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 if (map_value_has_timer(map
->inner_map_meta
))
1147 reg
->map_uid
= reg
->id
;
1148 } else if (map
->map_type
== BPF_MAP_TYPE_XSKMAP
) {
1149 reg
->type
= PTR_TO_XDP_SOCK
;
1150 } else if (map
->map_type
== BPF_MAP_TYPE_SOCKMAP
||
1151 map
->map_type
== BPF_MAP_TYPE_SOCKHASH
) {
1152 reg
->type
= PTR_TO_SOCKET
;
1154 reg
->type
= PTR_TO_MAP_VALUE
;
1158 case PTR_TO_SOCKET_OR_NULL
:
1159 reg
->type
= PTR_TO_SOCKET
;
1161 case PTR_TO_SOCK_COMMON_OR_NULL
:
1162 reg
->type
= PTR_TO_SOCK_COMMON
;
1164 case PTR_TO_TCP_SOCK_OR_NULL
:
1165 reg
->type
= PTR_TO_TCP_SOCK
;
1167 case PTR_TO_BTF_ID_OR_NULL
:
1168 reg
->type
= PTR_TO_BTF_ID
;
1170 case PTR_TO_MEM_OR_NULL
:
1171 reg
->type
= PTR_TO_MEM
;
1173 case PTR_TO_RDONLY_BUF_OR_NULL
:
1174 reg
->type
= PTR_TO_RDONLY_BUF
;
1176 case PTR_TO_RDWR_BUF_OR_NULL
:
1177 reg
->type
= PTR_TO_RDWR_BUF
;
1180 WARN_ONCE(1, "unknown nullable register type");
1184 static bool reg_is_pkt_pointer(const struct bpf_reg_state
*reg
)
1186 return type_is_pkt_pointer(reg
->type
);
1189 static bool reg_is_pkt_pointer_any(const struct bpf_reg_state
*reg
)
1191 return reg_is_pkt_pointer(reg
) ||
1192 reg
->type
== PTR_TO_PACKET_END
;
1195 /* Unmodified PTR_TO_PACKET[_META,_END] register from ctx access. */
1196 static bool reg_is_init_pkt_pointer(const struct bpf_reg_state
*reg
,
1197 enum bpf_reg_type which
)
1199 /* The register can already have a range from prior markings.
1200 * This is fine as long as it hasn't been advanced from its
1203 return reg
->type
== which
&&
1206 tnum_equals_const(reg
->var_off
, 0);
1209 /* Reset the min/max bounds of a register */
1210 static void __mark_reg_unbounded(struct bpf_reg_state
*reg
)
1212 reg
->smin_value
= S64_MIN
;
1213 reg
->smax_value
= S64_MAX
;
1214 reg
->umin_value
= 0;
1215 reg
->umax_value
= U64_MAX
;
1217 reg
->s32_min_value
= S32_MIN
;
1218 reg
->s32_max_value
= S32_MAX
;
1219 reg
->u32_min_value
= 0;
1220 reg
->u32_max_value
= U32_MAX
;
1223 static void __mark_reg64_unbounded(struct bpf_reg_state
*reg
)
1225 reg
->smin_value
= S64_MIN
;
1226 reg
->smax_value
= S64_MAX
;
1227 reg
->umin_value
= 0;
1228 reg
->umax_value
= U64_MAX
;
1231 static void __mark_reg32_unbounded(struct bpf_reg_state
*reg
)
1233 reg
->s32_min_value
= S32_MIN
;
1234 reg
->s32_max_value
= S32_MAX
;
1235 reg
->u32_min_value
= 0;
1236 reg
->u32_max_value
= U32_MAX
;
1239 static void __update_reg32_bounds(struct bpf_reg_state
*reg
)
1241 struct tnum var32_off
= tnum_subreg(reg
->var_off
);
1243 /* min signed is max(sign bit) | min(other bits) */
1244 reg
->s32_min_value
= max_t(s32
, reg
->s32_min_value
,
1245 var32_off
.value
| (var32_off
.mask
& S32_MIN
));
1246 /* max signed is min(sign bit) | max(other bits) */
1247 reg
->s32_max_value
= min_t(s32
, reg
->s32_max_value
,
1248 var32_off
.value
| (var32_off
.mask
& S32_MAX
));
1249 reg
->u32_min_value
= max_t(u32
, reg
->u32_min_value
, (u32
)var32_off
.value
);
1250 reg
->u32_max_value
= min(reg
->u32_max_value
,
1251 (u32
)(var32_off
.value
| var32_off
.mask
));
1254 static void __update_reg64_bounds(struct bpf_reg_state
*reg
)
1256 /* min signed is max(sign bit) | min(other bits) */
1257 reg
->smin_value
= max_t(s64
, reg
->smin_value
,
1258 reg
->var_off
.value
| (reg
->var_off
.mask
& S64_MIN
));
1259 /* max signed is min(sign bit) | max(other bits) */
1260 reg
->smax_value
= min_t(s64
, reg
->smax_value
,
1261 reg
->var_off
.value
| (reg
->var_off
.mask
& S64_MAX
));
1262 reg
->umin_value
= max(reg
->umin_value
, reg
->var_off
.value
);
1263 reg
->umax_value
= min(reg
->umax_value
,
1264 reg
->var_off
.value
| reg
->var_off
.mask
);
1267 static void __update_reg_bounds(struct bpf_reg_state
*reg
)
1269 __update_reg32_bounds(reg
);
1270 __update_reg64_bounds(reg
);
1273 /* Uses signed min/max values to inform unsigned, and vice-versa */
1274 static void __reg32_deduce_bounds(struct bpf_reg_state
*reg
)
1276 /* Learn sign from signed bounds.
1277 * If we cannot cross the sign boundary, then signed and unsigned bounds
1278 * are the same, so combine. This works even in the negative case, e.g.
1279 * -3 s<= x s<= -1 implies 0xf...fd u<= x u<= 0xf...ff.
1281 if (reg
->s32_min_value
>= 0 || reg
->s32_max_value
< 0) {
1282 reg
->s32_min_value
= reg
->u32_min_value
=
1283 max_t(u32
, reg
->s32_min_value
, reg
->u32_min_value
);
1284 reg
->s32_max_value
= reg
->u32_max_value
=
1285 min_t(u32
, reg
->s32_max_value
, reg
->u32_max_value
);
1288 /* Learn sign from unsigned bounds. Signed bounds cross the sign
1289 * boundary, so we must be careful.
1291 if ((s32
)reg
->u32_max_value
>= 0) {
1292 /* Positive. We can't learn anything from the smin, but smax
1293 * is positive, hence safe.
1295 reg
->s32_min_value
= reg
->u32_min_value
;
1296 reg
->s32_max_value
= reg
->u32_max_value
=
1297 min_t(u32
, reg
->s32_max_value
, reg
->u32_max_value
);
1298 } else if ((s32
)reg
->u32_min_value
< 0) {
1299 /* Negative. We can't learn anything from the smax, but smin
1300 * is negative, hence safe.
1302 reg
->s32_min_value
= reg
->u32_min_value
=
1303 max_t(u32
, reg
->s32_min_value
, reg
->u32_min_value
);
1304 reg
->s32_max_value
= reg
->u32_max_value
;
1308 static void __reg64_deduce_bounds(struct bpf_reg_state
*reg
)
1310 /* Learn sign from signed bounds.
1311 * If we cannot cross the sign boundary, then signed and unsigned bounds
1312 * are the same, so combine. This works even in the negative case, e.g.
1313 * -3 s<= x s<= -1 implies 0xf...fd u<= x u<= 0xf...ff.
1315 if (reg
->smin_value
>= 0 || reg
->smax_value
< 0) {
1316 reg
->smin_value
= reg
->umin_value
= max_t(u64
, reg
->smin_value
,
1318 reg
->smax_value
= reg
->umax_value
= min_t(u64
, reg
->smax_value
,
1322 /* Learn sign from unsigned bounds. Signed bounds cross the sign
1323 * boundary, so we must be careful.
1325 if ((s64
)reg
->umax_value
>= 0) {
1326 /* Positive. We can't learn anything from the smin, but smax
1327 * is positive, hence safe.
1329 reg
->smin_value
= reg
->umin_value
;
1330 reg
->smax_value
= reg
->umax_value
= min_t(u64
, reg
->smax_value
,
1332 } else if ((s64
)reg
->umin_value
< 0) {
1333 /* Negative. We can't learn anything from the smax, but smin
1334 * is negative, hence safe.
1336 reg
->smin_value
= reg
->umin_value
= max_t(u64
, reg
->smin_value
,
1338 reg
->smax_value
= reg
->umax_value
;
1342 static void __reg_deduce_bounds(struct bpf_reg_state
*reg
)
1344 __reg32_deduce_bounds(reg
);
1345 __reg64_deduce_bounds(reg
);
1348 /* Attempts to improve var_off based on unsigned min/max information */
1349 static void __reg_bound_offset(struct bpf_reg_state
*reg
)
1351 struct tnum var64_off
= tnum_intersect(reg
->var_off
,
1352 tnum_range(reg
->umin_value
,
1354 struct tnum var32_off
= tnum_intersect(tnum_subreg(reg
->var_off
),
1355 tnum_range(reg
->u32_min_value
,
1356 reg
->u32_max_value
));
1358 reg
->var_off
= tnum_or(tnum_clear_subreg(var64_off
), var32_off
);
1361 static bool __reg32_bound_s64(s32 a
)
1363 return a
>= 0 && a
<= S32_MAX
;
1366 static void __reg_assign_32_into_64(struct bpf_reg_state
*reg
)
1368 reg
->umin_value
= reg
->u32_min_value
;
1369 reg
->umax_value
= reg
->u32_max_value
;
1371 /* Attempt to pull 32-bit signed bounds into 64-bit bounds but must
1372 * be positive otherwise set to worse case bounds and refine later
1375 if (__reg32_bound_s64(reg
->s32_min_value
) &&
1376 __reg32_bound_s64(reg
->s32_max_value
)) {
1377 reg
->smin_value
= reg
->s32_min_value
;
1378 reg
->smax_value
= reg
->s32_max_value
;
1380 reg
->smin_value
= 0;
1381 reg
->smax_value
= U32_MAX
;
1385 static void __reg_combine_32_into_64(struct bpf_reg_state
*reg
)
1387 /* special case when 64-bit register has upper 32-bit register
1388 * zeroed. Typically happens after zext or <<32, >>32 sequence
1389 * allowing us to use 32-bit bounds directly,
1391 if (tnum_equals_const(tnum_clear_subreg(reg
->var_off
), 0)) {
1392 __reg_assign_32_into_64(reg
);
1394 /* Otherwise the best we can do is push lower 32bit known and
1395 * unknown bits into register (var_off set from jmp logic)
1396 * then learn as much as possible from the 64-bit tnum
1397 * known and unknown bits. The previous smin/smax bounds are
1398 * invalid here because of jmp32 compare so mark them unknown
1399 * so they do not impact tnum bounds calculation.
1401 __mark_reg64_unbounded(reg
);
1402 __update_reg_bounds(reg
);
1405 /* Intersecting with the old var_off might have improved our bounds
1406 * slightly. e.g. if umax was 0x7f...f and var_off was (0; 0xf...fc),
1407 * then new var_off is (0; 0x7f...fc) which improves our umax.
1409 __reg_deduce_bounds(reg
);
1410 __reg_bound_offset(reg
);
1411 __update_reg_bounds(reg
);
1414 static bool __reg64_bound_s32(s64 a
)
1416 return a
>= S32_MIN
&& a
<= S32_MAX
;
1419 static bool __reg64_bound_u32(u64 a
)
1421 return a
>= U32_MIN
&& a
<= U32_MAX
;
1424 static void __reg_combine_64_into_32(struct bpf_reg_state
*reg
)
1426 __mark_reg32_unbounded(reg
);
1428 if (__reg64_bound_s32(reg
->smin_value
) && __reg64_bound_s32(reg
->smax_value
)) {
1429 reg
->s32_min_value
= (s32
)reg
->smin_value
;
1430 reg
->s32_max_value
= (s32
)reg
->smax_value
;
1432 if (__reg64_bound_u32(reg
->umin_value
) && __reg64_bound_u32(reg
->umax_value
)) {
1433 reg
->u32_min_value
= (u32
)reg
->umin_value
;
1434 reg
->u32_max_value
= (u32
)reg
->umax_value
;
1437 /* Intersecting with the old var_off might have improved our bounds
1438 * slightly. e.g. if umax was 0x7f...f and var_off was (0; 0xf...fc),
1439 * then new var_off is (0; 0x7f...fc) which improves our umax.
1441 __reg_deduce_bounds(reg
);
1442 __reg_bound_offset(reg
);
1443 __update_reg_bounds(reg
);
1446 /* Mark a register as having a completely unknown (scalar) value. */
1447 static void __mark_reg_unknown(const struct bpf_verifier_env
*env
,
1448 struct bpf_reg_state
*reg
)
1451 * Clear type, id, off, and union(map_ptr, range) and
1452 * padding between 'type' and union
1454 memset(reg
, 0, offsetof(struct bpf_reg_state
, var_off
));
1455 reg
->type
= SCALAR_VALUE
;
1456 reg
->var_off
= tnum_unknown
;
1458 reg
->precise
= env
->subprog_cnt
> 1 || !env
->bpf_capable
;
1459 __mark_reg_unbounded(reg
);
1462 static void mark_reg_unknown(struct bpf_verifier_env
*env
,
1463 struct bpf_reg_state
*regs
, u32 regno
)
1465 if (WARN_ON(regno
>= MAX_BPF_REG
)) {
1466 verbose(env
, "mark_reg_unknown(regs, %u)\n", regno
);
1467 /* Something bad happened, let's kill all regs except FP */
1468 for (regno
= 0; regno
< BPF_REG_FP
; regno
++)
1469 __mark_reg_not_init(env
, regs
+ regno
);
1472 __mark_reg_unknown(env
, regs
+ regno
);
1475 static void __mark_reg_not_init(const struct bpf_verifier_env
*env
,
1476 struct bpf_reg_state
*reg
)
1478 __mark_reg_unknown(env
, reg
);
1479 reg
->type
= NOT_INIT
;
1482 static void mark_reg_not_init(struct bpf_verifier_env
*env
,
1483 struct bpf_reg_state
*regs
, u32 regno
)
1485 if (WARN_ON(regno
>= MAX_BPF_REG
)) {
1486 verbose(env
, "mark_reg_not_init(regs, %u)\n", regno
);
1487 /* Something bad happened, let's kill all regs except FP */
1488 for (regno
= 0; regno
< BPF_REG_FP
; regno
++)
1489 __mark_reg_not_init(env
, regs
+ regno
);
1492 __mark_reg_not_init(env
, regs
+ regno
);
1495 static void mark_btf_ld_reg(struct bpf_verifier_env
*env
,
1496 struct bpf_reg_state
*regs
, u32 regno
,
1497 enum bpf_reg_type reg_type
,
1498 struct btf
*btf
, u32 btf_id
)
1500 if (reg_type
== SCALAR_VALUE
) {
1501 mark_reg_unknown(env
, regs
, regno
);
1504 mark_reg_known_zero(env
, regs
, regno
);
1505 regs
[regno
].type
= PTR_TO_BTF_ID
;
1506 regs
[regno
].btf
= btf
;
1507 regs
[regno
].btf_id
= btf_id
;
1510 #define DEF_NOT_SUBREG (0)
1511 static void init_reg_state(struct bpf_verifier_env
*env
,
1512 struct bpf_func_state
*state
)
1514 struct bpf_reg_state
*regs
= state
->regs
;
1517 for (i
= 0; i
< MAX_BPF_REG
; i
++) {
1518 mark_reg_not_init(env
, regs
, i
);
1519 regs
[i
].live
= REG_LIVE_NONE
;
1520 regs
[i
].parent
= NULL
;
1521 regs
[i
].subreg_def
= DEF_NOT_SUBREG
;
1525 regs
[BPF_REG_FP
].type
= PTR_TO_STACK
;
1526 mark_reg_known_zero(env
, regs
, BPF_REG_FP
);
1527 regs
[BPF_REG_FP
].frameno
= state
->frameno
;
1530 #define BPF_MAIN_FUNC (-1)
1531 static void init_func_state(struct bpf_verifier_env
*env
,
1532 struct bpf_func_state
*state
,
1533 int callsite
, int frameno
, int subprogno
)
1535 state
->callsite
= callsite
;
1536 state
->frameno
= frameno
;
1537 state
->subprogno
= subprogno
;
1538 init_reg_state(env
, state
);
1541 /* Similar to push_stack(), but for async callbacks */
1542 static struct bpf_verifier_state
*push_async_cb(struct bpf_verifier_env
*env
,
1543 int insn_idx
, int prev_insn_idx
,
1546 struct bpf_verifier_stack_elem
*elem
;
1547 struct bpf_func_state
*frame
;
1549 elem
= kzalloc(sizeof(struct bpf_verifier_stack_elem
), GFP_KERNEL
);
1553 elem
->insn_idx
= insn_idx
;
1554 elem
->prev_insn_idx
= prev_insn_idx
;
1555 elem
->next
= env
->head
;
1556 elem
->log_pos
= env
->log
.len_used
;
1559 if (env
->stack_size
> BPF_COMPLEXITY_LIMIT_JMP_SEQ
) {
1561 "The sequence of %d jumps is too complex for async cb.\n",
1565 /* Unlike push_stack() do not copy_verifier_state().
1566 * The caller state doesn't matter.
1567 * This is async callback. It starts in a fresh stack.
1568 * Initialize it similar to do_check_common().
1570 elem
->st
.branches
= 1;
1571 frame
= kzalloc(sizeof(*frame
), GFP_KERNEL
);
1574 init_func_state(env
, frame
,
1575 BPF_MAIN_FUNC
/* callsite */,
1576 0 /* frameno within this callchain */,
1577 subprog
/* subprog number within this prog */);
1578 elem
->st
.frame
[0] = frame
;
1581 free_verifier_state(env
->cur_state
, true);
1582 env
->cur_state
= NULL
;
1583 /* pop all elements and return */
1584 while (!pop_stack(env
, NULL
, NULL
, false));
1590 SRC_OP
, /* register is used as source operand */
1591 DST_OP
, /* register is used as destination operand */
1592 DST_OP_NO_MARK
/* same as above, check only, don't mark */
1595 static int cmp_subprogs(const void *a
, const void *b
)
1597 return ((struct bpf_subprog_info
*)a
)->start
-
1598 ((struct bpf_subprog_info
*)b
)->start
;
1601 static int find_subprog(struct bpf_verifier_env
*env
, int off
)
1603 struct bpf_subprog_info
*p
;
1605 p
= bsearch(&off
, env
->subprog_info
, env
->subprog_cnt
,
1606 sizeof(env
->subprog_info
[0]), cmp_subprogs
);
1609 return p
- env
->subprog_info
;
1613 static int add_subprog(struct bpf_verifier_env
*env
, int off
)
1615 int insn_cnt
= env
->prog
->len
;
1618 if (off
>= insn_cnt
|| off
< 0) {
1619 verbose(env
, "call to invalid destination\n");
1622 ret
= find_subprog(env
, off
);
1625 if (env
->subprog_cnt
>= BPF_MAX_SUBPROGS
) {
1626 verbose(env
, "too many subprograms\n");
1629 /* determine subprog starts. The end is one before the next starts */
1630 env
->subprog_info
[env
->subprog_cnt
++].start
= off
;
1631 sort(env
->subprog_info
, env
->subprog_cnt
,
1632 sizeof(env
->subprog_info
[0]), cmp_subprogs
, NULL
);
1633 return env
->subprog_cnt
- 1;
1636 struct bpf_kfunc_desc
{
1637 struct btf_func_model func_model
;
1642 #define MAX_KFUNC_DESCS 256
1643 struct bpf_kfunc_desc_tab
{
1644 struct bpf_kfunc_desc descs
[MAX_KFUNC_DESCS
];
1648 static int kfunc_desc_cmp_by_id(const void *a
, const void *b
)
1650 const struct bpf_kfunc_desc
*d0
= a
;
1651 const struct bpf_kfunc_desc
*d1
= b
;
1653 /* func_id is not greater than BTF_MAX_TYPE */
1654 return d0
->func_id
- d1
->func_id
;
1657 static const struct bpf_kfunc_desc
*
1658 find_kfunc_desc(const struct bpf_prog
*prog
, u32 func_id
)
1660 struct bpf_kfunc_desc desc
= {
1663 struct bpf_kfunc_desc_tab
*tab
;
1665 tab
= prog
->aux
->kfunc_tab
;
1666 return bsearch(&desc
, tab
->descs
, tab
->nr_descs
,
1667 sizeof(tab
->descs
[0]), kfunc_desc_cmp_by_id
);
1670 static int add_kfunc_call(struct bpf_verifier_env
*env
, u32 func_id
)
1672 const struct btf_type
*func
, *func_proto
;
1673 struct bpf_kfunc_desc_tab
*tab
;
1674 struct bpf_prog_aux
*prog_aux
;
1675 struct bpf_kfunc_desc
*desc
;
1676 const char *func_name
;
1680 prog_aux
= env
->prog
->aux
;
1681 tab
= prog_aux
->kfunc_tab
;
1684 verbose(env
, "calling kernel function is not supported without CONFIG_DEBUG_INFO_BTF\n");
1688 if (!env
->prog
->jit_requested
) {
1689 verbose(env
, "JIT is required for calling kernel function\n");
1693 if (!bpf_jit_supports_kfunc_call()) {
1694 verbose(env
, "JIT does not support calling kernel function\n");
1698 if (!env
->prog
->gpl_compatible
) {
1699 verbose(env
, "cannot call kernel function from non-GPL compatible program\n");
1703 tab
= kzalloc(sizeof(*tab
), GFP_KERNEL
);
1706 prog_aux
->kfunc_tab
= tab
;
1709 if (find_kfunc_desc(env
->prog
, func_id
))
1712 if (tab
->nr_descs
== MAX_KFUNC_DESCS
) {
1713 verbose(env
, "too many different kernel function calls\n");
1717 func
= btf_type_by_id(btf_vmlinux
, func_id
);
1718 if (!func
|| !btf_type_is_func(func
)) {
1719 verbose(env
, "kernel btf_id %u is not a function\n",
1723 func_proto
= btf_type_by_id(btf_vmlinux
, func
->type
);
1724 if (!func_proto
|| !btf_type_is_func_proto(func_proto
)) {
1725 verbose(env
, "kernel function btf_id %u does not have a valid func_proto\n",
1730 func_name
= btf_name_by_offset(btf_vmlinux
, func
->name_off
);
1731 addr
= kallsyms_lookup_name(func_name
);
1733 verbose(env
, "cannot find address for kernel function %s\n",
1738 desc
= &tab
->descs
[tab
->nr_descs
++];
1739 desc
->func_id
= func_id
;
1740 desc
->imm
= BPF_CAST_CALL(addr
) - __bpf_call_base
;
1741 err
= btf_distill_func_proto(&env
->log
, btf_vmlinux
,
1742 func_proto
, func_name
,
1745 sort(tab
->descs
, tab
->nr_descs
, sizeof(tab
->descs
[0]),
1746 kfunc_desc_cmp_by_id
, NULL
);
1750 static int kfunc_desc_cmp_by_imm(const void *a
, const void *b
)
1752 const struct bpf_kfunc_desc
*d0
= a
;
1753 const struct bpf_kfunc_desc
*d1
= b
;
1755 if (d0
->imm
> d1
->imm
)
1757 else if (d0
->imm
< d1
->imm
)
1762 static void sort_kfunc_descs_by_imm(struct bpf_prog
*prog
)
1764 struct bpf_kfunc_desc_tab
*tab
;
1766 tab
= prog
->aux
->kfunc_tab
;
1770 sort(tab
->descs
, tab
->nr_descs
, sizeof(tab
->descs
[0]),
1771 kfunc_desc_cmp_by_imm
, NULL
);
1774 bool bpf_prog_has_kfunc_call(const struct bpf_prog
*prog
)
1776 return !!prog
->aux
->kfunc_tab
;
1779 const struct btf_func_model
*
1780 bpf_jit_find_kfunc_model(const struct bpf_prog
*prog
,
1781 const struct bpf_insn
*insn
)
1783 const struct bpf_kfunc_desc desc
= {
1786 const struct bpf_kfunc_desc
*res
;
1787 struct bpf_kfunc_desc_tab
*tab
;
1789 tab
= prog
->aux
->kfunc_tab
;
1790 res
= bsearch(&desc
, tab
->descs
, tab
->nr_descs
,
1791 sizeof(tab
->descs
[0]), kfunc_desc_cmp_by_imm
);
1793 return res
? &res
->func_model
: NULL
;
1796 static int add_subprog_and_kfunc(struct bpf_verifier_env
*env
)
1798 struct bpf_subprog_info
*subprog
= env
->subprog_info
;
1799 struct bpf_insn
*insn
= env
->prog
->insnsi
;
1800 int i
, ret
, insn_cnt
= env
->prog
->len
;
1802 /* Add entry function. */
1803 ret
= add_subprog(env
, 0);
1807 for (i
= 0; i
< insn_cnt
; i
++, insn
++) {
1808 if (!bpf_pseudo_func(insn
) && !bpf_pseudo_call(insn
) &&
1809 !bpf_pseudo_kfunc_call(insn
))
1812 if (!env
->bpf_capable
) {
1813 verbose(env
, "loading/calling other bpf or kernel functions are allowed for CAP_BPF and CAP_SYS_ADMIN\n");
1817 if (bpf_pseudo_func(insn
)) {
1818 ret
= add_subprog(env
, i
+ insn
->imm
+ 1);
1820 /* remember subprog */
1822 } else if (bpf_pseudo_call(insn
)) {
1823 ret
= add_subprog(env
, i
+ insn
->imm
+ 1);
1825 ret
= add_kfunc_call(env
, insn
->imm
);
1832 /* Add a fake 'exit' subprog which could simplify subprog iteration
1833 * logic. 'subprog_cnt' should not be increased.
1835 subprog
[env
->subprog_cnt
].start
= insn_cnt
;
1837 if (env
->log
.level
& BPF_LOG_LEVEL2
)
1838 for (i
= 0; i
< env
->subprog_cnt
; i
++)
1839 verbose(env
, "func#%d @%d\n", i
, subprog
[i
].start
);
1844 static int check_subprogs(struct bpf_verifier_env
*env
)
1846 int i
, subprog_start
, subprog_end
, off
, cur_subprog
= 0;
1847 struct bpf_subprog_info
*subprog
= env
->subprog_info
;
1848 struct bpf_insn
*insn
= env
->prog
->insnsi
;
1849 int insn_cnt
= env
->prog
->len
;
1851 /* now check that all jumps are within the same subprog */
1852 subprog_start
= subprog
[cur_subprog
].start
;
1853 subprog_end
= subprog
[cur_subprog
+ 1].start
;
1854 for (i
= 0; i
< insn_cnt
; i
++) {
1855 u8 code
= insn
[i
].code
;
1857 if (code
== (BPF_JMP
| BPF_CALL
) &&
1858 insn
[i
].imm
== BPF_FUNC_tail_call
&&
1859 insn
[i
].src_reg
!= BPF_PSEUDO_CALL
)
1860 subprog
[cur_subprog
].has_tail_call
= true;
1861 if (BPF_CLASS(code
) == BPF_LD
&&
1862 (BPF_MODE(code
) == BPF_ABS
|| BPF_MODE(code
) == BPF_IND
))
1863 subprog
[cur_subprog
].has_ld_abs
= true;
1864 if (BPF_CLASS(code
) != BPF_JMP
&& BPF_CLASS(code
) != BPF_JMP32
)
1866 if (BPF_OP(code
) == BPF_EXIT
|| BPF_OP(code
) == BPF_CALL
)
1868 off
= i
+ insn
[i
].off
+ 1;
1869 if (off
< subprog_start
|| off
>= subprog_end
) {
1870 verbose(env
, "jump out of range from insn %d to %d\n", i
, off
);
1874 if (i
== subprog_end
- 1) {
1875 /* to avoid fall-through from one subprog into another
1876 * the last insn of the subprog should be either exit
1877 * or unconditional jump back
1879 if (code
!= (BPF_JMP
| BPF_EXIT
) &&
1880 code
!= (BPF_JMP
| BPF_JA
)) {
1881 verbose(env
, "last insn is not an exit or jmp\n");
1884 subprog_start
= subprog_end
;
1886 if (cur_subprog
< env
->subprog_cnt
)
1887 subprog_end
= subprog
[cur_subprog
+ 1].start
;
1893 /* Parentage chain of this register (or stack slot) should take care of all
1894 * issues like callee-saved registers, stack slot allocation time, etc.
1896 static int mark_reg_read(struct bpf_verifier_env
*env
,
1897 const struct bpf_reg_state
*state
,
1898 struct bpf_reg_state
*parent
, u8 flag
)
1900 bool writes
= parent
== state
->parent
; /* Observe write marks */
1904 /* if read wasn't screened by an earlier write ... */
1905 if (writes
&& state
->live
& REG_LIVE_WRITTEN
)
1907 if (parent
->live
& REG_LIVE_DONE
) {
1908 verbose(env
, "verifier BUG type %s var_off %lld off %d\n",
1909 reg_type_str
[parent
->type
],
1910 parent
->var_off
.value
, parent
->off
);
1913 /* The first condition is more likely to be true than the
1914 * second, checked it first.
1916 if ((parent
->live
& REG_LIVE_READ
) == flag
||
1917 parent
->live
& REG_LIVE_READ64
)
1918 /* The parentage chain never changes and
1919 * this parent was already marked as LIVE_READ.
1920 * There is no need to keep walking the chain again and
1921 * keep re-marking all parents as LIVE_READ.
1922 * This case happens when the same register is read
1923 * multiple times without writes into it in-between.
1924 * Also, if parent has the stronger REG_LIVE_READ64 set,
1925 * then no need to set the weak REG_LIVE_READ32.
1928 /* ... then we depend on parent's value */
1929 parent
->live
|= flag
;
1930 /* REG_LIVE_READ64 overrides REG_LIVE_READ32. */
1931 if (flag
== REG_LIVE_READ64
)
1932 parent
->live
&= ~REG_LIVE_READ32
;
1934 parent
= state
->parent
;
1939 if (env
->longest_mark_read_walk
< cnt
)
1940 env
->longest_mark_read_walk
= cnt
;
1944 /* This function is supposed to be used by the following 32-bit optimization
1945 * code only. It returns TRUE if the source or destination register operates
1946 * on 64-bit, otherwise return FALSE.
1948 static bool is_reg64(struct bpf_verifier_env
*env
, struct bpf_insn
*insn
,
1949 u32 regno
, struct bpf_reg_state
*reg
, enum reg_arg_type t
)
1954 class = BPF_CLASS(code
);
1956 if (class == BPF_JMP
) {
1957 /* BPF_EXIT for "main" will reach here. Return TRUE
1962 if (op
== BPF_CALL
) {
1963 /* BPF to BPF call will reach here because of marking
1964 * caller saved clobber with DST_OP_NO_MARK for which we
1965 * don't care the register def because they are anyway
1966 * marked as NOT_INIT already.
1968 if (insn
->src_reg
== BPF_PSEUDO_CALL
)
1970 /* Helper call will reach here because of arg type
1971 * check, conservatively return TRUE.
1980 if (class == BPF_ALU64
|| class == BPF_JMP
||
1981 /* BPF_END always use BPF_ALU class. */
1982 (class == BPF_ALU
&& op
== BPF_END
&& insn
->imm
== 64))
1985 if (class == BPF_ALU
|| class == BPF_JMP32
)
1988 if (class == BPF_LDX
) {
1990 return BPF_SIZE(code
) == BPF_DW
;
1991 /* LDX source must be ptr. */
1995 if (class == BPF_STX
) {
1996 /* BPF_STX (including atomic variants) has multiple source
1997 * operands, one of which is a ptr. Check whether the caller is
2000 if (t
== SRC_OP
&& reg
->type
!= SCALAR_VALUE
)
2002 return BPF_SIZE(code
) == BPF_DW
;
2005 if (class == BPF_LD
) {
2006 u8 mode
= BPF_MODE(code
);
2009 if (mode
== BPF_IMM
)
2012 /* Both LD_IND and LD_ABS return 32-bit data. */
2016 /* Implicit ctx ptr. */
2017 if (regno
== BPF_REG_6
)
2020 /* Explicit source could be any width. */
2024 if (class == BPF_ST
)
2025 /* The only source register for BPF_ST is a ptr. */
2028 /* Conservatively return true at default. */
2032 /* Return the regno defined by the insn, or -1. */
2033 static int insn_def_regno(const struct bpf_insn
*insn
)
2035 switch (BPF_CLASS(insn
->code
)) {
2041 if (BPF_MODE(insn
->code
) == BPF_ATOMIC
&&
2042 (insn
->imm
& BPF_FETCH
)) {
2043 if (insn
->imm
== BPF_CMPXCHG
)
2046 return insn
->src_reg
;
2051 return insn
->dst_reg
;
2055 /* Return TRUE if INSN has defined any 32-bit value explicitly. */
2056 static bool insn_has_def32(struct bpf_verifier_env
*env
, struct bpf_insn
*insn
)
2058 int dst_reg
= insn_def_regno(insn
);
2063 return !is_reg64(env
, insn
, dst_reg
, NULL
, DST_OP
);
2066 static void mark_insn_zext(struct bpf_verifier_env
*env
,
2067 struct bpf_reg_state
*reg
)
2069 s32 def_idx
= reg
->subreg_def
;
2071 if (def_idx
== DEF_NOT_SUBREG
)
2074 env
->insn_aux_data
[def_idx
- 1].zext_dst
= true;
2075 /* The dst will be zero extended, so won't be sub-register anymore. */
2076 reg
->subreg_def
= DEF_NOT_SUBREG
;
2079 static int check_reg_arg(struct bpf_verifier_env
*env
, u32 regno
,
2080 enum reg_arg_type t
)
2082 struct bpf_verifier_state
*vstate
= env
->cur_state
;
2083 struct bpf_func_state
*state
= vstate
->frame
[vstate
->curframe
];
2084 struct bpf_insn
*insn
= env
->prog
->insnsi
+ env
->insn_idx
;
2085 struct bpf_reg_state
*reg
, *regs
= state
->regs
;
2088 if (regno
>= MAX_BPF_REG
) {
2089 verbose(env
, "R%d is invalid\n", regno
);
2094 rw64
= is_reg64(env
, insn
, regno
, reg
, t
);
2096 /* check whether register used as source operand can be read */
2097 if (reg
->type
== NOT_INIT
) {
2098 verbose(env
, "R%d !read_ok\n", regno
);
2101 /* We don't need to worry about FP liveness because it's read-only */
2102 if (regno
== BPF_REG_FP
)
2106 mark_insn_zext(env
, reg
);
2108 return mark_reg_read(env
, reg
, reg
->parent
,
2109 rw64
? REG_LIVE_READ64
: REG_LIVE_READ32
);
2111 /* check whether register used as dest operand can be written to */
2112 if (regno
== BPF_REG_FP
) {
2113 verbose(env
, "frame pointer is read only\n");
2116 reg
->live
|= REG_LIVE_WRITTEN
;
2117 reg
->subreg_def
= rw64
? DEF_NOT_SUBREG
: env
->insn_idx
+ 1;
2119 mark_reg_unknown(env
, regs
, regno
);
2124 /* for any branch, call, exit record the history of jmps in the given state */
2125 static int push_jmp_history(struct bpf_verifier_env
*env
,
2126 struct bpf_verifier_state
*cur
)
2128 u32 cnt
= cur
->jmp_history_cnt
;
2129 struct bpf_idx_pair
*p
;
2132 p
= krealloc(cur
->jmp_history
, cnt
* sizeof(*p
), GFP_USER
);
2135 p
[cnt
- 1].idx
= env
->insn_idx
;
2136 p
[cnt
- 1].prev_idx
= env
->prev_insn_idx
;
2137 cur
->jmp_history
= p
;
2138 cur
->jmp_history_cnt
= cnt
;
2142 /* Backtrack one insn at a time. If idx is not at the top of recorded
2143 * history then previous instruction came from straight line execution.
2145 static int get_prev_insn_idx(struct bpf_verifier_state
*st
, int i
,
2150 if (cnt
&& st
->jmp_history
[cnt
- 1].idx
== i
) {
2151 i
= st
->jmp_history
[cnt
- 1].prev_idx
;
2159 static const char *disasm_kfunc_name(void *data
, const struct bpf_insn
*insn
)
2161 const struct btf_type
*func
;
2163 if (insn
->src_reg
!= BPF_PSEUDO_KFUNC_CALL
)
2166 func
= btf_type_by_id(btf_vmlinux
, insn
->imm
);
2167 return btf_name_by_offset(btf_vmlinux
, func
->name_off
);
2170 /* For given verifier state backtrack_insn() is called from the last insn to
2171 * the first insn. Its purpose is to compute a bitmask of registers and
2172 * stack slots that needs precision in the parent verifier state.
2174 static int backtrack_insn(struct bpf_verifier_env
*env
, int idx
,
2175 u32
*reg_mask
, u64
*stack_mask
)
2177 const struct bpf_insn_cbs cbs
= {
2178 .cb_call
= disasm_kfunc_name
,
2179 .cb_print
= verbose
,
2180 .private_data
= env
,
2182 struct bpf_insn
*insn
= env
->prog
->insnsi
+ idx
;
2183 u8
class = BPF_CLASS(insn
->code
);
2184 u8 opcode
= BPF_OP(insn
->code
);
2185 u8 mode
= BPF_MODE(insn
->code
);
2186 u32 dreg
= 1u << insn
->dst_reg
;
2187 u32 sreg
= 1u << insn
->src_reg
;
2190 if (insn
->code
== 0)
2192 if (env
->log
.level
& BPF_LOG_LEVEL
) {
2193 verbose(env
, "regs=%x stack=%llx before ", *reg_mask
, *stack_mask
);
2194 verbose(env
, "%d: ", idx
);
2195 print_bpf_insn(&cbs
, insn
, env
->allow_ptr_leaks
);
2198 if (class == BPF_ALU
|| class == BPF_ALU64
) {
2199 if (!(*reg_mask
& dreg
))
2201 if (opcode
== BPF_MOV
) {
2202 if (BPF_SRC(insn
->code
) == BPF_X
) {
2204 * dreg needs precision after this insn
2205 * sreg needs precision before this insn
2211 * dreg needs precision after this insn.
2212 * Corresponding register is already marked
2213 * as precise=true in this verifier state.
2214 * No further markings in parent are necessary
2219 if (BPF_SRC(insn
->code
) == BPF_X
) {
2221 * both dreg and sreg need precision
2226 * dreg still needs precision before this insn
2229 } else if (class == BPF_LDX
) {
2230 if (!(*reg_mask
& dreg
))
2234 /* scalars can only be spilled into stack w/o losing precision.
2235 * Load from any other memory can be zero extended.
2236 * The desire to keep that precision is already indicated
2237 * by 'precise' mark in corresponding register of this state.
2238 * No further tracking necessary.
2240 if (insn
->src_reg
!= BPF_REG_FP
)
2242 if (BPF_SIZE(insn
->code
) != BPF_DW
)
2245 /* dreg = *(u64 *)[fp - off] was a fill from the stack.
2246 * that [fp - off] slot contains scalar that needs to be
2247 * tracked with precision
2249 spi
= (-insn
->off
- 1) / BPF_REG_SIZE
;
2251 verbose(env
, "BUG spi %d\n", spi
);
2252 WARN_ONCE(1, "verifier backtracking bug");
2255 *stack_mask
|= 1ull << spi
;
2256 } else if (class == BPF_STX
|| class == BPF_ST
) {
2257 if (*reg_mask
& dreg
)
2258 /* stx & st shouldn't be using _scalar_ dst_reg
2259 * to access memory. It means backtracking
2260 * encountered a case of pointer subtraction.
2263 /* scalars can only be spilled into stack */
2264 if (insn
->dst_reg
!= BPF_REG_FP
)
2266 if (BPF_SIZE(insn
->code
) != BPF_DW
)
2268 spi
= (-insn
->off
- 1) / BPF_REG_SIZE
;
2270 verbose(env
, "BUG spi %d\n", spi
);
2271 WARN_ONCE(1, "verifier backtracking bug");
2274 if (!(*stack_mask
& (1ull << spi
)))
2276 *stack_mask
&= ~(1ull << spi
);
2277 if (class == BPF_STX
)
2279 } else if (class == BPF_JMP
|| class == BPF_JMP32
) {
2280 if (opcode
== BPF_CALL
) {
2281 if (insn
->src_reg
== BPF_PSEUDO_CALL
)
2283 /* regular helper call sets R0 */
2285 if (*reg_mask
& 0x3f) {
2286 /* if backtracing was looking for registers R1-R5
2287 * they should have been found already.
2289 verbose(env
, "BUG regs %x\n", *reg_mask
);
2290 WARN_ONCE(1, "verifier backtracking bug");
2293 } else if (opcode
== BPF_EXIT
) {
2296 } else if (class == BPF_LD
) {
2297 if (!(*reg_mask
& dreg
))
2300 /* It's ld_imm64 or ld_abs or ld_ind.
2301 * For ld_imm64 no further tracking of precision
2302 * into parent is necessary
2304 if (mode
== BPF_IND
|| mode
== BPF_ABS
)
2305 /* to be analyzed */
2311 /* the scalar precision tracking algorithm:
2312 * . at the start all registers have precise=false.
2313 * . scalar ranges are tracked as normal through alu and jmp insns.
