subfunc22 -> sub rsp, 128
3664 * tailcall2 -> add rsp, 128
3665 * func3 -> sub rsp, 32 (total stack size 128 + 192 + 64 + 32 = 416)
3667 * tailcall will unwind the current stack frame but it will not get rid
3668 * of caller's stack as shown on the example above.
3670 if (idx
&& subprog
[idx
].has_tail_call
&& depth
>= 256) {
3672 "tail_calls are not allowed when call stack of previous frames is %d bytes. Too large\n",
3676 /* round up to 32-bytes, since this is granularity
3677 * of interpreter stack size
3679 depth
+= round_up(max_t(u32
, subprog
[idx
].stack_depth
, 1), 32);
3680 if (depth
> MAX_BPF_STACK
) {
3681 verbose(env
, "combined stack size of %d calls is %d. Too large\n",
3686 subprog_end
= subprog
[idx
+ 1].start
;
3687 for (; i
< subprog_end
; i
++) {
3690 if (!bpf_pseudo_call(insn
+ i
) && !bpf_pseudo_func(insn
+ i
))
3692 /* remember insn and function to return to */
3693 ret_insn
[frame
] = i
+ 1;
3694 ret_prog
[frame
] = idx
;
3696 /* find the callee */
3697 next_insn
= i
+ insn
[i
].imm
+ 1;
3698 idx
= find_subprog(env
, next_insn
);
3700 WARN_ONCE(1, "verifier bug. No program starts at insn %d\n",
3704 if (subprog
[idx
].is_async_cb
) {
3705 if (subprog
[idx
].has_tail_call
) {
3706 verbose(env
, "verifier bug. subprog has tail_call and async cb\n");
3709 /* async callbacks don't increase bpf prog stack size */
3714 if (subprog
[idx
].has_tail_call
)
3715 tail_call_reachable
= true;
3718 if (frame
>= MAX_CALL_FRAMES
) {
3719 verbose(env
, "the call stack of %d frames is too deep !\n",
3725 /* if tail call got detected across bpf2bpf calls then mark each of the
3726 * currently present subprog frames as tail call reachable subprogs;
3727 * this info will be utilized by JIT so that we will be preserving the
3728 * tail call counter throughout bpf2bpf calls combined with tailcalls
3730 if (tail_call_reachable
)
3731 for (j
= 0; j
< frame
; j
++)
3732 subprog
[ret_prog
[j
]].tail_call_reachable
= true;
3733 if (subprog
[0].tail_call_reachable
)
3734 env
->prog
->aux
->tail_call_reachable
= true;
3736 /* end of for() loop means the last insn of the 'subprog'
3737 * was reached. Doesn't matter whether it was JA or EXIT
3741 depth
-= round_up(max_t(u32
, subprog
[idx
].stack_depth
, 1), 32);
3743 i
= ret_insn
[frame
];
3744 idx
= ret_prog
[frame
];
3748 #ifndef CONFIG_BPF_JIT_ALWAYS_ON
3749 static int get_callee_stack_depth(struct bpf_verifier_env
*env
,
3750 const struct bpf_insn
*insn
, int idx
)
3752 int start
= idx
+ insn
->imm
+ 1, subprog
;
3754 subprog
= find_subprog(env
, start
);
3756 WARN_ONCE(1, "verifier bug. No program starts at insn %d\n",
3760 return env
->subprog_info
[subprog
].stack_depth
;
3764 int check_ctx_reg(struct bpf_verifier_env
*env
,
3765 const struct bpf_reg_state
*reg
, int regno
)
3767 /* Access to ctx or passing it to a helper is only allowed in
3768 * its original, unmodified form.
3772 verbose(env
, "dereference of modified ctx ptr R%d off=%d disallowed\n",
3777 if (!tnum_is_const(reg
->var_off
) || reg
->var_off
.value
) {
3780 tnum_strn(tn_buf
, sizeof(tn_buf
), reg
->var_off
);
3781 verbose(env
, "variable ctx access var_off=%s disallowed\n", tn_buf
);
3788 static int __check_buffer_access(struct bpf_verifier_env
*env
,
3789 const char *buf_info
,
3790 const struct bpf_reg_state
*reg
,
3791 int regno
, int off
, int size
)
3795 "R%d invalid %s buffer access: off=%d, size=%d\n",
3796 regno
, buf_info
, off
, size
);
3799 if (!tnum_is_const(reg
->var_off
) || reg
->var_off
.value
) {
3802 tnum_strn(tn_buf
, sizeof(tn_buf
), reg
->var_off
);
3804 "R%d invalid variable buffer offset: off=%d, var_off=%s\n",
3805 regno
, off
, tn_buf
);
3812 static int check_tp_buffer_access(struct bpf_verifier_env
*env
,
3813 const struct bpf_reg_state
*reg
,
3814 int regno
, int off
, int size
)
3818 err
= __check_buffer_access(env
, "tracepoint", reg
, regno
, off
, size
);
3822 if (off
+ size
> env
->prog
->aux
->max_tp_access
)
3823 env
->prog
->aux
->max_tp_access
= off
+ size
;
3828 static int check_buffer_access(struct bpf_verifier_env
*env
,
3829 const struct bpf_reg_state
*reg
,
3830 int regno
, int off
, int size
,
3831 bool zero_size_allowed
,
3832 const char *buf_info
,
3837 err
= __check_buffer_access(env
, buf_info
, reg
, regno
, off
, size
);
3841 if (off
+ size
> *max_access
)
3842 *max_access
= off
+ size
;
3847 /* BPF architecture zero extends alu32 ops into 64-bit registesr */
3848 static void zext_32_to_64(struct bpf_reg_state
*reg
)
3850 reg
->var_off
= tnum_subreg(reg
->var_off
);
3851 __reg_assign_32_into_64(reg
);
3854 /* truncate register to smaller size (in bytes)
3855 * must be called with size < BPF_REG_SIZE
3857 static void coerce_reg_to_size(struct bpf_reg_state
*reg
, int size
)
3861 /* clear high bits in bit representation */
3862 reg
->var_off
= tnum_cast(reg
->var_off
, size
);
3864 /* fix arithmetic bounds */
3865 mask
= ((u64
)1 << (size
* 8)) - 1;
3866 if ((reg
->umin_value
& ~mask
) == (reg
->umax_value
& ~mask
)) {
3867 reg
->umin_value
&= mask
;
3868 reg
->umax_value
&= mask
;
3870 reg
->umin_value
= 0;
3871 reg
->umax_value
= mask
;
3873 reg
->smin_value
= reg
->umin_value
;
3874 reg
->smax_value
= reg
->umax_value
;
3876 /* If size is smaller than 32bit register the 32bit register
3877 * values are also truncated so we push 64-bit bounds into
3878 * 32-bit bounds. Above were truncated < 32-bits already.
3882 __reg_combine_64_into_32(reg
);
3885 static bool bpf_map_is_rdonly(const struct bpf_map
*map
)
3887 return (map
->map_flags
& BPF_F_RDONLY_PROG
) && map
->frozen
;
3890 static int bpf_map_direct_read(struct bpf_map
*map
, int off
, int size
, u64
*val
)
3896 err
= map
->ops
->map_direct_value_addr(map
, &addr
, off
);
3899 ptr
= (void *)(long)addr
+ off
;
3903 *val
= (u64
)*(u8
*)ptr
;
3906 *val
= (u64
)*(u16
*)ptr
;
3909 *val
= (u64
)*(u32
*)ptr
;
3920 static int check_ptr_to_btf_access(struct bpf_verifier_env
*env
,
3921 struct bpf_reg_state
*regs
,
3922 int regno
, int off
, int size
,
3923 enum bpf_access_type atype
,
3926 struct bpf_reg_state
*reg
= regs
+ regno
;
3927 const struct btf_type
*t
= btf_type_by_id(reg
->btf
, reg
->btf_id
);
3928 const char *tname
= btf_name_by_offset(reg
->btf
, t
->name_off
);
3934 "R%d is ptr_%s invalid negative access: off=%d\n",
3938 if (!tnum_is_const(reg
->var_off
) || reg
->var_off
.value
) {
3941 tnum_strn(tn_buf
, sizeof(tn_buf
), reg
->var_off
);
3943 "R%d is ptr_%s invalid variable offset: off=%d, var_off=%s\n",
3944 regno
, tname
, off
, tn_buf
);
3948 if (env
->ops
->btf_struct_access
) {
3949 ret
= env
->ops
->btf_struct_access(&env
->log
, reg
->btf
, t
,
3950 off
, size
, atype
, &btf_id
);
3952 if (atype
!= BPF_READ
) {
3953 verbose(env
, "only read is supported\n");
3957 ret
= btf_struct_access(&env
->log
, reg
->btf
, t
, off
, size
,
3964 if (atype
== BPF_READ
&& value_regno
>= 0)
3965 mark_btf_ld_reg(env
, regs
, value_regno
, ret
, reg
->btf
, btf_id
);
3970 static int check_ptr_to_map_access(struct bpf_verifier_env
*env
,
3971 struct bpf_reg_state
*regs
,
3972 int regno
, int off
, int size
,
3973 enum bpf_access_type atype
,
3976 struct bpf_reg_state
*reg
= regs
+ regno
;
3977 struct bpf_map
*map
= reg
->map_ptr
;
3978 const struct btf_type
*t
;
3984 verbose(env
, "map_ptr access not supported without CONFIG_DEBUG_INFO_BTF\n");
3988 if (!map
->ops
->map_btf_id
|| !*map
->ops
->map_btf_id
) {
3989 verbose(env
, "map_ptr access not supported for map type %d\n",
3994 t
= btf_type_by_id(btf_vmlinux
, *map
->ops
->map_btf_id
);
3995 tname
= btf_name_by_offset(btf_vmlinux
, t
->name_off
);
3997 if (!env
->allow_ptr_to_map_access
) {
3999 "%s access is allowed only to CAP_PERFMON and CAP_SYS_ADMIN\n",
4005 verbose(env
, "R%d is %s invalid negative access: off=%d\n",
4010 if (atype
!= BPF_READ
) {
4011 verbose(env
, "only read from %s is supported\n", tname
);
4015 ret
= btf_struct_access(&env
->log
, btf_vmlinux
, t
, off
, size
, atype
, &btf_id
);
4019 if (value_regno
>= 0)
4020 mark_btf_ld_reg(env
, regs
, value_regno
, ret
, btf_vmlinux
, btf_id
);
4025 /* Check that the stack access at the given offset is within bounds. The
4026 * maximum valid offset is -1.
4028 * The minimum valid offset is -MAX_BPF_STACK for writes, and
4029 * -state->allocated_stack for reads.
4031 static int check_stack_slot_within_bounds(int off
,
4032 struct bpf_func_state
*state
,
4033 enum bpf_access_type t
)
4038 min_valid_off
= -MAX_BPF_STACK
;
4040 min_valid_off
= -state
->allocated_stack
;
4042 if (off
< min_valid_off
|| off
> -1)
4047 /* Check that the stack access at 'regno + off' falls within the maximum stack
4050 * 'off' includes `regno->offset`, but not its dynamic part (if any).
4052 static int check_stack_access_within_bounds(
4053 struct bpf_verifier_env
*env
,
4054 int regno
, int off
, int access_size
,
4055 enum stack_access_src src
, enum bpf_access_type type
)
4057 struct bpf_reg_state
*regs
= cur_regs(env
);
4058 struct bpf_reg_state
*reg
= regs
+ regno
;
4059 struct bpf_func_state
*state
= func(env
, reg
);
4060 int min_off
, max_off
;
4064 if (src
== ACCESS_HELPER
)
4065 /* We don't know if helpers are reading or writing (or both). */
4066 err_extra
= " indirect access to";
4067 else if (type
== BPF_READ
)
4068 err_extra
= " read from";
4070 err_extra
= " write to";
4072 if (tnum_is_const(reg
->var_off
)) {
4073 min_off
= reg
->var_off
.value
+ off
;
4074 if (access_size
> 0)
4075 max_off
= min_off
+ access_size
- 1;
4079 if (reg
->smax_value
>= BPF_MAX_VAR_OFF
||
4080 reg
->smin_value
<= -BPF_MAX_VAR_OFF
) {
4081 verbose(env
, "invalid unbounded variable-offset%s stack R%d\n",
4085 min_off
= reg
->smin_value
+ off
;
4086 if (access_size
> 0)
4087 max_off
= reg
->smax_value
+ off
+ access_size
- 1;
4092 err
= check_stack_slot_within_bounds(min_off
, state
, type
);
4094 err
= check_stack_slot_within_bounds(max_off
, state
, type
);
4097 if (tnum_is_const(reg
->var_off
)) {
4098 verbose(env
, "invalid%s stack R%d off=%d size=%d\n",
4099 err_extra
, regno
, off
, access_size
);
4103 tnum_strn(tn_buf
, sizeof(tn_buf
), reg
->var_off
);
4104 verbose(env
, "invalid variable-offset%s stack R%d var_off=%s size=%d\n",
4105 err_extra
, regno
, tn_buf
, access_size
);
4111 /* check whether memory at (regno + off) is accessible for t = (read | write)
4112 * if t==write, value_regno is a register which value is stored into memory
4113 * if t==read, value_regno is a register which will receive the value from memory
4114 * if t==write && value_regno==-1, some unknown value is stored into memory
4115 * if t==read && value_regno==-1, don't care what we read from memory
4117 static int check_mem_access(struct bpf_verifier_env
*env
, int insn_idx
, u32 regno
,
4118 int off
, int bpf_size
, enum bpf_access_type t
,
4119 int value_regno
, bool strict_alignment_once
)
4121 struct bpf_reg_state
*regs
= cur_regs(env
);
4122 struct bpf_reg_state
*reg
= regs
+ regno
;
4123 struct bpf_func_state
*state
;
4126 size
= bpf_size_to_bytes(bpf_size
);
4130 /* alignment checks will add in reg->off themselves */
4131 err
= check_ptr_alignment(env
, reg
, off
, size
, strict_alignment_once
);
4135 /* for access checks, reg->off is just part of off */
4138 if (reg
->type
== PTR_TO_MAP_KEY
) {
4139 if (t
== BPF_WRITE
) {
4140 verbose(env
, "write to change key R%d not allowed\n", regno
);
4144 err
= check_mem_region_access(env
, regno
, off
, size
,
4145 reg
->map_ptr
->key_size
, false);
4148 if (value_regno
>= 0)
4149 mark_reg_unknown(env
, regs
, value_regno
);
4150 } else if (reg
->type
== PTR_TO_MAP_VALUE
) {
4151 if (t
== BPF_WRITE
&& value_regno
>= 0 &&
4152 is_pointer_value(env
, value_regno
)) {
4153 verbose(env
, "R%d leaks addr into map\n", value_regno
);
4156 err
= check_map_access_type(env
, regno
, off
, size
, t
);
4159 err
= check_map_access(env
, regno
, off
, size
, false);
4160 if (!err
&& t
== BPF_READ
&& value_regno
>= 0) {
4161 struct bpf_map
*map
= reg
->map_ptr
;
4163 /* if map is read-only, track its contents as scalars */
4164 if (tnum_is_const(reg
->var_off
) &&
4165 bpf_map_is_rdonly(map
) &&
4166 map
->ops
->map_direct_value_addr
) {
4167 int map_off
= off
+ reg
->var_off
.value
;
4170 err
= bpf_map_direct_read(map
, map_off
, size
,
4175 regs
[value_regno
].type
= SCALAR_VALUE
;
4176 __mark_reg_known(®s
[value_regno
], val
);
4178 mark_reg_unknown(env
, regs
, value_regno
);
4181 } else if (reg
->type
== PTR_TO_MEM
) {
4182 if (t
== BPF_WRITE
&& value_regno
>= 0 &&
4183 is_pointer_value(env
, value_regno
)) {
4184 verbose(env
, "R%d leaks addr into mem\n", value_regno
);
4187 err
= check_mem_region_access(env
, regno
, off
, size
,
4188 reg
->mem_size
, false);
4189 if (!err
&& t
== BPF_READ
&& value_regno
>= 0)
4190 mark_reg_unknown(env
, regs
, value_regno
);
4191 } else if (reg
->type
== PTR_TO_CTX
) {
4192 enum bpf_reg_type reg_type
= SCALAR_VALUE
;
4193 struct btf
*btf
= NULL
;
4196 if (t
== BPF_WRITE
&& value_regno
>= 0 &&
4197 is_pointer_value(env
, value_regno
)) {
4198 verbose(env
, "R%d leaks addr into ctx\n", value_regno
);
4202 err
= check_ctx_reg(env
, reg
, regno
);
4206 err
= check_ctx_access(env
, insn_idx
, off
, size
, t
, ®_type
, &btf
, &btf_id
);
4208 verbose_linfo(env
, insn_idx
, "; ");
4209 if (!err
&& t
== BPF_READ
&& value_regno
>= 0) {
4210 /* ctx access returns either a scalar, or a
4211 * PTR_TO_PACKET[_META,_END]. In the latter
4212 * case, we know the offset is zero.
4214 if (reg_type
== SCALAR_VALUE
) {
4215 mark_reg_unknown(env
, regs
, value_regno
);
4217 mark_reg_known_zero(env
, regs
,
4219 if (reg_type_may_be_null(reg_type
))
4220 regs
[value_regno
].id
= ++env
->id_gen
;
4221 /* A load of ctx field could have different
4222 * actual load size with the one encoded in the
4223 * insn. When the dst is PTR, it is for sure not
4226 regs
[value_regno
].subreg_def
= DEF_NOT_SUBREG
;
4227 if (reg_type
== PTR_TO_BTF_ID
||
4228 reg_type
== PTR_TO_BTF_ID_OR_NULL
) {
4229 regs
[value_regno
].btf
= btf
;
4230 regs
[value_regno
].btf_id
= btf_id
;
4233 regs
[value_regno
].type
= reg_type
;
4236 } else if (reg
->type
== PTR_TO_STACK
) {
4237 /* Basic bounds checks. */
4238 err
= check_stack_access_within_bounds(env
, regno
, off
, size
, ACCESS_DIRECT
, t
);
4242 state
= func(env
, reg
);
4243 err
= update_stack_depth(env
, state
, off
);
4248 err
= check_stack_read(env
, regno
, off
, size
,
4251 err
= check_stack_write(env
, regno
, off
, size
,
4252 value_regno
, insn_idx
);
4253 } else if (reg_is_pkt_pointer(reg
)) {
4254 if (t
== BPF_WRITE
&& !may_access_direct_pkt_data(env
, NULL
, t
)) {
4255 verbose(env
, "cannot write into packet\n");
4258 if (t
== BPF_WRITE
&& value_regno
>= 0 &&
4259 is_pointer_value(env
, value_regno
)) {
4260 verbose(env
, "R%d leaks addr into packet\n",
4264 err
= check_packet_access(env
, regno
, off
, size
, false);
4265 if (!err
&& t
== BPF_READ
&& value_regno
>= 0)
4266 mark_reg_unknown(env
, regs
, value_regno
);
4267 } else if (reg
->type
== PTR_TO_FLOW_KEYS
) {
4268 if (t
== BPF_WRITE
&& value_regno
>= 0 &&
4269 is_pointer_value(env
, value_regno
)) {
4270 verbose(env
, "R%d leaks addr into flow keys\n",
4275 err
= check_flow_keys_access(env
, off
, size
);
4276 if (!err
&& t
== BPF_READ
&& value_regno
>= 0)
4277 mark_reg_unknown(env
, regs
, value_regno
);
4278 } else if (type_is_sk_pointer(reg
->type
)) {
4279 if (t
== BPF_WRITE
) {
4280 verbose(env
, "R%d cannot write into %s\n",
4281 regno
, reg_type_str
[reg
->type
]);
4284 err
= check_sock_access(env
, insn_idx
, regno
, off
, size
, t
);
4285 if (!err
&& value_regno
>= 0)
4286 mark_reg_unknown(env
, regs
, value_regno
);
4287 } else if (reg
->type
== PTR_TO_TP_BUFFER
) {
4288 err
= check_tp_buffer_access(env
, reg
, regno
, off
, size
);
4289 if (!err
&& t
== BPF_READ
&& value_regno
>= 0)
4290 mark_reg_unknown(env
, regs
, value_regno
);
4291 } else if (reg
->type
== PTR_TO_BTF_ID
) {
4292 err
= check_ptr_to_btf_access(env
, regs
, regno
, off
, size
, t
,
4294 } else if (reg
->type
== CONST_PTR_TO_MAP
) {
4295 err
= check_ptr_to_map_access(env
, regs
, regno
, off
, size
, t
,
4297 } else if (reg
->type
== PTR_TO_RDONLY_BUF
) {
4298 if (t
== BPF_WRITE
) {
4299 verbose(env
, "R%d cannot write into %s\n",
4300 regno
, reg_type_str
[reg
->type
]);
4303 err
= check_buffer_access(env
, reg
, regno
, off
, size
, false,
4305 &env
->prog
->aux
->max_rdonly_access
);
4306 if (!err
&& value_regno
>= 0)
4307 mark_reg_unknown(env
, regs
, value_regno
);
4308 } else if (reg
->type
== PTR_TO_RDWR_BUF
) {
4309 err
= check_buffer_access(env
, reg
, regno
, off
, size
, false,
4311 &env
->prog
->aux
->max_rdwr_access
);
4312 if (!err
&& t
== BPF_READ
&& value_regno
>= 0)
4313 mark_reg_unknown(env
, regs
, value_regno
);
4315 verbose(env
, "R%d invalid mem access '%s'\n", regno
,
4316 reg_type_str
[reg
->type
]);
4320 if (!err
&& size
< BPF_REG_SIZE
&& value_regno
>= 0 && t
== BPF_READ
&&
4321 regs
[value_regno
].type
== SCALAR_VALUE
) {
4322 /* b/h/w load zero-extends, mark upper bits as known 0 */
4323 coerce_reg_to_size(®s
[value_regno
], size
);
4328 static int check_atomic(struct bpf_verifier_env
*env
, int insn_idx
, struct bpf_insn
*insn
)
4333 switch (insn
->imm
) {
4335 case BPF_ADD
| BPF_FETCH
:
4337 case BPF_AND
| BPF_FETCH
:
4339 case BPF_OR
| BPF_FETCH
:
4341 case BPF_XOR
| BPF_FETCH
:
4346 verbose(env
, "BPF_ATOMIC uses invalid atomic opcode %02x\n", insn
->imm
);
4350 if (BPF_SIZE(insn
->code
) != BPF_W
&& BPF_SIZE(insn
->code
) != BPF_DW
) {
4351 verbose(env
, "invalid atomic operand size\n");
4355 /* check src1 operand */
4356 err
= check_reg_arg(env
, insn
->src_reg
, SRC_OP
);
4360 /* check src2 operand */
4361 err
= check_reg_arg(env
, insn
->dst_reg
, SRC_OP
);
4365 if (insn
->imm
== BPF_CMPXCHG
) {
4366 /* Check comparison of R0 with memory location */
4367 err
= check_reg_arg(env
, BPF_REG_0
, SRC_OP
);
4372 if (is_pointer_value(env
, insn
->src_reg
)) {
4373 verbose(env
, "R%d leaks addr into mem\n", insn
->src_reg
);
4377 if (is_ctx_reg(env
, insn
->dst_reg
) ||
4378 is_pkt_reg(env
, insn
->dst_reg
) ||
4379 is_flow_key_reg(env
, insn
->dst_reg
) ||
4380 is_sk_reg(env
, insn
->dst_reg
)) {
4381 verbose(env
, "BPF_ATOMIC stores into R%d %s is not allowed\n",
4383 reg_type_str
[reg_state(env
, insn
->dst_reg
)->type
]);
4387 if (insn
->imm
& BPF_FETCH
) {
4388 if (insn
->imm
== BPF_CMPXCHG
)
4389 load_reg
= BPF_REG_0
;
4391 load_reg
= insn
->src_reg
;
4393 /* check and record load of old value */
4394 err
= check_reg_arg(env
, load_reg
, DST_OP
);
4398 /* This instruction accesses a memory location but doesn't
4399 * actually load it into a register.
4404 /* check whether we can read the memory */
4405 err
= check_mem_access(env
, insn_idx
, insn
->dst_reg
, insn
->off
,
4406 BPF_SIZE(insn
->code
), BPF_READ
, load_reg
, true);
4410 /* check whether we can write into the same memory */
4411 err
= check_mem_access(env
, insn_idx
, insn
->dst_reg
, insn
->off
,
4412 BPF_SIZE(insn
->code
), BPF_WRITE
, -1, true);
4419 /* When register 'regno' is used to read the stack (either directly or through
4420 * a helper function) make sure that it's within stack boundary and, depending
4421 * on the access type, that all elements of the stack are initialized.
4423 * 'off' includes 'regno->off', but not its dynamic part (if any).
4425 * All registers that have been spilled on the stack in the slots within the
4426 * read offsets are marked as read.
4428 static int check_stack_range_initialized(
4429 struct bpf_verifier_env
*env
, int regno
, int off
,
4430 int access_size
, bool zero_size_allowed
,
4431 enum stack_access_src type
, struct bpf_call_arg_meta
*meta
)
4433 struct bpf_reg_state
*reg
= reg_state(env
, regno
);
4434 struct bpf_func_state
*state
= func(env
, reg
);
4435 int err
, min_off
, max_off
, i
, j
, slot
, spi
;
4436 char *err_extra
= type
== ACCESS_HELPER
? " indirect" : "";
4437 enum bpf_access_type bounds_check_type
;
4438 /* Some accesses can write anything into the stack, others are
4441 bool clobber
= false;
4443 if (access_size
== 0 && !zero_size_allowed
) {
4444 verbose(env
, "invalid zero-sized read\n");
4448 if (type
== ACCESS_HELPER
) {
4449 /* The bounds checks for writes are more permissive than for
4450 * reads. However, if raw_mode is not set, we'll do extra
4453 bounds_check_type
= BPF_WRITE
;
4456 bounds_check_type
= BPF_READ
;
4458 err
= check_stack_access_within_bounds(env
, regno
, off
, access_size
,
4459 type
, bounds_check_type
);
4464 if (tnum_is_const(reg
->var_off
)) {
4465 min_off
= max_off
= reg
->var_off
.value
+ off
;
4467 /* Variable offset is prohibited for unprivileged mode for
4468 * simplicity since it requires corresponding support in
4469 * Spectre masking for stack ALU.
4470 * See also retrieve_ptr_limit().
4472 if (!env
->bypass_spec_v1
) {
4475 tnum_strn(tn_buf
, sizeof(tn_buf
), reg
->var_off
);
4476 verbose(env
, "R%d%s variable offset stack access prohibited for !root, var_off=%s\n",
4477 regno
, err_extra
, tn_buf
);
4480 /* Only initialized buffer on stack is allowed to be accessed
4481 * with variable offset. With uninitialized buffer it's hard to
4482 * guarantee that whole memory is marked as initialized on
4483 * helper return since specific bounds are unknown what may
4484 * cause uninitialized stack leaking.
4486 if (meta
&& meta
->raw_mode
)
4489 min_off
= reg
->smin_value
+ off
;
4490 max_off
= reg
->smax_value
+ off
;
4493 if (meta
&& meta
->raw_mode
) {
4494 meta
->access_size
= access_size
;
4495 meta
->regno
= regno
;
4499 for (i
= min_off
; i
< max_off
+ access_size
; i
++) {
4503 spi
= slot
/ BPF_REG_SIZE
;
4504 if (state
->allocated_stack
<= slot
)
4506 stype
= &state
->stack
[spi
].slot_type
[slot
% BPF_REG_SIZE
];
4507 if (*stype
== STACK_MISC
)
4509 if (*stype
== STACK_ZERO
) {
4511 /* helper can write anything into the stack */
4512 *stype
= STACK_MISC
;
4517 if (state
->stack
[spi
].slot_type
[0] == STACK_SPILL
&&
4518 state
->stack
[spi
].spilled_ptr
.type
== PTR_TO_BTF_ID
)
4521 if (state
->stack
[spi
].slot_type
[0] == STACK_SPILL
&&
4522 (state
->stack
[spi
].spilled_ptr
.type
== SCALAR_VALUE
||
4523 env
->allow_ptr_leaks
)) {
4525 __mark_reg_unknown(env
, &state
->stack
[spi
].spilled_ptr
);
4526 for (j
= 0; j
< BPF_REG_SIZE
; j
++)
4527 state
->stack
[spi
].slot_type
[j
] = STACK_MISC
;
4533 if (tnum_is_const(reg
->var_off
)) {
4534 verbose(env
, "invalid%s read from stack R%d off %d+%d size %d\n",
4535 err_extra
, regno
, min_off
, i
- min_off
, access_size
);
4539 tnum_strn(tn_buf
, sizeof(tn_buf
), reg
->var_off
);
4540 verbose(env
, "invalid%s read from stack R%d var_off %s+%d size %d\n",
4541 err_extra
, regno
, tn_buf
, i
- min_off
, access_size
);
4545 /* reading any byte out of 8-byte 'spill_slot' will cause
4546 * the whole slot to be marked as 'read'
4548 mark_reg_read(env
, &state
->stack
[spi
].spilled_ptr
,
4549 state
->stack
[spi
].spilled_ptr
.parent
,
4552 return update_stack_depth(env
, state
, min_off
);
4555 static int check_helper_mem_access(struct bpf_verifier_env
*env
, int regno
,
4556 int access_size
, bool zero_size_allowed
,
4557 struct bpf_call_arg_meta
*meta
)
4559 struct bpf_reg_state
*regs
= cur_regs(env
), *reg
= ®s
[regno
];
4561 switch (reg
->type
) {
4563 case PTR_TO_PACKET_META
:
4564 return check_packet_access(env
, regno
, reg
->off
, access_size
,
4566 case PTR_TO_MAP_KEY
:
4567 return check_mem_region_access(env
, regno
, reg
->off
, access_size
,
4568 reg
->map_ptr
->key_size
, false);
4569 case PTR_TO_MAP_VALUE
:
4570 if (check_map_access_type(env
, regno
, reg
->off
, access_size
,
4571 meta
&& meta
->raw_mode
? BPF_WRITE
:
4574 return check_map_access(env
, regno
, reg
->off
, access_size
,
4577 return check_mem_region_access(env
, regno
, reg
->off
,
4578 access_size
, reg
->mem_size
,
4580 case PTR_TO_RDONLY_BUF
:
4581 if (meta
&& meta
->raw_mode
)
4583 return check_buffer_access(env
, reg
, regno
, reg
->off
,
4584 access_size
, zero_size_allowed
,
4586 &env
->prog
->aux
->max_rdonly_access
);
4587 case PTR_TO_RDWR_BUF
:
4588 return check_buffer_access(env
, reg
, regno
, reg
->off
,
4589 access_size
, zero_size_allowed
,
4591 &env
->prog
->aux
->max_rdwr_access
);
4593 return check_stack_range_initialized(
4595 regno
, reg
->off
, access_size
,
4596 zero_size_allowed
, ACCESS_HELPER
, meta
);
4597 default: /* scalar_value or invalid ptr */
4598 /* Allow zero-byte read from NULL, regardless of pointer type */
4599 if (zero_size_allowed
&& access_size
== 0 &&
4600 register_is_null(reg
))
4603 verbose(env
, "R%d type=%s expected=%s\n", regno
,
4604 reg_type_str
[reg
->type
],
4605 reg_type_str
[PTR_TO_STACK
]);
4610 int check_mem_reg(struct bpf_verifier_env
*env
, struct bpf_reg_state
*reg
,
4611 u32 regno
, u32 mem_size
)
4613 if (register_is_null(reg
))
4616 if (reg_type_may_be_null(reg
->type
)) {
4617 /* Assuming that the register contains a value check if the memory
4618 * access is safe. Temporarily save and restore the register's state as
4619 * the conversion shouldn't be visible to a caller.
4621 const struct bpf_reg_state saved_reg
= *reg
;
4624 mark_ptr_not_null_reg(reg
);
4625 rv
= check_helper_mem_access(env
, regno
, mem_size
, true, NULL
);
4630 return check_helper_mem_access(env
, regno
, mem_size
, true, NULL
);
4633 /* Implementation details:
4634 * bpf_map_lookup returns PTR_TO_MAP_VALUE_OR_NULL
4635 * Two bpf_map_lookups (even with the same key) will have different reg->id.
4636 * For traditional PTR_TO_MAP_VALUE the verifier clears reg->id after
4637 * value_or_null->value transition, since the verifier only cares about
4638 * the range of access to valid map value pointer and doesn't care about actual
4639 * address of the map element.
4640 * For maps with 'struct bpf_spin_lock' inside map value the verifier keeps
4641 * reg->id > 0 after value_or_null->value transition. By doing so
4642 * two bpf_map_lookups will be considered two different pointers that
4643 * point to different bpf_spin_locks.
4644 * The verifier allows taking only one bpf_spin_lock at a time to avoid
4646 * Since only one bpf_spin_lock is allowed the checks are simpler than
4647 * reg_is_refcounted() logic. The verifier needs to remember only
4648 * one spin_lock instead of array of acquired_refs.
4649 * cur_state->active_spin_lock remembers which map value element got locked
4650 * and clears it after bpf_spin_unlock.
4652 static int process_spin_lock(struct bpf_verifier_env
*env
, int regno
,
4655 struct bpf_reg_state
*regs
= cur_regs(env
), *reg
= ®s
[regno
];
4656 struct bpf_verifier_state
*cur
= env
->cur_state
;
4657 bool is_const
= tnum_is_const(reg
->var_off
);
4658 struct bpf_map
*map
= reg
->map_ptr
;
4659 u64 val
= reg
->var_off
.value
;
4663 "R%d doesn't have constant offset. bpf_spin_lock has to be at the constant offset\n",
4669 "map '%s' has to have BTF in order to use bpf_spin_lock\n",
4673 if (!map_value_has_spin_lock(map
)) {
4674 if (map
->spin_lock_off
== -E2BIG
)
4676 "map '%s' has more than one 'struct bpf_spin_lock'\n",
4678 else if (map
->spin_lock_off
== -ENOENT
)
4680 "map '%s' doesn't have 'struct bpf_spin_lock'\n",
4684 "map '%s' is not a struct type or bpf_spin_lock is mangled\n",
4688 if (map
->spin_lock_off
!= val
+ reg
->off
) {
4689 verbose(env
, "off %lld doesn't point to 'struct bpf_spin_lock'\n",
4694 if (cur
->active_spin_lock
) {
4696 "Locking two bpf_spin_locks are not allowed\n");
4699 cur
->active_spin_lock
= reg
->id
;
4701 if (!cur
->active_spin_lock
) {
4702 verbose(env
, "bpf_spin_unlock without taking a lock\n");
4705 if (cur
->active_spin_lock
!= reg
->id
) {
4706 verbose(env
, "bpf_spin_unlock of different lock\n");
4709 cur
->active_spin_lock
= 0;
4714 static int process_timer_func(struct bpf_verifier_env
*env
, int regno
,
4715 struct bpf_call_arg_meta
*meta
)
4717 struct bpf_reg_state
*regs
= cur_regs(env
), *reg
= ®s
[regno
];
4718 bool is_const
= tnum_is_const(reg
->var_off
);
4719 struct bpf_map
*map
= reg
->map_ptr
;
4720 u64 val
= reg
->var_off
.value
;
4724 "R%d doesn't have constant offset. bpf_timer has to be at the constant offset\n",
4729 verbose(env
, "map '%s' has to have BTF in order to use bpf_timer\n",
4733 if (!map_value_has_timer(map
)) {
4734 if (map
->timer_off
== -E2BIG
)
4736 "map '%s' has more than one 'struct bpf_timer'\n",
4738 else if (map
->timer_off
== -ENOENT
)
4740 "map '%s' doesn't have 'struct bpf_timer'\n",
4744 "map '%s' is not a struct type or bpf_timer is mangled\n",
4748 if (map
->timer_off
!= val
+ reg
->off
) {
4749 verbose(env
, "off %lld doesn't point to 'struct bpf_timer' that is at %d\n",
4750 val
+ reg
->off
, map
->timer_off
);
4753 if (meta
->map_ptr
) {
4754 verbose(env
, "verifier bug. Two map pointers in a timer helper\n");
4757 meta
->map_uid
= reg
->map_uid
;
4758 meta
->map_ptr
= map
;
4762 static bool arg_type_is_mem_ptr(enum bpf_arg_type type
)
4764 return type
== ARG_PTR_TO_MEM
||
4765 type
== ARG_PTR_TO_MEM_OR_NULL
||
4766 type
== ARG_PTR_TO_UNINIT_MEM
;
4769 static bool arg_type_is_mem_size(enum bpf_arg_type type
)
4771 return type
== ARG_CONST_SIZE
||
4772 type
== ARG_CONST_SIZE_OR_ZERO
;
4775 static bool arg_type_is_alloc_size(enum bpf_arg_type type
)
4777 return type
== ARG_CONST_ALLOC_SIZE_OR_ZERO
;
4780 static bool arg_type_is_int_ptr(enum bpf_arg_type type
)
4782 return type
== ARG_PTR_TO_INT
||
4783 type
== ARG_PTR_TO_LONG
;
4786 static int int_ptr_type_to_size(enum bpf_arg_type type
)
4788 if (type
== ARG_PTR_TO_INT
)
4790 else if (type
== ARG_PTR_TO_LONG
)
4796 static int resolve_map_arg_type(struct bpf_verifier_env
*env
,
4797 const struct bpf_call_arg_meta
*meta
,
4798 enum bpf_arg_type
*arg_type
)
4800 if (!meta
->map_ptr
) {
4801 /* kernel subsystem misconfigured verifier */
4802 verbose(env
, "invalid map_ptr to access map->type\n");
4806 switch (meta
->map_ptr
->map_type
) {
4807 case BPF_MAP_TYPE_SOCKMAP
:
4808 case BPF_MAP_TYPE_SOCKHASH
:
4809 if (*arg_type
== ARG_PTR_TO_MAP_VALUE
) {
4810 *arg_type
= ARG_PTR_TO_BTF_ID_SOCK_COMMON
;
4812 verbose(env
, "invalid arg_type for sockmap/sockhash\n");
4823 struct bpf_reg_types
{
4824 const enum bpf_reg_type types
[10];
4828 static const struct bpf_reg_types map_key_value_types
= {
4838 static const struct bpf_reg_types sock_types
= {
4848 static const struct bpf_reg_types btf_id_sock_common_types
= {
4856 .btf_id
= &btf_sock_ids
[BTF_SOCK_TYPE_SOCK_COMMON
],
4860 static const struct bpf_reg_types mem_types
= {
4873 static const struct bpf_reg_types int_ptr_types
= {
4883 static const struct bpf_reg_types fullsock_types
= { .types
= { PTR_TO_SOCKET
} };
4884 static const struct bpf_reg_types scalar_types
= { .types
= { SCALAR_VALUE
} };
4885 static const struct bpf_reg_types context_types
= { .types
= { PTR_TO_CTX
} };
4886 static const struct bpf_reg_types alloc_mem_types
= { .types
= { PTR_TO_MEM
} };
4887 static const struct bpf_reg_types const_map_ptr_types
= { .types
= { CONST_PTR_TO_MAP
} };
4888 static const struct bpf_reg_types btf_ptr_types
= { .types
= { PTR_TO_BTF_ID
} };
4889 static const struct bpf_reg_types spin_lock_types
= { .types
= { PTR_TO_MAP_VALUE
} };
4890 static const struct bpf_reg_types percpu_btf_ptr_types
= { .types
= { PTR_TO_PERCPU_BTF_ID
} };
4891 static const struct bpf_reg_types func_ptr_types
= { .types
= { PTR_TO_FUNC
} };
4892 static const struct bpf_reg_types stack_ptr_types
= { .types
= { PTR_TO_STACK
} };
4893 static const struct bpf_reg_types const_str_ptr_types
= { .types
= { PTR_TO_MAP_VALUE
} };
4894 static const struct bpf_reg_types timer_types
= { .types
= { PTR_TO_MAP_VALUE
} };
4896 static const struct bpf_reg_types
*compatible_reg_types
[__BPF_ARG_TYPE_MAX
] = {
4897 [ARG_PTR_TO_MAP_KEY
] = &map_key_value_types
,
4898 [ARG_PTR_TO_MAP_VALUE
] = &map_key_value_types
,
4899 [ARG_PTR_TO_UNINIT_MAP_VALUE
] = &map_key_value_types
,
4900 [ARG_PTR_TO_MAP_VALUE_OR_NULL
] = &map_key_value_types
,
4901 [ARG_CONST_SIZE
] = &scalar_types
,
4902 [ARG_CONST_SIZE_OR_ZERO
] = &scalar_types
,
4903 [ARG_CONST_ALLOC_SIZE_OR_ZERO
] = &scalar_types
,
4904 [ARG_CONST_MAP_PTR
] = &const_map_ptr_types
,
4905 [ARG_PTR_TO_CTX
] = &context_types
,
4906 [ARG_PTR_TO_CTX_OR_NULL
] = &context_types
,
4907 [ARG_PTR_TO_SOCK_COMMON
] = &sock_types
,
4909 [ARG_PTR_TO_BTF_ID_SOCK_COMMON
] = &btf_id_sock_common_types
,
4911 [ARG_PTR_TO_SOCKET
] = &fullsock_types
,
4912 [ARG_PTR_TO_SOCKET_OR_NULL
] = &fullsock_types
,
4913 [ARG_PTR_TO_BTF_ID
] = &btf_ptr_types
,
4914 [ARG_PTR_TO_SPIN_LOCK
] = &spin_lock_types
,
4915 [ARG_PTR_TO_MEM
] = &mem_types
,
4916 [ARG_PTR_TO_MEM_OR_NULL
] = &mem_types
,
4917 [ARG_PTR_TO_UNINIT_MEM
] = &mem_types
,
4918 [ARG_PTR_TO_ALLOC_MEM
] = &alloc_mem_types
,
4919 [ARG_PTR_TO_ALLOC_MEM_OR_NULL
] = &alloc_mem_types
,
4920 [ARG_PTR_TO_INT
] = &int_ptr_types
,
4921 [ARG_PTR_TO_LONG
] = &int_ptr_types
,
4922 [ARG_PTR_TO_PERCPU_BTF_ID
] = &percpu_btf_ptr_types
,
4923 [ARG_PTR_TO_FUNC
] = &func_ptr_types
,
4924 [ARG_PTR_TO_STACK_OR_NULL
] = &stack_ptr_types
,
4925 [ARG_PTR_TO_CONST_STR
] = &const_str_ptr_types
,
4926 [ARG_PTR_TO_TIMER
] = &timer_types
,
4929 static int check_reg_type(struct bpf_verifier_env
*env
, u32 regno
,
4930 enum bpf_arg_type arg_type
,
4931 const u32
*arg_btf_id
)
4933 struct bpf_reg_state
*regs
= cur_regs(env
), *reg
= ®s
[regno
];
4934 enum bpf_reg_type expected
, type
= reg
->type
;
4935 const struct bpf_reg_types
*compatible
;
4938 compatible
= compatible_reg_types
[arg_type
];
4940 verbose(env
, "verifier internal error: unsupported arg type %d\n", arg_type
);
4944 for (i
= 0; i
< ARRAY_SIZE(compatible
->types
); i
++) {
4945 expected
= compatible
->types
[i
];
4946 if (expected
== NOT_INIT
)
4949 if (type
== expected
)
4953 verbose(env
, "R%d type=%s expected=", regno
, reg_type_str
[type
]);
4954 for (j
= 0; j
+ 1 < i
; j
++)
4955 verbose(env
, "%s, ", reg_type_str
[compatible
->types
[j
]]);
4956 verbose(env
, "%s\n", reg_type_str
[compatible
->types
[j
]]);
4960 if (type
== PTR_TO_BTF_ID
) {
4962 if (!compatible
->btf_id
) {
4963 verbose(env
, "verifier internal error: missing arg compatible BTF ID\n");
4966 arg_btf_id
= compatible
->btf_id
;
4969 if (!btf_struct_ids_match(&env
->log
, reg
->btf
, reg
->btf_id
, reg
->off
,
4970 btf_vmlinux
, *arg_btf_id
)) {
4971 verbose(env
, "R%d is of type %s but %s is expected\n",
4972 regno
, kernel_type_name(reg
->btf
, reg
->btf_id
),
4973 kernel_type_name(btf_vmlinux
, *arg_btf_id
));
4977 if (!tnum_is_const(reg
->var_off
) || reg
->var_off
.value
) {
4978 verbose(env
, "R%d is a pointer to in-kernel struct with non-zero offset\n",
4987 static int check_func_arg(struct bpf_verifier_env
*env
, u32 arg
,
4988 struct bpf_call_arg_meta
*meta
,
4989 const struct bpf_func_proto
*fn
)
4991 u32 regno
= BPF_REG_1
+ arg
;
4992 struct bpf_reg_state
*regs
= cur_regs(env
), *reg
= ®s
[regno
];
4993 enum bpf_arg_type arg_type
= fn
->arg_type
[arg
];
4994 enum bpf_reg_type type
= reg
->type
;
4997 if (arg_type
== ARG_DONTCARE
)
5000 err
= check_reg_arg(env
, regno
, SRC_OP
);
5004 if (arg_type
== ARG_ANYTHING
) {
5005 if (is_pointer_value(env
, regno
)) {
5006 verbose(env
, "R%d leaks addr into helper function\n",
5013 if (type_is_pkt_pointer(type
) &&
5014 !may_access_direct_pkt_data(env
, meta
, BPF_READ
)) {
5015 verbose(env
, "helper access to the packet is not allowed\n");
5019 if (arg_type
== ARG_PTR_TO_MAP_VALUE
||
5020 arg_type
== ARG_PTR_TO_UNINIT_MAP_VALUE
||
5021 arg_type
== ARG_PTR_TO_MAP_VALUE_OR_NULL
) {
5022 err
= resolve_map_arg_type(env
, meta
, &arg_type
);
5027 if (register_is_null(reg
) && arg_type_may_be_null(arg_type
))
5028 /* A NULL register has a SCALAR_VALUE type, so skip
5031 goto skip_type_check
;
5033 err
= check_reg_type(env
, regno
, arg_type
, fn
->arg_btf_id
[arg
]);
5037 if (type
== PTR_TO_CTX
) {
5038 err
= check_ctx_reg(env
, reg
, regno
);
5044 if (reg
->ref_obj_id
) {
5045 if (meta
->ref_obj_id
) {
5046 verbose(env
, "verifier internal error: more than one arg with ref_obj_id R%d %u %u\n",
5047 regno
, reg
->ref_obj_id
,
5051 meta
->ref_obj_id
= reg
->ref_obj_id
;
5054 if (arg_type
== ARG_CONST_MAP_PTR
) {
5055 /* bpf_map_xxx(map_ptr) call: remember that map_ptr */
5056 if (meta
->map_ptr
) {
5057 /* Use map_uid (which is unique id of inner map) to reject:
5058 * inner_map1 = bpf_map_lookup_elem(outer_map, key1)
5059 * inner_map2 = bpf_map_lookup_elem(outer_map, key2)
5060 * if (inner_map1 && inner_map2) {
5061 * timer = bpf_map_lookup_elem(inner_map1);
5063 * // mismatch would have been allowed
5064 * bpf_timer_init(timer, inner_map2);
5067 * Comparing map_ptr is enough to distinguish normal and outer maps.