2314 * . once precise value of the scalar register is used in:
2315 * . ptr + scalar alu
2316 * . if (scalar cond K|scalar)
2317 * . helper_call(.., scalar, ...) where ARG_CONST is expected
2318 * backtrack through the verifier states and mark all registers and
2319 * stack slots with spilled constants that these scalar regisers
2320 * should be precise.
2321 * . during state pruning two registers (or spilled stack slots)
2322 * are equivalent if both are not precise.
2324 * Note the verifier cannot simply walk register parentage chain,
2325 * since many different registers and stack slots could have been
2326 * used to compute single precise scalar.
2328 * The approach of starting with precise=true for all registers and then
2329 * backtrack to mark a register as not precise when the verifier detects
2330 * that program doesn't care about specific value (e.g., when helper
2331 * takes register as ARG_ANYTHING parameter) is not safe.
2333 * It's ok to walk single parentage chain of the verifier states.
2334 * It's possible that this backtracking will go all the way till 1st insn.
2335 * All other branches will be explored for needing precision later.
2337 * The backtracking needs to deal with cases like:
2338 * 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)
2341 * if r5 > 0x79f goto pc+7
2342 * R5_w=inv(id=0,umax_value=1951,var_off=(0x0; 0x7ff))
2345 * call bpf_perf_event_output#25
2346 * where .arg5_type = ARG_CONST_SIZE_OR_ZERO
2350 * call foo // uses callee's r6 inside to compute r0
2354 * to track above reg_mask/stack_mask needs to be independent for each frame.
2356 * Also if parent's curframe > frame where backtracking started,
2357 * the verifier need to mark registers in both frames, otherwise callees
2358 * may incorrectly prune callers. This is similar to
2359 * commit 7640ead93924 ("bpf: verifier: make sure callees don't prune with caller differences")
2361 * For now backtracking falls back into conservative marking.
2363 static void mark_all_scalars_precise(struct bpf_verifier_env
*env
,
2364 struct bpf_verifier_state
*st
)
2366 struct bpf_func_state
*func
;
2367 struct bpf_reg_state
*reg
;
2370 /* big hammer: mark all scalars precise in this path.
2371 * pop_stack may still get !precise scalars.
2373 for (; st
; st
= st
->parent
)
2374 for (i
= 0; i
<= st
->curframe
; i
++) {
2375 func
= st
->frame
[i
];
2376 for (j
= 0; j
< BPF_REG_FP
; j
++) {
2377 reg
= &func
->regs
[j
];
2378 if (reg
->type
!= SCALAR_VALUE
)
2380 reg
->precise
= true;
2382 for (j
= 0; j
< func
->allocated_stack
/ BPF_REG_SIZE
; j
++) {
2383 if (func
->stack
[j
].slot_type
[0] != STACK_SPILL
)
2385 reg
= &func
->stack
[j
].spilled_ptr
;
2386 if (reg
->type
!= SCALAR_VALUE
)
2388 reg
->precise
= true;
2393 static int __mark_chain_precision(struct bpf_verifier_env
*env
, int regno
,
2396 struct bpf_verifier_state
*st
= env
->cur_state
;
2397 int first_idx
= st
->first_insn_idx
;
2398 int last_idx
= env
->insn_idx
;
2399 struct bpf_func_state
*func
;
2400 struct bpf_reg_state
*reg
;
2401 u32 reg_mask
= regno
>= 0 ? 1u << regno
: 0;
2402 u64 stack_mask
= spi
>= 0 ? 1ull << spi
: 0;
2403 bool skip_first
= true;
2404 bool new_marks
= false;
2407 if (!env
->bpf_capable
)
2410 func
= st
->frame
[st
->curframe
];
2412 reg
= &func
->regs
[regno
];
2413 if (reg
->type
!= SCALAR_VALUE
) {
2414 WARN_ONCE(1, "backtracing misuse");
2421 reg
->precise
= true;
2425 if (func
->stack
[spi
].slot_type
[0] != STACK_SPILL
) {
2429 reg
= &func
->stack
[spi
].spilled_ptr
;
2430 if (reg
->type
!= SCALAR_VALUE
) {
2438 reg
->precise
= true;
2444 if (!reg_mask
&& !stack_mask
)
2447 DECLARE_BITMAP(mask
, 64);
2448 u32 history
= st
->jmp_history_cnt
;
2450 if (env
->log
.level
& BPF_LOG_LEVEL
)
2451 verbose(env
, "last_idx %d first_idx %d\n", last_idx
, first_idx
);
2452 for (i
= last_idx
;;) {
2457 err
= backtrack_insn(env
, i
, ®_mask
, &stack_mask
);
2459 if (err
== -ENOTSUPP
) {
2460 mark_all_scalars_precise(env
, st
);
2465 if (!reg_mask
&& !stack_mask
)
2466 /* Found assignment(s) into tracked register in this state.
2467 * Since this state is already marked, just return.
2468 * Nothing to be tracked further in the parent state.
2473 i
= get_prev_insn_idx(st
, i
, &history
);
2474 if (i
>= env
->prog
->len
) {
2475 /* This can happen if backtracking reached insn 0
2476 * and there are still reg_mask or stack_mask
2478 * It means the backtracking missed the spot where
2479 * particular register was initialized with a constant.
2481 verbose(env
, "BUG backtracking idx %d\n", i
);
2482 WARN_ONCE(1, "verifier backtracking bug");
2491 func
= st
->frame
[st
->curframe
];
2492 bitmap_from_u64(mask
, reg_mask
);
2493 for_each_set_bit(i
, mask
, 32) {
2494 reg
= &func
->regs
[i
];
2495 if (reg
->type
!= SCALAR_VALUE
) {
2496 reg_mask
&= ~(1u << i
);
2501 reg
->precise
= true;
2504 bitmap_from_u64(mask
, stack_mask
);
2505 for_each_set_bit(i
, mask
, 64) {
2506 if (i
>= func
->allocated_stack
/ BPF_REG_SIZE
) {
2507 /* the sequence of instructions:
2509 * 3: (7b) *(u64 *)(r3 -8) = r0
2510 * 4: (79) r4 = *(u64 *)(r10 -8)
2511 * doesn't contain jmps. It's backtracked
2512 * as a single block.
2513 * During backtracking insn 3 is not recognized as
2514 * stack access, so at the end of backtracking
2515 * stack slot fp-8 is still marked in stack_mask.
2516 * However the parent state may not have accessed
2517 * fp-8 and it's "unallocated" stack space.
2518 * In such case fallback to conservative.
2520 mark_all_scalars_precise(env
, st
);
2524 if (func
->stack
[i
].slot_type
[0] != STACK_SPILL
) {
2525 stack_mask
&= ~(1ull << i
);
2528 reg
= &func
->stack
[i
].spilled_ptr
;
2529 if (reg
->type
!= SCALAR_VALUE
) {
2530 stack_mask
&= ~(1ull << i
);
2535 reg
->precise
= true;
2537 if (env
->log
.level
& BPF_LOG_LEVEL
) {
2538 print_verifier_state(env
, func
);
2539 verbose(env
, "parent %s regs=%x stack=%llx marks\n",
2540 new_marks
? "didn't have" : "already had",
2541 reg_mask
, stack_mask
);
2544 if (!reg_mask
&& !stack_mask
)
2549 last_idx
= st
->last_insn_idx
;
2550 first_idx
= st
->first_insn_idx
;
2555 static int mark_chain_precision(struct bpf_verifier_env
*env
, int regno
)
2557 return __mark_chain_precision(env
, regno
, -1);
2560 static int mark_chain_precision_stack(struct bpf_verifier_env
*env
, int spi
)
2562 return __mark_chain_precision(env
, -1, spi
);
2565 static bool is_spillable_regtype(enum bpf_reg_type type
)
2568 case PTR_TO_MAP_VALUE
:
2569 case PTR_TO_MAP_VALUE_OR_NULL
:
2573 case PTR_TO_PACKET_META
:
2574 case PTR_TO_PACKET_END
:
2575 case PTR_TO_FLOW_KEYS
:
2576 case CONST_PTR_TO_MAP
:
2578 case PTR_TO_SOCKET_OR_NULL
:
2579 case PTR_TO_SOCK_COMMON
:
2580 case PTR_TO_SOCK_COMMON_OR_NULL
:
2581 case PTR_TO_TCP_SOCK
:
2582 case PTR_TO_TCP_SOCK_OR_NULL
:
2583 case PTR_TO_XDP_SOCK
:
2585 case PTR_TO_BTF_ID_OR_NULL
:
2586 case PTR_TO_RDONLY_BUF
:
2587 case PTR_TO_RDONLY_BUF_OR_NULL
:
2588 case PTR_TO_RDWR_BUF
:
2589 case PTR_TO_RDWR_BUF_OR_NULL
:
2590 case PTR_TO_PERCPU_BTF_ID
:
2592 case PTR_TO_MEM_OR_NULL
:
2594 case PTR_TO_MAP_KEY
:
2601 /* Does this register contain a constant zero? */
2602 static bool register_is_null(struct bpf_reg_state
*reg
)
2604 return reg
->type
== SCALAR_VALUE
&& tnum_equals_const(reg
->var_off
, 0);
2607 static bool register_is_const(struct bpf_reg_state
*reg
)
2609 return reg
->type
== SCALAR_VALUE
&& tnum_is_const(reg
->var_off
);
2612 static bool __is_scalar_unbounded(struct bpf_reg_state
*reg
)
2614 return tnum_is_unknown(reg
->var_off
) &&
2615 reg
->smin_value
== S64_MIN
&& reg
->smax_value
== S64_MAX
&&
2616 reg
->umin_value
== 0 && reg
->umax_value
== U64_MAX
&&
2617 reg
->s32_min_value
== S32_MIN
&& reg
->s32_max_value
== S32_MAX
&&
2618 reg
->u32_min_value
== 0 && reg
->u32_max_value
== U32_MAX
;
2621 static bool register_is_bounded(struct bpf_reg_state
*reg
)
2623 return reg
->type
== SCALAR_VALUE
&& !__is_scalar_unbounded(reg
);
2626 static bool __is_pointer_value(bool allow_ptr_leaks
,
2627 const struct bpf_reg_state
*reg
)
2629 if (allow_ptr_leaks
)
2632 return reg
->type
!= SCALAR_VALUE
;
2635 static void save_register_state(struct bpf_func_state
*state
,
2636 int spi
, struct bpf_reg_state
*reg
)
2640 state
->stack
[spi
].spilled_ptr
= *reg
;
2641 state
->stack
[spi
].spilled_ptr
.live
|= REG_LIVE_WRITTEN
;
2643 for (i
= 0; i
< BPF_REG_SIZE
; i
++)
2644 state
->stack
[spi
].slot_type
[i
] = STACK_SPILL
;
2647 /* check_stack_{read,write}_fixed_off functions track spill/fill of registers,
2648 * stack boundary and alignment are checked in check_mem_access()
2650 static int check_stack_write_fixed_off(struct bpf_verifier_env
*env
,
2651 /* stack frame we're writing to */
2652 struct bpf_func_state
*state
,
2653 int off
, int size
, int value_regno
,
2656 struct bpf_func_state
*cur
; /* state of the current function */
2657 int i
, slot
= -off
- 1, spi
= slot
/ BPF_REG_SIZE
, err
;
2658 u32 dst_reg
= env
->prog
->insnsi
[insn_idx
].dst_reg
;
2659 struct bpf_reg_state
*reg
= NULL
;
2661 err
= grow_stack_state(state
, round_up(slot
+ 1, BPF_REG_SIZE
));
2664 /* caller checked that off % size == 0 and -MAX_BPF_STACK <= off < 0,
2665 * so it's aligned access and [off, off + size) are within stack limits
2667 if (!env
->allow_ptr_leaks
&&
2668 state
->stack
[spi
].slot_type
[0] == STACK_SPILL
&&
2669 size
!= BPF_REG_SIZE
) {
2670 verbose(env
, "attempt to corrupt spilled pointer on stack\n");
2674 cur
= env
->cur_state
->frame
[env
->cur_state
->curframe
];
2675 if (value_regno
>= 0)
2676 reg
= &cur
->regs
[value_regno
];
2677 if (!env
->bypass_spec_v4
) {
2678 bool sanitize
= reg
&& is_spillable_regtype(reg
->type
);
2680 for (i
= 0; i
< size
; i
++) {
2681 if (state
->stack
[spi
].slot_type
[i
] == STACK_INVALID
) {
2688 env
->insn_aux_data
[insn_idx
].sanitize_stack_spill
= true;
2691 if (reg
&& size
== BPF_REG_SIZE
&& register_is_bounded(reg
) &&
2692 !register_is_null(reg
) && env
->bpf_capable
) {
2693 if (dst_reg
!= BPF_REG_FP
) {
2694 /* The backtracking logic can only recognize explicit
2695 * stack slot address like [fp - 8]. Other spill of
2696 * scalar via different register has to be conservative.
2697 * Backtrack from here and mark all registers as precise
2698 * that contributed into 'reg' being a constant.
2700 err
= mark_chain_precision(env
, value_regno
);
2704 save_register_state(state
, spi
, reg
);
2705 } else if (reg
&& is_spillable_regtype(reg
->type
)) {
2706 /* register containing pointer is being spilled into stack */
2707 if (size
!= BPF_REG_SIZE
) {
2708 verbose_linfo(env
, insn_idx
, "; ");
2709 verbose(env
, "invalid size of register spill\n");
2712 if (state
!= cur
&& reg
->type
== PTR_TO_STACK
) {
2713 verbose(env
, "cannot spill pointers to stack into stack frame of the caller\n");
2716 save_register_state(state
, spi
, reg
);
2718 u8 type
= STACK_MISC
;
2720 /* regular write of data into stack destroys any spilled ptr */
2721 state
->stack
[spi
].spilled_ptr
.type
= NOT_INIT
;
2722 /* Mark slots as STACK_MISC if they belonged to spilled ptr. */
2723 if (state
->stack
[spi
].slot_type
[0] == STACK_SPILL
)
2724 for (i
= 0; i
< BPF_REG_SIZE
; i
++)
2725 state
->stack
[spi
].slot_type
[i
] = STACK_MISC
;
2727 /* only mark the slot as written if all 8 bytes were written
2728 * otherwise read propagation may incorrectly stop too soon
2729 * when stack slots are partially written.
2730 * This heuristic means that read propagation will be
2731 * conservative, since it will add reg_live_read marks
2732 * to stack slots all the way to first state when programs
2733 * writes+reads less than 8 bytes
2735 if (size
== BPF_REG_SIZE
)
2736 state
->stack
[spi
].spilled_ptr
.live
|= REG_LIVE_WRITTEN
;
2738 /* when we zero initialize stack slots mark them as such */
2739 if (reg
&& register_is_null(reg
)) {
2740 /* backtracking doesn't work for STACK_ZERO yet. */
2741 err
= mark_chain_precision(env
, value_regno
);
2747 /* Mark slots affected by this stack write. */
2748 for (i
= 0; i
< size
; i
++)
2749 state
->stack
[spi
].slot_type
[(slot
- i
) % BPF_REG_SIZE
] =
2755 /* Write the stack: 'stack[ptr_regno + off] = value_regno'. 'ptr_regno' is
2756 * known to contain a variable offset.
2757 * This function checks whether the write is permitted and conservatively
2758 * tracks the effects of the write, considering that each stack slot in the
2759 * dynamic range is potentially written to.
2761 * 'off' includes 'regno->off'.
2762 * 'value_regno' can be -1, meaning that an unknown value is being written to
2765 * Spilled pointers in range are not marked as written because we don't know
2766 * what's going to be actually written. This means that read propagation for
2767 * future reads cannot be terminated by this write.
2769 * For privileged programs, uninitialized stack slots are considered
2770 * initialized by this write (even though we don't know exactly what offsets
2771 * are going to be written to). The idea is that we don't want the verifier to
2772 * reject future reads that access slots written to through variable offsets.
2774 static int check_stack_write_var_off(struct bpf_verifier_env
*env
,
2775 /* func where register points to */
2776 struct bpf_func_state
*state
,
2777 int ptr_regno
, int off
, int size
,
2778 int value_regno
, int insn_idx
)
2780 struct bpf_func_state
*cur
; /* state of the current function */
2781 int min_off
, max_off
;
2783 struct bpf_reg_state
*ptr_reg
= NULL
, *value_reg
= NULL
;
2784 bool writing_zero
= false;
2785 /* set if the fact that we're writing a zero is used to let any
2786 * stack slots remain STACK_ZERO
2788 bool zero_used
= false;
2790 cur
= env
->cur_state
->frame
[env
->cur_state
->curframe
];
2791 ptr_reg
= &cur
->regs
[ptr_regno
];
2792 min_off
= ptr_reg
->smin_value
+ off
;
2793 max_off
= ptr_reg
->smax_value
+ off
+ size
;
2794 if (value_regno
>= 0)
2795 value_reg
= &cur
->regs
[value_regno
];
2796 if (value_reg
&& register_is_null(value_reg
))
2797 writing_zero
= true;
2799 err
= grow_stack_state(state
, round_up(-min_off
, BPF_REG_SIZE
));
2804 /* Variable offset writes destroy any spilled pointers in range. */
2805 for (i
= min_off
; i
< max_off
; i
++) {
2806 u8 new_type
, *stype
;
2810 spi
= slot
/ BPF_REG_SIZE
;
2811 stype
= &state
->stack
[spi
].slot_type
[slot
% BPF_REG_SIZE
];
2813 if (!env
->allow_ptr_leaks
2814 && *stype
!= NOT_INIT
2815 && *stype
!= SCALAR_VALUE
) {
2816 /* Reject the write if there's are spilled pointers in
2817 * range. If we didn't reject here, the ptr status
2818 * would be erased below (even though not all slots are
2819 * actually overwritten), possibly opening the door to
2822 verbose(env
, "spilled ptr in range of var-offset stack write; insn %d, ptr off: %d",
2827 /* Erase all spilled pointers. */
2828 state
->stack
[spi
].spilled_ptr
.type
= NOT_INIT
;
2830 /* Update the slot type. */
2831 new_type
= STACK_MISC
;
2832 if (writing_zero
&& *stype
== STACK_ZERO
) {
2833 new_type
= STACK_ZERO
;
2836 /* If the slot is STACK_INVALID, we check whether it's OK to
2837 * pretend that it will be initialized by this write. The slot
2838 * might not actually be written to, and so if we mark it as
2839 * initialized future reads might leak uninitialized memory.
2840 * For privileged programs, we will accept such reads to slots
2841 * that may or may not be written because, if we're reject
2842 * them, the error would be too confusing.
2844 if (*stype
== STACK_INVALID
&& !env
->allow_uninit_stack
) {
2845 verbose(env
, "uninit stack in range of var-offset write prohibited for !root; insn %d, off: %d",
2852 /* backtracking doesn't work for STACK_ZERO yet. */
2853 err
= mark_chain_precision(env
, value_regno
);
2860 /* When register 'dst_regno' is assigned some values from stack[min_off,
2861 * max_off), we set the register's type according to the types of the
2862 * respective stack slots. If all the stack values are known to be zeros, then
2863 * so is the destination reg. Otherwise, the register is considered to be
2864 * SCALAR. This function does not deal with register filling; the caller must
2865 * ensure that all spilled registers in the stack range have been marked as
2868 static void mark_reg_stack_read(struct bpf_verifier_env
*env
,
2869 /* func where src register points to */
2870 struct bpf_func_state
*ptr_state
,
2871 int min_off
, int max_off
, int dst_regno
)
2873 struct bpf_verifier_state
*vstate
= env
->cur_state
;
2874 struct bpf_func_state
*state
= vstate
->frame
[vstate
->curframe
];
2879 for (i
= min_off
; i
< max_off
; i
++) {
2881 spi
= slot
/ BPF_REG_SIZE
;
2882 stype
= ptr_state
->stack
[spi
].slot_type
;
2883 if (stype
[slot
% BPF_REG_SIZE
] != STACK_ZERO
)
2887 if (zeros
== max_off
- min_off
) {
2888 /* any access_size read into register is zero extended,
2889 * so the whole register == const_zero
2891 __mark_reg_const_zero(&state
->regs
[dst_regno
]);
2892 /* backtracking doesn't support STACK_ZERO yet,
2893 * so mark it precise here, so that later
2894 * backtracking can stop here.
2895 * Backtracking may not need this if this register
2896 * doesn't participate in pointer adjustment.
2897 * Forward propagation of precise flag is not
2898 * necessary either. This mark is only to stop
2899 * backtracking. Any register that contributed
2900 * to const 0 was marked precise before spill.
2902 state
->regs
[dst_regno
].precise
= true;
2904 /* have read misc data from the stack */
2905 mark_reg_unknown(env
, state
->regs
, dst_regno
);
2907 state
->regs
[dst_regno
].live
|= REG_LIVE_WRITTEN
;
2910 /* Read the stack at 'off' and put the results into the register indicated by
2911 * 'dst_regno'. It handles reg filling if the addressed stack slot is a
2914 * 'dst_regno' can be -1, meaning that the read value is not going to a
2917 * The access is assumed to be within the current stack bounds.
2919 static int check_stack_read_fixed_off(struct bpf_verifier_env
*env
,
2920 /* func where src register points to */
2921 struct bpf_func_state
*reg_state
,
2922 int off
, int size
, int dst_regno
)
2924 struct bpf_verifier_state
*vstate
= env
->cur_state
;
2925 struct bpf_func_state
*state
= vstate
->frame
[vstate
->curframe
];
2926 int i
, slot
= -off
- 1, spi
= slot
/ BPF_REG_SIZE
;
2927 struct bpf_reg_state
*reg
;
2930 stype
= reg_state
->stack
[spi
].slot_type
;
2931 reg
= ®_state
->stack
[spi
].spilled_ptr
;
2933 if (stype
[0] == STACK_SPILL
) {
2934 if (size
!= BPF_REG_SIZE
) {
2935 if (reg
->type
!= SCALAR_VALUE
) {
2936 verbose_linfo(env
, env
->insn_idx
, "; ");
2937 verbose(env
, "invalid size of register fill\n");
2940 if (dst_regno
>= 0) {
2941 mark_reg_unknown(env
, state
->regs
, dst_regno
);
2942 state
->regs
[dst_regno
].live
|= REG_LIVE_WRITTEN
;
2944 mark_reg_read(env
, reg
, reg
->parent
, REG_LIVE_READ64
);
2947 for (i
= 1; i
< BPF_REG_SIZE
; i
++) {
2948 if (stype
[(slot
- i
) % BPF_REG_SIZE
] != STACK_SPILL
) {
2949 verbose(env
, "corrupted spill memory\n");
2954 if (dst_regno
>= 0) {
2955 /* restore register state from stack */
2956 state
->regs
[dst_regno
] = *reg
;
2957 /* mark reg as written since spilled pointer state likely
2958 * has its liveness marks cleared by is_state_visited()
2959 * which resets stack/reg liveness for state transitions
2961 state
->regs
[dst_regno
].live
|= REG_LIVE_WRITTEN
;
2962 } else if (__is_pointer_value(env
->allow_ptr_leaks
, reg
)) {
2963 /* If dst_regno==-1, the caller is asking us whether
2964 * it is acceptable to use this value as a SCALAR_VALUE
2966 * We must not allow unprivileged callers to do that
2967 * with spilled pointers.
2969 verbose(env
, "leaking pointer from stack off %d\n",
2973 mark_reg_read(env
, reg
, reg
->parent
, REG_LIVE_READ64
);
2977 for (i
= 0; i
< size
; i
++) {
2978 type
= stype
[(slot
- i
) % BPF_REG_SIZE
];
2979 if (type
== STACK_MISC
)
2981 if (type
== STACK_ZERO
)
2983 verbose(env
, "invalid read from stack off %d+%d size %d\n",
2987 mark_reg_read(env
, reg
, reg
->parent
, REG_LIVE_READ64
);
2989 mark_reg_stack_read(env
, reg_state
, off
, off
+ size
, dst_regno
);
2994 enum stack_access_src
{
2995 ACCESS_DIRECT
= 1, /* the access is performed by an instruction */
2996 ACCESS_HELPER
= 2, /* the access is performed by a helper */
2999 static int check_stack_range_initialized(struct bpf_verifier_env
*env
,
3000 int regno
, int off
, int access_size
,
3001 bool zero_size_allowed
,
3002 enum stack_access_src type
,
3003 struct bpf_call_arg_meta
*meta
);
3005 static struct bpf_reg_state
*reg_state(struct bpf_verifier_env
*env
, int regno
)
3007 return cur_regs(env
) + regno
;
3010 /* Read the stack at 'ptr_regno + off' and put the result into the register
3012 * 'off' includes the pointer register's fixed offset(i.e. 'ptr_regno.off'),
3013 * but not its variable offset.
3014 * 'size' is assumed to be <= reg size and the access is assumed to be aligned.
3016 * As opposed to check_stack_read_fixed_off, this function doesn't deal with
3017 * filling registers (i.e. reads of spilled register cannot be detected when
3018 * the offset is not fixed). We conservatively mark 'dst_regno' as containing
3019 * SCALAR_VALUE. That's why we assert that the 'ptr_regno' has a variable
3020 * offset; for a fixed offset check_stack_read_fixed_off should be used
3023 static int check_stack_read_var_off(struct bpf_verifier_env
*env
,
3024 int ptr_regno
, int off
, int size
, int dst_regno
)
3026 /* The state of the source register. */
3027 struct bpf_reg_state
*reg
= reg_state(env
, ptr_regno
);
3028 struct bpf_func_state
*ptr_state
= func(env
, reg
);
3030 int min_off
, max_off
;
3032 /* Note that we pass a NULL meta, so raw access will not be permitted.
3034 err
= check_stack_range_initialized(env
, ptr_regno
, off
, size
,
3035 false, ACCESS_DIRECT
, NULL
);
3039 min_off
= reg
->smin_value
+ off
;
3040 max_off
= reg
->smax_value
+ off
;
3041 mark_reg_stack_read(env
, ptr_state
, min_off
, max_off
+ size
, dst_regno
);
3045 /* check_stack_read dispatches to check_stack_read_fixed_off or
3046 * check_stack_read_var_off.
3048 * The caller must ensure that the offset falls within the allocated stack
3051 * 'dst_regno' is a register which will receive the value from the stack. It
3052 * can be -1, meaning that the read value is not going to a register.
3054 static int check_stack_read(struct bpf_verifier_env
*env
,
3055 int ptr_regno
, int off
, int size
,
3058 struct bpf_reg_state
*reg
= reg_state(env
, ptr_regno
);
3059 struct bpf_func_state
*state
= func(env
, reg
);
3061 /* Some accesses are only permitted with a static offset. */
3062 bool var_off
= !tnum_is_const(reg
->var_off
);
3064 /* The offset is required to be static when reads don't go to a
3065 * register, in order to not leak pointers (see
3066 * check_stack_read_fixed_off).
3068 if (dst_regno
< 0 && var_off
) {
3071 tnum_strn(tn_buf
, sizeof(tn_buf
), reg
->var_off
);
3072 verbose(env
, "variable offset stack pointer cannot be passed into helper function; var_off=%s off=%d size=%d\n",
3076 /* Variable offset is prohibited for unprivileged mode for simplicity
3077 * since it requires corresponding support in Spectre masking for stack
3078 * ALU. See also retrieve_ptr_limit().
3080 if (!env
->bypass_spec_v1
&& var_off
) {
3083 tnum_strn(tn_buf
, sizeof(tn_buf
), reg
->var_off
);
3084 verbose(env
, "R%d variable offset stack access prohibited for !root, var_off=%s\n",
3090 off
+= reg
->var_off
.value
;
3091 err
= check_stack_read_fixed_off(env
, state
, off
, size
,
3094 /* Variable offset stack reads need more conservative handling
3095 * than fixed offset ones. Note that dst_regno >= 0 on this
3098 err
= check_stack_read_var_off(env
, ptr_regno
, off
, size
,
3105 /* check_stack_write dispatches to check_stack_write_fixed_off or
3106 * check_stack_write_var_off.
3108 * 'ptr_regno' is the register used as a pointer into the stack.
3109 * 'off' includes 'ptr_regno->off', but not its variable offset (if any).
3110 * 'value_regno' is the register whose value we're writing to the stack. It can
3111 * be -1, meaning that we're not writing from a register.
3113 * The caller must ensure that the offset falls within the maximum stack size.
3115 static int check_stack_write(struct bpf_verifier_env
*env
,
3116 int ptr_regno
, int off
, int size
,
3117 int value_regno
, int insn_idx
)
3119 struct bpf_reg_state
*reg
= reg_state(env
, ptr_regno
);
3120 struct bpf_func_state
*state
= func(env
, reg
);
3123 if (tnum_is_const(reg
->var_off
)) {
3124 off
+= reg
->var_off
.value
;
3125 err
= check_stack_write_fixed_off(env
, state
, off
, size
,
3126 value_regno
, insn_idx
);
3128 /* Variable offset stack reads need more conservative handling
3129 * than fixed offset ones.
3131 err
= check_stack_write_var_off(env
, state
,
3132 ptr_regno
, off
, size
,
3133 value_regno
, insn_idx
);
3138 static int check_map_access_type(struct bpf_verifier_env
*env
, u32 regno
,
3139 int off
, int size
, enum bpf_access_type type
)
3141 struct bpf_reg_state
*regs
= cur_regs(env
);
3142 struct bpf_map
*map
= regs
[regno
].map_ptr
;
3143 u32 cap
= bpf_map_flags_to_cap(map
);
3145 if (type
== BPF_WRITE
&& !(cap
& BPF_MAP_CAN_WRITE
)) {
3146 verbose(env
, "write into map forbidden, value_size=%d off=%d size=%d\n",
3147 map
->value_size
, off
, size
);
3151 if (type
== BPF_READ
&& !(cap
& BPF_MAP_CAN_READ
)) {
3152 verbose(env
, "read from map forbidden, value_size=%d off=%d size=%d\n",
3153 map
->value_size
, off
, size
);
3160 /* check read/write into memory region (e.g., map value, ringbuf sample, etc) */
3161 static int __check_mem_access(struct bpf_verifier_env
*env
, int regno
,
3162 int off
, int size
, u32 mem_size
,
3163 bool zero_size_allowed
)
3165 bool size_ok
= size
> 0 || (size
== 0 && zero_size_allowed
);
3166 struct bpf_reg_state
*reg
;
3168 if (off
>= 0 && size_ok
&& (u64
)off
+ size
<= mem_size
)
3171 reg
= &cur_regs(env
)[regno
];
3172 switch (reg
->type
) {
3173 case PTR_TO_MAP_KEY
:
3174 verbose(env
, "invalid access to map key, key_size=%d off=%d size=%d\n",
3175 mem_size
, off
, size
);
3177 case PTR_TO_MAP_VALUE
:
3178 verbose(env
, "invalid access to map value, value_size=%d off=%d size=%d\n",
3179 mem_size
, off
, size
);
3182 case PTR_TO_PACKET_META
:
3183 case PTR_TO_PACKET_END
:
3184 verbose(env
, "invalid access to packet, off=%d size=%d, R%d(id=%d,off=%d,r=%d)\n",
3185 off
, size
, regno
, reg
->id
, off
, mem_size
);
3189 verbose(env
, "invalid access to memory, mem_size=%u off=%d size=%d\n",
3190 mem_size
, off
, size
);
3196 /* check read/write into a memory region with possible variable offset */
3197 static int check_mem_region_access(struct bpf_verifier_env
*env
, u32 regno
,
3198 int off
, int size
, u32 mem_size
,
3199 bool zero_size_allowed
)
3201 struct bpf_verifier_state
*vstate
= env
->cur_state
;
3202 struct bpf_func_state
*state
= vstate
->frame
[vstate
->curframe
];
3203 struct bpf_reg_state
*reg
= &state
->regs
[regno
];
3206 /* We may have adjusted the register pointing to memory region, so we
3207 * need to try adding each of min_value and max_value to off
3208 * to make sure our theoretical access will be safe.
3210 if (env
->log
.level
& BPF_LOG_LEVEL
)
3211 print_verifier_state(env
, state
);
3213 /* The minimum value is only important with signed
3214 * comparisons where we can't assume the floor of a
3215 * value is 0. If we are using signed variables for our
3216 * index'es we need to make sure that whatever we use
3217 * will have a set floor within our range.
3219 if (reg
->smin_value
< 0 &&
3220 (reg
->smin_value
== S64_MIN
||
3221 (off
+ reg
->smin_value
!= (s64
)(s32
)(off
+ reg
->smin_value
)) ||
3222 reg
->smin_value
+ off
< 0)) {
3223 verbose(env
, "R%d min value is negative, either use unsigned index or do a if (index >=0) check.\n",
3227 err
= __check_mem_access(env
, regno
, reg
->smin_value
+ off
, size
,
3228 mem_size
, zero_size_allowed
);
3230 verbose(env
, "R%d min value is outside of the allowed memory range\n",
3235 /* If we haven't set a max value then we need to bail since we can't be
3236 * sure we won't do bad things.
3237 * If reg->umax_value + off could overflow, treat that as unbounded too.
3239 if (reg
->umax_value
>= BPF_MAX_VAR_OFF
) {
3240 verbose(env
, "R%d unbounded memory access, make sure to bounds check any such access\n",
3244 err
= __check_mem_access(env
, regno
, reg
->umax_value
+ off
, size
,
3245 mem_size
, zero_size_allowed
);
3247 verbose(env
, "R%d max value is outside of the allowed memory range\n",
3255 /* check read/write into a map element with possible variable offset */
3256 static int check_map_access(struct bpf_verifier_env
*env
, u32 regno
,
3257 int off
, int size
, bool zero_size_allowed
)
3259 struct bpf_verifier_state
*vstate
= env
->cur_state
;
3260 struct bpf_func_state
*state
= vstate
->frame
[vstate
->curframe
];
3261 struct bpf_reg_state
*reg
= &state
->regs
[regno
];
3262 struct bpf_map
*map
= reg
->map_ptr
;
3265 err
= check_mem_region_access(env
, regno
, off
, size
, map
->value_size
,
3270 if (map_value_has_spin_lock(map
)) {
3271 u32 lock
= map
->spin_lock_off
;
3273 /* if any part of struct bpf_spin_lock can be touched by
3274 * load/store reject this program.
3275 * To check that [x1, x2) overlaps with [y1, y2)
3276 * it is sufficient to check x1 < y2 && y1 < x2.
3278 if (reg
->smin_value
+ off
< lock
+ sizeof(struct bpf_spin_lock
) &&
3279 lock
< reg
->umax_value
+ off
+ size
) {
3280 verbose(env
, "bpf_spin_lock cannot be accessed directly by load/store\n");
3284 if (map_value_has_timer(map
)) {
3285 u32 t
= map
->timer_off
;
3287 if (reg
->smin_value
+ off
< t
+ sizeof(struct bpf_timer
) &&
3288 t
< reg
->umax_value
+ off
+ size
) {
3289 verbose(env
, "bpf_timer cannot be accessed directly by load/store\n");
3296 #define MAX_PACKET_OFF 0xffff
3298 static enum bpf_prog_type
resolve_prog_type(struct bpf_prog
*prog
)
3300 return prog
->aux
->dst_prog
? prog
->aux
->dst_prog
->type
: prog
->type
;
3303 static bool may_access_direct_pkt_data(struct bpf_verifier_env
*env
,
3304 const struct bpf_call_arg_meta
*meta
,
3305 enum bpf_access_type t
)
3307 enum bpf_prog_type prog_type
= resolve_prog_type(env
->prog
);
3309 switch (prog_type
) {
3310 /* Program types only with direct read access go here! */
3311 case BPF_PROG_TYPE_LWT_IN
:
3312 case BPF_PROG_TYPE_LWT_OUT
:
3313 case BPF_PROG_TYPE_LWT_SEG6LOCAL
:
3314 case BPF_PROG_TYPE_SK_REUSEPORT
:
3315 case BPF_PROG_TYPE_FLOW_DISSECTOR
:
3316 case BPF_PROG_TYPE_CGROUP_SKB
:
3321 /* Program types with direct read + write access go here! */
3322 case BPF_PROG_TYPE_SCHED_CLS
:
3323 case BPF_PROG_TYPE_SCHED_ACT
:
3324 case BPF_PROG_TYPE_XDP
:
3325 case BPF_PROG_TYPE_LWT_XMIT
:
3326 case BPF_PROG_TYPE_SK_SKB
:
3327 case BPF_PROG_TYPE_SK_MSG
:
3329 return meta
->pkt_access
;
3331 env
->seen_direct_write
= true;
3334 case BPF_PROG_TYPE_CGROUP_SOCKOPT
:
3336 env
->seen_direct_write
= true;
3345 static int check_packet_access(struct bpf_verifier_env
*env
, u32 regno
, int off
,
3346 int size
, bool zero_size_allowed
)
3348 struct bpf_reg_state
*regs
= cur_regs(env
);
3349 struct bpf_reg_state
*reg
= ®s
[regno
];
3352 /* We may have added a variable offset to the packet pointer; but any
3353 * reg->range we have comes after that. We are only checking the fixed
3357 /* We don't allow negative numbers, because we aren't tracking enough
3358 * detail to prove they're safe.
3360 if (reg
->smin_value
< 0) {
3361 verbose(env
, "R%d min value is negative, either use unsigned index or do a if (index >=0) check.\n",
3366 err
= reg
->range
< 0 ? -EINVAL
:
3367 __check_mem_access(env
, regno
, off
, size
, reg
->range
,
3370 verbose(env
, "R%d offset is outside of the packet\n", regno
);
3374 /* __check_mem_access has made sure "off + size - 1" is within u16.
3375 * reg->umax_value can't be bigger than MAX_PACKET_OFF which is 0xffff,
3376 * otherwise find_good_pkt_pointers would have refused to set range info
3377 * that __check_mem_access would have rejected this pkt access.
3378 * Therefore, "off + reg->umax_value + size - 1" won't overflow u32.
3380 env
->prog
->aux
->max_pkt_offset
=
3381 max_t(u32
, env
->prog
->aux
->max_pkt_offset
,
3382 off
+ reg
->umax_value
+ size
- 1);
3387 /* check access to 'struct bpf_context' fields. Supports fixed offsets only */
3388 static int check_ctx_access(struct bpf_verifier_env
*env
, int insn_idx
, int off
, int size
,
3389 enum bpf_access_type t
, enum bpf_reg_type
*reg_type
,
3390 struct btf
**btf
, u32
*btf_id
)
3392 struct bpf_insn_access_aux info
= {
3393 .reg_type
= *reg_type
,
3397 if (env
->ops
->is_valid_access
&&
3398 env
->ops
->is_valid_access(off
, size
, t
, env
->prog
, &info
)) {
3399 /* A non zero info.ctx_field_size indicates that this field is a
3400 * candidate for later verifier transformation to load the whole
3401 * field and then apply a mask when accessed with a narrower
3402 * access than actual ctx access size. A zero info.ctx_field_size
3403 * will only allow for whole field access and rejects any other
3404 * type of narrower access.
3406 *reg_type
= info
.reg_type
;
3408 if (*reg_type
== PTR_TO_BTF_ID
|| *reg_type
== PTR_TO_BTF_ID_OR_NULL
) {
3410 *btf_id
= info
.btf_id
;
3412 env
->insn_aux_data
[insn_idx
].ctx_field_size
= info
.ctx_field_size
;
3414 /* remember the offset of last byte accessed in ctx */
3415 if (env
->prog
->aux
->max_ctx_offset
< off
+ size
)
3416 env
->prog
->aux
->max_ctx_offset
= off
+ size
;
3420 verbose(env
, "invalid bpf_context access off=%d size=%d\n", off
, size
);
3424 static int check_flow_keys_access(struct bpf_verifier_env
*env
, int off
,
3427 if (size
< 0 || off
< 0 ||
3428 (u64
)off
+ size
> sizeof(struct bpf_flow_keys
)) {
3429 verbose(env
, "invalid access to flow keys off=%d size=%d\n",
3436 static int check_sock_access(struct bpf_verifier_env
*env
, int insn_idx
,
3437 u32 regno
, int off
, int size
,
3438 enum bpf_access_type t
)
3440 struct bpf_reg_state
*regs
= cur_regs(env
);
3441 struct bpf_reg_state
*reg
= ®s
[regno
];
3442 struct bpf_insn_access_aux info
= {};
3445 if (reg
->smin_value
< 0) {
3446 verbose(env
, "R%d min value is negative, either use unsigned index or do a if (index >=0) check.\n",
3451 switch (reg
->type
) {
3452 case PTR_TO_SOCK_COMMON
:
3453 valid
= bpf_sock_common_is_valid_access(off
, size
, t
, &info
);
3456 valid
= bpf_sock_is_valid_access(off
, size
, t
, &info
);
3458 case PTR_TO_TCP_SOCK
:
3459 valid
= bpf_tcp_sock_is_valid_access(off
, size
, t
, &info
);
3461 case PTR_TO_XDP_SOCK
:
3462 valid
= bpf_xdp_sock_is_valid_access(off
, size
, t
, &info
);
3470 env
->insn_aux_data
[insn_idx
].ctx_field_size
=
3471 info
.ctx_field_size
;
3475 verbose(env
, "R%d invalid %s access off=%d size=%d\n",
3476 regno
, reg_type_str
[reg
->type
], off
, size
);
3481 static bool is_pointer_value(struct bpf_verifier_env
*env
, int regno
)
3483 return __is_pointer_value(env
->allow_ptr_leaks
, reg_state(env
, regno
));
3486 static bool is_ctx_reg(struct bpf_verifier_env
*env
, int regno
)
3488 const struct bpf_reg_state
*reg
= reg_state(env
, regno
);
3490 return reg
->type
== PTR_TO_CTX
;
3493 static bool is_sk_reg(struct bpf_verifier_env
*env
, int regno
)
3495 const struct bpf_reg_state
*reg
= reg_state(env
, regno
);
3497 return type_is_sk_pointer(reg
->type
);
3500 static bool is_pkt_reg(struct bpf_verifier_env
*env
, int regno
)
3502 const struct bpf_reg_state
*reg
= reg_state(env
, regno
);
3504 return type_is_pkt_pointer(reg
->type
);
3507 static bool is_flow_key_reg(struct bpf_verifier_env
*env
, int regno
)
3509 const struct bpf_reg_state
*reg
= reg_state(env
, regno
);
3511 /* Separate to is_ctx_reg() since we still want to allow BPF_ST here. */
3512 return reg
->type
== PTR_TO_FLOW_KEYS
;
3515 static int check_pkt_ptr_alignment(struct bpf_verifier_env
*env
,
3516 const struct bpf_reg_state
*reg
,
3517 int off
, int size
, bool strict
)
3519 struct tnum reg_off
;
3522 /* Byte size accesses are always allowed. */
3523 if (!strict
|| size
== 1)
3526 /* For platforms that do not have a Kconfig enabling
3527 * CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS the value of
3528 * NET_IP_ALIGN is universally set to '2'. And on platforms
3529 * that do set CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS, we get
3530 * to this code only in strict mode where we want to emulate
3531 * the NET_IP_ALIGN==2 checking. Therefore use an
3532 * unconditional IP align value of '2'.
3536 reg_off
= tnum_add(reg
->var_off
, tnum_const(ip_align
+ reg
->off
+ off
));
3537 if (!tnum_is_aligned(reg_off
, size
)) {
3540 tnum_strn(tn_buf
, sizeof(tn_buf
), reg
->var_off
);
3542 "misaligned packet access off %d+%s+%d+%d size %d\n",
3543 ip_align
, tn_buf
, reg
->off
, off
, size
);
3550 static int check_generic_ptr_alignment(struct bpf_verifier_env
*env
,
3551 const struct bpf_reg_state
*reg
,
3552 const char *pointer_desc
,
3553 int off
, int size
, bool strict
)
3555 struct tnum reg_off
;
3557 /* Byte size accesses are always allowed. */
3558 if (!strict
|| size
== 1)
3561 reg_off
= tnum_add(reg
->var_off
, tnum_const(reg
->off
+ off
));
3562 if (!tnum_is_aligned(reg_off
, size
)) {
3565 tnum_strn(tn_buf
, sizeof(tn_buf
), reg
->var_off
);
3566 verbose(env
, "misaligned %saccess off %s+%d+%d size %d\n",
3567 pointer_desc
, tn_buf
, reg
->off
, off
, size
);
3574 static int check_ptr_alignment(struct bpf_verifier_env
*env
,
3575 const struct bpf_reg_state
*reg
, int off
,
3576 int size
, bool strict_alignment_once
)
3578 bool strict
= env
->strict_alignment
|| strict_alignment_once
;
3579 const char *pointer_desc
= "";
3581 switch (reg
->type
) {
3583 case PTR_TO_PACKET_META
:
3584 /* Special case, because of NET_IP_ALIGN. Given metadata sits
3585 * right in front, treat it the very same way.
3587 return check_pkt_ptr_alignment(env
, reg
, off
, size
, strict
);
3588 case PTR_TO_FLOW_KEYS
:
3589 pointer_desc
= "flow keys ";
3591 case PTR_TO_MAP_KEY
:
3592 pointer_desc
= "key ";
3594 case PTR_TO_MAP_VALUE
:
3595 pointer_desc
= "value ";
3598 pointer_desc
= "context ";
3601 pointer_desc
= "stack ";
3602 /* The stack spill tracking logic in check_stack_write_fixed_off()
3603 * and check_stack_read_fixed_off() relies on stack accesses being
3609 pointer_desc
= "sock ";
3611 case PTR_TO_SOCK_COMMON
:
3612 pointer_desc
= "sock_common ";
3614 case PTR_TO_TCP_SOCK
:
3615 pointer_desc
= "tcp_sock ";
3617 case PTR_TO_XDP_SOCK
:
3618 pointer_desc
= "xdp_sock ";
3623 return check_generic_ptr_alignment(env
, reg
, pointer_desc
, off
, size
,
3627 static int update_stack_depth(struct bpf_verifier_env
*env
,
3628 const struct bpf_func_state
*func
,
3631 u16 stack
= env
->subprog_info
[func
->subprogno
].stack_depth
;
3636 /* update known max for given subprogram */
3637 env
->subprog_info
[func
->subprogno
].stack_depth
= -off
;
3641 /* starting from main bpf function walk all instructions of the function
3642 * and recursively walk all callees that given function can call.
3643 * Ignore jump and exit insns.
3644 * Since recursion is prevented by check_cfg() this algorithm
3645 * only needs a local stack of MAX_CALL_FRAMES to remember callsites
3647 static int check_max_stack_depth(struct bpf_verifier_env
*env
)
3649 int depth
= 0, frame
= 0, idx
= 0, i
= 0, subprog_end
;
3650 struct bpf_subprog_info
*subprog
= env
->subprog_info
;
3651 struct bpf_insn
*insn
= env
->prog
->insnsi
;
3652 bool tail_call_reachable
= false;
3653 int ret_insn
[MAX_CALL_FRAMES
];
3654 int ret_prog
[MAX_CALL_FRAMES
];
3658 /* protect against potential stack overflow that might happen when
3659 * bpf2bpf calls get combined with tailcalls. Limit the caller's stack
3660 * depth for such case down to 256 so that the worst case scenario
3661 * would result in 8k stack size (32 which is tailcall limit * 256 =
3664 * To get the idea what might happen, see an example:
3665 * func1 -> sub rsp, 128
3666 * subfunc1 -> sub rsp, 256
3667 * tailcall1 -> add rsp, 256
3668 * func2 -> sub rsp, 192 (total stack size = 128 + 192 = 320)
3669 * subfunc2 -> sub rsp, 64
3670 * subfunc22 -> sub rsp, 128
3671 * tailcall2 -> add rsp, 128
3672 * func3 -> sub rsp, 32 (total stack size 128 + 192 + 64 + 32 = 416)
3674 * tailcall will unwind the current stack frame but it will not get rid
3675 * of caller's stack as shown on the example above.
3677 if (idx
&& subprog
[idx
].has_tail_call
&& depth
>= 256) {
3679 "tail_calls are not allowed when call stack of previous frames is %d bytes. Too large\n",
3683 /* round up to 32-bytes, since this is granularity
3684 * of interpreter stack size
3686 depth
+= round_up(max_t(u32
, subprog
[idx
].stack_depth
, 1), 32);
3687 if (depth
> MAX_BPF_STACK
) {
3688 verbose(env
, "combined stack size of %d calls is %d. Too large\n",
3693 subprog_end
= subprog
[idx
+ 1].start
;
3694 for (; i
< subprog_end
; i
++) {
3697 if (!bpf_pseudo_call(insn
+ i
) && !bpf_pseudo_func(insn
+ i
))
3699 /* remember insn and function to return to */
3700 ret_insn
[frame
] = i
+ 1;
3701 ret_prog
[frame
] = idx
;
3703 /* find the callee */
3704 next_insn
= i
+ insn
[i
].imm
+ 1;
3705 idx
= find_subprog(env
, next_insn
);
3707 WARN_ONCE(1, "verifier bug. No program starts at insn %d\n",
3711 if (subprog
[idx
].is_async_cb
) {
3712 if (subprog
[idx
].has_tail_call
) {
3713 verbose(env
, "verifier bug. subprog has tail_call and async cb\n");
3716 /* async callbacks don't increase bpf prog stack size */
3721 if (subprog
[idx
].has_tail_call
)
3722 tail_call_reachable
= true;
3725 if (frame
>= MAX_CALL_FRAMES
) {
3726 verbose(env
, "the call stack of %d frames is too deep !\n",
3732 /* if tail call got detected across bpf2bpf calls then mark each of the
3733 * currently present subprog frames as tail call reachable subprogs;
3734 * this info will be utilized by JIT so that we will be preserving the
3735 * tail call counter throughout bpf2bpf calls combined with tailcalls
3737 if (tail_call_reachable
)
3738 for (j
= 0; j
< frame
; j
++)
3739 subprog
[ret_prog
[j
]].tail_call_reachable
= true;
3740 if (subprog
[0].tail_call_reachable
)
3741 env
->prog
->aux
->tail_call_reachable
= true;
3743 /* end of for() loop means the last insn of the 'subprog'
3744 * was reached. Doesn't matter whether it was JA or EXIT
3748 depth
-= round_up(max_t(u32
, subprog
[idx
].stack_depth
, 1), 32);
3750 i
= ret_insn
[frame
];
3751 idx
= ret_prog
[frame
];
3755 #ifndef CONFIG_BPF_JIT_ALWAYS_ON
3756 static int get_callee_stack_depth(struct bpf_verifier_env
*env
,
3757 const struct bpf_insn
*insn
, int idx
)
3759 int start
= idx
+ insn
->imm
+ 1, subprog
;
3761 subprog
= find_subprog(env
, start
);
3763 WARN_ONCE(1, "verifier bug. No program starts at insn %d\n",
3767 return env
->subprog_info
[subprog
].stack_depth
;
3771 int check_ctx_reg(struct bpf_verifier_env
*env
,
3772 const struct bpf_reg_state
*reg
, int regno
)
3774 /* Access to ctx or passing it to a helper is only allowed in
3775 * its original, unmodified form.
3779 verbose(env
, "dereference of modified ctx ptr R%d off=%d disallowed\n",
3784 if (!tnum_is_const(reg
->var_off
) || reg
->var_off
.value
) {
3787 tnum_strn(tn_buf
, sizeof(tn_buf
), reg
->var_off
);
3788 verbose(env
, "variable ctx access var_off=%s disallowed\n", tn_buf
);
3795 static int __check_buffer_access(struct bpf_verifier_env
*env
,
3796 const char *buf_info
,
3797 const struct bpf_reg_state
*reg
,
3798 int regno
, int off
, int size
)
3802 "R%d invalid %s buffer access: off=%d, size=%d\n",
3803 regno
, buf_info
, off
, size
);
3806 if (!tnum_is_const(reg
->var_off
) || reg
->var_off
.value
) {
3809 tnum_strn(tn_buf
, sizeof(tn_buf
), reg
->var_off
);
3811 "R%d invalid variable buffer offset: off=%d, var_off=%s\n",
3812 regno
, off
, tn_buf
);
3819 static int check_tp_buffer_access(struct bpf_verifier_env
*env
,
3820 const struct bpf_reg_state
*reg
,
3821 int regno
, int off
, int size
)
3825 err
= __check_buffer_access(env
, "tracepoint", reg
, regno
, off
, size
);
3829 if (off
+ size
> env
->prog
->aux
->max_tp_access
)
3830 env
->prog
->aux
->max_tp_access
= off
+ size
;
3835 static int check_buffer_access(struct bpf_verifier_env
*env
,
3836 const struct bpf_reg_state
*reg
,
3837 int regno
, int off
, int size
,
3838 bool zero_size_allowed
,
3839 const char *buf_info
,
3844 err
= __check_buffer_access(env
, buf_info
, reg
, regno
, off
, size
);
3848 if (off
+ size
> *max_access
)
3849 *max_access
= off
+ size
;
3854 /* BPF architecture zero extends alu32 ops into 64-bit registesr */
3855 static void zext_32_to_64(struct bpf_reg_state
*reg
)
3857 reg
->var_off
= tnum_subreg(reg
->var_off
);
3858 __reg_assign_32_into_64(reg
);
3861 /* truncate register to smaller size (in bytes)
3862 * must be called with size < BPF_REG_SIZE
3864 static void coerce_reg_to_size(struct bpf_reg_state
*reg
, int size
)
3868 /* clear high bits in bit representation */
3869 reg
->var_off
= tnum_cast(reg
->var_off
, size
);
3871 /* fix arithmetic bounds */
3872 mask
= ((u64
)1 << (size
* 8)) - 1;
3873 if ((reg
->umin_value
& ~mask
) == (reg
->umax_value
& ~mask
)) {
3874 reg
->umin_value
&= mask
;
3875 reg
->umax_value
&= mask
;
3877 reg
->umin_value
= 0;
3878 reg
->umax_value
= mask
;
3880 reg
->smin_value
= reg
->umin_value
;
3881 reg
->smax_value
= reg
->umax_value
;
3883 /* If size is smaller than 32bit register the 32bit register
3884 * values are also truncated so we push 64-bit bounds into
3885 * 32-bit bounds. Above were truncated < 32-bits already.
3889 __reg_combine_64_into_32(reg
);
3892 static bool bpf_map_is_rdonly(const struct bpf_map
*map
)
3894 /* A map is considered read-only if the following condition are true:
3896 * 1) BPF program side cannot change any of the map content. The
3897 * BPF_F_RDONLY_PROG flag is throughout the lifetime of a map
3898 * and was set at map creation time.
3899 * 2) The map value(s) have been initialized from user space by a
3900 * loader and then "frozen", such that no new map update/delete
3901 * operations from syscall side are possible for the rest of
3902 * the map's lifetime from that point onwards.
3903 * 3) Any parallel/pending map update/delete operations from syscall
3904 * side have been completed. Only after that point, it's safe to
3905 * assume that map value(s) are immutable.
3907 return (map
->map_flags
& BPF_F_RDONLY_PROG
) &&
3908 READ_ONCE(map
->frozen
) &&
3909 !bpf_map_write_active(map
);
3912 static int bpf_map_direct_read(struct bpf_map
*map
, int off
, int size
, u64
*val
)
3918 err
= map
->ops
->map_direct_value_addr(map
, &addr
, off
);
3921 ptr
= (void *)(long)addr
+ off
;
3925 *val
= (u64
)*(u8
*)ptr
;
3928 *val
= (u64
)*(u16
*)ptr
;
3931 *val
= (u64
)*(u32
*)ptr
;
3942 static int check_ptr_to_btf_access(struct bpf_verifier_env
*env
,
3943 struct bpf_reg_state
*regs
,
3944 int regno
, int off
, int size
,
3945 enum bpf_access_type atype
,
3948 struct bpf_reg_state
*reg
= regs
+ regno
;
3949 const struct btf_type
*t
= btf_type_by_id(reg
->btf
, reg
->btf_id
);
3950 const char *tname
= btf_name_by_offset(reg
->btf
, t
->name_off
);
3956 "R%d is ptr_%s invalid negative access: off=%d\n",
3960 if (!tnum_is_const(reg
->var_off
) || reg
->var_off
.value
) {
3963 tnum_strn(tn_buf
, sizeof(tn_buf
), reg
->var_off
);
3965 "R%d is ptr_%s invalid variable offset: off=%d, var_off=%s\n",
3966 regno
, tname
, off
, tn_buf
);
3970 if (env
->ops
->btf_struct_access
) {
3971 ret
= env
->ops
->btf_struct_access(&env
->log
, reg
->btf
, t
,
3972 off
, size
, atype
, &btf_id
);
3974 if (atype
!= BPF_READ
) {
3975 verbose(env
, "only read is supported\n");
3979 ret
= btf_struct_access(&env
->log
, reg
->btf
, t
, off
, size
,
3986 if (atype
== BPF_READ
&& value_regno
>= 0)
3987 mark_btf_ld_reg(env
, regs
, value_regno
, ret
, reg
->btf
, btf_id
);
3992 static int check_ptr_to_map_access(struct bpf_verifier_env
*env
,
3993 struct bpf_reg_state
*regs
,
3994 int regno
, int off
, int size
,
3995 enum bpf_access_type atype
,
3998 struct bpf_reg_state
*reg
= regs
+ regno
;
3999 struct bpf_map
*map
= reg
->map_ptr
;
4000 const struct btf_type
*t
;
4006 verbose(env
, "map_ptr access not supported without CONFIG_DEBUG_INFO_BTF\n");
4010 if (!map
->ops
->map_btf_id
|| !*map
->ops
->map_btf_id
) {
4011 verbose(env
, "map_ptr access not supported for map type %d\n",
4016 t
= btf_type_by_id(btf_vmlinux
, *map
->ops
->map_btf_id
);
4017 tname
= btf_name_by_offset(btf_vmlinux
, t
->name_off
);
4019 if (!env
->allow_ptr_to_map_access
) {
4021 "%s access is allowed only to CAP_PERFMON and CAP_SYS_ADMIN\n",
4027 verbose(env
, "R%d is %s invalid negative access: off=%d\n",
4032 if (atype
!= BPF_READ
) {
4033 verbose(env
, "only read from %s is supported\n", tname
);
4037 ret
= btf_struct_access(&env
->log
, btf_vmlinux
, t
, off
, size
, atype
, &btf_id
);
4041 if (value_regno
>= 0)
4042 mark_btf_ld_reg(env
, regs
, value_regno
, ret
, btf_vmlinux
, btf_id
);
4047 /* Check that the stack access at the given offset is within bounds. The
4048 * maximum valid offset is -1.
4050 * The minimum valid offset is -MAX_BPF_STACK for writes, and
4051 * -state->allocated_stack for reads.
4053 static int check_stack_slot_within_bounds(int off
,
4054 struct bpf_func_state
*state
,
4055 enum bpf_access_type t
)
4060 min_valid_off
= -MAX_BPF_STACK
;
4062 min_valid_off
= -state
->allocated_stack
;
4064 if (off
< min_valid_off
|| off
> -1)
4069 /* Check that the stack access at 'regno + off' falls within the maximum stack
4072 * 'off' includes `regno->offset`, but not its dynamic part (if any).
4074 static int check_stack_access_within_bounds(
4075 struct bpf_verifier_env
*env
,
4076 int regno
, int off
, int access_size
,
4077 enum stack_access_src src
, enum bpf_access_type type
)
4079 struct bpf_reg_state
*regs
= cur_regs(env
);
4080 struct bpf_reg_state
*reg
= regs
+ regno
;
4081 struct bpf_func_state
*state
= func(env
, reg
);
4082 int min_off
, max_off
;
4086 if (src
== ACCESS_HELPER
)
4087 /* We don't know if helpers are reading or writing (or both). */
4088 err_extra
= " indirect access to";
4089 else if (type
== BPF_READ
)
4090 err_extra
= " read from";
4092 err_extra
= " write to";
4094 if (tnum_is_const(reg
->var_off
)) {
4095 min_off
= reg
->var_off
.value
+ off
;
4096 if (access_size
> 0)
4097 max_off
= min_off
+ access_size
- 1;
4101 if (reg
->smax_value
>= BPF_MAX_VAR_OFF
||
4102 reg
->smin_value
<= -BPF_MAX_VAR_OFF
) {
4103 verbose(env
, "invalid unbounded variable-offset%s stack R%d\n",
4107 min_off
= reg
->smin_value
+ off
;
4108 if (access_size
> 0)
4109 max_off
= reg
->smax_value
+ off
+ access_size
- 1;
4114 err
= check_stack_slot_within_bounds(min_off
, state
, type
);
4116 err
= check_stack_slot_within_bounds(max_off
, state
, type
);
4119 if (tnum_is_const(reg
->var_off
)) {
4120 verbose(env
, "invalid%s stack R%d off=%d size=%d\n",
4121 err_extra
, regno
, off
, access_size
);
4125 tnum_strn(tn_buf
, sizeof(tn_buf
), reg
->var_off
);
4126 verbose(env
, "invalid variable-offset%s stack R%d var_off=%s size=%d\n",
4127 err_extra
, regno
, tn_buf
, access_size
);
4133 /* check whether memory at (regno + off) is accessible for t = (read | write)
4134 * if t==write, value_regno is a register which value is stored into memory
4135 * if t==read, value_regno is a register which will receive the value from memory
4136 * if t==write && value_regno==-1, some unknown value is stored into memory
4137 * if t==read && value_regno==-1, don't care what we read from memory
4139 static int check_mem_access(struct bpf_verifier_env
*env
, int insn_idx
, u32 regno
,
4140 int off
, int bpf_size
, enum bpf_access_type t
,
4141 int value_regno
, bool strict_alignment_once
)
4143 struct bpf_reg_state
*regs
= cur_regs(env
);
4144 struct bpf_reg_state
*reg
= regs
+ regno
;
4145 struct bpf_func_state
*state
;
4148 size
= bpf_size_to_bytes(bpf_size
);
4152 /* alignment checks will add in reg->off themselves */
4153 err
= check_ptr_alignment(env
, reg
, off
, size
, strict_alignment_once
);
4157 /* for access checks, reg->off is just part of off */
4160 if (reg
->type
== PTR_TO_MAP_KEY
) {
4161 if (t
== BPF_WRITE
) {
4162 verbose(env
, "write to change key R%d not allowed\n", regno
);
4166 err
= check_mem_region_access(env
, regno
, off
, size
,
4167 reg
->map_ptr
->key_size
, false);
4170 if (value_regno
>= 0)
4171 mark_reg_unknown(env
, regs
, value_regno
);
4172 } else if (reg
->type
== PTR_TO_MAP_VALUE
) {
4173 if (t
== BPF_WRITE
&& value_regno
>= 0 &&
4174 is_pointer_value(env
, value_regno
)) {
4175 verbose(env
, "R%d leaks addr into map\n", value_regno
);
4178 err
= check_map_access_type(env
, regno
, off
, size
, t
);
4181 err
= check_map_access(env
, regno
, off
, size
, false);
4182 if (!err
&& t
== BPF_READ
&& value_regno
>= 0) {
4183 struct bpf_map
*map
= reg
->map_ptr
;
4185 /* if map is read-only, track its contents as scalars */
4186 if (tnum_is_const(reg
->var_off
) &&
4187 bpf_map_is_rdonly(map
) &&
4188 map
->ops
->map_direct_value_addr
) {
4189 int map_off
= off
+ reg
->var_off
.value
;
4192 err
= bpf_map_direct_read(map
, map_off
, size
,
4197 regs
[value_regno
].type
= SCALAR_VALUE
;
4198 __mark_reg_known(®s
[value_regno
], val
);
4200 mark_reg_unknown(env
, regs
, value_regno
);
4203 } else if (reg
->type
== PTR_TO_MEM
) {
4204 if (t
== BPF_WRITE
&& value_regno
>= 0 &&
4205 is_pointer_value(env
, value_regno
)) {
4206 verbose(env
, "R%d leaks addr into mem\n", value_regno
);
4209 err
= check_mem_region_access(env
, regno
, off
, size
,
4210 reg
->mem_size
, false);
4211 if (!err
&& t
== BPF_READ
&& value_regno
>= 0)
4212 mark_reg_unknown(env
, regs
, value_regno
);
4213 } else if (reg
->type
== PTR_TO_CTX
) {
4214 enum bpf_reg_type reg_type
= SCALAR_VALUE
;
4215 struct btf
*btf
= NULL
;
4218 if (t
== BPF_WRITE
&& value_regno
>= 0 &&
4219 is_pointer_value(env
, value_regno
)) {
4220 verbose(env
, "R%d leaks addr into ctx\n", value_regno
);
4224 err
= check_ctx_reg(env
, reg
, regno
);
4228 err
= check_ctx_access(env
, insn_idx
, off
, size
, t
, ®_type
, &btf
, &btf_id
);
4230 verbose_linfo(env
, insn_idx
, "; ");
4231 if (!err
&& t
== BPF_READ
&& value_regno
>= 0) {
4232 /* ctx access returns either a scalar, or a
4233 * PTR_TO_PACKET[_META,_END]. In the latter
4234 * case, we know the offset is zero.
4236 if (reg_type
== SCALAR_VALUE
) {
4237 mark_reg_unknown(env
, regs
, value_regno
);
4239 mark_reg_known_zero(env
, regs
,
4241 if (reg_type_may_be_null(reg_type
))
4242 regs
[value_regno
].id
= ++env
->id_gen
;
4243 /* A load of ctx field could have different
4244 * actual load size with the one encoded in the
4245 * insn. When the dst is PTR, it is for sure not
4248 regs
[value_regno
].subreg_def
= DEF_NOT_SUBREG
;
4249 if (reg_type
== PTR_TO_BTF_ID
||
4250 reg_type
== PTR_TO_BTF_ID_OR_NULL
) {
4251 regs
[value_regno
].btf
= btf
;
4252 regs
[value_regno
].btf_id
= btf_id
;
4255 regs
[value_regno
].type
= reg_type
;
4258 } else if (reg
->type
== PTR_TO_STACK
) {
4259 /* Basic bounds checks. */
4260 err
= check_stack_access_within_bounds(env
, regno
, off
, size
, ACCESS_DIRECT
, t
);
4264 state
= func(env
, reg
);
4265 err
= update_stack_depth(env
, state
, off
);
4270 err
= check_stack_read(env
, regno
, off
, size
,
4273 err
= check_stack_write(env
, regno
, off
, size
,
4274 value_regno
, insn_idx
);
4275 } else if (reg_is_pkt_pointer(reg
)) {
4276 if (t
== BPF_WRITE
&& !may_access_direct_pkt_data(env
, NULL
, t
)) {
4277 verbose(env
, "cannot write into packet\n");
4280 if (t
== BPF_WRITE
&& value_regno
>= 0 &&
4281 is_pointer_value(env
, value_regno
)) {
4282 verbose(env
, "R%d leaks addr into packet\n",
4286 err
= check_packet_access(env
, regno
, off
, size
, false);
4287 if (!err
&& t
== BPF_READ
&& value_regno
>= 0)
4288 mark_reg_unknown(env
, regs
, value_regno
);
4289 } else if (reg
->type
== PTR_TO_FLOW_KEYS
) {
4290 if (t
== BPF_WRITE
&& value_regno
>= 0 &&
4291 is_pointer_value(env
, value_regno
)) {
4292 verbose(env
, "R%d leaks addr into flow keys\n",
4297 err
= check_flow_keys_access(env
, off
, size
);
4298 if (!err
&& t
== BPF_READ
&& value_regno
>= 0)
4299 mark_reg_unknown(env
, regs
, value_regno
);
4300 } else if (type_is_sk_pointer(reg
->type
)) {
4301 if (t
== BPF_WRITE
) {
4302 verbose(env
, "R%d cannot write into %s\n",
4303 regno
, reg_type_str
[reg
->type
]);
4306 err
= check_sock_access(env
, insn_idx
, regno
, off
, size
, t
);
4307 if (!err
&& value_regno
>= 0)
4308 mark_reg_unknown(env
, regs
, value_regno
);
4309 } else if (reg
->type
== PTR_TO_TP_BUFFER
) {
4310 err
= check_tp_buffer_access(env
, reg
, regno
, off
, size
);
4311 if (!err
&& t
== BPF_READ
&& value_regno
>= 0)
4312 mark_reg_unknown(env
, regs
, value_regno
);
4313 } else if (reg
->type
== PTR_TO_BTF_ID
) {
4314 err
= check_ptr_to_btf_access(env
, regs
, regno
, off
, size
, t
,
4316 } else if (reg
->type
== CONST_PTR_TO_MAP
) {
4317 err
= check_ptr_to_map_access(env
, regs
, regno
, off
, size
, t
,
4319 } else if (reg
->type
== PTR_TO_RDONLY_BUF
) {
4320 if (t
== BPF_WRITE
) {
4321 verbose(env
, "R%d cannot write into %s\n",
4322 regno
, reg_type_str
[reg
->type
]);
4325 err
= check_buffer_access(env
, reg
, regno
, off
, size
, false,
4327 &env
->prog
->aux
->max_rdonly_access
);
4328 if (!err
&& value_regno
>= 0)
4329 mark_reg_unknown(env
, regs
, value_regno
);
4330 } else if (reg
->type
== PTR_TO_RDWR_BUF
) {
4331 err
= check_buffer_access(env
, reg
, regno
, off
, size
, false,
4333 &env
->prog
->aux
->max_rdwr_access
);
4334 if (!err
&& t
== BPF_READ
&& value_regno
>= 0)
4335 mark_reg_unknown(env
, regs
, value_regno
);
4337 verbose(env
, "R%d invalid mem access '%s'\n", regno
,
4338 reg_type_str
[reg
->type
]);
4342 if (!err
&& size
< BPF_REG_SIZE
&& value_regno
>= 0 && t
== BPF_READ
&&
4343 regs
[value_regno
].type
== SCALAR_VALUE
) {
4344 /* b/h/w load zero-extends, mark upper bits as known 0 */
4345 coerce_reg_to_size(®s
[value_regno
], size
);
4350 static int check_atomic(struct bpf_verifier_env
*env
, int insn_idx
, struct bpf_insn
*insn
)
4355 switch (insn
->imm
) {
4357 case BPF_ADD
| BPF_FETCH
:
4359 case BPF_AND
| BPF_FETCH
:
4361 case BPF_OR
| BPF_FETCH
:
4363 case BPF_XOR
| BPF_FETCH
:
4368 verbose(env
, "BPF_ATOMIC uses invalid atomic opcode %02x\n", insn
->imm
);
4372 if (BPF_SIZE(insn
->code
) != BPF_W
&& BPF_SIZE(insn
->code
) != BPF_DW
) {
4373 verbose(env
, "invalid atomic operand size\n");
4377 /* check src1 operand */
4378 err
= check_reg_arg(env
, insn
->src_reg
, SRC_OP
);
4382 /* check src2 operand */
4383 err
= check_reg_arg(env
, insn
->dst_reg
, SRC_OP
);
4387 if (insn
->imm
== BPF_CMPXCHG
) {
4388 /* Check comparison of R0 with memory location */
4389 const u32 aux_reg
= BPF_REG_0
;
4391 err
= check_reg_arg(env
, aux_reg
, SRC_OP
);
4395 if (is_pointer_value(env
, aux_reg
)) {
4396 verbose(env
, "R%d leaks addr into mem\n", aux_reg
);
4401 if (is_pointer_value(env
, insn
->src_reg
)) {
4402 verbose(env
, "R%d leaks addr into mem\n", insn
->src_reg
);
4406 if (is_ctx_reg(env
, insn
->dst_reg
) ||
4407 is_pkt_reg(env
, insn
->dst_reg
) ||
4408 is_flow_key_reg(env
, insn
->dst_reg
) ||
4409 is_sk_reg(env
, insn
->dst_reg
)) {
4410 verbose(env
, "BPF_ATOMIC stores into R%d %s is not allowed\n",
4412 reg_type_str
[reg_state(env
, insn
->dst_reg
)->type
]);
4416 if (insn
->imm
& BPF_FETCH
) {
4417 if (insn
->imm
== BPF_CMPXCHG
)
4418 load_reg
= BPF_REG_0
;
4420 load_reg
= insn
->src_reg
;
4422 /* check and record load of old value */
4423 err
= check_reg_arg(env
, load_reg
, DST_OP
);
4427 /* This instruction accesses a memory location but doesn't
4428 * actually load it into a register.
4433 /* Check whether we can read the memory, with second call for fetch
4434 * case to simulate the register fill.
4436 err
= check_mem_access(env
, insn_idx
, insn
->dst_reg
, insn
->off
,
4437 BPF_SIZE(insn
->code
), BPF_READ
, -1, true);
4438 if (!err
&& load_reg
>= 0)
4439 err
= check_mem_access(env
, insn_idx
, insn
->dst_reg
, insn
->off
,
4440 BPF_SIZE(insn
->code
), BPF_READ
, load_reg
,
4445 /* Check whether we can write into the same memory. */
4446 err
= check_mem_access(env
, insn_idx
, insn
->dst_reg
, insn
->off
,
4447 BPF_SIZE(insn
->code
), BPF_WRITE
, -1, true);
4454 /* When register 'regno' is used to read the stack (either directly or through
4455 * a helper function) make sure that it's within stack boundary and, depending
4456 * on the access type, that all elements of the stack are initialized.
4458 * 'off' includes 'regno->off', but not its dynamic part (if any).
4460 * All registers that have been spilled on the stack in the slots within the
4461 * read offsets are marked as read.
4463 static int check_stack_range_initialized(
4464 struct bpf_verifier_env
*env
, int regno
, int off
,
4465 int access_size
, bool zero_size_allowed
,
4466 enum stack_access_src type
, struct bpf_call_arg_meta
*meta
)
4468 struct bpf_reg_state
*reg
= reg_state(env
, regno
);
4469 struct bpf_func_state
*state
= func(env
, reg
);
4470 int err
, min_off
, max_off
, i
, j
, slot
, spi
;
4471 char *err_extra
= type
== ACCESS_HELPER
? " indirect" : "";
4472 enum bpf_access_type bounds_check_type
;
4473 /* Some accesses can write anything into the stack, others are
4476 bool clobber
= false;
4478 if (access_size
== 0 && !zero_size_allowed
) {
4479 verbose(env
, "invalid zero-sized read\n");
4483 if (type
== ACCESS_HELPER
) {
4484 /* The bounds checks for writes are more permissive than for
4485 * reads. However, if raw_mode is not set, we'll do extra
4488 bounds_check_type
= BPF_WRITE
;
4491 bounds_check_type
= BPF_READ
;
4493 err
= check_stack_access_within_bounds(env
, regno
, off
, access_size
,
4494 type
, bounds_check_type
);
4499 if (tnum_is_const(reg
->var_off
)) {
4500 min_off
= max_off
= reg
->var_off
.value
+ off
;
4502 /* Variable offset is prohibited for unprivileged mode for
4503 * simplicity since it requires corresponding support in
4504 * Spectre masking for stack ALU.
4505 * See also retrieve_ptr_limit().
4507 if (!env
->bypass_spec_v1
) {
4510 tnum_strn(tn_buf
, sizeof(tn_buf
), reg
->var_off
);
4511 verbose(env
, "R%d%s variable offset stack access prohibited for !root, var_off=%s\n",
4512 regno
, err_extra
, tn_buf
);
4515 /* Only initialized buffer on stack is allowed to be accessed
4516 * with variable offset. With uninitialized buffer it's hard to
4517 * guarantee that whole memory is marked as initialized on
4518 * helper return since specific bounds are unknown what may
4519 * cause uninitialized stack leaking.
4521 if (meta
&& meta
->raw_mode
)
4524 min_off
= reg
->smin_value
+ off
;
4525 max_off
= reg
->smax_value
+ off
;
4528 if (meta
&& meta
->raw_mode
) {
4529 meta
->access_size
= access_size
;
4530 meta
->regno
= regno
;
4534 for (i
= min_off
; i
< max_off
+ access_size
; i
++) {
4538 spi
= slot
/ BPF_REG_SIZE
;
4539 if (state
->allocated_stack
<= slot
)
4541 stype
= &state
->stack
[spi
].slot_type
[slot
% BPF_REG_SIZE
];
4542 if (*stype
== STACK_MISC
)
4544 if (*stype
== STACK_ZERO
) {
4546 /* helper can write anything into the stack */
4547 *stype
= STACK_MISC
;
4552 if (state
->stack
[spi
].slot_type
[0] == STACK_SPILL
&&
4553 state
->stack
[spi
].spilled_ptr
.type
== PTR_TO_BTF_ID
)
4556 if (state
->stack
[spi
].slot_type
[0] == STACK_SPILL
&&
4557 (state
->stack
[spi
].spilled_ptr
.type
== SCALAR_VALUE
||
4558 env
->allow_ptr_leaks
)) {
4560 __mark_reg_unknown(env
, &state
->stack
[spi
].spilled_ptr
);
4561 for (j
= 0; j
< BPF_REG_SIZE
; j
++)
4562 state
->stack
[spi
].slot_type
[j
] = STACK_MISC
;
4568 if (tnum_is_const(reg
->var_off
)) {
4569 verbose(env
, "invalid%s read from stack R%d off %d+%d size %d\n",
4570 err_extra
, regno
, min_off
, i
- min_off
, access_size
);
4574 tnum_strn(tn_buf
, sizeof(tn_buf
), reg
->var_off
);
4575 verbose(env
, "invalid%s read from stack R%d var_off %s+%d size %d\n",
4576 err_extra
, regno
, tn_buf
, i
- min_off
, access_size
);
4580 /* reading any byte out of 8-byte 'spill_slot' will cause
4581 * the whole slot to be marked as 'read'
4583 mark_reg_read(env
, &state
->stack
[spi
].spilled_ptr
,
4584 state
->stack
[spi
].spilled_ptr
.parent
,
4587 return update_stack_depth(env
, state
, min_off
);
4590 static int check_helper_mem_access(struct bpf_verifier_env
*env
, int regno
,
4591 int access_size
, bool zero_size_allowed
,
4592 struct bpf_call_arg_meta
*meta
)
4594 struct bpf_reg_state
*regs
= cur_regs(env
), *reg
= ®s
[regno
];
4596 switch (reg
->type
) {
4598 case PTR_TO_PACKET_META
:
4599 return check_packet_access(env
, regno
, reg
->off
, access_size
,
4601 case PTR_TO_MAP_KEY
:
4602 return check_mem_region_access(env
, regno
, reg
->off
, access_size
,
4603 reg
->map_ptr
->key_size
, false);
4604 case PTR_TO_MAP_VALUE
:
4605 if (check_map_access_type(env
, regno
, reg
->off
, access_size
,
4606 meta
&& meta
->raw_mode
? BPF_WRITE
:
4609 return check_map_access(env
, regno
, reg
->off
, access_size
,
4612 return check_mem_region_access(env
, regno
, reg
->off
,
4613 access_size
, reg
->mem_size
,
4615 case PTR_TO_RDONLY_BUF
:
4616 if (meta
&& meta
->raw_mode
)
4618 return check_buffer_access(env
, reg
, regno
, reg
->off
,
4619 access_size
, zero_size_allowed
,
4621 &env
->prog
->aux
->max_rdonly_access
);
4622 case PTR_TO_RDWR_BUF
:
4623 return check_buffer_access(env
, reg
, regno
, reg
->off
,
4624 access_size
, zero_size_allowed
,
4626 &env
->prog
->aux
->max_rdwr_access
);
4628 return check_stack_range_initialized(
4630 regno
, reg
->off
, access_size
,
4631 zero_size_allowed
, ACCESS_HELPER
, meta
);
4632 default: /* scalar_value or invalid ptr */
4633 /* Allow zero-byte read from NULL, regardless of pointer type */
4634 if (zero_size_allowed
&& access_size
== 0 &&
4635 register_is_null(reg
))
4638 verbose(env
, "R%d type=%s expected=%s\n", regno
,
4639 reg_type_str
[reg
->type
],
4640 reg_type_str
[PTR_TO_STACK
]);
4645 int check_mem_reg(struct bpf_verifier_env
*env
, struct bpf_reg_state
*reg
,
4646 u32 regno
, u32 mem_size
)
4648 if (register_is_null(reg
))
4651 if (reg_type_may_be_null(reg
->type
)) {
4652 /* Assuming that the register contains a value check if the memory
4653 * access is safe. Temporarily save and restore the register's state as
4654 * the conversion shouldn't be visible to a caller.
4656 const struct bpf_reg_state saved_reg
= *reg
;
4659 mark_ptr_not_null_reg(reg
);
4660 rv
= check_helper_mem_access(env
, regno
, mem_size
, true, NULL
);
4665 return check_helper_mem_access(env
, regno
, mem_size
, true, NULL
);
4668 /* Implementation details:
4669 * bpf_map_lookup returns PTR_TO_MAP_VALUE_OR_NULL
4670 * Two bpf_map_lookups (even with the same key) will have different reg->id.
4671 * For traditional PTR_TO_MAP_VALUE the verifier clears reg->id after
4672 * value_or_null->value transition, since the verifier only cares about
4673 * the range of access to valid map value pointer and doesn't care about actual
4674 * address of the map element.
4675 * For maps with 'struct bpf_spin_lock' inside map value the verifier keeps
4676 * reg->id > 0 after value_or_null->value transition. By doing so
4677 * two bpf_map_lookups will be considered two different pointers that
4678 * point to different bpf_spin_locks.
4679 * The verifier allows taking only one bpf_spin_lock at a time to avoid
4681 * Since only one bpf_spin_lock is allowed the checks are simpler than
4682 * reg_is_refcounted() logic. The verifier needs to remember only
4683 * one spin_lock instead of array of acquired_refs.
4684 * cur_state->active_spin_lock remembers which map value element got locked
4685 * and clears it after bpf_spin_unlock.