5069 if (meta
->map_ptr
!= reg
->map_ptr
||
5070 meta
->map_uid
!= reg
->map_uid
) {
5072 "timer pointer in R1 map_uid=%d doesn't match map pointer in R2 map_uid=%d\n",
5073 meta
->map_uid
, reg
->map_uid
);
5077 meta
->map_ptr
= reg
->map_ptr
;
5078 meta
->map_uid
= reg
->map_uid
;
5079 } else if (arg_type
== ARG_PTR_TO_MAP_KEY
) {
5080 /* bpf_map_xxx(..., map_ptr, ..., key) call:
5081 * check that [key, key + map->key_size) are within
5082 * stack limits and initialized
5084 if (!meta
->map_ptr
) {
5085 /* in function declaration map_ptr must come before
5086 * map_key, so that it's verified and known before
5087 * we have to check map_key here. Otherwise it means
5088 * that kernel subsystem misconfigured verifier
5090 verbose(env
, "invalid map_ptr to access map->key\n");
5093 err
= check_helper_mem_access(env
, regno
,
5094 meta
->map_ptr
->key_size
, false,
5096 } else if (arg_type
== ARG_PTR_TO_MAP_VALUE
||
5097 (arg_type
== ARG_PTR_TO_MAP_VALUE_OR_NULL
&&
5098 !register_is_null(reg
)) ||
5099 arg_type
== ARG_PTR_TO_UNINIT_MAP_VALUE
) {
5100 /* bpf_map_xxx(..., map_ptr, ..., value) call:
5101 * check [value, value + map->value_size) validity
5103 if (!meta
->map_ptr
) {
5104 /* kernel subsystem misconfigured verifier */
5105 verbose(env
, "invalid map_ptr to access map->value\n");
5108 meta
->raw_mode
= (arg_type
== ARG_PTR_TO_UNINIT_MAP_VALUE
);
5109 err
= check_helper_mem_access(env
, regno
,
5110 meta
->map_ptr
->value_size
, false,
5112 } else if (arg_type
== ARG_PTR_TO_PERCPU_BTF_ID
) {
5114 verbose(env
, "Helper has invalid btf_id in R%d\n", regno
);
5117 meta
->ret_btf
= reg
->btf
;
5118 meta
->ret_btf_id
= reg
->btf_id
;
5119 } else if (arg_type
== ARG_PTR_TO_SPIN_LOCK
) {
5120 if (meta
->func_id
== BPF_FUNC_spin_lock
) {
5121 if (process_spin_lock(env
, regno
, true))
5123 } else if (meta
->func_id
== BPF_FUNC_spin_unlock
) {
5124 if (process_spin_lock(env
, regno
, false))
5127 verbose(env
, "verifier internal error\n");
5130 } else if (arg_type
== ARG_PTR_TO_TIMER
) {
5131 if (process_timer_func(env
, regno
, meta
))
5133 } else if (arg_type
== ARG_PTR_TO_FUNC
) {
5134 meta
->subprogno
= reg
->subprogno
;
5135 } else if (arg_type_is_mem_ptr(arg_type
)) {
5136 /* The access to this pointer is only checked when we hit the
5137 * next is_mem_size argument below.
5139 meta
->raw_mode
= (arg_type
== ARG_PTR_TO_UNINIT_MEM
);
5140 } else if (arg_type_is_mem_size(arg_type
)) {
5141 bool zero_size_allowed
= (arg_type
== ARG_CONST_SIZE_OR_ZERO
);
5143 /* This is used to refine r0 return value bounds for helpers
5144 * that enforce this value as an upper bound on return values.
5145 * See do_refine_retval_range() for helpers that can refine
5146 * the return value. C type of helper is u32 so we pull register
5147 * bound from umax_value however, if negative verifier errors
5148 * out. Only upper bounds can be learned because retval is an
5149 * int type and negative retvals are allowed.
5151 meta
->msize_max_value
= reg
->umax_value
;
5153 /* The register is SCALAR_VALUE; the access check
5154 * happens using its boundaries.
5156 if (!tnum_is_const(reg
->var_off
))
5157 /* For unprivileged variable accesses, disable raw
5158 * mode so that the program is required to
5159 * initialize all the memory that the helper could
5160 * just partially fill up.
5164 if (reg
->smin_value
< 0) {
5165 verbose(env
, "R%d min value is negative, either use unsigned or 'var &= const'\n",
5170 if (reg
->umin_value
== 0) {
5171 err
= check_helper_mem_access(env
, regno
- 1, 0,
5178 if (reg
->umax_value
>= BPF_MAX_VAR_SIZ
) {
5179 verbose(env
, "R%d unbounded memory access, use 'var &= const' or 'if (var < const)'\n",
5183 err
= check_helper_mem_access(env
, regno
- 1,
5185 zero_size_allowed
, meta
);
5187 err
= mark_chain_precision(env
, regno
);
5188 } else if (arg_type_is_alloc_size(arg_type
)) {
5189 if (!tnum_is_const(reg
->var_off
)) {
5190 verbose(env
, "R%d is not a known constant'\n",
5194 meta
->mem_size
= reg
->var_off
.value
;
5195 } else if (arg_type_is_int_ptr(arg_type
)) {
5196 int size
= int_ptr_type_to_size(arg_type
);
5198 err
= check_helper_mem_access(env
, regno
, size
, false, meta
);
5201 err
= check_ptr_alignment(env
, reg
, 0, size
, true);
5202 } else if (arg_type
== ARG_PTR_TO_CONST_STR
) {
5203 struct bpf_map
*map
= reg
->map_ptr
;
5208 if (!bpf_map_is_rdonly(map
)) {
5209 verbose(env
, "R%d does not point to a readonly map'\n", regno
);
5213 if (!tnum_is_const(reg
->var_off
)) {
5214 verbose(env
, "R%d is not a constant address'\n", regno
);
5218 if (!map
->ops
->map_direct_value_addr
) {
5219 verbose(env
, "no direct value access support for this map type\n");
5223 err
= check_map_access(env
, regno
, reg
->off
,
5224 map
->value_size
- reg
->off
, false);
5228 map_off
= reg
->off
+ reg
->var_off
.value
;
5229 err
= map
->ops
->map_direct_value_addr(map
, &map_addr
, map_off
);
5231 verbose(env
, "direct value access on string failed\n");
5235 str_ptr
= (char *)(long)(map_addr
);
5236 if (!strnchr(str_ptr
+ map_off
, map
->value_size
- map_off
, 0)) {
5237 verbose(env
, "string is not zero-terminated\n");
5245 static bool may_update_sockmap(struct bpf_verifier_env
*env
, int func_id
)
5247 enum bpf_attach_type eatype
= env
->prog
->expected_attach_type
;
5248 enum bpf_prog_type type
= resolve_prog_type(env
->prog
);
5250 if (func_id
!= BPF_FUNC_map_update_elem
)
5253 /* It's not possible to get access to a locked struct sock in these
5254 * contexts, so updating is safe.
5257 case BPF_PROG_TYPE_TRACING
:
5258 if (eatype
== BPF_TRACE_ITER
)
5261 case BPF_PROG_TYPE_SOCKET_FILTER
:
5262 case BPF_PROG_TYPE_SCHED_CLS
:
5263 case BPF_PROG_TYPE_SCHED_ACT
:
5264 case BPF_PROG_TYPE_XDP
:
5265 case BPF_PROG_TYPE_SK_REUSEPORT
:
5266 case BPF_PROG_TYPE_FLOW_DISSECTOR
:
5267 case BPF_PROG_TYPE_SK_LOOKUP
:
5273 verbose(env
, "cannot update sockmap in this context\n");
5277 static bool allow_tail_call_in_subprogs(struct bpf_verifier_env
*env
)
5279 return env
->prog
->jit_requested
&& IS_ENABLED(CONFIG_X86_64
);
5282 static int check_map_func_compatibility(struct bpf_verifier_env
*env
,
5283 struct bpf_map
*map
, int func_id
)
5288 /* We need a two way check, first is from map perspective ... */
5289 switch (map
->map_type
) {
5290 case BPF_MAP_TYPE_PROG_ARRAY
:
5291 if (func_id
!= BPF_FUNC_tail_call
)
5294 case BPF_MAP_TYPE_PERF_EVENT_ARRAY
:
5295 if (func_id
!= BPF_FUNC_perf_event_read
&&
5296 func_id
!= BPF_FUNC_perf_event_output
&&
5297 func_id
!= BPF_FUNC_skb_output
&&
5298 func_id
!= BPF_FUNC_perf_event_read_value
&&
5299 func_id
!= BPF_FUNC_xdp_output
)
5302 case BPF_MAP_TYPE_RINGBUF
:
5303 if (func_id
!= BPF_FUNC_ringbuf_output
&&
5304 func_id
!= BPF_FUNC_ringbuf_reserve
&&
5305 func_id
!= BPF_FUNC_ringbuf_query
)
5308 case BPF_MAP_TYPE_STACK_TRACE
:
5309 if (func_id
!= BPF_FUNC_get_stackid
)
5312 case BPF_MAP_TYPE_CGROUP_ARRAY
:
5313 if (func_id
!= BPF_FUNC_skb_under_cgroup
&&
5314 func_id
!= BPF_FUNC_current_task_under_cgroup
)
5317 case BPF_MAP_TYPE_CGROUP_STORAGE
:
5318 case BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE
:
5319 if (func_id
!= BPF_FUNC_get_local_storage
)
5322 case BPF_MAP_TYPE_DEVMAP
:
5323 case BPF_MAP_TYPE_DEVMAP_HASH
:
5324 if (func_id
!= BPF_FUNC_redirect_map
&&
5325 func_id
!= BPF_FUNC_map_lookup_elem
)
5328 /* Restrict bpf side of cpumap and xskmap, open when use-cases
5331 case BPF_MAP_TYPE_CPUMAP
:
5332 if (func_id
!= BPF_FUNC_redirect_map
)
5335 case BPF_MAP_TYPE_XSKMAP
:
5336 if (func_id
!= BPF_FUNC_redirect_map
&&
5337 func_id
!= BPF_FUNC_map_lookup_elem
)
5340 case BPF_MAP_TYPE_ARRAY_OF_MAPS
:
5341 case BPF_MAP_TYPE_HASH_OF_MAPS
:
5342 if (func_id
!= BPF_FUNC_map_lookup_elem
)
5345 case BPF_MAP_TYPE_SOCKMAP
:
5346 if (func_id
!= BPF_FUNC_sk_redirect_map
&&
5347 func_id
!= BPF_FUNC_sock_map_update
&&
5348 func_id
!= BPF_FUNC_map_delete_elem
&&
5349 func_id
!= BPF_FUNC_msg_redirect_map
&&
5350 func_id
!= BPF_FUNC_sk_select_reuseport
&&
5351 func_id
!= BPF_FUNC_map_lookup_elem
&&
5352 !may_update_sockmap(env
, func_id
))
5355 case BPF_MAP_TYPE_SOCKHASH
:
5356 if (func_id
!= BPF_FUNC_sk_redirect_hash
&&
5357 func_id
!= BPF_FUNC_sock_hash_update
&&
5358 func_id
!= BPF_FUNC_map_delete_elem
&&
5359 func_id
!= BPF_FUNC_msg_redirect_hash
&&
5360 func_id
!= BPF_FUNC_sk_select_reuseport
&&
5361 func_id
!= BPF_FUNC_map_lookup_elem
&&
5362 !may_update_sockmap(env
, func_id
))
5365 case BPF_MAP_TYPE_REUSEPORT_SOCKARRAY
:
5366 if (func_id
!= BPF_FUNC_sk_select_reuseport
)
5369 case BPF_MAP_TYPE_QUEUE
:
5370 case BPF_MAP_TYPE_STACK
:
5371 if (func_id
!= BPF_FUNC_map_peek_elem
&&
5372 func_id
!= BPF_FUNC_map_pop_elem
&&
5373 func_id
!= BPF_FUNC_map_push_elem
)
5376 case BPF_MAP_TYPE_SK_STORAGE
:
5377 if (func_id
!= BPF_FUNC_sk_storage_get
&&
5378 func_id
!= BPF_FUNC_sk_storage_delete
)
5381 case BPF_MAP_TYPE_INODE_STORAGE
:
5382 if (func_id
!= BPF_FUNC_inode_storage_get
&&
5383 func_id
!= BPF_FUNC_inode_storage_delete
)
5386 case BPF_MAP_TYPE_TASK_STORAGE
:
5387 if (func_id
!= BPF_FUNC_task_storage_get
&&
5388 func_id
!= BPF_FUNC_task_storage_delete
)
5395 /* ... and second from the function itself. */
5397 case BPF_FUNC_tail_call
:
5398 if (map
->map_type
!= BPF_MAP_TYPE_PROG_ARRAY
)
5400 if (env
->subprog_cnt
> 1 && !allow_tail_call_in_subprogs(env
)) {
5401 verbose(env
, "tail_calls are not allowed in non-JITed programs with bpf-to-bpf calls\n");
5405 case BPF_FUNC_perf_event_read
:
5406 case BPF_FUNC_perf_event_output
:
5407 case BPF_FUNC_perf_event_read_value
:
5408 case BPF_FUNC_skb_output
:
5409 case BPF_FUNC_xdp_output
:
5410 if (map
->map_type
!= BPF_MAP_TYPE_PERF_EVENT_ARRAY
)
5413 case BPF_FUNC_ringbuf_output
:
5414 case BPF_FUNC_ringbuf_reserve
:
5415 case BPF_FUNC_ringbuf_query
:
5416 if (map
->map_type
!= BPF_MAP_TYPE_RINGBUF
)
5419 case BPF_FUNC_get_stackid
:
5420 if (map
->map_type
!= BPF_MAP_TYPE_STACK_TRACE
)
5423 case BPF_FUNC_current_task_under_cgroup
:
5424 case BPF_FUNC_skb_under_cgroup
:
5425 if (map
->map_type
!= BPF_MAP_TYPE_CGROUP_ARRAY
)
5428 case BPF_FUNC_redirect_map
:
5429 if (map
->map_type
!= BPF_MAP_TYPE_DEVMAP
&&
5430 map
->map_type
!= BPF_MAP_TYPE_DEVMAP_HASH
&&
5431 map
->map_type
!= BPF_MAP_TYPE_CPUMAP
&&
5432 map
->map_type
!= BPF_MAP_TYPE_XSKMAP
)
5435 case BPF_FUNC_sk_redirect_map
:
5436 case BPF_FUNC_msg_redirect_map
:
5437 case BPF_FUNC_sock_map_update
:
5438 if (map
->map_type
!= BPF_MAP_TYPE_SOCKMAP
)
5441 case BPF_FUNC_sk_redirect_hash
:
5442 case BPF_FUNC_msg_redirect_hash
:
5443 case BPF_FUNC_sock_hash_update
:
5444 if (map
->map_type
!= BPF_MAP_TYPE_SOCKHASH
)
5447 case BPF_FUNC_get_local_storage
:
5448 if (map
->map_type
!= BPF_MAP_TYPE_CGROUP_STORAGE
&&
5449 map
->map_type
!= BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE
)
5452 case BPF_FUNC_sk_select_reuseport
:
5453 if (map
->map_type
!= BPF_MAP_TYPE_REUSEPORT_SOCKARRAY
&&
5454 map
->map_type
!= BPF_MAP_TYPE_SOCKMAP
&&
5455 map
->map_type
!= BPF_MAP_TYPE_SOCKHASH
)
5458 case BPF_FUNC_map_peek_elem
:
5459 case BPF_FUNC_map_pop_elem
:
5460 case BPF_FUNC_map_push_elem
:
5461 if (map
->map_type
!= BPF_MAP_TYPE_QUEUE
&&
5462 map
->map_type
!= BPF_MAP_TYPE_STACK
)
5465 case BPF_FUNC_sk_storage_get
:
5466 case BPF_FUNC_sk_storage_delete
:
5467 if (map
->map_type
!= BPF_MAP_TYPE_SK_STORAGE
)
5470 case BPF_FUNC_inode_storage_get
:
5471 case BPF_FUNC_inode_storage_delete
:
5472 if (map
->map_type
!= BPF_MAP_TYPE_INODE_STORAGE
)
5475 case BPF_FUNC_task_storage_get
:
5476 case BPF_FUNC_task_storage_delete
:
5477 if (map
->map_type
!= BPF_MAP_TYPE_TASK_STORAGE
)
5486 verbose(env
, "cannot pass map_type %d into func %s#%d\n",
5487 map
->map_type
, func_id_name(func_id
), func_id
);
5491 static bool check_raw_mode_ok(const struct bpf_func_proto
*fn
)
5495 if (fn
->arg1_type
== ARG_PTR_TO_UNINIT_MEM
)
5497 if (fn
->arg2_type
== ARG_PTR_TO_UNINIT_MEM
)
5499 if (fn
->arg3_type
== ARG_PTR_TO_UNINIT_MEM
)
5501 if (fn
->arg4_type
== ARG_PTR_TO_UNINIT_MEM
)
5503 if (fn
->arg5_type
== ARG_PTR_TO_UNINIT_MEM
)
5506 /* We only support one arg being in raw mode at the moment,
5507 * which is sufficient for the helper functions we have
5513 static bool check_args_pair_invalid(enum bpf_arg_type arg_curr
,
5514 enum bpf_arg_type arg_next
)
5516 return (arg_type_is_mem_ptr(arg_curr
) &&
5517 !arg_type_is_mem_size(arg_next
)) ||
5518 (!arg_type_is_mem_ptr(arg_curr
) &&
5519 arg_type_is_mem_size(arg_next
));
5522 static bool check_arg_pair_ok(const struct bpf_func_proto
*fn
)
5524 /* bpf_xxx(..., buf, len) call will access 'len'
5525 * bytes from memory 'buf'. Both arg types need
5526 * to be paired, so make sure there's no buggy
5527 * helper function specification.
5529 if (arg_type_is_mem_size(fn
->arg1_type
) ||
5530 arg_type_is_mem_ptr(fn
->arg5_type
) ||
5531 check_args_pair_invalid(fn
->arg1_type
, fn
->arg2_type
) ||
5532 check_args_pair_invalid(fn
->arg2_type
, fn
->arg3_type
) ||
5533 check_args_pair_invalid(fn
->arg3_type
, fn
->arg4_type
) ||
5534 check_args_pair_invalid(fn
->arg4_type
, fn
->arg5_type
))
5540 static bool check_refcount_ok(const struct bpf_func_proto
*fn
, int func_id
)
5544 if (arg_type_may_be_refcounted(fn
->arg1_type
))
5546 if (arg_type_may_be_refcounted(fn
->arg2_type
))
5548 if (arg_type_may_be_refcounted(fn
->arg3_type
))
5550 if (arg_type_may_be_refcounted(fn
->arg4_type
))
5552 if (arg_type_may_be_refcounted(fn
->arg5_type
))
5555 /* A reference acquiring function cannot acquire
5556 * another refcounted ptr.
5558 if (may_be_acquire_function(func_id
) && count
)
5561 /* We only support one arg being unreferenced at the moment,
5562 * which is sufficient for the helper functions we have right now.
5567 static bool check_btf_id_ok(const struct bpf_func_proto
*fn
)
5571 for (i
= 0; i
< ARRAY_SIZE(fn
->arg_type
); i
++) {
5572 if (fn
->arg_type
[i
] == ARG_PTR_TO_BTF_ID
&& !fn
->arg_btf_id
[i
])
5575 if (fn
->arg_type
[i
] != ARG_PTR_TO_BTF_ID
&& fn
->arg_btf_id
[i
])
5582 static int check_func_proto(const struct bpf_func_proto
*fn
, int func_id
)
5584 return check_raw_mode_ok(fn
) &&
5585 check_arg_pair_ok(fn
) &&
5586 check_btf_id_ok(fn
) &&
5587 check_refcount_ok(fn
, func_id
) ? 0 : -EINVAL
;
5590 /* Packet data might have moved, any old PTR_TO_PACKET[_META,_END]
5591 * are now invalid, so turn them into unknown SCALAR_VALUE.
5593 static void __clear_all_pkt_pointers(struct bpf_verifier_env
*env
,
5594 struct bpf_func_state
*state
)
5596 struct bpf_reg_state
*regs
= state
->regs
, *reg
;
5599 for (i
= 0; i
< MAX_BPF_REG
; i
++)
5600 if (reg_is_pkt_pointer_any(®s
[i
]))
5601 mark_reg_unknown(env
, regs
, i
);
5603 bpf_for_each_spilled_reg(i
, state
, reg
) {
5606 if (reg_is_pkt_pointer_any(reg
))
5607 __mark_reg_unknown(env
, reg
);
5611 static void clear_all_pkt_pointers(struct bpf_verifier_env
*env
)
5613 struct bpf_verifier_state
*vstate
= env
->cur_state
;
5616 for (i
= 0; i
<= vstate
->curframe
; i
++)
5617 __clear_all_pkt_pointers(env
, vstate
->frame
[i
]);
5622 BEYOND_PKT_END
= -2,
5625 static void mark_pkt_end(struct bpf_verifier_state
*vstate
, int regn
, bool range_open
)
5627 struct bpf_func_state
*state
= vstate
->frame
[vstate
->curframe
];
5628 struct bpf_reg_state
*reg
= &state
->regs
[regn
];
5630 if (reg
->type
!= PTR_TO_PACKET
)
5631 /* PTR_TO_PACKET_META is not supported yet */
5634 /* The 'reg' is pkt > pkt_end or pkt >= pkt_end.
5635 * How far beyond pkt_end it goes is unknown.
5636 * if (!range_open) it's the case of pkt >= pkt_end
5637 * if (range_open) it's the case of pkt > pkt_end
5638 * hence this pointer is at least 1 byte bigger than pkt_end
5641 reg
->range
= BEYOND_PKT_END
;
5643 reg
->range
= AT_PKT_END
;
5646 static void release_reg_references(struct bpf_verifier_env
*env
,
5647 struct bpf_func_state
*state
,
5650 struct bpf_reg_state
*regs
= state
->regs
, *reg
;
5653 for (i
= 0; i
< MAX_BPF_REG
; i
++)
5654 if (regs
[i
].ref_obj_id
== ref_obj_id
)
5655 mark_reg_unknown(env
, regs
, i
);
5657 bpf_for_each_spilled_reg(i
, state
, reg
) {
5660 if (reg
->ref_obj_id
== ref_obj_id
)
5661 __mark_reg_unknown(env
, reg
);
5665 /* The pointer with the specified id has released its reference to kernel
5666 * resources. Identify all copies of the same pointer and clear the reference.
5668 static int release_reference(struct bpf_verifier_env
*env
,
5671 struct bpf_verifier_state
*vstate
= env
->cur_state
;
5675 err
= release_reference_state(cur_func(env
), ref_obj_id
);
5679 for (i
= 0; i
<= vstate
->curframe
; i
++)
5680 release_reg_references(env
, vstate
->frame
[i
], ref_obj_id
);
5685 static void clear_caller_saved_regs(struct bpf_verifier_env
*env
,
5686 struct bpf_reg_state
*regs
)
5690 /* after the call registers r0 - r5 were scratched */
5691 for (i
= 0; i
< CALLER_SAVED_REGS
; i
++) {
5692 mark_reg_not_init(env
, regs
, caller_saved
[i
]);
5693 check_reg_arg(env
, caller_saved
[i
], DST_OP_NO_MARK
);
5697 typedef int (*set_callee_state_fn
)(struct bpf_verifier_env
*env
,
5698 struct bpf_func_state
*caller
,
5699 struct bpf_func_state
*callee
,
5702 static int __check_func_call(struct bpf_verifier_env
*env
, struct bpf_insn
*insn
,
5703 int *insn_idx
, int subprog
,
5704 set_callee_state_fn set_callee_state_cb
)
5706 struct bpf_verifier_state
*state
= env
->cur_state
;
5707 struct bpf_func_info_aux
*func_info_aux
;
5708 struct bpf_func_state
*caller
, *callee
;
5710 bool is_global
= false;
5712 if (state
->curframe
+ 1 >= MAX_CALL_FRAMES
) {
5713 verbose(env
, "the call stack of %d frames is too deep\n",
5714 state
->curframe
+ 2);
5718 caller
= state
->frame
[state
->curframe
];
5719 if (state
->frame
[state
->curframe
+ 1]) {
5720 verbose(env
, "verifier bug. Frame %d already allocated\n",
5721 state
->curframe
+ 1);
5725 func_info_aux
= env
->prog
->aux
->func_info_aux
;
5727 is_global
= func_info_aux
[subprog
].linkage
== BTF_FUNC_GLOBAL
;
5728 err
= btf_check_subprog_arg_match(env
, subprog
, caller
->regs
);
5733 verbose(env
, "Caller passes invalid args into func#%d\n",
5737 if (env
->log
.level
& BPF_LOG_LEVEL
)
5739 "Func#%d is global and valid. Skipping.\n",
5741 clear_caller_saved_regs(env
, caller
->regs
);
5743 /* All global functions return a 64-bit SCALAR_VALUE */
5744 mark_reg_unknown(env
, caller
->regs
, BPF_REG_0
);
5745 caller
->regs
[BPF_REG_0
].subreg_def
= DEF_NOT_SUBREG
;
5747 /* continue with next insn after call */
5752 if (insn
->code
== (BPF_JMP
| BPF_CALL
) &&
5753 insn
->imm
== BPF_FUNC_timer_set_callback
) {
5754 struct bpf_verifier_state
*async_cb
;
5756 /* there is no real recursion here. timer callbacks are async */
5757 env
->subprog_info
[subprog
].is_async_cb
= true;
5758 async_cb
= push_async_cb(env
, env
->subprog_info
[subprog
].start
,
5759 *insn_idx
, subprog
);
5762 callee
= async_cb
->frame
[0];
5763 callee
->async_entry_cnt
= caller
->async_entry_cnt
+ 1;
5765 /* Convert bpf_timer_set_callback() args into timer callback args */
5766 err
= set_callee_state_cb(env
, caller
, callee
, *insn_idx
);
5770 clear_caller_saved_regs(env
, caller
->regs
);
5771 mark_reg_unknown(env
, caller
->regs
, BPF_REG_0
);
5772 caller
->regs
[BPF_REG_0
].subreg_def
= DEF_NOT_SUBREG
;
5773 /* continue with next insn after call */
5777 callee
= kzalloc(sizeof(*callee
), GFP_KERNEL
);
5780 state
->frame
[state
->curframe
+ 1] = callee
;
5782 /* callee cannot access r0, r6 - r9 for reading and has to write
5783 * into its own stack before reading from it.
5784 * callee can read/write into caller's stack
5786 init_func_state(env
, callee
,
5787 /* remember the callsite, it will be used by bpf_exit */
5788 *insn_idx
/* callsite */,
5789 state
->curframe
+ 1 /* frameno within this callchain */,
5790 subprog
/* subprog number within this prog */);
5792 /* Transfer references to the callee */
5793 err
= copy_reference_state(callee
, caller
);
5797 err
= set_callee_state_cb(env
, caller
, callee
, *insn_idx
);
5801 clear_caller_saved_regs(env
, caller
->regs
);
5803 /* only increment it after check_reg_arg() finished */
5806 /* and go analyze first insn of the callee */
5807 *insn_idx
= env
->subprog_info
[subprog
].start
- 1;
5809 if (env
->log
.level
& BPF_LOG_LEVEL
) {
5810 verbose(env
, "caller:\n");
5811 print_verifier_state(env
, caller
);
5812 verbose(env
, "callee:\n");
5813 print_verifier_state(env
, callee
);
5818 int map_set_for_each_callback_args(struct bpf_verifier_env
*env
,
5819 struct bpf_func_state
*caller
,
5820 struct bpf_func_state
*callee
)
5822 /* bpf_for_each_map_elem(struct bpf_map *map, void *callback_fn,
5823 * void *callback_ctx, u64 flags);
5824 * callback_fn(struct bpf_map *map, void *key, void *value,
5825 * void *callback_ctx);
5827 callee
->regs
[BPF_REG_1
] = caller
->regs
[BPF_REG_1
];
5829 callee
->regs
[BPF_REG_2
].type
= PTR_TO_MAP_KEY
;
5830 __mark_reg_known_zero(&callee
->regs
[BPF_REG_2
]);
5831 callee
->regs
[BPF_REG_2
].map_ptr
= caller
->regs
[BPF_REG_1
].map_ptr
;
5833 callee
->regs
[BPF_REG_3
].type
= PTR_TO_MAP_VALUE
;
5834 __mark_reg_known_zero(&callee
->regs
[BPF_REG_3
]);
5835 callee
->regs
[BPF_REG_3
].map_ptr
= caller
->regs
[BPF_REG_1
].map_ptr
;
5837 /* pointer to stack or null */
5838 callee
->regs
[BPF_REG_4
] = caller
->regs
[BPF_REG_3
];
5841 __mark_reg_not_init(env
, &callee
->regs
[BPF_REG_5
]);
5845 static int set_callee_state(struct bpf_verifier_env
*env
,
5846 struct bpf_func_state
*caller
,
5847 struct bpf_func_state
*callee
, int insn_idx
)
5851 /* copy r1 - r5 args that callee can access. The copy includes parent
5852 * pointers, which connects us up to the liveness chain
5854 for (i
= BPF_REG_1
; i
<= BPF_REG_5
; i
++)
5855 callee
->regs
[i
] = caller
->regs
[i
];
5859 static int check_func_call(struct bpf_verifier_env
*env
, struct bpf_insn
*insn
,
5862 int subprog
, target_insn
;
5864 target_insn
= *insn_idx
+ insn
->imm
+ 1;
5865 subprog
= find_subprog(env
, target_insn
);
5867 verbose(env
, "verifier bug. No program starts at insn %d\n",
5872 return __check_func_call(env
, insn
, insn_idx
, subprog
, set_callee_state
);
5875 static int set_map_elem_callback_state(struct bpf_verifier_env
*env
,
5876 struct bpf_func_state
*caller
,
5877 struct bpf_func_state
*callee
,
5880 struct bpf_insn_aux_data
*insn_aux
= &env
->insn_aux_data
[insn_idx
];
5881 struct bpf_map
*map
;
5884 if (bpf_map_ptr_poisoned(insn_aux
)) {
5885 verbose(env
, "tail_call abusing map_ptr\n");
5889 map
= BPF_MAP_PTR(insn_aux
->map_ptr_state
);
5890 if (!map
->ops
->map_set_for_each_callback_args
||
5891 !map
->ops
->map_for_each_callback
) {
5892 verbose(env
, "callback function not allowed for map\n");
5896 err
= map
->ops
->map_set_for_each_callback_args(env
, caller
, callee
);
5900 callee
->in_callback_fn
= true;
5904 static int set_timer_callback_state(struct bpf_verifier_env
*env
,
5905 struct bpf_func_state
*caller
,
5906 struct bpf_func_state
*callee
,
5909 struct bpf_map
*map_ptr
= caller
->regs
[BPF_REG_1
].map_ptr
;
5911 /* bpf_timer_set_callback(struct bpf_timer *timer, void *callback_fn);
5912 * callback_fn(struct bpf_map *map, void *key, void *value);
5914 callee
->regs
[BPF_REG_1
].type
= CONST_PTR_TO_MAP
;
5915 __mark_reg_known_zero(&callee
->regs
[BPF_REG_1
]);
5916 callee
->regs
[BPF_REG_1
].map_ptr
= map_ptr
;
5918 callee
->regs
[BPF_REG_2
].type
= PTR_TO_MAP_KEY
;
5919 __mark_reg_known_zero(&callee
->regs
[BPF_REG_2
]);
5920 callee
->regs
[BPF_REG_2
].map_ptr
= map_ptr
;
5922 callee
->regs
[BPF_REG_3
].type
= PTR_TO_MAP_VALUE
;
5923 __mark_reg_known_zero(&callee
->regs
[BPF_REG_3
]);
5924 callee
->regs
[BPF_REG_3
].map_ptr
= map_ptr
;
5927 __mark_reg_not_init(env
, &callee
->regs
[BPF_REG_4
]);
5928 __mark_reg_not_init(env
, &callee
->regs
[BPF_REG_5
]);
5929 callee
->in_async_callback_fn
= true;
5933 static int prepare_func_exit(struct bpf_verifier_env
*env
, int *insn_idx
)
5935 struct bpf_verifier_state
*state
= env
->cur_state
;
5936 struct bpf_func_state
*caller
, *callee
;
5937 struct bpf_reg_state
*r0
;
5940 callee
= state
->frame
[state
->curframe
];
5941 r0
= &callee
->regs
[BPF_REG_0
];
5942 if (r0
->type
== PTR_TO_STACK
) {
5943 /* technically it's ok to return caller's stack pointer
5944 * (or caller's caller's pointer) back to the caller,
5945 * since these pointers are valid. Only current stack
5946 * pointer will be invalid as soon as function exits,
5947 * but let's be conservative
5949 verbose(env
, "cannot return stack pointer to the caller\n");
5954 caller
= state
->frame
[state
->curframe
];
5955 if (callee
->in_callback_fn
) {
5956 /* enforce R0 return value range [0, 1]. */
5957 struct tnum range
= tnum_range(0, 1);
5959 if (r0
->type
!= SCALAR_VALUE
) {
5960 verbose(env
, "R0 not a scalar value\n");
5963 if (!tnum_in(range
, r0
->var_off
)) {
5964 verbose_invalid_scalar(env
, r0
, &range
, "callback return", "R0");
5968 /* return to the caller whatever r0 had in the callee */
5969 caller
->regs
[BPF_REG_0
] = *r0
;
5972 /* Transfer references to the caller */
5973 err
= copy_reference_state(caller
, callee
);
5977 *insn_idx
= callee
->callsite
+ 1;
5978 if (env
->log
.level
& BPF_LOG_LEVEL
) {
5979 verbose(env
, "returning from callee:\n");
5980 print_verifier_state(env
, callee
);
5981 verbose(env
, "to caller at %d:\n", *insn_idx
);
5982 print_verifier_state(env
, caller
);
5984 /* clear everything in the callee */
5985 free_func_state(callee
);
5986 state
->frame
[state
->curframe
+ 1] = NULL
;
5990 static void do_refine_retval_range(struct bpf_reg_state
*regs
, int ret_type
,
5992 struct bpf_call_arg_meta
*meta
)
5994 struct bpf_reg_state
*ret_reg
= ®s
[BPF_REG_0
];
5996 if (ret_type
!= RET_INTEGER
||
5997 (func_id
!= BPF_FUNC_get_stack
&&
5998 func_id
!= BPF_FUNC_get_task_stack
&&
5999 func_id
!= BPF_FUNC_probe_read_str
&&
6000 func_id
!= BPF_FUNC_probe_read_kernel_str
&&
6001 func_id
!= BPF_FUNC_probe_read_user_str
))
6004 ret_reg
->smax_value
= meta
->msize_max_value
;
6005 ret_reg
->s32_max_value
= meta
->msize_max_value
;
6006 ret_reg
->smin_value
= -MAX_ERRNO
;
6007 ret_reg
->s32_min_value
= -MAX_ERRNO
;
6008 __reg_deduce_bounds(ret_reg
);
6009 __reg_bound_offset(ret_reg
);
6010 __update_reg_bounds(ret_reg
);
6014 record_func_map(struct bpf_verifier_env
*env
, struct bpf_call_arg_meta
*meta
,
6015 int func_id
, int insn_idx
)
6017 struct bpf_insn_aux_data
*aux
= &env
->insn_aux_data
[insn_idx
];
6018 struct bpf_map
*map
= meta
->map_ptr
;
6020 if (func_id
!= BPF_FUNC_tail_call
&&
6021 func_id
!= BPF_FUNC_map_lookup_elem
&&
6022 func_id
!= BPF_FUNC_map_update_elem
&&
6023 func_id
!= BPF_FUNC_map_delete_elem
&&
6024 func_id
!= BPF_FUNC_map_push_elem
&&
6025 func_id
!= BPF_FUNC_map_pop_elem
&&
6026 func_id
!= BPF_FUNC_map_peek_elem
&&
6027 func_id
!= BPF_FUNC_for_each_map_elem
&&
6028 func_id
!= BPF_FUNC_redirect_map
)
6032 verbose(env
, "kernel subsystem misconfigured verifier\n");
6036 /* In case of read-only, some additional restrictions
6037 * need to be applied in order to prevent altering the
6038 * state of the map from program side.
6040 if ((map
->map_flags
& BPF_F_RDONLY_PROG
) &&
6041 (func_id
== BPF_FUNC_map_delete_elem
||
6042 func_id
== BPF_FUNC_map_update_elem
||
6043 func_id
== BPF_FUNC_map_push_elem
||
6044 func_id
== BPF_FUNC_map_pop_elem
)) {
6045 verbose(env
, "write into map forbidden\n");
6049 if (!BPF_MAP_PTR(aux
->map_ptr_state
))
6050 bpf_map_ptr_store(aux
, meta
->map_ptr
,
6051 !meta
->map_ptr
->bypass_spec_v1
);
6052 else if (BPF_MAP_PTR(aux
->map_ptr_state
) != meta
->map_ptr
)
6053 bpf_map_ptr_store(aux
, BPF_MAP_PTR_POISON
,
6054 !meta
->map_ptr
->bypass_spec_v1
);
6059 record_func_key(struct bpf_verifier_env
*env
, struct bpf_call_arg_meta
*meta
,
6060 int func_id
, int insn_idx
)
6062 struct bpf_insn_aux_data
*aux
= &env
->insn_aux_data
[insn_idx
];
6063 struct bpf_reg_state
*regs
= cur_regs(env
), *reg
;
6064 struct bpf_map
*map
= meta
->map_ptr
;
6069 if (func_id
!= BPF_FUNC_tail_call
)
6071 if (!map
|| map
->map_type
!= BPF_MAP_TYPE_PROG_ARRAY
) {
6072 verbose(env
, "kernel subsystem misconfigured verifier\n");
6076 range
= tnum_range(0, map
->max_entries
- 1);
6077 reg
= ®s
[BPF_REG_3
];
6079 if (!register_is_const(reg
) || !tnum_in(range
, reg
->var_off
)) {
6080 bpf_map_key_store(aux
, BPF_MAP_KEY_POISON
);
6084 err
= mark_chain_precision(env
, BPF_REG_3
);
6088 val
= reg
->var_off
.value
;
6089 if (bpf_map_key_unseen(aux
))
6090 bpf_map_key_store(aux
, val
);
6091 else if (!bpf_map_key_poisoned(aux
) &&
6092 bpf_map_key_immediate(aux
) != val
)
6093 bpf_map_key_store(aux
, BPF_MAP_KEY_POISON
);
6097 static int check_reference_leak(struct bpf_verifier_env
*env
)
6099 struct bpf_func_state
*state
= cur_func(env
);
6102 for (i
= 0; i
< state
->acquired_refs
; i
++) {
6103 verbose(env
, "Unreleased reference id=%d alloc_insn=%d\n",
6104 state
->refs
[i
].id
, state
->refs
[i
].insn_idx
);
6106 return state
->acquired_refs
? -EINVAL
: 0;
6109 static int check_bpf_snprintf_call(struct bpf_verifier_env
*env
,
6110 struct bpf_reg_state
*regs
)
6112 struct bpf_reg_state
*fmt_reg
= ®s
[BPF_REG_3
];
6113 struct bpf_reg_state
*data_len_reg
= ®s
[BPF_REG_5
];
6114 struct bpf_map
*fmt_map
= fmt_reg
->map_ptr
;
6115 int err
, fmt_map_off
, num_args
;
6119 /* data must be an array of u64 */
6120 if (data_len_reg
->var_off
.value
% 8)
6122 num_args
= data_len_reg
->var_off
.value
/ 8;
6124 /* fmt being ARG_PTR_TO_CONST_STR guarantees that var_off is const
6125 * and map_direct_value_addr is set.
6127 fmt_map_off
= fmt_reg
->off
+ fmt_reg
->var_off
.value
;
6128 err
= fmt_map
->ops
->map_direct_value_addr(fmt_map
, &fmt_addr
,
6131 verbose(env
, "verifier bug\n");
6134 fmt
= (char *)(long)fmt_addr
+ fmt_map_off
;
6136 /* We are also guaranteed that fmt+fmt_map_off is NULL terminated, we
6137 * can focus on validating the format specifiers.
6139 err
= bpf_bprintf_prepare(fmt
, UINT_MAX
, NULL
, NULL
, num_args
);
6141 verbose(env
, "Invalid format string\n");
6146 static int check_get_func_ip(struct bpf_verifier_env
*env
)
6148 enum bpf_attach_type eatype
= env
->prog
->expected_attach_type
;
6149 enum bpf_prog_type type
= resolve_prog_type(env
->prog
);
6150 int func_id
= BPF_FUNC_get_func_ip
;
6152 if (type
== BPF_PROG_TYPE_TRACING
) {
6153 if (eatype
!= BPF_TRACE_FENTRY
&& eatype
!= BPF_TRACE_FEXIT
&&
6154 eatype
!= BPF_MODIFY_RETURN
) {
6155 verbose(env
, "func %s#%d supported only for fentry/fexit/fmod_ret programs\n",
6156 func_id_name(func_id
), func_id
);
6160 } else if (type
== BPF_PROG_TYPE_KPROBE
) {
6164 verbose(env
, "func %s#%d not supported for program type %d\n",
6165 func_id_name(func_id
), func_id
, type
);
6169 static int check_helper_call(struct bpf_verifier_env
*env
, struct bpf_insn
*insn
,
6172 const struct bpf_func_proto
*fn
= NULL
;
6173 struct bpf_reg_state
*regs
;
6174 struct bpf_call_arg_meta meta
;
6175 int insn_idx
= *insn_idx_p
;
6177 int i
, err
, func_id
;
6179 /* find function prototype */
6180 func_id
= insn
->imm
;
6181 if (func_id
< 0 || func_id
>= __BPF_FUNC_MAX_ID
) {
6182 verbose(env
, "invalid func %s#%d\n", func_id_name(func_id
),
6187 if (env
->ops
->get_func_proto
)
6188 fn
= env
->ops
->get_func_proto(func_id
, env
->prog
);
6190 verbose(env
, "unknown func %s#%d\n", func_id_name(func_id
),
6195 /* eBPF programs must be GPL compatible to use GPL-ed functions */
6196 if (!env
->prog
->gpl_compatible
&& fn
->gpl_only
) {
6197 verbose(env
, "cannot call GPL-restricted function from non-GPL compatible program\n");
6201 if (fn
->allowed
&& !fn
->allowed(env
->prog
)) {
6202 verbose(env
, "helper call is not allowed in probe\n");
6206 /* With LD_ABS/IND some JITs save/restore skb from r1. */
6207 changes_data
= bpf_helper_changes_pkt_data(fn
->func
);
6208 if (changes_data
&& fn
->arg1_type
!= ARG_PTR_TO_CTX
) {
6209 verbose(env
, "kernel subsystem misconfigured func %s#%d: r1 != ctx\n",
6210 func_id_name(func_id
), func_id
);
6214 memset(&meta
, 0, sizeof(meta
));
6215 meta
.pkt_access
= fn
->pkt_access
;
6217 err
= check_func_proto(fn
, func_id
);
6219 verbose(env
, "kernel subsystem misconfigured func %s#%d\n",
6220 func_id_name(func_id
), func_id
);
6224 meta
.func_id
= func_id
;
6226 for (i
= 0; i
< MAX_BPF_FUNC_REG_ARGS
; i
++) {
6227 err
= check_func_arg(env
, i
, &meta
, fn
);
6232 err
= record_func_map(env
, &meta
, func_id
, insn_idx
);
6236 err
= record_func_key(env
, &meta
, func_id
, insn_idx
);
6240 /* Mark slots with STACK_MISC in case of raw mode, stack offset
6241 * is inferred from register state.
6243 for (i
= 0; i
< meta
.access_size
; i
++) {
6244 err
= check_mem_access(env
, insn_idx
, meta
.regno
, i
, BPF_B
,
6245 BPF_WRITE
, -1, false);
6250 if (func_id
== BPF_FUNC_tail_call
) {
6251 err
= check_reference_leak(env
);
6253 verbose(env
, "tail_call would lead to reference leak\n");
6256 } else if (is_release_function(func_id
)) {
6257 err
= release_reference(env
, meta
.ref_obj_id
);
6259 verbose(env
, "func %s#%d reference has not been acquired before\n",
6260 func_id_name(func_id
), func_id
);
6265 regs
= cur_regs(env
);
6267 /* check that flags argument in get_local_storage(map, flags) is 0,
6268 * this is required because get_local_storage() can't return an error.