4687 static int process_spin_lock(struct bpf_verifier_env
*env
, int regno
,
4690 struct bpf_reg_state
*regs
= cur_regs(env
), *reg
= ®s
[regno
];
4691 struct bpf_verifier_state
*cur
= env
->cur_state
;
4692 bool is_const
= tnum_is_const(reg
->var_off
);
4693 struct bpf_map
*map
= reg
->map_ptr
;
4694 u64 val
= reg
->var_off
.value
;
4698 "R%d doesn't have constant offset. bpf_spin_lock has to be at the constant offset\n",
4704 "map '%s' has to have BTF in order to use bpf_spin_lock\n",
4708 if (!map_value_has_spin_lock(map
)) {
4709 if (map
->spin_lock_off
== -E2BIG
)
4711 "map '%s' has more than one 'struct bpf_spin_lock'\n",
4713 else if (map
->spin_lock_off
== -ENOENT
)
4715 "map '%s' doesn't have 'struct bpf_spin_lock'\n",
4719 "map '%s' is not a struct type or bpf_spin_lock is mangled\n",
4723 if (map
->spin_lock_off
!= val
+ reg
->off
) {
4724 verbose(env
, "off %lld doesn't point to 'struct bpf_spin_lock'\n",
4729 if (cur
->active_spin_lock
) {
4731 "Locking two bpf_spin_locks are not allowed\n");
4734 cur
->active_spin_lock
= reg
->id
;
4736 if (!cur
->active_spin_lock
) {
4737 verbose(env
, "bpf_spin_unlock without taking a lock\n");
4740 if (cur
->active_spin_lock
!= reg
->id
) {
4741 verbose(env
, "bpf_spin_unlock of different lock\n");
4744 cur
->active_spin_lock
= 0;
4749 static int process_timer_func(struct bpf_verifier_env
*env
, int regno
,
4750 struct bpf_call_arg_meta
*meta
)
4752 struct bpf_reg_state
*regs
= cur_regs(env
), *reg
= ®s
[regno
];
4753 bool is_const
= tnum_is_const(reg
->var_off
);
4754 struct bpf_map
*map
= reg
->map_ptr
;
4755 u64 val
= reg
->var_off
.value
;
4759 "R%d doesn't have constant offset. bpf_timer has to be at the constant offset\n",
4764 verbose(env
, "map '%s' has to have BTF in order to use bpf_timer\n",
4768 if (!map_value_has_timer(map
)) {
4769 if (map
->timer_off
== -E2BIG
)
4771 "map '%s' has more than one 'struct bpf_timer'\n",
4773 else if (map
->timer_off
== -ENOENT
)
4775 "map '%s' doesn't have 'struct bpf_timer'\n",
4779 "map '%s' is not a struct type or bpf_timer is mangled\n",
4783 if (map
->timer_off
!= val
+ reg
->off
) {
4784 verbose(env
, "off %lld doesn't point to 'struct bpf_timer' that is at %d\n",
4785 val
+ reg
->off
, map
->timer_off
);
4788 if (meta
->map_ptr
) {
4789 verbose(env
, "verifier bug. Two map pointers in a timer helper\n");
4792 meta
->map_uid
= reg
->map_uid
;
4793 meta
->map_ptr
= map
;
4797 static bool arg_type_is_mem_ptr(enum bpf_arg_type type
)
4799 return type
== ARG_PTR_TO_MEM
||
4800 type
== ARG_PTR_TO_MEM_OR_NULL
||
4801 type
== ARG_PTR_TO_UNINIT_MEM
;
4804 static bool arg_type_is_mem_size(enum bpf_arg_type type
)
4806 return type
== ARG_CONST_SIZE
||
4807 type
== ARG_CONST_SIZE_OR_ZERO
;
4810 static bool arg_type_is_alloc_size(enum bpf_arg_type type
)
4812 return type
== ARG_CONST_ALLOC_SIZE_OR_ZERO
;
4815 static bool arg_type_is_int_ptr(enum bpf_arg_type type
)
4817 return type
== ARG_PTR_TO_INT
||
4818 type
== ARG_PTR_TO_LONG
;
4821 static int int_ptr_type_to_size(enum bpf_arg_type type
)
4823 if (type
== ARG_PTR_TO_INT
)
4825 else if (type
== ARG_PTR_TO_LONG
)
4831 static int resolve_map_arg_type(struct bpf_verifier_env
*env
,
4832 const struct bpf_call_arg_meta
*meta
,
4833 enum bpf_arg_type
*arg_type
)
4835 if (!meta
->map_ptr
) {
4836 /* kernel subsystem misconfigured verifier */
4837 verbose(env
, "invalid map_ptr to access map->type\n");
4841 switch (meta
->map_ptr
->map_type
) {
4842 case BPF_MAP_TYPE_SOCKMAP
:
4843 case BPF_MAP_TYPE_SOCKHASH
:
4844 if (*arg_type
== ARG_PTR_TO_MAP_VALUE
) {
4845 *arg_type
= ARG_PTR_TO_BTF_ID_SOCK_COMMON
;
4847 verbose(env
, "invalid arg_type for sockmap/sockhash\n");
4858 struct bpf_reg_types
{
4859 const enum bpf_reg_type types
[10];
4863 static const struct bpf_reg_types map_key_value_types
= {
4873 static const struct bpf_reg_types sock_types
= {
4883 static const struct bpf_reg_types btf_id_sock_common_types
= {
4891 .btf_id
= &btf_sock_ids
[BTF_SOCK_TYPE_SOCK_COMMON
],
4895 static const struct bpf_reg_types mem_types
= {
4908 static const struct bpf_reg_types int_ptr_types
= {
4918 static const struct bpf_reg_types fullsock_types
= { .types
= { PTR_TO_SOCKET
} };
4919 static const struct bpf_reg_types scalar_types
= { .types
= { SCALAR_VALUE
} };
4920 static const struct bpf_reg_types context_types
= { .types
= { PTR_TO_CTX
} };
4921 static const struct bpf_reg_types alloc_mem_types
= { .types
= { PTR_TO_MEM
} };
4922 static const struct bpf_reg_types const_map_ptr_types
= { .types
= { CONST_PTR_TO_MAP
} };
4923 static const struct bpf_reg_types btf_ptr_types
= { .types
= { PTR_TO_BTF_ID
} };
4924 static const struct bpf_reg_types spin_lock_types
= { .types
= { PTR_TO_MAP_VALUE
} };
4925 static const struct bpf_reg_types percpu_btf_ptr_types
= { .types
= { PTR_TO_PERCPU_BTF_ID
} };
4926 static const struct bpf_reg_types func_ptr_types
= { .types
= { PTR_TO_FUNC
} };
4927 static const struct bpf_reg_types stack_ptr_types
= { .types
= { PTR_TO_STACK
} };
4928 static const struct bpf_reg_types const_str_ptr_types
= { .types
= { PTR_TO_MAP_VALUE
} };
4929 static const struct bpf_reg_types timer_types
= { .types
= { PTR_TO_MAP_VALUE
} };
4931 static const struct bpf_reg_types
*compatible_reg_types
[__BPF_ARG_TYPE_MAX
] = {
4932 [ARG_PTR_TO_MAP_KEY
] = &map_key_value_types
,
4933 [ARG_PTR_TO_MAP_VALUE
] = &map_key_value_types
,
4934 [ARG_PTR_TO_UNINIT_MAP_VALUE
] = &map_key_value_types
,
4935 [ARG_PTR_TO_MAP_VALUE_OR_NULL
] = &map_key_value_types
,
4936 [ARG_CONST_SIZE
] = &scalar_types
,
4937 [ARG_CONST_SIZE_OR_ZERO
] = &scalar_types
,
4938 [ARG_CONST_ALLOC_SIZE_OR_ZERO
] = &scalar_types
,
4939 [ARG_CONST_MAP_PTR
] = &const_map_ptr_types
,
4940 [ARG_PTR_TO_CTX
] = &context_types
,
4941 [ARG_PTR_TO_CTX_OR_NULL
] = &context_types
,
4942 [ARG_PTR_TO_SOCK_COMMON
] = &sock_types
,
4944 [ARG_PTR_TO_BTF_ID_SOCK_COMMON
] = &btf_id_sock_common_types
,
4946 [ARG_PTR_TO_SOCKET
] = &fullsock_types
,
4947 [ARG_PTR_TO_SOCKET_OR_NULL
] = &fullsock_types
,
4948 [ARG_PTR_TO_BTF_ID
] = &btf_ptr_types
,
4949 [ARG_PTR_TO_SPIN_LOCK
] = &spin_lock_types
,
4950 [ARG_PTR_TO_MEM
] = &mem_types
,
4951 [ARG_PTR_TO_MEM_OR_NULL
] = &mem_types
,
4952 [ARG_PTR_TO_UNINIT_MEM
] = &mem_types
,
4953 [ARG_PTR_TO_ALLOC_MEM
] = &alloc_mem_types
,
4954 [ARG_PTR_TO_ALLOC_MEM_OR_NULL
] = &alloc_mem_types
,
4955 [ARG_PTR_TO_INT
] = &int_ptr_types
,
4956 [ARG_PTR_TO_LONG
] = &int_ptr_types
,
4957 [ARG_PTR_TO_PERCPU_BTF_ID
] = &percpu_btf_ptr_types
,
4958 [ARG_PTR_TO_FUNC
] = &func_ptr_types
,
4959 [ARG_PTR_TO_STACK_OR_NULL
] = &stack_ptr_types
,
4960 [ARG_PTR_TO_CONST_STR
] = &const_str_ptr_types
,
4961 [ARG_PTR_TO_TIMER
] = &timer_types
,
4964 static int check_reg_type(struct bpf_verifier_env
*env
, u32 regno
,
4965 enum bpf_arg_type arg_type
,
4966 const u32
*arg_btf_id
)
4968 struct bpf_reg_state
*regs
= cur_regs(env
), *reg
= ®s
[regno
];
4969 enum bpf_reg_type expected
, type
= reg
->type
;
4970 const struct bpf_reg_types
*compatible
;
4973 compatible
= compatible_reg_types
[arg_type
];
4975 verbose(env
, "verifier internal error: unsupported arg type %d\n", arg_type
);
4979 for (i
= 0; i
< ARRAY_SIZE(compatible
->types
); i
++) {
4980 expected
= compatible
->types
[i
];
4981 if (expected
== NOT_INIT
)
4984 if (type
== expected
)
4988 verbose(env
, "R%d type=%s expected=", regno
, reg_type_str
[type
]);
4989 for (j
= 0; j
+ 1 < i
; j
++)
4990 verbose(env
, "%s, ", reg_type_str
[compatible
->types
[j
]]);
4991 verbose(env
, "%s\n", reg_type_str
[compatible
->types
[j
]]);
4995 if (type
== PTR_TO_BTF_ID
) {
4997 if (!compatible
->btf_id
) {
4998 verbose(env
, "verifier internal error: missing arg compatible BTF ID\n");
5001 arg_btf_id
= compatible
->btf_id
;
5004 if (!btf_struct_ids_match(&env
->log
, reg
->btf
, reg
->btf_id
, reg
->off
,
5005 btf_vmlinux
, *arg_btf_id
)) {
5006 verbose(env
, "R%d is of type %s but %s is expected\n",
5007 regno
, kernel_type_name(reg
->btf
, reg
->btf_id
),
5008 kernel_type_name(btf_vmlinux
, *arg_btf_id
));
5012 if (!tnum_is_const(reg
->var_off
) || reg
->var_off
.value
) {
5013 verbose(env
, "R%d is a pointer to in-kernel struct with non-zero offset\n",
5022 static int check_func_arg(struct bpf_verifier_env
*env
, u32 arg
,
5023 struct bpf_call_arg_meta
*meta
,
5024 const struct bpf_func_proto
*fn
)
5026 u32 regno
= BPF_REG_1
+ arg
;
5027 struct bpf_reg_state
*regs
= cur_regs(env
), *reg
= ®s
[regno
];
5028 enum bpf_arg_type arg_type
= fn
->arg_type
[arg
];
5029 enum bpf_reg_type type
= reg
->type
;
5032 if (arg_type
== ARG_DONTCARE
)
5035 err
= check_reg_arg(env
, regno
, SRC_OP
);
5039 if (arg_type
== ARG_ANYTHING
) {
5040 if (is_pointer_value(env
, regno
)) {
5041 verbose(env
, "R%d leaks addr into helper function\n",
5048 if (type_is_pkt_pointer(type
) &&
5049 !may_access_direct_pkt_data(env
, meta
, BPF_READ
)) {
5050 verbose(env
, "helper access to the packet is not allowed\n");
5054 if (arg_type
== ARG_PTR_TO_MAP_VALUE
||
5055 arg_type
== ARG_PTR_TO_UNINIT_MAP_VALUE
||
5056 arg_type
== ARG_PTR_TO_MAP_VALUE_OR_NULL
) {
5057 err
= resolve_map_arg_type(env
, meta
, &arg_type
);
5062 if (register_is_null(reg
) && arg_type_may_be_null(arg_type
))
5063 /* A NULL register has a SCALAR_VALUE type, so skip
5066 goto skip_type_check
;
5068 /* We already checked for NULL above */
5069 if (arg_type
== ARG_PTR_TO_ALLOC_MEM
) {
5070 if (reg
->off
!= 0 || !tnum_is_const(reg
->var_off
)) {
5071 verbose(env
, "helper wants pointer to allocated memory\n");
5076 err
= check_reg_type(env
, regno
, arg_type
, fn
->arg_btf_id
[arg
]);
5080 if (type
== PTR_TO_CTX
) {
5081 err
= check_ctx_reg(env
, reg
, regno
);
5087 if (reg
->ref_obj_id
) {
5088 if (meta
->ref_obj_id
) {
5089 verbose(env
, "verifier internal error: more than one arg with ref_obj_id R%d %u %u\n",
5090 regno
, reg
->ref_obj_id
,
5094 meta
->ref_obj_id
= reg
->ref_obj_id
;
5097 if (arg_type
== ARG_CONST_MAP_PTR
) {
5098 /* bpf_map_xxx(map_ptr) call: remember that map_ptr */
5099 if (meta
->map_ptr
) {
5100 /* Use map_uid (which is unique id of inner map) to reject:
5101 * inner_map1 = bpf_map_lookup_elem(outer_map, key1)
5102 * inner_map2 = bpf_map_lookup_elem(outer_map, key2)
5103 * if (inner_map1 && inner_map2) {
5104 * timer = bpf_map_lookup_elem(inner_map1);
5106 * // mismatch would have been allowed
5107 * bpf_timer_init(timer, inner_map2);
5110 * Comparing map_ptr is enough to distinguish normal and outer maps.
5112 if (meta
->map_ptr
!= reg
->map_ptr
||
5113 meta
->map_uid
!= reg
->map_uid
) {
5115 "timer pointer in R1 map_uid=%d doesn't match map pointer in R2 map_uid=%d\n",
5116 meta
->map_uid
, reg
->map_uid
);
5120 meta
->map_ptr
= reg
->map_ptr
;
5121 meta
->map_uid
= reg
->map_uid
;
5122 } else if (arg_type
== ARG_PTR_TO_MAP_KEY
) {
5123 /* bpf_map_xxx(..., map_ptr, ..., key) call:
5124 * check that [key, key + map->key_size) are within
5125 * stack limits and initialized
5127 if (!meta
->map_ptr
) {
5128 /* in function declaration map_ptr must come before
5129 * map_key, so that it's verified and known before
5130 * we have to check map_key here. Otherwise it means
5131 * that kernel subsystem misconfigured verifier
5133 verbose(env
, "invalid map_ptr to access map->key\n");
5136 err
= check_helper_mem_access(env
, regno
,
5137 meta
->map_ptr
->key_size
, false,
5139 } else if (arg_type
== ARG_PTR_TO_MAP_VALUE
||
5140 (arg_type
== ARG_PTR_TO_MAP_VALUE_OR_NULL
&&
5141 !register_is_null(reg
)) ||
5142 arg_type
== ARG_PTR_TO_UNINIT_MAP_VALUE
) {
5143 /* bpf_map_xxx(..., map_ptr, ..., value) call:
5144 * check [value, value + map->value_size) validity
5146 if (!meta
->map_ptr
) {
5147 /* kernel subsystem misconfigured verifier */
5148 verbose(env
, "invalid map_ptr to access map->value\n");
5151 meta
->raw_mode
= (arg_type
== ARG_PTR_TO_UNINIT_MAP_VALUE
);
5152 err
= check_helper_mem_access(env
, regno
,
5153 meta
->map_ptr
->value_size
, false,
5155 } else if (arg_type
== ARG_PTR_TO_PERCPU_BTF_ID
) {
5157 verbose(env
, "Helper has invalid btf_id in R%d\n", regno
);
5160 meta
->ret_btf
= reg
->btf
;
5161 meta
->ret_btf_id
= reg
->btf_id
;
5162 } else if (arg_type
== ARG_PTR_TO_SPIN_LOCK
) {
5163 if (meta
->func_id
== BPF_FUNC_spin_lock
) {
5164 if (process_spin_lock(env
, regno
, true))
5166 } else if (meta
->func_id
== BPF_FUNC_spin_unlock
) {
5167 if (process_spin_lock(env
, regno
, false))
5170 verbose(env
, "verifier internal error\n");
5173 } else if (arg_type
== ARG_PTR_TO_TIMER
) {
5174 if (process_timer_func(env
, regno
, meta
))
5176 } else if (arg_type
== ARG_PTR_TO_FUNC
) {
5177 meta
->subprogno
= reg
->subprogno
;
5178 } else if (arg_type_is_mem_ptr(arg_type
)) {
5179 /* The access to this pointer is only checked when we hit the
5180 * next is_mem_size argument below.
5182 meta
->raw_mode
= (arg_type
== ARG_PTR_TO_UNINIT_MEM
);
5183 } else if (arg_type_is_mem_size(arg_type
)) {
5184 bool zero_size_allowed
= (arg_type
== ARG_CONST_SIZE_OR_ZERO
);
5186 /* This is used to refine r0 return value bounds for helpers
5187 * that enforce this value as an upper bound on return values.
5188 * See do_refine_retval_range() for helpers that can refine
5189 * the return value. C type of helper is u32 so we pull register
5190 * bound from umax_value however, if negative verifier errors
5191 * out. Only upper bounds can be learned because retval is an
5192 * int type and negative retvals are allowed.
5194 meta
->msize_max_value
= reg
->umax_value
;
5196 /* The register is SCALAR_VALUE; the access check
5197 * happens using its boundaries.
5199 if (!tnum_is_const(reg
->var_off
))
5200 /* For unprivileged variable accesses, disable raw
5201 * mode so that the program is required to
5202 * initialize all the memory that the helper could
5203 * just partially fill up.
5207 if (reg
->smin_value
< 0) {
5208 verbose(env
, "R%d min value is negative, either use unsigned or 'var &= const'\n",
5213 if (reg
->umin_value
== 0) {
5214 err
= check_helper_mem_access(env
, regno
- 1, 0,
5221 if (reg
->umax_value
>= BPF_MAX_VAR_SIZ
) {
5222 verbose(env
, "R%d unbounded memory access, use 'var &= const' or 'if (var < const)'\n",
5226 err
= check_helper_mem_access(env
, regno
- 1,
5228 zero_size_allowed
, meta
);
5230 err
= mark_chain_precision(env
, regno
);
5231 } else if (arg_type_is_alloc_size(arg_type
)) {
5232 if (!tnum_is_const(reg
->var_off
)) {
5233 verbose(env
, "R%d is not a known constant'\n",
5237 meta
->mem_size
= reg
->var_off
.value
;
5238 } else if (arg_type_is_int_ptr(arg_type
)) {
5239 int size
= int_ptr_type_to_size(arg_type
);
5241 err
= check_helper_mem_access(env
, regno
, size
, false, meta
);
5244 err
= check_ptr_alignment(env
, reg
, 0, size
, true);
5245 } else if (arg_type
== ARG_PTR_TO_CONST_STR
) {
5246 struct bpf_map
*map
= reg
->map_ptr
;
5251 if (!bpf_map_is_rdonly(map
)) {
5252 verbose(env
, "R%d does not point to a readonly map'\n", regno
);
5256 if (!tnum_is_const(reg
->var_off
)) {
5257 verbose(env
, "R%d is not a constant address'\n", regno
);
5261 if (!map
->ops
->map_direct_value_addr
) {
5262 verbose(env
, "no direct value access support for this map type\n");
5266 err
= check_map_access(env
, regno
, reg
->off
,
5267 map
->value_size
- reg
->off
, false);
5271 map_off
= reg
->off
+ reg
->var_off
.value
;
5272 err
= map
->ops
->map_direct_value_addr(map
, &map_addr
, map_off
);
5274 verbose(env
, "direct value access on string failed\n");
5278 str_ptr
= (char *)(long)(map_addr
);
5279 if (!strnchr(str_ptr
+ map_off
, map
->value_size
- map_off
, 0)) {
5280 verbose(env
, "string is not zero-terminated\n");
5288 static bool may_update_sockmap(struct bpf_verifier_env
*env
, int func_id
)
5290 enum bpf_attach_type eatype
= env
->prog
->expected_attach_type
;
5291 enum bpf_prog_type type
= resolve_prog_type(env
->prog
);
5293 if (func_id
!= BPF_FUNC_map_update_elem
)
5296 /* It's not possible to get access to a locked struct sock in these
5297 * contexts, so updating is safe.
5300 case BPF_PROG_TYPE_TRACING
:
5301 if (eatype
== BPF_TRACE_ITER
)
5304 case BPF_PROG_TYPE_SOCKET_FILTER
:
5305 case BPF_PROG_TYPE_SCHED_CLS
:
5306 case BPF_PROG_TYPE_SCHED_ACT
:
5307 case BPF_PROG_TYPE_XDP
:
5308 case BPF_PROG_TYPE_SK_REUSEPORT
:
5309 case BPF_PROG_TYPE_FLOW_DISSECTOR
:
5310 case BPF_PROG_TYPE_SK_LOOKUP
:
5316 verbose(env
, "cannot update sockmap in this context\n");
5320 static bool allow_tail_call_in_subprogs(struct bpf_verifier_env
*env
)
5322 return env
->prog
->jit_requested
&& IS_ENABLED(CONFIG_X86_64
);
5325 static int check_map_func_compatibility(struct bpf_verifier_env
*env
,
5326 struct bpf_map
*map
, int func_id
)
5331 /* We need a two way check, first is from map perspective ... */
5332 switch (map
->map_type
) {
5333 case BPF_MAP_TYPE_PROG_ARRAY
:
5334 if (func_id
!= BPF_FUNC_tail_call
)
5337 case BPF_MAP_TYPE_PERF_EVENT_ARRAY
:
5338 if (func_id
!= BPF_FUNC_perf_event_read
&&
5339 func_id
!= BPF_FUNC_perf_event_output
&&
5340 func_id
!= BPF_FUNC_skb_output
&&
5341 func_id
!= BPF_FUNC_perf_event_read_value
&&
5342 func_id
!= BPF_FUNC_xdp_output
)
5345 case BPF_MAP_TYPE_RINGBUF
:
5346 if (func_id
!= BPF_FUNC_ringbuf_output
&&
5347 func_id
!= BPF_FUNC_ringbuf_reserve
&&
5348 func_id
!= BPF_FUNC_ringbuf_query
)
5351 case BPF_MAP_TYPE_STACK_TRACE
:
5352 if (func_id
!= BPF_FUNC_get_stackid
)
5355 case BPF_MAP_TYPE_CGROUP_ARRAY
:
5356 if (func_id
!= BPF_FUNC_skb_under_cgroup
&&
5357 func_id
!= BPF_FUNC_current_task_under_cgroup
)
5360 case BPF_MAP_TYPE_CGROUP_STORAGE
:
5361 case BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE
:
5362 if (func_id
!= BPF_FUNC_get_local_storage
)
5365 case BPF_MAP_TYPE_DEVMAP
:
5366 case BPF_MAP_TYPE_DEVMAP_HASH
:
5367 if (func_id
!= BPF_FUNC_redirect_map
&&
5368 func_id
!= BPF_FUNC_map_lookup_elem
)
5371 /* Restrict bpf side of cpumap and xskmap, open when use-cases
5374 case BPF_MAP_TYPE_CPUMAP
:
5375 if (func_id
!= BPF_FUNC_redirect_map
)
5378 case BPF_MAP_TYPE_XSKMAP
:
5379 if (func_id
!= BPF_FUNC_redirect_map
&&
5380 func_id
!= BPF_FUNC_map_lookup_elem
)
5383 case BPF_MAP_TYPE_ARRAY_OF_MAPS
:
5384 case BPF_MAP_TYPE_HASH_OF_MAPS
:
5385 if (func_id
!= BPF_FUNC_map_lookup_elem
)
5388 case BPF_MAP_TYPE_SOCKMAP
:
5389 if (func_id
!= BPF_FUNC_sk_redirect_map
&&
5390 func_id
!= BPF_FUNC_sock_map_update
&&
5391 func_id
!= BPF_FUNC_map_delete_elem
&&
5392 func_id
!= BPF_FUNC_msg_redirect_map
&&
5393 func_id
!= BPF_FUNC_sk_select_reuseport
&&
5394 func_id
!= BPF_FUNC_map_lookup_elem
&&
5395 !may_update_sockmap(env
, func_id
))
5398 case BPF_MAP_TYPE_SOCKHASH
:
5399 if (func_id
!= BPF_FUNC_sk_redirect_hash
&&
5400 func_id
!= BPF_FUNC_sock_hash_update
&&
5401 func_id
!= BPF_FUNC_map_delete_elem
&&
5402 func_id
!= BPF_FUNC_msg_redirect_hash
&&
5403 func_id
!= BPF_FUNC_sk_select_reuseport
&&
5404 func_id
!= BPF_FUNC_map_lookup_elem
&&
5405 !may_update_sockmap(env
, func_id
))
5408 case BPF_MAP_TYPE_REUSEPORT_SOCKARRAY
:
5409 if (func_id
!= BPF_FUNC_sk_select_reuseport
)
5412 case BPF_MAP_TYPE_QUEUE
:
5413 case BPF_MAP_TYPE_STACK
:
5414 if (func_id
!= BPF_FUNC_map_peek_elem
&&
5415 func_id
!= BPF_FUNC_map_pop_elem
&&
5416 func_id
!= BPF_FUNC_map_push_elem
)
5419 case BPF_MAP_TYPE_SK_STORAGE
:
5420 if (func_id
!= BPF_FUNC_sk_storage_get
&&
5421 func_id
!= BPF_FUNC_sk_storage_delete
)
5424 case BPF_MAP_TYPE_INODE_STORAGE
:
5425 if (func_id
!= BPF_FUNC_inode_storage_get
&&
5426 func_id
!= BPF_FUNC_inode_storage_delete
)
5429 case BPF_MAP_TYPE_TASK_STORAGE
:
5430 if (func_id
!= BPF_FUNC_task_storage_get
&&
5431 func_id
!= BPF_FUNC_task_storage_delete
)
5438 /* ... and second from the function itself. */
5440 case BPF_FUNC_tail_call
:
5441 if (map
->map_type
!= BPF_MAP_TYPE_PROG_ARRAY
)
5443 if (env
->subprog_cnt
> 1 && !allow_tail_call_in_subprogs(env
)) {
5444 verbose(env
, "tail_calls are not allowed in non-JITed programs with bpf-to-bpf calls\n");
5448 case BPF_FUNC_perf_event_read
:
5449 case BPF_FUNC_perf_event_output
:
5450 case BPF_FUNC_perf_event_read_value
:
5451 case BPF_FUNC_skb_output
:
5452 case BPF_FUNC_xdp_output
:
5453 if (map
->map_type
!= BPF_MAP_TYPE_PERF_EVENT_ARRAY
)
5456 case BPF_FUNC_ringbuf_output
:
5457 case BPF_FUNC_ringbuf_reserve
:
5458 case BPF_FUNC_ringbuf_query
:
5459 if (map
->map_type
!= BPF_MAP_TYPE_RINGBUF
)
5462 case BPF_FUNC_get_stackid
:
5463 if (map
->map_type
!= BPF_MAP_TYPE_STACK_TRACE
)
5466 case BPF_FUNC_current_task_under_cgroup
:
5467 case BPF_FUNC_skb_under_cgroup
:
5468 if (map
->map_type
!= BPF_MAP_TYPE_CGROUP_ARRAY
)
5471 case BPF_FUNC_redirect_map
:
5472 if (map
->map_type
!= BPF_MAP_TYPE_DEVMAP
&&
5473 map
->map_type
!= BPF_MAP_TYPE_DEVMAP_HASH
&&
5474 map
->map_type
!= BPF_MAP_TYPE_CPUMAP
&&
5475 map
->map_type
!= BPF_MAP_TYPE_XSKMAP
)
5478 case BPF_FUNC_sk_redirect_map
:
5479 case BPF_FUNC_msg_redirect_map
:
5480 case BPF_FUNC_sock_map_update
:
5481 if (map
->map_type
!= BPF_MAP_TYPE_SOCKMAP
)
5484 case BPF_FUNC_sk_redirect_hash
:
5485 case BPF_FUNC_msg_redirect_hash
:
5486 case BPF_FUNC_sock_hash_update
:
5487 if (map
->map_type
!= BPF_MAP_TYPE_SOCKHASH
)
5490 case BPF_FUNC_get_local_storage
:
5491 if (map
->map_type
!= BPF_MAP_TYPE_CGROUP_STORAGE
&&
5492 map
->map_type
!= BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE
)
5495 case BPF_FUNC_sk_select_reuseport
:
5496 if (map
->map_type
!= BPF_MAP_TYPE_REUSEPORT_SOCKARRAY
&&
5497 map
->map_type
!= BPF_MAP_TYPE_SOCKMAP
&&
5498 map
->map_type
!= BPF_MAP_TYPE_SOCKHASH
)
5501 case BPF_FUNC_map_peek_elem
:
5502 case BPF_FUNC_map_pop_elem
:
5503 case BPF_FUNC_map_push_elem
:
5504 if (map
->map_type
!= BPF_MAP_TYPE_QUEUE
&&
5505 map
->map_type
!= BPF_MAP_TYPE_STACK
)
5508 case BPF_FUNC_sk_storage_get
:
5509 case BPF_FUNC_sk_storage_delete
:
5510 if (map
->map_type
!= BPF_MAP_TYPE_SK_STORAGE
)
5513 case BPF_FUNC_inode_storage_get
:
5514 case BPF_FUNC_inode_storage_delete
:
5515 if (map
->map_type
!= BPF_MAP_TYPE_INODE_STORAGE
)
5518 case BPF_FUNC_task_storage_get
:
5519 case BPF_FUNC_task_storage_delete
:
5520 if (map
->map_type
!= BPF_MAP_TYPE_TASK_STORAGE
)
5529 verbose(env
, "cannot pass map_type %d into func %s#%d\n",
5530 map
->map_type
, func_id_name(func_id
), func_id
);
5534 static bool check_raw_mode_ok(const struct bpf_func_proto
*fn
)
5538 if (fn
->arg1_type
== ARG_PTR_TO_UNINIT_MEM
)
5540 if (fn
->arg2_type
== ARG_PTR_TO_UNINIT_MEM
)
5542 if (fn
->arg3_type
== ARG_PTR_TO_UNINIT_MEM
)
5544 if (fn
->arg4_type
== ARG_PTR_TO_UNINIT_MEM
)
5546 if (fn
->arg5_type
== ARG_PTR_TO_UNINIT_MEM
)
5549 /* We only support one arg being in raw mode at the moment,
5550 * which is sufficient for the helper functions we have
5556 static bool check_args_pair_invalid(enum bpf_arg_type arg_curr
,
5557 enum bpf_arg_type arg_next
)
5559 return (arg_type_is_mem_ptr(arg_curr
) &&
5560 !arg_type_is_mem_size(arg_next
)) ||
5561 (!arg_type_is_mem_ptr(arg_curr
) &&
5562 arg_type_is_mem_size(arg_next
));
5565 static bool check_arg_pair_ok(const struct bpf_func_proto
*fn
)
5567 /* bpf_xxx(..., buf, len) call will access 'len'
5568 * bytes from memory 'buf'. Both arg types need
5569 * to be paired, so make sure there's no buggy
5570 * helper function specification.
5572 if (arg_type_is_mem_size(fn
->arg1_type
) ||
5573 arg_type_is_mem_ptr(fn
->arg5_type
) ||
5574 check_args_pair_invalid(fn
->arg1_type
, fn
->arg2_type
) ||
5575 check_args_pair_invalid(fn
->arg2_type
, fn
->arg3_type
) ||
5576 check_args_pair_invalid(fn
->arg3_type
, fn
->arg4_type
) ||
5577 check_args_pair_invalid(fn
->arg4_type
, fn
->arg5_type
))
5583 static bool check_refcount_ok(const struct bpf_func_proto
*fn
, int func_id
)
5587 if (arg_type_may_be_refcounted(fn
->arg1_type
))
5589 if (arg_type_may_be_refcounted(fn
->arg2_type
))
5591 if (arg_type_may_be_refcounted(fn
->arg3_type
))
5593 if (arg_type_may_be_refcounted(fn
->arg4_type
))
5595 if (arg_type_may_be_refcounted(fn
->arg5_type
))
5598 /* A reference acquiring function cannot acquire
5599 * another refcounted ptr.
5601 if (may_be_acquire_function(func_id
) && count
)
5604 /* We only support one arg being unreferenced at the moment,
5605 * which is sufficient for the helper functions we have right now.
5610 static bool check_btf_id_ok(const struct bpf_func_proto
*fn
)
5614 for (i
= 0; i
< ARRAY_SIZE(fn
->arg_type
); i
++) {
5615 if (fn
->arg_type
[i
] == ARG_PTR_TO_BTF_ID
&& !fn
->arg_btf_id
[i
])
5618 if (fn
->arg_type
[i
] != ARG_PTR_TO_BTF_ID
&& fn
->arg_btf_id
[i
])
5625 static int check_func_proto(const struct bpf_func_proto
*fn
, int func_id
)
5627 return check_raw_mode_ok(fn
) &&
5628 check_arg_pair_ok(fn
) &&
5629 check_btf_id_ok(fn
) &&
5630 check_refcount_ok(fn
, func_id
) ? 0 : -EINVAL
;
5633 /* Packet data might have moved, any old PTR_TO_PACKET[_META,_END]
5634 * are now invalid, so turn them into unknown SCALAR_VALUE.
5636 static void __clear_all_pkt_pointers(struct bpf_verifier_env
*env
,
5637 struct bpf_func_state
*state
)
5639 struct bpf_reg_state
*regs
= state
->regs
, *reg
;
5642 for (i
= 0; i
< MAX_BPF_REG
; i
++)
5643 if (reg_is_pkt_pointer_any(®s
[i
]))
5644 mark_reg_unknown(env
, regs
, i
);
5646 bpf_for_each_spilled_reg(i
, state
, reg
) {
5649 if (reg_is_pkt_pointer_any(reg
))
5650 __mark_reg_unknown(env
, reg
);
5654 static void clear_all_pkt_pointers(struct bpf_verifier_env
*env
)
5656 struct bpf_verifier_state
*vstate
= env
->cur_state
;
5659 for (i
= 0; i
<= vstate
->curframe
; i
++)
5660 __clear_all_pkt_pointers(env
, vstate
->frame
[i
]);
5665 BEYOND_PKT_END
= -2,
5668 static void mark_pkt_end(struct bpf_verifier_state
*vstate
, int regn
, bool range_open
)
5670 struct bpf_func_state
*state
= vstate
->frame
[vstate
->curframe
];
5671 struct bpf_reg_state
*reg
= &state
->regs
[regn
];
5673 if (reg
->type
!= PTR_TO_PACKET
)
5674 /* PTR_TO_PACKET_META is not supported yet */
5677 /* The 'reg' is pkt > pkt_end or pkt >= pkt_end.
5678 * How far beyond pkt_end it goes is unknown.
5679 * if (!range_open) it's the case of pkt >= pkt_end
5680 * if (range_open) it's the case of pkt > pkt_end
5681 * hence this pointer is at least 1 byte bigger than pkt_end
5684 reg
->range
= BEYOND_PKT_END
;
5686 reg
->range
= AT_PKT_END
;
5689 static void release_reg_references(struct bpf_verifier_env
*env
,
5690 struct bpf_func_state
*state
,
5693 struct bpf_reg_state
*regs
= state
->regs
, *reg
;
5696 for (i
= 0; i
< MAX_BPF_REG
; i
++)
5697 if (regs
[i
].ref_obj_id
== ref_obj_id
)
5698 mark_reg_unknown(env
, regs
, i
);
5700 bpf_for_each_spilled_reg(i
, state
, reg
) {
5703 if (reg
->ref_obj_id
== ref_obj_id
)
5704 __mark_reg_unknown(env
, reg
);
5708 /* The pointer with the specified id has released its reference to kernel
5709 * resources. Identify all copies of the same pointer and clear the reference.
5711 static int release_reference(struct bpf_verifier_env
*env
,
5714 struct bpf_verifier_state
*vstate
= env
->cur_state
;
5718 err
= release_reference_state(cur_func(env
), ref_obj_id
);
5722 for (i
= 0; i
<= vstate
->curframe
; i
++)
5723 release_reg_references(env
, vstate
->frame
[i
], ref_obj_id
);
5728 static void clear_caller_saved_regs(struct bpf_verifier_env
*env
,
5729 struct bpf_reg_state
*regs
)
5733 /* after the call registers r0 - r5 were scratched */
5734 for (i
= 0; i
< CALLER_SAVED_REGS
; i
++) {
5735 mark_reg_not_init(env
, regs
, caller_saved
[i
]);
5736 check_reg_arg(env
, caller_saved
[i
], DST_OP_NO_MARK
);
5740 typedef int (*set_callee_state_fn
)(struct bpf_verifier_env
*env
,
5741 struct bpf_func_state
*caller
,
5742 struct bpf_func_state
*callee
,
5745 static int __check_func_call(struct bpf_verifier_env
*env
, struct bpf_insn
*insn
,
5746 int *insn_idx
, int subprog
,
5747 set_callee_state_fn set_callee_state_cb
)
5749 struct bpf_verifier_state
*state
= env
->cur_state
;
5750 struct bpf_func_info_aux
*func_info_aux
;
5751 struct bpf_func_state
*caller
, *callee
;
5753 bool is_global
= false;
5755 if (state
->curframe
+ 1 >= MAX_CALL_FRAMES
) {
5756 verbose(env
, "the call stack of %d frames is too deep\n",
5757 state
->curframe
+ 2);
5761 caller
= state
->frame
[state
->curframe
];
5762 if (state
->frame
[state
->curframe
+ 1]) {
5763 verbose(env
, "verifier bug. Frame %d already allocated\n",
5764 state
->curframe
+ 1);
5768 func_info_aux
= env
->prog
->aux
->func_info_aux
;
5770 is_global
= func_info_aux
[subprog
].linkage
== BTF_FUNC_GLOBAL
;
5771 err
= btf_check_subprog_arg_match(env
, subprog
, caller
->regs
);
5776 verbose(env
, "Caller passes invalid args into func#%d\n",
5780 if (env
->log
.level
& BPF_LOG_LEVEL
)
5782 "Func#%d is global and valid. Skipping.\n",
5784 clear_caller_saved_regs(env
, caller
->regs
);
5786 /* All global functions return a 64-bit SCALAR_VALUE */
5787 mark_reg_unknown(env
, caller
->regs
, BPF_REG_0
);
5788 caller
->regs
[BPF_REG_0
].subreg_def
= DEF_NOT_SUBREG
;
5790 /* continue with next insn after call */
5795 if (insn
->code
== (BPF_JMP
| BPF_CALL
) &&
5796 insn
->src_reg
== 0 &&
5797 insn
->imm
== BPF_FUNC_timer_set_callback
) {
5798 struct bpf_verifier_state
*async_cb
;
5800 /* there is no real recursion here. timer callbacks are async */
5801 env
->subprog_info
[subprog
].is_async_cb
= true;
5802 async_cb
= push_async_cb(env
, env
->subprog_info
[subprog
].start
,
5803 *insn_idx
, subprog
);
5806 callee
= async_cb
->frame
[0];
5807 callee
->async_entry_cnt
= caller
->async_entry_cnt
+ 1;
5809 /* Convert bpf_timer_set_callback() args into timer callback args */
5810 err
= set_callee_state_cb(env
, caller
, callee
, *insn_idx
);
5814 clear_caller_saved_regs(env
, caller
->regs
);
5815 mark_reg_unknown(env
, caller
->regs
, BPF_REG_0
);
5816 caller
->regs
[BPF_REG_0
].subreg_def
= DEF_NOT_SUBREG
;
5817 /* continue with next insn after call */
5821 callee
= kzalloc(sizeof(*callee
), GFP_KERNEL
);
5824 state
->frame
[state
->curframe
+ 1] = callee
;
5826 /* callee cannot access r0, r6 - r9 for reading and has to write
5827 * into its own stack before reading from it.
5828 * callee can read/write into caller's stack
5830 init_func_state(env
, callee
,
5831 /* remember the callsite, it will be used by bpf_exit */
5832 *insn_idx
/* callsite */,
5833 state
->curframe
+ 1 /* frameno within this callchain */,
5834 subprog
/* subprog number within this prog */);
5836 /* Transfer references to the callee */
5837 err
= copy_reference_state(callee
, caller
);
5841 err
= set_callee_state_cb(env
, caller
, callee
, *insn_idx
);
5845 clear_caller_saved_regs(env
, caller
->regs
);
5847 /* only increment it after check_reg_arg() finished */
5850 /* and go analyze first insn of the callee */
5851 *insn_idx
= env
->subprog_info
[subprog
].start
- 1;
5853 if (env
->log
.level
& BPF_LOG_LEVEL
) {
5854 verbose(env
, "caller:\n");
5855 print_verifier_state(env
, caller
);
5856 verbose(env
, "callee:\n");
5857 print_verifier_state(env
, callee
);
5862 int map_set_for_each_callback_args(struct bpf_verifier_env
*env
,
5863 struct bpf_func_state
*caller
,
5864 struct bpf_func_state
*callee
)
5866 /* bpf_for_each_map_elem(struct bpf_map *map, void *callback_fn,
5867 * void *callback_ctx, u64 flags);
5868 * callback_fn(struct bpf_map *map, void *key, void *value,
5869 * void *callback_ctx);
5871 callee
->regs
[BPF_REG_1
] = caller
->regs
[BPF_REG_1
];
5873 callee
->regs
[BPF_REG_2
].type
= PTR_TO_MAP_KEY
;
5874 __mark_reg_known_zero(&callee
->regs
[BPF_REG_2
]);
5875 callee
->regs
[BPF_REG_2
].map_ptr
= caller
->regs
[BPF_REG_1
].map_ptr
;
5877 callee
->regs
[BPF_REG_3
].type
= PTR_TO_MAP_VALUE
;
5878 __mark_reg_known_zero(&callee
->regs
[BPF_REG_3
]);
5879 callee
->regs
[BPF_REG_3
].map_ptr
= caller
->regs
[BPF_REG_1
].map_ptr
;
5881 /* pointer to stack or null */
5882 callee
->regs
[BPF_REG_4
] = caller
->regs
[BPF_REG_3
];
5885 __mark_reg_not_init(env
, &callee
->regs
[BPF_REG_5
]);
5889 static int set_callee_state(struct bpf_verifier_env
*env
,
5890 struct bpf_func_state
*caller
,
5891 struct bpf_func_state
*callee
, int insn_idx
)
5895 /* copy r1 - r5 args that callee can access. The copy includes parent
5896 * pointers, which connects us up to the liveness chain
5898 for (i
= BPF_REG_1
; i
<= BPF_REG_5
; i
++)
5899 callee
->regs
[i
] = caller
->regs
[i
];
5903 static int check_func_call(struct bpf_verifier_env
*env
, struct bpf_insn
*insn
,
5906 int subprog
, target_insn
;
5908 target_insn
= *insn_idx
+ insn
->imm
+ 1;
5909 subprog
= find_subprog(env
, target_insn
);
5911 verbose(env
, "verifier bug. No program starts at insn %d\n",
5916 return __check_func_call(env
, insn
, insn_idx
, subprog
, set_callee_state
);
5919 static int set_map_elem_callback_state(struct bpf_verifier_env
*env
,
5920 struct bpf_func_state
*caller
,
5921 struct bpf_func_state
*callee
,
5924 struct bpf_insn_aux_data
*insn_aux
= &env
->insn_aux_data
[insn_idx
];
5925 struct bpf_map
*map
;
5928 if (bpf_map_ptr_poisoned(insn_aux
)) {
5929 verbose(env
, "tail_call abusing map_ptr\n");
5933 map
= BPF_MAP_PTR(insn_aux
->map_ptr_state
);
5934 if (!map
->ops
->map_set_for_each_callback_args
||
5935 !map
->ops
->map_for_each_callback
) {
5936 verbose(env
, "callback function not allowed for map\n");
5940 err
= map
->ops
->map_set_for_each_callback_args(env
, caller
, callee
);
5944 callee
->in_callback_fn
= true;
5948 static int set_timer_callback_state(struct bpf_verifier_env
*env
,
5949 struct bpf_func_state
*caller
,
5950 struct bpf_func_state
*callee
,
5953 struct bpf_map
*map_ptr
= caller
->regs
[BPF_REG_1
].map_ptr
;
5955 /* bpf_timer_set_callback(struct bpf_timer *timer, void *callback_fn);
5956 * callback_fn(struct bpf_map *map, void *key, void *value);
5958 callee
->regs
[BPF_REG_1
].type
= CONST_PTR_TO_MAP
;
5959 __mark_reg_known_zero(&callee
->regs
[BPF_REG_1
]);
5960 callee
->regs
[BPF_REG_1
].map_ptr
= map_ptr
;
5962 callee
->regs
[BPF_REG_2
].type
= PTR_TO_MAP_KEY
;
5963 __mark_reg_known_zero(&callee
->regs
[BPF_REG_2
]);
5964 callee
->regs
[BPF_REG_2
].map_ptr
= map_ptr
;
5966 callee
->regs
[BPF_REG_3
].type
= PTR_TO_MAP_VALUE
;
5967 __mark_reg_known_zero(&callee
->regs
[BPF_REG_3
]);
5968 callee
->regs
[BPF_REG_3
].map_ptr
= map_ptr
;
5971 __mark_reg_not_init(env
, &callee
->regs
[BPF_REG_4
]);
5972 __mark_reg_not_init(env
, &callee
->regs
[BPF_REG_5
]);
5973 callee
->in_async_callback_fn
= true;
5977 static int prepare_func_exit(struct bpf_verifier_env
*env
, int *insn_idx
)
5979 struct bpf_verifier_state
*state
= env
->cur_state
;
5980 struct bpf_func_state
*caller
, *callee
;
5981 struct bpf_reg_state
*r0
;
5984 callee
= state
->frame
[state
->curframe
];
5985 r0
= &callee
->regs
[BPF_REG_0
];
5986 if (r0
->type
== PTR_TO_STACK
) {
5987 /* technically it's ok to return caller's stack pointer
5988 * (or caller's caller's pointer) back to the caller,
5989 * since these pointers are valid. Only current stack
5990 * pointer will be invalid as soon as function exits,
5991 * but let's be conservative
5993 verbose(env
, "cannot return stack pointer to the caller\n");
5998 caller
= state
->frame
[state
->curframe
];
5999 if (callee
->in_callback_fn
) {
6000 /* enforce R0 return value range [0, 1]. */
6001 struct tnum range
= tnum_range(0, 1);
6003 if (r0
->type
!= SCALAR_VALUE
) {
6004 verbose(env
, "R0 not a scalar value\n");
6007 if (!tnum_in(range
, r0
->var_off
)) {
6008 verbose_invalid_scalar(env
, r0
, &range
, "callback return", "R0");
6012 /* return to the caller whatever r0 had in the callee */
6013 caller
->regs
[BPF_REG_0
] = *r0
;
6016 /* Transfer references to the caller */
6017 err
= copy_reference_state(caller
, callee
);
6021 *insn_idx
= callee
->callsite
+ 1;
6022 if (env
->log
.level
& BPF_LOG_LEVEL
) {
6023 verbose(env
, "returning from callee:\n");
6024 print_verifier_state(env
, callee
);
6025 verbose(env
, "to caller at %d:\n", *insn_idx
);
6026 print_verifier_state(env
, caller
);
6028 /* clear everything in the callee */
6029 free_func_state(callee
);
6030 state
->frame
[state
->curframe
+ 1] = NULL
;
6034 static void do_refine_retval_range(struct bpf_reg_state
*regs
, int ret_type
,
6036 struct bpf_call_arg_meta
*meta
)
6038 struct bpf_reg_state
*ret_reg
= ®s
[BPF_REG_0
];
6040 if (ret_type
!= RET_INTEGER
||
6041 (func_id
!= BPF_FUNC_get_stack
&&
6042 func_id
!= BPF_FUNC_get_task_stack
&&
6043 func_id
!= BPF_FUNC_probe_read_str
&&
6044 func_id
!= BPF_FUNC_probe_read_kernel_str
&&
6045 func_id
!= BPF_FUNC_probe_read_user_str
))
6048 ret_reg
->smax_value
= meta
->msize_max_value
;
6049 ret_reg
->s32_max_value
= meta
->msize_max_value
;
6050 ret_reg
->smin_value
= -MAX_ERRNO
;
6051 ret_reg
->s32_min_value
= -MAX_ERRNO
;
6052 __reg_deduce_bounds(ret_reg
);
6053 __reg_bound_offset(ret_reg
);
6054 __update_reg_bounds(ret_reg
);
6058 record_func_map(struct bpf_verifier_env
*env
, struct bpf_call_arg_meta
*meta
,
6059 int func_id
, int insn_idx
)
6061 struct bpf_insn_aux_data
*aux
= &env
->insn_aux_data
[insn_idx
];
6062 struct bpf_map
*map
= meta
->map_ptr
;
6064 if (func_id
!= BPF_FUNC_tail_call
&&
6065 func_id
!= BPF_FUNC_map_lookup_elem
&&
6066 func_id
!= BPF_FUNC_map_update_elem
&&
6067 func_id
!= BPF_FUNC_map_delete_elem
&&
6068 func_id
!= BPF_FUNC_map_push_elem
&&
6069 func_id
!= BPF_FUNC_map_pop_elem
&&
6070 func_id
!= BPF_FUNC_map_peek_elem
&&
6071 func_id
!= BPF_FUNC_for_each_map_elem
&&
6072 func_id
!= BPF_FUNC_redirect_map
)
6076 verbose(env
, "kernel subsystem misconfigured verifier\n");
6080 /* In case of read-only, some additional restrictions
6081 * need to be applied in order to prevent altering the
6082 * state of the map from program side.
6084 if ((map
->map_flags
& BPF_F_RDONLY_PROG
) &&
6085 (func_id
== BPF_FUNC_map_delete_elem
||
6086 func_id
== BPF_FUNC_map_update_elem
||
6087 func_id
== BPF_FUNC_map_push_elem
||
6088 func_id
== BPF_FUNC_map_pop_elem
)) {
6089 verbose(env
, "write into map forbidden\n");
6093 if (!BPF_MAP_PTR(aux
->map_ptr_state
))
6094 bpf_map_ptr_store(aux
, meta
->map_ptr
,
6095 !meta
->map_ptr
->bypass_spec_v1
);
6096 else if (BPF_MAP_PTR(aux
->map_ptr_state
) != meta
->map_ptr
)
6097 bpf_map_ptr_store(aux
, BPF_MAP_PTR_POISON
,
6098 !meta
->map_ptr
->bypass_spec_v1
);
6103 record_func_key(struct bpf_verifier_env
*env
, struct bpf_call_arg_meta
*meta
,
6104 int func_id
, int insn_idx
)
6106 struct bpf_insn_aux_data
*aux
= &env
->insn_aux_data
[insn_idx
];
6107 struct bpf_reg_state
*regs
= cur_regs(env
), *reg
;
6108 struct bpf_map
*map
= meta
->map_ptr
;
6113 if (func_id
!= BPF_FUNC_tail_call
)
6115 if (!map
|| map
->map_type
!= BPF_MAP_TYPE_PROG_ARRAY
) {
6116 verbose(env
, "kernel subsystem misconfigured verifier\n");
6120 range
= tnum_range(0, map
->max_entries
- 1);
6121 reg
= ®s
[BPF_REG_3
];
6123 if (!register_is_const(reg
) || !tnum_in(range
, reg
->var_off
)) {
6124 bpf_map_key_store(aux
, BPF_MAP_KEY_POISON
);
6128 err
= mark_chain_precision(env
, BPF_REG_3
);
6132 val
= reg
->var_off
.value
;
6133 if (bpf_map_key_unseen(aux
))
6134 bpf_map_key_store(aux
, val
);
6135 else if (!bpf_map_key_poisoned(aux
) &&
6136 bpf_map_key_immediate(aux
) != val
)
6137 bpf_map_key_store(aux
, BPF_MAP_KEY_POISON
);
6141 static int check_reference_leak(struct bpf_verifier_env
*env
)
6143 struct bpf_func_state
*state
= cur_func(env
);
6146 for (i
= 0; i
< state
->acquired_refs
; i
++) {
6147 verbose(env
, "Unreleased reference id=%d alloc_insn=%d\n",
6148 state
->refs
[i
].id
, state
->refs
[i
].insn_idx
);
6150 return state
->acquired_refs
? -EINVAL
: 0;
6153 static int check_bpf_snprintf_call(struct bpf_verifier_env
*env
,
6154 struct bpf_reg_state
*regs
)
6156 struct bpf_reg_state
*fmt_reg
= ®s
[BPF_REG_3
];
6157 struct bpf_reg_state
*data_len_reg
= ®s
[BPF_REG_5
];
6158 struct bpf_map
*fmt_map
= fmt_reg
->map_ptr
;
6159 int err
, fmt_map_off
, num_args
;
6163 /* data must be an array of u64 */
6164 if (data_len_reg
->var_off
.value
% 8)
6166 num_args
= data_len_reg
->var_off
.value
/ 8;
6168 /* fmt being ARG_PTR_TO_CONST_STR guarantees that var_off is const
6169 * and map_direct_value_addr is set.
6171 fmt_map_off
= fmt_reg
->off
+ fmt_reg
->var_off
.value
;
6172 err
= fmt_map
->ops
->map_direct_value_addr(fmt_map
, &fmt_addr
,
6175 verbose(env
, "verifier bug\n");
6178 fmt
= (char *)(long)fmt_addr
+ fmt_map_off
;
6180 /* We are also guaranteed that fmt+fmt_map_off is NULL terminated, we
6181 * can focus on validating the format specifiers.
6183 err
= bpf_bprintf_prepare(fmt
, UINT_MAX
, NULL
, NULL
, num_args
);
6185 verbose(env
, "Invalid format string\n");
6190 static int check_get_func_ip(struct bpf_verifier_env
*env
)
6192 enum bpf_attach_type eatype
= env
->prog
->expected_attach_type
;
6193 enum bpf_prog_type type
= resolve_prog_type(env
->prog
);
6194 int func_id
= BPF_FUNC_get_func_ip
;
6196 if (type
== BPF_PROG_TYPE_TRACING
) {
6197 if (eatype
!= BPF_TRACE_FENTRY
&& eatype
!= BPF_TRACE_FEXIT
&&
6198 eatype
!= BPF_MODIFY_RETURN
) {
6199 verbose(env
, "func %s#%d supported only for fentry/fexit/fmod_ret programs\n",
6200 func_id_name(func_id
), func_id
);
6204 } else if (type
== BPF_PROG_TYPE_KPROBE
) {
6208 verbose(env
, "func %s#%d not supported for program type %d\n",
6209 func_id_name(func_id
), func_id
, type
);
6213 static int check_helper_call(struct bpf_verifier_env
*env
, struct bpf_insn
*insn
,
6216 const struct bpf_func_proto
*fn
= NULL
;
6217 struct bpf_reg_state
*regs
;
6218 struct bpf_call_arg_meta meta
;
6219 int insn_idx
= *insn_idx_p
;
6221 int i
, err
, func_id
;
6223 /* find function prototype */
6224 func_id
= insn
->imm
;
6225 if (func_id
< 0 || func_id
>= __BPF_FUNC_MAX_ID
) {
6226 verbose(env
, "invalid func %s#%d\n", func_id_name(func_id
),
6231 if (env
->ops
->get_func_proto
)
6232 fn
= env
->ops
->get_func_proto(func_id
, env
->prog
);
6234 verbose(env
, "unknown func %s#%d\n", func_id_name(func_id
),
6239 /* eBPF programs must be GPL compatible to use GPL-ed functions */
6240 if (!env
->prog
->gpl_compatible
&& fn
->gpl_only
) {
6241 verbose(env
, "cannot call GPL-restricted function from non-GPL compatible program\n");
6245 if (fn
->allowed
&& !fn
->allowed(env
->prog
)) {
6246 verbose(env
, "helper call is not allowed in probe\n");
6250 /* With LD_ABS/IND some JITs save/restore skb from r1. */
6251 changes_data
= bpf_helper_changes_pkt_data(fn
->func
);
6252 if (changes_data
&& fn
->arg1_type
!= ARG_PTR_TO_CTX
) {
6253 verbose(env
, "kernel subsystem misconfigured func %s#%d: r1 != ctx\n",
6254 func_id_name(func_id
), func_id
);
6258 memset(&meta
, 0, sizeof(meta
));
6259 meta
.pkt_access
= fn
->pkt_access
;
6261 err
= check_func_proto(fn
, func_id
);
6263 verbose(env
, "kernel subsystem misconfigured func %s#%d\n",
6264 func_id_name(func_id
), func_id
);
6268 meta
.func_id
= func_id
;
6270 for (i
= 0; i
< MAX_BPF_FUNC_REG_ARGS
; i
++) {
6271 err
= check_func_arg(env
, i
, &meta
, fn
);
6276 err
= record_func_map(env
, &meta
, func_id
, insn_idx
);
6280 err
= record_func_key(env
, &meta
, func_id
, insn_idx
);
6284 /* Mark slots with STACK_MISC in case of raw mode, stack offset
6285 * is inferred from register state.
6287 for (i
= 0; i
< meta
.access_size
; i
++) {
6288 err
= check_mem_access(env
, insn_idx
, meta
.regno
, i
, BPF_B
,
6289 BPF_WRITE
, -1, false);
6294 if (func_id
== BPF_FUNC_tail_call
) {
6295 err
= check_reference_leak(env
);
6297 verbose(env
, "tail_call would lead to reference leak\n");
6300 } else if (is_release_function(func_id
)) {
6301 err
= release_reference(env
, meta
.ref_obj_id
);
6303 verbose(env
, "func %s#%d reference has not been acquired before\n",
6304 func_id_name(func_id
), func_id
);
6309 regs
= cur_regs(env
);
6311 /* check that flags argument in get_local_storage(map, flags) is 0,
6312 * this is required because get_local_storage() can't return an error.
6314 if (func_id
== BPF_FUNC_get_local_storage
&&
6315 !register_is_null(®s
[BPF_REG_2
])) {
6316 verbose(env
, "get_local_storage() doesn't support non-zero flags\n");
6320 if (func_id
== BPF_FUNC_for_each_map_elem
) {
6321 err
= __check_func_call(env
, insn
, insn_idx_p
, meta
.subprogno
,
6322 set_map_elem_callback_state
);
6327 if (func_id
== BPF_FUNC_timer_set_callback
) {
6328 err
= __check_func_call(env
, insn
, insn_idx_p
, meta
.subprogno
,
6329 set_timer_callback_state
);
6334 if (func_id
== BPF_FUNC_snprintf
) {
6335 err
= check_bpf_snprintf_call(env
, regs
);
6340 /* reset caller saved regs */
6341 for (i
= 0; i
< CALLER_SAVED_REGS
; i
++) {
6342 mark_reg_not_init(env
, regs
, caller_saved
[i
]);
6343 check_reg_arg(env
, caller_saved
[i
], DST_OP_NO_MARK
);
6346 /* helper call returns 64-bit value. */
6347 regs
[BPF_REG_0
].subreg_def
= DEF_NOT_SUBREG
;
6349 /* update return register (already marked as written above) */
6350 if (fn
->ret_type
== RET_INTEGER
) {
6351 /* sets type to SCALAR_VALUE */
6352 mark_reg_unknown(env
, regs
, BPF_REG_0
);
6353 } else if (fn
->ret_type
== RET_VOID
) {
6354 regs
[BPF_REG_0
].type
= NOT_INIT
;
6355 } else if (fn
->ret_type
== RET_PTR_TO_MAP_VALUE_OR_NULL
||
6356 fn
->ret_type
== RET_PTR_TO_MAP_VALUE
) {
6357 /* There is no offset yet applied, variable or fixed */
6358 mark_reg_known_zero(env
, regs
, BPF_REG_0
);
6359 /* remember map_ptr, so that check_map_access()
6360 * can check 'value_size' boundary of memory access
6361 * to map element returned from bpf_map_lookup_elem()
6363 if (meta
.map_ptr
== NULL
) {
6365 "kernel subsystem misconfigured verifier\n");
6368 regs
[BPF_REG_0
].map_ptr
= meta
.map_ptr
;
6369 regs
[BPF_REG_0
].map_uid
= meta
.map_uid
;
6370 if (fn
->ret_type
== RET_PTR_TO_MAP_VALUE
) {
6371 regs
[BPF_REG_0
].type
= PTR_TO_MAP_VALUE
;
6372 if (map_value_has_spin_lock(meta
.map_ptr
))
6373 regs
[BPF_REG_0
].id
= ++env
->id_gen
;
6375 regs
[BPF_REG_0
].type
= PTR_TO_MAP_VALUE_OR_NULL
;
6377 } else if (fn
->ret_type
== RET_PTR_TO_SOCKET_OR_NULL
) {
6378 mark_reg_known_zero(env
, regs
, BPF_REG_0
);
6379 regs
[BPF_REG_0
].type
= PTR_TO_SOCKET_OR_NULL
;
6380 } else if (fn
->ret_type
== RET_PTR_TO_SOCK_COMMON_OR_NULL
) {
6381 mark_reg_known_zero(env
, regs
, BPF_REG_0
);
6382 regs
[BPF_REG_0
].type
= PTR_TO_SOCK_COMMON_OR_NULL
;
6383 } else if (fn
->ret_type
== RET_PTR_TO_TCP_SOCK_OR_NULL
) {
6384 mark_reg_known_zero(env
, regs
, BPF_REG_0
);
6385 regs
[BPF_REG_0
].type
= PTR_TO_TCP_SOCK_OR_NULL
;
6386 } else if (fn
->ret_type
== RET_PTR_TO_ALLOC_MEM_OR_NULL
) {
6387 mark_reg_known_zero(env
, regs
, BPF_REG_0
);
6388 regs
[BPF_REG_0
].type
= PTR_TO_MEM_OR_NULL
;
6389 regs
[BPF_REG_0
].mem_size
= meta
.mem_size
;
6390 } else if (fn
->ret_type
== RET_PTR_TO_MEM_OR_BTF_ID_OR_NULL
||
6391 fn
->ret_type
== RET_PTR_TO_MEM_OR_BTF_ID
) {
6392 const struct btf_type
*t
;
6394 mark_reg_known_zero(env
, regs
, BPF_REG_0
);
6395 t
= btf_type_skip_modifiers(meta
.ret_btf
, meta
.ret_btf_id
, NULL
);
6396 if (!btf_type_is_struct(t
)) {
6398 const struct btf_type
*ret
;
6401 /* resolve the type size of ksym. */
6402 ret
= btf_resolve_size(meta
.ret_btf
, t
, &tsize
);
6404 tname
= btf_name_by_offset(meta
.ret_btf
, t
->name_off
);
6405 verbose(env
, "unable to resolve the size of type '%s': %ld\n",
6406 tname
, PTR_ERR(ret
));
6409 regs
[BPF_REG_0
].type
=
6410 fn
->ret_type
== RET_PTR_TO_MEM_OR_BTF_ID
?
6411 PTR_TO_MEM
: PTR_TO_MEM_OR_NULL
;
6412 regs
[BPF_REG_0
].mem_size
= tsize
;
6414 regs
[BPF_REG_0
].type
=
6415 fn
->ret_type
== RET_PTR_TO_MEM_OR_BTF_ID
?
6416 PTR_TO_BTF_ID
: PTR_TO_BTF_ID_OR_NULL
;
6417 regs
[BPF_REG_0
].btf
= meta
.ret_btf
;
6418 regs
[BPF_REG_0
].btf_id
= meta
.ret_btf_id
;
6420 } else if (fn
->ret_type
== RET_PTR_TO_BTF_ID_OR_NULL
||
6421 fn
->ret_type
== RET_PTR_TO_BTF_ID
) {
6424 mark_reg_known_zero(env
, regs
, BPF_REG_0
);
6425 regs
[BPF_REG_0
].type
= fn
->ret_type
== RET_PTR_TO_BTF_ID
?
6427 PTR_TO_BTF_ID_OR_NULL
;
6428 ret_btf_id
= *fn
->ret_btf_id
;
6429 if (ret_btf_id
== 0) {
6430 verbose(env
, "invalid return type %d of func %s#%d\n",
6431 fn
->ret_type
, func_id_name(func_id
), func_id
);
6434 /* current BPF helper definitions are only coming from
6435 * built-in code with type IDs from vmlinux BTF
6437 regs
[BPF_REG_0
].btf
= btf_vmlinux
;
6438 regs
[BPF_REG_0
].btf_id
= ret_btf_id
;
6440 verbose(env
, "unknown return type %d of func %s#%d\n",
6441 fn
->ret_type
, func_id_name(func_id
), func_id
);
6445 if (reg_type_may_be_null(regs
[BPF_REG_0
].type
))
6446 regs
[BPF_REG_0
].id
= ++env
->id_gen
;
6448 if (is_ptr_cast_function(func_id
)) {
6449 /* For release_reference() */
6450 regs
[BPF_REG_0
].ref_obj_id
= meta
.ref_obj_id
;
6451 } else if (is_acquire_function(func_id
, meta
.map_ptr
)) {
6452 int id
= acquire_reference_state(env
, insn_idx
);
6456 /* For mark_ptr_or_null_reg() */
6457 regs
[BPF_REG_0
].id
= id
;
6458 /* For release_reference() */
6459 regs
[BPF_REG_0
].ref_obj_id
= id
;
6462 do_refine_retval_range(regs
, fn
->ret_type
, func_id
, &meta
);
6464 err
= check_map_func_compatibility(env
, meta
.map_ptr
, func_id
);
6468 if ((func_id
== BPF_FUNC_get_stack
||
6469 func_id
== BPF_FUNC_get_task_stack
) &&
6470 !env
->prog
->has_callchain_buf
) {
6471 const char *err_str
;
6473 #ifdef CONFIG_PERF_EVENTS
6474 err
= get_callchain_buffers(sysctl_perf_event_max_stack
);
6475 err_str
= "cannot get callchain buffer for func %s#%d\n";
6478 err_str
= "func %s#%d not supported without CONFIG_PERF_EVENTS\n";
6481 verbose(env
, err_str
, func_id_name(func_id
), func_id
);
6485 env
->prog
->has_callchain_buf
= true;
6488 if (func_id
== BPF_FUNC_get_stackid
|| func_id
== BPF_FUNC_get_stack
)
6489 env
->prog
->call_get_stack
= true;
6491 if (func_id
== BPF_FUNC_get_func_ip
) {
6492 if (check_get_func_ip(env
))
6494 env
->prog
->call_get_func_ip
= true;
6498 clear_all_pkt_pointers(env
);
6502 /* mark_btf_func_reg_size() is used when the reg size is determined by
6503 * the BTF func_proto's return value size and argument.
6505 static void mark_btf_func_reg_size(struct bpf_verifier_env
*env
, u32 regno
,
6508 struct bpf_reg_state
*reg
= &cur_regs(env
)[regno
];
6510 if (regno
== BPF_REG_0
) {
6511 /* Function return value */
6512 reg
->live
|= REG_LIVE_WRITTEN
;
6513 reg
->subreg_def
= reg_size
== sizeof(u64
) ?
6514 DEF_NOT_SUBREG
: env
->insn_idx
+ 1;
6516 /* Function argument */
6517 if (reg_size
== sizeof(u64
)) {
6518 mark_insn_zext(env
, reg
);
6519 mark_reg_read(env
, reg
, reg
->parent
, REG_LIVE_READ64
);
6521 mark_reg_read(env
, reg
, reg
->parent
, REG_LIVE_READ32
);
6526 static int check_kfunc_call(struct bpf_verifier_env
*env
, struct bpf_insn
*insn
)
6528 const struct btf_type
*t
, *func
, *func_proto
, *ptr_type
;
6529 struct bpf_reg_state
*regs
= cur_regs(env
);
6530 const char *func_name
, *ptr_type_name
;
6531 u32 i
, nargs
, func_id
, ptr_type_id
;
6532 const struct btf_param
*args
;
6535 func_id
= insn
->imm
;
6536 func
= btf_type_by_id(btf_vmlinux
, func_id
);
6537 func_name
= btf_name_by_offset(btf_vmlinux
, func
->name_off
);
6538 func_proto
= btf_type_by_id(btf_vmlinux
, func
->type
);
6540 if (!env
->ops
->check_kfunc_call
||
6541 !env
->ops
->check_kfunc_call(func_id
)) {
6542 verbose(env
, "calling kernel function %s is not allowed\n",
6547 /* Check the arguments */
6548 err
= btf_check_kfunc_arg_match(env
, btf_vmlinux
, func_id
, regs
);
6552 for (i
= 0; i
< CALLER_SAVED_REGS
; i
++)
6553 mark_reg_not_init(env
, regs
, caller_saved
[i
]);
6555 /* Check return type */
6556 t
= btf_type_skip_modifiers(btf_vmlinux
, func_proto
->type
, NULL
);
6557 if (btf_type_is_scalar(t
)) {
6558 mark_reg_unknown(env
, regs
, BPF_REG_0
);
6559 mark_btf_func_reg_size(env
, BPF_REG_0
, t
->size
);
6560 } else if (btf_type_is_ptr(t
)) {
6561 ptr_type
= btf_type_skip_modifiers(btf_vmlinux
, t
->type
,
6563 if (!btf_type_is_struct(ptr_type
)) {
6564 ptr_type_name
= btf_name_by_offset(btf_vmlinux
,
6565 ptr_type
->name_off
);
6566 verbose(env
, "kernel function %s returns pointer type %s %s is not supported\n",
6567 func_name
, btf_type_str(ptr_type
),
6571 mark_reg_known_zero(env
, regs
, BPF_REG_0
);
6572 regs
[BPF_REG_0
].btf
= btf_vmlinux
;
6573 regs
[BPF_REG_0
].type
= PTR_TO_BTF_ID
;
6574 regs
[BPF_REG_0
].btf_id
= ptr_type_id
;
6575 mark_btf_func_reg_size(env
, BPF_REG_0
, sizeof(void *));
6576 } /* else { add_kfunc_call() ensures it is btf_type_is_void(t) } */
6578 nargs
= btf_type_vlen(func_proto
);
6579 args
= (const struct btf_param
*)(func_proto
+ 1);
6580 for (i
= 0; i
< nargs
; i
++) {
6583 t
= btf_type_skip_modifiers(btf_vmlinux
, args
[i
].type
, NULL
);
6584 if (btf_type_is_ptr(t
))
6585 mark_btf_func_reg_size(env
, regno
, sizeof(void *));
6587 /* scalar. ensured by btf_check_kfunc_arg_match() */
6588 mark_btf_func_reg_size(env
, regno
, t
->size
);
6594 static bool signed_add_overflows(s64 a
, s64 b
)
6596 /* Do the add in u64, where overflow is well-defined */
6597 s64 res
= (s64
)((u64
)a
+ (u64
)b
);
6604 static bool signed_add32_overflows(s32 a
, s32 b
)
6606 /* Do the add in u32, where overflow is well-defined */
6607 s32 res
= (s32
)((u32
)a
+ (u32
)b
);
6614 static bool signed_sub_overflows(s64 a
, s64 b
)
6616 /* Do the sub in u64, where overflow is well-defined */
6617 s64 res
= (s64
)((u64
)a
- (u64
)b
);
6624 static bool signed_sub32_overflows(s32 a
, s32 b
)
6626 /* Do the sub in u32, where overflow is well-defined */
6627 s32 res
= (s32
)((u32
)a
- (u32
)b
);
6634 static bool check_reg_sane_offset(struct bpf_verifier_env
*env
,
6635 const struct bpf_reg_state
*reg
,
6636 enum bpf_reg_type type
)
6638 bool known
= tnum_is_const(reg
->var_off
);
6639 s64 val
= reg
->var_off
.value
;
6640 s64 smin
= reg
->smin_value
;
6642 if (known
&& (val
>= BPF_MAX_VAR_OFF
|| val
<= -BPF_MAX_VAR_OFF
)) {
6643 verbose(env
, "math between %s pointer and %lld is not allowed\n",
6644 reg_type_str
[type
], val
);
6648 if (reg
->off
>= BPF_MAX_VAR_OFF
|| reg
->off
<= -BPF_MAX_VAR_OFF
) {
6649 verbose(env
, "%s pointer offset %d is not allowed\n",
6650 reg_type_str
[type
], reg
->off
);
6654 if (smin
== S64_MIN
) {
6655 verbose(env
, "math between %s pointer and register with unbounded min value is not allowed\n",
6656 reg_type_str
[type
]);
6660 if (smin
>= BPF_MAX_VAR_OFF
|| smin
<= -BPF_MAX_VAR_OFF
) {
6661 verbose(env
, "value %lld makes %s pointer be out of bounds\n",
6662 smin
, reg_type_str
[type
]);
6669 static struct bpf_insn_aux_data
*cur_aux(struct bpf_verifier_env
*env
)
6671 return &env
->insn_aux_data
[env
->insn_idx
];
6682 static int retrieve_ptr_limit(const struct bpf_reg_state
*ptr_reg
,
6683 u32
*alu_limit
, bool mask_to_left
)
6685 u32 max
= 0, ptr_limit
= 0;
6687 switch (ptr_reg
->type
) {
6689 /* Offset 0 is out-of-bounds, but acceptable start for the
6690 * left direction, see BPF_REG_FP. Also, unknown scalar
6691 * offset where we would need to deal with min/max bounds is
6692 * currently prohibited for unprivileged.
6694 max
= MAX_BPF_STACK
+ mask_to_left
;
6695 ptr_limit
= -(ptr_reg
->var_off
.value
+ ptr_reg
->off
);
6697 case PTR_TO_MAP_VALUE
:
6698 max
= ptr_reg
->map_ptr
->value_size
;
6699 ptr_limit
= (mask_to_left
?
6700 ptr_reg
->smin_value
:
6701 ptr_reg
->umax_value
) + ptr_reg
->off
;
6707 if (ptr_limit
>= max
)
6708 return REASON_LIMIT
;
6709 *alu_limit
= ptr_limit
;
6713 static bool can_skip_alu_sanitation(const struct bpf_verifier_env
*env
,
6714 const struct bpf_insn
*insn
)
6716 return env
->bypass_spec_v1
|| BPF_SRC(insn
->code
) == BPF_K
;
6719 static int update_alu_sanitation_state(struct bpf_insn_aux_data
*aux
,
6720 u32 alu_state
, u32 alu_limit
)
6722 /* If we arrived here from different branches with different
6723 * state or limits to sanitize, then this won't work.
6725 if (aux
->alu_state
&&
6726 (aux
->alu_state
!= alu_state
||
6727 aux
->alu_limit
!= alu_limit
))
6728 return REASON_PATHS
;
6730 /* Corresponding fixup done in do_misc_fixups(). */
6731 aux
->alu_state
= alu_state
;
6732 aux
->alu_limit
= alu_limit
;
6736 static int sanitize_val_alu(struct bpf_verifier_env
*env
,
6737 struct bpf_insn
*insn
)
6739 struct bpf_insn_aux_data
*aux
= cur_aux(env
);
6741 if (can_skip_alu_sanitation(env
, insn
))
6744 return update_alu_sanitation_state(aux
, BPF_ALU_NON_POINTER
, 0);
6747 static bool sanitize_needed(u8 opcode
)
6749 return opcode
== BPF_ADD
|| opcode
== BPF_SUB
;
6752 struct bpf_sanitize_info
{
6753 struct bpf_insn_aux_data aux
;
6757 static struct bpf_verifier_state
*
6758 sanitize_speculative_path(struct bpf_verifier_env
*env
,
6759 const struct bpf_insn
*insn
,
6760 u32 next_idx
, u32 curr_idx
)
6762 struct bpf_verifier_state
*branch
;
6763 struct bpf_reg_state
*regs
;
6765 branch
= push_stack(env
, next_idx
, curr_idx
, true);
6766 if (branch
&& insn
) {
6767 regs
= branch
->frame
[branch
->curframe
]->regs
;
6768 if (BPF_SRC(insn
->code
) == BPF_K
) {
6769 mark_reg_unknown(env
, regs
, insn
->dst_reg
);
6770 } else if (BPF_SRC(insn
->code
) == BPF_X
) {
6771 mark_reg_unknown(env
, regs
, insn
->dst_reg
);
6772 mark_reg_unknown(env
, regs
, insn
->src_reg
);
6778 static int sanitize_ptr_alu(struct bpf_verifier_env
*env
,
6779 struct bpf_insn
*insn
,
6780 const struct bpf_reg_state
*ptr_reg
,
6781 const struct bpf_reg_state
*off_reg
,
6782 struct bpf_reg_state
*dst_reg
,
6783 struct bpf_sanitize_info
*info
,
6784 const bool commit_window
)
6786 struct bpf_insn_aux_data
*aux
= commit_window
? cur_aux(env
) : &info
->aux
;
6787 struct bpf_verifier_state
*vstate
= env
->cur_state
;
6788 bool off_is_imm
= tnum_is_const(off_reg
->var_off
);
6789 bool off_is_neg
= off_reg
->smin_value
< 0;
6790 bool ptr_is_dst_reg
= ptr_reg
== dst_reg
;
6791 u8 opcode
= BPF_OP(insn
->code
);
6792 u32 alu_state
, alu_limit
;
6793 struct bpf_reg_state tmp
;
6797 if (can_skip_alu_sanitation(env
, insn
))
6800 /* We already marked aux for masking from non-speculative
6801 * paths, thus we got here in the first place. We only care
6802 * to explore bad access from here.
6804 if (vstate
->speculative
)
6807 if (!commit_window
) {
6808 if (!tnum_is_const(off_reg
->var_off
) &&
6809 (off_reg
->smin_value
< 0) != (off_reg
->smax_value
< 0))
6810 return REASON_BOUNDS
;
6812 info
->mask_to_left
= (opcode
== BPF_ADD
&& off_is_neg
) ||
6813 (opcode
== BPF_SUB
&& !off_is_neg
);
6816 err
= retrieve_ptr_limit(ptr_reg
, &alu_limit
, info
->mask_to_left
);
6820 if (commit_window
) {
6821 /* In commit phase we narrow the masking window based on
6822 * the observed pointer move after the simulated operation.
6824 alu_state
= info
->aux
.alu_state
;
6825 alu_limit
= abs(info
->aux
.alu_limit
- alu_limit
);
6827 alu_state
= off_is_neg
? BPF_ALU_NEG_VALUE
: 0;
6828 alu_state
|= off_is_imm
? BPF_ALU_IMMEDIATE
: 0;
6829 alu_state
|= ptr_is_dst_reg
?
6830 BPF_ALU_SANITIZE_SRC
: BPF_ALU_SANITIZE_DST
;
6832 /* Limit pruning on unknown scalars to enable deep search for
6833 * potential masking differences from other program paths.
6836 env
->explore_alu_limits
= true;
6839 err
= update_alu_sanitation_state(aux
, alu_state
, alu_limit
);
6843 /* If we're in commit phase, we're done here given we already
6844 * pushed the truncated dst_reg into the speculative verification
6847 * Also, when register is a known constant, we rewrite register-based
6848 * operation to immediate-based, and thus do not need masking (and as
6849 * a consequence, do not need to simulate the zero-truncation either).
6851 if (commit_window
|| off_is_imm
)
6854 /* Simulate and find potential out-of-bounds access under
6855 * speculative execution from truncation as a result of
6856 * masking when off was not within expected range. If off
6857 * sits in dst, then we temporarily need to move ptr there
6858 * to simulate dst (== 0) +/-= ptr. Needed, for example,
6859 * for cases where we use K-based arithmetic in one direction
6860 * and truncated reg-based in the other in order to explore
6863 if (!ptr_is_dst_reg
) {
6865 *dst_reg
= *ptr_reg
;
6867 ret
= sanitize_speculative_path(env
, NULL
, env
->insn_idx
+ 1,
6869 if (!ptr_is_dst_reg
&& ret
)
6871 return !ret
? REASON_STACK
: 0;
6874 static void sanitize_mark_insn_seen(struct bpf_verifier_env
*env
)
6876 struct bpf_verifier_state
*vstate
= env
->cur_state
;
6878 /* If we simulate paths under speculation, we don't update the
6879 * insn as 'seen' such that when we verify unreachable paths in
6880 * the non-speculative domain, sanitize_dead_code() can still
6881 * rewrite/sanitize them.
6883 if (!vstate
->speculative
)
6884 env
->insn_aux_data
[env
->insn_idx
].seen
= env
->pass_cnt
;
6887 static int sanitize_err(struct bpf_verifier_env
*env
,
6888 const struct bpf_insn
*insn
, int reason
,
6889 const struct bpf_reg_state
*off_reg
,
6890 const struct bpf_reg_state
*dst_reg
)
6892 static const char *err
= "pointer arithmetic with it prohibited for !root";
6893 const char *op
= BPF_OP(insn
->code
) == BPF_ADD
? "add" : "sub";
6894 u32 dst
= insn
->dst_reg
, src
= insn
->src_reg
;
6898 verbose(env
, "R%d has unknown scalar with mixed signed bounds, %s\n",
6899 off_reg
== dst_reg
? dst
: src
, err
);
6902 verbose(env
, "R%d has pointer with unsupported alu operation, %s\n",
6903 off_reg
== dst_reg
? src
: dst
, err
);
6906 verbose(env
, "R%d tried to %s from different maps, paths or scalars, %s\n",
6910 verbose(env
, "R%d tried to %s beyond pointer bounds, %s\n",
6914 verbose(env
, "R%d could not be pushed for speculative verification, %s\n",
6918 verbose(env
, "verifier internal error: unknown reason (%d)\n",
6926 /* check that stack access falls within stack limits and that 'reg' doesn't
6927 * have a variable offset.
6929 * Variable offset is prohibited for unprivileged mode for simplicity since it
6930 * requires corresponding support in Spectre masking for stack ALU. See also
6931 * retrieve_ptr_limit().
6934 * 'off' includes 'reg->off'.
6936 static int check_stack_access_for_ptr_arithmetic(
6937 struct bpf_verifier_env
*env
,
6939 const struct bpf_reg_state
*reg
,
6942 if (!tnum_is_const(reg
->var_off
)) {
6945 tnum_strn(tn_buf
, sizeof(tn_buf
), reg
->var_off
);
6946 verbose(env
, "R%d variable stack access prohibited for !root, var_off=%s off=%d\n",
6947 regno
, tn_buf
, off
);
6951 if (off
>= 0 || off
< -MAX_BPF_STACK
) {
6952 verbose(env
, "R%d stack pointer arithmetic goes out of range, "
6953 "prohibited for !root; off=%d\n", regno
, off
);
6960 static int sanitize_check_bounds(struct bpf_verifier_env
*env
,
6961 const struct bpf_insn
*insn
,
6962 const struct bpf_reg_state
*dst_reg
)
6964 u32 dst
= insn
->dst_reg
;
6966 /* For unprivileged we require that resulting offset must be in bounds
6967 * in order to be able to sanitize access later on.