6270 if (func_id
== BPF_FUNC_get_local_storage
&&
6271 !register_is_null(®s
[BPF_REG_2
])) {
6272 verbose(env
, "get_local_storage() doesn't support non-zero flags\n");
6276 if (func_id
== BPF_FUNC_for_each_map_elem
) {
6277 err
= __check_func_call(env
, insn
, insn_idx_p
, meta
.subprogno
,
6278 set_map_elem_callback_state
);
6283 if (func_id
== BPF_FUNC_timer_set_callback
) {
6284 err
= __check_func_call(env
, insn
, insn_idx_p
, meta
.subprogno
,
6285 set_timer_callback_state
);
6290 if (func_id
== BPF_FUNC_snprintf
) {
6291 err
= check_bpf_snprintf_call(env
, regs
);
6296 /* reset caller saved regs */
6297 for (i
= 0; i
< CALLER_SAVED_REGS
; i
++) {
6298 mark_reg_not_init(env
, regs
, caller_saved
[i
]);
6299 check_reg_arg(env
, caller_saved
[i
], DST_OP_NO_MARK
);
6302 /* helper call returns 64-bit value. */
6303 regs
[BPF_REG_0
].subreg_def
= DEF_NOT_SUBREG
;
6305 /* update return register (already marked as written above) */
6306 if (fn
->ret_type
== RET_INTEGER
) {
6307 /* sets type to SCALAR_VALUE */
6308 mark_reg_unknown(env
, regs
, BPF_REG_0
);
6309 } else if (fn
->ret_type
== RET_VOID
) {
6310 regs
[BPF_REG_0
].type
= NOT_INIT
;
6311 } else if (fn
->ret_type
== RET_PTR_TO_MAP_VALUE_OR_NULL
||
6312 fn
->ret_type
== RET_PTR_TO_MAP_VALUE
) {
6313 /* There is no offset yet applied, variable or fixed */
6314 mark_reg_known_zero(env
, regs
, BPF_REG_0
);
6315 /* remember map_ptr, so that check_map_access()
6316 * can check 'value_size' boundary of memory access
6317 * to map element returned from bpf_map_lookup_elem()
6319 if (meta
.map_ptr
== NULL
) {
6321 "kernel subsystem misconfigured verifier\n");
6324 regs
[BPF_REG_0
].map_ptr
= meta
.map_ptr
;
6325 regs
[BPF_REG_0
].map_uid
= meta
.map_uid
;
6326 if (fn
->ret_type
== RET_PTR_TO_MAP_VALUE
) {
6327 regs
[BPF_REG_0
].type
= PTR_TO_MAP_VALUE
;
6328 if (map_value_has_spin_lock(meta
.map_ptr
))
6329 regs
[BPF_REG_0
].id
= ++env
->id_gen
;
6331 regs
[BPF_REG_0
].type
= PTR_TO_MAP_VALUE_OR_NULL
;
6333 } else if (fn
->ret_type
== RET_PTR_TO_SOCKET_OR_NULL
) {
6334 mark_reg_known_zero(env
, regs
, BPF_REG_0
);
6335 regs
[BPF_REG_0
].type
= PTR_TO_SOCKET_OR_NULL
;
6336 } else if (fn
->ret_type
== RET_PTR_TO_SOCK_COMMON_OR_NULL
) {
6337 mark_reg_known_zero(env
, regs
, BPF_REG_0
);
6338 regs
[BPF_REG_0
].type
= PTR_TO_SOCK_COMMON_OR_NULL
;
6339 } else if (fn
->ret_type
== RET_PTR_TO_TCP_SOCK_OR_NULL
) {
6340 mark_reg_known_zero(env
, regs
, BPF_REG_0
);
6341 regs
[BPF_REG_0
].type
= PTR_TO_TCP_SOCK_OR_NULL
;
6342 } else if (fn
->ret_type
== RET_PTR_TO_ALLOC_MEM_OR_NULL
) {
6343 mark_reg_known_zero(env
, regs
, BPF_REG_0
);
6344 regs
[BPF_REG_0
].type
= PTR_TO_MEM_OR_NULL
;
6345 regs
[BPF_REG_0
].mem_size
= meta
.mem_size
;
6346 } else if (fn
->ret_type
== RET_PTR_TO_MEM_OR_BTF_ID_OR_NULL
||
6347 fn
->ret_type
== RET_PTR_TO_MEM_OR_BTF_ID
) {
6348 const struct btf_type
*t
;
6350 mark_reg_known_zero(env
, regs
, BPF_REG_0
);
6351 t
= btf_type_skip_modifiers(meta
.ret_btf
, meta
.ret_btf_id
, NULL
);
6352 if (!btf_type_is_struct(t
)) {
6354 const struct btf_type
*ret
;
6357 /* resolve the type size of ksym. */
6358 ret
= btf_resolve_size(meta
.ret_btf
, t
, &tsize
);
6360 tname
= btf_name_by_offset(meta
.ret_btf
, t
->name_off
);
6361 verbose(env
, "unable to resolve the size of type '%s': %ld\n",
6362 tname
, PTR_ERR(ret
));
6365 regs
[BPF_REG_0
].type
=
6366 fn
->ret_type
== RET_PTR_TO_MEM_OR_BTF_ID
?
6367 PTR_TO_MEM
: PTR_TO_MEM_OR_NULL
;
6368 regs
[BPF_REG_0
].mem_size
= tsize
;
6370 regs
[BPF_REG_0
].type
=
6371 fn
->ret_type
== RET_PTR_TO_MEM_OR_BTF_ID
?
6372 PTR_TO_BTF_ID
: PTR_TO_BTF_ID_OR_NULL
;
6373 regs
[BPF_REG_0
].btf
= meta
.ret_btf
;
6374 regs
[BPF_REG_0
].btf_id
= meta
.ret_btf_id
;
6376 } else if (fn
->ret_type
== RET_PTR_TO_BTF_ID_OR_NULL
||
6377 fn
->ret_type
== RET_PTR_TO_BTF_ID
) {
6380 mark_reg_known_zero(env
, regs
, BPF_REG_0
);
6381 regs
[BPF_REG_0
].type
= fn
->ret_type
== RET_PTR_TO_BTF_ID
?
6383 PTR_TO_BTF_ID_OR_NULL
;
6384 ret_btf_id
= *fn
->ret_btf_id
;
6385 if (ret_btf_id
== 0) {
6386 verbose(env
, "invalid return type %d of func %s#%d\n",
6387 fn
->ret_type
, func_id_name(func_id
), func_id
);
6390 /* current BPF helper definitions are only coming from
6391 * built-in code with type IDs from vmlinux BTF
6393 regs
[BPF_REG_0
].btf
= btf_vmlinux
;
6394 regs
[BPF_REG_0
].btf_id
= ret_btf_id
;
6396 verbose(env
, "unknown return type %d of func %s#%d\n",
6397 fn
->ret_type
, func_id_name(func_id
), func_id
);
6401 if (reg_type_may_be_null(regs
[BPF_REG_0
].type
))
6402 regs
[BPF_REG_0
].id
= ++env
->id_gen
;
6404 if (is_ptr_cast_function(func_id
)) {
6405 /* For release_reference() */
6406 regs
[BPF_REG_0
].ref_obj_id
= meta
.ref_obj_id
;
6407 } else if (is_acquire_function(func_id
, meta
.map_ptr
)) {
6408 int id
= acquire_reference_state(env
, insn_idx
);
6412 /* For mark_ptr_or_null_reg() */
6413 regs
[BPF_REG_0
].id
= id
;
6414 /* For release_reference() */
6415 regs
[BPF_REG_0
].ref_obj_id
= id
;
6418 do_refine_retval_range(regs
, fn
->ret_type
, func_id
, &meta
);
6420 err
= check_map_func_compatibility(env
, meta
.map_ptr
, func_id
);
6424 if ((func_id
== BPF_FUNC_get_stack
||
6425 func_id
== BPF_FUNC_get_task_stack
) &&
6426 !env
->prog
->has_callchain_buf
) {
6427 const char *err_str
;
6429 #ifdef CONFIG_PERF_EVENTS
6430 err
= get_callchain_buffers(sysctl_perf_event_max_stack
);
6431 err_str
= "cannot get callchain buffer for func %s#%d\n";
6434 err_str
= "func %s#%d not supported without CONFIG_PERF_EVENTS\n";
6437 verbose(env
, err_str
, func_id_name(func_id
), func_id
);
6441 env
->prog
->has_callchain_buf
= true;
6444 if (func_id
== BPF_FUNC_get_stackid
|| func_id
== BPF_FUNC_get_stack
)
6445 env
->prog
->call_get_stack
= true;
6447 if (func_id
== BPF_FUNC_get_func_ip
) {
6448 if (check_get_func_ip(env
))
6450 env
->prog
->call_get_func_ip
= true;
6454 clear_all_pkt_pointers(env
);
6458 /* mark_btf_func_reg_size() is used when the reg size is determined by
6459 * the BTF func_proto's return value size and argument.
6461 static void mark_btf_func_reg_size(struct bpf_verifier_env
*env
, u32 regno
,
6464 struct bpf_reg_state
*reg
= &cur_regs(env
)[regno
];
6466 if (regno
== BPF_REG_0
) {
6467 /* Function return value */
6468 reg
->live
|= REG_LIVE_WRITTEN
;
6469 reg
->subreg_def
= reg_size
== sizeof(u64
) ?
6470 DEF_NOT_SUBREG
: env
->insn_idx
+ 1;
6472 /* Function argument */
6473 if (reg_size
== sizeof(u64
)) {
6474 mark_insn_zext(env
, reg
);
6475 mark_reg_read(env
, reg
, reg
->parent
, REG_LIVE_READ64
);
6477 mark_reg_read(env
, reg
, reg
->parent
, REG_LIVE_READ32
);
6482 static int check_kfunc_call(struct bpf_verifier_env
*env
, struct bpf_insn
*insn
)
6484 const struct btf_type
*t
, *func
, *func_proto
, *ptr_type
;
6485 struct bpf_reg_state
*regs
= cur_regs(env
);
6486 const char *func_name
, *ptr_type_name
;
6487 u32 i
, nargs
, func_id
, ptr_type_id
;
6488 const struct btf_param
*args
;
6491 func_id
= insn
->imm
;
6492 func
= btf_type_by_id(btf_vmlinux
, func_id
);
6493 func_name
= btf_name_by_offset(btf_vmlinux
, func
->name_off
);
6494 func_proto
= btf_type_by_id(btf_vmlinux
, func
->type
);
6496 if (!env
->ops
->check_kfunc_call
||
6497 !env
->ops
->check_kfunc_call(func_id
)) {
6498 verbose(env
, "calling kernel function %s is not allowed\n",
6503 /* Check the arguments */
6504 err
= btf_check_kfunc_arg_match(env
, btf_vmlinux
, func_id
, regs
);
6508 for (i
= 0; i
< CALLER_SAVED_REGS
; i
++)
6509 mark_reg_not_init(env
, regs
, caller_saved
[i
]);
6511 /* Check return type */
6512 t
= btf_type_skip_modifiers(btf_vmlinux
, func_proto
->type
, NULL
);
6513 if (btf_type_is_scalar(t
)) {
6514 mark_reg_unknown(env
, regs
, BPF_REG_0
);
6515 mark_btf_func_reg_size(env
, BPF_REG_0
, t
->size
);
6516 } else if (btf_type_is_ptr(t
)) {
6517 ptr_type
= btf_type_skip_modifiers(btf_vmlinux
, t
->type
,
6519 if (!btf_type_is_struct(ptr_type
)) {
6520 ptr_type_name
= btf_name_by_offset(btf_vmlinux
,
6521 ptr_type
->name_off
);
6522 verbose(env
, "kernel function %s returns pointer type %s %s is not supported\n",
6523 func_name
, btf_type_str(ptr_type
),
6527 mark_reg_known_zero(env
, regs
, BPF_REG_0
);
6528 regs
[BPF_REG_0
].btf
= btf_vmlinux
;
6529 regs
[BPF_REG_0
].type
= PTR_TO_BTF_ID
;
6530 regs
[BPF_REG_0
].btf_id
= ptr_type_id
;
6531 mark_btf_func_reg_size(env
, BPF_REG_0
, sizeof(void *));
6532 } /* else { add_kfunc_call() ensures it is btf_type_is_void(t) } */
6534 nargs
= btf_type_vlen(func_proto
);
6535 args
= (const struct btf_param
*)(func_proto
+ 1);
6536 for (i
= 0; i
< nargs
; i
++) {
6539 t
= btf_type_skip_modifiers(btf_vmlinux
, args
[i
].type
, NULL
);
6540 if (btf_type_is_ptr(t
))
6541 mark_btf_func_reg_size(env
, regno
, sizeof(void *));
6543 /* scalar. ensured by btf_check_kfunc_arg_match() */
6544 mark_btf_func_reg_size(env
, regno
, t
->size
);
6550 static bool signed_add_overflows(s64 a
, s64 b
)
6552 /* Do the add in u64, where overflow is well-defined */
6553 s64 res
= (s64
)((u64
)a
+ (u64
)b
);
6560 static bool signed_add32_overflows(s32 a
, s32 b
)
6562 /* Do the add in u32, where overflow is well-defined */
6563 s32 res
= (s32
)((u32
)a
+ (u32
)b
);
6570 static bool signed_sub_overflows(s64 a
, s64 b
)
6572 /* Do the sub in u64, where overflow is well-defined */
6573 s64 res
= (s64
)((u64
)a
- (u64
)b
);
6580 static bool signed_sub32_overflows(s32 a
, s32 b
)
6582 /* Do the sub in u32, where overflow is well-defined */
6583 s32 res
= (s32
)((u32
)a
- (u32
)b
);
6590 static bool check_reg_sane_offset(struct bpf_verifier_env
*env
,
6591 const struct bpf_reg_state
*reg
,
6592 enum bpf_reg_type type
)
6594 bool known
= tnum_is_const(reg
->var_off
);
6595 s64 val
= reg
->var_off
.value
;
6596 s64 smin
= reg
->smin_value
;
6598 if (known
&& (val
>= BPF_MAX_VAR_OFF
|| val
<= -BPF_MAX_VAR_OFF
)) {
6599 verbose(env
, "math between %s pointer and %lld is not allowed\n",
6600 reg_type_str
[type
], val
);
6604 if (reg
->off
>= BPF_MAX_VAR_OFF
|| reg
->off
<= -BPF_MAX_VAR_OFF
) {
6605 verbose(env
, "%s pointer offset %d is not allowed\n",
6606 reg_type_str
[type
], reg
->off
);
6610 if (smin
== S64_MIN
) {
6611 verbose(env
, "math between %s pointer and register with unbounded min value is not allowed\n",
6612 reg_type_str
[type
]);
6616 if (smin
>= BPF_MAX_VAR_OFF
|| smin
<= -BPF_MAX_VAR_OFF
) {
6617 verbose(env
, "value %lld makes %s pointer be out of bounds\n",
6618 smin
, reg_type_str
[type
]);
6625 static struct bpf_insn_aux_data
*cur_aux(struct bpf_verifier_env
*env
)
6627 return &env
->insn_aux_data
[env
->insn_idx
];
6638 static int retrieve_ptr_limit(const struct bpf_reg_state
*ptr_reg
,
6639 u32
*alu_limit
, bool mask_to_left
)
6641 u32 max
= 0, ptr_limit
= 0;
6643 switch (ptr_reg
->type
) {
6645 /* Offset 0 is out-of-bounds, but acceptable start for the
6646 * left direction, see BPF_REG_FP. Also, unknown scalar
6647 * offset where we would need to deal with min/max bounds is
6648 * currently prohibited for unprivileged.
6650 max
= MAX_BPF_STACK
+ mask_to_left
;
6651 ptr_limit
= -(ptr_reg
->var_off
.value
+ ptr_reg
->off
);
6653 case PTR_TO_MAP_VALUE
:
6654 max
= ptr_reg
->map_ptr
->value_size
;
6655 ptr_limit
= (mask_to_left
?
6656 ptr_reg
->smin_value
:
6657 ptr_reg
->umax_value
) + ptr_reg
->off
;
6663 if (ptr_limit
>= max
)
6664 return REASON_LIMIT
;
6665 *alu_limit
= ptr_limit
;
6669 static bool can_skip_alu_sanitation(const struct bpf_verifier_env
*env
,
6670 const struct bpf_insn
*insn
)
6672 return env
->bypass_spec_v1
|| BPF_SRC(insn
->code
) == BPF_K
;
6675 static int update_alu_sanitation_state(struct bpf_insn_aux_data
*aux
,
6676 u32 alu_state
, u32 alu_limit
)
6678 /* If we arrived here from different branches with different
6679 * state or limits to sanitize, then this won't work.
6681 if (aux
->alu_state
&&
6682 (aux
->alu_state
!= alu_state
||
6683 aux
->alu_limit
!= alu_limit
))
6684 return REASON_PATHS
;
6686 /* Corresponding fixup done in do_misc_fixups(). */
6687 aux
->alu_state
= alu_state
;
6688 aux
->alu_limit
= alu_limit
;
6692 static int sanitize_val_alu(struct bpf_verifier_env
*env
,
6693 struct bpf_insn
*insn
)
6695 struct bpf_insn_aux_data
*aux
= cur_aux(env
);
6697 if (can_skip_alu_sanitation(env
, insn
))
6700 return update_alu_sanitation_state(aux
, BPF_ALU_NON_POINTER
, 0);
6703 static bool sanitize_needed(u8 opcode
)
6705 return opcode
== BPF_ADD
|| opcode
== BPF_SUB
;
6708 struct bpf_sanitize_info
{
6709 struct bpf_insn_aux_data aux
;
6713 static struct bpf_verifier_state
*
6714 sanitize_speculative_path(struct bpf_verifier_env
*env
,
6715 const struct bpf_insn
*insn
,
6716 u32 next_idx
, u32 curr_idx
)
6718 struct bpf_verifier_state
*branch
;
6719 struct bpf_reg_state
*regs
;
6721 branch
= push_stack(env
, next_idx
, curr_idx
, true);
6722 if (branch
&& insn
) {
6723 regs
= branch
->frame
[branch
->curframe
]->regs
;
6724 if (BPF_SRC(insn
->code
) == BPF_K
) {
6725 mark_reg_unknown(env
, regs
, insn
->dst_reg
);
6726 } else if (BPF_SRC(insn
->code
) == BPF_X
) {
6727 mark_reg_unknown(env
, regs
, insn
->dst_reg
);
6728 mark_reg_unknown(env
, regs
, insn
->src_reg
);
6734 static int sanitize_ptr_alu(struct bpf_verifier_env
*env
,
6735 struct bpf_insn
*insn
,
6736 const struct bpf_reg_state
*ptr_reg
,
6737 const struct bpf_reg_state
*off_reg
,
6738 struct bpf_reg_state
*dst_reg
,
6739 struct bpf_sanitize_info
*info
,
6740 const bool commit_window
)
6742 struct bpf_insn_aux_data
*aux
= commit_window
? cur_aux(env
) : &info
->aux
;
6743 struct bpf_verifier_state
*vstate
= env
->cur_state
;
6744 bool off_is_imm
= tnum_is_const(off_reg
->var_off
);
6745 bool off_is_neg
= off_reg
->smin_value
< 0;
6746 bool ptr_is_dst_reg
= ptr_reg
== dst_reg
;
6747 u8 opcode
= BPF_OP(insn
->code
);
6748 u32 alu_state
, alu_limit
;
6749 struct bpf_reg_state tmp
;
6753 if (can_skip_alu_sanitation(env
, insn
))
6756 /* We already marked aux for masking from non-speculative
6757 * paths, thus we got here in the first place. We only care
6758 * to explore bad access from here.
6760 if (vstate
->speculative
)
6763 if (!commit_window
) {
6764 if (!tnum_is_const(off_reg
->var_off
) &&
6765 (off_reg
->smin_value
< 0) != (off_reg
->smax_value
< 0))
6766 return REASON_BOUNDS
;
6768 info
->mask_to_left
= (opcode
== BPF_ADD
&& off_is_neg
) ||
6769 (opcode
== BPF_SUB
&& !off_is_neg
);
6772 err
= retrieve_ptr_limit(ptr_reg
, &alu_limit
, info
->mask_to_left
);
6776 if (commit_window
) {
6777 /* In commit phase we narrow the masking window based on
6778 * the observed pointer move after the simulated operation.
6780 alu_state
= info
->aux
.alu_state
;
6781 alu_limit
= abs(info
->aux
.alu_limit
- alu_limit
);
6783 alu_state
= off_is_neg
? BPF_ALU_NEG_VALUE
: 0;
6784 alu_state
|= off_is_imm
? BPF_ALU_IMMEDIATE
: 0;
6785 alu_state
|= ptr_is_dst_reg
?
6786 BPF_ALU_SANITIZE_SRC
: BPF_ALU_SANITIZE_DST
;
6788 /* Limit pruning on unknown scalars to enable deep search for
6789 * potential masking differences from other program paths.
6792 env
->explore_alu_limits
= true;
6795 err
= update_alu_sanitation_state(aux
, alu_state
, alu_limit
);
6799 /* If we're in commit phase, we're done here given we already
6800 * pushed the truncated dst_reg into the speculative verification
6803 * Also, when register is a known constant, we rewrite register-based
6804 * operation to immediate-based, and thus do not need masking (and as
6805 * a consequence, do not need to simulate the zero-truncation either).
6807 if (commit_window
|| off_is_imm
)
6810 /* Simulate and find potential out-of-bounds access under
6811 * speculative execution from truncation as a result of
6812 * masking when off was not within expected range. If off
6813 * sits in dst, then we temporarily need to move ptr there
6814 * to simulate dst (== 0) +/-= ptr. Needed, for example,
6815 * for cases where we use K-based arithmetic in one direction
6816 * and truncated reg-based in the other in order to explore
6819 if (!ptr_is_dst_reg
) {
6821 *dst_reg
= *ptr_reg
;
6823 ret
= sanitize_speculative_path(env
, NULL
, env
->insn_idx
+ 1,
6825 if (!ptr_is_dst_reg
&& ret
)
6827 return !ret
? REASON_STACK
: 0;
6830 static void sanitize_mark_insn_seen(struct bpf_verifier_env
*env
)
6832 struct bpf_verifier_state
*vstate
= env
->cur_state
;
6834 /* If we simulate paths under speculation, we don't update the
6835 * insn as 'seen' such that when we verify unreachable paths in
6836 * the non-speculative domain, sanitize_dead_code() can still
6837 * rewrite/sanitize them.
6839 if (!vstate
->speculative
)
6840 env
->insn_aux_data
[env
->insn_idx
].seen
= env
->pass_cnt
;
6843 static int sanitize_err(struct bpf_verifier_env
*env
,
6844 const struct bpf_insn
*insn
, int reason
,
6845 const struct bpf_reg_state
*off_reg
,
6846 const struct bpf_reg_state
*dst_reg
)
6848 static const char *err
= "pointer arithmetic with it prohibited for !root";
6849 const char *op
= BPF_OP(insn
->code
) == BPF_ADD
? "add" : "sub";
6850 u32 dst
= insn
->dst_reg
, src
= insn
->src_reg
;
6854 verbose(env
, "R%d has unknown scalar with mixed signed bounds, %s\n",
6855 off_reg
== dst_reg
? dst
: src
, err
);
6858 verbose(env
, "R%d has pointer with unsupported alu operation, %s\n",
6859 off_reg
== dst_reg
? src
: dst
, err
);
6862 verbose(env
, "R%d tried to %s from different maps, paths or scalars, %s\n",
6866 verbose(env
, "R%d tried to %s beyond pointer bounds, %s\n",
6870 verbose(env
, "R%d could not be pushed for speculative verification, %s\n",
6874 verbose(env
, "verifier internal error: unknown reason (%d)\n",
6882 /* check that stack access falls within stack limits and that 'reg' doesn't
6883 * have a variable offset.
6885 * Variable offset is prohibited for unprivileged mode for simplicity since it
6886 * requires corresponding support in Spectre masking for stack ALU. See also
6887 * retrieve_ptr_limit().
6890 * 'off' includes 'reg->off'.
6892 static int check_stack_access_for_ptr_arithmetic(
6893 struct bpf_verifier_env
*env
,
6895 const struct bpf_reg_state
*reg
,
6898 if (!tnum_is_const(reg
->var_off
)) {
6901 tnum_strn(tn_buf
, sizeof(tn_buf
), reg
->var_off
);
6902 verbose(env
, "R%d variable stack access prohibited for !root, var_off=%s off=%d\n",
6903 regno
, tn_buf
, off
);
6907 if (off
>= 0 || off
< -MAX_BPF_STACK
) {
6908 verbose(env
, "R%d stack pointer arithmetic goes out of range, "
6909 "prohibited for !root; off=%d\n", regno
, off
);
6916 static int sanitize_check_bounds(struct bpf_verifier_env
*env
,
6917 const struct bpf_insn
*insn
,
6918 const struct bpf_reg_state
*dst_reg
)
6920 u32 dst
= insn
->dst_reg
;
6922 /* For unprivileged we require that resulting offset must be in bounds
6923 * in order to be able to sanitize access later on.
6925 if (env
->bypass_spec_v1
)
6928 switch (dst_reg
->type
) {
6930 if (check_stack_access_for_ptr_arithmetic(env
, dst
, dst_reg
,
6931 dst_reg
->off
+ dst_reg
->var_off
.value
))
6934 case PTR_TO_MAP_VALUE
:
6935 if (check_map_access(env
, dst
, dst_reg
->off
, 1, false)) {
6936 verbose(env
, "R%d pointer arithmetic of map value goes out of range, "
6937 "prohibited for !root\n", dst
);
6948 /* Handles arithmetic on a pointer and a scalar: computes new min/max and var_off.
6949 * Caller should also handle BPF_MOV case separately.
6950 * If we return -EACCES, caller may want to try again treating pointer as a
6951 * scalar. So we only emit a diagnostic if !env->allow_ptr_leaks.
6953 static int adjust_ptr_min_max_vals(struct bpf_verifier_env
*env
,
6954 struct bpf_insn
*insn
,
6955 const struct bpf_reg_state
*ptr_reg
,
6956 const struct bpf_reg_state
*off_reg
)
6958 struct bpf_verifier_state
*vstate
= env
->cur_state
;
6959 struct bpf_func_state
*state
= vstate
->frame
[vstate
->curframe
];
6960 struct bpf_reg_state
*regs
= state
->regs
, *dst_reg
;
6961 bool known
= tnum_is_const(off_reg
->var_off
);
6962 s64 smin_val
= off_reg
->smin_value
, smax_val
= off_reg
->smax_value
,
6963 smin_ptr
= ptr_reg
->smin_value
, smax_ptr
= ptr_reg
->smax_value
;
6964 u64 umin_val
= off_reg
->umin_value
, umax_val
= off_reg
->umax_value
,
6965 umin_ptr
= ptr_reg
->umin_value
, umax_ptr
= ptr_reg
->umax_value
;
6966 struct bpf_sanitize_info info
= {};
6967 u8 opcode
= BPF_OP(insn
->code
);
6968 u32 dst
= insn
->dst_reg
;
6971 dst_reg
= ®s
[dst
];
6973 if ((known
&& (smin_val
!= smax_val
|| umin_val
!= umax_val
)) ||
6974 smin_val
> smax_val
|| umin_val
> umax_val
) {
6975 /* Taint dst register if offset had invalid bounds derived from
6976 * e.g. dead branches.
6978 __mark_reg_unknown(env
, dst_reg
);
6982 if (BPF_CLASS(insn
->code
) != BPF_ALU64
) {
6983 /* 32-bit ALU ops on pointers produce (meaningless) scalars */
6984 if (opcode
== BPF_SUB
&& env
->allow_ptr_leaks
) {
6985 __mark_reg_unknown(env
, dst_reg
);
6990 "R%d 32-bit pointer arithmetic prohibited\n",
6995 switch (ptr_reg
->type
) {
6996 case PTR_TO_MAP_VALUE_OR_NULL
:
6997 verbose(env
, "R%d pointer arithmetic on %s prohibited, null-check it first\n",
6998 dst
, reg_type_str
[ptr_reg
->type
]);
7000 case CONST_PTR_TO_MAP
:
7001 /* smin_val represents the known value */
7002 if (known
&& smin_val
== 0 && opcode
== BPF_ADD
)
7005 case PTR_TO_PACKET_END
:
7007 case PTR_TO_SOCKET_OR_NULL
:
7008 case PTR_TO_SOCK_COMMON
:
7009 case PTR_TO_SOCK_COMMON_OR_NULL
:
7010 case PTR_TO_TCP_SOCK
:
7011 case PTR_TO_TCP_SOCK_OR_NULL
:
7012 case PTR_TO_XDP_SOCK
:
7013 verbose(env
, "R%d pointer arithmetic on %s prohibited\n",
7014 dst
, reg_type_str
[ptr_reg
->type
]);
7020 /* In case of 'scalar += pointer', dst_reg inherits pointer type and id.
7021 * The id may be overwritten later if we create a new variable offset.
7023 dst_reg
->type
= ptr_reg
->type
;
7024 dst_reg
->id
= ptr_reg
->id
;
7026 if (!check_reg_sane_offset(env
, off_reg
, ptr_reg
->type
) ||
7027 !check_reg_sane_offset(env
, ptr_reg
, ptr_reg
->type
))
7030 /* pointer types do not carry 32-bit bounds at the moment. */
7031 __mark_reg32_unbounded(dst_reg
);
7033 if (sanitize_needed(opcode
)) {
7034 ret
= sanitize_ptr_alu(env
, insn
, ptr_reg
, off_reg
, dst_reg
,
7037 return sanitize_err(env
, insn
, ret
, off_reg
, dst_reg
);
7042 /* We can take a fixed offset as long as it doesn't overflow
7043 * the s32 'off' field
7045 if (known
&& (ptr_reg
->off
+ smin_val
==
7046 (s64
)(s32
)(ptr_reg
->off
+ smin_val
))) {
7047 /* pointer += K. Accumulate it into fixed offset */
7048 dst_reg
->smin_value
= smin_ptr
;
7049 dst_reg
->smax_value
= smax_ptr
;
7050 dst_reg
->umin_value
= umin_ptr
;
7051 dst_reg
->umax_value
= umax_ptr
;
7052 dst_reg
->var_off
= ptr_reg
->var_off
;
7053 dst_reg
->off
= ptr_reg
->off
+ smin_val
;
7054 dst_reg
->raw
= ptr_reg
->raw
;
7057 /* A new variable offset is created. Note that off_reg->off
7058 * == 0, since it's a scalar.
7059 * dst_reg gets the pointer type and since some positive
7060 * integer value was added to the pointer, give it a new 'id'
7061 * if it's a PTR_TO_PACKET.
7062 * this creates a new 'base' pointer, off_reg (variable) gets
7063 * added into the variable offset, and we copy the fixed offset
7066 if (signed_add_overflows(smin_ptr
, smin_val
) ||
7067 signed_add_overflows(smax_ptr
, smax_val
)) {
7068 dst_reg
->smin_value
= S64_MIN
;
7069 dst_reg
->smax_value
= S64_MAX
;
7071 dst_reg
->smin_value
= smin_ptr
+ smin_val
;
7072 dst_reg
->smax_value
= smax_ptr
+ smax_val
;
7074 if (umin_ptr
+ umin_val
< umin_ptr
||
7075 umax_ptr
+ umax_val
< umax_ptr
) {
7076 dst_reg
->umin_value
= 0;
7077 dst_reg
->umax_value
= U64_MAX
;
7079 dst_reg
->umin_value
= umin_ptr
+ umin_val
;
7080 dst_reg
->umax_value
= umax_ptr
+ umax_val
;
7082 dst_reg
->var_off
= tnum_add(ptr_reg
->var_off
, off_reg
->var_off
);
7083 dst_reg
->off
= ptr_reg
->off
;
7084 dst_reg
->raw
= ptr_reg
->raw
;
7085 if (reg_is_pkt_pointer(ptr_reg
)) {
7086 dst_reg
->id
= ++env
->id_gen
;
7087 /* something was added to pkt_ptr, set range to zero */
7088 memset(&dst_reg
->raw
, 0, sizeof(dst_reg
->raw
));
7092 if (dst_reg
== off_reg
) {
7093 /* scalar -= pointer. Creates an unknown scalar */
7094 verbose(env
, "R%d tried to subtract pointer from scalar\n",
7098 /* We don't allow subtraction from FP, because (according to
7099 * test_verifier.c test "invalid fp arithmetic", JITs might not
7100 * be able to deal with it.
7102 if (ptr_reg
->type
== PTR_TO_STACK
) {
7103 verbose(env
, "R%d subtraction from stack pointer prohibited\n",
7107 if (known
&& (ptr_reg
->off
- smin_val
==
7108 (s64
)(s32
)(ptr_reg
->off
- smin_val
))) {
7109 /* pointer -= K. Subtract it from fixed offset */
7110 dst_reg
->smin_value
= smin_ptr
;
7111 dst_reg
->smax_value
= smax_ptr
;
7112 dst_reg
->umin_value
= umin_ptr
;
7113 dst_reg
->umax_value
= umax_ptr
;
7114 dst_reg
->var_off
= ptr_reg
->var_off
;
7115 dst_reg
->id
= ptr_reg
->id
;
7116 dst_reg
->off
= ptr_reg
->off
- smin_val
;
7117 dst_reg
->raw
= ptr_reg
->raw
;
7120 /* A new variable offset is created. If the subtrahend is known
7121 * nonnegative, then any reg->range we had before is still good.
7123 if (signed_sub_overflows(smin_ptr
, smax_val
) ||
7124 signed_sub_overflows(smax_ptr
, smin_val
)) {
7125 /* Overflow possible, we know nothing */
7126 dst_reg
->smin_value
= S64_MIN
;
7127 dst_reg
->smax_value
= S64_MAX
;
7129 dst_reg
->smin_value
= smin_ptr
- smax_val
;
7130 dst_reg
->smax_value
= smax_ptr
- smin_val
;
7132 if (umin_ptr
< umax_val
) {
7133 /* Overflow possible, we know nothing */
7134 dst_reg
->umin_value
= 0;
7135 dst_reg
->umax_value
= U64_MAX
;
7137 /* Cannot overflow (as long as bounds are consistent) */
7138 dst_reg
->umin_value
= umin_ptr
- umax_val
;
7139 dst_reg
->umax_value
= umax_ptr
- umin_val
;
7141 dst_reg
->var_off
= tnum_sub(ptr_reg
->var_off
, off_reg
->var_off
);
7142 dst_reg
->off
= ptr_reg
->off
;
7143 dst_reg
->raw
= ptr_reg
->raw
;
7144 if (reg_is_pkt_pointer(ptr_reg
)) {
7145 dst_reg
->id
= ++env
->id_gen
;
7146 /* something was added to pkt_ptr, set range to zero */
7148 memset(&dst_reg
->raw
, 0, sizeof(dst_reg
->raw
));
7154 /* bitwise ops on pointers are troublesome, prohibit. */
7155 verbose(env
, "R%d bitwise operator %s on pointer prohibited\n",
7156 dst
, bpf_alu_string
[opcode
>> 4]);
7159 /* other operators (e.g. MUL,LSH) produce non-pointer results */
7160 verbose(env
, "R%d pointer arithmetic with %s operator prohibited\n",
7161 dst
, bpf_alu_string
[opcode
>> 4]);
7165 if (!check_reg_sane_offset(env
, dst_reg
, ptr_reg
->type
))
7168 __update_reg_bounds(dst_reg
);
7169 __reg_deduce_bounds(dst_reg
);
7170 __reg_bound_offset(dst_reg
);
7172 if (sanitize_check_bounds(env
, insn
, dst_reg
) < 0)
7174 if (sanitize_needed(opcode
)) {
7175 ret
= sanitize_ptr_alu(env
, insn
, dst_reg
, off_reg
, dst_reg
,
7178 return sanitize_err(env
, insn
, ret
, off_reg
, dst_reg
);
7184 static void scalar32_min_max_add(struct bpf_reg_state
*dst_reg
,
7185 struct bpf_reg_state
*src_reg
)
7187 s32 smin_val
= src_reg
->s32_min_value
;
7188 s32 smax_val
= src_reg
->s32_max_value
;
7189 u32 umin_val
= src_reg
->u32_min_value
;
7190 u32 umax_val
= src_reg
->u32_max_value
;
7192 if (signed_add32_overflows(dst_reg
->s32_min_value
, smin_val
) ||
7193 signed_add32_overflows(dst_reg
->s32_max_value
, smax_val
)) {
7194 dst_reg
->s32_min_value
= S32_MIN
;
7195 dst_reg
->s32_max_value
= S32_MAX
;
7197 dst_reg
->s32_min_value
+= smin_val
;
7198 dst_reg
->s32_max_value
+= smax_val
;
7200 if (dst_reg
->u32_min_value
+ umin_val
< umin_val
||
7201 dst_reg
->u32_max_value
+ umax_val
< umax_val
) {
7202 dst_reg
->u32_min_value
= 0;
7203 dst_reg
->u32_max_value
= U32_MAX
;
7205 dst_reg
->u32_min_value
+= umin_val
;
7206 dst_reg
->u32_max_value
+= umax_val
;
7210 static void scalar_min_max_add(struct bpf_reg_state
*dst_reg
,
7211 struct bpf_reg_state
*src_reg
)
7213 s64 smin_val
= src_reg
->smin_value
;
7214 s64 smax_val
= src_reg
->smax_value
;
7215 u64 umin_val
= src_reg
->umin_value
;
7216 u64 umax_val
= src_reg
->umax_value
;
7218 if (signed_add_overflows(dst_reg
->smin_value
, smin_val
) ||
7219 signed_add_overflows(dst_reg
->smax_value
, smax_val
)) {
7220 dst_reg
->smin_value
= S64_MIN
;
7221 dst_reg
->smax_value
= S64_MAX
;
7223 dst_reg
->smin_value
+= smin_val
;
7224 dst_reg
->smax_value
+= smax_val
;
7226 if (dst_reg
->umin_value
+ umin_val
< umin_val
||
7227 dst_reg
->umax_value
+ umax_val
< umax_val
) {
7228 dst_reg
->umin_value
= 0;
7229 dst_reg
->umax_value
= U64_MAX
;
7231 dst_reg
->umin_value
+= umin_val
;
7232 dst_reg
->umax_value
+= umax_val
;
7236 static void scalar32_min_max_sub(struct bpf_reg_state
*dst_reg
,
7237 struct bpf_reg_state
*src_reg
)
7239 s32 smin_val
= src_reg
->s32_min_value
;
7240 s32 smax_val
= src_reg
->s32_max_value
;
7241 u32 umin_val
= src_reg
->u32_min_value
;
7242 u32 umax_val
= src_reg
->u32_max_value
;
7244 if (signed_sub32_overflows(dst_reg
->s32_min_value
, smax_val
) ||
7245 signed_sub32_overflows(dst_reg
->s32_max_value
, smin_val
)) {
7246 /* Overflow possible, we know nothing */
7247 dst_reg
->s32_min_value
= S32_MIN
;
7248 dst_reg
->s32_max_value
= S32_MAX
;
7250 dst_reg
->s32_min_value
-= smax_val
;
7251 dst_reg
->s32_max_value
-= smin_val
;
7253 if (dst_reg
->u32_min_value
< umax_val
) {
7254 /* Overflow possible, we know nothing */
7255 dst_reg
->u32_min_value
= 0;
7256 dst_reg
->u32_max_value
= U32_MAX
;
7258 /* Cannot overflow (as long as bounds are consistent) */
7259 dst_reg
->u32_min_value
-= umax_val
;
7260 dst_reg
->u32_max_value
-= umin_val
;
7264 static void scalar_min_max_sub(struct bpf_reg_state
*dst_reg
,
7265 struct bpf_reg_state
*src_reg
)
7267 s64 smin_val
= src_reg
->smin_value
;
7268 s64 smax_val
= src_reg
->smax_value
;
7269 u64 umin_val
= src_reg
->umin_value
;
7270 u64 umax_val
= src_reg
->umax_value
;
7272 if (signed_sub_overflows(dst_reg
->smin_value
, smax_val
) ||
7273 signed_sub_overflows(dst_reg
->smax_value
, smin_val
)) {
7274 /* Overflow possible, we know nothing */
7275 dst_reg
->smin_value
= S64_MIN
;
7276 dst_reg
->smax_value
= S64_MAX
;
7278 dst_reg
->smin_value
-= smax_val
;
7279 dst_reg
->smax_value
-= smin_val
;
7281 if (dst_reg
->umin_value
< umax_val
) {
7282 /* Overflow possible, we know nothing */
7283 dst_reg
->umin_value
= 0;
7284 dst_reg
->umax_value
= U64_MAX
;
7286 /* Cannot overflow (as long as bounds are consistent) */
7287 dst_reg
->umin_value
-= umax_val
;
7288 dst_reg
->umax_value
-= umin_val
;
7292 static void scalar32_min_max_mul(struct bpf_reg_state
*dst_reg
,
7293 struct bpf_reg_state
*src_reg
)
7295 s32 smin_val
= src_reg
->s32_min_value
;
7296 u32 umin_val
= src_reg
->u32_min_value
;
7297 u32 umax_val
= src_reg
->u32_max_value
;
7299 if (smin_val
< 0 || dst_reg
->s32_min_value
< 0) {
7300 /* Ain't nobody got time to multiply that sign */
7301 __mark_reg32_unbounded(dst_reg
);
7304 /* Both values are positive, so we can work with unsigned and
7305 * copy the result to signed (unless it exceeds S32_MAX).
7307 if (umax_val
> U16_MAX
|| dst_reg
->u32_max_value
> U16_MAX
) {
7308 /* Potential overflow, we know nothing */
7309 __mark_reg32_unbounded(dst_reg
);
7312 dst_reg
->u32_min_value
*= umin_val
;
7313 dst_reg
->u32_max_value
*= umax_val
;
7314 if (dst_reg
->u32_max_value
> S32_MAX
) {
7315 /* Overflow possible, we know nothing */
7316 dst_reg
->s32_min_value
= S32_MIN
;
7317 dst_reg
->s32_max_value
= S32_MAX
;
7319 dst_reg
->s32_min_value
= dst_reg
->u32_min_value
;
7320 dst_reg
->s32_max_value
= dst_reg
->u32_max_value
;
7324 static void scalar_min_max_mul(struct bpf_reg_state
*dst_reg
,
7325 struct bpf_reg_state
*src_reg
)
7327 s64 smin_val
= src_reg
->smin_value
;
7328 u64 umin_val
= src_reg
->umin_value
;
7329 u64 umax_val
= src_reg
->umax_value
;
7331 if (smin_val
< 0 || dst_reg
->smin_value
< 0) {
7332 /* Ain't nobody got time to multiply that sign */
7333 __mark_reg64_unbounded(dst_reg
);
7336 /* Both values are positive, so we can work with unsigned and
7337 * copy the result to signed (unless it exceeds S64_MAX).
7339 if (umax_val
> U32_MAX
|| dst_reg
->umax_value
> U32_MAX
) {
7340 /* Potential overflow, we know nothing */
7341 __mark_reg64_unbounded(dst_reg
);
7344 dst_reg
->umin_value
*= umin_val
;
7345 dst_reg
->umax_value
*= umax_val
;
7346 if (dst_reg
->umax_value
> S64_MAX
) {
7347 /* Overflow possible, we know nothing */
7348 dst_reg
->smin_value
= S64_MIN
;
7349 dst_reg
->smax_value
= S64_MAX
;
7351 dst_reg
->smin_value
= dst_reg
->umin_value
;
7352 dst_reg
->smax_value
= dst_reg
->umax_value
;
7356 static void scalar32_min_max_and(struct bpf_reg_state
*dst_reg
,
7357 struct bpf_reg_state
*src_reg
)
7359 bool src_known
= tnum_subreg_is_const(src_reg
->var_off
);
7360 bool dst_known
= tnum_subreg_is_const(dst_reg
->var_off
);
7361 struct tnum var32_off
= tnum_subreg(dst_reg
->var_off
);
7362 s32 smin_val
= src_reg
->s32_min_value
;
7363 u32 umax_val
= src_reg
->u32_max_value
;
7365 if (src_known
&& dst_known
) {
7366 __mark_reg32_known(dst_reg
, var32_off
.value
);
7370 /* We get our minimum from the var_off, since that's inherently
7371 * bitwise. Our maximum is the minimum of the operands' maxima.
7373 dst_reg
->u32_min_value
= var32_off
.value
;
7374 dst_reg
->u32_max_value
= min(dst_reg
->u32_max_value
, umax_val
);
7375 if (dst_reg
->s32_min_value
< 0 || smin_val
< 0) {
7376 /* Lose signed bounds when ANDing negative numbers,
7377 * ain't nobody got time for that.
7379 dst_reg
->s32_min_value
= S32_MIN
;
7380 dst_reg
->s32_max_value
= S32_MAX
;
7382 /* ANDing two positives gives a positive, so safe to
7383 * cast result into s64.
7385 dst_reg
->s32_min_value
= dst_reg
->u32_min_value
;
7386 dst_reg
->s32_max_value
= dst_reg
->u32_max_value
;
7390 static void scalar_min_max_and(struct bpf_reg_state
*dst_reg
,
7391 struct bpf_reg_state
*src_reg
)
7393 bool src_known
= tnum_is_const(src_reg
->var_off
);
7394 bool dst_known
= tnum_is_const(dst_reg
->var_off
);
7395 s64 smin_val
= src_reg
->smin_value
;
7396 u64 umax_val
= src_reg
->umax_value
;
7398 if (src_known
&& dst_known
) {
7399 __mark_reg_known(dst_reg
, dst_reg
->var_off
.value
);
7403 /* We get our minimum from the var_off, since that's inherently
7404 * bitwise. Our maximum is the minimum of the operands' maxima.
7406 dst_reg
->umin_value
= dst_reg
->var_off
.value
;
7407 dst_reg
->umax_value
= min(dst_reg
->umax_value
, umax_val
);
7408 if (dst_reg
->smin_value
< 0 || smin_val
< 0) {
7409 /* Lose signed bounds when ANDing negative numbers,
7410 * ain't nobody got time for that.
7412 dst_reg
->smin_value
= S64_MIN
;
7413 dst_reg
->smax_value
= S64_MAX
;
7415 /* ANDing two positives gives a positive, so safe to
7416 * cast result into s64.
7418 dst_reg
->smin_value
= dst_reg
->umin_value
;
7419 dst_reg
->smax_value
= dst_reg
->umax_value
;
7421 /* We may learn something more from the var_off */
7422 __update_reg_bounds(dst_reg
);
7425 static void scalar32_min_max_or(struct bpf_reg_state
*dst_reg
,
7426 struct bpf_reg_state
*src_reg
)
7428 bool src_known
= tnum_subreg_is_const(src_reg
->var_off
);
7429 bool dst_known
= tnum_subreg_is_const(dst_reg
->var_off
);
7430 struct tnum var32_off
= tnum_subreg(dst_reg
->var_off
);
7431 s32 smin_val
= src_reg
->s32_min_value
;
7432 u32 umin_val
= src_reg
->u32_min_value
;
7434 if (src_known
&& dst_known
) {
7435 __mark_reg32_known(dst_reg
, var32_off
.value
);
7439 /* We get our maximum from the var_off, and our minimum is the
7440 * maximum of the operands' minima
7442 dst_reg
->u32_min_value
= max(dst_reg
->u32_min_value
, umin_val
);
7443 dst_reg
->u32_max_value
= var32_off
.value
| var32_off
.mask
;
7444 if (dst_reg
->s32_min_value
< 0 || smin_val
< 0) {
7445 /* Lose signed bounds when ORing negative numbers,
7446 * ain't nobody got time for that.
7448 dst_reg
->s32_min_value
= S32_MIN
;
7449 dst_reg
->s32_max_value
= S32_MAX
;
7451 /* ORing two positives gives a positive, so safe to
7452 * cast result into s64.
7454 dst_reg
->s32_min_value
= dst_reg
->u32_min_value
;
7455 dst_reg
->s32_max_value
= dst_reg
->u32_max_value
;
7459 static void scalar_min_max_or(struct bpf_reg_state
*dst_reg
,
7460 struct bpf_reg_state
*src_reg
)
7462 bool src_known
= tnum_is_const(src_reg
->var_off
);
7463 bool dst_known
= tnum_is_const(dst_reg
->var_off
);
7464 s64 smin_val
= src_reg
->smin_value
;
7465 u64 umin_val
= src_reg
->umin_value
;
7467 if (src_known
&& dst_known
) {
7468 __mark_reg_known(dst_reg
, dst_reg
->var_off
.value
);
7472 /* We get our maximum from the var_off, and our minimum is the
7473 * maximum of the operands' minima
7475 dst_reg
->umin_value
= max(dst_reg
->umin_value
, umin_val
);
7476 dst_reg
->umax_value
= dst_reg
->var_off
.value
| dst_reg
->var_off
.mask
;
7477 if (dst_reg
->smin_value
< 0 || smin_val
< 0) {
7478 /* Lose signed bounds when ORing negative numbers,
7479 * ain't nobody got time for that.
7481 dst_reg
->smin_value
= S64_MIN
;
7482 dst_reg
->smax_value
= S64_MAX
;
7484 /* ORing two positives gives a positive, so safe to
7485 * cast result into s64.
7487 dst_reg
->smin_value
= dst_reg
->umin_value
;
7488 dst_reg
->smax_value
= dst_reg
->umax_value
;
7490 /* We may learn something more from the var_off */
7491 __update_reg_bounds(dst_reg
);
7494 static void scalar32_min_max_xor(struct bpf_reg_state
*dst_reg
,
7495 struct bpf_reg_state
*src_reg
)
7497 bool src_known
= tnum_subreg_is_const(src_reg
->var_off
);
7498 bool dst_known
= tnum_subreg_is_const(dst_reg
->var_off
);
7499 struct tnum var32_off
= tnum_subreg(dst_reg
->var_off
);
7500 s32 smin_val
= src_reg
->s32_min_value
;
7502 if (src_known
&& dst_known
) {
7503 __mark_reg32_known(dst_reg
, var32_off
.value
);
7507 /* We get both minimum and maximum from the var32_off. */
7508 dst_reg
->u32_min_value
= var32_off
.value
;
7509 dst_reg
->u32_max_value
= var32_off
.value
| var32_off
.mask
;
7511 if (dst_reg
->s32_min_value
>= 0 && smin_val
>= 0) {
7512 /* XORing two positive sign numbers gives a positive,
7513 * so safe to cast u32 result into s32.
7515 dst_reg
->s32_min_value
= dst_reg
->u32_min_value
;
7516 dst_reg
->s32_max_value
= dst_reg
->u32_max_value
;
7518 dst_reg
->s32_min_value
= S32_MIN
;
7519 dst_reg
->s32_max_value
= S32_MAX
;
7523 static void scalar_min_max_xor(struct bpf_reg_state
*dst_reg
,
7524 struct bpf_reg_state
*src_reg
)
7526 bool src_known
= tnum_is_const(src_reg
->var_off
);
7527 bool dst_known
= tnum_is_const(dst_reg
->var_off
);
7528 s64 smin_val
= src_reg
->smin_value
;
7530 if (src_known
&& dst_known
) {
7531 /* dst_reg->var_off.value has been updated earlier */
7532 __mark_reg_known(dst_reg
, dst_reg
->var_off
.value
);
7536 /* We get both minimum and maximum from the var_off. */
7537 dst_reg
->umin_value
= dst_reg
->var_off
.value
;
7538 dst_reg
->umax_value
= dst_reg
->var_off
.value
| dst_reg
->var_off
.mask
;
7540 if (dst_reg
->smin_value
>= 0 && smin_val
>= 0) {
7541 /* XORing two positive sign numbers gives a positive,
7542 * so safe to cast u64 result into s64.
7544 dst_reg
->smin_value
= dst_reg
->umin_value
;
7545 dst_reg
->smax_value
= dst_reg
->umax_value
;
7547 dst_reg
->smin_value
= S64_MIN
;
7548 dst_reg
->smax_value
= S64_MAX
;
7551 __update_reg_bounds(dst_reg
);
7554 static void __scalar32_min_max_lsh(struct bpf_reg_state
*dst_reg
,
7555 u64 umin_val
, u64 umax_val
)
7557 /* We lose all sign bit information (except what we can pick
7560 dst_reg
->s32_min_value
= S32_MIN
;
7561 dst_reg
->s32_max_value
= S32_MAX
;
7562 /* If we might shift our top bit out, then we know nothing */
7563 if (umax_val
> 31 || dst_reg
->u32_max_value
> 1ULL << (31 - umax_val
)) {
7564 dst_reg
->u32_min_value
= 0;
7565 dst_reg
->u32_max_value
= U32_MAX
;
7567 dst_reg
->u32_min_value
<<= umin_val
;
7568 dst_reg
->u32_max_value
<<= umax_val
;
7572 static void scalar32_min_max_lsh(struct bpf_reg_state
*dst_reg
,
7573 struct bpf_reg_state
*src_reg
)
7575 u32 umax_val
= src_reg
->u32_max_value
;
7576 u32 umin_val
= src_reg
->u32_min_value
;
7577 /* u32 alu operation will zext upper bits */
7578 struct tnum subreg
= tnum_subreg(dst_reg
->var_off
);
7580 __scalar32_min_max_lsh(dst_reg
, umin_val
, umax_val
);
7581 dst_reg
->var_off
= tnum_subreg(tnum_lshift(subreg
, umin_val
));
7582 /* Not required but being careful mark reg64 bounds as unknown so
7583 * that we are forced to pick them up from tnum and zext later and
7584 * if some path skips this step we are still safe.
7586 __mark_reg64_unbounded(dst_reg
);
7587 __update_reg32_bounds(dst_reg
);
7590 static void __scalar64_min_max_lsh(struct bpf_reg_state
*dst_reg
,
7591 u64 umin_val
, u64 umax_val
)
7593 /* Special case <<32 because it is a common compiler pattern to sign
7594 * extend subreg by doing <<32 s>>32. In this case if 32bit bounds are
7595 * positive we know this shift will also be positive so we can track
7596 * bounds correctly. Otherwise we lose all sign bit information except
7597 * what we can pick up from var_off. Perhaps we can generalize this
7598 * later to shifts of any length.
7600 if (umin_val
== 32 && umax_val
== 32 && dst_reg
->s32_max_value
>= 0)
7601 dst_reg
->smax_value
= (s64
)dst_reg
->s32_max_value
<< 32;
7603 dst_reg
->smax_value
= S64_MAX
;
7605 if (umin_val
== 32 && umax_val
== 32 && dst_reg
->s32_min_value
>= 0)
7606 dst_reg
->smin_value
= (s64
)dst_reg
->s32_min_value
<< 32;
7608 dst_reg
->smin_value
= S64_MIN
;
7610 /* If we might shift our top bit out, then we know nothing */
7611 if (dst_reg
->umax_value
> 1ULL << (63 - umax_val
)) {
7612 dst_reg
->umin_value
= 0;
7613 dst_reg
->umax_value
= U64_MAX
;
7615 dst_reg
->umin_value
<<= umin_val
;
7616 dst_reg
->umax_value
<<= umax_val
;
7620 static void scalar_min_max_lsh(struct bpf_reg_state
*dst_reg
,
7621 struct bpf_reg_state
*src_reg
)
7623 u64 umax_val
= src_reg
->umax_value
;
7624 u64 umin_val
= src_reg
->umin_value
;
7626 /* scalar64 calc uses 32bit unshifted bounds so must be called first */
7627 __scalar64_min_max_lsh(dst_reg
, umin_val
, umax_val
);
7628 __scalar32_min_max_lsh(dst_reg
, umin_val
, umax_val
);
7630 dst_reg
->var_off
= tnum_lshift(dst_reg
->var_off
, umin_val
);
7631 /* We may learn something more from the var_off */
7632 __update_reg_bounds(dst_reg
);
7635 static void scalar32_min_max_rsh(struct bpf_reg_state
*dst_reg
,
7636 struct bpf_reg_state
*src_reg
)
7638 struct tnum subreg
= tnum_subreg(dst_reg
->var_off
);
7639 u32 umax_val
= src_reg
->u32_max_value
;
7640 u32 umin_val
= src_reg
->u32_min_value
;
7642 /* BPF_RSH is an unsigned shift. If the value in dst_reg might
7643 * be negative, then either:
7644 * 1) src_reg might be zero, so the sign bit of the result is
7645 * unknown, so we lose our signed bounds
7646 * 2) it's known negative, thus the unsigned bounds capture the
7648 * 3) the signed bounds cross zero, so they tell us nothing
7650 * If the value in dst_reg is known nonnegative, then again the
7651 * unsigned bounds capture the signed bounds.
7652 * Thus, in all cases it suffices to blow away our signed bounds
7653 * and rely on inferring new ones from the unsigned bounds and
7654 * var_off of the result.
7656 dst_reg
->s32_min_value
= S32_MIN
;
7657 dst_reg
->s32_max_value
= S32_MAX
;
7659 dst_reg
->var_off
= tnum_rshift(subreg
, umin_val
);
7660 dst_reg
->u32_min_value
>>= umax_val
;
7661 dst_reg
->u32_max_value
>>= umin_val
;
7663 __mark_reg64_unbounded(dst_reg
);
7664 __update_reg32_bounds(dst_reg
);
7667 static void scalar_min_max_rsh(struct bpf_reg_state
*dst_reg
,
7668 struct bpf_reg_state
*src_reg
)
7670 u64 umax_val
= src_reg
->umax_value
;
7671 u64 umin_val
= src_reg
->umin_value
;
7673 /* BPF_RSH is an unsigned shift. If the value in dst_reg might
7674 * be negative, then either:
7675 * 1) src_reg might be zero, so the sign bit of the result is
7676 * unknown, so we lose our signed bounds
7677 * 2) it's known negative, thus the unsigned bounds capture the
7679 * 3) the signed bounds cross zero, so they tell us nothing
7681 * If the value in dst_reg is known nonnegative, then again the
7682 * unsigned bounds capture the signed bounds.
7683 * Thus, in all cases it suffices to blow away our signed bounds
7684 * and rely on inferring new ones from the unsigned bounds and
7685 * var_off of the result.
7687 dst_reg
->smin_value
= S64_MIN
;
7688 dst_reg
->smax_value
= S64_MAX
;
7689 dst_reg
->var_off
= tnum_rshift(dst_reg
->var_off
, umin_val
);
7690 dst_reg
->umin_value
>>= umax_val
;
7691 dst_reg
->umax_value
>>= umin_val
;
7693 /* Its not easy to operate on alu32 bounds here because it depends
7694 * on bits being shifted in. Take easy way out and mark unbounded
7695 * so we can recalculate later from tnum.
7697 __mark_reg32_unbounded(dst_reg
);
7698 __update_reg_bounds(dst_reg
);
7701 static void scalar32_min_max_arsh(struct bpf_reg_state
*dst_reg
,
7702 struct bpf_reg_state
*src_reg
)
7704 u64 umin_val
= src_reg
->u32_min_value
;
7706 /* Upon reaching here, src_known is true and
7707 * umax_val is equal to umin_val.
7709 dst_reg
->s32_min_value
= (u32
)(((s32
)dst_reg
->s32_min_value
) >> umin_val
);
7710 dst_reg
->s32_max_value
= (u32
)(((s32
)dst_reg
->s32_max_value
) >> umin_val
);
7712 dst_reg
->var_off
= tnum_arshift(tnum_subreg(dst_reg
->var_off
), umin_val
, 32);
7714 /* blow away the dst_reg umin_value/umax_value and rely on
7715 * dst_reg var_off to refine the result.
7717 dst_reg
->u32_min_value
= 0;
7718 dst_reg
->u32_max_value
= U32_MAX
;
7720 __mark_reg64_unbounded(dst_reg
);
7721 __update_reg32_bounds(dst_reg
);
7724 static void scalar_min_max_arsh(struct bpf_reg_state
*dst_reg
,
7725 struct bpf_reg_state
*src_reg
)
7727 u64 umin_val
= src_reg
->umin_value
;
7729 /* Upon reaching here, src_known is true and umax_val is equal
7732 dst_reg
->smin_value
>>= umin_val
;
7733 dst_reg
->smax_value
>>= umin_val
;
7735 dst_reg
->var_off
= tnum_arshift(dst_reg
->var_off
, umin_val
, 64);
7737 /* blow away the dst_reg umin_value/umax_value and rely on
7738 * dst_reg var_off to refine the result.
7740 dst_reg
->umin_value
= 0;
7741 dst_reg
->umax_value
= U64_MAX
;
7743 /* Its not easy to operate on alu32 bounds here because it depends
7744 * on bits being shifted in from upper 32-bits. Take easy way out
7745 * and mark unbounded so we can recalculate later from tnum.
7747 __mark_reg32_unbounded(dst_reg
);
7748 __update_reg_bounds(dst_reg
);
7751 /* WARNING: This function does calculations on 64-bit values, but the actual
7752 * execution may occur on 32-bit values. Therefore, things like bitshifts
7753 * need extra checks in the 32-bit case.
7755 static int adjust_scalar_min_max_vals(struct bpf_verifier_env
*env
,
7756 struct bpf_insn
*insn
,
7757 struct bpf_reg_state
*dst_reg
,
7758 struct bpf_reg_state src_reg
)
7760 struct bpf_reg_state
*regs
= cur_regs(env
);
7761 u8 opcode
= BPF_OP(insn
->code
);
7763 s64 smin_val
, smax_val
;
7764 u64 umin_val
, umax_val
;
7765 s32 s32_min_val
, s32_max_val
;
7766 u32 u32_min_val
, u32_max_val
;
7767 u64 insn_bitness
= (BPF_CLASS(insn
->code
) == BPF_ALU64
) ? 64 : 32;
7768 bool alu32
= (BPF_CLASS(insn
->code
) != BPF_ALU64
);
7771 smin_val
= src_reg
.smin_value
;
7772 smax_val
= src_reg
.smax_value
;
7773 umin_val
= src_reg
.umin_value
;
7774 umax_val
= src_reg
.umax_value
;
7776 s32_min_val
= src_reg
.s32_min_value
;
7777 s32_max_val
= src_reg
.s32_max_value
;
7778 u32_min_val
= src_reg
.u32_min_value
;
7779 u32_max_val
= src_reg
.u32_max_value
;
7782 src_known
= tnum_subreg_is_const(src_reg
.var_off
);
7784 (s32_min_val
!= s32_max_val
|| u32_min_val
!= u32_max_val
)) ||
7785 s32_min_val
> s32_max_val
|| u32_min_val
> u32_max_val
) {
7786 /* Taint dst register if offset had invalid bounds
7787 * derived from e.g. dead branches.
7789 __mark_reg_unknown(env
, dst_reg
);
7793 src_known
= tnum_is_const(src_reg
.var_off
);
7795 (smin_val
!= smax_val
|| umin_val
!= umax_val
)) ||
7796 smin_val
> smax_val
|| umin_val
> umax_val
) {
7797 /* Taint dst register if offset had invalid bounds
7798 * derived from e.g. dead branches.
7800 __mark_reg_unknown(env
, dst_reg
);
7806 opcode
!= BPF_ADD
&& opcode
!= BPF_SUB
&& opcode
!= BPF_AND
) {
7807 __mark_reg_unknown(env
, dst_reg
);
7811 if (sanitize_needed(opcode
)) {
7812 ret
= sanitize_val_alu(env
, insn
);
7814 return sanitize_err(env
, insn
, ret
, NULL
, NULL
);
7817 /* Calculate sign/unsigned bounds and tnum for alu32 and alu64 bit ops.
7818 * There are two classes of instructions: The first class we track both
7819 * alu32 and alu64 sign/unsigned bounds independently this provides the
7820 * greatest amount of precision when alu operations are mixed with jmp32
7821 * operations. These operations are BPF_ADD, BPF_SUB, BPF_MUL, BPF_ADD,
7822 * and BPF_OR. This is possible because these ops have fairly easy to
7823 * understand and calculate behavior in both 32-bit and 64-bit alu ops.
7824 * See alu32 verifier tests for examples. The second class of
7825 * operations, BPF_LSH, BPF_RSH, and BPF_ARSH, however are not so easy
7826 * with regards to tracking sign/unsigned bounds because the bits may
7827 * cross subreg boundaries in the alu64 case. When this happens we mark
7828 * the reg unbounded in the subreg bound space and use the resulting
7829 * tnum to calculate an approximation of the sign/unsigned bounds.
7833 scalar32_min_max_add(dst_reg
, &src_reg
);
7834 scalar_min_max_add(dst_reg
, &src_reg
);
7835 dst_reg
->var_off
= tnum_add(dst_reg
->var_off
, src_reg
.var_off
);
7838 scalar32_min_max_sub(dst_reg
, &src_reg
);
7839 scalar_min_max_sub(dst_reg
, &src_reg
);
7840 dst_reg
->var_off
= tnum_sub(dst_reg
->var_off
, src_reg
.var_off
);
7843 dst_reg
->var_off
= tnum_mul(dst_reg
->var_off
, src_reg
.var_off
);
7844 scalar32_min_max_mul(dst_reg
, &src_reg
);
7845 scalar_min_max_mul(dst_reg
, &src_reg
);
7848 dst_reg
->var_off
= tnum_and(dst_reg
->var_off
, src_reg
.var_off
);
7849 scalar32_min_max_and(dst_reg
, &src_reg
);
7850 scalar_min_max_and(dst_reg
, &src_reg
);
7853 dst_reg
->var_off
= tnum_or(dst_reg
->var_off
, src_reg
.var_off
);
7854 scalar32_min_max_or(dst_reg
, &src_reg
);
7855 scalar_min_max_or(dst_reg
, &src_reg
);
7858 dst_reg
->var_off
= tnum_xor(dst_reg
->var_off
, src_reg
.var_off
);
7859 scalar32_min_max_xor(dst_reg
, &src_reg
);
7860 scalar_min_max_xor(dst_reg
, &src_reg
);
7863 if (umax_val
>= insn_bitness
) {
7864 /* Shifts greater than 31 or 63 are undefined.
7865 * This includes shifts by a negative number.
7867 mark_reg_unknown(env
, regs
, insn
->dst_reg
);
7871 scalar32_min_max_lsh(dst_reg
, &src_reg
);
7873 scalar_min_max_lsh(dst_reg
, &src_reg
);
7876 if (umax_val
>= insn_bitness
) {
7877 /* Shifts greater than 31 or 63 are undefined.
7878 * This includes shifts by a negative number.
7880 mark_reg_unknown(env
, regs
, insn
->dst_reg
);
7884 scalar32_min_max_rsh(dst_reg
, &src_reg
);
7886 scalar_min_max_rsh(dst_reg
, &src_reg
);
7889 if (umax_val
>= insn_bitness
) {
7890 /* Shifts greater than 31 or 63 are undefined.
7891 * This includes shifts by a negative number.
7893 mark_reg_unknown(env
, regs
, insn
->dst_reg
);
7897 scalar32_min_max_arsh(dst_reg
, &src_reg
);
7899 scalar_min_max_arsh(dst_reg
, &src_reg
);
7902 mark_reg_unknown(env
, regs
, insn
->dst_reg
);
7906 /* ALU32 ops are zero extended into 64bit register */
7908 zext_32_to_64(dst_reg
);
7910 __update_reg_bounds(dst_reg
);
7911 __reg_deduce_bounds(dst_reg
);
7912 __reg_bound_offset(dst_reg
);
7916 /* Handles ALU ops other than BPF_END, BPF_NEG and BPF_MOV: computes new min/max
7919 static int adjust_reg_min_max_vals(struct bpf_verifier_env
*env
,
7920 struct bpf_insn
*insn
)
7922 struct bpf_verifier_state
*vstate
= env
->cur_state
;
7923 struct bpf_func_state
*state
= vstate
->frame
[vstate
->curframe
];
7924 struct bpf_reg_state
*regs
= state
->regs
, *dst_reg
, *src_reg
;
7925 struct bpf_reg_state
*ptr_reg
= NULL
, off_reg
= {0};
7926 u8 opcode
= BPF_OP(insn
->code
);
7929 dst_reg
= ®s
[insn
->dst_reg
];
7931 if (dst_reg
->type
!= SCALAR_VALUE
)
7934 /* Make sure ID is cleared otherwise dst_reg min/max could be
7935 * incorrectly propagated into other registers by find_equal_scalars()
7938 if (BPF_SRC(insn
->code
) == BPF_X
) {
7939 src_reg
= ®s
[insn
->src_reg
];
7940 if (src_reg
->type
!= SCALAR_VALUE
) {
7941 if (dst_reg
->type
!= SCALAR_VALUE
) {
7942 /* Combining two pointers by any ALU op yields
7943 * an arbitrary scalar. Disallow all math except
7944 * pointer subtraction
7946 if (opcode
== BPF_SUB
&& env
->allow_ptr_leaks
) {
7947 mark_reg_unknown(env
, regs
, insn
->dst_reg
);
7950 verbose(env
, "R%d pointer %s pointer prohibited\n",
7952 bpf_alu_string
[opcode
>> 4]);
7955 /* scalar += pointer
7956 * This is legal, but we have to reverse our
7957 * src/dest handling in computing the range
7959 err
= mark_chain_precision(env
, insn
->dst_reg
);
7962 return adjust_ptr_min_max_vals(env
, insn
,
7965 } else if (ptr_reg
) {
7966 /* pointer += scalar */
7967 err
= mark_chain_precision(env
, insn
->src_reg
);
7970 return adjust_ptr_min_max_vals(env
, insn
,
7974 /* Pretend the src is a reg with a known value, since we only
7975 * need to be able to read from this state.
7977 off_reg
.type
= SCALAR_VALUE
;
7978 __mark_reg_known(&off_reg
, insn
->imm
);
7980 if (ptr_reg
) /* pointer += K */
7981 return adjust_ptr_min_max_vals(env
, insn
,
7985 /* Got here implies adding two SCALAR_VALUEs */
7986 if (WARN_ON_ONCE(ptr_reg
)) {
7987 print_verifier_state(env
, state
);
7988 verbose(env
, "verifier internal error: unexpected ptr_reg\n");
7991 if (WARN_ON(!src_reg
)) {
7992 print_verifier_state(env
, state
);
7993 verbose(env
, "verifier internal error: no src_reg\n");
7996 return adjust_scalar_min_max_vals(env
, insn
, dst_reg
, *src_reg
);
7999 /* check validity of 32-bit and 64-bit arithmetic operations */
8000 static int check_alu_op(struct bpf_verifier_env
*env
, struct bpf_insn
*insn
)
8002 struct bpf_reg_state
*regs
= cur_regs(env
);
8003 u8 opcode
= BPF_OP(insn
->code
);
8006 if (opcode
== BPF_END
|| opcode
== BPF_NEG
) {
8007 if (opcode
== BPF_NEG
) {
8008 if (BPF_SRC(insn
->code
) != 0 ||
8009 insn
->src_reg
!= BPF_REG_0
||
8010 insn
->off
!= 0 || insn
->imm
!= 0) {
8011 verbose(env
, "BPF_NEG uses reserved fields\n");
8015 if (insn
->src_reg
!= BPF_REG_0
|| insn
->off
!= 0 ||
8016 (insn
->imm
!= 16 && insn
->imm
!= 32 && insn
->imm
!= 64) ||
8017 BPF_CLASS(insn
->code
) == BPF_ALU64
) {
8018 verbose(env
, "BPF_END uses reserved fields\n");
8023 /* check src operand */
8024 err
= check_reg_arg(env
, insn
->dst_reg
, SRC_OP
);
8028 if (is_pointer_value(env
, insn
->dst_reg
)) {
8029 verbose(env
, "R%d pointer arithmetic prohibited\n",
8034 /* check dest operand */
8035 err
= check_reg_arg(env
, insn
->dst_reg
, DST_OP
);
8039 } else if (opcode
== BPF_MOV
) {
8041 if (BPF_SRC(insn
->code
) == BPF_X
) {
8042 if (insn
->imm
!= 0 || insn
->off
!= 0) {
8043 verbose(env
, "BPF_MOV uses reserved fields\n");
8047 /* check src operand */
8048 err
= check_reg_arg(env
, insn
->src_reg
, SRC_OP
);
8052 if (insn
->src_reg
!= BPF_REG_0
|| insn
->off
!= 0) {
8053 verbose(env
, "BPF_MOV uses reserved fields\n");
8058 /* check dest operand, mark as required later */
8059 err
= check_reg_arg(env
, insn
->dst_reg
, DST_OP_NO_MARK
);
8063 if (BPF_SRC(insn
->code
) == BPF_X
) {
8064 struct bpf_reg_state
*src_reg
= regs
+ insn
->src_reg
;
8065 struct bpf_reg_state
*dst_reg
= regs
+ insn
->dst_reg
;
8067 if (BPF_CLASS(insn
->code
) == BPF_ALU64
) {
8069 * copy register state to dest reg
8071 if (src_reg
->type
== SCALAR_VALUE
&& !src_reg
->id
)
8072 /* Assign src and dst registers the same ID
8073 * that will be used by find_equal_scalars()
8074 * to propagate min/max range.
8076 src_reg
->id
= ++env
->id_gen
;
8077 *dst_reg
= *src_reg
;
8078 dst_reg
->live
|= REG_LIVE_WRITTEN
;
8079 dst_reg
->subreg_def
= DEF_NOT_SUBREG
;
8082 if (is_pointer_value(env
, insn
->src_reg
)) {
8084 "R%d partial copy of pointer\n",
8087 } else if (src_reg
->type
== SCALAR_VALUE
) {
8088 *dst_reg
= *src_reg
;
8089 /* Make sure ID is cleared otherwise
8090 * dst_reg min/max could be incorrectly
8091 * propagated into src_reg by find_equal_scalars()
8094 dst_reg
->live
|= REG_LIVE_WRITTEN
;
8095 dst_reg
->subreg_def
= env
->insn_idx
+ 1;
8097 mark_reg_unknown(env
, regs
,
8100 zext_32_to_64(dst_reg
);
8104 * remember the value we stored into this reg
8106 /* clear any state __mark_reg_known doesn't set */
8107 mark_reg_unknown(env
, regs
, insn
->dst_reg
);
8108 regs
[insn
->dst_reg
].type
= SCALAR_VALUE
;
8109 if (BPF_CLASS(insn
->code
) == BPF_ALU64
) {
8110 __mark_reg_known(regs
+ insn
->dst_reg
,
8113 __mark_reg_known(regs
+ insn
->dst_reg
,
8118 } else if (opcode
> BPF_END
) {
8119 verbose(env
, "invalid BPF_ALU opcode %x\n", opcode
);
8122 } else { /* all other ALU ops: and, sub, xor, add, ... */
8124 if (BPF_SRC(insn
->code
) == BPF_X
) {
8125 if (insn
->imm
!= 0 || insn
->off
!= 0) {
8126 verbose(env
, "BPF_ALU uses reserved fields\n");
8129 /* check src1 operand */
8130 err
= check_reg_arg(env
, insn
->src_reg
, SRC_OP
);
8134 if (insn
->src_reg
!= BPF_REG_0
|| insn
->off
!= 0) {
8135 verbose(env
, "BPF_ALU uses reserved fields\n");
8140 /* check src2 operand */
8141 err
= check_reg_arg(env
, insn
->dst_reg
, SRC_OP
);
8145 if ((opcode
== BPF_MOD
|| opcode
== BPF_DIV
) &&
8146 BPF_SRC(insn
->code
) == BPF_K
&& insn
->imm
== 0) {
8147 verbose(env
, "div by zero\n");
8151 if ((opcode
== BPF_LSH
|| opcode
== BPF_RSH
||
8152 opcode
== BPF_ARSH
) && BPF_SRC(insn
->code
) == BPF_K
) {
8153 int size
= BPF_CLASS(insn
->code
) == BPF_ALU64
? 64 : 32;
8155 if (insn
->imm
< 0 || insn
->imm
>= size
) {
8156 verbose(env
, "invalid shift %d\n", insn
->imm
);
8161 /* check dest operand */
8162 err
= check_reg_arg(env
, insn
->dst_reg
, DST_OP_NO_MARK
);
8166 return adjust_reg_min_max_vals(env
, insn
);
8172 static void __find_good_pkt_pointers(struct bpf_func_state
*state
,
8173 struct bpf_reg_state
*dst_reg
,
8174 enum bpf_reg_type type
, int new_range
)
8176 struct bpf_reg_state
*reg
;
8179 for (i
= 0; i
< MAX_BPF_REG
; i
++) {
8180 reg
= &state
->regs
[i
];
8181 if (reg
->type
== type
&& reg
->id
== dst_reg
->id
)
8182 /* keep the maximum range already checked */
8183 reg
->range
= max(reg
->range
, new_range
);
8186 bpf_for_each_spilled_reg(i
, state
, reg
) {
8189 if (reg
->type
== type
&& reg
->id
== dst_reg
->id
)
8190 reg
->range
= max(reg
->range
, new_range
);
8194 static void find_good_pkt_pointers(struct bpf_verifier_state
*vstate
,
8195 struct bpf_reg_state
*dst_reg
,
8196 enum bpf_reg_type type
,
8197 bool range_right_open
)
8201 if (dst_reg
->off
< 0 ||
8202 (dst_reg
->off
== 0 && range_right_open
))
8203 /* This doesn't give us any range */
8206 if (dst_reg
->umax_value
> MAX_PACKET_OFF
||
8207 dst_reg
->umax_value
+ dst_reg
->off
> MAX_PACKET_OFF
)
8208 /* Risk of overflow. For instance, ptr + (1<<63) may be less
8209 * than pkt_end, but that's because it's also less than pkt.
8213 new_range
= dst_reg
->off
;
8214 if (range_right_open
)
8217 /* Examples for register markings:
8219 * pkt_data in dst register:
8223 * if (r2 > pkt_end) goto <handle exception>
8228 * if (r2 < pkt_end) goto <access okay>
8229 * <handle exception>
8232 * r2 == dst_reg, pkt_end == src_reg
8233 * r2=pkt(id=n,off=8,r=0)
8234 * r3=pkt(id=n,off=0,r=0)
8236 * pkt_data in src register:
8240 * if (pkt_end >= r2) goto <access okay>
8241 * <handle exception>
8245 * if (pkt_end <= r2) goto <handle exception>
8249 * pkt_end == dst_reg, r2 == src_reg
8250 * r2=pkt(id=n,off=8,r=0)
8251 * r3=pkt(id=n,off=0,r=0)
8253 * Find register r3 and mark its range as r3=pkt(id=n,off=0,r=8)
8254 * or r3=pkt(id=n,off=0,r=8-1), so that range of bytes [r3, r3 + 8)
8255 * and [r3, r3 + 8-1) respectively is safe to access depending on
8259 /* If our ids match, then we must have the same max_value. And we
8260 * don't care about the other reg's fixed offset, since if it's too big
8261 * the range won't allow anything.
8262 * dst_reg->off is known < MAX_PACKET_OFF, therefore it fits in a u16.
8264 for (i
= 0; i
<= vstate
->curframe
; i
++)
8265 __find_good_pkt_pointers(vstate
->frame
[i
], dst_reg
, type
,
8269 static int is_branch32_taken(struct bpf_reg_state
*reg
, u32 val
, u8 opcode
)
8271 struct tnum subreg
= tnum_subreg(reg
->var_off
);
8272 s32 sval
= (s32
)val
;
8276 if (tnum_is_const(subreg
))
8277 return !!tnum_equals_const(subreg
, val
);
8280 if (tnum_is_const(subreg
))
8281 return !tnum_equals_const(subreg
, val
);
8284 if ((~subreg
.mask
& subreg
.value
) & val
)
8286 if (!((subreg
.mask
| subreg
.value
) & val
))
8290 if (reg
->u32_min_value
> val
)
8292 else if (reg
->u32_max_value
<= val
)
8296 if (reg
->s32_min_value
> sval
)
8298 else if (reg
->s32_max_value
<= sval
)
8302 if (reg
->u32_max_value
< val
)
8304 else if (reg
->u32_min_value
>= val
)
8308 if (reg
->s32_max_value
< sval
)
8310 else if (reg
->s32_min_value
>= sval
)
8314 if (reg
->u32_min_value
>= val
)
8316 else if (reg
->u32_max_value
< val
)
8320 if (reg
->s32_min_value
>= sval
)
8322 else if (reg
->s32_max_value
< sval
)
8326 if (reg
->u32_max_value
<= val
)
8328 else if (reg
->u32_min_value
> val
)
8332 if (reg
->s32_max_value
<= sval
)
8334 else if (reg
->s32_min_value
> sval
)
8343 static int is_branch64_taken(struct bpf_reg_state
*reg
, u64 val
, u8 opcode
)
8345 s64 sval
= (s64
)val
;
8349 if (tnum_is_const(reg
->var_off
))
8350 return !!tnum_equals_const(reg
->var_off
, val
);
8353 if (tnum_is_const(reg
->var_off
))
8354 return !tnum_equals_const(reg
->var_off
, val
);
8357 if ((~reg
->var_off
.mask
& reg
->var_off
.value
) & val
)
8359 if (!((reg
->var_off
.mask
| reg
->var_off
.value
) & val
))
8363 if (reg
->umin_value
> val
)
8365 else if (reg
->umax_value
<= val
)
8369 if (reg
->smin_value
> sval
)
8371 else if (reg
->smax_value
<= sval
)
8375 if (reg
->umax_value
< val
)
8377 else if (reg
->umin_value
>= val
)
8381 if (reg
->smax_value
< sval
)
8383 else if (reg
->smin_value
>= sval
)
8387 if (reg
->umin_value
>= val
)
8389 else if (reg
->umax_value
< val
)
8393 if (reg
->smin_value
>= sval
)
8395 else if (reg
->smax_value
< sval
)
8399 if (reg
->umax_value
<= val
)
8401 else if (reg
->umin_value
> val
)
8405 if (reg
->smax_value
<= sval
)
8407 else if (reg
->smin_value
> sval
)
8415 /* compute branch direction of the expression "if (reg opcode val) goto target;"
8417 * 1 - branch will be taken and "goto target" will be executed
8418 * 0 - branch will not be taken and fall-through to next insn
8419 * -1 - unknown. Example: "if (reg < 5)" is unknown when register value
8422 static int is_branch_taken(struct bpf_reg_state
*reg
, u64 val
, u8 opcode
,
8425 if (__is_pointer_value(false, reg
)) {
8426 if (!reg_type_not_null(reg
->type
))
8429 /* If pointer is valid tests against zero will fail so we can
8430 * use this to direct branch taken.
8446 return is_branch32_taken(reg
, val
, opcode
);
8447 return is_branch64_taken(reg
, val
, opcode
);
8450 static int flip_opcode(u32 opcode
)
8452 /* How can we transform "a <op> b" into "b <op> a"? */
8453 static const u8 opcode_flip
[16] = {
8454 /* these stay the same */
8455 [BPF_JEQ
>> 4] = BPF_JEQ
,
8456 [BPF_JNE
>> 4] = BPF_JNE
,
8457 [BPF_JSET
>> 4] = BPF_JSET
,
8458 /* these swap "lesser" and "greater" (L and G in the opcodes) */
8459 [BPF_JGE
>> 4] = BPF_JLE
,
8460 [BPF_JGT
>> 4] = BPF_JLT
,
8461 [BPF_JLE
>> 4] = BPF_JGE
,
8462 [BPF_JLT
>> 4] = BPF_JGT
,
8463 [BPF_JSGE
>> 4] = BPF_JSLE
,
8464 [BPF_JSGT
>> 4] = BPF_JSLT
,
8465 [BPF_JSLE
>> 4] = BPF_JSGE
,
8466 [BPF_JSLT
>> 4] = BPF_JSGT
8468 return opcode_flip
[opcode
>> 4];
8471 static int is_pkt_ptr_branch_taken(struct bpf_reg_state
*dst_reg
,
8472 struct bpf_reg_state
*src_reg
,
8475 struct bpf_reg_state
*pkt
;
8477 if (src_reg
->type
== PTR_TO_PACKET_END
) {
8479 } else if (dst_reg
->type
== PTR_TO_PACKET_END
) {
8481 opcode
= flip_opcode(opcode
);
8486 if (pkt
->range
>= 0)
8491 /* pkt <= pkt_end */
8495 if (pkt
->range
== BEYOND_PKT_END
)
8496 /* pkt has at last one extra byte beyond pkt_end */
8497 return opcode
== BPF_JGT
;
8503 /* pkt >= pkt_end */
8504 if (pkt
->range
== BEYOND_PKT_END
|| pkt
->range
== AT_PKT_END
)
8505 return opcode
== BPF_JGE
;
8511 /* Adjusts the register min/max values in the case that the dst_reg is the
8512 * variable register that we are working on, and src_reg is a constant or we're
8513 * simply doing a BPF_K check.
8514 * In JEQ/JNE cases we also adjust the var_off values.
8516 static void reg_set_min_max(struct bpf_reg_state
*true_reg
,
8517 struct bpf_reg_state
*false_reg
,
8519 u8 opcode
, bool is_jmp32
)
8521 struct tnum false_32off
= tnum_subreg(false_reg
->var_off
);
8522 struct tnum false_64off
= false_reg
->var_off
;
8523 struct tnum true_32off
= tnum_subreg(true_reg
->var_off
);
8524 struct tnum true_64off
= true_reg
->var_off
;
8525 s64 sval
= (s64
)val
;
8526 s32 sval32
= (s32
)val32
;
8528 /* If the dst_reg is a pointer, we can't learn anything about its
8529 * variable offset from the compare (unless src_reg were a pointer into
8530 * the same object, but we don't bother with that.
8531 * Since false_reg and true_reg have the same type by construction, we
8532 * only need to check one of them for pointerness.
8534 if (__is_pointer_value(false, false_reg
))
8541 struct bpf_reg_state
*reg
=
8542 opcode
== BPF_JEQ
? true_reg
: false_reg
;
8544 /* JEQ/JNE comparison doesn't change the register equivalence.
8546 * if (r1 == 42) goto label;
8548 * label: // here both r1 and r2 are known to be 42.
8550 * Hence when marking register as known preserve it's ID.
8553 __mark_reg32_known(reg
, val32
);
8555 ___mark_reg_known(reg
, val
);
8560 false_32off
= tnum_and(false_32off
, tnum_const(~val32
));
8561 if (is_power_of_2(val32
))
8562 true_32off
= tnum_or(true_32off
,
8565 false_64off
= tnum_and(false_64off
, tnum_const(~val
));
8566 if (is_power_of_2(val
))
8567 true_64off
= tnum_or(true_64off
,
8575 u32 false_umax
= opcode
== BPF_JGT
? val32
: val32
- 1;
8576 u32 true_umin
= opcode
== BPF_JGT
? val32
+ 1 : val32
;
8578 false_reg
->u32_max_value
= min(false_reg
->u32_max_value
,
8580 true_reg
->u32_min_value
= max(true_reg
->u32_min_value
,
8583 u64 false_umax
= opcode
== BPF_JGT
? val
: val
- 1;
8584 u64 true_umin
= opcode
== BPF_JGT
? val
+ 1 : val
;
8586 false_reg
->umax_value
= min(false_reg
->umax_value
, false_umax
);
8587 true_reg
->umin_value
= max(true_reg
->umin_value
, true_umin
);
8595 s32 false_smax
= opcode
== BPF_JSGT
? sval32
: sval32
- 1;
8596 s32 true_smin
= opcode
== BPF_JSGT
? sval32
+ 1 : sval32
;
8598 false_reg
->s32_max_value
= min(false_reg
->s32_max_value
, false_smax
);
8599 true_reg
->s32_min_value
= max(true_reg
->s32_min_value
, true_smin
);
8601 s64 false_smax
= opcode
== BPF_JSGT
? sval
: sval
- 1;
8602 s64 true_smin
= opcode
== BPF_JSGT
? sval
+ 1 : sval
;
8604 false_reg
->smax_value
= min(false_reg
->smax_value
, false_smax
);
8605 true_reg
->smin_value
= max(true_reg
->smin_value
, true_smin
);
8613 u32 false_umin
= opcode
== BPF_JLT
? val32
: val32
+ 1;
8614 u32 true_umax
= opcode
== BPF_JLT
? val32
- 1 : val32
;
8616 false_reg
->u32_min_value
= max(false_reg
->u32_min_value
,
8618 true_reg
->u32_max_value
= min(true_reg
->u32_max_value
,
8621 u64 false_umin
= opcode
== BPF_JLT
? val
: val
+ 1;
8622 u64 true_umax
= opcode
== BPF_JLT
? val
- 1 : val
;
8624 false_reg
->umin_value
= max(false_reg
->umin_value
, false_umin
);
8625 true_reg
->umax_value
= min(true_reg
->umax_value
, true_umax
);
8633 s32 false_smin
= opcode
== BPF_JSLT
? sval32
: sval32
+ 1;
8634 s32 true_smax
= opcode
== BPF_JSLT
? sval32
- 1 : sval32
;
8636 false_reg
->s32_min_value
= max(false_reg
->s32_min_value
, false_smin
);
8637 true_reg
->s32_max_value
= min(true_reg
->s32_max_value
, true_smax
);
8639 s64 false_smin
= opcode
== BPF_JSLT
? sval
: sval
+ 1;
8640 s64 true_smax
= opcode
== BPF_JSLT
? sval
- 1 : sval
;
8642 false_reg
->smin_value
= max(false_reg
->smin_value
, false_smin
);
8643 true_reg
->smax_value
= min(true_reg
->smax_value
, true_smax
);
8652 false_reg
->var_off
= tnum_or(tnum_clear_subreg(false_64off
),
8653 tnum_subreg(false_32off
));
8654 true_reg
->var_off
= tnum_or(tnum_clear_subreg(true_64off
),
8655 tnum_subreg(true_32off
));
8656 __reg_combine_32_into_64(false_reg
);
8657 __reg_combine_32_into_64(true_reg
);
8659 false_reg
->var_off
= false_64off
;
8660 true_reg
->var_off
= true_64off
;
8661 __reg_combine_64_into_32(false_reg
);
8662 __reg_combine_64_into_32(true_reg
);
8666 /* Same as above, but for the case that dst_reg holds a constant and src_reg is
8669 static void reg_set_min_max_inv(struct bpf_reg_state
*true_reg
,
8670 struct bpf_reg_state
*false_reg
,
8672 u8 opcode
, bool is_jmp32
)
8674 opcode
= flip_opcode(opcode
);
8675 /* This uses zero as "not present in table"; luckily the zero opcode,
8676 * BPF_JA, can't get here.