6969 if (env
->bypass_spec_v1
)
6972 switch (dst_reg
->type
) {
6974 if (check_stack_access_for_ptr_arithmetic(env
, dst
, dst_reg
,
6975 dst_reg
->off
+ dst_reg
->var_off
.value
))
6978 case PTR_TO_MAP_VALUE
:
6979 if (check_map_access(env
, dst
, dst_reg
->off
, 1, false)) {
6980 verbose(env
, "R%d pointer arithmetic of map value goes out of range, "
6981 "prohibited for !root\n", dst
);
6992 /* Handles arithmetic on a pointer and a scalar: computes new min/max and var_off.
6993 * Caller should also handle BPF_MOV case separately.
6994 * If we return -EACCES, caller may want to try again treating pointer as a
6995 * scalar. So we only emit a diagnostic if !env->allow_ptr_leaks.
6997 static int adjust_ptr_min_max_vals(struct bpf_verifier_env
*env
,
6998 struct bpf_insn
*insn
,
6999 const struct bpf_reg_state
*ptr_reg
,
7000 const struct bpf_reg_state
*off_reg
)
7002 struct bpf_verifier_state
*vstate
= env
->cur_state
;
7003 struct bpf_func_state
*state
= vstate
->frame
[vstate
->curframe
];
7004 struct bpf_reg_state
*regs
= state
->regs
, *dst_reg
;
7005 bool known
= tnum_is_const(off_reg
->var_off
);
7006 s64 smin_val
= off_reg
->smin_value
, smax_val
= off_reg
->smax_value
,
7007 smin_ptr
= ptr_reg
->smin_value
, smax_ptr
= ptr_reg
->smax_value
;
7008 u64 umin_val
= off_reg
->umin_value
, umax_val
= off_reg
->umax_value
,
7009 umin_ptr
= ptr_reg
->umin_value
, umax_ptr
= ptr_reg
->umax_value
;
7010 struct bpf_sanitize_info info
= {};
7011 u8 opcode
= BPF_OP(insn
->code
);
7012 u32 dst
= insn
->dst_reg
;
7015 dst_reg
= ®s
[dst
];
7017 if ((known
&& (smin_val
!= smax_val
|| umin_val
!= umax_val
)) ||
7018 smin_val
> smax_val
|| umin_val
> umax_val
) {
7019 /* Taint dst register if offset had invalid bounds derived from
7020 * e.g. dead branches.
7022 __mark_reg_unknown(env
, dst_reg
);
7026 if (BPF_CLASS(insn
->code
) != BPF_ALU64
) {
7027 /* 32-bit ALU ops on pointers produce (meaningless) scalars */
7028 if (opcode
== BPF_SUB
&& env
->allow_ptr_leaks
) {
7029 __mark_reg_unknown(env
, dst_reg
);
7034 "R%d 32-bit pointer arithmetic prohibited\n",
7039 switch (ptr_reg
->type
) {
7040 case PTR_TO_MAP_VALUE_OR_NULL
:
7041 verbose(env
, "R%d pointer arithmetic on %s prohibited, null-check it first\n",
7042 dst
, reg_type_str
[ptr_reg
->type
]);
7044 case CONST_PTR_TO_MAP
:
7045 /* smin_val represents the known value */
7046 if (known
&& smin_val
== 0 && opcode
== BPF_ADD
)
7049 case PTR_TO_PACKET_END
:
7051 case PTR_TO_SOCK_COMMON
:
7052 case PTR_TO_TCP_SOCK
:
7053 case PTR_TO_XDP_SOCK
:
7055 verbose(env
, "R%d pointer arithmetic on %s prohibited\n",
7056 dst
, reg_type_str
[ptr_reg
->type
]);
7059 if (reg_type_may_be_null(ptr_reg
->type
))
7064 /* In case of 'scalar += pointer', dst_reg inherits pointer type and id.
7065 * The id may be overwritten later if we create a new variable offset.
7067 dst_reg
->type
= ptr_reg
->type
;
7068 dst_reg
->id
= ptr_reg
->id
;
7070 if (!check_reg_sane_offset(env
, off_reg
, ptr_reg
->type
) ||
7071 !check_reg_sane_offset(env
, ptr_reg
, ptr_reg
->type
))
7074 /* pointer types do not carry 32-bit bounds at the moment. */
7075 __mark_reg32_unbounded(dst_reg
);
7077 if (sanitize_needed(opcode
)) {
7078 ret
= sanitize_ptr_alu(env
, insn
, ptr_reg
, off_reg
, dst_reg
,
7081 return sanitize_err(env
, insn
, ret
, off_reg
, dst_reg
);
7086 /* We can take a fixed offset as long as it doesn't overflow
7087 * the s32 'off' field
7089 if (known
&& (ptr_reg
->off
+ smin_val
==
7090 (s64
)(s32
)(ptr_reg
->off
+ smin_val
))) {
7091 /* pointer += K. Accumulate it into fixed offset */
7092 dst_reg
->smin_value
= smin_ptr
;
7093 dst_reg
->smax_value
= smax_ptr
;
7094 dst_reg
->umin_value
= umin_ptr
;
7095 dst_reg
->umax_value
= umax_ptr
;
7096 dst_reg
->var_off
= ptr_reg
->var_off
;
7097 dst_reg
->off
= ptr_reg
->off
+ smin_val
;
7098 dst_reg
->raw
= ptr_reg
->raw
;
7101 /* A new variable offset is created. Note that off_reg->off
7102 * == 0, since it's a scalar.
7103 * dst_reg gets the pointer type and since some positive
7104 * integer value was added to the pointer, give it a new 'id'
7105 * if it's a PTR_TO_PACKET.
7106 * this creates a new 'base' pointer, off_reg (variable) gets
7107 * added into the variable offset, and we copy the fixed offset
7110 if (signed_add_overflows(smin_ptr
, smin_val
) ||
7111 signed_add_overflows(smax_ptr
, smax_val
)) {
7112 dst_reg
->smin_value
= S64_MIN
;
7113 dst_reg
->smax_value
= S64_MAX
;
7115 dst_reg
->smin_value
= smin_ptr
+ smin_val
;
7116 dst_reg
->smax_value
= smax_ptr
+ smax_val
;
7118 if (umin_ptr
+ umin_val
< umin_ptr
||
7119 umax_ptr
+ umax_val
< umax_ptr
) {
7120 dst_reg
->umin_value
= 0;
7121 dst_reg
->umax_value
= U64_MAX
;
7123 dst_reg
->umin_value
= umin_ptr
+ umin_val
;
7124 dst_reg
->umax_value
= umax_ptr
+ umax_val
;
7126 dst_reg
->var_off
= tnum_add(ptr_reg
->var_off
, off_reg
->var_off
);
7127 dst_reg
->off
= ptr_reg
->off
;
7128 dst_reg
->raw
= ptr_reg
->raw
;
7129 if (reg_is_pkt_pointer(ptr_reg
)) {
7130 dst_reg
->id
= ++env
->id_gen
;
7131 /* something was added to pkt_ptr, set range to zero */
7132 memset(&dst_reg
->raw
, 0, sizeof(dst_reg
->raw
));
7136 if (dst_reg
== off_reg
) {
7137 /* scalar -= pointer. Creates an unknown scalar */
7138 verbose(env
, "R%d tried to subtract pointer from scalar\n",
7142 /* We don't allow subtraction from FP, because (according to
7143 * test_verifier.c test "invalid fp arithmetic", JITs might not
7144 * be able to deal with it.
7146 if (ptr_reg
->type
== PTR_TO_STACK
) {
7147 verbose(env
, "R%d subtraction from stack pointer prohibited\n",
7151 if (known
&& (ptr_reg
->off
- smin_val
==
7152 (s64
)(s32
)(ptr_reg
->off
- smin_val
))) {
7153 /* pointer -= K. Subtract it from fixed offset */
7154 dst_reg
->smin_value
= smin_ptr
;
7155 dst_reg
->smax_value
= smax_ptr
;
7156 dst_reg
->umin_value
= umin_ptr
;
7157 dst_reg
->umax_value
= umax_ptr
;
7158 dst_reg
->var_off
= ptr_reg
->var_off
;
7159 dst_reg
->id
= ptr_reg
->id
;
7160 dst_reg
->off
= ptr_reg
->off
- smin_val
;
7161 dst_reg
->raw
= ptr_reg
->raw
;
7164 /* A new variable offset is created. If the subtrahend is known
7165 * nonnegative, then any reg->range we had before is still good.
7167 if (signed_sub_overflows(smin_ptr
, smax_val
) ||
7168 signed_sub_overflows(smax_ptr
, smin_val
)) {
7169 /* Overflow possible, we know nothing */
7170 dst_reg
->smin_value
= S64_MIN
;
7171 dst_reg
->smax_value
= S64_MAX
;
7173 dst_reg
->smin_value
= smin_ptr
- smax_val
;
7174 dst_reg
->smax_value
= smax_ptr
- smin_val
;
7176 if (umin_ptr
< umax_val
) {
7177 /* Overflow possible, we know nothing */
7178 dst_reg
->umin_value
= 0;
7179 dst_reg
->umax_value
= U64_MAX
;
7181 /* Cannot overflow (as long as bounds are consistent) */
7182 dst_reg
->umin_value
= umin_ptr
- umax_val
;
7183 dst_reg
->umax_value
= umax_ptr
- umin_val
;
7185 dst_reg
->var_off
= tnum_sub(ptr_reg
->var_off
, off_reg
->var_off
);
7186 dst_reg
->off
= ptr_reg
->off
;
7187 dst_reg
->raw
= ptr_reg
->raw
;
7188 if (reg_is_pkt_pointer(ptr_reg
)) {
7189 dst_reg
->id
= ++env
->id_gen
;
7190 /* something was added to pkt_ptr, set range to zero */
7192 memset(&dst_reg
->raw
, 0, sizeof(dst_reg
->raw
));
7198 /* bitwise ops on pointers are troublesome, prohibit. */
7199 verbose(env
, "R%d bitwise operator %s on pointer prohibited\n",
7200 dst
, bpf_alu_string
[opcode
>> 4]);
7203 /* other operators (e.g. MUL,LSH) produce non-pointer results */
7204 verbose(env
, "R%d pointer arithmetic with %s operator prohibited\n",
7205 dst
, bpf_alu_string
[opcode
>> 4]);
7209 if (!check_reg_sane_offset(env
, dst_reg
, ptr_reg
->type
))
7212 __update_reg_bounds(dst_reg
);
7213 __reg_deduce_bounds(dst_reg
);
7214 __reg_bound_offset(dst_reg
);
7216 if (sanitize_check_bounds(env
, insn
, dst_reg
) < 0)
7218 if (sanitize_needed(opcode
)) {
7219 ret
= sanitize_ptr_alu(env
, insn
, dst_reg
, off_reg
, dst_reg
,
7222 return sanitize_err(env
, insn
, ret
, off_reg
, dst_reg
);
7228 static void scalar32_min_max_add(struct bpf_reg_state
*dst_reg
,
7229 struct bpf_reg_state
*src_reg
)
7231 s32 smin_val
= src_reg
->s32_min_value
;
7232 s32 smax_val
= src_reg
->s32_max_value
;
7233 u32 umin_val
= src_reg
->u32_min_value
;
7234 u32 umax_val
= src_reg
->u32_max_value
;
7236 if (signed_add32_overflows(dst_reg
->s32_min_value
, smin_val
) ||
7237 signed_add32_overflows(dst_reg
->s32_max_value
, smax_val
)) {
7238 dst_reg
->s32_min_value
= S32_MIN
;
7239 dst_reg
->s32_max_value
= S32_MAX
;
7241 dst_reg
->s32_min_value
+= smin_val
;
7242 dst_reg
->s32_max_value
+= smax_val
;
7244 if (dst_reg
->u32_min_value
+ umin_val
< umin_val
||
7245 dst_reg
->u32_max_value
+ umax_val
< umax_val
) {
7246 dst_reg
->u32_min_value
= 0;
7247 dst_reg
->u32_max_value
= U32_MAX
;
7249 dst_reg
->u32_min_value
+= umin_val
;
7250 dst_reg
->u32_max_value
+= umax_val
;
7254 static void scalar_min_max_add(struct bpf_reg_state
*dst_reg
,
7255 struct bpf_reg_state
*src_reg
)
7257 s64 smin_val
= src_reg
->smin_value
;
7258 s64 smax_val
= src_reg
->smax_value
;
7259 u64 umin_val
= src_reg
->umin_value
;
7260 u64 umax_val
= src_reg
->umax_value
;
7262 if (signed_add_overflows(dst_reg
->smin_value
, smin_val
) ||
7263 signed_add_overflows(dst_reg
->smax_value
, smax_val
)) {
7264 dst_reg
->smin_value
= S64_MIN
;
7265 dst_reg
->smax_value
= S64_MAX
;
7267 dst_reg
->smin_value
+= smin_val
;
7268 dst_reg
->smax_value
+= smax_val
;
7270 if (dst_reg
->umin_value
+ umin_val
< umin_val
||
7271 dst_reg
->umax_value
+ umax_val
< umax_val
) {
7272 dst_reg
->umin_value
= 0;
7273 dst_reg
->umax_value
= U64_MAX
;
7275 dst_reg
->umin_value
+= umin_val
;
7276 dst_reg
->umax_value
+= umax_val
;
7280 static void scalar32_min_max_sub(struct bpf_reg_state
*dst_reg
,
7281 struct bpf_reg_state
*src_reg
)
7283 s32 smin_val
= src_reg
->s32_min_value
;
7284 s32 smax_val
= src_reg
->s32_max_value
;
7285 u32 umin_val
= src_reg
->u32_min_value
;
7286 u32 umax_val
= src_reg
->u32_max_value
;
7288 if (signed_sub32_overflows(dst_reg
->s32_min_value
, smax_val
) ||
7289 signed_sub32_overflows(dst_reg
->s32_max_value
, smin_val
)) {
7290 /* Overflow possible, we know nothing */
7291 dst_reg
->s32_min_value
= S32_MIN
;
7292 dst_reg
->s32_max_value
= S32_MAX
;
7294 dst_reg
->s32_min_value
-= smax_val
;
7295 dst_reg
->s32_max_value
-= smin_val
;
7297 if (dst_reg
->u32_min_value
< umax_val
) {
7298 /* Overflow possible, we know nothing */
7299 dst_reg
->u32_min_value
= 0;
7300 dst_reg
->u32_max_value
= U32_MAX
;
7302 /* Cannot overflow (as long as bounds are consistent) */
7303 dst_reg
->u32_min_value
-= umax_val
;
7304 dst_reg
->u32_max_value
-= umin_val
;
7308 static void scalar_min_max_sub(struct bpf_reg_state
*dst_reg
,
7309 struct bpf_reg_state
*src_reg
)
7311 s64 smin_val
= src_reg
->smin_value
;
7312 s64 smax_val
= src_reg
->smax_value
;
7313 u64 umin_val
= src_reg
->umin_value
;
7314 u64 umax_val
= src_reg
->umax_value
;
7316 if (signed_sub_overflows(dst_reg
->smin_value
, smax_val
) ||
7317 signed_sub_overflows(dst_reg
->smax_value
, smin_val
)) {
7318 /* Overflow possible, we know nothing */
7319 dst_reg
->smin_value
= S64_MIN
;
7320 dst_reg
->smax_value
= S64_MAX
;
7322 dst_reg
->smin_value
-= smax_val
;
7323 dst_reg
->smax_value
-= smin_val
;
7325 if (dst_reg
->umin_value
< umax_val
) {
7326 /* Overflow possible, we know nothing */
7327 dst_reg
->umin_value
= 0;
7328 dst_reg
->umax_value
= U64_MAX
;
7330 /* Cannot overflow (as long as bounds are consistent) */
7331 dst_reg
->umin_value
-= umax_val
;
7332 dst_reg
->umax_value
-= umin_val
;
7336 static void scalar32_min_max_mul(struct bpf_reg_state
*dst_reg
,
7337 struct bpf_reg_state
*src_reg
)
7339 s32 smin_val
= src_reg
->s32_min_value
;
7340 u32 umin_val
= src_reg
->u32_min_value
;
7341 u32 umax_val
= src_reg
->u32_max_value
;
7343 if (smin_val
< 0 || dst_reg
->s32_min_value
< 0) {
7344 /* Ain't nobody got time to multiply that sign */
7345 __mark_reg32_unbounded(dst_reg
);
7348 /* Both values are positive, so we can work with unsigned and
7349 * copy the result to signed (unless it exceeds S32_MAX).
7351 if (umax_val
> U16_MAX
|| dst_reg
->u32_max_value
> U16_MAX
) {
7352 /* Potential overflow, we know nothing */
7353 __mark_reg32_unbounded(dst_reg
);
7356 dst_reg
->u32_min_value
*= umin_val
;
7357 dst_reg
->u32_max_value
*= umax_val
;
7358 if (dst_reg
->u32_max_value
> S32_MAX
) {
7359 /* Overflow possible, we know nothing */
7360 dst_reg
->s32_min_value
= S32_MIN
;
7361 dst_reg
->s32_max_value
= S32_MAX
;
7363 dst_reg
->s32_min_value
= dst_reg
->u32_min_value
;
7364 dst_reg
->s32_max_value
= dst_reg
->u32_max_value
;
7368 static void scalar_min_max_mul(struct bpf_reg_state
*dst_reg
,
7369 struct bpf_reg_state
*src_reg
)
7371 s64 smin_val
= src_reg
->smin_value
;
7372 u64 umin_val
= src_reg
->umin_value
;
7373 u64 umax_val
= src_reg
->umax_value
;
7375 if (smin_val
< 0 || dst_reg
->smin_value
< 0) {
7376 /* Ain't nobody got time to multiply that sign */
7377 __mark_reg64_unbounded(dst_reg
);
7380 /* Both values are positive, so we can work with unsigned and
7381 * copy the result to signed (unless it exceeds S64_MAX).
7383 if (umax_val
> U32_MAX
|| dst_reg
->umax_value
> U32_MAX
) {
7384 /* Potential overflow, we know nothing */
7385 __mark_reg64_unbounded(dst_reg
);
7388 dst_reg
->umin_value
*= umin_val
;
7389 dst_reg
->umax_value
*= umax_val
;
7390 if (dst_reg
->umax_value
> S64_MAX
) {
7391 /* Overflow possible, we know nothing */
7392 dst_reg
->smin_value
= S64_MIN
;
7393 dst_reg
->smax_value
= S64_MAX
;
7395 dst_reg
->smin_value
= dst_reg
->umin_value
;
7396 dst_reg
->smax_value
= dst_reg
->umax_value
;
7400 static void scalar32_min_max_and(struct bpf_reg_state
*dst_reg
,
7401 struct bpf_reg_state
*src_reg
)
7403 bool src_known
= tnum_subreg_is_const(src_reg
->var_off
);
7404 bool dst_known
= tnum_subreg_is_const(dst_reg
->var_off
);
7405 struct tnum var32_off
= tnum_subreg(dst_reg
->var_off
);
7406 s32 smin_val
= src_reg
->s32_min_value
;
7407 u32 umax_val
= src_reg
->u32_max_value
;
7409 if (src_known
&& dst_known
) {
7410 __mark_reg32_known(dst_reg
, var32_off
.value
);
7414 /* We get our minimum from the var_off, since that's inherently
7415 * bitwise. Our maximum is the minimum of the operands' maxima.
7417 dst_reg
->u32_min_value
= var32_off
.value
;
7418 dst_reg
->u32_max_value
= min(dst_reg
->u32_max_value
, umax_val
);
7419 if (dst_reg
->s32_min_value
< 0 || smin_val
< 0) {
7420 /* Lose signed bounds when ANDing negative numbers,
7421 * ain't nobody got time for that.
7423 dst_reg
->s32_min_value
= S32_MIN
;
7424 dst_reg
->s32_max_value
= S32_MAX
;
7426 /* ANDing two positives gives a positive, so safe to
7427 * cast result into s64.
7429 dst_reg
->s32_min_value
= dst_reg
->u32_min_value
;
7430 dst_reg
->s32_max_value
= dst_reg
->u32_max_value
;
7434 static void scalar_min_max_and(struct bpf_reg_state
*dst_reg
,
7435 struct bpf_reg_state
*src_reg
)
7437 bool src_known
= tnum_is_const(src_reg
->var_off
);
7438 bool dst_known
= tnum_is_const(dst_reg
->var_off
);
7439 s64 smin_val
= src_reg
->smin_value
;
7440 u64 umax_val
= src_reg
->umax_value
;
7442 if (src_known
&& dst_known
) {
7443 __mark_reg_known(dst_reg
, dst_reg
->var_off
.value
);
7447 /* We get our minimum from the var_off, since that's inherently
7448 * bitwise. Our maximum is the minimum of the operands' maxima.
7450 dst_reg
->umin_value
= dst_reg
->var_off
.value
;
7451 dst_reg
->umax_value
= min(dst_reg
->umax_value
, umax_val
);
7452 if (dst_reg
->smin_value
< 0 || smin_val
< 0) {
7453 /* Lose signed bounds when ANDing negative numbers,
7454 * ain't nobody got time for that.
7456 dst_reg
->smin_value
= S64_MIN
;
7457 dst_reg
->smax_value
= S64_MAX
;
7459 /* ANDing two positives gives a positive, so safe to
7460 * cast result into s64.
7462 dst_reg
->smin_value
= dst_reg
->umin_value
;
7463 dst_reg
->smax_value
= dst_reg
->umax_value
;
7465 /* We may learn something more from the var_off */
7466 __update_reg_bounds(dst_reg
);
7469 static void scalar32_min_max_or(struct bpf_reg_state
*dst_reg
,
7470 struct bpf_reg_state
*src_reg
)
7472 bool src_known
= tnum_subreg_is_const(src_reg
->var_off
);
7473 bool dst_known
= tnum_subreg_is_const(dst_reg
->var_off
);
7474 struct tnum var32_off
= tnum_subreg(dst_reg
->var_off
);
7475 s32 smin_val
= src_reg
->s32_min_value
;
7476 u32 umin_val
= src_reg
->u32_min_value
;
7478 if (src_known
&& dst_known
) {
7479 __mark_reg32_known(dst_reg
, var32_off
.value
);
7483 /* We get our maximum from the var_off, and our minimum is the
7484 * maximum of the operands' minima
7486 dst_reg
->u32_min_value
= max(dst_reg
->u32_min_value
, umin_val
);
7487 dst_reg
->u32_max_value
= var32_off
.value
| var32_off
.mask
;
7488 if (dst_reg
->s32_min_value
< 0 || smin_val
< 0) {
7489 /* Lose signed bounds when ORing negative numbers,
7490 * ain't nobody got time for that.
7492 dst_reg
->s32_min_value
= S32_MIN
;
7493 dst_reg
->s32_max_value
= S32_MAX
;
7495 /* ORing two positives gives a positive, so safe to
7496 * cast result into s64.
7498 dst_reg
->s32_min_value
= dst_reg
->u32_min_value
;
7499 dst_reg
->s32_max_value
= dst_reg
->u32_max_value
;
7503 static void scalar_min_max_or(struct bpf_reg_state
*dst_reg
,
7504 struct bpf_reg_state
*src_reg
)
7506 bool src_known
= tnum_is_const(src_reg
->var_off
);
7507 bool dst_known
= tnum_is_const(dst_reg
->var_off
);
7508 s64 smin_val
= src_reg
->smin_value
;
7509 u64 umin_val
= src_reg
->umin_value
;
7511 if (src_known
&& dst_known
) {
7512 __mark_reg_known(dst_reg
, dst_reg
->var_off
.value
);
7516 /* We get our maximum from the var_off, and our minimum is the
7517 * maximum of the operands' minima
7519 dst_reg
->umin_value
= max(dst_reg
->umin_value
, umin_val
);
7520 dst_reg
->umax_value
= dst_reg
->var_off
.value
| dst_reg
->var_off
.mask
;
7521 if (dst_reg
->smin_value
< 0 || smin_val
< 0) {
7522 /* Lose signed bounds when ORing negative numbers,
7523 * ain't nobody got time for that.
7525 dst_reg
->smin_value
= S64_MIN
;
7526 dst_reg
->smax_value
= S64_MAX
;
7528 /* ORing two positives gives a positive, so safe to
7529 * cast result into s64.
7531 dst_reg
->smin_value
= dst_reg
->umin_value
;
7532 dst_reg
->smax_value
= dst_reg
->umax_value
;
7534 /* We may learn something more from the var_off */
7535 __update_reg_bounds(dst_reg
);
7538 static void scalar32_min_max_xor(struct bpf_reg_state
*dst_reg
,
7539 struct bpf_reg_state
*src_reg
)
7541 bool src_known
= tnum_subreg_is_const(src_reg
->var_off
);
7542 bool dst_known
= tnum_subreg_is_const(dst_reg
->var_off
);
7543 struct tnum var32_off
= tnum_subreg(dst_reg
->var_off
);
7544 s32 smin_val
= src_reg
->s32_min_value
;
7546 if (src_known
&& dst_known
) {
7547 __mark_reg32_known(dst_reg
, var32_off
.value
);
7551 /* We get both minimum and maximum from the var32_off. */
7552 dst_reg
->u32_min_value
= var32_off
.value
;
7553 dst_reg
->u32_max_value
= var32_off
.value
| var32_off
.mask
;
7555 if (dst_reg
->s32_min_value
>= 0 && smin_val
>= 0) {
7556 /* XORing two positive sign numbers gives a positive,
7557 * so safe to cast u32 result into s32.
7559 dst_reg
->s32_min_value
= dst_reg
->u32_min_value
;
7560 dst_reg
->s32_max_value
= dst_reg
->u32_max_value
;
7562 dst_reg
->s32_min_value
= S32_MIN
;
7563 dst_reg
->s32_max_value
= S32_MAX
;
7567 static void scalar_min_max_xor(struct bpf_reg_state
*dst_reg
,
7568 struct bpf_reg_state
*src_reg
)
7570 bool src_known
= tnum_is_const(src_reg
->var_off
);
7571 bool dst_known
= tnum_is_const(dst_reg
->var_off
);
7572 s64 smin_val
= src_reg
->smin_value
;
7574 if (src_known
&& dst_known
) {
7575 /* dst_reg->var_off.value has been updated earlier */
7576 __mark_reg_known(dst_reg
, dst_reg
->var_off
.value
);
7580 /* We get both minimum and maximum from the var_off. */
7581 dst_reg
->umin_value
= dst_reg
->var_off
.value
;
7582 dst_reg
->umax_value
= dst_reg
->var_off
.value
| dst_reg
->var_off
.mask
;
7584 if (dst_reg
->smin_value
>= 0 && smin_val
>= 0) {
7585 /* XORing two positive sign numbers gives a positive,
7586 * so safe to cast u64 result into s64.
7588 dst_reg
->smin_value
= dst_reg
->umin_value
;
7589 dst_reg
->smax_value
= dst_reg
->umax_value
;
7591 dst_reg
->smin_value
= S64_MIN
;
7592 dst_reg
->smax_value
= S64_MAX
;
7595 __update_reg_bounds(dst_reg
);
7598 static void __scalar32_min_max_lsh(struct bpf_reg_state
*dst_reg
,
7599 u64 umin_val
, u64 umax_val
)
7601 /* We lose all sign bit information (except what we can pick
7604 dst_reg
->s32_min_value
= S32_MIN
;
7605 dst_reg
->s32_max_value
= S32_MAX
;
7606 /* If we might shift our top bit out, then we know nothing */
7607 if (umax_val
> 31 || dst_reg
->u32_max_value
> 1ULL << (31 - umax_val
)) {
7608 dst_reg
->u32_min_value
= 0;
7609 dst_reg
->u32_max_value
= U32_MAX
;
7611 dst_reg
->u32_min_value
<<= umin_val
;
7612 dst_reg
->u32_max_value
<<= umax_val
;
7616 static void scalar32_min_max_lsh(struct bpf_reg_state
*dst_reg
,
7617 struct bpf_reg_state
*src_reg
)
7619 u32 umax_val
= src_reg
->u32_max_value
;
7620 u32 umin_val
= src_reg
->u32_min_value
;
7621 /* u32 alu operation will zext upper bits */
7622 struct tnum subreg
= tnum_subreg(dst_reg
->var_off
);
7624 __scalar32_min_max_lsh(dst_reg
, umin_val
, umax_val
);
7625 dst_reg
->var_off
= tnum_subreg(tnum_lshift(subreg
, umin_val
));
7626 /* Not required but being careful mark reg64 bounds as unknown so
7627 * that we are forced to pick them up from tnum and zext later and
7628 * if some path skips this step we are still safe.
7630 __mark_reg64_unbounded(dst_reg
);
7631 __update_reg32_bounds(dst_reg
);
7634 static void __scalar64_min_max_lsh(struct bpf_reg_state
*dst_reg
,
7635 u64 umin_val
, u64 umax_val
)
7637 /* Special case <<32 because it is a common compiler pattern to sign
7638 * extend subreg by doing <<32 s>>32. In this case if 32bit bounds are
7639 * positive we know this shift will also be positive so we can track
7640 * bounds correctly. Otherwise we lose all sign bit information except
7641 * what we can pick up from var_off. Perhaps we can generalize this
7642 * later to shifts of any length.
7644 if (umin_val
== 32 && umax_val
== 32 && dst_reg
->s32_max_value
>= 0)
7645 dst_reg
->smax_value
= (s64
)dst_reg
->s32_max_value
<< 32;
7647 dst_reg
->smax_value
= S64_MAX
;
7649 if (umin_val
== 32 && umax_val
== 32 && dst_reg
->s32_min_value
>= 0)
7650 dst_reg
->smin_value
= (s64
)dst_reg
->s32_min_value
<< 32;
7652 dst_reg
->smin_value
= S64_MIN
;
7654 /* If we might shift our top bit out, then we know nothing */
7655 if (dst_reg
->umax_value
> 1ULL << (63 - umax_val
)) {
7656 dst_reg
->umin_value
= 0;
7657 dst_reg
->umax_value
= U64_MAX
;
7659 dst_reg
->umin_value
<<= umin_val
;
7660 dst_reg
->umax_value
<<= umax_val
;
7664 static void scalar_min_max_lsh(struct bpf_reg_state
*dst_reg
,
7665 struct bpf_reg_state
*src_reg
)
7667 u64 umax_val
= src_reg
->umax_value
;
7668 u64 umin_val
= src_reg
->umin_value
;
7670 /* scalar64 calc uses 32bit unshifted bounds so must be called first */
7671 __scalar64_min_max_lsh(dst_reg
, umin_val
, umax_val
);
7672 __scalar32_min_max_lsh(dst_reg
, umin_val
, umax_val
);
7674 dst_reg
->var_off
= tnum_lshift(dst_reg
->var_off
, umin_val
);
7675 /* We may learn something more from the var_off */
7676 __update_reg_bounds(dst_reg
);
7679 static void scalar32_min_max_rsh(struct bpf_reg_state
*dst_reg
,
7680 struct bpf_reg_state
*src_reg
)
7682 struct tnum subreg
= tnum_subreg(dst_reg
->var_off
);
7683 u32 umax_val
= src_reg
->u32_max_value
;
7684 u32 umin_val
= src_reg
->u32_min_value
;
7686 /* BPF_RSH is an unsigned shift. If the value in dst_reg might
7687 * be negative, then either:
7688 * 1) src_reg might be zero, so the sign bit of the result is
7689 * unknown, so we lose our signed bounds
7690 * 2) it's known negative, thus the unsigned bounds capture the
7692 * 3) the signed bounds cross zero, so they tell us nothing
7694 * If the value in dst_reg is known nonnegative, then again the
7695 * unsigned bounds capture the signed bounds.
7696 * Thus, in all cases it suffices to blow away our signed bounds
7697 * and rely on inferring new ones from the unsigned bounds and
7698 * var_off of the result.
7700 dst_reg
->s32_min_value
= S32_MIN
;
7701 dst_reg
->s32_max_value
= S32_MAX
;
7703 dst_reg
->var_off
= tnum_rshift(subreg
, umin_val
);
7704 dst_reg
->u32_min_value
>>= umax_val
;
7705 dst_reg
->u32_max_value
>>= umin_val
;
7707 __mark_reg64_unbounded(dst_reg
);
7708 __update_reg32_bounds(dst_reg
);
7711 static void scalar_min_max_rsh(struct bpf_reg_state
*dst_reg
,
7712 struct bpf_reg_state
*src_reg
)
7714 u64 umax_val
= src_reg
->umax_value
;
7715 u64 umin_val
= src_reg
->umin_value
;
7717 /* BPF_RSH is an unsigned shift. If the value in dst_reg might
7718 * be negative, then either:
7719 * 1) src_reg might be zero, so the sign bit of the result is
7720 * unknown, so we lose our signed bounds
7721 * 2) it's known negative, thus the unsigned bounds capture the
7723 * 3) the signed bounds cross zero, so they tell us nothing
7725 * If the value in dst_reg is known nonnegative, then again the
7726 * unsigned bounds capture the signed bounds.
7727 * Thus, in all cases it suffices to blow away our signed bounds
7728 * and rely on inferring new ones from the unsigned bounds and
7729 * var_off of the result.
7731 dst_reg
->smin_value
= S64_MIN
;
7732 dst_reg
->smax_value
= S64_MAX
;
7733 dst_reg
->var_off
= tnum_rshift(dst_reg
->var_off
, umin_val
);
7734 dst_reg
->umin_value
>>= umax_val
;
7735 dst_reg
->umax_value
>>= umin_val
;
7737 /* Its not easy to operate on alu32 bounds here because it depends
7738 * on bits being shifted in. Take easy way out and mark unbounded
7739 * so we can recalculate later from tnum.
7741 __mark_reg32_unbounded(dst_reg
);
7742 __update_reg_bounds(dst_reg
);
7745 static void scalar32_min_max_arsh(struct bpf_reg_state
*dst_reg
,
7746 struct bpf_reg_state
*src_reg
)
7748 u64 umin_val
= src_reg
->u32_min_value
;
7750 /* Upon reaching here, src_known is true and
7751 * umax_val is equal to umin_val.
7753 dst_reg
->s32_min_value
= (u32
)(((s32
)dst_reg
->s32_min_value
) >> umin_val
);
7754 dst_reg
->s32_max_value
= (u32
)(((s32
)dst_reg
->s32_max_value
) >> umin_val
);
7756 dst_reg
->var_off
= tnum_arshift(tnum_subreg(dst_reg
->var_off
), umin_val
, 32);
7758 /* blow away the dst_reg umin_value/umax_value and rely on
7759 * dst_reg var_off to refine the result.
7761 dst_reg
->u32_min_value
= 0;
7762 dst_reg
->u32_max_value
= U32_MAX
;
7764 __mark_reg64_unbounded(dst_reg
);
7765 __update_reg32_bounds(dst_reg
);
7768 static void scalar_min_max_arsh(struct bpf_reg_state
*dst_reg
,
7769 struct bpf_reg_state
*src_reg
)
7771 u64 umin_val
= src_reg
->umin_value
;
7773 /* Upon reaching here, src_known is true and umax_val is equal
7776 dst_reg
->smin_value
>>= umin_val
;
7777 dst_reg
->smax_value
>>= umin_val
;
7779 dst_reg
->var_off
= tnum_arshift(dst_reg
->var_off
, umin_val
, 64);
7781 /* blow away the dst_reg umin_value/umax_value and rely on
7782 * dst_reg var_off to refine the result.
7784 dst_reg
->umin_value
= 0;
7785 dst_reg
->umax_value
= U64_MAX
;
7787 /* Its not easy to operate on alu32 bounds here because it depends
7788 * on bits being shifted in from upper 32-bits. Take easy way out
7789 * and mark unbounded so we can recalculate later from tnum.
7791 __mark_reg32_unbounded(dst_reg
);
7792 __update_reg_bounds(dst_reg
);
7795 /* WARNING: This function does calculations on 64-bit values, but the actual
7796 * execution may occur on 32-bit values. Therefore, things like bitshifts
7797 * need extra checks in the 32-bit case.
7799 static int adjust_scalar_min_max_vals(struct bpf_verifier_env
*env
,
7800 struct bpf_insn
*insn
,
7801 struct bpf_reg_state
*dst_reg
,
7802 struct bpf_reg_state src_reg
)
7804 struct bpf_reg_state
*regs
= cur_regs(env
);
7805 u8 opcode
= BPF_OP(insn
->code
);
7807 s64 smin_val
, smax_val
;
7808 u64 umin_val
, umax_val
;
7809 s32 s32_min_val
, s32_max_val
;
7810 u32 u32_min_val
, u32_max_val
;
7811 u64 insn_bitness
= (BPF_CLASS(insn
->code
) == BPF_ALU64
) ? 64 : 32;
7812 bool alu32
= (BPF_CLASS(insn
->code
) != BPF_ALU64
);
7815 smin_val
= src_reg
.smin_value
;
7816 smax_val
= src_reg
.smax_value
;
7817 umin_val
= src_reg
.umin_value
;
7818 umax_val
= src_reg
.umax_value
;
7820 s32_min_val
= src_reg
.s32_min_value
;
7821 s32_max_val
= src_reg
.s32_max_value
;
7822 u32_min_val
= src_reg
.u32_min_value
;
7823 u32_max_val
= src_reg
.u32_max_value
;
7826 src_known
= tnum_subreg_is_const(src_reg
.var_off
);
7828 (s32_min_val
!= s32_max_val
|| u32_min_val
!= u32_max_val
)) ||
7829 s32_min_val
> s32_max_val
|| u32_min_val
> u32_max_val
) {
7830 /* Taint dst register if offset had invalid bounds
7831 * derived from e.g. dead branches.
7833 __mark_reg_unknown(env
, dst_reg
);
7837 src_known
= tnum_is_const(src_reg
.var_off
);
7839 (smin_val
!= smax_val
|| umin_val
!= umax_val
)) ||
7840 smin_val
> smax_val
|| umin_val
> umax_val
) {
7841 /* Taint dst register if offset had invalid bounds
7842 * derived from e.g. dead branches.
7844 __mark_reg_unknown(env
, dst_reg
);
7850 opcode
!= BPF_ADD
&& opcode
!= BPF_SUB
&& opcode
!= BPF_AND
) {
7851 __mark_reg_unknown(env
, dst_reg
);
7855 if (sanitize_needed(opcode
)) {
7856 ret
= sanitize_val_alu(env
, insn
);
7858 return sanitize_err(env
, insn
, ret
, NULL
, NULL
);
7861 /* Calculate sign/unsigned bounds and tnum for alu32 and alu64 bit ops.
7862 * There are two classes of instructions: The first class we track both
7863 * alu32 and alu64 sign/unsigned bounds independently this provides the
7864 * greatest amount of precision when alu operations are mixed with jmp32
7865 * operations. These operations are BPF_ADD, BPF_SUB, BPF_MUL, BPF_ADD,
7866 * and BPF_OR. This is possible because these ops have fairly easy to
7867 * understand and calculate behavior in both 32-bit and 64-bit alu ops.
7868 * See alu32 verifier tests for examples. The second class of
7869 * operations, BPF_LSH, BPF_RSH, and BPF_ARSH, however are not so easy
7870 * with regards to tracking sign/unsigned bounds because the bits may
7871 * cross subreg boundaries in the alu64 case. When this happens we mark
7872 * the reg unbounded in the subreg bound space and use the resulting
7873 * tnum to calculate an approximation of the sign/unsigned bounds.
7877 scalar32_min_max_add(dst_reg
, &src_reg
);
7878 scalar_min_max_add(dst_reg
, &src_reg
);
7879 dst_reg
->var_off
= tnum_add(dst_reg
->var_off
, src_reg
.var_off
);
7882 scalar32_min_max_sub(dst_reg
, &src_reg
);
7883 scalar_min_max_sub(dst_reg
, &src_reg
);
7884 dst_reg
->var_off
= tnum_sub(dst_reg
->var_off
, src_reg
.var_off
);
7887 dst_reg
->var_off
= tnum_mul(dst_reg
->var_off
, src_reg
.var_off
);
7888 scalar32_min_max_mul(dst_reg
, &src_reg
);
7889 scalar_min_max_mul(dst_reg
, &src_reg
);
7892 dst_reg
->var_off
= tnum_and(dst_reg
->var_off
, src_reg
.var_off
);
7893 scalar32_min_max_and(dst_reg
, &src_reg
);
7894 scalar_min_max_and(dst_reg
, &src_reg
);
7897 dst_reg
->var_off
= tnum_or(dst_reg
->var_off
, src_reg
.var_off
);
7898 scalar32_min_max_or(dst_reg
, &src_reg
);
7899 scalar_min_max_or(dst_reg
, &src_reg
);
7902 dst_reg
->var_off
= tnum_xor(dst_reg
->var_off
, src_reg
.var_off
);
7903 scalar32_min_max_xor(dst_reg
, &src_reg
);
7904 scalar_min_max_xor(dst_reg
, &src_reg
);
7907 if (umax_val
>= insn_bitness
) {
7908 /* Shifts greater than 31 or 63 are undefined.
7909 * This includes shifts by a negative number.
7911 mark_reg_unknown(env
, regs
, insn
->dst_reg
);
7915 scalar32_min_max_lsh(dst_reg
, &src_reg
);
7917 scalar_min_max_lsh(dst_reg
, &src_reg
);
7920 if (umax_val
>= insn_bitness
) {
7921 /* Shifts greater than 31 or 63 are undefined.
7922 * This includes shifts by a negative number.
7924 mark_reg_unknown(env
, regs
, insn
->dst_reg
);
7928 scalar32_min_max_rsh(dst_reg
, &src_reg
);
7930 scalar_min_max_rsh(dst_reg
, &src_reg
);
7933 if (umax_val
>= insn_bitness
) {
7934 /* Shifts greater than 31 or 63 are undefined.
7935 * This includes shifts by a negative number.