8679 reg_set_min_max(true_reg
, false_reg
, val
, val32
, opcode
, is_jmp32
);
8682 /* Regs are known to be equal, so intersect their min/max/var_off */
8683 static void __reg_combine_min_max(struct bpf_reg_state
*src_reg
,
8684 struct bpf_reg_state
*dst_reg
)
8686 src_reg
->umin_value
= dst_reg
->umin_value
= max(src_reg
->umin_value
,
8687 dst_reg
->umin_value
);
8688 src_reg
->umax_value
= dst_reg
->umax_value
= min(src_reg
->umax_value
,
8689 dst_reg
->umax_value
);
8690 src_reg
->smin_value
= dst_reg
->smin_value
= max(src_reg
->smin_value
,
8691 dst_reg
->smin_value
);
8692 src_reg
->smax_value
= dst_reg
->smax_value
= min(src_reg
->smax_value
,
8693 dst_reg
->smax_value
);
8694 src_reg
->var_off
= dst_reg
->var_off
= tnum_intersect(src_reg
->var_off
,
8696 /* We might have learned new bounds from the var_off. */
8697 __update_reg_bounds(src_reg
);
8698 __update_reg_bounds(dst_reg
);
8699 /* We might have learned something about the sign bit. */
8700 __reg_deduce_bounds(src_reg
);
8701 __reg_deduce_bounds(dst_reg
);
8702 /* We might have learned some bits from the bounds. */
8703 __reg_bound_offset(src_reg
);
8704 __reg_bound_offset(dst_reg
);
8705 /* Intersecting with the old var_off might have improved our bounds
8706 * slightly. e.g. if umax was 0x7f...f and var_off was (0; 0xf...fc),
8707 * then new var_off is (0; 0x7f...fc) which improves our umax.
8709 __update_reg_bounds(src_reg
);
8710 __update_reg_bounds(dst_reg
);
8713 static void reg_combine_min_max(struct bpf_reg_state
*true_src
,
8714 struct bpf_reg_state
*true_dst
,
8715 struct bpf_reg_state
*false_src
,
8716 struct bpf_reg_state
*false_dst
,
8721 __reg_combine_min_max(true_src
, true_dst
);
8724 __reg_combine_min_max(false_src
, false_dst
);
8729 static void mark_ptr_or_null_reg(struct bpf_func_state
*state
,
8730 struct bpf_reg_state
*reg
, u32 id
,
8733 if (reg_type_may_be_null(reg
->type
) && reg
->id
== id
&&
8734 !WARN_ON_ONCE(!reg
->id
)) {
8735 /* Old offset (both fixed and variable parts) should
8736 * have been known-zero, because we don't allow pointer
8737 * arithmetic on pointers that might be NULL.
8739 if (WARN_ON_ONCE(reg
->smin_value
|| reg
->smax_value
||
8740 !tnum_equals_const(reg
->var_off
, 0) ||
8742 __mark_reg_known_zero(reg
);
8746 reg
->type
= SCALAR_VALUE
;
8747 /* We don't need id and ref_obj_id from this point
8748 * onwards anymore, thus we should better reset it,
8749 * so that state pruning has chances to take effect.
8752 reg
->ref_obj_id
= 0;
8757 mark_ptr_not_null_reg(reg
);
8759 if (!reg_may_point_to_spin_lock(reg
)) {
8760 /* For not-NULL ptr, reg->ref_obj_id will be reset
8761 * in release_reg_references().
8763 * reg->id is still used by spin_lock ptr. Other
8764 * than spin_lock ptr type, reg->id can be reset.
8771 static void __mark_ptr_or_null_regs(struct bpf_func_state
*state
, u32 id
,
8774 struct bpf_reg_state
*reg
;
8777 for (i
= 0; i
< MAX_BPF_REG
; i
++)
8778 mark_ptr_or_null_reg(state
, &state
->regs
[i
], id
, is_null
);
8780 bpf_for_each_spilled_reg(i
, state
, reg
) {
8783 mark_ptr_or_null_reg(state
, reg
, id
, is_null
);
8787 /* The logic is similar to find_good_pkt_pointers(), both could eventually
8788 * be folded together at some point.
8790 static void mark_ptr_or_null_regs(struct bpf_verifier_state
*vstate
, u32 regno
,
8793 struct bpf_func_state
*state
= vstate
->frame
[vstate
->curframe
];
8794 struct bpf_reg_state
*regs
= state
->regs
;
8795 u32 ref_obj_id
= regs
[regno
].ref_obj_id
;
8796 u32 id
= regs
[regno
].id
;
8799 if (ref_obj_id
&& ref_obj_id
== id
&& is_null
)
8800 /* regs[regno] is in the " == NULL" branch.
8801 * No one could have freed the reference state before
8802 * doing the NULL check.
8804 WARN_ON_ONCE(release_reference_state(state
, id
));
8806 for (i
= 0; i
<= vstate
->curframe
; i
++)
8807 __mark_ptr_or_null_regs(vstate
->frame
[i
], id
, is_null
);
8810 static bool try_match_pkt_pointers(const struct bpf_insn
*insn
,
8811 struct bpf_reg_state
*dst_reg
,
8812 struct bpf_reg_state
*src_reg
,
8813 struct bpf_verifier_state
*this_branch
,
8814 struct bpf_verifier_state
*other_branch
)
8816 if (BPF_SRC(insn
->code
) != BPF_X
)
8819 /* Pointers are always 64-bit. */
8820 if (BPF_CLASS(insn
->code
) == BPF_JMP32
)
8823 switch (BPF_OP(insn
->code
)) {
8825 if ((dst_reg
->type
== PTR_TO_PACKET
&&
8826 src_reg
->type
== PTR_TO_PACKET_END
) ||
8827 (dst_reg
->type
== PTR_TO_PACKET_META
&&
8828 reg_is_init_pkt_pointer(src_reg
, PTR_TO_PACKET
))) {
8829 /* pkt_data' > pkt_end, pkt_meta' > pkt_data */
8830 find_good_pkt_pointers(this_branch
, dst_reg
,
8831 dst_reg
->type
, false);
8832 mark_pkt_end(other_branch
, insn
->dst_reg
, true);
8833 } else if ((dst_reg
->type
== PTR_TO_PACKET_END
&&
8834 src_reg
->type
== PTR_TO_PACKET
) ||
8835 (reg_is_init_pkt_pointer(dst_reg
, PTR_TO_PACKET
) &&
8836 src_reg
->type
== PTR_TO_PACKET_META
)) {
8837 /* pkt_end > pkt_data', pkt_data > pkt_meta' */
8838 find_good_pkt_pointers(other_branch
, src_reg
,
8839 src_reg
->type
, true);
8840 mark_pkt_end(this_branch
, insn
->src_reg
, false);
8846 if ((dst_reg
->type
== PTR_TO_PACKET
&&
8847 src_reg
->type
== PTR_TO_PACKET_END
) ||
8848 (dst_reg
->type
== PTR_TO_PACKET_META
&&
8849 reg_is_init_pkt_pointer(src_reg
, PTR_TO_PACKET
))) {
8850 /* pkt_data' < pkt_end, pkt_meta' < pkt_data */
8851 find_good_pkt_pointers(other_branch
, dst_reg
,
8852 dst_reg
->type
, true);
8853 mark_pkt_end(this_branch
, insn
->dst_reg
, false);
8854 } else if ((dst_reg
->type
== PTR_TO_PACKET_END
&&
8855 src_reg
->type
== PTR_TO_PACKET
) ||
8856 (reg_is_init_pkt_pointer(dst_reg
, PTR_TO_PACKET
) &&
8857 src_reg
->type
== PTR_TO_PACKET_META
)) {
8858 /* pkt_end < pkt_data', pkt_data > pkt_meta' */
8859 find_good_pkt_pointers(this_branch
, src_reg
,
8860 src_reg
->type
, false);
8861 mark_pkt_end(other_branch
, insn
->src_reg
, true);
8867 if ((dst_reg
->type
== PTR_TO_PACKET
&&
8868 src_reg
->type
== PTR_TO_PACKET_END
) ||
8869 (dst_reg
->type
== PTR_TO_PACKET_META
&&
8870 reg_is_init_pkt_pointer(src_reg
, PTR_TO_PACKET
))) {
8871 /* pkt_data' >= pkt_end, pkt_meta' >= pkt_data */
8872 find_good_pkt_pointers(this_branch
, dst_reg
,
8873 dst_reg
->type
, true);
8874 mark_pkt_end(other_branch
, insn
->dst_reg
, false);
8875 } else if ((dst_reg
->type
== PTR_TO_PACKET_END
&&
8876 src_reg
->type
== PTR_TO_PACKET
) ||
8877 (reg_is_init_pkt_pointer(dst_reg
, PTR_TO_PACKET
) &&
8878 src_reg
->type
== PTR_TO_PACKET_META
)) {
8879 /* pkt_end >= pkt_data', pkt_data >= pkt_meta' */
8880 find_good_pkt_pointers(other_branch
, src_reg
,
8881 src_reg
->type
, false);
8882 mark_pkt_end(this_branch
, insn
->src_reg
, true);
8888 if ((dst_reg
->type
== PTR_TO_PACKET
&&
8889 src_reg
->type
== PTR_TO_PACKET_END
) ||
8890 (dst_reg
->type
== PTR_TO_PACKET_META
&&
8891 reg_is_init_pkt_pointer(src_reg
, PTR_TO_PACKET
))) {
8892 /* pkt_data' <= pkt_end, pkt_meta' <= pkt_data */
8893 find_good_pkt_pointers(other_branch
, dst_reg
,
8894 dst_reg
->type
, false);
8895 mark_pkt_end(this_branch
, insn
->dst_reg
, true);
8896 } else if ((dst_reg
->type
== PTR_TO_PACKET_END
&&
8897 src_reg
->type
== PTR_TO_PACKET
) ||
8898 (reg_is_init_pkt_pointer(dst_reg
, PTR_TO_PACKET
) &&
8899 src_reg
->type
== PTR_TO_PACKET_META
)) {
8900 /* pkt_end <= pkt_data', pkt_data <= pkt_meta' */
8901 find_good_pkt_pointers(this_branch
, src_reg
,
8902 src_reg
->type
, true);
8903 mark_pkt_end(other_branch
, insn
->src_reg
, false);
8915 static void find_equal_scalars(struct bpf_verifier_state
*vstate
,
8916 struct bpf_reg_state
*known_reg
)
8918 struct bpf_func_state
*state
;
8919 struct bpf_reg_state
*reg
;
8922 for (i
= 0; i
<= vstate
->curframe
; i
++) {
8923 state
= vstate
->frame
[i
];
8924 for (j
= 0; j
< MAX_BPF_REG
; j
++) {
8925 reg
= &state
->regs
[j
];
8926 if (reg
->type
== SCALAR_VALUE
&& reg
->id
== known_reg
->id
)
8930 bpf_for_each_spilled_reg(j
, state
, reg
) {
8933 if (reg
->type
== SCALAR_VALUE
&& reg
->id
== known_reg
->id
)
8939 static int check_cond_jmp_op(struct bpf_verifier_env
*env
,
8940 struct bpf_insn
*insn
, int *insn_idx
)
8942 struct bpf_verifier_state
*this_branch
= env
->cur_state
;
8943 struct bpf_verifier_state
*other_branch
;
8944 struct bpf_reg_state
*regs
= this_branch
->frame
[this_branch
->curframe
]->regs
;
8945 struct bpf_reg_state
*dst_reg
, *other_branch_regs
, *src_reg
= NULL
;
8946 u8 opcode
= BPF_OP(insn
->code
);
8951 /* Only conditional jumps are expected to reach here. */
8952 if (opcode
== BPF_JA
|| opcode
> BPF_JSLE
) {
8953 verbose(env
, "invalid BPF_JMP/JMP32 opcode %x\n", opcode
);
8957 if (BPF_SRC(insn
->code
) == BPF_X
) {
8958 if (insn
->imm
!= 0) {
8959 verbose(env
, "BPF_JMP/JMP32 uses reserved fields\n");
8963 /* check src1 operand */
8964 err
= check_reg_arg(env
, insn
->src_reg
, SRC_OP
);
8968 if (is_pointer_value(env
, insn
->src_reg
)) {
8969 verbose(env
, "R%d pointer comparison prohibited\n",
8973 src_reg
= ®s
[insn
->src_reg
];
8975 if (insn
->src_reg
!= BPF_REG_0
) {
8976 verbose(env
, "BPF_JMP/JMP32 uses reserved fields\n");
8981 /* check src2 operand */
8982 err
= check_reg_arg(env
, insn
->dst_reg
, SRC_OP
);
8986 dst_reg
= ®s
[insn
->dst_reg
];
8987 is_jmp32
= BPF_CLASS(insn
->code
) == BPF_JMP32
;
8989 if (BPF_SRC(insn
->code
) == BPF_K
) {
8990 pred
= is_branch_taken(dst_reg
, insn
->imm
, opcode
, is_jmp32
);
8991 } else if (src_reg
->type
== SCALAR_VALUE
&&
8992 is_jmp32
&& tnum_is_const(tnum_subreg(src_reg
->var_off
))) {
8993 pred
= is_branch_taken(dst_reg
,
8994 tnum_subreg(src_reg
->var_off
).value
,
8997 } else if (src_reg
->type
== SCALAR_VALUE
&&
8998 !is_jmp32
&& tnum_is_const(src_reg
->var_off
)) {
8999 pred
= is_branch_taken(dst_reg
,
9000 src_reg
->var_off
.value
,
9003 } else if (reg_is_pkt_pointer_any(dst_reg
) &&
9004 reg_is_pkt_pointer_any(src_reg
) &&
9006 pred
= is_pkt_ptr_branch_taken(dst_reg
, src_reg
, opcode
);
9010 /* If we get here with a dst_reg pointer type it is because
9011 * above is_branch_taken() special cased the 0 comparison.
9013 if (!__is_pointer_value(false, dst_reg
))
9014 err
= mark_chain_precision(env
, insn
->dst_reg
);
9015 if (BPF_SRC(insn
->code
) == BPF_X
&& !err
&&
9016 !__is_pointer_value(false, src_reg
))
9017 err
= mark_chain_precision(env
, insn
->src_reg
);
9023 /* Only follow the goto, ignore fall-through. If needed, push
9024 * the fall-through branch for simulation under speculative
9027 if (!env
->bypass_spec_v1
&&
9028 !sanitize_speculative_path(env
, insn
, *insn_idx
+ 1,
9031 *insn_idx
+= insn
->off
;
9033 } else if (pred
== 0) {
9034 /* Only follow the fall-through branch, since that's where the
9035 * program will go. If needed, push the goto branch for
9036 * simulation under speculative execution.
9038 if (!env
->bypass_spec_v1
&&
9039 !sanitize_speculative_path(env
, insn
,
9040 *insn_idx
+ insn
->off
+ 1,
9046 other_branch
= push_stack(env
, *insn_idx
+ insn
->off
+ 1, *insn_idx
,
9050 other_branch_regs
= other_branch
->frame
[other_branch
->curframe
]->regs
;
9052 /* detect if we are comparing against a constant value so we can adjust
9053 * our min/max values for our dst register.
9054 * this is only legit if both are scalars (or pointers to the same
9055 * object, I suppose, but we don't support that right now), because
9056 * otherwise the different base pointers mean the offsets aren't
9059 if (BPF_SRC(insn
->code
) == BPF_X
) {
9060 struct bpf_reg_state
*src_reg
= ®s
[insn
->src_reg
];
9062 if (dst_reg
->type
== SCALAR_VALUE
&&
9063 src_reg
->type
== SCALAR_VALUE
) {
9064 if (tnum_is_const(src_reg
->var_off
) ||
9066 tnum_is_const(tnum_subreg(src_reg
->var_off
))))
9067 reg_set_min_max(&other_branch_regs
[insn
->dst_reg
],
9069 src_reg
->var_off
.value
,
9070 tnum_subreg(src_reg
->var_off
).value
,
9072 else if (tnum_is_const(dst_reg
->var_off
) ||
9074 tnum_is_const(tnum_subreg(dst_reg
->var_off
))))
9075 reg_set_min_max_inv(&other_branch_regs
[insn
->src_reg
],
9077 dst_reg
->var_off
.value
,
9078 tnum_subreg(dst_reg
->var_off
).value
,
9080 else if (!is_jmp32
&&
9081 (opcode
== BPF_JEQ
|| opcode
== BPF_JNE
))
9082 /* Comparing for equality, we can combine knowledge */
9083 reg_combine_min_max(&other_branch_regs
[insn
->src_reg
],
9084 &other_branch_regs
[insn
->dst_reg
],
9085 src_reg
, dst_reg
, opcode
);
9087 !WARN_ON_ONCE(src_reg
->id
!= other_branch_regs
[insn
->src_reg
].id
)) {
9088 find_equal_scalars(this_branch
, src_reg
);
9089 find_equal_scalars(other_branch
, &other_branch_regs
[insn
->src_reg
]);
9093 } else if (dst_reg
->type
== SCALAR_VALUE
) {
9094 reg_set_min_max(&other_branch_regs
[insn
->dst_reg
],
9095 dst_reg
, insn
->imm
, (u32
)insn
->imm
,
9099 if (dst_reg
->type
== SCALAR_VALUE
&& dst_reg
->id
&&
9100 !WARN_ON_ONCE(dst_reg
->id
!= other_branch_regs
[insn
->dst_reg
].id
)) {
9101 find_equal_scalars(this_branch
, dst_reg
);
9102 find_equal_scalars(other_branch
, &other_branch_regs
[insn
->dst_reg
]);
9105 /* detect if R == 0 where R is returned from bpf_map_lookup_elem().
9106 * NOTE: these optimizations below are related with pointer comparison
9107 * which will never be JMP32.
9109 if (!is_jmp32
&& BPF_SRC(insn
->code
) == BPF_K
&&
9110 insn
->imm
== 0 && (opcode
== BPF_JEQ
|| opcode
== BPF_JNE
) &&
9111 reg_type_may_be_null(dst_reg
->type
)) {
9112 /* Mark all identical registers in each branch as either
9113 * safe or unknown depending R == 0 or R != 0 conditional.
9115 mark_ptr_or_null_regs(this_branch
, insn
->dst_reg
,
9117 mark_ptr_or_null_regs(other_branch
, insn
->dst_reg
,
9119 } else if (!try_match_pkt_pointers(insn
, dst_reg
, ®s
[insn
->src_reg
],
9120 this_branch
, other_branch
) &&
9121 is_pointer_value(env
, insn
->dst_reg
)) {
9122 verbose(env
, "R%d pointer comparison prohibited\n",
9126 if (env
->log
.level
& BPF_LOG_LEVEL
)
9127 print_verifier_state(env
, this_branch
->frame
[this_branch
->curframe
]);
9131 /* verify BPF_LD_IMM64 instruction */
9132 static int check_ld_imm(struct bpf_verifier_env
*env
, struct bpf_insn
*insn
)
9134 struct bpf_insn_aux_data
*aux
= cur_aux(env
);
9135 struct bpf_reg_state
*regs
= cur_regs(env
);
9136 struct bpf_reg_state
*dst_reg
;
9137 struct bpf_map
*map
;
9140 if (BPF_SIZE(insn
->code
) != BPF_DW
) {
9141 verbose(env
, "invalid BPF_LD_IMM insn\n");
9144 if (insn
->off
!= 0) {
9145 verbose(env
, "BPF_LD_IMM64 uses reserved fields\n");
9149 err
= check_reg_arg(env
, insn
->dst_reg
, DST_OP
);
9153 dst_reg
= ®s
[insn
->dst_reg
];
9154 if (insn
->src_reg
== 0) {
9155 u64 imm
= ((u64
)(insn
+ 1)->imm
<< 32) | (u32
)insn
->imm
;
9157 dst_reg
->type
= SCALAR_VALUE
;
9158 __mark_reg_known(®s
[insn
->dst_reg
], imm
);
9162 if (insn
->src_reg
== BPF_PSEUDO_BTF_ID
) {
9163 mark_reg_known_zero(env
, regs
, insn
->dst_reg
);
9165 dst_reg
->type
= aux
->btf_var
.reg_type
;
9166 switch (dst_reg
->type
) {
9168 dst_reg
->mem_size
= aux
->btf_var
.mem_size
;
9171 case PTR_TO_PERCPU_BTF_ID
:
9172 dst_reg
->btf
= aux
->btf_var
.btf
;
9173 dst_reg
->btf_id
= aux
->btf_var
.btf_id
;
9176 verbose(env
, "bpf verifier is misconfigured\n");
9182 if (insn
->src_reg
== BPF_PSEUDO_FUNC
) {
9183 struct bpf_prog_aux
*aux
= env
->prog
->aux
;
9184 u32 subprogno
= insn
[1].imm
;
9186 if (!aux
->func_info
) {
9187 verbose(env
, "missing btf func_info\n");
9190 if (aux
->func_info_aux
[subprogno
].linkage
!= BTF_FUNC_STATIC
) {
9191 verbose(env
, "callback function not static\n");
9195 dst_reg
->type
= PTR_TO_FUNC
;
9196 dst_reg
->subprogno
= subprogno
;
9200 map
= env
->used_maps
[aux
->map_index
];
9201 mark_reg_known_zero(env
, regs
, insn
->dst_reg
);
9202 dst_reg
->map_ptr
= map
;
9204 if (insn
->src_reg
== BPF_PSEUDO_MAP_VALUE
||
9205 insn
->src_reg
== BPF_PSEUDO_MAP_IDX_VALUE
) {
9206 dst_reg
->type
= PTR_TO_MAP_VALUE
;
9207 dst_reg
->off
= aux
->map_off
;
9208 if (map_value_has_spin_lock(map
))
9209 dst_reg
->id
= ++env
->id_gen
;
9210 } else if (insn
->src_reg
== BPF_PSEUDO_MAP_FD
||
9211 insn
->src_reg
== BPF_PSEUDO_MAP_IDX
) {
9212 dst_reg
->type
= CONST_PTR_TO_MAP
;
9214 verbose(env
, "bpf verifier is misconfigured\n");
9221 static bool may_access_skb(enum bpf_prog_type type
)
9224 case BPF_PROG_TYPE_SOCKET_FILTER
:
9225 case BPF_PROG_TYPE_SCHED_CLS
:
9226 case BPF_PROG_TYPE_SCHED_ACT
:
9233 /* verify safety of LD_ABS|LD_IND instructions:
9234 * - they can only appear in the programs where ctx == skb
9235 * - since they are wrappers of function calls, they scratch R1-R5 registers,
9236 * preserve R6-R9, and store return value into R0
9239 * ctx == skb == R6 == CTX
9242 * SRC == any register
9243 * IMM == 32-bit immediate
9246 * R0 - 8/16/32-bit skb data converted to cpu endianness
9248 static int check_ld_abs(struct bpf_verifier_env
*env
, struct bpf_insn
*insn
)
9250 struct bpf_reg_state
*regs
= cur_regs(env
);
9251 static const int ctx_reg
= BPF_REG_6
;
9252 u8 mode
= BPF_MODE(insn
->code
);
9255 if (!may_access_skb(resolve_prog_type(env
->prog
))) {
9256 verbose(env
, "BPF_LD_[ABS|IND] instructions not allowed for this program type\n");
9260 if (!env
->ops
->gen_ld_abs
) {
9261 verbose(env
, "bpf verifier is misconfigured\n");
9265 if (insn
->dst_reg
!= BPF_REG_0
|| insn
->off
!= 0 ||
9266 BPF_SIZE(insn
->code
) == BPF_DW
||
9267 (mode
== BPF_ABS
&& insn
->src_reg
!= BPF_REG_0
)) {
9268 verbose(env
, "BPF_LD_[ABS|IND] uses reserved fields\n");
9272 /* check whether implicit source operand (register R6) is readable */
9273 err
= check_reg_arg(env
, ctx_reg
, SRC_OP
);
9277 /* Disallow usage of BPF_LD_[ABS|IND] with reference tracking, as
9278 * gen_ld_abs() may terminate the program at runtime, leading to
9281 err
= check_reference_leak(env
);
9283 verbose(env
, "BPF_LD_[ABS|IND] cannot be mixed with socket references\n");
9287 if (env
->cur_state
->active_spin_lock
) {
9288 verbose(env
, "BPF_LD_[ABS|IND] cannot be used inside bpf_spin_lock-ed region\n");
9292 if (regs
[ctx_reg
].type
!= PTR_TO_CTX
) {
9294 "at the time of BPF_LD_ABS|IND R6 != pointer to skb\n");
9298 if (mode
== BPF_IND
) {
9299 /* check explicit source operand */
9300 err
= check_reg_arg(env
, insn
->src_reg
, SRC_OP
);
9305 err
= check_ctx_reg(env
, ®s
[ctx_reg
], ctx_reg
);
9309 /* reset caller saved regs to unreadable */
9310 for (i
= 0; i
< CALLER_SAVED_REGS
; i
++) {
9311 mark_reg_not_init(env
, regs
, caller_saved
[i
]);
9312 check_reg_arg(env
, caller_saved
[i
], DST_OP_NO_MARK
);
9315 /* mark destination R0 register as readable, since it contains
9316 * the value fetched from the packet.
9317 * Already marked as written above.
9319 mark_reg_unknown(env
, regs
, BPF_REG_0
);
9320 /* ld_abs load up to 32-bit skb data. */
9321 regs
[BPF_REG_0
].subreg_def
= env
->insn_idx
+ 1;
9325 static int check_return_code(struct bpf_verifier_env
*env
)
9327 struct tnum enforce_attach_type_range
= tnum_unknown
;
9328 const struct bpf_prog
*prog
= env
->prog
;
9329 struct bpf_reg_state
*reg
;
9330 struct tnum range
= tnum_range(0, 1);
9331 enum bpf_prog_type prog_type
= resolve_prog_type(env
->prog
);
9333 struct bpf_func_state
*frame
= env
->cur_state
->frame
[0];
9334 const bool is_subprog
= frame
->subprogno
;
9336 /* LSM and struct_ops func-ptr's return type could be "void" */
9338 (prog_type
== BPF_PROG_TYPE_STRUCT_OPS
||
9339 prog_type
== BPF_PROG_TYPE_LSM
) &&
9340 !prog
->aux
->attach_func_proto
->type
)
9343 /* eBPF calling convention is such that R0 is used
9344 * to return the value from eBPF program.
9345 * Make sure that it's readable at this time
9346 * of bpf_exit, which means that program wrote
9347 * something into it earlier
9349 err
= check_reg_arg(env
, BPF_REG_0
, SRC_OP
);
9353 if (is_pointer_value(env
, BPF_REG_0
)) {
9354 verbose(env
, "R0 leaks addr as return value\n");
9358 reg
= cur_regs(env
) + BPF_REG_0
;
9360 if (frame
->in_async_callback_fn
) {
9361 /* enforce return zero from async callbacks like timer */
9362 if (reg
->type
!= SCALAR_VALUE
) {
9363 verbose(env
, "In async callback the register R0 is not a known value (%s)\n",
9364 reg_type_str
[reg
->type
]);
9368 if (!tnum_in(tnum_const(0), reg
->var_off
)) {
9369 verbose_invalid_scalar(env
, reg
, &range
, "async callback", "R0");
9376 if (reg
->type
!= SCALAR_VALUE
) {
9377 verbose(env
, "At subprogram exit the register R0 is not a scalar value (%s)\n",
9378 reg_type_str
[reg
->type
]);
9384 switch (prog_type
) {
9385 case BPF_PROG_TYPE_CGROUP_SOCK_ADDR
:
9386 if (env
->prog
->expected_attach_type
== BPF_CGROUP_UDP4_RECVMSG
||
9387 env
->prog
->expected_attach_type
== BPF_CGROUP_UDP6_RECVMSG
||
9388 env
->prog
->expected_attach_type
== BPF_CGROUP_INET4_GETPEERNAME
||
9389 env
->prog
->expected_attach_type
== BPF_CGROUP_INET6_GETPEERNAME
||
9390 env
->prog
->expected_attach_type
== BPF_CGROUP_INET4_GETSOCKNAME
||
9391 env
->prog
->expected_attach_type
== BPF_CGROUP_INET6_GETSOCKNAME
)
9392 range
= tnum_range(1, 1);
9393 if (env
->prog
->expected_attach_type
== BPF_CGROUP_INET4_BIND
||
9394 env
->prog
->expected_attach_type
== BPF_CGROUP_INET6_BIND
)
9395 range
= tnum_range(0, 3);
9397 case BPF_PROG_TYPE_CGROUP_SKB
:
9398 if (env
->prog
->expected_attach_type
== BPF_CGROUP_INET_EGRESS
) {
9399 range
= tnum_range(0, 3);
9400 enforce_attach_type_range
= tnum_range(2, 3);
9403 case BPF_PROG_TYPE_CGROUP_SOCK
:
9404 case BPF_PROG_TYPE_SOCK_OPS
:
9405 case BPF_PROG_TYPE_CGROUP_DEVICE
:
9406 case BPF_PROG_TYPE_CGROUP_SYSCTL
:
9407 case BPF_PROG_TYPE_CGROUP_SOCKOPT
:
9409 case BPF_PROG_TYPE_RAW_TRACEPOINT
:
9410 if (!env
->prog
->aux
->attach_btf_id
)
9412 range
= tnum_const(0);
9414 case BPF_PROG_TYPE_TRACING
:
9415 switch (env
->prog
->expected_attach_type
) {
9416 case BPF_TRACE_FENTRY
:
9417 case BPF_TRACE_FEXIT
:
9418 range
= tnum_const(0);
9420 case BPF_TRACE_RAW_TP
:
9421 case BPF_MODIFY_RETURN
:
9423 case BPF_TRACE_ITER
:
9429 case BPF_PROG_TYPE_SK_LOOKUP
:
9430 range
= tnum_range(SK_DROP
, SK_PASS
);
9432 case BPF_PROG_TYPE_EXT
:
9433 /* freplace program can return anything as its return value
9434 * depends on the to-be-replaced kernel func or bpf program.
9440 if (reg
->type
!= SCALAR_VALUE
) {
9441 verbose(env
, "At program exit the register R0 is not a known value (%s)\n",
9442 reg_type_str
[reg
->type
]);
9446 if (!tnum_in(range
, reg
->var_off
)) {
9447 verbose_invalid_scalar(env
, reg
, &range
, "program exit", "R0");
9451 if (!tnum_is_unknown(enforce_attach_type_range
) &&
9452 tnum_in(enforce_attach_type_range
, reg
->var_off
))
9453 env
->prog
->enforce_expected_attach_type
= 1;
9457 /* non-recursive DFS pseudo code
9458 * 1 procedure DFS-iterative(G,v):
9459 * 2 label v as discovered
9460 * 3 let S be a stack
9462 * 5 while S is not empty
9464 * 7 if t is what we're looking for:
9466 * 9 for all edges e in G.adjacentEdges(t) do
9467 * 10 if edge e is already labelled
9468 * 11 continue with the next edge
9469 * 12 w <- G.adjacentVertex(t,e)
9470 * 13 if vertex w is not discovered and not explored
9471 * 14 label e as tree-edge
9472 * 15 label w as discovered
9475 * 18 else if vertex w is discovered
9476 * 19 label e as back-edge
9478 * 21 // vertex w is explored
9479 * 22 label e as forward- or cross-edge
9480 * 23 label t as explored
9485 * 0x11 - discovered and fall-through edge labelled
9486 * 0x12 - discovered and fall-through and branch edges labelled
9497 static u32
state_htab_size(struct bpf_verifier_env
*env
)
9499 return env
->prog
->len
;
9502 static struct bpf_verifier_state_list
**explored_state(
9503 struct bpf_verifier_env
*env
,
9506 struct bpf_verifier_state
*cur
= env
->cur_state
;
9507 struct bpf_func_state
*state
= cur
->frame
[cur
->curframe
];
9509 return &env
->explored_states
[(idx
^ state
->callsite
) % state_htab_size(env
)];
9512 static void init_explored_state(struct bpf_verifier_env
*env
, int idx
)
9514 env
->insn_aux_data
[idx
].prune_point
= true;
9522 /* t, w, e - match pseudo-code above:
9523 * t - index of current instruction
9524 * w - next instruction
9527 static int push_insn(int t
, int w
, int e
, struct bpf_verifier_env
*env
,
9530 int *insn_stack
= env
->cfg
.insn_stack
;
9531 int *insn_state
= env
->cfg
.insn_state
;
9533 if (e
== FALLTHROUGH
&& insn_state
[t
] >= (DISCOVERED
| FALLTHROUGH
))
9534 return DONE_EXPLORING
;
9536 if (e
== BRANCH
&& insn_state
[t
] >= (DISCOVERED
| BRANCH
))
9537 return DONE_EXPLORING
;
9539 if (w
< 0 || w
>= env
->prog
->len
) {
9540 verbose_linfo(env
, t
, "%d: ", t
);
9541 verbose(env
, "jump out of range from insn %d to %d\n", t
, w
);
9546 /* mark branch target for state pruning */
9547 init_explored_state(env
, w
);
9549 if (insn_state
[w
] == 0) {
9551 insn_state
[t
] = DISCOVERED
| e
;
9552 insn_state
[w
] = DISCOVERED
;
9553 if (env
->cfg
.cur_stack
>= env
->prog
->len
)
9555 insn_stack
[env
->cfg
.cur_stack
++] = w
;
9556 return KEEP_EXPLORING
;
9557 } else if ((insn_state
[w
] & 0xF0) == DISCOVERED
) {
9558 if (loop_ok
&& env
->bpf_capable
)
9559 return DONE_EXPLORING
;
9560 verbose_linfo(env
, t
, "%d: ", t
);
9561 verbose_linfo(env
, w
, "%d: ", w
);
9562 verbose(env
, "back-edge from insn %d to %d\n", t
, w
);
9564 } else if (insn_state
[w
] == EXPLORED
) {
9565 /* forward- or cross-edge */
9566 insn_state
[t
] = DISCOVERED
| e
;
9568 verbose(env
, "insn state internal bug\n");
9571 return DONE_EXPLORING
;
9574 static int visit_func_call_insn(int t
, int insn_cnt
,
9575 struct bpf_insn
*insns
,
9576 struct bpf_verifier_env
*env
,
9581 ret
= push_insn(t
, t
+ 1, FALLTHROUGH
, env
, false);
9585 if (t
+ 1 < insn_cnt
)
9586 init_explored_state(env
, t
+ 1);
9588 init_explored_state(env
, t
);
9589 ret
= push_insn(t
, t
+ insns
[t
].imm
+ 1, BRANCH
, env
,
9590 /* It's ok to allow recursion from CFG point of
9591 * view. __check_func_call() will do the actual
9594 bpf_pseudo_func(insns
+ t
));
9599 /* Visits the instruction at index t and returns one of the following:
9600 * < 0 - an error occurred
9601 * DONE_EXPLORING - the instruction was fully explored
9602 * KEEP_EXPLORING - there is still work to be done before it is fully explored
9604 static int visit_insn(int t
, int insn_cnt
, struct bpf_verifier_env
*env
)
9606 struct bpf_insn
*insns
= env
->prog
->insnsi
;
9609 if (bpf_pseudo_func(insns
+ t
))
9610 return visit_func_call_insn(t
, insn_cnt
, insns
, env
, true);
9612 /* All non-branch instructions have a single fall-through edge. */
9613 if (BPF_CLASS(insns
[t
].code
) != BPF_JMP
&&
9614 BPF_CLASS(insns
[t
].code
) != BPF_JMP32
)
9615 return push_insn(t
, t
+ 1, FALLTHROUGH
, env
, false);
9617 switch (BPF_OP(insns
[t
].code
)) {
9619 return DONE_EXPLORING
;
9622 if (insns
[t
].imm
== BPF_FUNC_timer_set_callback
)
9623 /* Mark this call insn to trigger is_state_visited() check
9624 * before call itself is processed by __check_func_call().
9625 * Otherwise new async state will be pushed for further
9628 init_explored_state(env
, t
);
9629 return visit_func_call_insn(t
, insn_cnt
, insns
, env
,
9630 insns
[t
].src_reg
== BPF_PSEUDO_CALL
);
9633 if (BPF_SRC(insns
[t
].code
) != BPF_K
)
9636 /* unconditional jump with single edge */
9637 ret
= push_insn(t
, t
+ insns
[t
].off
+ 1, FALLTHROUGH
, env
,
9642 /* unconditional jmp is not a good pruning point,
9643 * but it's marked, since backtracking needs
9644 * to record jmp history in is_state_visited().
9646 init_explored_state(env
, t
+ insns
[t
].off
+ 1);
9647 /* tell verifier to check for equivalent states
9648 * after every call and jump
9650 if (t
+ 1 < insn_cnt
)
9651 init_explored_state(env
, t
+ 1);
9656 /* conditional jump with two edges */
9657 init_explored_state(env
, t
);
9658 ret
= push_insn(t
, t
+ 1, FALLTHROUGH
, env
, true);
9662 return push_insn(t
, t
+ insns
[t
].off
+ 1, BRANCH
, env
, true);
9666 /* non-recursive depth-first-search to detect loops in BPF program
9667 * loop == back-edge in directed graph
9669 static int check_cfg(struct bpf_verifier_env
*env
)
9671 int insn_cnt
= env
->prog
->len
;
9672 int *insn_stack
, *insn_state
;
9676 insn_state
= env
->cfg
.insn_state
= kvcalloc(insn_cnt
, sizeof(int), GFP_KERNEL
);
9680 insn_stack
= env
->cfg
.insn_stack
= kvcalloc(insn_cnt
, sizeof(int), GFP_KERNEL
);
9686 insn_state
[0] = DISCOVERED
; /* mark 1st insn as discovered */
9687 insn_stack
[0] = 0; /* 0 is the first instruction */
9688 env
->cfg
.cur_stack
= 1;
9690 while (env
->cfg
.cur_stack
> 0) {
9691 int t
= insn_stack
[env
->cfg
.cur_stack
- 1];
9693 ret
= visit_insn(t
, insn_cnt
, env
);
9695 case DONE_EXPLORING
:
9696 insn_state
[t
] = EXPLORED
;
9697 env
->cfg
.cur_stack
--;
9699 case KEEP_EXPLORING
:
9703 verbose(env
, "visit_insn internal bug\n");
9710 if (env
->cfg
.cur_stack
< 0) {
9711 verbose(env
, "pop stack internal bug\n");
9716 for (i
= 0; i
< insn_cnt
; i
++) {
9717 if (insn_state
[i
] != EXPLORED
) {
9718 verbose(env
, "unreachable insn %d\n", i
);
9723 ret
= 0; /* cfg looks good */
9728 env
->cfg
.insn_state
= env
->cfg
.insn_stack
= NULL
;
9732 static int check_abnormal_return(struct bpf_verifier_env
*env
)
9736 for (i
= 1; i
< env
->subprog_cnt
; i
++) {
9737 if (env
->subprog_info
[i
].has_ld_abs
) {
9738 verbose(env
, "LD_ABS is not allowed in subprogs without BTF\n");
9741 if (env
->subprog_info
[i
].has_tail_call
) {
9742 verbose(env
, "tail_call is not allowed in subprogs without BTF\n");
9749 /* The minimum supported BTF func info size */
9750 #define MIN_BPF_FUNCINFO_SIZE 8
9751 #define MAX_FUNCINFO_REC_SIZE 252
9753 static int check_btf_func(struct bpf_verifier_env
*env
,
9754 const union bpf_attr
*attr
,
9757 const struct btf_type
*type
, *func_proto
, *ret_type
;
9758 u32 i
, nfuncs
, urec_size
, min_size
;
9759 u32 krec_size
= sizeof(struct bpf_func_info
);
9760 struct bpf_func_info
*krecord
;
9761 struct bpf_func_info_aux
*info_aux
= NULL
;
9762 struct bpf_prog
*prog
;
9763 const struct btf
*btf
;
9765 u32 prev_offset
= 0;
9769 nfuncs
= attr
->func_info_cnt
;
9771 if (check_abnormal_return(env
))
9776 if (nfuncs
!= env
->subprog_cnt
) {
9777 verbose(env
, "number of funcs in func_info doesn't match number of subprogs\n");
9781 urec_size
= attr
->func_info_rec_size
;
9782 if (urec_size
< MIN_BPF_FUNCINFO_SIZE
||
9783 urec_size
> MAX_FUNCINFO_REC_SIZE
||
9784 urec_size
% sizeof(u32
)) {
9785 verbose(env
, "invalid func info rec size %u\n", urec_size
);
9790 btf
= prog
->aux
->btf
;
9792 urecord
= make_bpfptr(attr
->func_info
, uattr
.is_kernel
);
9793 min_size
= min_t(u32
, krec_size
, urec_size
);
9795 krecord
= kvcalloc(nfuncs
, krec_size
, GFP_KERNEL
| __GFP_NOWARN
);
9798 info_aux
= kcalloc(nfuncs
, sizeof(*info_aux
), GFP_KERNEL
| __GFP_NOWARN
);
9802 for (i
= 0; i
< nfuncs
; i
++) {
9803 ret
= bpf_check_uarg_tail_zero(urecord
, krec_size
, urec_size
);
9805 if (ret
== -E2BIG
) {
9806 verbose(env
, "nonzero tailing record in func info");
9807 /* set the size kernel expects so loader can zero
9808 * out the rest of the record.
9810 if (copy_to_bpfptr_offset(uattr
,
9811 offsetof(union bpf_attr
, func_info_rec_size
),
9812 &min_size
, sizeof(min_size
)))
9818 if (copy_from_bpfptr(&krecord
[i
], urecord
, min_size
)) {
9823 /* check insn_off */
9826 if (krecord
[i
].insn_off
) {
9828 "nonzero insn_off %u for the first func info record",
9829 krecord
[i
].insn_off
);
9832 } else if (krecord
[i
].insn_off
<= prev_offset
) {
9834 "same or smaller insn offset (%u) than previous func info record (%u)",
9835 krecord
[i
].insn_off
, prev_offset
);
9839 if (env
->subprog_info
[i
].start
!= krecord
[i
].insn_off
) {
9840 verbose(env
, "func_info BTF section doesn't match subprog layout in BPF program\n");
9845 type
= btf_type_by_id(btf
, krecord
[i
].type_id
);
9846 if (!type
|| !btf_type_is_func(type
)) {
9847 verbose(env
, "invalid type id %d in func info",
9848 krecord
[i
].type_id
);
9851 info_aux
[i
].linkage
= BTF_INFO_VLEN(type
->info
);
9853 func_proto
= btf_type_by_id(btf
, type
->type
);
9854 if (unlikely(!func_proto
|| !btf_type_is_func_proto(func_proto
)))
9855 /* btf_func_check() already verified it during BTF load */
9857 ret_type
= btf_type_skip_modifiers(btf
, func_proto
->type
, NULL
);
9859 btf_type_is_small_int(ret_type
) || btf_type_is_enum(ret_type
);
9860 if (i
&& !scalar_return
&& env
->subprog_info
[i
].has_ld_abs
) {
9861 verbose(env
, "LD_ABS is only allowed in functions that return 'int'.\n");
9864 if (i
&& !scalar_return
&& env
->subprog_info
[i
].has_tail_call
) {
9865 verbose(env
, "tail_call is only allowed in functions that return 'int'.\n");
9869 prev_offset
= krecord
[i
].insn_off
;
9870 bpfptr_add(&urecord
, urec_size
);
9873 prog
->aux
->func_info
= krecord
;
9874 prog
->aux
->func_info_cnt
= nfuncs
;
9875 prog
->aux
->func_info_aux
= info_aux
;
9884 static void adjust_btf_func(struct bpf_verifier_env
*env
)
9886 struct bpf_prog_aux
*aux
= env
->prog
->aux
;
9889 if (!aux
->func_info
)
9892 for (i
= 0; i
< env
->subprog_cnt
; i
++)
9893 aux
->func_info
[i
].insn_off
= env
->subprog_info
[i
].start
;
9896 #define MIN_BPF_LINEINFO_SIZE (offsetof(struct bpf_line_info, line_col) + \
9897 sizeof(((struct bpf_line_info *)(0))->line_col))
9898 #define MAX_LINEINFO_REC_SIZE MAX_FUNCINFO_REC_SIZE
9900 static int check_btf_line(struct bpf_verifier_env
*env
,
9901 const union bpf_attr
*attr
,
9904 u32 i
, s
, nr_linfo
, ncopy
, expected_size
, rec_size
, prev_offset
= 0;
9905 struct bpf_subprog_info
*sub
;
9906 struct bpf_line_info
*linfo
;
9907 struct bpf_prog
*prog
;
9908 const struct btf
*btf
;
9912 nr_linfo
= attr
->line_info_cnt
;
9915 if (nr_linfo
> INT_MAX
/ sizeof(struct bpf_line_info
))
9918 rec_size
= attr
->line_info_rec_size
;
9919 if (rec_size
< MIN_BPF_LINEINFO_SIZE
||
9920 rec_size
> MAX_LINEINFO_REC_SIZE
||
9921 rec_size
& (sizeof(u32
) - 1))
9924 /* Need to zero it in case the userspace may
9925 * pass in a smaller bpf_line_info object.
9927 linfo
= kvcalloc(nr_linfo
, sizeof(struct bpf_line_info
),
9928 GFP_KERNEL
| __GFP_NOWARN
);
9933 btf
= prog
->aux
->btf
;
9936 sub
= env
->subprog_info
;
9937 ulinfo
= make_bpfptr(attr
->line_info
, uattr
.is_kernel
);
9938 expected_size
= sizeof(struct bpf_line_info
);
9939 ncopy
= min_t(u32
, expected_size
, rec_size
);
9940 for (i
= 0; i
< nr_linfo
; i
++) {
9941 err
= bpf_check_uarg_tail_zero(ulinfo
, expected_size
, rec_size
);
9943 if (err
== -E2BIG
) {
9944 verbose(env
, "nonzero tailing record in line_info");
9945 if (copy_to_bpfptr_offset(uattr
,
9946 offsetof(union bpf_attr
, line_info_rec_size
),
9947 &expected_size
, sizeof(expected_size
)))
9953 if (copy_from_bpfptr(&linfo
[i
], ulinfo
, ncopy
)) {
9959 * Check insn_off to ensure
9960 * 1) strictly increasing AND
9961 * 2) bounded by prog->len
9963 * The linfo[0].insn_off == 0 check logically falls into
9964 * the later "missing bpf_line_info for func..." case
9965 * because the first linfo[0].insn_off must be the
9966 * first sub also and the first sub must have
9967 * subprog_info[0].start == 0.
9969 if ((i
&& linfo
[i
].insn_off
<= prev_offset
) ||
9970 linfo
[i
].insn_off
>= prog
->len
) {
9971 verbose(env
, "Invalid line_info[%u].insn_off:%u (prev_offset:%u prog->len:%u)\n",
9972 i
, linfo
[i
].insn_off
, prev_offset
,
9978 if (!prog
->insnsi
[linfo
[i
].insn_off
].code
) {
9980 "Invalid insn code at line_info[%u].insn_off\n",
9986 if (!btf_name_by_offset(btf
, linfo
[i
].line_off
) ||
9987 !btf_name_by_offset(btf
, linfo
[i
].file_name_off
)) {
9988 verbose(env
, "Invalid line_info[%u].line_off or .file_name_off\n", i
);
9993 if (s
!= env
->subprog_cnt
) {
9994 if (linfo
[i
].insn_off
== sub
[s
].start
) {
9995 sub
[s
].linfo_idx
= i
;
9997 } else if (sub
[s
].start
< linfo
[i
].insn_off
) {
9998 verbose(env
, "missing bpf_line_info for func#%u\n", s
);
10004 prev_offset
= linfo
[i
].insn_off
;
10005 bpfptr_add(&ulinfo
, rec_size
);
10008 if (s
!= env
->subprog_cnt
) {
10009 verbose(env
, "missing bpf_line_info for %u funcs starting from func#%u\n",
10010 env
->subprog_cnt
- s
, s
);
10015 prog
->aux
->linfo
= linfo
;
10016 prog
->aux
->nr_linfo
= nr_linfo
;
10025 static int check_btf_info(struct bpf_verifier_env
*env
,
10026 const union bpf_attr
*attr
,
10032 if (!attr
->func_info_cnt
&& !attr
->line_info_cnt
) {
10033 if (check_abnormal_return(env
))
10038 btf
= btf_get_by_fd(attr
->prog_btf_fd
);
10040 return PTR_ERR(btf
);
10041 if (btf_is_kernel(btf
)) {
10045 env
->prog
->aux
->btf
= btf
;
10047 err
= check_btf_func(env
, attr
, uattr
);
10051 err
= check_btf_line(env
, attr
, uattr
);
10058 /* check %cur's range satisfies %old's */
10059 static bool range_within(struct bpf_reg_state
*old
,
10060 struct bpf_reg_state
*cur
)
10062 return old
->umin_value
<= cur
->umin_value
&&
10063 old
->umax_value
>= cur
->umax_value
&&
10064 old
->smin_value
<= cur
->smin_value
&&
10065 old
->smax_value
>= cur
->smax_value
&&
10066 old
->u32_min_value
<= cur
->u32_min_value
&&
10067 old
->u32_max_value
>= cur
->u32_max_value
&&
10068 old
->s32_min_value
<= cur
->s32_min_value
&&
10069 old
->s32_max_value
>= cur
->s32_max_value
;
10072 /* If in the old state two registers had the same id, then they need to have
10073 * the same id in the new state as well. But that id could be different from
10074 * the old state, so we need to track the mapping from old to new ids.
10075 * Once we have seen that, say, a reg with old id 5 had new id 9, any subsequent
10076 * regs with old id 5 must also have new id 9 for the new state to be safe. But
10077 * regs with a different old id could still have new id 9, we don't care about
10079 * So we look through our idmap to see if this old id has been seen before. If
10080 * so, we require the new id to match; otherwise, we add the id pair to the map.
10082 static bool check_ids(u32 old_id
, u32 cur_id
, struct bpf_id_pair
*idmap
)
10086 for (i
= 0; i
< BPF_ID_MAP_SIZE
; i
++) {
10087 if (!idmap
[i
].old
) {
10088 /* Reached an empty slot; haven't seen this id before */
10089 idmap
[i
].old
= old_id
;
10090 idmap
[i
].cur
= cur_id
;
10093 if (idmap
[i
].old
== old_id
)
10094 return idmap
[i
].cur
== cur_id
;
10096 /* We ran out of idmap slots, which should be impossible */
10101 static void clean_func_state(struct bpf_verifier_env
*env
,
10102 struct bpf_func_state
*st
)
10104 enum bpf_reg_liveness live
;
10107 for (i
= 0; i
< BPF_REG_FP
; i
++) {
10108 live
= st
->regs
[i
].live
;
10109 /* liveness must not touch this register anymore */
10110 st
->regs
[i
].live
|= REG_LIVE_DONE
;
10111 if (!(live
& REG_LIVE_READ
))
10112 /* since the register is unused, clear its state
10113 * to make further comparison simpler
10115 __mark_reg_not_init(env
, &st
->regs
[i
]);
10118 for (i
= 0; i
< st
->allocated_stack
/ BPF_REG_SIZE
; i
++) {
10119 live
= st
->stack
[i
].spilled_ptr
.live
;
10120 /* liveness must not touch this stack slot anymore */
10121 st
->stack
[i
].spilled_ptr
.live
|= REG_LIVE_DONE
;
10122 if (!(live
& REG_LIVE_READ
)) {
10123 __mark_reg_not_init(env
, &st
->stack
[i
].spilled_ptr
);
10124 for (j
= 0; j
< BPF_REG_SIZE
; j
++)
10125 st
->stack
[i
].slot_type
[j
] = STACK_INVALID
;
10130 static void clean_verifier_state(struct bpf_verifier_env
*env
,
10131 struct bpf_verifier_state
*st
)
10135 if (st
->frame
[0]->regs
[0].live
& REG_LIVE_DONE
)
10136 /* all regs in this state in all frames were already marked */
10139 for (i
= 0; i
<= st
->curframe
; i
++)
10140 clean_func_state(env
, st
->frame
[i
]);
10143 /* the parentage chains form a tree.
10144 * the verifier states are added to state lists at given insn and
10145 * pushed into state stack for future exploration.
10146 * when the verifier reaches bpf_exit insn some of the verifer states
10147 * stored in the state lists have their final liveness state already,
10148 * but a lot of states will get revised from liveness point of view when
10149 * the verifier explores other branches.
10152 * 2: if r1 == 100 goto pc+1
10155 * when the verifier reaches exit insn the register r0 in the state list of
10156 * insn 2 will be seen as !REG_LIVE_READ. Then the verifier pops the other_branch
10157 * of insn 2 and goes exploring further. At the insn 4 it will walk the
10158 * parentage chain from insn 4 into insn 2 and will mark r0 as REG_LIVE_READ.
10160 * Since the verifier pushes the branch states as it sees them while exploring
10161 * the program the condition of walking the branch instruction for the second
10162 * time means that all states below this branch were already explored and
10163 * their final liveness marks are already propagated.
10164 * Hence when the verifier completes the search of state list in is_state_visited()
10165 * we can call this clean_live_states() function to mark all liveness states
10166 * as REG_LIVE_DONE to indicate that 'parent' pointers of 'struct bpf_reg_state'
10167 * will not be used.
10168 * This function also clears the registers and stack for states that !READ
10169 * to simplify state merging.
10171 * Important note here that walking the same branch instruction in the callee
10172 * doesn't meant that the states are DONE. The verifier has to compare
10175 static void clean_live_states(struct bpf_verifier_env
*env
, int insn
,
10176 struct bpf_verifier_state
*cur
)
10178 struct bpf_verifier_state_list
*sl
;
10181 sl
= *explored_state(env
, insn
);
10183 if (sl
->state
.branches
)
10185 if (sl
->state
.insn_idx
!= insn
||
10186 sl
->state
.curframe
!= cur
->curframe
)
10188 for (i
= 0; i
<= cur
->curframe
; i
++)
10189 if (sl
->state
.frame
[i
]->callsite
!= cur
->frame
[i
]->callsite
)
10191 clean_verifier_state(env
, &sl
->state
);
10197 /* Returns true if (rold safe implies rcur safe) */
10198 static bool regsafe(struct bpf_verifier_env
*env
, struct bpf_reg_state
*rold
,
10199 struct bpf_reg_state
*rcur
, struct bpf_id_pair
*idmap
)
10203 if (!(rold
->live
& REG_LIVE_READ
))
10204 /* explored state didn't use this */
10207 equal
= memcmp(rold
, rcur
, offsetof(struct bpf_reg_state
, parent
)) == 0;
10209 if (rold
->type
== PTR_TO_STACK
)
10210 /* two stack pointers are equal only if they're pointing to
10211 * the same stack frame, since fp-8 in foo != fp-8 in bar
10213 return equal
&& rold
->frameno
== rcur
->frameno
;
10218 if (rold
->type
== NOT_INIT
)
10219 /* explored state can't have used this */
10221 if (rcur
->type
== NOT_INIT
)
10223 switch (rold
->type
) {
10225 if (env
->explore_alu_limits
)
10227 if (rcur
->type
== SCALAR_VALUE
) {
10228 if (!rold
->precise
&& !rcur
->precise
)
10230 /* new val must satisfy old val knowledge */
10231 return range_within(rold
, rcur
) &&
10232 tnum_in(rold
->var_off
, rcur
->var_off
);
10234 /* We're trying to use a pointer in place of a scalar.
10235 * Even if the scalar was unbounded, this could lead to
10236 * pointer leaks because scalars are allowed to leak
10237 * while pointers are not. We could make this safe in
10238 * special cases if root is calling us, but it's
10239 * probably not worth the hassle.
10243 case PTR_TO_MAP_KEY
:
10244 case PTR_TO_MAP_VALUE
:
10245 /* If the new min/max/var_off satisfy the old ones and
10246 * everything else matches, we are OK.
10247 * 'id' is not compared, since it's only used for maps with
10248 * bpf_spin_lock inside map element and in such cases if
10249 * the rest of the prog is valid for one map element then
10250 * it's valid for all map elements regardless of the key
10251 * used in bpf_map_lookup()
10253 return memcmp(rold
, rcur
, offsetof(struct bpf_reg_state
, id
)) == 0 &&
10254 range_within(rold
, rcur
) &&
10255 tnum_in(rold
->var_off
, rcur
->var_off
);
10256 case PTR_TO_MAP_VALUE_OR_NULL
:
10257 /* a PTR_TO_MAP_VALUE could be safe to use as a
10258 * PTR_TO_MAP_VALUE_OR_NULL into the same map.
10259 * However, if the old PTR_TO_MAP_VALUE_OR_NULL then got NULL-
10260 * checked, doing so could have affected others with the same
10261 * id, and we can't check for that because we lost the id when
10262 * we converted to a PTR_TO_MAP_VALUE.
10264 if (rcur
->type
!= PTR_TO_MAP_VALUE_OR_NULL
)
10266 if (memcmp(rold
, rcur
, offsetof(struct bpf_reg_state
, id
)))
10268 /* Check our ids match any regs they're supposed to */
10269 return check_ids(rold
->id
, rcur
->id
, idmap
);
10270 case PTR_TO_PACKET_META
:
10271 case PTR_TO_PACKET
:
10272 if (rcur
->type
!= rold
->type
)
10274 /* We must have at least as much range as the old ptr
10275 * did, so that any accesses which were safe before are
10276 * still safe. This is true even if old range < old off,
10277 * since someone could have accessed through (ptr - k), or
10278 * even done ptr -= k in a register, to get a safe access.
10280 if (rold
->range
> rcur
->range
)
10282 /* If the offsets don't match, we can't trust our alignment;
10283 * nor can we be sure that we won't fall out of range.
10285 if (rold
->off
!= rcur
->off
)
10287 /* id relations must be preserved */
10288 if (rold
->id
&& !check_ids(rold
->id
, rcur
->id
, idmap
))
10290 /* new val must satisfy old val knowledge */
10291 return range_within(rold
, rcur
) &&
10292 tnum_in(rold
->var_off
, rcur
->var_off
);
10294 case CONST_PTR_TO_MAP
:
10295 case PTR_TO_PACKET_END
:
10296 case PTR_TO_FLOW_KEYS
:
10297 case PTR_TO_SOCKET
:
10298 case PTR_TO_SOCKET_OR_NULL
:
10299 case PTR_TO_SOCK_COMMON
:
10300 case PTR_TO_SOCK_COMMON_OR_NULL
:
10301 case PTR_TO_TCP_SOCK
:
10302 case PTR_TO_TCP_SOCK_OR_NULL
:
10303 case PTR_TO_XDP_SOCK
:
10304 /* Only valid matches are exact, which memcmp() above
10305 * would have accepted
10308 /* Don't know what's going on, just say it's not safe */
10312 /* Shouldn't get here; if we do, say it's not safe */
10317 static bool stacksafe(struct bpf_verifier_env
*env
, struct bpf_func_state
*old
,
10318 struct bpf_func_state
*cur
, struct bpf_id_pair
*idmap
)
10322 /* walk slots of the explored stack and ignore any additional
10323 * slots in the current stack, since explored(safe) state
10326 for (i
= 0; i
< old
->allocated_stack
; i
++) {
10327 spi
= i
/ BPF_REG_SIZE
;
10329 if (!(old
->stack
[spi
].spilled_ptr
.live
& REG_LIVE_READ
)) {
10330 i
+= BPF_REG_SIZE
- 1;
10331 /* explored state didn't use this */
10335 if (old
->stack
[spi
].slot_type
[i
% BPF_REG_SIZE
] == STACK_INVALID
)
10338 /* explored stack has more populated slots than current stack
10339 * and these slots were used
10341 if (i
>= cur
->allocated_stack
)
10344 /* if old state was safe with misc data in the stack
10345 * it will be safe with zero-initialized stack.
10346 * The opposite is not true
10348 if (old
->stack
[spi
].slot_type
[i
% BPF_REG_SIZE
] == STACK_MISC
&&
10349 cur
->stack
[spi
].slot_type
[i
% BPF_REG_SIZE
] == STACK_ZERO
)
10351 if (old
->stack
[spi
].slot_type
[i
% BPF_REG_SIZE
] !=
10352 cur
->stack
[spi
].slot_type
[i
% BPF_REG_SIZE
])
10353 /* Ex: old explored (safe) state has STACK_SPILL in
10354 * this stack slot, but current has STACK_MISC ->
10355 * this verifier states are not equivalent,
10356 * return false to continue verification of this path
10359 if (i
% BPF_REG_SIZE
)
10361 if (old
->stack
[spi
].slot_type
[0] != STACK_SPILL
)
10363 if (!regsafe(env
, &old
->stack
[spi
].spilled_ptr
,
10364 &cur
->stack
[spi
].spilled_ptr
, idmap
))
10365 /* when explored and current stack slot are both storing
10366 * spilled registers, check that stored pointers types
10367 * are the same as well.
10368 * Ex: explored safe path could have stored
10369 * (bpf_reg_state) {.type = PTR_TO_STACK, .off = -8}
10370 * but current path has stored:
10371 * (bpf_reg_state) {.type = PTR_TO_STACK, .off = -16}
10372 * such verifier states are not equivalent.
10373 * return false to continue verification of this path
10380 static bool refsafe(struct bpf_func_state
*old
, struct bpf_func_state
*cur
)
10382 if (old
->acquired_refs
!= cur
->acquired_refs
)
10384 return !memcmp(old
->refs
, cur
->refs
,
10385 sizeof(*old
->refs
) * old
->acquired_refs
);
10388 /* compare two verifier states
10390 * all states stored in state_list are known to be valid, since
10391 * verifier reached 'bpf_exit' instruction through them
10393 * this function is called when verifier exploring different branches of
10394 * execution popped from the state stack. If it sees an old state that has
10395 * more strict register state and more strict stack state then this execution
10396 * branch doesn't need to be explored further, since verifier already
10397 * concluded that more strict state leads to valid finish.
10399 * Therefore two states are equivalent if register state is more conservative
10400 * and explored stack state is more conservative than the current one.
10403 * (slot1=INV slot2=MISC) == (slot1=MISC slot2=MISC)
10404 * (slot1=MISC slot2=MISC) != (slot1=INV slot2=MISC)
10406 * In other words if current stack state (one being explored) has more
10407 * valid slots than old one that already passed validation, it means
10408 * the verifier can stop exploring and conclude that current state is valid too
10410 * Similarly with registers. If explored state has register type as invalid
10411 * whereas register type in current state is meaningful, it means that
10412 * the current state will reach 'bpf_exit' instruction safely
10414 static bool func_states_equal(struct bpf_verifier_env
*env
, struct bpf_func_state
*old
,
10415 struct bpf_func_state
*cur
)
10419 memset(env
->idmap_scratch
, 0, sizeof(env
->idmap_scratch
));
10420 for (i
= 0; i
< MAX_BPF_REG
; i
++)
10421 if (!regsafe(env
, &old
->regs
[i
], &cur
->regs
[i
],
10422 env
->idmap_scratch
))
10425 if (!stacksafe(env
, old
, cur
, env
->idmap_scratch
))
10428 if (!refsafe(old
, cur
))
10434 static bool states_equal(struct bpf_verifier_env
*env
,
10435 struct bpf_verifier_state
*old
,
10436 struct bpf_verifier_state
*cur
)
10440 if (old
->curframe
!= cur
->curframe
)
10443 /* Verification state from speculative execution simulation
10444 * must never prune a non-speculative execution one.
10446 if (old
->speculative
&& !cur
->speculative
)
10449 if (old
->active_spin_lock
!= cur
->active_spin_lock
)
10452 /* for states to be equal callsites have to be the same
10453 * and all frame states need to be equivalent
10455 for (i
= 0; i
<= old
->curframe
; i
++) {
10456 if (old
->frame
[i
]->callsite
!= cur
->frame
[i
]->callsite
)
10458 if (!func_states_equal(env
, old
->frame
[i
], cur
->frame
[i
]))
10464 /* Return 0 if no propagation happened. Return negative error code if error
10465 * happened. Otherwise, return the propagated bit.
10467 static int propagate_liveness_reg(struct bpf_verifier_env
*env
,
10468 struct bpf_reg_state
*reg
,
10469 struct bpf_reg_state
*parent_reg
)
10471 u8 parent_flag
= parent_reg
->live
& REG_LIVE_READ
;
10472 u8 flag
= reg
->live
& REG_LIVE_READ
;
10475 /* When comes here, read flags of PARENT_REG or REG could be any of
10476 * REG_LIVE_READ64, REG_LIVE_READ32, REG_LIVE_NONE. There is no need
10477 * of propagation if PARENT_REG has strongest REG_LIVE_READ64.
10479 if (parent_flag
== REG_LIVE_READ64
||
10480 /* Or if there is no read flag from REG. */
10482 /* Or if the read flag from REG is the same as PARENT_REG. */
10483 parent_flag
== flag
)
10486 err
= mark_reg_read(env
, reg
, parent_reg
, flag
);
10493 /* A write screens off any subsequent reads; but write marks come from the
10494 * straight-line code between a state and its parent. When we arrive at an
10495 * equivalent state (jump target or such) we didn't arrive by the straight-line
10496 * code, so read marks in the state must propagate to the parent regardless
10497 * of the state's write marks. That's what 'parent == state->parent' comparison
10498 * in mark_reg_read() is for.
10500 static int propagate_liveness(struct bpf_verifier_env
*env
,
10501 const struct bpf_verifier_state
*vstate
,
10502 struct bpf_verifier_state
*vparent
)
10504 struct bpf_reg_state
*state_reg
, *parent_reg
;
10505 struct bpf_func_state
*state
, *parent
;
10506 int i
, frame
, err
= 0;
10508 if (vparent
->curframe
!= vstate
->curframe
) {
10509 WARN(1, "propagate_live: parent frame %d current frame %d\n",
10510 vparent
->curframe
, vstate
->curframe
);
10513 /* Propagate read liveness of registers... */
10514 BUILD_BUG_ON(BPF_REG_FP
+ 1 != MAX_BPF_REG
);
10515 for (frame
= 0; frame
<= vstate
->curframe
; frame
++) {
10516 parent
= vparent
->frame
[frame
];
10517 state
= vstate
->frame
[frame
];
10518 parent_reg
= parent
->regs
;
10519 state_reg
= state
->regs
;
10520 /* We don't need to worry about FP liveness, it's read-only */
10521 for (i
= frame
< vstate
->curframe
? BPF_REG_6
: 0; i
< BPF_REG_FP
; i
++) {
10522 err
= propagate_liveness_reg(env
, &state_reg
[i
],
10526 if (err
== REG_LIVE_READ64
)
10527 mark_insn_zext(env
, &parent_reg
[i
]);
10530 /* Propagate stack slots. */
10531 for (i
= 0; i
< state
->allocated_stack
/ BPF_REG_SIZE
&&
10532 i
< parent
->allocated_stack
/ BPF_REG_SIZE
; i
++) {
10533 parent_reg
= &parent
->stack
[i
].spilled_ptr
;
10534 state_reg
= &state
->stack
[i
].spilled_ptr
;
10535 err
= propagate_liveness_reg(env
, state_reg
,
10544 /* find precise scalars in the previous equivalent state and
10545 * propagate them into the current state
10547 static int propagate_precision(struct bpf_verifier_env
*env
,
10548 const struct bpf_verifier_state
*old
)
10550 struct bpf_reg_state
*state_reg
;
10551 struct bpf_func_state
*state
;
10554 state
= old
->frame
[old
->curframe
];
10555 state_reg
= state
->regs
;
10556 for (i
= 0; i
< BPF_REG_FP
; i
++, state_reg
++) {
10557 if (state_reg
->type
!= SCALAR_VALUE
||
10558 !state_reg
->precise
)
10560 if (env
->log
.level
& BPF_LOG_LEVEL2
)
10561 verbose(env
, "propagating r%d\n", i
);
10562 err
= mark_chain_precision(env
, i
);
10567 for (i
= 0; i
< state
->allocated_stack
/ BPF_REG_SIZE
; i
++) {
10568 if (state
->stack
[i
].slot_type
[0] != STACK_SPILL
)
10570 state_reg
= &state
->stack
[i
].spilled_ptr
;
10571 if (state_reg
->type
!= SCALAR_VALUE
||
10572 !state_reg
->precise
)
10574 if (env
->log
.level
& BPF_LOG_LEVEL2
)
10575 verbose(env
, "propagating fp%d\n",
10576 (-i
- 1) * BPF_REG_SIZE
);
10577 err
= mark_chain_precision_stack(env
, i
);
10584 static bool states_maybe_looping(struct bpf_verifier_state
*old
,
10585 struct bpf_verifier_state
*cur
)
10587 struct bpf_func_state
*fold
, *fcur
;
10588 int i
, fr
= cur
->curframe
;
10590 if (old
->curframe
!= fr
)
10593 fold
= old
->frame
[fr
];
10594 fcur
= cur
->frame
[fr
];
10595 for (i
= 0; i
< MAX_BPF_REG
; i
++)
10596 if (memcmp(&fold
->regs
[i
], &fcur
->regs
[i
],
10597 offsetof(struct bpf_reg_state
, parent
)))
10603 static int is_state_visited(struct bpf_verifier_env
*env
, int insn_idx
)
10605 struct bpf_verifier_state_list
*new_sl
;
10606 struct bpf_verifier_state_list
*sl
, **pprev
;
10607 struct bpf_verifier_state
*cur
= env
->cur_state
, *new;
10608 int i
, j
, err
, states_cnt
= 0;
10609 bool add_new_state
= env
->test_state_freq
? true : false;
10611 cur
->last_insn_idx
= env
->prev_insn_idx
;
10612 if (!env
->insn_aux_data
[insn_idx
].prune_point
)
10613 /* this 'insn_idx' instruction wasn't marked, so we will not
10614 * be doing state search here
10618 /* bpf progs typically have pruning point every 4 instructions
10619 * http://vger.kernel.org/bpfconf2019.html#session-1
10620 * Do not add new state for future pruning if the verifier hasn't seen
10621 * at least 2 jumps and at least 8 instructions.
10622 * This heuristics helps decrease 'total_states' and 'peak_states' metric.
10623 * In tests that amounts to up to 50% reduction into total verifier
10624 * memory consumption and 20% verifier time speedup.
10626 if (env
->jmps_processed
- env
->prev_jmps_processed
>= 2 &&
10627 env
->insn_processed
- env
->prev_insn_processed
>= 8)
10628 add_new_state
= true;
10630 pprev
= explored_state(env
, insn_idx
);
10633 clean_live_states(env
, insn_idx
, cur
);
10637 if (sl
->state
.insn_idx
!= insn_idx
)
10640 if (sl
->state
.branches
) {
10641 struct bpf_func_state
*frame
= sl
->state
.frame
[sl
->state
.curframe
];
10643 if (frame
->in_async_callback_fn
&&
10644 frame
->async_entry_cnt
!= cur
->frame
[cur
->curframe
]->async_entry_cnt
) {
10645 /* Different async_entry_cnt means that the verifier is
10646 * processing another entry into async callback.
10647 * Seeing the same state is not an indication of infinite
10648 * loop or infinite recursion.
10649 * But finding the same state doesn't mean that it's safe
10650 * to stop processing the current state. The previous state
10651 * hasn't yet reached bpf_exit, since state.branches > 0.
10652 * Checking in_async_callback_fn alone is not enough either.
10653 * Since the verifier still needs to catch infinite loops
10654 * inside async callbacks.
10656 } else if (states_maybe_looping(&sl
->state
, cur
) &&
10657 states_equal(env
, &sl
->state
, cur
)) {
10658 verbose_linfo(env
, insn_idx
, "; ");
10659 verbose(env
, "infinite loop detected at insn %d\n", insn_idx
);
10662 /* if the verifier is processing a loop, avoid adding new state
10663 * too often, since different loop iterations have distinct
10664 * states and may not help future pruning.
10665 * This threshold shouldn't be too low to make sure that
10666 * a loop with large bound will be rejected quickly.
10667 * The most abusive loop will be:
10669 * if r1 < 1000000 goto pc-2
10670 * 1M insn_procssed limit / 100 == 10k peak states.
10671 * This threshold shouldn't be too high either, since states
10672 * at the end of the loop are likely to be useful in pruning.
10674 if (env
->jmps_processed
- env
->prev_jmps_processed
< 20 &&
10675 env
->insn_processed
- env
->prev_insn_processed
< 100)
10676 add_new_state
= false;
10679 if (states_equal(env
, &sl
->state
, cur
)) {
10681 /* reached equivalent register/stack state,
10682 * prune the search.
10683 * Registers read by the continuation are read by us.
10684 * If we have any write marks in env->cur_state, they
10685 * will prevent corresponding reads in the continuation
10686 * from reaching our parent (an explored_state). Our
10687 * own state will get the read marks recorded, but
10688 * they'll be immediately forgotten as we're pruning
10689 * this state and will pop a new one.
10691 err
= propagate_liveness(env
, &sl
->state
, cur
);
10693 /* if previous state reached the exit with precision and
10694 * current state is equivalent to it (except precsion marks)
10695 * the precision needs to be propagated back in
10696 * the current state.
10698 err
= err
? : push_jmp_history(env
, cur
);
10699 err
= err
? : propagate_precision(env
, &sl
->state
);
10705 /* when new state is not going to be added do not increase miss count.
10706 * Otherwise several loop iterations will remove the state
10707 * recorded earlier. The goal of these heuristics is to have
10708 * states from some iterations of the loop (some in the beginning
10709 * and some at the end) to help pruning.
10713 /* heuristic to determine whether this state is beneficial
10714 * to keep checking from state equivalence point of view.
10715 * Higher numbers increase max_states_per_insn and verification time,
10716 * but do not meaningfully decrease insn_processed.
10718 if (sl
->miss_cnt
> sl
->hit_cnt
* 3 + 3) {
10719 /* the state is unlikely to be useful. Remove it to
10720 * speed up verification
10723 if (sl
->state
.frame
[0]->regs
[0].live
& REG_LIVE_DONE
) {
10724 u32 br
= sl
->state
.branches
;
10727 "BUG live_done but branches_to_explore %d\n",
10729 free_verifier_state(&sl
->state
, false);
10731 env
->peak_states
--;
10733 /* cannot free this state, since parentage chain may
10734 * walk it later. Add it for free_list instead to
10735 * be freed at the end of verification
10737 sl
->next
= env
->free_list
;
10738 env
->free_list
= sl
;
10748 if (env
->max_states_per_insn
< states_cnt
)
10749 env
->max_states_per_insn
= states_cnt
;
10751 if (!env
->bpf_capable
&& states_cnt
> BPF_COMPLEXITY_LIMIT_STATES
)
10752 return push_jmp_history(env
, cur
);
10754 if (!add_new_state
)
10755 return push_jmp_history(env
, cur
);
10757 /* There were no equivalent states, remember the current one.
10758 * Technically the current state is not proven to be safe yet,
10759 * but it will either reach outer most bpf_exit (which means it's safe)
10760 * or it will be rejected. When there are no loops the verifier won't be
10761 * seeing this tuple (frame[0].callsite, frame[1].callsite, .. insn_idx)
10762 * again on the way to bpf_exit.
10763 * When looping the sl->state.branches will be > 0 and this state
10764 * will not be considered for equivalence until branches == 0.
10766 new_sl
= kzalloc(sizeof(struct bpf_verifier_state_list
), GFP_KERNEL
);
10769 env
->total_states
++;
10770 env
->peak_states
++;
10771 env
->prev_jmps_processed
= env
->jmps_processed
;
10772 env
->prev_insn_processed
= env
->insn_processed
;
10774 /* add new state to the head of linked list */
10775 new = &new_sl
->state
;
10776 err
= copy_verifier_state(new, cur
);
10778 free_verifier_state(new, false);
10782 new->insn_idx
= insn_idx
;
10783 WARN_ONCE(new->branches
!= 1,
10784 "BUG is_state_visited:branches_to_explore=%d insn %d\n", new->branches
, insn_idx
);
10787 cur
->first_insn_idx
= insn_idx
;
10788 clear_jmp_history(cur
);
10789 new_sl
->next
= *explored_state(env
, insn_idx
);
10790 *explored_state(env
, insn_idx
) = new_sl
;
10791 /* connect new state to parentage chain. Current frame needs all
10792 * registers connected. Only r6 - r9 of the callers are alive (pushed
10793 * to the stack implicitly by JITs) so in callers' frames connect just
10794 * r6 - r9 as an optimization. Callers will have r1 - r5 connected to
10795 * the state of the call instruction (with WRITTEN set), and r0 comes
10796 * from callee with its full parentage chain, anyway.
10798 /* clear write marks in current state: the writes we did are not writes
10799 * our child did, so they don't screen off its reads from us.
10800 * (There are no read marks in current state, because reads always mark
10801 * their parent and current state never has children yet. Only
10802 * explored_states can get read marks.)
10804 for (j
= 0; j
<= cur
->curframe
; j
++) {
10805 for (i
= j
< cur
->curframe
? BPF_REG_6
: 0; i
< BPF_REG_FP
; i
++)
10806 cur
->frame
[j
]->regs
[i
].parent
= &new->frame
[j
]->regs
[i
];
10807 for (i
= 0; i
< BPF_REG_FP
; i
++)
10808 cur
->frame
[j
]->regs
[i
].live
= REG_LIVE_NONE
;
10811 /* all stack frames are accessible from callee, clear them all */
10812 for (j
= 0; j
<= cur
->curframe
; j
++) {
10813 struct bpf_func_state
*frame
= cur
->frame
[j
];
10814 struct bpf_func_state
*newframe
= new->frame
[j
];
10816 for (i
= 0; i
< frame
->allocated_stack
/ BPF_REG_SIZE
; i
++) {
10817 frame
->stack
[i
].spilled_ptr
.live
= REG_LIVE_NONE
;
10818 frame
->stack
[i
].spilled_ptr
.parent
=
10819 &newframe
->stack
[i
].spilled_ptr
;
10825 /* Return true if it's OK to have the same insn return a different type. */
10826 static bool reg_type_mismatch_ok(enum bpf_reg_type type
)
10830 case PTR_TO_SOCKET
:
10831 case PTR_TO_SOCKET_OR_NULL
:
10832 case PTR_TO_SOCK_COMMON
:
10833 case PTR_TO_SOCK_COMMON_OR_NULL
:
10834 case PTR_TO_TCP_SOCK
:
10835 case PTR_TO_TCP_SOCK_OR_NULL
:
10836 case PTR_TO_XDP_SOCK
:
10837 case PTR_TO_BTF_ID
:
10838 case PTR_TO_BTF_ID_OR_NULL
:
10845 /* If an instruction was previously used with particular pointer types, then we
10846 * need to be careful to avoid cases such as the below, where it may be ok
10847 * for one branch accessing the pointer, but not ok for the other branch:
10852 * R1 = some_other_valid_ptr;
10855 * R2 = *(u32 *)(R1 + 0);
10857 static bool reg_type_mismatch(enum bpf_reg_type src
, enum bpf_reg_type prev
)
10859 return src
!= prev
&& (!reg_type_mismatch_ok(src
) ||
10860 !reg_type_mismatch_ok(prev
));
10863 static int do_check(struct bpf_verifier_env
*env
)
10865 bool pop_log
= !(env
->log
.level
& BPF_LOG_LEVEL2
);
10866 struct bpf_verifier_state
*state
= env
->cur_state
;
10867 struct bpf_insn
*insns
= env
->prog
->insnsi
;
10868 struct bpf_reg_state
*regs
;
10869 int insn_cnt
= env
->prog
->len
;
10870 bool do_print_state
= false;
10871 int prev_insn_idx
= -1;
10874 struct bpf_insn
*insn
;
10878 env
->prev_insn_idx
= prev_insn_idx
;
10879 if (env
->insn_idx
>= insn_cnt
) {
10880 verbose(env
, "invalid insn idx %d insn_cnt %d\n",
10881 env
->insn_idx
, insn_cnt
);
10885 insn
= &insns
[env
->insn_idx
];
10886 class = BPF_CLASS(insn
->code
);
10888 if (++env
->insn_processed
> BPF_COMPLEXITY_LIMIT_INSNS
) {
10890 "BPF program is too large. Processed %d insn\n",
10891 env
->insn_processed
);
10895 err
= is_state_visited(env
, env
->insn_idx
);
10899 /* found equivalent state, can prune the search */
10900 if (env
->log
.level
& BPF_LOG_LEVEL
) {
10901 if (do_print_state
)
10902 verbose(env
, "\nfrom %d to %d%s: safe\n",
10903 env
->prev_insn_idx
, env
->insn_idx
,
10904 env
->cur_state
->speculative
?
10905 " (speculative execution)" : "");
10907 verbose(env
, "%d: safe\n", env
->insn_idx
);
10909 goto process_bpf_exit
;
10912 if (signal_pending(current
))
10915 if (need_resched())
10918 if (env
->log
.level
& BPF_LOG_LEVEL2
||
10919 (env
->log
.level
& BPF_LOG_LEVEL
&& do_print_state
)) {
10920 if (env
->log
.level
& BPF_LOG_LEVEL2
)
10921 verbose(env
, "%d:", env
->insn_idx
);
10923 verbose(env
, "\nfrom %d to %d%s:",
10924 env
->prev_insn_idx
, env
->insn_idx
,
10925 env
->cur_state
->speculative
?
10926 " (speculative execution)" : "");
10927 print_verifier_state(env
, state
->frame
[state
->curframe
]);
10928 do_print_state
= false;
10931 if (env
->log
.level
& BPF_LOG_LEVEL
) {
10932 const struct bpf_insn_cbs cbs
= {
10933 .cb_call
= disasm_kfunc_name
,
10934 .cb_print
= verbose
,
10935 .private_data
= env
,
10938 verbose_linfo(env
, env
->insn_idx
, "; ");
10939 verbose(env
, "%d: ", env
->insn_idx
);
10940 print_bpf_insn(&cbs
, insn
, env
->allow_ptr_leaks
);
10943 if (bpf_prog_is_dev_bound(env
->prog
->aux
)) {
10944 err
= bpf_prog_offload_verify_insn(env
, env
->insn_idx
,
10945 env
->prev_insn_idx
);
10950 regs
= cur_regs(env
);
10951 sanitize_mark_insn_seen(env
);
10952 prev_insn_idx
= env
->insn_idx
;
10954 if (class == BPF_ALU
|| class == BPF_ALU64
) {
10955 err
= check_alu_op(env
, insn
);
10959 } else if (class == BPF_LDX
) {
10960 enum bpf_reg_type
*prev_src_type
, src_reg_type
;
10962 /* check for reserved fields is already done */
10964 /* check src operand */
10965 err
= check_reg_arg(env
, insn
->src_reg
, SRC_OP
);
10969 err
= check_reg_arg(env
, insn
->dst_reg
, DST_OP_NO_MARK
);
10973 src_reg_type
= regs
[insn
->src_reg
].type
;
10975 /* check that memory (src_reg + off) is readable,
10976 * the state of dst_reg will be updated by this func
10978 err
= check_mem_access(env
, env
->insn_idx
, insn
->src_reg
,
10979 insn
->off
, BPF_SIZE(insn
->code
),
10980 BPF_READ
, insn
->dst_reg
, false);
10984 prev_src_type
= &env
->insn_aux_data
[env
->insn_idx
].ptr_type
;
10986 if (*prev_src_type
== NOT_INIT
) {
10987 /* saw a valid insn
10988 * dst_reg = *(u32 *)(src_reg + off)
10989 * save type to validate intersecting paths
10991 *prev_src_type
= src_reg_type
;
10993 } else if (reg_type_mismatch(src_reg_type
, *prev_src_type
)) {
10994 /* ABuser program is trying to use the same insn
10995 * dst_reg = *(u32*) (src_reg + off)
10996 * with different pointer types:
10997 * src_reg == ctx in one branch and
10998 * src_reg == stack|map in some other branch.
11001 verbose(env
, "same insn cannot be used with different pointers\n");
11005 } else if (class == BPF_STX
) {
11006 enum bpf_reg_type
*prev_dst_type
, dst_reg_type
;
11008 if (BPF_MODE(insn
->code
) == BPF_ATOMIC
) {
11009 err
= check_atomic(env
, env
->insn_idx
, insn
);
11016 if (BPF_MODE(insn
->code
) != BPF_MEM
|| insn
->imm
!= 0) {
11017 verbose(env
, "BPF_STX uses reserved fields\n");
11021 /* check src1 operand */
11022 err
= check_reg_arg(env
, insn
->src_reg
, SRC_OP
);
11025 /* check src2 operand */
11026 err
= check_reg_arg(env
, insn
->dst_reg
, SRC_OP
);
11030 dst_reg_type
= regs
[insn
->dst_reg
].type
;
11032 /* check that memory (dst_reg + off) is writeable */
11033 err
= check_mem_access(env
, env
->insn_idx
, insn
->dst_reg
,
11034 insn
->off
, BPF_SIZE(insn
->code
),
11035 BPF_WRITE
, insn
->src_reg
, false);
11039 prev_dst_type
= &env
->insn_aux_data
[env
->insn_idx
].ptr_type
;
11041 if (*prev_dst_type
== NOT_INIT
) {
11042 *prev_dst_type
= dst_reg_type
;
11043 } else if (reg_type_mismatch(dst_reg_type
, *prev_dst_type
)) {
11044 verbose(env
, "same insn cannot be used with different pointers\n");
11048 } else if (class == BPF_ST
) {
11049 if (BPF_MODE(insn
->code
) != BPF_MEM
||
11050 insn
->src_reg
!= BPF_REG_0
) {
11051 verbose(env
, "BPF_ST uses reserved fields\n");
11054 /* check src operand */
11055 err
= check_reg_arg(env
, insn
->dst_reg
, SRC_OP
);
11059 if (is_ctx_reg(env
, insn
->dst_reg
)) {
11060 verbose(env
, "BPF_ST stores into R%d %s is not allowed\n",
11062 reg_type_str
[reg_state(env
, insn
->dst_reg
)->type
]);
11066 /* check that memory (dst_reg + off) is writeable */
11067 err
= check_mem_access(env
, env
->insn_idx
, insn
->dst_reg
,
11068 insn
->off
, BPF_SIZE(insn
->code
),
11069 BPF_WRITE
, -1, false);
11073 } else if (class == BPF_JMP
|| class == BPF_JMP32
) {
11074 u8 opcode
= BPF_OP(insn
->code
);
11076 env
->jmps_processed
++;
11077 if (opcode
== BPF_CALL
) {
11078 if (BPF_SRC(insn
->code
) != BPF_K
||
11080 (insn
->src_reg
!= BPF_REG_0
&&
11081 insn
->src_reg
!= BPF_PSEUDO_CALL
&&
11082 insn
->src_reg
!= BPF_PSEUDO_KFUNC_CALL
) ||
11083 insn
->dst_reg
!= BPF_REG_0
||
11084 class == BPF_JMP32
) {
11085 verbose(env
, "BPF_CALL uses reserved fields\n");
11089 if (env
->cur_state
->active_spin_lock
&&
11090 (insn
->src_reg
== BPF_PSEUDO_CALL
||
11091 insn
->imm
!= BPF_FUNC_spin_unlock
)) {
11092 verbose(env
, "function calls are not allowed while holding a lock\n");
11095 if (insn
->src_reg
== BPF_PSEUDO_CALL
)
11096 err
= check_func_call(env
, insn
, &env
->insn_idx
);
11097 else if (insn
->src_reg
== BPF_PSEUDO_KFUNC_CALL
)
11098 err
= check_kfunc_call(env
, insn
);
11100 err
= check_helper_call(env
, insn
, &env
->insn_idx
);
11103 } else if (opcode
== BPF_JA
) {
11104 if (BPF_SRC(insn
->code
) != BPF_K
||
11106 insn
->src_reg
!= BPF_REG_0
||
11107 insn
->dst_reg
!= BPF_REG_0
||
11108 class == BPF_JMP32
) {
11109 verbose(env
, "BPF_JA uses reserved fields\n");
11113 env
->insn_idx
+= insn
->off
+ 1;
11116 } else if (opcode
== BPF_EXIT
) {
11117 if (BPF_SRC(insn
->code
) != BPF_K
||
11119 insn
->src_reg
!= BPF_REG_0
||
11120 insn
->dst_reg
!= BPF_REG_0
||
11121 class == BPF_JMP32
) {
11122 verbose(env
, "BPF_EXIT uses reserved fields\n");
11126 if (env
->cur_state
->active_spin_lock
) {
11127 verbose(env
, "bpf_spin_unlock is missing\n");
11131 if (state
->curframe
) {
11132 /* exit from nested function */
11133 err
= prepare_func_exit(env
, &env
->insn_idx
);
11136 do_print_state
= true;
11140 err
= check_reference_leak(env
);
11144 err
= check_return_code(env
);
11148 update_branch_counts(env
, env
->cur_state
);
11149 err
= pop_stack(env
, &prev_insn_idx
,
11150 &env
->insn_idx
, pop_log
);
11152 if (err
!= -ENOENT
)
11156 do_print_state
= true;
11160 err
= check_cond_jmp_op(env
, insn
, &env
->insn_idx
);
11164 } else if (class == BPF_LD
) {
11165 u8 mode
= BPF_MODE(insn
->code
);
11167 if (mode
== BPF_ABS
|| mode
== BPF_IND
) {
11168 err
= check_ld_abs(env
, insn
);
11172 } else if (mode
== BPF_IMM
) {
11173 err
= check_ld_imm(env
, insn
);
11178 sanitize_mark_insn_seen(env
);
11180 verbose(env
, "invalid BPF_LD mode\n");
11184 verbose(env
, "unknown insn class %d\n", class);
11194 static int find_btf_percpu_datasec(struct btf
*btf
)
11196 const struct btf_type
*t
;
11201 * Both vmlinux and module each have their own ".data..percpu"
11202 * DATASECs in BTF. So for module's case, we need to skip vmlinux BTF
11203 * types to look at only module's own BTF types.