7937 mark_reg_unknown(env
, regs
, insn
->dst_reg
);
7941 scalar32_min_max_arsh(dst_reg
, &src_reg
);
7943 scalar_min_max_arsh(dst_reg
, &src_reg
);
7946 mark_reg_unknown(env
, regs
, insn
->dst_reg
);
7950 /* ALU32 ops are zero extended into 64bit register */
7952 zext_32_to_64(dst_reg
);
7954 __update_reg_bounds(dst_reg
);
7955 __reg_deduce_bounds(dst_reg
);
7956 __reg_bound_offset(dst_reg
);
7960 /* Handles ALU ops other than BPF_END, BPF_NEG and BPF_MOV: computes new min/max
7963 static int adjust_reg_min_max_vals(struct bpf_verifier_env
*env
,
7964 struct bpf_insn
*insn
)
7966 struct bpf_verifier_state
*vstate
= env
->cur_state
;
7967 struct bpf_func_state
*state
= vstate
->frame
[vstate
->curframe
];
7968 struct bpf_reg_state
*regs
= state
->regs
, *dst_reg
, *src_reg
;
7969 struct bpf_reg_state
*ptr_reg
= NULL
, off_reg
= {0};
7970 u8 opcode
= BPF_OP(insn
->code
);
7973 dst_reg
= ®s
[insn
->dst_reg
];
7975 if (dst_reg
->type
!= SCALAR_VALUE
)
7978 /* Make sure ID is cleared otherwise dst_reg min/max could be
7979 * incorrectly propagated into other registers by find_equal_scalars()
7982 if (BPF_SRC(insn
->code
) == BPF_X
) {
7983 src_reg
= ®s
[insn
->src_reg
];
7984 if (src_reg
->type
!= SCALAR_VALUE
) {
7985 if (dst_reg
->type
!= SCALAR_VALUE
) {
7986 /* Combining two pointers by any ALU op yields
7987 * an arbitrary scalar. Disallow all math except
7988 * pointer subtraction
7990 if (opcode
== BPF_SUB
&& env
->allow_ptr_leaks
) {
7991 mark_reg_unknown(env
, regs
, insn
->dst_reg
);
7994 verbose(env
, "R%d pointer %s pointer prohibited\n",
7996 bpf_alu_string
[opcode
>> 4]);
7999 /* scalar += pointer
8000 * This is legal, but we have to reverse our
8001 * src/dest handling in computing the range
8003 err
= mark_chain_precision(env
, insn
->dst_reg
);
8006 return adjust_ptr_min_max_vals(env
, insn
,
8009 } else if (ptr_reg
) {
8010 /* pointer += scalar */
8011 err
= mark_chain_precision(env
, insn
->src_reg
);
8014 return adjust_ptr_min_max_vals(env
, insn
,
8018 /* Pretend the src is a reg with a known value, since we only
8019 * need to be able to read from this state.
8021 off_reg
.type
= SCALAR_VALUE
;
8022 __mark_reg_known(&off_reg
, insn
->imm
);
8024 if (ptr_reg
) /* pointer += K */
8025 return adjust_ptr_min_max_vals(env
, insn
,
8029 /* Got here implies adding two SCALAR_VALUEs */
8030 if (WARN_ON_ONCE(ptr_reg
)) {
8031 print_verifier_state(env
, state
);
8032 verbose(env
, "verifier internal error: unexpected ptr_reg\n");
8035 if (WARN_ON(!src_reg
)) {
8036 print_verifier_state(env
, state
);
8037 verbose(env
, "verifier internal error: no src_reg\n");
8040 return adjust_scalar_min_max_vals(env
, insn
, dst_reg
, *src_reg
);
8043 /* check validity of 32-bit and 64-bit arithmetic operations */
8044 static int check_alu_op(struct bpf_verifier_env
*env
, struct bpf_insn
*insn
)
8046 struct bpf_reg_state
*regs
= cur_regs(env
);
8047 u8 opcode
= BPF_OP(insn
->code
);
8050 if (opcode
== BPF_END
|| opcode
== BPF_NEG
) {
8051 if (opcode
== BPF_NEG
) {
8052 if (BPF_SRC(insn
->code
) != 0 ||
8053 insn
->src_reg
!= BPF_REG_0
||
8054 insn
->off
!= 0 || insn
->imm
!= 0) {
8055 verbose(env
, "BPF_NEG uses reserved fields\n");
8059 if (insn
->src_reg
!= BPF_REG_0
|| insn
->off
!= 0 ||
8060 (insn
->imm
!= 16 && insn
->imm
!= 32 && insn
->imm
!= 64) ||
8061 BPF_CLASS(insn
->code
) == BPF_ALU64
) {
8062 verbose(env
, "BPF_END uses reserved fields\n");
8067 /* check src operand */
8068 err
= check_reg_arg(env
, insn
->dst_reg
, SRC_OP
);
8072 if (is_pointer_value(env
, insn
->dst_reg
)) {
8073 verbose(env
, "R%d pointer arithmetic prohibited\n",
8078 /* check dest operand */
8079 err
= check_reg_arg(env
, insn
->dst_reg
, DST_OP
);
8083 } else if (opcode
== BPF_MOV
) {
8085 if (BPF_SRC(insn
->code
) == BPF_X
) {
8086 if (insn
->imm
!= 0 || insn
->off
!= 0) {
8087 verbose(env
, "BPF_MOV uses reserved fields\n");
8091 /* check src operand */
8092 err
= check_reg_arg(env
, insn
->src_reg
, SRC_OP
);
8096 if (insn
->src_reg
!= BPF_REG_0
|| insn
->off
!= 0) {
8097 verbose(env
, "BPF_MOV uses reserved fields\n");
8102 /* check dest operand, mark as required later */
8103 err
= check_reg_arg(env
, insn
->dst_reg
, DST_OP_NO_MARK
);
8107 if (BPF_SRC(insn
->code
) == BPF_X
) {
8108 struct bpf_reg_state
*src_reg
= regs
+ insn
->src_reg
;
8109 struct bpf_reg_state
*dst_reg
= regs
+ insn
->dst_reg
;
8111 if (BPF_CLASS(insn
->code
) == BPF_ALU64
) {
8113 * copy register state to dest reg
8115 if (src_reg
->type
== SCALAR_VALUE
&& !src_reg
->id
)
8116 /* Assign src and dst registers the same ID
8117 * that will be used by find_equal_scalars()
8118 * to propagate min/max range.
8120 src_reg
->id
= ++env
->id_gen
;
8121 *dst_reg
= *src_reg
;
8122 dst_reg
->live
|= REG_LIVE_WRITTEN
;
8123 dst_reg
->subreg_def
= DEF_NOT_SUBREG
;
8126 if (is_pointer_value(env
, insn
->src_reg
)) {
8128 "R%d partial copy of pointer\n",
8131 } else if (src_reg
->type
== SCALAR_VALUE
) {
8132 *dst_reg
= *src_reg
;
8133 /* Make sure ID is cleared otherwise
8134 * dst_reg min/max could be incorrectly
8135 * propagated into src_reg by find_equal_scalars()
8138 dst_reg
->live
|= REG_LIVE_WRITTEN
;
8139 dst_reg
->subreg_def
= env
->insn_idx
+ 1;
8141 mark_reg_unknown(env
, regs
,
8144 zext_32_to_64(dst_reg
);
8146 __update_reg_bounds(dst_reg
);
8147 __reg_deduce_bounds(dst_reg
);
8148 __reg_bound_offset(dst_reg
);
8152 * remember the value we stored into this reg
8154 /* clear any state __mark_reg_known doesn't set */
8155 mark_reg_unknown(env
, regs
, insn
->dst_reg
);
8156 regs
[insn
->dst_reg
].type
= SCALAR_VALUE
;
8157 if (BPF_CLASS(insn
->code
) == BPF_ALU64
) {
8158 __mark_reg_known(regs
+ insn
->dst_reg
,
8161 __mark_reg_known(regs
+ insn
->dst_reg
,
8166 } else if (opcode
> BPF_END
) {
8167 verbose(env
, "invalid BPF_ALU opcode %x\n", opcode
);
8170 } else { /* all other ALU ops: and, sub, xor, add, ... */
8172 if (BPF_SRC(insn
->code
) == BPF_X
) {
8173 if (insn
->imm
!= 0 || insn
->off
!= 0) {
8174 verbose(env
, "BPF_ALU uses reserved fields\n");
8177 /* check src1 operand */
8178 err
= check_reg_arg(env
, insn
->src_reg
, SRC_OP
);
8182 if (insn
->src_reg
!= BPF_REG_0
|| insn
->off
!= 0) {
8183 verbose(env
, "BPF_ALU uses reserved fields\n");
8188 /* check src2 operand */
8189 err
= check_reg_arg(env
, insn
->dst_reg
, SRC_OP
);
8193 if ((opcode
== BPF_MOD
|| opcode
== BPF_DIV
) &&
8194 BPF_SRC(insn
->code
) == BPF_K
&& insn
->imm
== 0) {
8195 verbose(env
, "div by zero\n");
8199 if ((opcode
== BPF_LSH
|| opcode
== BPF_RSH
||
8200 opcode
== BPF_ARSH
) && BPF_SRC(insn
->code
) == BPF_K
) {
8201 int size
= BPF_CLASS(insn
->code
) == BPF_ALU64
? 64 : 32;
8203 if (insn
->imm
< 0 || insn
->imm
>= size
) {
8204 verbose(env
, "invalid shift %d\n", insn
->imm
);
8209 /* check dest operand */
8210 err
= check_reg_arg(env
, insn
->dst_reg
, DST_OP_NO_MARK
);
8214 return adjust_reg_min_max_vals(env
, insn
);
8220 static void __find_good_pkt_pointers(struct bpf_func_state
*state
,
8221 struct bpf_reg_state
*dst_reg
,
8222 enum bpf_reg_type type
, int new_range
)
8224 struct bpf_reg_state
*reg
;
8227 for (i
= 0; i
< MAX_BPF_REG
; i
++) {
8228 reg
= &state
->regs
[i
];
8229 if (reg
->type
== type
&& reg
->id
== dst_reg
->id
)
8230 /* keep the maximum range already checked */
8231 reg
->range
= max(reg
->range
, new_range
);
8234 bpf_for_each_spilled_reg(i
, state
, reg
) {
8237 if (reg
->type
== type
&& reg
->id
== dst_reg
->id
)
8238 reg
->range
= max(reg
->range
, new_range
);
8242 static void find_good_pkt_pointers(struct bpf_verifier_state
*vstate
,
8243 struct bpf_reg_state
*dst_reg
,
8244 enum bpf_reg_type type
,
8245 bool range_right_open
)
8249 if (dst_reg
->off
< 0 ||
8250 (dst_reg
->off
== 0 && range_right_open
))
8251 /* This doesn't give us any range */
8254 if (dst_reg
->umax_value
> MAX_PACKET_OFF
||
8255 dst_reg
->umax_value
+ dst_reg
->off
> MAX_PACKET_OFF
)
8256 /* Risk of overflow. For instance, ptr + (1<<63) may be less
8257 * than pkt_end, but that's because it's also less than pkt.
8261 new_range
= dst_reg
->off
;
8262 if (range_right_open
)
8265 /* Examples for register markings:
8267 * pkt_data in dst register:
8271 * if (r2 > pkt_end) goto <handle exception>
8276 * if (r2 < pkt_end) goto <access okay>
8277 * <handle exception>
8280 * r2 == dst_reg, pkt_end == src_reg
8281 * r2=pkt(id=n,off=8,r=0)
8282 * r3=pkt(id=n,off=0,r=0)
8284 * pkt_data in src register:
8288 * if (pkt_end >= r2) goto <access okay>
8289 * <handle exception>
8293 * if (pkt_end <= r2) goto <handle exception>
8297 * pkt_end == dst_reg, r2 == src_reg
8298 * r2=pkt(id=n,off=8,r=0)
8299 * r3=pkt(id=n,off=0,r=0)
8301 * Find register r3 and mark its range as r3=pkt(id=n,off=0,r=8)
8302 * or r3=pkt(id=n,off=0,r=8-1), so that range of bytes [r3, r3 + 8)
8303 * and [r3, r3 + 8-1) respectively is safe to access depending on
8307 /* If our ids match, then we must have the same max_value. And we
8308 * don't care about the other reg's fixed offset, since if it's too big
8309 * the range won't allow anything.
8310 * dst_reg->off is known < MAX_PACKET_OFF, therefore it fits in a u16.
8312 for (i
= 0; i
<= vstate
->curframe
; i
++)
8313 __find_good_pkt_pointers(vstate
->frame
[i
], dst_reg
, type
,
8317 static int is_branch32_taken(struct bpf_reg_state
*reg
, u32 val
, u8 opcode
)
8319 struct tnum subreg
= tnum_subreg(reg
->var_off
);
8320 s32 sval
= (s32
)val
;
8324 if (tnum_is_const(subreg
))
8325 return !!tnum_equals_const(subreg
, val
);
8328 if (tnum_is_const(subreg
))
8329 return !tnum_equals_const(subreg
, val
);
8332 if ((~subreg
.mask
& subreg
.value
) & val
)
8334 if (!((subreg
.mask
| subreg
.value
) & val
))
8338 if (reg
->u32_min_value
> val
)
8340 else if (reg
->u32_max_value
<= val
)
8344 if (reg
->s32_min_value
> sval
)
8346 else if (reg
->s32_max_value
<= sval
)
8350 if (reg
->u32_max_value
< val
)
8352 else if (reg
->u32_min_value
>= val
)
8356 if (reg
->s32_max_value
< sval
)
8358 else if (reg
->s32_min_value
>= sval
)
8362 if (reg
->u32_min_value
>= val
)
8364 else if (reg
->u32_max_value
< val
)
8368 if (reg
->s32_min_value
>= sval
)
8370 else if (reg
->s32_max_value
< sval
)
8374 if (reg
->u32_max_value
<= val
)
8376 else if (reg
->u32_min_value
> val
)
8380 if (reg
->s32_max_value
<= sval
)
8382 else if (reg
->s32_min_value
> sval
)
8391 static int is_branch64_taken(struct bpf_reg_state
*reg
, u64 val
, u8 opcode
)
8393 s64 sval
= (s64
)val
;
8397 if (tnum_is_const(reg
->var_off
))
8398 return !!tnum_equals_const(reg
->var_off
, val
);
8401 if (tnum_is_const(reg
->var_off
))
8402 return !tnum_equals_const(reg
->var_off
, val
);
8405 if ((~reg
->var_off
.mask
& reg
->var_off
.value
) & val
)
8407 if (!((reg
->var_off
.mask
| reg
->var_off
.value
) & val
))
8411 if (reg
->umin_value
> val
)
8413 else if (reg
->umax_value
<= val
)
8417 if (reg
->smin_value
> sval
)
8419 else if (reg
->smax_value
<= sval
)
8423 if (reg
->umax_value
< val
)
8425 else if (reg
->umin_value
>= val
)
8429 if (reg
->smax_value
< sval
)
8431 else if (reg
->smin_value
>= sval
)
8435 if (reg
->umin_value
>= val
)
8437 else if (reg
->umax_value
< val
)
8441 if (reg
->smin_value
>= sval
)
8443 else if (reg
->smax_value
< sval
)
8447 if (reg
->umax_value
<= val
)
8449 else if (reg
->umin_value
> val
)
8453 if (reg
->smax_value
<= sval
)
8455 else if (reg
->smin_value
> sval
)
8463 /* compute branch direction of the expression "if (reg opcode val) goto target;"
8465 * 1 - branch will be taken and "goto target" will be executed
8466 * 0 - branch will not be taken and fall-through to next insn
8467 * -1 - unknown. Example: "if (reg < 5)" is unknown when register value
8470 static int is_branch_taken(struct bpf_reg_state
*reg
, u64 val
, u8 opcode
,
8473 if (__is_pointer_value(false, reg
)) {
8474 if (!reg_type_not_null(reg
->type
))
8477 /* If pointer is valid tests against zero will fail so we can
8478 * use this to direct branch taken.
8494 return is_branch32_taken(reg
, val
, opcode
);
8495 return is_branch64_taken(reg
, val
, opcode
);
8498 static int flip_opcode(u32 opcode
)
8500 /* How can we transform "a <op> b" into "b <op> a"? */
8501 static const u8 opcode_flip
[16] = {
8502 /* these stay the same */
8503 [BPF_JEQ
>> 4] = BPF_JEQ
,
8504 [BPF_JNE
>> 4] = BPF_JNE
,
8505 [BPF_JSET
>> 4] = BPF_JSET
,
8506 /* these swap "lesser" and "greater" (L and G in the opcodes) */
8507 [BPF_JGE
>> 4] = BPF_JLE
,
8508 [BPF_JGT
>> 4] = BPF_JLT
,
8509 [BPF_JLE
>> 4] = BPF_JGE
,
8510 [BPF_JLT
>> 4] = BPF_JGT
,
8511 [BPF_JSGE
>> 4] = BPF_JSLE
,
8512 [BPF_JSGT
>> 4] = BPF_JSLT
,
8513 [BPF_JSLE
>> 4] = BPF_JSGE
,
8514 [BPF_JSLT
>> 4] = BPF_JSGT
8516 return opcode_flip
[opcode
>> 4];
8519 static int is_pkt_ptr_branch_taken(struct bpf_reg_state
*dst_reg
,
8520 struct bpf_reg_state
*src_reg
,
8523 struct bpf_reg_state
*pkt
;
8525 if (src_reg
->type
== PTR_TO_PACKET_END
) {
8527 } else if (dst_reg
->type
== PTR_TO_PACKET_END
) {
8529 opcode
= flip_opcode(opcode
);
8534 if (pkt
->range
>= 0)
8539 /* pkt <= pkt_end */
8543 if (pkt
->range
== BEYOND_PKT_END
)
8544 /* pkt has at last one extra byte beyond pkt_end */
8545 return opcode
== BPF_JGT
;
8551 /* pkt >= pkt_end */
8552 if (pkt
->range
== BEYOND_PKT_END
|| pkt
->range
== AT_PKT_END
)
8553 return opcode
== BPF_JGE
;
8559 /* Adjusts the register min/max values in the case that the dst_reg is the
8560 * variable register that we are working on, and src_reg is a constant or we're
8561 * simply doing a BPF_K check.
8562 * In JEQ/JNE cases we also adjust the var_off values.
8564 static void reg_set_min_max(struct bpf_reg_state
*true_reg
,
8565 struct bpf_reg_state
*false_reg
,
8567 u8 opcode
, bool is_jmp32
)
8569 struct tnum false_32off
= tnum_subreg(false_reg
->var_off
);
8570 struct tnum false_64off
= false_reg
->var_off
;
8571 struct tnum true_32off
= tnum_subreg(true_reg
->var_off
);
8572 struct tnum true_64off
= true_reg
->var_off
;
8573 s64 sval
= (s64
)val
;
8574 s32 sval32
= (s32
)val32
;
8576 /* If the dst_reg is a pointer, we can't learn anything about its
8577 * variable offset from the compare (unless src_reg were a pointer into
8578 * the same object, but we don't bother with that.
8579 * Since false_reg and true_reg have the same type by construction, we
8580 * only need to check one of them for pointerness.
8582 if (__is_pointer_value(false, false_reg
))
8589 struct bpf_reg_state
*reg
=
8590 opcode
== BPF_JEQ
? true_reg
: false_reg
;
8592 /* JEQ/JNE comparison doesn't change the register equivalence.
8594 * if (r1 == 42) goto label;
8596 * label: // here both r1 and r2 are known to be 42.
8598 * Hence when marking register as known preserve it's ID.
8601 __mark_reg32_known(reg
, val32
);
8603 ___mark_reg_known(reg
, val
);
8608 false_32off
= tnum_and(false_32off
, tnum_const(~val32
));
8609 if (is_power_of_2(val32
))
8610 true_32off
= tnum_or(true_32off
,
8613 false_64off
= tnum_and(false_64off
, tnum_const(~val
));
8614 if (is_power_of_2(val
))
8615 true_64off
= tnum_or(true_64off
,
8623 u32 false_umax
= opcode
== BPF_JGT
? val32
: val32
- 1;
8624 u32 true_umin
= opcode
== BPF_JGT
? val32
+ 1 : val32
;
8626 false_reg
->u32_max_value
= min(false_reg
->u32_max_value
,
8628 true_reg
->u32_min_value
= max(true_reg
->u32_min_value
,
8631 u64 false_umax
= opcode
== BPF_JGT
? val
: val
- 1;
8632 u64 true_umin
= opcode
== BPF_JGT
? val
+ 1 : val
;
8634 false_reg
->umax_value
= min(false_reg
->umax_value
, false_umax
);
8635 true_reg
->umin_value
= max(true_reg
->umin_value
, true_umin
);
8643 s32 false_smax
= opcode
== BPF_JSGT
? sval32
: sval32
- 1;
8644 s32 true_smin
= opcode
== BPF_JSGT
? sval32
+ 1 : sval32
;
8646 false_reg
->s32_max_value
= min(false_reg
->s32_max_value
, false_smax
);
8647 true_reg
->s32_min_value
= max(true_reg
->s32_min_value
, true_smin
);
8649 s64 false_smax
= opcode
== BPF_JSGT
? sval
: sval
- 1;
8650 s64 true_smin
= opcode
== BPF_JSGT
? sval
+ 1 : sval
;
8652 false_reg
->smax_value
= min(false_reg
->smax_value
, false_smax
);
8653 true_reg
->smin_value
= max(true_reg
->smin_value
, true_smin
);
8661 u32 false_umin
= opcode
== BPF_JLT
? val32
: val32
+ 1;
8662 u32 true_umax
= opcode
== BPF_JLT
? val32
- 1 : val32
;
8664 false_reg
->u32_min_value
= max(false_reg
->u32_min_value
,
8666 true_reg
->u32_max_value
= min(true_reg
->u32_max_value
,
8669 u64 false_umin
= opcode
== BPF_JLT
? val
: val
+ 1;
8670 u64 true_umax
= opcode
== BPF_JLT
? val
- 1 : val
;
8672 false_reg
->umin_value
= max(false_reg
->umin_value
, false_umin
);
8673 true_reg
->umax_value
= min(true_reg
->umax_value
, true_umax
);
8681 s32 false_smin
= opcode
== BPF_JSLT
? sval32
: sval32
+ 1;
8682 s32 true_smax
= opcode
== BPF_JSLT
? sval32
- 1 : sval32
;
8684 false_reg
->s32_min_value
= max(false_reg
->s32_min_value
, false_smin
);
8685 true_reg
->s32_max_value
= min(true_reg
->s32_max_value
, true_smax
);
8687 s64 false_smin
= opcode
== BPF_JSLT
? sval
: sval
+ 1;
8688 s64 true_smax
= opcode
== BPF_JSLT
? sval
- 1 : sval
;
8690 false_reg
->smin_value
= max(false_reg
->smin_value
, false_smin
);
8691 true_reg
->smax_value
= min(true_reg
->smax_value
, true_smax
);
8700 false_reg
->var_off
= tnum_or(tnum_clear_subreg(false_64off
),
8701 tnum_subreg(false_32off
));
8702 true_reg
->var_off
= tnum_or(tnum_clear_subreg(true_64off
),
8703 tnum_subreg(true_32off
));
8704 __reg_combine_32_into_64(false_reg
);
8705 __reg_combine_32_into_64(true_reg
);
8707 false_reg
->var_off
= false_64off
;
8708 true_reg
->var_off
= true_64off
;
8709 __reg_combine_64_into_32(false_reg
);
8710 __reg_combine_64_into_32(true_reg
);
8714 /* Same as above, but for the case that dst_reg holds a constant and src_reg is
8717 static void reg_set_min_max_inv(struct bpf_reg_state
*true_reg
,
8718 struct bpf_reg_state
*false_reg
,
8720 u8 opcode
, bool is_jmp32
)
8722 opcode
= flip_opcode(opcode
);
8723 /* This uses zero as "not present in table"; luckily the zero opcode,
8724 * BPF_JA, can't get here.
8727 reg_set_min_max(true_reg
, false_reg
, val
, val32
, opcode
, is_jmp32
);
8730 /* Regs are known to be equal, so intersect their min/max/var_off */
8731 static void __reg_combine_min_max(struct bpf_reg_state
*src_reg
,
8732 struct bpf_reg_state
*dst_reg
)
8734 src_reg
->umin_value
= dst_reg
->umin_value
= max(src_reg
->umin_value
,
8735 dst_reg
->umin_value
);
8736 src_reg
->umax_value
= dst_reg
->umax_value
= min(src_reg
->umax_value
,
8737 dst_reg
->umax_value
);
8738 src_reg
->smin_value
= dst_reg
->smin_value
= max(src_reg
->smin_value
,
8739 dst_reg
->smin_value
);
8740 src_reg
->smax_value
= dst_reg
->smax_value
= min(src_reg
->smax_value
,
8741 dst_reg
->smax_value
);
8742 src_reg
->var_off
= dst_reg
->var_off
= tnum_intersect(src_reg
->var_off
,
8744 /* We might have learned new bounds from the var_off. */
8745 __update_reg_bounds(src_reg
);
8746 __update_reg_bounds(dst_reg
);
8747 /* We might have learned something about the sign bit. */
8748 __reg_deduce_bounds(src_reg
);
8749 __reg_deduce_bounds(dst_reg
);
8750 /* We might have learned some bits from the bounds. */
8751 __reg_bound_offset(src_reg
);
8752 __reg_bound_offset(dst_reg
);
8753 /* Intersecting with the old var_off might have improved our bounds
8754 * slightly. e.g. if umax was 0x7f...f and var_off was (0; 0xf...fc),
8755 * then new var_off is (0; 0x7f...fc) which improves our umax.
8757 __update_reg_bounds(src_reg
);
8758 __update_reg_bounds(dst_reg
);
8761 static void reg_combine_min_max(struct bpf_reg_state
*true_src
,
8762 struct bpf_reg_state
*true_dst
,
8763 struct bpf_reg_state
*false_src
,
8764 struct bpf_reg_state
*false_dst
,
8769 __reg_combine_min_max(true_src
, true_dst
);
8772 __reg_combine_min_max(false_src
, false_dst
);
8777 static void mark_ptr_or_null_reg(struct bpf_func_state
*state
,
8778 struct bpf_reg_state
*reg
, u32 id
,
8781 if (reg_type_may_be_null(reg
->type
) && reg
->id
== id
&&
8782 !WARN_ON_ONCE(!reg
->id
)) {
8783 if (WARN_ON_ONCE(reg
->smin_value
|| reg
->smax_value
||
8784 !tnum_equals_const(reg
->var_off
, 0) ||
8786 /* Old offset (both fixed and variable parts) should
8787 * have been known-zero, because we don't allow pointer
8788 * arithmetic on pointers that might be NULL. If we
8789 * see this happening, don't convert the register.
8794 reg
->type
= SCALAR_VALUE
;
8795 /* We don't need id and ref_obj_id from this point
8796 * onwards anymore, thus we should better reset it,
8797 * so that state pruning has chances to take effect.
8800 reg
->ref_obj_id
= 0;
8805 mark_ptr_not_null_reg(reg
);
8807 if (!reg_may_point_to_spin_lock(reg
)) {
8808 /* For not-NULL ptr, reg->ref_obj_id will be reset
8809 * in release_reg_references().
8811 * reg->id is still used by spin_lock ptr. Other
8812 * than spin_lock ptr type, reg->id can be reset.
8819 static void __mark_ptr_or_null_regs(struct bpf_func_state
*state
, u32 id
,
8822 struct bpf_reg_state
*reg
;
8825 for (i
= 0; i
< MAX_BPF_REG
; i
++)
8826 mark_ptr_or_null_reg(state
, &state
->regs
[i
], id
, is_null
);
8828 bpf_for_each_spilled_reg(i
, state
, reg
) {
8831 mark_ptr_or_null_reg(state
, reg
, id
, is_null
);
8835 /* The logic is similar to find_good_pkt_pointers(), both could eventually
8836 * be folded together at some point.
8838 static void mark_ptr_or_null_regs(struct bpf_verifier_state
*vstate
, u32 regno
,
8841 struct bpf_func_state
*state
= vstate
->frame
[vstate
->curframe
];
8842 struct bpf_reg_state
*regs
= state
->regs
;
8843 u32 ref_obj_id
= regs
[regno
].ref_obj_id
;
8844 u32 id
= regs
[regno
].id
;
8847 if (ref_obj_id
&& ref_obj_id
== id
&& is_null
)
8848 /* regs[regno] is in the " == NULL" branch.
8849 * No one could have freed the reference state before
8850 * doing the NULL check.
8852 WARN_ON_ONCE(release_reference_state(state
, id
));
8854 for (i
= 0; i
<= vstate
->curframe
; i
++)
8855 __mark_ptr_or_null_regs(vstate
->frame
[i
], id
, is_null
);
8858 static bool try_match_pkt_pointers(const struct bpf_insn
*insn
,
8859 struct bpf_reg_state
*dst_reg
,
8860 struct bpf_reg_state
*src_reg
,
8861 struct bpf_verifier_state
*this_branch
,
8862 struct bpf_verifier_state
*other_branch
)
8864 if (BPF_SRC(insn
->code
) != BPF_X
)
8867 /* Pointers are always 64-bit. */
8868 if (BPF_CLASS(insn
->code
) == BPF_JMP32
)
8871 switch (BPF_OP(insn
->code
)) {
8873 if ((dst_reg
->type
== PTR_TO_PACKET
&&
8874 src_reg
->type
== PTR_TO_PACKET_END
) ||
8875 (dst_reg
->type
== PTR_TO_PACKET_META
&&
8876 reg_is_init_pkt_pointer(src_reg
, PTR_TO_PACKET
))) {
8877 /* pkt_data' > pkt_end, pkt_meta' > pkt_data */
8878 find_good_pkt_pointers(this_branch
, dst_reg
,
8879 dst_reg
->type
, false);
8880 mark_pkt_end(other_branch
, insn
->dst_reg
, true);
8881 } else if ((dst_reg
->type
== PTR_TO_PACKET_END
&&
8882 src_reg
->type
== PTR_TO_PACKET
) ||
8883 (reg_is_init_pkt_pointer(dst_reg
, PTR_TO_PACKET
) &&
8884 src_reg
->type
== PTR_TO_PACKET_META
)) {
8885 /* pkt_end > pkt_data', pkt_data > pkt_meta' */
8886 find_good_pkt_pointers(other_branch
, src_reg
,
8887 src_reg
->type
, true);
8888 mark_pkt_end(this_branch
, insn
->src_reg
, false);
8894 if ((dst_reg
->type
== PTR_TO_PACKET
&&
8895 src_reg
->type
== PTR_TO_PACKET_END
) ||
8896 (dst_reg
->type
== PTR_TO_PACKET_META
&&
8897 reg_is_init_pkt_pointer(src_reg
, PTR_TO_PACKET
))) {
8898 /* pkt_data' < pkt_end, pkt_meta' < pkt_data */
8899 find_good_pkt_pointers(other_branch
, dst_reg
,
8900 dst_reg
->type
, true);
8901 mark_pkt_end(this_branch
, insn
->dst_reg
, false);
8902 } else if ((dst_reg
->type
== PTR_TO_PACKET_END
&&
8903 src_reg
->type
== PTR_TO_PACKET
) ||
8904 (reg_is_init_pkt_pointer(dst_reg
, PTR_TO_PACKET
) &&
8905 src_reg
->type
== PTR_TO_PACKET_META
)) {
8906 /* pkt_end < pkt_data', pkt_data > pkt_meta' */
8907 find_good_pkt_pointers(this_branch
, src_reg
,
8908 src_reg
->type
, false);
8909 mark_pkt_end(other_branch
, insn
->src_reg
, true);
8915 if ((dst_reg
->type
== PTR_TO_PACKET
&&
8916 src_reg
->type
== PTR_TO_PACKET_END
) ||
8917 (dst_reg
->type
== PTR_TO_PACKET_META
&&
8918 reg_is_init_pkt_pointer(src_reg
, PTR_TO_PACKET
))) {
8919 /* pkt_data' >= pkt_end, pkt_meta' >= pkt_data */
8920 find_good_pkt_pointers(this_branch
, dst_reg
,
8921 dst_reg
->type
, true);
8922 mark_pkt_end(other_branch
, insn
->dst_reg
, false);
8923 } else if ((dst_reg
->type
== PTR_TO_PACKET_END
&&
8924 src_reg
->type
== PTR_TO_PACKET
) ||
8925 (reg_is_init_pkt_pointer(dst_reg
, PTR_TO_PACKET
) &&
8926 src_reg
->type
== PTR_TO_PACKET_META
)) {
8927 /* pkt_end >= pkt_data', pkt_data >= pkt_meta' */
8928 find_good_pkt_pointers(other_branch
, src_reg
,
8929 src_reg
->type
, false);
8930 mark_pkt_end(this_branch
, insn
->src_reg
, true);
8936 if ((dst_reg
->type
== PTR_TO_PACKET
&&
8937 src_reg
->type
== PTR_TO_PACKET_END
) ||
8938 (dst_reg
->type
== PTR_TO_PACKET_META
&&
8939 reg_is_init_pkt_pointer(src_reg
, PTR_TO_PACKET
))) {
8940 /* pkt_data' <= pkt_end, pkt_meta' <= pkt_data */
8941 find_good_pkt_pointers(other_branch
, dst_reg
,
8942 dst_reg
->type
, false);
8943 mark_pkt_end(this_branch
, insn
->dst_reg
, true);
8944 } else if ((dst_reg
->type
== PTR_TO_PACKET_END
&&
8945 src_reg
->type
== PTR_TO_PACKET
) ||
8946 (reg_is_init_pkt_pointer(dst_reg
, PTR_TO_PACKET
) &&
8947 src_reg
->type
== PTR_TO_PACKET_META
)) {
8948 /* pkt_end <= pkt_data', pkt_data <= pkt_meta' */
8949 find_good_pkt_pointers(this_branch
, src_reg
,
8950 src_reg
->type
, true);
8951 mark_pkt_end(other_branch
, insn
->src_reg
, false);
8963 static void find_equal_scalars(struct bpf_verifier_state
*vstate
,
8964 struct bpf_reg_state
*known_reg
)
8966 struct bpf_func_state
*state
;
8967 struct bpf_reg_state
*reg
;
8970 for (i
= 0; i
<= vstate
->curframe
; i
++) {
8971 state
= vstate
->frame
[i
];
8972 for (j
= 0; j
< MAX_BPF_REG
; j
++) {
8973 reg
= &state
->regs
[j
];
8974 if (reg
->type
== SCALAR_VALUE
&& reg
->id
== known_reg
->id
)
8978 bpf_for_each_spilled_reg(j
, state
, reg
) {
8981 if (reg
->type
== SCALAR_VALUE
&& reg
->id
== known_reg
->id
)
8987 static int check_cond_jmp_op(struct bpf_verifier_env
*env
,
8988 struct bpf_insn
*insn
, int *insn_idx
)
8990 struct bpf_verifier_state
*this_branch
= env
->cur_state
;
8991 struct bpf_verifier_state
*other_branch
;
8992 struct bpf_reg_state
*regs
= this_branch
->frame
[this_branch
->curframe
]->regs
;
8993 struct bpf_reg_state
*dst_reg
, *other_branch_regs
, *src_reg
= NULL
;
8994 u8 opcode
= BPF_OP(insn
->code
);
8999 /* Only conditional jumps are expected to reach here. */
9000 if (opcode
== BPF_JA
|| opcode
> BPF_JSLE
) {
9001 verbose(env
, "invalid BPF_JMP/JMP32 opcode %x\n", opcode
);
9005 if (BPF_SRC(insn
->code
) == BPF_X
) {
9006 if (insn
->imm
!= 0) {
9007 verbose(env
, "BPF_JMP/JMP32 uses reserved fields\n");
9011 /* check src1 operand */
9012 err
= check_reg_arg(env
, insn
->src_reg
, SRC_OP
);
9016 if (is_pointer_value(env
, insn
->src_reg
)) {
9017 verbose(env
, "R%d pointer comparison prohibited\n",
9021 src_reg
= ®s
[insn
->src_reg
];
9023 if (insn
->src_reg
!= BPF_REG_0
) {
9024 verbose(env
, "BPF_JMP/JMP32 uses reserved fields\n");
9029 /* check src2 operand */
9030 err
= check_reg_arg(env
, insn
->dst_reg
, SRC_OP
);
9034 dst_reg
= ®s
[insn
->dst_reg
];
9035 is_jmp32
= BPF_CLASS(insn
->code
) == BPF_JMP32
;
9037 if (BPF_SRC(insn
->code
) == BPF_K
) {
9038 pred
= is_branch_taken(dst_reg
, insn
->imm
, opcode
, is_jmp32
);
9039 } else if (src_reg
->type
== SCALAR_VALUE
&&
9040 is_jmp32
&& tnum_is_const(tnum_subreg(src_reg
->var_off
))) {
9041 pred
= is_branch_taken(dst_reg
,
9042 tnum_subreg(src_reg
->var_off
).value
,
9045 } else if (src_reg
->type
== SCALAR_VALUE
&&
9046 !is_jmp32
&& tnum_is_const(src_reg
->var_off
)) {
9047 pred
= is_branch_taken(dst_reg
,
9048 src_reg
->var_off
.value
,
9051 } else if (reg_is_pkt_pointer_any(dst_reg
) &&
9052 reg_is_pkt_pointer_any(src_reg
) &&
9054 pred
= is_pkt_ptr_branch_taken(dst_reg
, src_reg
, opcode
);
9058 /* If we get here with a dst_reg pointer type it is because
9059 * above is_branch_taken() special cased the 0 comparison.
9061 if (!__is_pointer_value(false, dst_reg
))
9062 err
= mark_chain_precision(env
, insn
->dst_reg
);
9063 if (BPF_SRC(insn
->code
) == BPF_X
&& !err
&&
9064 !__is_pointer_value(false, src_reg
))
9065 err
= mark_chain_precision(env
, insn
->src_reg
);
9071 /* Only follow the goto, ignore fall-through. If needed, push
9072 * the fall-through branch for simulation under speculative
9075 if (!env
->bypass_spec_v1
&&
9076 !sanitize_speculative_path(env
, insn
, *insn_idx
+ 1,
9079 *insn_idx
+= insn
->off
;
9081 } else if (pred
== 0) {
9082 /* Only follow the fall-through branch, since that's where the
9083 * program will go. If needed, push the goto branch for
9084 * simulation under speculative execution.
9086 if (!env
->bypass_spec_v1
&&
9087 !sanitize_speculative_path(env
, insn
,
9088 *insn_idx
+ insn
->off
+ 1,
9094 other_branch
= push_stack(env
, *insn_idx
+ insn
->off
+ 1, *insn_idx
,
9098 other_branch_regs
= other_branch
->frame
[other_branch
->curframe
]->regs
;
9100 /* detect if we are comparing against a constant value so we can adjust
9101 * our min/max values for our dst register.
9102 * this is only legit if both are scalars (or pointers to the same
9103 * object, I suppose, but we don't support that right now), because
9104 * otherwise the different base pointers mean the offsets aren't
9107 if (BPF_SRC(insn
->code
) == BPF_X
) {
9108 struct bpf_reg_state
*src_reg
= ®s
[insn
->src_reg
];
9110 if (dst_reg
->type
== SCALAR_VALUE
&&
9111 src_reg
->type
== SCALAR_VALUE
) {
9112 if (tnum_is_const(src_reg
->var_off
) ||
9114 tnum_is_const(tnum_subreg(src_reg
->var_off
))))
9115 reg_set_min_max(&other_branch_regs
[insn
->dst_reg
],
9117 src_reg
->var_off
.value
,
9118 tnum_subreg(src_reg
->var_off
).value
,
9120 else if (tnum_is_const(dst_reg
->var_off
) ||
9122 tnum_is_const(tnum_subreg(dst_reg
->var_off
))))
9123 reg_set_min_max_inv(&other_branch_regs
[insn
->src_reg
],
9125 dst_reg
->var_off
.value
,
9126 tnum_subreg(dst_reg
->var_off
).value
,
9128 else if (!is_jmp32
&&
9129 (opcode
== BPF_JEQ
|| opcode
== BPF_JNE
))
9130 /* Comparing for equality, we can combine knowledge */
9131 reg_combine_min_max(&other_branch_regs
[insn
->src_reg
],
9132 &other_branch_regs
[insn
->dst_reg
],
9133 src_reg
, dst_reg
, opcode
);
9135 !WARN_ON_ONCE(src_reg
->id
!= other_branch_regs
[insn
->src_reg
].id
)) {
9136 find_equal_scalars(this_branch
, src_reg
);
9137 find_equal_scalars(other_branch
, &other_branch_regs
[insn
->src_reg
]);
9141 } else if (dst_reg
->type
== SCALAR_VALUE
) {
9142 reg_set_min_max(&other_branch_regs
[insn
->dst_reg
],
9143 dst_reg
, insn
->imm
, (u32
)insn
->imm
,
9147 if (dst_reg
->type
== SCALAR_VALUE
&& dst_reg
->id
&&
9148 !WARN_ON_ONCE(dst_reg
->id
!= other_branch_regs
[insn
->dst_reg
].id
)) {
9149 find_equal_scalars(this_branch
, dst_reg
);
9150 find_equal_scalars(other_branch
, &other_branch_regs
[insn
->dst_reg
]);
9153 /* detect if R == 0 where R is returned from bpf_map_lookup_elem().
9154 * NOTE: these optimizations below are related with pointer comparison
9155 * which will never be JMP32.