11205 n
= btf_nr_types(btf
);
11206 if (btf_is_module(btf
))
11207 i
= btf_nr_types(btf_vmlinux
);
11211 for(; i
< n
; i
++) {
11212 t
= btf_type_by_id(btf
, i
);
11213 if (BTF_INFO_KIND(t
->info
) != BTF_KIND_DATASEC
)
11216 tname
= btf_name_by_offset(btf
, t
->name_off
);
11217 if (!strcmp(tname
, ".data..percpu"))
11224 /* replace pseudo btf_id with kernel symbol address */
11225 static int check_pseudo_btf_id(struct bpf_verifier_env
*env
,
11226 struct bpf_insn
*insn
,
11227 struct bpf_insn_aux_data
*aux
)
11229 const struct btf_var_secinfo
*vsi
;
11230 const struct btf_type
*datasec
;
11231 struct btf_mod_pair
*btf_mod
;
11232 const struct btf_type
*t
;
11233 const char *sym_name
;
11234 bool percpu
= false;
11235 u32 type
, id
= insn
->imm
;
11239 int i
, btf_fd
, err
;
11241 btf_fd
= insn
[1].imm
;
11243 btf
= btf_get_by_fd(btf_fd
);
11245 verbose(env
, "invalid module BTF object FD specified.\n");
11249 if (!btf_vmlinux
) {
11250 verbose(env
, "kernel is missing BTF, make sure CONFIG_DEBUG_INFO_BTF=y is specified in Kconfig.\n");
11257 t
= btf_type_by_id(btf
, id
);
11259 verbose(env
, "ldimm64 insn specifies invalid btf_id %d.\n", id
);
11264 if (!btf_type_is_var(t
)) {
11265 verbose(env
, "pseudo btf_id %d in ldimm64 isn't KIND_VAR.\n", id
);
11270 sym_name
= btf_name_by_offset(btf
, t
->name_off
);
11271 addr
= kallsyms_lookup_name(sym_name
);
11273 verbose(env
, "ldimm64 failed to find the address for kernel symbol '%s'.\n",
11279 datasec_id
= find_btf_percpu_datasec(btf
);
11280 if (datasec_id
> 0) {
11281 datasec
= btf_type_by_id(btf
, datasec_id
);
11282 for_each_vsi(i
, datasec
, vsi
) {
11283 if (vsi
->type
== id
) {
11290 insn
[0].imm
= (u32
)addr
;
11291 insn
[1].imm
= addr
>> 32;
11294 t
= btf_type_skip_modifiers(btf
, type
, NULL
);
11296 aux
->btf_var
.reg_type
= PTR_TO_PERCPU_BTF_ID
;
11297 aux
->btf_var
.btf
= btf
;
11298 aux
->btf_var
.btf_id
= type
;
11299 } else if (!btf_type_is_struct(t
)) {
11300 const struct btf_type
*ret
;
11304 /* resolve the type size of ksym. */
11305 ret
= btf_resolve_size(btf
, t
, &tsize
);
11307 tname
= btf_name_by_offset(btf
, t
->name_off
);
11308 verbose(env
, "ldimm64 unable to resolve the size of type '%s': %ld\n",
11309 tname
, PTR_ERR(ret
));
11313 aux
->btf_var
.reg_type
= PTR_TO_MEM
;
11314 aux
->btf_var
.mem_size
= tsize
;
11316 aux
->btf_var
.reg_type
= PTR_TO_BTF_ID
;
11317 aux
->btf_var
.btf
= btf
;
11318 aux
->btf_var
.btf_id
= type
;
11321 /* check whether we recorded this BTF (and maybe module) already */
11322 for (i
= 0; i
< env
->used_btf_cnt
; i
++) {
11323 if (env
->used_btfs
[i
].btf
== btf
) {
11329 if (env
->used_btf_cnt
>= MAX_USED_BTFS
) {
11334 btf_mod
= &env
->used_btfs
[env
->used_btf_cnt
];
11335 btf_mod
->btf
= btf
;
11336 btf_mod
->module
= NULL
;
11338 /* if we reference variables from kernel module, bump its refcount */
11339 if (btf_is_module(btf
)) {
11340 btf_mod
->module
= btf_try_get_module(btf
);
11341 if (!btf_mod
->module
) {
11347 env
->used_btf_cnt
++;
11355 static int check_map_prealloc(struct bpf_map
*map
)
11357 return (map
->map_type
!= BPF_MAP_TYPE_HASH
&&
11358 map
->map_type
!= BPF_MAP_TYPE_PERCPU_HASH
&&
11359 map
->map_type
!= BPF_MAP_TYPE_HASH_OF_MAPS
) ||
11360 !(map
->map_flags
& BPF_F_NO_PREALLOC
);
11363 static bool is_tracing_prog_type(enum bpf_prog_type type
)
11366 case BPF_PROG_TYPE_KPROBE
:
11367 case BPF_PROG_TYPE_TRACEPOINT
:
11368 case BPF_PROG_TYPE_PERF_EVENT
:
11369 case BPF_PROG_TYPE_RAW_TRACEPOINT
:
11376 static bool is_preallocated_map(struct bpf_map
*map
)
11378 if (!check_map_prealloc(map
))
11380 if (map
->inner_map_meta
&& !check_map_prealloc(map
->inner_map_meta
))
11385 static int check_map_prog_compatibility(struct bpf_verifier_env
*env
,
11386 struct bpf_map
*map
,
11387 struct bpf_prog
*prog
)
11390 enum bpf_prog_type prog_type
= resolve_prog_type(prog
);
11392 * Validate that trace type programs use preallocated hash maps.
11394 * For programs attached to PERF events this is mandatory as the
11395 * perf NMI can hit any arbitrary code sequence.
11397 * All other trace types using preallocated hash maps are unsafe as
11398 * well because tracepoint or kprobes can be inside locked regions
11399 * of the memory allocator or at a place where a recursion into the
11400 * memory allocator would see inconsistent state.
11402 * On RT enabled kernels run-time allocation of all trace type
11403 * programs is strictly prohibited due to lock type constraints. On
11404 * !RT kernels it is allowed for backwards compatibility reasons for
11405 * now, but warnings are emitted so developers are made aware of
11406 * the unsafety and can fix their programs before this is enforced.
11408 if (is_tracing_prog_type(prog_type
) && !is_preallocated_map(map
)) {
11409 if (prog_type
== BPF_PROG_TYPE_PERF_EVENT
) {
11410 verbose(env
, "perf_event programs can only use preallocated hash map\n");
11413 if (IS_ENABLED(CONFIG_PREEMPT_RT
)) {
11414 verbose(env
, "trace type programs can only use preallocated hash map\n");
11417 WARN_ONCE(1, "trace type BPF program uses run-time allocation\n");
11418 verbose(env
, "trace type programs with run-time allocated hash maps are unsafe. Switch to preallocated hash maps.\n");
11421 if (map_value_has_spin_lock(map
)) {
11422 if (prog_type
== BPF_PROG_TYPE_SOCKET_FILTER
) {
11423 verbose(env
, "socket filter progs cannot use bpf_spin_lock yet\n");
11427 if (is_tracing_prog_type(prog_type
)) {
11428 verbose(env
, "tracing progs cannot use bpf_spin_lock yet\n");
11432 if (prog
->aux
->sleepable
) {
11433 verbose(env
, "sleepable progs cannot use bpf_spin_lock yet\n");
11438 if ((bpf_prog_is_dev_bound(prog
->aux
) || bpf_map_is_dev_bound(map
)) &&
11439 !bpf_offload_prog_map_match(prog
, map
)) {
11440 verbose(env
, "offload device mismatch between prog and map\n");
11444 if (map
->map_type
== BPF_MAP_TYPE_STRUCT_OPS
) {
11445 verbose(env
, "bpf_struct_ops map cannot be used in prog\n");
11449 if (prog
->aux
->sleepable
)
11450 switch (map
->map_type
) {
11451 case BPF_MAP_TYPE_HASH
:
11452 case BPF_MAP_TYPE_LRU_HASH
:
11453 case BPF_MAP_TYPE_ARRAY
:
11454 case BPF_MAP_TYPE_PERCPU_HASH
:
11455 case BPF_MAP_TYPE_PERCPU_ARRAY
:
11456 case BPF_MAP_TYPE_LRU_PERCPU_HASH
:
11457 case BPF_MAP_TYPE_ARRAY_OF_MAPS
:
11458 case BPF_MAP_TYPE_HASH_OF_MAPS
:
11459 if (!is_preallocated_map(map
)) {
11461 "Sleepable programs can only use preallocated maps\n");
11465 case BPF_MAP_TYPE_RINGBUF
:
11469 "Sleepable programs can only use array, hash, and ringbuf maps\n");
11476 static bool bpf_map_is_cgroup_storage(struct bpf_map
*map
)
11478 return (map
->map_type
== BPF_MAP_TYPE_CGROUP_STORAGE
||
11479 map
->map_type
== BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE
);
11482 /* find and rewrite pseudo imm in ld_imm64 instructions:
11484 * 1. if it accesses map FD, replace it with actual map pointer.
11485 * 2. if it accesses btf_id of a VAR, replace it with pointer to the var.
11487 * NOTE: btf_vmlinux is required for converting pseudo btf_id.
11489 static int resolve_pseudo_ldimm64(struct bpf_verifier_env
*env
)
11491 struct bpf_insn
*insn
= env
->prog
->insnsi
;
11492 int insn_cnt
= env
->prog
->len
;
11495 err
= bpf_prog_calc_tag(env
->prog
);
11499 for (i
= 0; i
< insn_cnt
; i
++, insn
++) {
11500 if (BPF_CLASS(insn
->code
) == BPF_LDX
&&
11501 (BPF_MODE(insn
->code
) != BPF_MEM
|| insn
->imm
!= 0)) {
11502 verbose(env
, "BPF_LDX uses reserved fields\n");
11506 if (insn
[0].code
== (BPF_LD
| BPF_IMM
| BPF_DW
)) {
11507 struct bpf_insn_aux_data
*aux
;
11508 struct bpf_map
*map
;
11513 if (i
== insn_cnt
- 1 || insn
[1].code
!= 0 ||
11514 insn
[1].dst_reg
!= 0 || insn
[1].src_reg
!= 0 ||
11515 insn
[1].off
!= 0) {
11516 verbose(env
, "invalid bpf_ld_imm64 insn\n");
11520 if (insn
[0].src_reg
== 0)
11521 /* valid generic load 64-bit imm */
11524 if (insn
[0].src_reg
== BPF_PSEUDO_BTF_ID
) {
11525 aux
= &env
->insn_aux_data
[i
];
11526 err
= check_pseudo_btf_id(env
, insn
, aux
);
11532 if (insn
[0].src_reg
== BPF_PSEUDO_FUNC
) {
11533 aux
= &env
->insn_aux_data
[i
];
11534 aux
->ptr_type
= PTR_TO_FUNC
;
11538 /* In final convert_pseudo_ld_imm64() step, this is
11539 * converted into regular 64-bit imm load insn.
11541 switch (insn
[0].src_reg
) {
11542 case BPF_PSEUDO_MAP_VALUE
:
11543 case BPF_PSEUDO_MAP_IDX_VALUE
:
11545 case BPF_PSEUDO_MAP_FD
:
11546 case BPF_PSEUDO_MAP_IDX
:
11547 if (insn
[1].imm
== 0)
11551 verbose(env
, "unrecognized bpf_ld_imm64 insn\n");
11555 switch (insn
[0].src_reg
) {
11556 case BPF_PSEUDO_MAP_IDX_VALUE
:
11557 case BPF_PSEUDO_MAP_IDX
:
11558 if (bpfptr_is_null(env
->fd_array
)) {
11559 verbose(env
, "fd_idx without fd_array is invalid\n");
11562 if (copy_from_bpfptr_offset(&fd
, env
->fd_array
,
11563 insn
[0].imm
* sizeof(fd
),
11573 map
= __bpf_map_get(f
);
11575 verbose(env
, "fd %d is not pointing to valid bpf_map\n",
11577 return PTR_ERR(map
);
11580 err
= check_map_prog_compatibility(env
, map
, env
->prog
);
11586 aux
= &env
->insn_aux_data
[i
];
11587 if (insn
[0].src_reg
== BPF_PSEUDO_MAP_FD
||
11588 insn
[0].src_reg
== BPF_PSEUDO_MAP_IDX
) {
11589 addr
= (unsigned long)map
;
11591 u32 off
= insn
[1].imm
;
11593 if (off
>= BPF_MAX_VAR_OFF
) {
11594 verbose(env
, "direct value offset of %u is not allowed\n", off
);
11599 if (!map
->ops
->map_direct_value_addr
) {
11600 verbose(env
, "no direct value access support for this map type\n");
11605 err
= map
->ops
->map_direct_value_addr(map
, &addr
, off
);
11607 verbose(env
, "invalid access to map value pointer, value_size=%u off=%u\n",
11608 map
->value_size
, off
);
11613 aux
->map_off
= off
;
11617 insn
[0].imm
= (u32
)addr
;
11618 insn
[1].imm
= addr
>> 32;
11620 /* check whether we recorded this map already */
11621 for (j
= 0; j
< env
->used_map_cnt
; j
++) {
11622 if (env
->used_maps
[j
] == map
) {
11623 aux
->map_index
= j
;
11629 if (env
->used_map_cnt
>= MAX_USED_MAPS
) {
11634 /* hold the map. If the program is rejected by verifier,
11635 * the map will be released by release_maps() or it
11636 * will be used by the valid program until it's unloaded
11637 * and all maps are released in free_used_maps()
11641 aux
->map_index
= env
->used_map_cnt
;
11642 env
->used_maps
[env
->used_map_cnt
++] = map
;
11644 if (bpf_map_is_cgroup_storage(map
) &&
11645 bpf_cgroup_storage_assign(env
->prog
->aux
, map
)) {
11646 verbose(env
, "only one cgroup storage of each type is allowed\n");
11658 /* Basic sanity check before we invest more work here. */
11659 if (!bpf_opcode_in_insntable(insn
->code
)) {
11660 verbose(env
, "unknown opcode %02x\n", insn
->code
);
11665 /* now all pseudo BPF_LD_IMM64 instructions load valid
11666 * 'struct bpf_map *' into a register instead of user map_fd.
11667 * These pointers will be used later by verifier to validate map access.
11672 /* drop refcnt of maps used by the rejected program */
11673 static void release_maps(struct bpf_verifier_env
*env
)
11675 __bpf_free_used_maps(env
->prog
->aux
, env
->used_maps
,
11676 env
->used_map_cnt
);
11679 /* drop refcnt of maps used by the rejected program */
11680 static void release_btfs(struct bpf_verifier_env
*env
)
11682 __bpf_free_used_btfs(env
->prog
->aux
, env
->used_btfs
,
11683 env
->used_btf_cnt
);
11686 /* convert pseudo BPF_LD_IMM64 into generic BPF_LD_IMM64 */
11687 static void convert_pseudo_ld_imm64(struct bpf_verifier_env
*env
)
11689 struct bpf_insn
*insn
= env
->prog
->insnsi
;
11690 int insn_cnt
= env
->prog
->len
;
11693 for (i
= 0; i
< insn_cnt
; i
++, insn
++) {
11694 if (insn
->code
!= (BPF_LD
| BPF_IMM
| BPF_DW
))
11696 if (insn
->src_reg
== BPF_PSEUDO_FUNC
)
11702 /* single env->prog->insni[off] instruction was replaced with the range
11703 * insni[off, off + cnt). Adjust corresponding insn_aux_data by copying
11704 * [0, off) and [off, end) to new locations, so the patched range stays zero
11706 static void adjust_insn_aux_data(struct bpf_verifier_env
*env
,
11707 struct bpf_insn_aux_data
*new_data
,
11708 struct bpf_prog
*new_prog
, u32 off
, u32 cnt
)
11710 struct bpf_insn_aux_data
*old_data
= env
->insn_aux_data
;
11711 struct bpf_insn
*insn
= new_prog
->insnsi
;
11712 u32 old_seen
= old_data
[off
].seen
;
11716 /* aux info at OFF always needs adjustment, no matter fast path
11717 * (cnt == 1) is taken or not. There is no guarantee INSN at OFF is the
11718 * original insn at old prog.
11720 old_data
[off
].zext_dst
= insn_has_def32(env
, insn
+ off
+ cnt
- 1);
11724 prog_len
= new_prog
->len
;
11726 memcpy(new_data
, old_data
, sizeof(struct bpf_insn_aux_data
) * off
);
11727 memcpy(new_data
+ off
+ cnt
- 1, old_data
+ off
,
11728 sizeof(struct bpf_insn_aux_data
) * (prog_len
- off
- cnt
+ 1));
11729 for (i
= off
; i
< off
+ cnt
- 1; i
++) {
11730 /* Expand insni[off]'s seen count to the patched range. */
11731 new_data
[i
].seen
= old_seen
;
11732 new_data
[i
].zext_dst
= insn_has_def32(env
, insn
+ i
);
11734 env
->insn_aux_data
= new_data
;
11738 static void adjust_subprog_starts(struct bpf_verifier_env
*env
, u32 off
, u32 len
)
11744 /* NOTE: fake 'exit' subprog should be updated as well. */
11745 for (i
= 0; i
<= env
->subprog_cnt
; i
++) {
11746 if (env
->subprog_info
[i
].start
<= off
)
11748 env
->subprog_info
[i
].start
+= len
- 1;
11752 static void adjust_poke_descs(struct bpf_prog
*prog
, u32 off
, u32 len
)
11754 struct bpf_jit_poke_descriptor
*tab
= prog
->aux
->poke_tab
;
11755 int i
, sz
= prog
->aux
->size_poke_tab
;
11756 struct bpf_jit_poke_descriptor
*desc
;
11758 for (i
= 0; i
< sz
; i
++) {
11760 if (desc
->insn_idx
<= off
)
11762 desc
->insn_idx
+= len
- 1;
11766 static struct bpf_prog
*bpf_patch_insn_data(struct bpf_verifier_env
*env
, u32 off
,
11767 const struct bpf_insn
*patch
, u32 len
)
11769 struct bpf_prog
*new_prog
;
11770 struct bpf_insn_aux_data
*new_data
= NULL
;
11773 new_data
= vzalloc(array_size(env
->prog
->len
+ len
- 1,
11774 sizeof(struct bpf_insn_aux_data
)));
11779 new_prog
= bpf_patch_insn_single(env
->prog
, off
, patch
, len
);
11780 if (IS_ERR(new_prog
)) {
11781 if (PTR_ERR(new_prog
) == -ERANGE
)
11783 "insn %d cannot be patched due to 16-bit range\n",
11784 env
->insn_aux_data
[off
].orig_idx
);
11788 adjust_insn_aux_data(env
, new_data
, new_prog
, off
, len
);
11789 adjust_subprog_starts(env
, off
, len
);
11790 adjust_poke_descs(new_prog
, off
, len
);
11794 static int adjust_subprog_starts_after_remove(struct bpf_verifier_env
*env
,
11799 /* find first prog starting at or after off (first to remove) */
11800 for (i
= 0; i
< env
->subprog_cnt
; i
++)
11801 if (env
->subprog_info
[i
].start
>= off
)
11803 /* find first prog starting at or after off + cnt (first to stay) */
11804 for (j
= i
; j
< env
->subprog_cnt
; j
++)
11805 if (env
->subprog_info
[j
].start
>= off
+ cnt
)
11807 /* if j doesn't start exactly at off + cnt, we are just removing
11808 * the front of previous prog
11810 if (env
->subprog_info
[j
].start
!= off
+ cnt
)
11814 struct bpf_prog_aux
*aux
= env
->prog
->aux
;
11817 /* move fake 'exit' subprog as well */
11818 move
= env
->subprog_cnt
+ 1 - j
;
11820 memmove(env
->subprog_info
+ i
,
11821 env
->subprog_info
+ j
,
11822 sizeof(*env
->subprog_info
) * move
);
11823 env
->subprog_cnt
-= j
- i
;
11825 /* remove func_info */
11826 if (aux
->func_info
) {
11827 move
= aux
->func_info_cnt
- j
;
11829 memmove(aux
->func_info
+ i
,
11830 aux
->func_info
+ j
,
11831 sizeof(*aux
->func_info
) * move
);
11832 aux
->func_info_cnt
-= j
- i
;
11833 /* func_info->insn_off is set after all code rewrites,
11834 * in adjust_btf_func() - no need to adjust
11838 /* convert i from "first prog to remove" to "first to adjust" */
11839 if (env
->subprog_info
[i
].start
== off
)
11843 /* update fake 'exit' subprog as well */
11844 for (; i
<= env
->subprog_cnt
; i
++)
11845 env
->subprog_info
[i
].start
-= cnt
;
11850 static int bpf_adj_linfo_after_remove(struct bpf_verifier_env
*env
, u32 off
,
11853 struct bpf_prog
*prog
= env
->prog
;
11854 u32 i
, l_off
, l_cnt
, nr_linfo
;
11855 struct bpf_line_info
*linfo
;
11857 nr_linfo
= prog
->aux
->nr_linfo
;
11861 linfo
= prog
->aux
->linfo
;
11863 /* find first line info to remove, count lines to be removed */
11864 for (i
= 0; i
< nr_linfo
; i
++)
11865 if (linfo
[i
].insn_off
>= off
)
11870 for (; i
< nr_linfo
; i
++)
11871 if (linfo
[i
].insn_off
< off
+ cnt
)
11876 /* First live insn doesn't match first live linfo, it needs to "inherit"
11877 * last removed linfo. prog is already modified, so prog->len == off
11878 * means no live instructions after (tail of the program was removed).
11880 if (prog
->len
!= off
&& l_cnt
&&
11881 (i
== nr_linfo
|| linfo
[i
].insn_off
!= off
+ cnt
)) {
11883 linfo
[--i
].insn_off
= off
+ cnt
;
11886 /* remove the line info which refer to the removed instructions */
11888 memmove(linfo
+ l_off
, linfo
+ i
,
11889 sizeof(*linfo
) * (nr_linfo
- i
));
11891 prog
->aux
->nr_linfo
-= l_cnt
;
11892 nr_linfo
= prog
->aux
->nr_linfo
;
11895 /* pull all linfo[i].insn_off >= off + cnt in by cnt */
11896 for (i
= l_off
; i
< nr_linfo
; i
++)
11897 linfo
[i
].insn_off
-= cnt
;
11899 /* fix up all subprogs (incl. 'exit') which start >= off */
11900 for (i
= 0; i
<= env
->subprog_cnt
; i
++)
11901 if (env
->subprog_info
[i
].linfo_idx
> l_off
) {
11902 /* program may have started in the removed region but
11903 * may not be fully removed
11905 if (env
->subprog_info
[i
].linfo_idx
>= l_off
+ l_cnt
)
11906 env
->subprog_info
[i
].linfo_idx
-= l_cnt
;
11908 env
->subprog_info
[i
].linfo_idx
= l_off
;
11914 static int verifier_remove_insns(struct bpf_verifier_env
*env
, u32 off
, u32 cnt
)
11916 struct bpf_insn_aux_data
*aux_data
= env
->insn_aux_data
;
11917 unsigned int orig_prog_len
= env
->prog
->len
;
11920 if (bpf_prog_is_dev_bound(env
->prog
->aux
))
11921 bpf_prog_offload_remove_insns(env
, off
, cnt
);
11923 err
= bpf_remove_insns(env
->prog
, off
, cnt
);
11927 err
= adjust_subprog_starts_after_remove(env
, off
, cnt
);
11931 err
= bpf_adj_linfo_after_remove(env
, off
, cnt
);
11935 memmove(aux_data
+ off
, aux_data
+ off
+ cnt
,
11936 sizeof(*aux_data
) * (orig_prog_len
- off
- cnt
));
11941 /* The verifier does more data flow analysis than llvm and will not
11942 * explore branches that are dead at run time. Malicious programs can
11943 * have dead code too. Therefore replace all dead at-run-time code
11946 * Just nops are not optimal, e.g. if they would sit at the end of the
11947 * program and through another bug we would manage to jump there, then
11948 * we'd execute beyond program memory otherwise. Returning exception
11949 * code also wouldn't work since we can have subprogs where the dead
11950 * code could be located.
11952 static void sanitize_dead_code(struct bpf_verifier_env
*env
)
11954 struct bpf_insn_aux_data
*aux_data
= env
->insn_aux_data
;
11955 struct bpf_insn trap
= BPF_JMP_IMM(BPF_JA
, 0, 0, -1);
11956 struct bpf_insn
*insn
= env
->prog
->insnsi
;
11957 const int insn_cnt
= env
->prog
->len
;
11960 for (i
= 0; i
< insn_cnt
; i
++) {
11961 if (aux_data
[i
].seen
)
11963 memcpy(insn
+ i
, &trap
, sizeof(trap
));
11964 aux_data
[i
].zext_dst
= false;
11968 static bool insn_is_cond_jump(u8 code
)
11972 if (BPF_CLASS(code
) == BPF_JMP32
)
11975 if (BPF_CLASS(code
) != BPF_JMP
)
11979 return op
!= BPF_JA
&& op
!= BPF_EXIT
&& op
!= BPF_CALL
;
11982 static void opt_hard_wire_dead_code_branches(struct bpf_verifier_env
*env
)
11984 struct bpf_insn_aux_data
*aux_data
= env
->insn_aux_data
;
11985 struct bpf_insn ja
= BPF_JMP_IMM(BPF_JA
, 0, 0, 0);
11986 struct bpf_insn
*insn
= env
->prog
->insnsi
;
11987 const int insn_cnt
= env
->prog
->len
;
11990 for (i
= 0; i
< insn_cnt
; i
++, insn
++) {
11991 if (!insn_is_cond_jump(insn
->code
))
11994 if (!aux_data
[i
+ 1].seen
)
11995 ja
.off
= insn
->off
;
11996 else if (!aux_data
[i
+ 1 + insn
->off
].seen
)
12001 if (bpf_prog_is_dev_bound(env
->prog
->aux
))
12002 bpf_prog_offload_replace_insn(env
, i
, &ja
);
12004 memcpy(insn
, &ja
, sizeof(ja
));
12008 static int opt_remove_dead_code(struct bpf_verifier_env
*env
)
12010 struct bpf_insn_aux_data
*aux_data
= env
->insn_aux_data
;
12011 int insn_cnt
= env
->prog
->len
;
12014 for (i
= 0; i
< insn_cnt
; i
++) {
12018 while (i
+ j
< insn_cnt
&& !aux_data
[i
+ j
].seen
)
12023 err
= verifier_remove_insns(env
, i
, j
);
12026 insn_cnt
= env
->prog
->len
;
12032 static int opt_remove_nops(struct bpf_verifier_env
*env
)
12034 const struct bpf_insn ja
= BPF_JMP_IMM(BPF_JA
, 0, 0, 0);
12035 struct bpf_insn
*insn
= env
->prog
->insnsi
;
12036 int insn_cnt
= env
->prog
->len
;
12039 for (i
= 0; i
< insn_cnt
; i
++) {
12040 if (memcmp(&insn
[i
], &ja
, sizeof(ja
)))
12043 err
= verifier_remove_insns(env
, i
, 1);
12053 static int opt_subreg_zext_lo32_rnd_hi32(struct bpf_verifier_env
*env
,
12054 const union bpf_attr
*attr
)
12056 struct bpf_insn
*patch
, zext_patch
[2], rnd_hi32_patch
[4];
12057 struct bpf_insn_aux_data
*aux
= env
->insn_aux_data
;
12058 int i
, patch_len
, delta
= 0, len
= env
->prog
->len
;
12059 struct bpf_insn
*insns
= env
->prog
->insnsi
;
12060 struct bpf_prog
*new_prog
;
12063 rnd_hi32
= attr
->prog_flags
& BPF_F_TEST_RND_HI32
;
12064 zext_patch
[1] = BPF_ZEXT_REG(0);
12065 rnd_hi32_patch
[1] = BPF_ALU64_IMM(BPF_MOV
, BPF_REG_AX
, 0);
12066 rnd_hi32_patch
[2] = BPF_ALU64_IMM(BPF_LSH
, BPF_REG_AX
, 32);
12067 rnd_hi32_patch
[3] = BPF_ALU64_REG(BPF_OR
, 0, BPF_REG_AX
);
12068 for (i
= 0; i
< len
; i
++) {
12069 int adj_idx
= i
+ delta
;
12070 struct bpf_insn insn
;
12073 insn
= insns
[adj_idx
];
12074 load_reg
= insn_def_regno(&insn
);
12075 if (!aux
[adj_idx
].zext_dst
) {
12083 class = BPF_CLASS(code
);
12084 if (load_reg
== -1)
12087 /* NOTE: arg "reg" (the fourth one) is only used for
12088 * BPF_STX + SRC_OP, so it is safe to pass NULL
12091 if (is_reg64(env
, &insn
, load_reg
, NULL
, DST_OP
)) {
12092 if (class == BPF_LD
&&
12093 BPF_MODE(code
) == BPF_IMM
)
12098 /* ctx load could be transformed into wider load. */
12099 if (class == BPF_LDX
&&
12100 aux
[adj_idx
].ptr_type
== PTR_TO_CTX
)
12103 imm_rnd
= get_random_int();
12104 rnd_hi32_patch
[0] = insn
;
12105 rnd_hi32_patch
[1].imm
= imm_rnd
;
12106 rnd_hi32_patch
[3].dst_reg
= load_reg
;
12107 patch
= rnd_hi32_patch
;
12109 goto apply_patch_buffer
;
12112 /* Add in an zero-extend instruction if a) the JIT has requested
12113 * it or b) it's a CMPXCHG.
12115 * The latter is because: BPF_CMPXCHG always loads a value into
12116 * R0, therefore always zero-extends. However some archs'
12117 * equivalent instruction only does this load when the
12118 * comparison is successful. This detail of CMPXCHG is
12119 * orthogonal to the general zero-extension behaviour of the
12120 * CPU, so it's treated independently of bpf_jit_needs_zext.
12122 if (!bpf_jit_needs_zext() && !is_cmpxchg_insn(&insn
))
12125 if (WARN_ON(load_reg
== -1)) {
12126 verbose(env
, "verifier bug. zext_dst is set, but no reg is defined\n");
12130 zext_patch
[0] = insn
;
12131 zext_patch
[1].dst_reg
= load_reg
;
12132 zext_patch
[1].src_reg
= load_reg
;
12133 patch
= zext_patch
;
12135 apply_patch_buffer
:
12136 new_prog
= bpf_patch_insn_data(env
, adj_idx
, patch
, patch_len
);
12139 env
->prog
= new_prog
;
12140 insns
= new_prog
->insnsi
;
12141 aux
= env
->insn_aux_data
;
12142 delta
+= patch_len
- 1;
12148 /* convert load instructions that access fields of a context type into a
12149 * sequence of instructions that access fields of the underlying structure:
12150 * struct __sk_buff -> struct sk_buff
12151 * struct bpf_sock_ops -> struct sock
12153 static int convert_ctx_accesses(struct bpf_verifier_env
*env
)
12155 const struct bpf_verifier_ops
*ops
= env
->ops
;
12156 int i
, cnt
, size
, ctx_field_size
, delta
= 0;
12157 const int insn_cnt
= env
->prog
->len
;
12158 struct bpf_insn insn_buf
[16], *insn
;
12159 u32 target_size
, size_default
, off
;
12160 struct bpf_prog
*new_prog
;
12161 enum bpf_access_type type
;
12162 bool is_narrower_load
;
12164 if (ops
->gen_prologue
|| env
->seen_direct_write
) {
12165 if (!ops
->gen_prologue
) {
12166 verbose(env
, "bpf verifier is misconfigured\n");
12169 cnt
= ops
->gen_prologue(insn_buf
, env
->seen_direct_write
,
12171 if (cnt
>= ARRAY_SIZE(insn_buf
)) {
12172 verbose(env
, "bpf verifier is misconfigured\n");
12175 new_prog
= bpf_patch_insn_data(env
, 0, insn_buf
, cnt
);
12179 env
->prog
= new_prog
;
12184 if (bpf_prog_is_dev_bound(env
->prog
->aux
))
12187 insn
= env
->prog
->insnsi
+ delta
;
12189 for (i
= 0; i
< insn_cnt
; i
++, insn
++) {
12190 bpf_convert_ctx_access_t convert_ctx_access
;
12193 if (insn
->code
== (BPF_LDX
| BPF_MEM
| BPF_B
) ||
12194 insn
->code
== (BPF_LDX
| BPF_MEM
| BPF_H
) ||
12195 insn
->code
== (BPF_LDX
| BPF_MEM
| BPF_W
) ||
12196 insn
->code
== (BPF_LDX
| BPF_MEM
| BPF_DW
)) {
12199 } else if (insn
->code
== (BPF_STX
| BPF_MEM
| BPF_B
) ||
12200 insn
->code
== (BPF_STX
| BPF_MEM
| BPF_H
) ||
12201 insn
->code
== (BPF_STX
| BPF_MEM
| BPF_W
) ||
12202 insn
->code
== (BPF_STX
| BPF_MEM
| BPF_DW
) ||
12203 insn
->code
== (BPF_ST
| BPF_MEM
| BPF_B
) ||
12204 insn
->code
== (BPF_ST
| BPF_MEM
| BPF_H
) ||
12205 insn
->code
== (BPF_ST
| BPF_MEM
| BPF_W
) ||
12206 insn
->code
== (BPF_ST
| BPF_MEM
| BPF_DW
)) {
12208 ctx_access
= BPF_CLASS(insn
->code
) == BPF_STX
;
12213 if (type
== BPF_WRITE
&&
12214 env
->insn_aux_data
[i
+ delta
].sanitize_stack_spill
) {
12215 struct bpf_insn patch
[] = {
12220 cnt
= ARRAY_SIZE(patch
);
12221 new_prog
= bpf_patch_insn_data(env
, i
+ delta
, patch
, cnt
);
12226 env
->prog
= new_prog
;
12227 insn
= new_prog
->insnsi
+ i
+ delta
;
12234 switch (env
->insn_aux_data
[i
+ delta
].ptr_type
) {
12236 if (!ops
->convert_ctx_access
)
12238 convert_ctx_access
= ops
->convert_ctx_access
;
12240 case PTR_TO_SOCKET
:
12241 case PTR_TO_SOCK_COMMON
:
12242 convert_ctx_access
= bpf_sock_convert_ctx_access
;
12244 case PTR_TO_TCP_SOCK
:
12245 convert_ctx_access
= bpf_tcp_sock_convert_ctx_access
;
12247 case PTR_TO_XDP_SOCK
:
12248 convert_ctx_access
= bpf_xdp_sock_convert_ctx_access
;
12250 case PTR_TO_BTF_ID
:
12251 if (type
== BPF_READ
) {
12252 insn
->code
= BPF_LDX
| BPF_PROBE_MEM
|
12253 BPF_SIZE((insn
)->code
);
12254 env
->prog
->aux
->num_exentries
++;
12255 } else if (resolve_prog_type(env
->prog
) != BPF_PROG_TYPE_STRUCT_OPS
) {
12256 verbose(env
, "Writes through BTF pointers are not allowed\n");
12264 ctx_field_size
= env
->insn_aux_data
[i
+ delta
].ctx_field_size
;
12265 size
= BPF_LDST_BYTES(insn
);
12267 /* If the read access is a narrower load of the field,
12268 * convert to a 4/8-byte load, to minimum program type specific
12269 * convert_ctx_access changes. If conversion is successful,
12270 * we will apply proper mask to the result.
12272 is_narrower_load
= size
< ctx_field_size
;
12273 size_default
= bpf_ctx_off_adjust_machine(ctx_field_size
);
12275 if (is_narrower_load
) {
12278 if (type
== BPF_WRITE
) {
12279 verbose(env
, "bpf verifier narrow ctx access misconfigured\n");
12284 if (ctx_field_size
== 4)
12286 else if (ctx_field_size
== 8)
12287 size_code
= BPF_DW
;
12289 insn
->off
= off
& ~(size_default
- 1);
12290 insn
->code
= BPF_LDX
| BPF_MEM
| size_code
;
12294 cnt
= convert_ctx_access(type
, insn
, insn_buf
, env
->prog
,
12296 if (cnt
== 0 || cnt
>= ARRAY_SIZE(insn_buf
) ||
12297 (ctx_field_size
&& !target_size
)) {
12298 verbose(env
, "bpf verifier is misconfigured\n");
12302 if (is_narrower_load
&& size
< target_size
) {
12303 u8 shift
= bpf_ctx_narrow_access_offset(
12304 off
, size
, size_default
) * 8;
12305 if (shift
&& cnt
+ 1 >= ARRAY_SIZE(insn_buf
)) {
12306 verbose(env
, "bpf verifier narrow ctx load misconfigured\n");
12309 if (ctx_field_size
<= 4) {
12311 insn_buf
[cnt
++] = BPF_ALU32_IMM(BPF_RSH
,
12314 insn_buf
[cnt
++] = BPF_ALU32_IMM(BPF_AND
, insn
->dst_reg
,
12315 (1 << size
* 8) - 1);
12318 insn_buf
[cnt
++] = BPF_ALU64_IMM(BPF_RSH
,
12321 insn_buf
[cnt
++] = BPF_ALU64_IMM(BPF_AND
, insn
->dst_reg
,
12322 (1ULL << size
* 8) - 1);
12326 new_prog
= bpf_patch_insn_data(env
, i
+ delta
, insn_buf
, cnt
);
12332 /* keep walking new program and skip insns we just inserted */
12333 env
->prog
= new_prog
;
12334 insn
= new_prog
->insnsi
+ i
+ delta
;
12340 static int jit_subprogs(struct bpf_verifier_env
*env
)
12342 struct bpf_prog
*prog
= env
->prog
, **func
, *tmp
;
12343 int i
, j
, subprog_start
, subprog_end
= 0, len
, subprog
;
12344 struct bpf_map
*map_ptr
;
12345 struct bpf_insn
*insn
;
12346 void *old_bpf_func
;
12347 int err
, num_exentries
;
12349 if (env
->subprog_cnt
<= 1)
12352 for (i
= 0, insn
= prog
->insnsi
; i
< prog
->len
; i
++, insn
++) {
12353 if (bpf_pseudo_func(insn
)) {
12354 env
->insn_aux_data
[i
].call_imm
= insn
->imm
;
12355 /* subprog is encoded in insn[1].imm */
12359 if (!bpf_pseudo_call(insn
))
12361 /* Upon error here we cannot fall back to interpreter but
12362 * need a hard reject of the program. Thus -EFAULT is
12363 * propagated in any case.
12365 subprog
= find_subprog(env
, i
+ insn
->imm
+ 1);
12367 WARN_ONCE(1, "verifier bug. No program starts at insn %d\n",
12368 i
+ insn
->imm
+ 1);
12371 /* temporarily remember subprog id inside insn instead of
12372 * aux_data, since next loop will split up all insns into funcs
12374 insn
->off
= subprog
;
12375 /* remember original imm in case JIT fails and fallback
12376 * to interpreter will be needed
12378 env
->insn_aux_data
[i
].call_imm
= insn
->imm
;
12379 /* point imm to __bpf_call_base+1 from JITs point of view */
12383 err
= bpf_prog_alloc_jited_linfo(prog
);
12385 goto out_undo_insn
;
12388 func
= kcalloc(env
->subprog_cnt
, sizeof(prog
), GFP_KERNEL
);
12390 goto out_undo_insn
;
12392 for (i
= 0; i
< env
->subprog_cnt
; i
++) {
12393 subprog_start
= subprog_end
;
12394 subprog_end
= env
->subprog_info
[i
+ 1].start
;
12396 len
= subprog_end
- subprog_start
;
12397 /* bpf_prog_run() doesn't call subprogs directly,
12398 * hence main prog stats include the runtime of subprogs.
12399 * subprogs don't have IDs and not reachable via prog_get_next_id
12400 * func[i]->stats will never be accessed and stays NULL
12402 func
[i
] = bpf_prog_alloc_no_stats(bpf_prog_size(len
), GFP_USER
);
12405 memcpy(func
[i
]->insnsi
, &prog
->insnsi
[subprog_start
],
12406 len
* sizeof(struct bpf_insn
));
12407 func
[i
]->type
= prog
->type
;
12408 func
[i
]->len
= len
;
12409 if (bpf_prog_calc_tag(func
[i
]))
12411 func
[i
]->is_func
= 1;
12412 func
[i
]->aux
->func_idx
= i
;
12413 /* Below members will be freed only at prog->aux */
12414 func
[i
]->aux
->btf
= prog
->aux
->btf
;
12415 func
[i
]->aux
->func_info
= prog
->aux
->func_info
;
12416 func
[i
]->aux
->poke_tab
= prog
->aux
->poke_tab
;
12417 func
[i
]->aux
->size_poke_tab
= prog
->aux
->size_poke_tab
;
12419 for (j
= 0; j
< prog
->aux
->size_poke_tab
; j
++) {
12420 struct bpf_jit_poke_descriptor
*poke
;
12422 poke
= &prog
->aux
->poke_tab
[j
];
12423 if (poke
->insn_idx
< subprog_end
&&
12424 poke
->insn_idx
>= subprog_start
)
12425 poke
->aux
= func
[i
]->aux
;
12428 /* Use bpf_prog_F_tag to indicate functions in stack traces.
12429 * Long term would need debug info to populate names
12431 func
[i
]->aux
->name
[0] = 'F';
12432 func
[i
]->aux
->stack_depth
= env
->subprog_info
[i
].stack_depth
;
12433 func
[i
]->jit_requested
= 1;
12434 func
[i
]->aux
->kfunc_tab
= prog
->aux
->kfunc_tab
;
12435 func
[i
]->aux
->linfo
= prog
->aux
->linfo
;
12436 func
[i
]->aux
->nr_linfo
= prog
->aux
->nr_linfo
;
12437 func
[i
]->aux
->jited_linfo
= prog
->aux
->jited_linfo
;
12438 func
[i
]->aux
->linfo_idx
= env
->subprog_info
[i
].linfo_idx
;
12440 insn
= func
[i
]->insnsi
;
12441 for (j
= 0; j
< func
[i
]->len
; j
++, insn
++) {
12442 if (BPF_CLASS(insn
->code
) == BPF_LDX
&&
12443 BPF_MODE(insn
->code
) == BPF_PROBE_MEM
)
12446 func
[i
]->aux
->num_exentries
= num_exentries
;
12447 func
[i
]->aux
->tail_call_reachable
= env
->subprog_info
[i
].tail_call_reachable
;
12448 func
[i
] = bpf_int_jit_compile(func
[i
]);
12449 if (!func
[i
]->jited
) {
12456 /* at this point all bpf functions were successfully JITed
12457 * now populate all bpf_calls with correct addresses and
12458 * run last pass of JIT
12460 for (i
= 0; i
< env
->subprog_cnt
; i
++) {
12461 insn
= func
[i
]->insnsi
;
12462 for (j
= 0; j
< func
[i
]->len
; j
++, insn
++) {
12463 if (bpf_pseudo_func(insn
)) {
12464 subprog
= insn
[1].imm
;
12465 insn
[0].imm
= (u32
)(long)func
[subprog
]->bpf_func
;
12466 insn
[1].imm
= ((u64
)(long)func
[subprog
]->bpf_func
) >> 32;
12469 if (!bpf_pseudo_call(insn
))
12471 subprog
= insn
->off
;
12472 insn
->imm
= BPF_CAST_CALL(func
[subprog
]->bpf_func
) -
12476 /* we use the aux data to keep a list of the start addresses
12477 * of the JITed images for each function in the program
12479 * for some architectures, such as powerpc64, the imm field
12480 * might not be large enough to hold the offset of the start
12481 * address of the callee's JITed image from __bpf_call_base
12483 * in such cases, we can lookup the start address of a callee
12484 * by using its subprog id, available from the off field of
12485 * the call instruction, as an index for this list
12487 func
[i
]->aux
->func
= func
;
12488 func
[i
]->aux
->func_cnt
= env
->subprog_cnt
;
12490 for (i
= 0; i
< env
->subprog_cnt
; i
++) {
12491 old_bpf_func
= func
[i
]->bpf_func
;
12492 tmp
= bpf_int_jit_compile(func
[i
]);
12493 if (tmp
!= func
[i
] || func
[i
]->bpf_func
!= old_bpf_func
) {
12494 verbose(env
, "JIT doesn't support bpf-to-bpf calls\n");
12501 /* finally lock prog and jit images for all functions and
12502 * populate kallsysm
12504 for (i
= 0; i
< env
->subprog_cnt
; i
++) {
12505 bpf_prog_lock_ro(func
[i
]);
12506 bpf_prog_kallsyms_add(func
[i
]);
12509 /* Last step: make now unused interpreter insns from main
12510 * prog consistent for later dump requests, so they can
12511 * later look the same as if they were interpreted only.
12513 for (i
= 0, insn
= prog
->insnsi
; i
< prog
->len
; i
++, insn
++) {
12514 if (bpf_pseudo_func(insn
)) {
12515 insn
[0].imm
= env
->insn_aux_data
[i
].call_imm
;
12516 insn
[1].imm
= find_subprog(env
, i
+ insn
[0].imm
+ 1);
12519 if (!bpf_pseudo_call(insn
))
12521 insn
->off
= env
->insn_aux_data
[i
].call_imm
;
12522 subprog
= find_subprog(env
, i
+ insn
->off
+ 1);
12523 insn
->imm
= subprog
;
12527 prog
->bpf_func
= func
[0]->bpf_func
;
12528 prog
->aux
->func
= func
;
12529 prog
->aux
->func_cnt
= env
->subprog_cnt
;
12530 bpf_prog_jit_attempt_done(prog
);
12533 /* We failed JIT'ing, so at this point we need to unregister poke
12534 * descriptors from subprogs, so that kernel is not attempting to
12535 * patch it anymore as we're freeing the subprog JIT memory.
12537 for (i
= 0; i
< prog
->aux
->size_poke_tab
; i
++) {
12538 map_ptr
= prog
->aux
->poke_tab
[i
].tail_call
.map
;
12539 map_ptr
->ops
->map_poke_untrack(map_ptr
, prog
->aux
);
12541 /* At this point we're guaranteed that poke descriptors are not
12542 * live anymore. We can just unlink its descriptor table as it's
12543 * released with the main prog.
12545 for (i
= 0; i
< env
->subprog_cnt
; i
++) {
12548 func
[i
]->aux
->poke_tab
= NULL
;
12549 bpf_jit_free(func
[i
]);
12553 /* cleanup main prog to be interpreted */
12554 prog
->jit_requested
= 0;
12555 for (i
= 0, insn
= prog
->insnsi
; i
< prog
->len
; i
++, insn
++) {
12556 if (!bpf_pseudo_call(insn
))
12559 insn
->imm
= env
->insn_aux_data
[i
].call_imm
;
12561 bpf_prog_jit_attempt_done(prog
);
12565 static int fixup_call_args(struct bpf_verifier_env
*env
)
12567 #ifndef CONFIG_BPF_JIT_ALWAYS_ON
12568 struct bpf_prog
*prog
= env
->prog
;
12569 struct bpf_insn
*insn
= prog
->insnsi
;
12570 bool has_kfunc_call
= bpf_prog_has_kfunc_call(prog
);
12575 if (env
->prog
->jit_requested
&&
12576 !bpf_prog_is_dev_bound(env
->prog
->aux
)) {
12577 err
= jit_subprogs(env
);
12580 if (err
== -EFAULT
)
12583 #ifndef CONFIG_BPF_JIT_ALWAYS_ON
12584 if (has_kfunc_call
) {
12585 verbose(env
, "calling kernel functions are not allowed in non-JITed programs\n");
12588 if (env
->subprog_cnt
> 1 && env
->prog
->aux
->tail_call_reachable
) {
12589 /* When JIT fails the progs with bpf2bpf calls and tail_calls
12590 * have to be rejected, since interpreter doesn't support them yet.
12592 verbose(env
, "tail_calls are not allowed in non-JITed programs with bpf-to-bpf calls\n");
12595 for (i
= 0; i
< prog
->len
; i
++, insn
++) {
12596 if (bpf_pseudo_func(insn
)) {
12597 /* When JIT fails the progs with callback calls
12598 * have to be rejected, since interpreter doesn't support them yet.
12600 verbose(env
, "callbacks are not allowed in non-JITed programs\n");
12604 if (!bpf_pseudo_call(insn
))
12606 depth
= get_callee_stack_depth(env
, insn
, i
);
12609 bpf_patch_call_args(insn
, depth
);
12616 static int fixup_kfunc_call(struct bpf_verifier_env
*env
,
12617 struct bpf_insn
*insn
)
12619 const struct bpf_kfunc_desc
*desc
;
12621 /* insn->imm has the btf func_id. Replace it with
12622 * an address (relative to __bpf_base_call).
12624 desc
= find_kfunc_desc(env
->prog
, insn
->imm
);
12626 verbose(env
, "verifier internal error: kernel function descriptor not found for func_id %u\n",
12631 insn
->imm
= desc
->imm
;
12636 /* Do various post-verification rewrites in a single program pass.
12637 * These rewrites simplify JIT and interpreter implementations.
12639 static int do_misc_fixups(struct bpf_verifier_env
*env
)
12641 struct bpf_prog
*prog
= env
->prog
;
12642 bool expect_blinding
= bpf_jit_blinding_enabled(prog
);
12643 enum bpf_prog_type prog_type
= resolve_prog_type(prog
);
12644 struct bpf_insn
*insn
= prog
->insnsi
;
12645 const struct bpf_func_proto
*fn
;
12646 const int insn_cnt
= prog
->len
;
12647 const struct bpf_map_ops
*ops
;
12648 struct bpf_insn_aux_data
*aux
;
12649 struct bpf_insn insn_buf
[16];
12650 struct bpf_prog
*new_prog
;
12651 struct bpf_map
*map_ptr
;
12652 int i
, ret
, cnt
, delta
= 0;
12654 for (i
= 0; i
< insn_cnt
; i
++, insn
++) {
12655 /* Make divide-by-zero exceptions impossible. */
12656 if (insn
->code
== (BPF_ALU64
| BPF_MOD
| BPF_X
) ||
12657 insn
->code
== (BPF_ALU64
| BPF_DIV
| BPF_X
) ||
12658 insn
->code
== (BPF_ALU
| BPF_MOD
| BPF_X
) ||
12659 insn
->code
== (BPF_ALU
| BPF_DIV
| BPF_X
)) {
12660 bool is64
= BPF_CLASS(insn
->code
) == BPF_ALU64
;
12661 bool isdiv
= BPF_OP(insn
->code
) == BPF_DIV
;
12662 struct bpf_insn
*patchlet
;
12663 struct bpf_insn chk_and_div
[] = {
12664 /* [R,W]x div 0 -> 0 */
12665 BPF_RAW_INSN((is64
? BPF_JMP
: BPF_JMP32
) |
12666 BPF_JNE
| BPF_K
, insn
->src_reg
,
12668 BPF_ALU32_REG(BPF_XOR
, insn
->dst_reg
, insn
->dst_reg
),
12669 BPF_JMP_IMM(BPF_JA
, 0, 0, 1),
12672 struct bpf_insn chk_and_mod
[] = {
12673 /* [R,W]x mod 0 -> [R,W]x */
12674 BPF_RAW_INSN((is64
? BPF_JMP
: BPF_JMP32
) |
12675 BPF_JEQ
| BPF_K
, insn
->src_reg
,
12676 0, 1 + (is64
? 0 : 1), 0),
12678 BPF_JMP_IMM(BPF_JA
, 0, 0, 1),
12679 BPF_MOV32_REG(insn
->dst_reg
, insn
->dst_reg
),
12682 patchlet
= isdiv
? chk_and_div
: chk_and_mod
;
12683 cnt
= isdiv
? ARRAY_SIZE(chk_and_div
) :
12684 ARRAY_SIZE(chk_and_mod
) - (is64
? 2 : 0);
12686 new_prog
= bpf_patch_insn_data(env
, i
+ delta
, patchlet
, cnt
);
12691 env
->prog
= prog
= new_prog
;
12692 insn
= new_prog
->insnsi
+ i
+ delta
;
12696 /* Implement LD_ABS and LD_IND with a rewrite, if supported by the program type. */
12697 if (BPF_CLASS(insn
->code
) == BPF_LD
&&
12698 (BPF_MODE(insn
->code
) == BPF_ABS
||
12699 BPF_MODE(insn
->code
) == BPF_IND
)) {
12700 cnt
= env
->ops
->gen_ld_abs(insn
, insn_buf
);
12701 if (cnt
== 0 || cnt
>= ARRAY_SIZE(insn_buf
)) {
12702 verbose(env
, "bpf verifier is misconfigured\n");
12706 new_prog
= bpf_patch_insn_data(env
, i
+ delta
, insn_buf
, cnt
);
12711 env
->prog
= prog
= new_prog
;
12712 insn
= new_prog
->insnsi
+ i
+ delta
;
12716 /* Rewrite pointer arithmetic to mitigate speculation attacks. */
12717 if (insn
->code
== (BPF_ALU64
| BPF_ADD
| BPF_X
) ||
12718 insn
->code
== (BPF_ALU64
| BPF_SUB
| BPF_X
)) {
12719 const u8 code_add
= BPF_ALU64
| BPF_ADD
| BPF_X
;
12720 const u8 code_sub
= BPF_ALU64
| BPF_SUB
| BPF_X
;
12721 struct bpf_insn
*patch
= &insn_buf
[0];
12722 bool issrc
, isneg
, isimm
;
12725 aux
= &env
->insn_aux_data
[i
+ delta
];
12726 if (!aux
->alu_state
||
12727 aux
->alu_state
== BPF_ALU_NON_POINTER
)
12730 isneg
= aux
->alu_state
& BPF_ALU_NEG_VALUE
;
12731 issrc
= (aux
->alu_state
& BPF_ALU_SANITIZE
) ==
12732 BPF_ALU_SANITIZE_SRC
;
12733 isimm
= aux
->alu_state
& BPF_ALU_IMMEDIATE
;
12735 off_reg
= issrc
? insn
->src_reg
: insn
->dst_reg
;
12737 *patch
++ = BPF_MOV32_IMM(BPF_REG_AX
, aux
->alu_limit
);
12740 *patch
++ = BPF_ALU64_IMM(BPF_MUL
, off_reg
, -1);
12741 *patch
++ = BPF_MOV32_IMM(BPF_REG_AX
, aux
->alu_limit
);
12742 *patch
++ = BPF_ALU64_REG(BPF_SUB
, BPF_REG_AX
, off_reg
);
12743 *patch
++ = BPF_ALU64_REG(BPF_OR
, BPF_REG_AX
, off_reg
);
12744 *patch
++ = BPF_ALU64_IMM(BPF_NEG
, BPF_REG_AX
, 0);
12745 *patch
++ = BPF_ALU64_IMM(BPF_ARSH
, BPF_REG_AX
, 63);
12746 *patch
++ = BPF_ALU64_REG(BPF_AND
, BPF_REG_AX
, off_reg
);
12749 *patch
++ = BPF_MOV64_REG(insn
->dst_reg
, insn
->src_reg
);
12750 insn
->src_reg
= BPF_REG_AX
;
12752 insn
->code
= insn
->code
== code_add
?
12753 code_sub
: code_add
;
12755 if (issrc
&& isneg
&& !isimm
)
12756 *patch
++ = BPF_ALU64_IMM(BPF_MUL
, off_reg
, -1);
12757 cnt
= patch
- insn_buf
;
12759 new_prog
= bpf_patch_insn_data(env
, i
+ delta
, insn_buf
, cnt
);
12764 env
->prog
= prog
= new_prog
;
12765 insn
= new_prog
->insnsi
+ i
+ delta
;
12769 if (insn
->code
!= (BPF_JMP
| BPF_CALL
))
12771 if (insn
->src_reg
== BPF_PSEUDO_CALL
)
12773 if (insn
->src_reg
== BPF_PSEUDO_KFUNC_CALL
) {
12774 ret
= fixup_kfunc_call(env
, insn
);
12780 if (insn
->imm
== BPF_FUNC_get_route_realm
)
12781 prog
->dst_needed
= 1;
12782 if (insn
->imm
== BPF_FUNC_get_prandom_u32
)
12783 bpf_user_rnd_init_once();
12784 if (insn
->imm
== BPF_FUNC_override_return
)
12785 prog
->kprobe_override
= 1;
12786 if (insn
->imm
== BPF_FUNC_tail_call
) {
12787 /* If we tail call into other programs, we
12788 * cannot make any assumptions since they can
12789 * be replaced dynamically during runtime in
12790 * the program array.
12792 prog
->cb_access
= 1;
12793 if (!allow_tail_call_in_subprogs(env
))
12794 prog
->aux
->stack_depth
= MAX_BPF_STACK
;
12795 prog
->aux
->max_pkt_offset
= MAX_PACKET_OFF
;
12797 /* mark bpf_tail_call as different opcode to avoid
12798 * conditional branch in the interpreter for every normal
12799 * call and to prevent accidental JITing by JIT compiler
12800 * that doesn't support bpf_tail_call yet
12803 insn
->code
= BPF_JMP
| BPF_TAIL_CALL
;
12805 aux
= &env
->insn_aux_data
[i
+ delta
];
12806 if (env
->bpf_capable
&& !expect_blinding
&&
12807 prog
->jit_requested
&&
12808 !bpf_map_key_poisoned(aux
) &&
12809 !bpf_map_ptr_poisoned(aux
) &&
12810 !bpf_map_ptr_unpriv(aux
)) {
12811 struct bpf_jit_poke_descriptor desc
= {
12812 .reason
= BPF_POKE_REASON_TAIL_CALL
,
12813 .tail_call
.map
= BPF_MAP_PTR(aux
->map_ptr_state
),
12814 .tail_call
.key
= bpf_map_key_immediate(aux
),
12815 .insn_idx
= i
+ delta
,
12818 ret
= bpf_jit_add_poke_descriptor(prog
, &desc
);
12820 verbose(env
, "adding tail call poke descriptor failed\n");
12824 insn
->imm
= ret
+ 1;
12828 if (!bpf_map_ptr_unpriv(aux
))
12831 /* instead of changing every JIT dealing with tail_call
12832 * emit two extra insns:
12833 * if (index >= max_entries) goto out;
12834 * index &= array->index_mask;
12835 * to avoid out-of-bounds cpu speculation
12837 if (bpf_map_ptr_poisoned(aux
)) {
12838 verbose(env
, "tail_call abusing map_ptr\n");
12842 map_ptr
= BPF_MAP_PTR(aux
->map_ptr_state
);
12843 insn_buf
[0] = BPF_JMP_IMM(BPF_JGE
, BPF_REG_3
,
12844 map_ptr
->max_entries
, 2);
12845 insn_buf
[1] = BPF_ALU32_IMM(BPF_AND
, BPF_REG_3
,
12846 container_of(map_ptr
,
12849 insn_buf
[2] = *insn
;
12851 new_prog
= bpf_patch_insn_data(env
, i
+ delta
, insn_buf
, cnt
);
12856 env
->prog
= prog
= new_prog
;
12857 insn
= new_prog
->insnsi
+ i
+ delta
;
12861 if (insn
->imm
== BPF_FUNC_timer_set_callback
) {
12862 /* The verifier will process callback_fn as many times as necessary
12863 * with different maps and the register states prepared by
12864 * set_timer_callback_state will be accurate.
12866 * The following use case is valid:
12867 * map1 is shared by prog1, prog2, prog3.
12868 * prog1 calls bpf_timer_init for some map1 elements
12869 * prog2 calls bpf_timer_set_callback for some map1 elements.
12870 * Those that were not bpf_timer_init-ed will return -EINVAL.
12871 * prog3 calls bpf_timer_start for some map1 elements.
12872 * Those that were not both bpf_timer_init-ed and
12873 * bpf_timer_set_callback-ed will return -EINVAL.
12875 struct bpf_insn ld_addrs
[2] = {
12876 BPF_LD_IMM64(BPF_REG_3
, (long)prog
->aux
),
12879 insn_buf
[0] = ld_addrs
[0];
12880 insn_buf
[1] = ld_addrs
[1];
12881 insn_buf
[2] = *insn
;
12884 new_prog
= bpf_patch_insn_data(env
, i
+ delta
, insn_buf
, cnt
);
12889 env
->prog
= prog
= new_prog
;
12890 insn
= new_prog
->insnsi
+ i
+ delta
;
12891 goto patch_call_imm
;
12894 /* BPF_EMIT_CALL() assumptions in some of the map_gen_lookup
12895 * and other inlining handlers are currently limited to 64 bit
12898 if (prog
->jit_requested
&& BITS_PER_LONG
== 64 &&
12899 (insn
->imm
== BPF_FUNC_map_lookup_elem
||
12900 insn
->imm
== BPF_FUNC_map_update_elem
||
12901 insn
->imm
== BPF_FUNC_map_delete_elem
||
12902 insn
->imm
== BPF_FUNC_map_push_elem
||
12903 insn
->imm
== BPF_FUNC_map_pop_elem
||
12904 insn
->imm
== BPF_FUNC_map_peek_elem
||
12905 insn
->imm
== BPF_FUNC_redirect_map
)) {
12906 aux
= &env
->insn_aux_data
[i
+ delta
];
12907 if (bpf_map_ptr_poisoned(aux
))
12908 goto patch_call_imm
;
12910 map_ptr
= BPF_MAP_PTR(aux
->map_ptr_state
);
12911 ops
= map_ptr
->ops
;
12912 if (insn
->imm
== BPF_FUNC_map_lookup_elem
&&
12913 ops
->map_gen_lookup
) {
12914 cnt
= ops
->map_gen_lookup(map_ptr
, insn_buf
);
12915 if (cnt
== -EOPNOTSUPP
)
12916 goto patch_map_ops_generic
;
12917 if (cnt
<= 0 || cnt
>= ARRAY_SIZE(insn_buf
)) {
12918 verbose(env
, "bpf verifier is misconfigured\n");
12922 new_prog
= bpf_patch_insn_data(env
, i
+ delta
,
12928 env
->prog
= prog
= new_prog
;
12929 insn
= new_prog
->insnsi
+ i
+ delta
;
12933 BUILD_BUG_ON(!__same_type(ops
->map_lookup_elem
,
12934 (void *(*)(struct bpf_map
*map
, void *key
))NULL
));
12935 BUILD_BUG_ON(!__same_type(ops
->map_delete_elem
,
12936 (int (*)(struct bpf_map
*map
, void *key
))NULL
));
12937 BUILD_BUG_ON(!__same_type(ops
->map_update_elem
,
12938 (int (*)(struct bpf_map
*map
, void *key
, void *value
,
12940 BUILD_BUG_ON(!__same_type(ops
->map_push_elem
,
12941 (int (*)(struct bpf_map
*map
, void *value
,
12943 BUILD_BUG_ON(!__same_type(ops
->map_pop_elem
,
12944 (int (*)(struct bpf_map
*map
, void *value
))NULL
));
12945 BUILD_BUG_ON(!__same_type(ops
->map_peek_elem
,
12946 (int (*)(struct bpf_map
*map
, void *value
))NULL
));
12947 BUILD_BUG_ON(!__same_type(ops
->map_redirect
,
12948 (int (*)(struct bpf_map
*map
, u32 ifindex
, u64 flags
))NULL
));
12950 patch_map_ops_generic
:
12951 switch (insn
->imm
) {
12952 case BPF_FUNC_map_lookup_elem
:
12953 insn
->imm
= BPF_CAST_CALL(ops
->map_lookup_elem
) -
12956 case BPF_FUNC_map_update_elem
:
12957 insn
->imm
= BPF_CAST_CALL(ops
->map_update_elem
) -
12960 case BPF_FUNC_map_delete_elem
:
12961 insn
->imm
= BPF_CAST_CALL(ops
->map_delete_elem
) -
12964 case BPF_FUNC_map_push_elem
:
12965 insn
->imm
= BPF_CAST_CALL(ops
->map_push_elem
) -
12968 case BPF_FUNC_map_pop_elem
:
12969 insn
->imm
= BPF_CAST_CALL(ops
->map_pop_elem
) -
12972 case BPF_FUNC_map_peek_elem
:
12973 insn
->imm
= BPF_CAST_CALL(ops
->map_peek_elem
) -
12976 case BPF_FUNC_redirect_map
:
12977 insn
->imm
= BPF_CAST_CALL(ops
->map_redirect
) -
12982 goto patch_call_imm
;
12985 /* Implement bpf_jiffies64 inline. */
12986 if (prog
->jit_requested
&& BITS_PER_LONG
== 64 &&
12987 insn
->imm
== BPF_FUNC_jiffies64
) {
12988 struct bpf_insn ld_jiffies_addr
[2] = {
12989 BPF_LD_IMM64(BPF_REG_0
,
12990 (unsigned long)&jiffies
),
12993 insn_buf
[0] = ld_jiffies_addr
[0];
12994 insn_buf
[1] = ld_jiffies_addr
[1];
12995 insn_buf
[2] = BPF_LDX_MEM(BPF_DW
, BPF_REG_0
,
12999 new_prog
= bpf_patch_insn_data(env
, i
+ delta
, insn_buf
,
13005 env
->prog
= prog
= new_prog
;
13006 insn
= new_prog
->insnsi
+ i
+ delta
;
13010 /* Implement bpf_get_func_ip inline. */
13011 if (prog_type
== BPF_PROG_TYPE_TRACING
&&
13012 insn
->imm
== BPF_FUNC_get_func_ip
) {
13013 /* Load IP address from ctx - 8 */
13014 insn_buf
[0] = BPF_LDX_MEM(BPF_DW
, BPF_REG_0
, BPF_REG_1
, -8);
13016 new_prog
= bpf_patch_insn_data(env
, i
+ delta
, insn_buf
, 1);
13020 env
->prog
= prog
= new_prog
;
13021 insn
= new_prog
->insnsi
+ i
+ delta
;
13026 fn
= env
->ops
->get_func_proto(insn
->imm
, env
->prog
);
13027 /* all functions that have prototype and verifier allowed
13028 * programs to call them, must be real in-kernel functions
13032 "kernel subsystem misconfigured func %s#%d\n",
13033 func_id_name(insn
->imm
), insn
->imm
);
13036 insn
->imm
= fn
->func
- __bpf_call_base
;
13039 /* Since poke tab is now finalized, publish aux to tracker. */
13040 for (i
= 0; i
< prog
->aux
->size_poke_tab
; i
++) {
13041 map_ptr
= prog
->aux
->poke_tab
[i
].tail_call
.map
;
13042 if (!map_ptr
->ops
->map_poke_track
||
13043 !map_ptr
->ops
->map_poke_untrack
||
13044 !map_ptr
->ops
->map_poke_run
) {
13045 verbose(env
, "bpf verifier is misconfigured\n");
13049 ret
= map_ptr
->ops
->map_poke_track(map_ptr
, prog
->aux
);
13051 verbose(env
, "tracking tail call prog failed\n");
13056 sort_kfunc_descs_by_imm(env
->prog
);
13061 static void free_states(struct bpf_verifier_env
*env
)
13063 struct bpf_verifier_state_list
*sl
, *sln
;
13066 sl
= env
->free_list
;
13069 free_verifier_state(&sl
->state
, false);
13073 env
->free_list
= NULL
;
13075 if (!env
->explored_states
)
13078 for (i
= 0; i
< state_htab_size(env
); i
++) {
13079 sl
= env
->explored_states
[i
];
13083 free_verifier_state(&sl
->state
, false);
13087 env
->explored_states
[i
] = NULL
;
13091 static int do_check_common(struct bpf_verifier_env
*env
, int subprog
)
13093 bool pop_log
= !(env
->log
.level
& BPF_LOG_LEVEL2
);
13094 struct bpf_verifier_state
*state
;
13095 struct bpf_reg_state
*regs
;
13098 env
->prev_linfo
= NULL
;
13101 state
= kzalloc(sizeof(struct bpf_verifier_state
), GFP_KERNEL
);
13104 state
->curframe
= 0;
13105 state
->speculative
= false;
13106 state
->branches
= 1;
13107 state
->frame
[0] = kzalloc(sizeof(struct bpf_func_state
), GFP_KERNEL
);
13108 if (!state
->frame
[0]) {
13112 env
->cur_state
= state
;
13113 init_func_state(env
, state
->frame
[0],
13114 BPF_MAIN_FUNC
/* callsite */,
13118 regs
= state
->frame
[state
->curframe
]->regs
;
13119 if (subprog
|| env
->prog
->type
== BPF_PROG_TYPE_EXT
) {
13120 ret
= btf_prepare_func_args(env
, subprog
, regs
);
13123 for (i
= BPF_REG_1
; i
<= BPF_REG_5
; i
++) {
13124 if (regs
[i
].type
== PTR_TO_CTX
)
13125 mark_reg_known_zero(env
, regs
, i
);
13126 else if (regs
[i
].type
== SCALAR_VALUE
)
13127 mark_reg_unknown(env
, regs
, i
);
13128 else if (regs
[i
].type
== PTR_TO_MEM_OR_NULL
) {
13129 const u32 mem_size
= regs
[i
].mem_size
;
13131 mark_reg_known_zero(env
, regs
, i
);
13132 regs
[i
].mem_size
= mem_size
;
13133 regs
[i
].id
= ++env
->id_gen
;
13137 /* 1st arg to a function */
13138 regs
[BPF_REG_1
].type
= PTR_TO_CTX
;
13139 mark_reg_known_zero(env
, regs
, BPF_REG_1
);
13140 ret
= btf_check_subprog_arg_match(env
, subprog
, regs
);
13141 if (ret
== -EFAULT
)
13142 /* unlikely verifier bug. abort.
13143 * ret == 0 and ret < 0 are sadly acceptable for
13144 * main() function due to backward compatibility.
13145 * Like socket filter program may be written as:
13146 * int bpf_prog(struct pt_regs *ctx)
13147 * and never dereference that ctx in the program.
13148 * 'struct pt_regs' is a type mismatch for socket
13149 * filter that should be using 'struct __sk_buff'.
13154 ret
= do_check(env
);
13156 /* check for NULL is necessary, since cur_state can be freed inside
13157 * do_check() under memory pressure.
13159 if (env
->cur_state
) {
13160 free_verifier_state(env
->cur_state
, true);
13161 env
->cur_state
= NULL
;
13163 while (!pop_stack(env
, NULL
, NULL
, false));
13164 if (!ret
&& pop_log
)
13165 bpf_vlog_reset(&env
->log
, 0);
13170 /* Verify all global functions in a BPF program one by one based on their BTF.
13171 * All global functions must pass verification. Otherwise the whole program is rejected.
13182 * foo() will be verified first for R1=any_scalar_value. During verification it
13183 * will be assumed that bar() already verified successfully and call to bar()
13184 * from foo() will be checked for type match only. Later bar() will be verified
13185 * independently to check that it's safe for R1=any_scalar_value.
13187 static int do_check_subprogs(struct bpf_verifier_env
*env
)
13189 struct bpf_prog_aux
*aux
= env
->prog
->aux
;
13192 if (!aux
->func_info
)
13195 for (i
= 1; i
< env
->subprog_cnt
; i
++) {
13196 if (aux
->func_info_aux
[i
].linkage
!= BTF_FUNC_GLOBAL
)
13198 env
->insn_idx
= env
->subprog_info
[i
].start
;
13199 WARN_ON_ONCE(env
->insn_idx
== 0);
13200 ret
= do_check_common(env
, i
);
13203 } else if (env
->log
.level
& BPF_LOG_LEVEL
) {
13205 "Func#%d is safe for any args that match its prototype\n",
13212 static int do_check_main(struct bpf_verifier_env
*env
)
13217 ret
= do_check_common(env
, 0);
13219 env
->prog
->aux
->stack_depth
= env
->subprog_info
[0].stack_depth
;
13224 static void print_verification_stats(struct bpf_verifier_env
*env
)
13228 if (env
->log
.level
& BPF_LOG_STATS
) {
13229 verbose(env
, "verification time %lld usec\n",
13230 div_u64(env
->verification_time
, 1000));
13231 verbose(env
, "stack depth ");
13232 for (i
= 0; i
< env
->subprog_cnt
; i
++) {
13233 u32 depth
= env
->subprog_info
[i
].stack_depth
;
13235 verbose(env
, "%d", depth
);
13236 if (i
+ 1 < env
->subprog_cnt
)
13239 verbose(env
, "\n");
13241 verbose(env
, "processed %d insns (limit %d) max_states_per_insn %d "
13242 "total_states %d peak_states %d mark_read %d\n",
13243 env
->insn_processed
, BPF_COMPLEXITY_LIMIT_INSNS
,
13244 env
->max_states_per_insn
, env
->total_states
,
13245 env
->peak_states
, env
->longest_mark_read_walk
);
13248 static int check_struct_ops_btf_id(struct bpf_verifier_env
*env
)
13250 const struct btf_type
*t
, *func_proto
;
13251 const struct bpf_struct_ops
*st_ops
;
13252 const struct btf_member
*member
;
13253 struct bpf_prog
*prog
= env
->prog
;
13254 u32 btf_id
, member_idx
;
13257 if (!prog
->gpl_compatible
) {
13258 verbose(env
, "struct ops programs must have a GPL compatible license\n");
13262 btf_id
= prog
->aux
->attach_btf_id
;
13263 st_ops
= bpf_struct_ops_find(btf_id
);
13265 verbose(env
, "attach_btf_id %u is not a supported struct\n",
13271 member_idx
= prog
->expected_attach_type
;
13272 if (member_idx
>= btf_type_vlen(t
)) {
13273 verbose(env
, "attach to invalid member idx %u of struct %s\n",
13274 member_idx
, st_ops
->name
);
13278 member
= &btf_type_member(t
)[member_idx
];
13279 mname
= btf_name_by_offset(btf_vmlinux
, member
->name_off
);
13280 func_proto
= btf_type_resolve_func_ptr(btf_vmlinux
, member
->type
,
13283 verbose(env
, "attach to invalid member %s(@idx %u) of struct %s\n",
13284 mname
, member_idx
, st_ops
->name
);
13288 if (st_ops
->check_member
) {
13289 int err
= st_ops
->check_member(t
, member
);
13292 verbose(env
, "attach to unsupported member %s of struct %s\n",
13293 mname
, st_ops
->name
);
13298 prog
->aux
->attach_func_proto
= func_proto
;
13299 prog
->aux
->attach_func_name
= mname
;
13300 env
->ops
= st_ops
->verifier_ops
;
13304 #define SECURITY_PREFIX "security_"
13306 static int check_attach_modify_return(unsigned long addr
, const char *func_name
)
13308 if (within_error_injection_list(addr
) ||
13309 !strncmp(SECURITY_PREFIX
, func_name
, sizeof(SECURITY_PREFIX
) - 1))
13315 /* list of non-sleepable functions that are otherwise on
13316 * ALLOW_ERROR_INJECTION list
13318 BTF_SET_START(btf_non_sleepable_error_inject
)
13319 /* Three functions below can be called from sleepable and non-sleepable context.
13320 * Assume non-sleepable from bpf safety point of view.
13322 BTF_ID(func
, __add_to_page_cache_locked
)
13323 BTF_ID(func
, should_fail_alloc_page
)
13324 BTF_ID(func
, should_failslab
)
13325 BTF_SET_END(btf_non_sleepable_error_inject
)
13327 static int check_non_sleepable_error_inject(u32 btf_id
)
13329 return btf_id_set_contains(&btf_non_sleepable_error_inject
, btf_id
);
13332 int bpf_check_attach_target(struct bpf_verifier_log
*log
,
13333 const struct bpf_prog
*prog
,
13334 const struct bpf_prog
*tgt_prog
,
13336 struct bpf_attach_target_info
*tgt_info
)
13338 bool prog_extension
= prog
->type
== BPF_PROG_TYPE_EXT
;
13339 const char prefix
[] = "btf_trace_";
13340 int ret
= 0, subprog
= -1, i
;
13341 const struct btf_type
*t
;
13342 bool conservative
= true;
13348 bpf_log(log
, "Tracing programs must provide btf_id\n");
13351 btf
= tgt_prog
? tgt_prog
->aux
->btf
: prog
->aux
->attach_btf
;
13354 "FENTRY/FEXIT program can only be attached to another program annotated with BTF\n");
13357 t
= btf_type_by_id(btf
, btf_id
);
13359 bpf_log(log
, "attach_btf_id %u is invalid\n", btf_id
);
13362 tname
= btf_name_by_offset(btf
, t
->name_off
);
13364 bpf_log(log
, "attach_btf_id %u doesn't have a name\n", btf_id
);
13368 struct bpf_prog_aux
*aux
= tgt_prog
->aux
;
13370 for (i
= 0; i
< aux
->func_info_cnt
; i
++)
13371 if (aux
->func_info
[i
].type_id
== btf_id
) {
13375 if (subprog
== -1) {
13376 bpf_log(log
, "Subprog %s doesn't exist\n", tname
);
13379 conservative
= aux
->func_info_aux
[subprog
].unreliable
;
13380 if (prog_extension
) {
13381 if (conservative
) {
13383 "Cannot replace static functions\n");
13386 if (!prog
->jit_requested
) {
13388 "Extension programs should be JITed\n");
13392 if (!tgt_prog
->jited
) {
13393 bpf_log(log
, "Can attach to only JITed progs\n");
13396 if (tgt_prog
->type
== prog
->type
) {
13397 /* Cannot fentry/fexit another fentry/fexit program.
13398 * Cannot attach program extension to another extension.
13399 * It's ok to attach fentry/fexit to extension program.
13401 bpf_log(log
, "Cannot recursively attach\n");
13404 if (tgt_prog
->type
== BPF_PROG_TYPE_TRACING
&&
13406 (tgt_prog
->expected_attach_type
== BPF_TRACE_FENTRY
||
13407 tgt_prog
->expected_attach_type
== BPF_TRACE_FEXIT
)) {
13408 /* Program extensions can extend all program types
13409 * except fentry/fexit. The reason is the following.
13410 * The fentry/fexit programs are used for performance
13411 * analysis, stats and can be attached to any program
13412 * type except themselves. When extension program is
13413 * replacing XDP function it is necessary to allow
13414 * performance analysis of all functions. Both original
13415 * XDP program and its program extension. Hence
13416 * attaching fentry/fexit to BPF_PROG_TYPE_EXT is
13417 * allowed. If extending of fentry/fexit was allowed it
13418 * would be possible to create long call chain
13419 * fentry->extension->fentry->extension beyond
13420 * reasonable stack size. Hence extending fentry is not
13423 bpf_log(log
, "Cannot extend fentry/fexit\n");
13427 if (prog_extension
) {
13428 bpf_log(log
, "Cannot replace kernel functions\n");
13433 switch (prog
->expected_attach_type
) {
13434 case BPF_TRACE_RAW_TP
:
13437 "Only FENTRY/FEXIT progs are attachable to another BPF prog\n");
13440 if (!btf_type_is_typedef(t
)) {
13441 bpf_log(log
, "attach_btf_id %u is not a typedef\n",
13445 if (strncmp(prefix
, tname
, sizeof(prefix
) - 1)) {
13446 bpf_log(log
, "attach_btf_id %u points to wrong type name %s\n",
13450 tname
+= sizeof(prefix
) - 1;
13451 t
= btf_type_by_id(btf
, t
->type
);
13452 if (!btf_type_is_ptr(t
))
13453 /* should never happen in valid vmlinux build */
13455 t
= btf_type_by_id(btf
, t
->type
);
13456 if (!btf_type_is_func_proto(t
))
13457 /* should never happen in valid vmlinux build */
13461 case BPF_TRACE_ITER
:
13462 if (!btf_type_is_func(t
)) {
13463 bpf_log(log
, "attach_btf_id %u is not a function\n",
13467 t
= btf_type_by_id(btf
, t
->type
);
13468 if (!btf_type_is_func_proto(t
))
13470 ret
= btf_distill_func_proto(log
, btf
, t
, tname
, &tgt_info
->fmodel
);
13475 if (!prog_extension
)
13478 case BPF_MODIFY_RETURN
:
13480 case BPF_TRACE_FENTRY
:
13481 case BPF_TRACE_FEXIT
:
13482 if (!btf_type_is_func(t
)) {
13483 bpf_log(log
, "attach_btf_id %u is not a function\n",
13487 if (prog_extension
&&
13488 btf_check_type_match(log
, prog
, btf
, t
))
13490 t
= btf_type_by_id(btf
, t
->type
);
13491 if (!btf_type_is_func_proto(t
))
13494 if ((prog
->aux
->saved_dst_prog_type
|| prog
->aux
->saved_dst_attach_type
) &&
13495 (!tgt_prog
|| prog
->aux
->saved_dst_prog_type
!= tgt_prog
->type
||
13496 prog
->aux
->saved_dst_attach_type
!= tgt_prog
->expected_attach_type
))
13499 if (tgt_prog
&& conservative
)
13502 ret
= btf_distill_func_proto(log
, btf
, t
, tname
, &tgt_info
->fmodel
);
13508 addr
= (long) tgt_prog
->bpf_func
;
13510 addr
= (long) tgt_prog
->aux
->func
[subprog
]->bpf_func
;
13512 addr
= kallsyms_lookup_name(tname
);
13515 "The address of function %s cannot be found\n",
13521 if (prog
->aux
->sleepable
) {
13523 switch (prog
->type
) {
13524 case BPF_PROG_TYPE_TRACING
:
13525 /* fentry/fexit/fmod_ret progs can be sleepable only if they are
13526 * attached to ALLOW_ERROR_INJECTION and are not in denylist.
13528 if (!check_non_sleepable_error_inject(btf_id
) &&
13529 within_error_injection_list(addr
))
13532 case BPF_PROG_TYPE_LSM
:
13533 /* LSM progs check that they are attached to bpf_lsm_*() funcs.
13534 * Only some of them are sleepable.
13536 if (bpf_lsm_is_sleepable_hook(btf_id
))
13543 bpf_log(log
, "%s is not sleepable\n", tname
);
13546 } else if (prog
->expected_attach_type
== BPF_MODIFY_RETURN
) {
13548 bpf_log(log
, "can't modify return codes of BPF programs\n");
13551 ret
= check_attach_modify_return(addr
, tname
);
13553 bpf_log(log
, "%s() is not modifiable\n", tname
);
13560 tgt_info
->tgt_addr
= addr
;
13561 tgt_info
->tgt_name
= tname
;
13562 tgt_info
->tgt_type
= t
;
13566 BTF_SET_START(btf_id_deny
)
13569 BTF_ID(func
, migrate_disable
)
13570 BTF_ID(func
, migrate_enable
)
13572 #if !defined CONFIG_PREEMPT_RCU && !defined CONFIG_TINY_RCU
13573 BTF_ID(func
, rcu_read_unlock_strict
)
13575 BTF_SET_END(btf_id_deny
)
13577 static int check_attach_btf_id(struct bpf_verifier_env
*env
)
13579 struct bpf_prog
*prog
= env
->prog
;
13580 struct bpf_prog
*tgt_prog
= prog
->aux
->dst_prog
;
13581 struct bpf_attach_target_info tgt_info
= {};
13582 u32 btf_id
= prog
->aux
->attach_btf_id
;
13583 struct bpf_trampoline
*tr
;
13587 if (prog
->type
== BPF_PROG_TYPE_SYSCALL
) {
13588 if (prog
->aux
->sleepable
)
13589 /* attach_btf_id checked to be zero already */
13591 verbose(env
, "Syscall programs can only be sleepable\n");
13595 if (prog
->aux
->sleepable
&& prog
->type
!= BPF_PROG_TYPE_TRACING
&&
13596 prog
->type
!= BPF_PROG_TYPE_LSM
) {
13597 verbose(env
, "Only fentry/fexit/fmod_ret and lsm programs can be sleepable\n");
13601 if (prog
->type
== BPF_PROG_TYPE_STRUCT_OPS
)
13602 return check_struct_ops_btf_id(env
);
13604 if (prog
->type
!= BPF_PROG_TYPE_TRACING
&&
13605 prog
->type
!= BPF_PROG_TYPE_LSM
&&
13606 prog
->type
!= BPF_PROG_TYPE_EXT
)
13609 ret
= bpf_check_attach_target(&env
->log
, prog
, tgt_prog
, btf_id
, &tgt_info
);
13613 if (tgt_prog
&& prog
->type
== BPF_PROG_TYPE_EXT
) {
13614 /* to make freplace equivalent to their targets, they need to
13615 * inherit env->ops and expected_attach_type for the rest of the
13618 env
->ops
= bpf_verifier_ops
[tgt_prog
->type
];
13619 prog
->expected_attach_type
= tgt_prog
->expected_attach_type
;
13622 /* store info about the attachment target that will be used later */
13623 prog
->aux
->attach_func_proto
= tgt_info
.tgt_type
;
13624 prog
->aux
->attach_func_name
= tgt_info
.tgt_name
;
13627 prog
->aux
->saved_dst_prog_type
= tgt_prog
->type
;
13628 prog
->aux
->saved_dst_attach_type
= tgt_prog
->expected_attach_type
;
13631 if (prog
->expected_attach_type
== BPF_TRACE_RAW_TP
) {
13632 prog
->aux
->attach_btf_trace
= true;
13634 } else if (prog
->expected_attach_type
== BPF_TRACE_ITER
) {
13635 if (!bpf_iter_prog_supported(prog
))
13640 if (prog
->type
== BPF_PROG_TYPE_LSM
) {
13641 ret
= bpf_lsm_verify_prog(&env
->log
, prog
);
13644 } else if (prog
->type
== BPF_PROG_TYPE_TRACING
&&
13645 btf_id_set_contains(&btf_id_deny
, btf_id
)) {
13649 key
= bpf_trampoline_compute_key(tgt_prog
, prog
->aux
->attach_btf
, btf_id
);
13650 tr
= bpf_trampoline_get(key
, &tgt_info
);
13654 prog
->aux
->dst_trampoline
= tr
;
13658 struct btf
*bpf_get_btf_vmlinux(void)
13660 if (!btf_vmlinux
&& IS_ENABLED(CONFIG_DEBUG_INFO_BTF
)) {
13661 mutex_lock(&bpf_verifier_lock
);
13663 btf_vmlinux
= btf_parse_vmlinux();
13664 mutex_unlock(&bpf_verifier_lock
);
13666 return btf_vmlinux
;
13669 int bpf_check(struct bpf_prog
**prog
, union bpf_attr
*attr
, bpfptr_t uattr
)
13671 u64 start_time
= ktime_get_ns();
13672 struct bpf_verifier_env
*env
;
13673 struct bpf_verifier_log
*log
;
13674 int i
, len
, ret
= -EINVAL
;
13677 /* no program is valid */
13678 if (ARRAY_SIZE(bpf_verifier_ops
) == 0)
13681 /* 'struct bpf_verifier_env' can be global, but since it's not small,
13682 * allocate/free it every time bpf_check() is called
13684 env
= kzalloc(sizeof(struct bpf_verifier_env
), GFP_KERNEL
);
13689 len
= (*prog
)->len
;
13690 env
->insn_aux_data
=
13691 vzalloc(array_size(sizeof(struct bpf_insn_aux_data
), len
));
13693 if (!env
->insn_aux_data
)
13695 for (i
= 0; i
< len
; i
++)
13696 env
->insn_aux_data
[i
].orig_idx
= i
;
13698 env
->ops
= bpf_verifier_ops
[env
->prog
->type
];
13699 env
->fd_array
= make_bpfptr(attr
->fd_array
, uattr
.is_kernel
);
13700 is_priv
= bpf_capable();
13702 bpf_get_btf_vmlinux();
13704 /* grab the mutex to protect few globals used by verifier */
13706 mutex_lock(&bpf_verifier_lock
);
13708 if (attr
->log_level
|| attr
->log_buf
|| attr
->log_size
) {
13709 /* user requested verbose verifier output
13710 * and supplied buffer to store the verification trace
13712 log
->level
= attr
->log_level
;
13713 log
->ubuf
= (char __user
*) (unsigned long) attr
->log_buf
;
13714 log
->len_total
= attr
->log_size
;
13717 /* log attributes have to be sane */
13718 if (log
->len_total
< 128 || log
->len_total
> UINT_MAX
>> 2 ||
13719 !log
->level
|| !log
->ubuf
|| log
->level
& ~BPF_LOG_MASK
)
13723 if (IS_ERR(btf_vmlinux
)) {
13724 /* Either gcc or pahole or kernel are broken. */
13725 verbose(env
, "in-kernel BTF is malformed\n");
13726 ret
= PTR_ERR(btf_vmlinux
);
13727 goto skip_full_check
;
13730 env
->strict_alignment
= !!(attr
->prog_flags
& BPF_F_STRICT_ALIGNMENT
);
13731 if (!IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS
))
13732 env
->strict_alignment
= true;
13733 if (attr
->prog_flags
& BPF_F_ANY_ALIGNMENT
)
13734 env
->strict_alignment
= false;
13736 env
->allow_ptr_leaks
= bpf_allow_ptr_leaks();
13737 env
->allow_uninit_stack
= bpf_allow_uninit_stack();
13738 env
->allow_ptr_to_map_access
= bpf_allow_ptr_to_map_access();
13739 env
->bypass_spec_v1
= bpf_bypass_spec_v1();
13740 env
->bypass_spec_v4
= bpf_bypass_spec_v4();
13741 env
->bpf_capable
= bpf_capable();
13744 env
->test_state_freq
= attr
->prog_flags
& BPF_F_TEST_STATE_FREQ
;
13746 env
->explored_states
= kvcalloc(state_htab_size(env
),
13747 sizeof(struct bpf_verifier_state_list
*),
13750 if (!env
->explored_states
)
13751 goto skip_full_check
;
13753 ret
= add_subprog_and_kfunc(env
);
13755 goto skip_full_check
;
13757 ret
= check_subprogs(env
);
13759 goto skip_full_check
;
13761 ret
= check_btf_info(env
, attr
, uattr
);
13763 goto skip_full_check
;
13765 ret
= check_attach_btf_id(env
);
13767 goto skip_full_check
;
13769 ret
= resolve_pseudo_ldimm64(env
);
13771 goto skip_full_check
;
13773 if (bpf_prog_is_dev_bound(env
->prog
->aux
)) {
13774 ret
= bpf_prog_offload_verifier_prep(env
->prog
);
13776 goto skip_full_check
;
13779 ret
= check_cfg(env
);
13781 goto skip_full_check
;
13783 ret
= do_check_subprogs(env
);
13784 ret
= ret
?: do_check_main(env
);
13786 if (ret
== 0 && bpf_prog_is_dev_bound(env
->prog
->aux
))
13787 ret
= bpf_prog_offload_finalize(env
);
13790 kvfree(env
->explored_states
);
13793 ret
= check_max_stack_depth(env
);
13795 /* instruction rewrites happen after this point */
13798 opt_hard_wire_dead_code_branches(env
);
13800 ret
= opt_remove_dead_code(env
);
13802 ret
= opt_remove_nops(env
);
13805 sanitize_dead_code(env
);
13809 /* program is valid, convert *(u32*)(ctx + off) accesses */
13810 ret
= convert_ctx_accesses(env
);
13813 ret
= do_misc_fixups(env
);
13815 /* do 32-bit optimization after insn patching has done so those patched
13816 * insns could be handled correctly.
13818 if (ret
== 0 && !bpf_prog_is_dev_bound(env
->prog
->aux
)) {
13819 ret
= opt_subreg_zext_lo32_rnd_hi32(env
, attr
);
13820 env
->prog
->aux
->verifier_zext
= bpf_jit_needs_zext() ? !ret
13825 ret
= fixup_call_args(env
);
13827 env
->verification_time
= ktime_get_ns() - start_time
;
13828 print_verification_stats(env
);
13830 if (log
->level
&& bpf_verifier_log_full(log
))
13832 if (log
->level
&& !log
->ubuf
) {
13834 goto err_release_maps
;
13838 goto err_release_maps
;
13840 if (env
->used_map_cnt
) {
13841 /* if program passed verifier, update used_maps in bpf_prog_info */
13842 env
->prog
->aux
->used_maps
= kmalloc_array(env
->used_map_cnt
,
13843 sizeof(env
->used_maps
[0]),
13846 if (!env
->prog
->aux
->used_maps
) {
13848 goto err_release_maps
;
13851 memcpy(env
->prog
->aux
->used_maps
, env
->used_maps
,
13852 sizeof(env
->used_maps
[0]) * env
->used_map_cnt
);
13853 env
->prog
->aux
->used_map_cnt
= env
->used_map_cnt
;
13855 if (env
->used_btf_cnt
) {
13856 /* if program passed verifier, update used_btfs in bpf_prog_aux */
13857 env
->prog
->aux
->used_btfs
= kmalloc_array(env
->used_btf_cnt
,
13858 sizeof(env
->used_btfs
[0]),
13860 if (!env
->prog
->aux
->used_btfs
) {
13862 goto err_release_maps
;
13865 memcpy(env
->prog
->aux
->used_btfs
, env
->used_btfs
,
13866 sizeof(env
->used_btfs
[0]) * env
->used_btf_cnt
);
13867 env
->prog
->aux
->used_btf_cnt
= env
->used_btf_cnt
;
13869 if (env
->used_map_cnt
|| env
->used_btf_cnt
) {
13870 /* program is valid. Convert pseudo bpf_ld_imm64 into generic
13871 * bpf_ld_imm64 instructions
13873 convert_pseudo_ld_imm64(env
);
13876 adjust_btf_func(env
);
13879 if (!env
->prog
->aux
->used_maps
)
13880 /* if we didn't copy map pointers into bpf_prog_info, release
13881 * them now. Otherwise free_used_maps() will release them.
13884 if (!env
->prog
->aux
->used_btfs
)
13887 /* extension progs temporarily inherit the attach_type of their targets
13888 for verification purposes, so set it back to zero before returning
13890 if (env
->prog
->type
== BPF_PROG_TYPE_EXT
)
13891 env
->prog
->expected_attach_type
= 0;
13896 mutex_unlock(&bpf_verifier_lock
);
13897 vfree(env
->insn_aux_data
);