9157 if (!is_jmp32
&& BPF_SRC(insn
->code
) == BPF_K
&&
9158 insn
->imm
== 0 && (opcode
== BPF_JEQ
|| opcode
== BPF_JNE
) &&
9159 reg_type_may_be_null(dst_reg
->type
)) {
9160 /* Mark all identical registers in each branch as either
9161 * safe or unknown depending R == 0 or R != 0 conditional.
9163 mark_ptr_or_null_regs(this_branch
, insn
->dst_reg
,
9165 mark_ptr_or_null_regs(other_branch
, insn
->dst_reg
,
9167 } else if (!try_match_pkt_pointers(insn
, dst_reg
, ®s
[insn
->src_reg
],
9168 this_branch
, other_branch
) &&
9169 is_pointer_value(env
, insn
->dst_reg
)) {
9170 verbose(env
, "R%d pointer comparison prohibited\n",
9174 if (env
->log
.level
& BPF_LOG_LEVEL
)
9175 print_verifier_state(env
, this_branch
->frame
[this_branch
->curframe
]);
9179 /* verify BPF_LD_IMM64 instruction */
9180 static int check_ld_imm(struct bpf_verifier_env
*env
, struct bpf_insn
*insn
)
9182 struct bpf_insn_aux_data
*aux
= cur_aux(env
);
9183 struct bpf_reg_state
*regs
= cur_regs(env
);
9184 struct bpf_reg_state
*dst_reg
;
9185 struct bpf_map
*map
;
9188 if (BPF_SIZE(insn
->code
) != BPF_DW
) {
9189 verbose(env
, "invalid BPF_LD_IMM insn\n");
9192 if (insn
->off
!= 0) {
9193 verbose(env
, "BPF_LD_IMM64 uses reserved fields\n");
9197 err
= check_reg_arg(env
, insn
->dst_reg
, DST_OP
);
9201 dst_reg
= ®s
[insn
->dst_reg
];
9202 if (insn
->src_reg
== 0) {
9203 u64 imm
= ((u64
)(insn
+ 1)->imm
<< 32) | (u32
)insn
->imm
;
9205 dst_reg
->type
= SCALAR_VALUE
;
9206 __mark_reg_known(®s
[insn
->dst_reg
], imm
);
9210 /* All special src_reg cases are listed below. From this point onwards
9211 * we either succeed and assign a corresponding dst_reg->type after
9212 * zeroing the offset, or fail and reject the program.
9214 mark_reg_known_zero(env
, regs
, insn
->dst_reg
);
9216 if (insn
->src_reg
== BPF_PSEUDO_BTF_ID
) {
9217 dst_reg
->type
= aux
->btf_var
.reg_type
;
9218 switch (dst_reg
->type
) {
9220 dst_reg
->mem_size
= aux
->btf_var
.mem_size
;
9223 case PTR_TO_PERCPU_BTF_ID
:
9224 dst_reg
->btf
= aux
->btf_var
.btf
;
9225 dst_reg
->btf_id
= aux
->btf_var
.btf_id
;
9228 verbose(env
, "bpf verifier is misconfigured\n");
9234 if (insn
->src_reg
== BPF_PSEUDO_FUNC
) {
9235 struct bpf_prog_aux
*aux
= env
->prog
->aux
;
9236 u32 subprogno
= insn
[1].imm
;
9238 if (!aux
->func_info
) {
9239 verbose(env
, "missing btf func_info\n");
9242 if (aux
->func_info_aux
[subprogno
].linkage
!= BTF_FUNC_STATIC
) {
9243 verbose(env
, "callback function not static\n");
9247 dst_reg
->type
= PTR_TO_FUNC
;
9248 dst_reg
->subprogno
= subprogno
;
9252 map
= env
->used_maps
[aux
->map_index
];
9253 dst_reg
->map_ptr
= map
;
9255 if (insn
->src_reg
== BPF_PSEUDO_MAP_VALUE
||
9256 insn
->src_reg
== BPF_PSEUDO_MAP_IDX_VALUE
) {
9257 dst_reg
->type
= PTR_TO_MAP_VALUE
;
9258 dst_reg
->off
= aux
->map_off
;
9259 if (map_value_has_spin_lock(map
))
9260 dst_reg
->id
= ++env
->id_gen
;
9261 } else if (insn
->src_reg
== BPF_PSEUDO_MAP_FD
||
9262 insn
->src_reg
== BPF_PSEUDO_MAP_IDX
) {
9263 dst_reg
->type
= CONST_PTR_TO_MAP
;
9265 verbose(env
, "bpf verifier is misconfigured\n");
9272 static bool may_access_skb(enum bpf_prog_type type
)
9275 case BPF_PROG_TYPE_SOCKET_FILTER
:
9276 case BPF_PROG_TYPE_SCHED_CLS
:
9277 case BPF_PROG_TYPE_SCHED_ACT
:
9284 /* verify safety of LD_ABS|LD_IND instructions:
9285 * - they can only appear in the programs where ctx == skb
9286 * - since they are wrappers of function calls, they scratch R1-R5 registers,
9287 * preserve R6-R9, and store return value into R0
9290 * ctx == skb == R6 == CTX
9293 * SRC == any register
9294 * IMM == 32-bit immediate
9297 * R0 - 8/16/32-bit skb data converted to cpu endianness
9299 static int check_ld_abs(struct bpf_verifier_env
*env
, struct bpf_insn
*insn
)
9301 struct bpf_reg_state
*regs
= cur_regs(env
);
9302 static const int ctx_reg
= BPF_REG_6
;
9303 u8 mode
= BPF_MODE(insn
->code
);
9306 if (!may_access_skb(resolve_prog_type(env
->prog
))) {
9307 verbose(env
, "BPF_LD_[ABS|IND] instructions not allowed for this program type\n");
9311 if (!env
->ops
->gen_ld_abs
) {
9312 verbose(env
, "bpf verifier is misconfigured\n");
9316 if (insn
->dst_reg
!= BPF_REG_0
|| insn
->off
!= 0 ||
9317 BPF_SIZE(insn
->code
) == BPF_DW
||
9318 (mode
== BPF_ABS
&& insn
->src_reg
!= BPF_REG_0
)) {
9319 verbose(env
, "BPF_LD_[ABS|IND] uses reserved fields\n");
9323 /* check whether implicit source operand (register R6) is readable */
9324 err
= check_reg_arg(env
, ctx_reg
, SRC_OP
);
9328 /* Disallow usage of BPF_LD_[ABS|IND] with reference tracking, as
9329 * gen_ld_abs() may terminate the program at runtime, leading to
9332 err
= check_reference_leak(env
);
9334 verbose(env
, "BPF_LD_[ABS|IND] cannot be mixed with socket references\n");
9338 if (env
->cur_state
->active_spin_lock
) {
9339 verbose(env
, "BPF_LD_[ABS|IND] cannot be used inside bpf_spin_lock-ed region\n");
9343 if (regs
[ctx_reg
].type
!= PTR_TO_CTX
) {
9345 "at the time of BPF_LD_ABS|IND R6 != pointer to skb\n");
9349 if (mode
== BPF_IND
) {
9350 /* check explicit source operand */
9351 err
= check_reg_arg(env
, insn
->src_reg
, SRC_OP
);
9356 err
= check_ctx_reg(env
, ®s
[ctx_reg
], ctx_reg
);
9360 /* reset caller saved regs to unreadable */
9361 for (i
= 0; i
< CALLER_SAVED_REGS
; i
++) {
9362 mark_reg_not_init(env
, regs
, caller_saved
[i
]);
9363 check_reg_arg(env
, caller_saved
[i
], DST_OP_NO_MARK
);
9366 /* mark destination R0 register as readable, since it contains
9367 * the value fetched from the packet.
9368 * Already marked as written above.
9370 mark_reg_unknown(env
, regs
, BPF_REG_0
);
9371 /* ld_abs load up to 32-bit skb data. */
9372 regs
[BPF_REG_0
].subreg_def
= env
->insn_idx
+ 1;
9376 static int check_return_code(struct bpf_verifier_env
*env
)
9378 struct tnum enforce_attach_type_range
= tnum_unknown
;
9379 const struct bpf_prog
*prog
= env
->prog
;
9380 struct bpf_reg_state
*reg
;
9381 struct tnum range
= tnum_range(0, 1);
9382 enum bpf_prog_type prog_type
= resolve_prog_type(env
->prog
);
9384 struct bpf_func_state
*frame
= env
->cur_state
->frame
[0];
9385 const bool is_subprog
= frame
->subprogno
;
9387 /* LSM and struct_ops func-ptr's return type could be "void" */
9389 (prog_type
== BPF_PROG_TYPE_STRUCT_OPS
||
9390 prog_type
== BPF_PROG_TYPE_LSM
) &&
9391 !prog
->aux
->attach_func_proto
->type
)
9394 /* eBPF calling convention is such that R0 is used
9395 * to return the value from eBPF program.
9396 * Make sure that it's readable at this time
9397 * of bpf_exit, which means that program wrote
9398 * something into it earlier
9400 err
= check_reg_arg(env
, BPF_REG_0
, SRC_OP
);
9404 if (is_pointer_value(env
, BPF_REG_0
)) {
9405 verbose(env
, "R0 leaks addr as return value\n");
9409 reg
= cur_regs(env
) + BPF_REG_0
;
9411 if (frame
->in_async_callback_fn
) {
9412 /* enforce return zero from async callbacks like timer */
9413 if (reg
->type
!= SCALAR_VALUE
) {
9414 verbose(env
, "In async callback the register R0 is not a known value (%s)\n",
9415 reg_type_str
[reg
->type
]);
9419 if (!tnum_in(tnum_const(0), reg
->var_off
)) {
9420 verbose_invalid_scalar(env
, reg
, &range
, "async callback", "R0");
9427 if (reg
->type
!= SCALAR_VALUE
) {
9428 verbose(env
, "At subprogram exit the register R0 is not a scalar value (%s)\n",
9429 reg_type_str
[reg
->type
]);
9435 switch (prog_type
) {
9436 case BPF_PROG_TYPE_CGROUP_SOCK_ADDR
:
9437 if (env
->prog
->expected_attach_type
== BPF_CGROUP_UDP4_RECVMSG
||
9438 env
->prog
->expected_attach_type
== BPF_CGROUP_UDP6_RECVMSG
||
9439 env
->prog
->expected_attach_type
== BPF_CGROUP_INET4_GETPEERNAME
||
9440 env
->prog
->expected_attach_type
== BPF_CGROUP_INET6_GETPEERNAME
||
9441 env
->prog
->expected_attach_type
== BPF_CGROUP_INET4_GETSOCKNAME
||
9442 env
->prog
->expected_attach_type
== BPF_CGROUP_INET6_GETSOCKNAME
)
9443 range
= tnum_range(1, 1);
9444 if (env
->prog
->expected_attach_type
== BPF_CGROUP_INET4_BIND
||
9445 env
->prog
->expected_attach_type
== BPF_CGROUP_INET6_BIND
)
9446 range
= tnum_range(0, 3);
9448 case BPF_PROG_TYPE_CGROUP_SKB
:
9449 if (env
->prog
->expected_attach_type
== BPF_CGROUP_INET_EGRESS
) {
9450 range
= tnum_range(0, 3);
9451 enforce_attach_type_range
= tnum_range(2, 3);
9454 case BPF_PROG_TYPE_CGROUP_SOCK
:
9455 case BPF_PROG_TYPE_SOCK_OPS
:
9456 case BPF_PROG_TYPE_CGROUP_DEVICE
:
9457 case BPF_PROG_TYPE_CGROUP_SYSCTL
:
9458 case BPF_PROG_TYPE_CGROUP_SOCKOPT
:
9460 case BPF_PROG_TYPE_RAW_TRACEPOINT
:
9461 if (!env
->prog
->aux
->attach_btf_id
)
9463 range
= tnum_const(0);
9465 case BPF_PROG_TYPE_TRACING
:
9466 switch (env
->prog
->expected_attach_type
) {
9467 case BPF_TRACE_FENTRY
:
9468 case BPF_TRACE_FEXIT
:
9469 range
= tnum_const(0);
9471 case BPF_TRACE_RAW_TP
:
9472 case BPF_MODIFY_RETURN
:
9474 case BPF_TRACE_ITER
:
9480 case BPF_PROG_TYPE_SK_LOOKUP
:
9481 range
= tnum_range(SK_DROP
, SK_PASS
);
9483 case BPF_PROG_TYPE_EXT
:
9484 /* freplace program can return anything as its return value
9485 * depends on the to-be-replaced kernel func or bpf program.
9491 if (reg
->type
!= SCALAR_VALUE
) {
9492 verbose(env
, "At program exit the register R0 is not a known value (%s)\n",
9493 reg_type_str
[reg
->type
]);
9497 if (!tnum_in(range
, reg
->var_off
)) {
9498 verbose_invalid_scalar(env
, reg
, &range
, "program exit", "R0");
9502 if (!tnum_is_unknown(enforce_attach_type_range
) &&
9503 tnum_in(enforce_attach_type_range
, reg
->var_off
))
9504 env
->prog
->enforce_expected_attach_type
= 1;
9508 /* non-recursive DFS pseudo code
9509 * 1 procedure DFS-iterative(G,v):
9510 * 2 label v as discovered
9511 * 3 let S be a stack
9513 * 5 while S is not empty
9515 * 7 if t is what we're looking for:
9517 * 9 for all edges e in G.adjacentEdges(t) do
9518 * 10 if edge e is already labelled
9519 * 11 continue with the next edge
9520 * 12 w <- G.adjacentVertex(t,e)
9521 * 13 if vertex w is not discovered and not explored
9522 * 14 label e as tree-edge
9523 * 15 label w as discovered
9526 * 18 else if vertex w is discovered
9527 * 19 label e as back-edge
9529 * 21 // vertex w is explored
9530 * 22 label e as forward- or cross-edge
9531 * 23 label t as explored
9536 * 0x11 - discovered and fall-through edge labelled
9537 * 0x12 - discovered and fall-through and branch edges labelled
9548 static u32
state_htab_size(struct bpf_verifier_env
*env
)
9550 return env
->prog
->len
;
9553 static struct bpf_verifier_state_list
**explored_state(
9554 struct bpf_verifier_env
*env
,
9557 struct bpf_verifier_state
*cur
= env
->cur_state
;
9558 struct bpf_func_state
*state
= cur
->frame
[cur
->curframe
];
9560 return &env
->explored_states
[(idx
^ state
->callsite
) % state_htab_size(env
)];
9563 static void init_explored_state(struct bpf_verifier_env
*env
, int idx
)
9565 env
->insn_aux_data
[idx
].prune_point
= true;
9573 /* t, w, e - match pseudo-code above:
9574 * t - index of current instruction
9575 * w - next instruction
9578 static int push_insn(int t
, int w
, int e
, struct bpf_verifier_env
*env
,
9581 int *insn_stack
= env
->cfg
.insn_stack
;
9582 int *insn_state
= env
->cfg
.insn_state
;
9584 if (e
== FALLTHROUGH
&& insn_state
[t
] >= (DISCOVERED
| FALLTHROUGH
))
9585 return DONE_EXPLORING
;
9587 if (e
== BRANCH
&& insn_state
[t
] >= (DISCOVERED
| BRANCH
))
9588 return DONE_EXPLORING
;
9590 if (w
< 0 || w
>= env
->prog
->len
) {
9591 verbose_linfo(env
, t
, "%d: ", t
);
9592 verbose(env
, "jump out of range from insn %d to %d\n", t
, w
);
9597 /* mark branch target for state pruning */
9598 init_explored_state(env
, w
);
9600 if (insn_state
[w
] == 0) {
9602 insn_state
[t
] = DISCOVERED
| e
;
9603 insn_state
[w
] = DISCOVERED
;
9604 if (env
->cfg
.cur_stack
>= env
->prog
->len
)
9606 insn_stack
[env
->cfg
.cur_stack
++] = w
;
9607 return KEEP_EXPLORING
;
9608 } else if ((insn_state
[w
] & 0xF0) == DISCOVERED
) {
9609 if (loop_ok
&& env
->bpf_capable
)
9610 return DONE_EXPLORING
;
9611 verbose_linfo(env
, t
, "%d: ", t
);
9612 verbose_linfo(env
, w
, "%d: ", w
);
9613 verbose(env
, "back-edge from insn %d to %d\n", t
, w
);
9615 } else if (insn_state
[w
] == EXPLORED
) {
9616 /* forward- or cross-edge */
9617 insn_state
[t
] = DISCOVERED
| e
;
9619 verbose(env
, "insn state internal bug\n");
9622 return DONE_EXPLORING
;
9625 static int visit_func_call_insn(int t
, int insn_cnt
,
9626 struct bpf_insn
*insns
,
9627 struct bpf_verifier_env
*env
,
9632 ret
= push_insn(t
, t
+ 1, FALLTHROUGH
, env
, false);
9636 if (t
+ 1 < insn_cnt
)
9637 init_explored_state(env
, t
+ 1);
9639 init_explored_state(env
, t
);
9640 ret
= push_insn(t
, t
+ insns
[t
].imm
+ 1, BRANCH
, env
,
9641 /* It's ok to allow recursion from CFG point of
9642 * view. __check_func_call() will do the actual
9645 bpf_pseudo_func(insns
+ t
));
9650 /* Visits the instruction at index t and returns one of the following:
9651 * < 0 - an error occurred
9652 * DONE_EXPLORING - the instruction was fully explored
9653 * KEEP_EXPLORING - there is still work to be done before it is fully explored
9655 static int visit_insn(int t
, int insn_cnt
, struct bpf_verifier_env
*env
)
9657 struct bpf_insn
*insns
= env
->prog
->insnsi
;
9660 if (bpf_pseudo_func(insns
+ t
))
9661 return visit_func_call_insn(t
, insn_cnt
, insns
, env
, true);
9663 /* All non-branch instructions have a single fall-through edge. */
9664 if (BPF_CLASS(insns
[t
].code
) != BPF_JMP
&&
9665 BPF_CLASS(insns
[t
].code
) != BPF_JMP32
)
9666 return push_insn(t
, t
+ 1, FALLTHROUGH
, env
, false);
9668 switch (BPF_OP(insns
[t
].code
)) {
9670 return DONE_EXPLORING
;
9673 if (insns
[t
].imm
== BPF_FUNC_timer_set_callback
)
9674 /* Mark this call insn to trigger is_state_visited() check
9675 * before call itself is processed by __check_func_call().
9676 * Otherwise new async state will be pushed for further
9679 init_explored_state(env
, t
);
9680 return visit_func_call_insn(t
, insn_cnt
, insns
, env
,
9681 insns
[t
].src_reg
== BPF_PSEUDO_CALL
);
9684 if (BPF_SRC(insns
[t
].code
) != BPF_K
)
9687 /* unconditional jump with single edge */
9688 ret
= push_insn(t
, t
+ insns
[t
].off
+ 1, FALLTHROUGH
, env
,
9693 /* unconditional jmp is not a good pruning point,
9694 * but it's marked, since backtracking needs
9695 * to record jmp history in is_state_visited().
9697 init_explored_state(env
, t
+ insns
[t
].off
+ 1);
9698 /* tell verifier to check for equivalent states
9699 * after every call and jump
9701 if (t
+ 1 < insn_cnt
)
9702 init_explored_state(env
, t
+ 1);
9707 /* conditional jump with two edges */
9708 init_explored_state(env
, t
);
9709 ret
= push_insn(t
, t
+ 1, FALLTHROUGH
, env
, true);
9713 return push_insn(t
, t
+ insns
[t
].off
+ 1, BRANCH
, env
, true);
9717 /* non-recursive depth-first-search to detect loops in BPF program
9718 * loop == back-edge in directed graph
9720 static int check_cfg(struct bpf_verifier_env
*env
)
9722 int insn_cnt
= env
->prog
->len
;
9723 int *insn_stack
, *insn_state
;
9727 insn_state
= env
->cfg
.insn_state
= kvcalloc(insn_cnt
, sizeof(int), GFP_KERNEL
);
9731 insn_stack
= env
->cfg
.insn_stack
= kvcalloc(insn_cnt
, sizeof(int), GFP_KERNEL
);
9737 insn_state
[0] = DISCOVERED
; /* mark 1st insn as discovered */
9738 insn_stack
[0] = 0; /* 0 is the first instruction */
9739 env
->cfg
.cur_stack
= 1;
9741 while (env
->cfg
.cur_stack
> 0) {
9742 int t
= insn_stack
[env
->cfg
.cur_stack
- 1];
9744 ret
= visit_insn(t
, insn_cnt
, env
);
9746 case DONE_EXPLORING
:
9747 insn_state
[t
] = EXPLORED
;
9748 env
->cfg
.cur_stack
--;
9750 case KEEP_EXPLORING
:
9754 verbose(env
, "visit_insn internal bug\n");
9761 if (env
->cfg
.cur_stack
< 0) {
9762 verbose(env
, "pop stack internal bug\n");
9767 for (i
= 0; i
< insn_cnt
; i
++) {
9768 if (insn_state
[i
] != EXPLORED
) {
9769 verbose(env
, "unreachable insn %d\n", i
);
9774 ret
= 0; /* cfg looks good */
9779 env
->cfg
.insn_state
= env
->cfg
.insn_stack
= NULL
;
9783 static int check_abnormal_return(struct bpf_verifier_env
*env
)
9787 for (i
= 1; i
< env
->subprog_cnt
; i
++) {
9788 if (env
->subprog_info
[i
].has_ld_abs
) {
9789 verbose(env
, "LD_ABS is not allowed in subprogs without BTF\n");
9792 if (env
->subprog_info
[i
].has_tail_call
) {
9793 verbose(env
, "tail_call is not allowed in subprogs without BTF\n");
9800 /* The minimum supported BTF func info size */
9801 #define MIN_BPF_FUNCINFO_SIZE 8
9802 #define MAX_FUNCINFO_REC_SIZE 252
9804 static int check_btf_func(struct bpf_verifier_env
*env
,
9805 const union bpf_attr
*attr
,
9808 const struct btf_type
*type
, *func_proto
, *ret_type
;
9809 u32 i
, nfuncs
, urec_size
, min_size
;
9810 u32 krec_size
= sizeof(struct bpf_func_info
);
9811 struct bpf_func_info
*krecord
;
9812 struct bpf_func_info_aux
*info_aux
= NULL
;
9813 struct bpf_prog
*prog
;
9814 const struct btf
*btf
;
9816 u32 prev_offset
= 0;
9820 nfuncs
= attr
->func_info_cnt
;
9822 if (check_abnormal_return(env
))
9827 if (nfuncs
!= env
->subprog_cnt
) {
9828 verbose(env
, "number of funcs in func_info doesn't match number of subprogs\n");
9832 urec_size
= attr
->func_info_rec_size
;
9833 if (urec_size
< MIN_BPF_FUNCINFO_SIZE
||
9834 urec_size
> MAX_FUNCINFO_REC_SIZE
||
9835 urec_size
% sizeof(u32
)) {
9836 verbose(env
, "invalid func info rec size %u\n", urec_size
);
9841 btf
= prog
->aux
->btf
;
9843 urecord
= make_bpfptr(attr
->func_info
, uattr
.is_kernel
);
9844 min_size
= min_t(u32
, krec_size
, urec_size
);
9846 krecord
= kvcalloc(nfuncs
, krec_size
, GFP_KERNEL
| __GFP_NOWARN
);
9849 info_aux
= kcalloc(nfuncs
, sizeof(*info_aux
), GFP_KERNEL
| __GFP_NOWARN
);
9853 for (i
= 0; i
< nfuncs
; i
++) {
9854 ret
= bpf_check_uarg_tail_zero(urecord
, krec_size
, urec_size
);
9856 if (ret
== -E2BIG
) {
9857 verbose(env
, "nonzero tailing record in func info");
9858 /* set the size kernel expects so loader can zero
9859 * out the rest of the record.
9861 if (copy_to_bpfptr_offset(uattr
,
9862 offsetof(union bpf_attr
, func_info_rec_size
),
9863 &min_size
, sizeof(min_size
)))
9869 if (copy_from_bpfptr(&krecord
[i
], urecord
, min_size
)) {
9874 /* check insn_off */
9877 if (krecord
[i
].insn_off
) {
9879 "nonzero insn_off %u for the first func info record",
9880 krecord
[i
].insn_off
);
9883 } else if (krecord
[i
].insn_off
<= prev_offset
) {
9885 "same or smaller insn offset (%u) than previous func info record (%u)",
9886 krecord
[i
].insn_off
, prev_offset
);
9890 if (env
->subprog_info
[i
].start
!= krecord
[i
].insn_off
) {
9891 verbose(env
, "func_info BTF section doesn't match subprog layout in BPF program\n");
9896 type
= btf_type_by_id(btf
, krecord
[i
].type_id
);
9897 if (!type
|| !btf_type_is_func(type
)) {
9898 verbose(env
, "invalid type id %d in func info",
9899 krecord
[i
].type_id
);
9902 info_aux
[i
].linkage
= BTF_INFO_VLEN(type
->info
);
9904 func_proto
= btf_type_by_id(btf
, type
->type
);
9905 if (unlikely(!func_proto
|| !btf_type_is_func_proto(func_proto
)))
9906 /* btf_func_check() already verified it during BTF load */
9908 ret_type
= btf_type_skip_modifiers(btf
, func_proto
->type
, NULL
);
9910 btf_type_is_small_int(ret_type
) || btf_type_is_enum(ret_type
);
9911 if (i
&& !scalar_return
&& env
->subprog_info
[i
].has_ld_abs
) {
9912 verbose(env
, "LD_ABS is only allowed in functions that return 'int'.\n");
9915 if (i
&& !scalar_return
&& env
->subprog_info
[i
].has_tail_call
) {
9916 verbose(env
, "tail_call is only allowed in functions that return 'int'.\n");
9920 prev_offset
= krecord
[i
].insn_off
;
9921 bpfptr_add(&urecord
, urec_size
);
9924 prog
->aux
->func_info
= krecord
;
9925 prog
->aux
->func_info_cnt
= nfuncs
;
9926 prog
->aux
->func_info_aux
= info_aux
;
9935 static void adjust_btf_func(struct bpf_verifier_env
*env
)
9937 struct bpf_prog_aux
*aux
= env
->prog
->aux
;
9940 if (!aux
->func_info
)
9943 for (i
= 0; i
< env
->subprog_cnt
; i
++)
9944 aux
->func_info
[i
].insn_off
= env
->subprog_info
[i
].start
;
9947 #define MIN_BPF_LINEINFO_SIZE (offsetof(struct bpf_line_info, line_col) + \
9948 sizeof(((struct bpf_line_info *)(0))->line_col))
9949 #define MAX_LINEINFO_REC_SIZE MAX_FUNCINFO_REC_SIZE
9951 static int check_btf_line(struct bpf_verifier_env
*env
,
9952 const union bpf_attr
*attr
,
9955 u32 i
, s
, nr_linfo
, ncopy
, expected_size
, rec_size
, prev_offset
= 0;
9956 struct bpf_subprog_info
*sub
;
9957 struct bpf_line_info
*linfo
;
9958 struct bpf_prog
*prog
;
9959 const struct btf
*btf
;
9963 nr_linfo
= attr
->line_info_cnt
;
9966 if (nr_linfo
> INT_MAX
/ sizeof(struct bpf_line_info
))
9969 rec_size
= attr
->line_info_rec_size
;
9970 if (rec_size
< MIN_BPF_LINEINFO_SIZE
||
9971 rec_size
> MAX_LINEINFO_REC_SIZE
||
9972 rec_size
& (sizeof(u32
) - 1))
9975 /* Need to zero it in case the userspace may
9976 * pass in a smaller bpf_line_info object.
9978 linfo
= kvcalloc(nr_linfo
, sizeof(struct bpf_line_info
),
9979 GFP_KERNEL
| __GFP_NOWARN
);
9984 btf
= prog
->aux
->btf
;
9987 sub
= env
->subprog_info
;
9988 ulinfo
= make_bpfptr(attr
->line_info
, uattr
.is_kernel
);
9989 expected_size
= sizeof(struct bpf_line_info
);
9990 ncopy
= min_t(u32
, expected_size
, rec_size
);
9991 for (i
= 0; i
< nr_linfo
; i
++) {
9992 err
= bpf_check_uarg_tail_zero(ulinfo
, expected_size
, rec_size
);
9994 if (err
== -E2BIG
) {
9995 verbose(env
, "nonzero tailing record in line_info");
9996 if (copy_to_bpfptr_offset(uattr
,
9997 offsetof(union bpf_attr
, line_info_rec_size
),
9998 &expected_size
, sizeof(expected_size
)))
10004 if (copy_from_bpfptr(&linfo
[i
], ulinfo
, ncopy
)) {
10010 * Check insn_off to ensure
10011 * 1) strictly increasing AND
10012 * 2) bounded by prog->len
10014 * The linfo[0].insn_off == 0 check logically falls into
10015 * the later "missing bpf_line_info for func..." case
10016 * because the first linfo[0].insn_off must be the
10017 * first sub also and the first sub must have
10018 * subprog_info[0].start == 0.
10020 if ((i
&& linfo
[i
].insn_off
<= prev_offset
) ||
10021 linfo
[i
].insn_off
>= prog
->len
) {
10022 verbose(env
, "Invalid line_info[%u].insn_off:%u (prev_offset:%u prog->len:%u)\n",
10023 i
, linfo
[i
].insn_off
, prev_offset
,
10029 if (!prog
->insnsi
[linfo
[i
].insn_off
].code
) {
10031 "Invalid insn code at line_info[%u].insn_off\n",
10037 if (!btf_name_by_offset(btf
, linfo
[i
].line_off
) ||
10038 !btf_name_by_offset(btf
, linfo
[i
].file_name_off
)) {
10039 verbose(env
, "Invalid line_info[%u].line_off or .file_name_off\n", i
);
10044 if (s
!= env
->subprog_cnt
) {
10045 if (linfo
[i
].insn_off
== sub
[s
].start
) {
10046 sub
[s
].linfo_idx
= i
;
10048 } else if (sub
[s
].start
< linfo
[i
].insn_off
) {
10049 verbose(env
, "missing bpf_line_info for func#%u\n", s
);
10055 prev_offset
= linfo
[i
].insn_off
;
10056 bpfptr_add(&ulinfo
, rec_size
);
10059 if (s
!= env
->subprog_cnt
) {
10060 verbose(env
, "missing bpf_line_info for %u funcs starting from func#%u\n",
10061 env
->subprog_cnt
- s
, s
);
10066 prog
->aux
->linfo
= linfo
;
10067 prog
->aux
->nr_linfo
= nr_linfo
;
10076 static int check_btf_info(struct bpf_verifier_env
*env
,
10077 const union bpf_attr
*attr
,
10083 if (!attr
->func_info_cnt
&& !attr
->line_info_cnt
) {
10084 if (check_abnormal_return(env
))
10089 btf
= btf_get_by_fd(attr
->prog_btf_fd
);
10091 return PTR_ERR(btf
);
10092 if (btf_is_kernel(btf
)) {
10096 env
->prog
->aux
->btf
= btf
;
10098 err
= check_btf_func(env
, attr
, uattr
);
10102 err
= check_btf_line(env
, attr
, uattr
);
10109 /* check %cur's range satisfies %old's */
10110 static bool range_within(struct bpf_reg_state
*old
,
10111 struct bpf_reg_state
*cur
)
10113 return old
->umin_value
<= cur
->umin_value
&&
10114 old
->umax_value
>= cur
->umax_value
&&
10115 old
->smin_value
<= cur
->smin_value
&&
10116 old
->smax_value
>= cur
->smax_value
&&
10117 old
->u32_min_value
<= cur
->u32_min_value
&&
10118 old
->u32_max_value
>= cur
->u32_max_value
&&
10119 old
->s32_min_value
<= cur
->s32_min_value
&&
10120 old
->s32_max_value
>= cur
->s32_max_value
;
10123 /* If in the old state two registers had the same id, then they need to have
10124 * the same id in the new state as well. But that id could be different from
10125 * the old state, so we need to track the mapping from old to new ids.
10126 * Once we have seen that, say, a reg with old id 5 had new id 9, any subsequent
10127 * regs with old id 5 must also have new id 9 for the new state to be safe. But
10128 * regs with a different old id could still have new id 9, we don't care about
10130 * So we look through our idmap to see if this old id has been seen before. If
10131 * so, we require the new id to match; otherwise, we add the id pair to the map.
10133 static bool check_ids(u32 old_id
, u32 cur_id
, struct bpf_id_pair
*idmap
)
10137 for (i
= 0; i
< BPF_ID_MAP_SIZE
; i
++) {
10138 if (!idmap
[i
].old
) {
10139 /* Reached an empty slot; haven't seen this id before */
10140 idmap
[i
].old
= old_id
;
10141 idmap
[i
].cur
= cur_id
;
10144 if (idmap
[i
].old
== old_id
)
10145 return idmap
[i
].cur
== cur_id
;
10147 /* We ran out of idmap slots, which should be impossible */
10152 static void clean_func_state(struct bpf_verifier_env
*env
,
10153 struct bpf_func_state
*st
)
10155 enum bpf_reg_liveness live
;
10158 for (i
= 0; i
< BPF_REG_FP
; i
++) {
10159 live
= st
->regs
[i
].live
;
10160 /* liveness must not touch this register anymore */
10161 st
->regs
[i
].live
|= REG_LIVE_DONE
;
10162 if (!(live
& REG_LIVE_READ
))
10163 /* since the register is unused, clear its state
10164 * to make further comparison simpler
10166 __mark_reg_not_init(env
, &st
->regs
[i
]);
10169 for (i
= 0; i
< st
->allocated_stack
/ BPF_REG_SIZE
; i
++) {
10170 live
= st
->stack
[i
].spilled_ptr
.live
;
10171 /* liveness must not touch this stack slot anymore */
10172 st
->stack
[i
].spilled_ptr
.live
|= REG_LIVE_DONE
;
10173 if (!(live
& REG_LIVE_READ
)) {
10174 __mark_reg_not_init(env
, &st
->stack
[i
].spilled_ptr
);
10175 for (j
= 0; j
< BPF_REG_SIZE
; j
++)
10176 st
->stack
[i
].slot_type
[j
] = STACK_INVALID
;
10181 static void clean_verifier_state(struct bpf_verifier_env
*env
,
10182 struct bpf_verifier_state
*st
)
10186 if (st
->frame
[0]->regs
[0].live
& REG_LIVE_DONE
)
10187 /* all regs in this state in all frames were already marked */
10190 for (i
= 0; i
<= st
->curframe
; i
++)
10191 clean_func_state(env
, st
->frame
[i
]);
10194 /* the parentage chains form a tree.
10195 * the verifier states are added to state lists at given insn and
10196 * pushed into state stack for future exploration.
10197 * when the verifier reaches bpf_exit insn some of the verifer states
10198 * stored in the state lists have their final liveness state already,
10199 * but a lot of states will get revised from liveness point of view when
10200 * the verifier explores other branches.
10203 * 2: if r1 == 100 goto pc+1
10206 * when the verifier reaches exit insn the register r0 in the state list of
10207 * insn 2 will be seen as !REG_LIVE_READ. Then the verifier pops the other_branch
10208 * of insn 2 and goes exploring further. At the insn 4 it will walk the
10209 * parentage chain from insn 4 into insn 2 and will mark r0 as REG_LIVE_READ.
10211 * Since the verifier pushes the branch states as it sees them while exploring
10212 * the program the condition of walking the branch instruction for the second
10213 * time means that all states below this branch were already explored and
10214 * their final liveness marks are already propagated.
10215 * Hence when the verifier completes the search of state list in is_state_visited()
10216 * we can call this clean_live_states() function to mark all liveness states
10217 * as REG_LIVE_DONE to indicate that 'parent' pointers of 'struct bpf_reg_state'
10218 * will not be used.
10219 * This function also clears the registers and stack for states that !READ
10220 * to simplify state merging.
10222 * Important note here that walking the same branch instruction in the callee
10223 * doesn't meant that the states are DONE. The verifier has to compare
10226 static void clean_live_states(struct bpf_verifier_env
*env
, int insn
,
10227 struct bpf_verifier_state
*cur
)
10229 struct bpf_verifier_state_list
*sl
;
10232 sl
= *explored_state(env
, insn
);
10234 if (sl
->state
.branches
)
10236 if (sl
->state
.insn_idx
!= insn
||
10237 sl
->state
.curframe
!= cur
->curframe
)
10239 for (i
= 0; i
<= cur
->curframe
; i
++)
10240 if (sl
->state
.frame
[i
]->callsite
!= cur
->frame
[i
]->callsite
)
10242 clean_verifier_state(env
, &sl
->state
);
10248 /* Returns true if (rold safe implies rcur safe) */
10249 static bool regsafe(struct bpf_verifier_env
*env
, struct bpf_reg_state
*rold
,
10250 struct bpf_reg_state
*rcur
, struct bpf_id_pair
*idmap
)
10254 if (!(rold
->live
& REG_LIVE_READ
))
10255 /* explored state didn't use this */
10258 equal
= memcmp(rold
, rcur
, offsetof(struct bpf_reg_state
, parent
)) == 0;
10260 if (rold
->type
== PTR_TO_STACK
)
10261 /* two stack pointers are equal only if they're pointing to
10262 * the same stack frame, since fp-8 in foo != fp-8 in bar
10264 return equal
&& rold
->frameno
== rcur
->frameno
;
10269 if (rold
->type
== NOT_INIT
)
10270 /* explored state can't have used this */
10272 if (rcur
->type
== NOT_INIT
)
10274 switch (rold
->type
) {
10276 if (env
->explore_alu_limits
)
10278 if (rcur
->type
== SCALAR_VALUE
) {
10279 if (!rold
->precise
&& !rcur
->precise
)
10281 /* new val must satisfy old val knowledge */
10282 return range_within(rold
, rcur
) &&
10283 tnum_in(rold
->var_off
, rcur
->var_off
);
10285 /* We're trying to use a pointer in place of a scalar.
10286 * Even if the scalar was unbounded, this could lead to
10287 * pointer leaks because scalars are allowed to leak
10288 * while pointers are not. We could make this safe in
10289 * special cases if root is calling us, but it's
10290 * probably not worth the hassle.
10294 case PTR_TO_MAP_KEY
:
10295 case PTR_TO_MAP_VALUE
:
10296 /* If the new min/max/var_off satisfy the old ones and
10297 * everything else matches, we are OK.
10298 * 'id' is not compared, since it's only used for maps with
10299 * bpf_spin_lock inside map element and in such cases if
10300 * the rest of the prog is valid for one map element then
10301 * it's valid for all map elements regardless of the key
10302 * used in bpf_map_lookup()
10304 return memcmp(rold
, rcur
, offsetof(struct bpf_reg_state
, id
)) == 0 &&
10305 range_within(rold
, rcur
) &&
10306 tnum_in(rold
->var_off
, rcur
->var_off
);
10307 case PTR_TO_MAP_VALUE_OR_NULL
:
10308 /* a PTR_TO_MAP_VALUE could be safe to use as a
10309 * PTR_TO_MAP_VALUE_OR_NULL into the same map.
10310 * However, if the old PTR_TO_MAP_VALUE_OR_NULL then got NULL-
10311 * checked, doing so could have affected others with the same
10312 * id, and we can't check for that because we lost the id when
10313 * we converted to a PTR_TO_MAP_VALUE.
10315 if (rcur
->type
!= PTR_TO_MAP_VALUE_OR_NULL
)
10317 if (memcmp(rold
, rcur
, offsetof(struct bpf_reg_state
, id
)))
10319 /* Check our ids match any regs they're supposed to */
10320 return check_ids(rold
->id
, rcur
->id
, idmap
);
10321 case PTR_TO_PACKET_META
:
10322 case PTR_TO_PACKET
:
10323 if (rcur
->type
!= rold
->type
)
10325 /* We must have at least as much range as the old ptr
10326 * did, so that any accesses which were safe before are
10327 * still safe. This is true even if old range < old off,
10328 * since someone could have accessed through (ptr - k), or
10329 * even done ptr -= k in a register, to get a safe access.
10331 if (rold
->range
> rcur
->range
)
10333 /* If the offsets don't match, we can't trust our alignment;
10334 * nor can we be sure that we won't fall out of range.
10336 if (rold
->off
!= rcur
->off
)
10338 /* id relations must be preserved */
10339 if (rold
->id
&& !check_ids(rold
->id
, rcur
->id
, idmap
))
10341 /* new val must satisfy old val knowledge */
10342 return range_within(rold
, rcur
) &&
10343 tnum_in(rold
->var_off
, rcur
->var_off
);
10345 case CONST_PTR_TO_MAP
:
10346 case PTR_TO_PACKET_END
:
10347 case PTR_TO_FLOW_KEYS
:
10348 case PTR_TO_SOCKET
:
10349 case PTR_TO_SOCKET_OR_NULL
:
10350 case PTR_TO_SOCK_COMMON
:
10351 case PTR_TO_SOCK_COMMON_OR_NULL
:
10352 case PTR_TO_TCP_SOCK
:
10353 case PTR_TO_TCP_SOCK_OR_NULL
:
10354 case PTR_TO_XDP_SOCK
:
10355 /* Only valid matches are exact, which memcmp() above
10356 * would have accepted
10359 /* Don't know what's going on, just say it's not safe */
10363 /* Shouldn't get here; if we do, say it's not safe */
10368 static bool stacksafe(struct bpf_verifier_env
*env
, struct bpf_func_state
*old
,
10369 struct bpf_func_state
*cur
, struct bpf_id_pair
*idmap
)
10373 /* walk slots of the explored stack and ignore any additional
10374 * slots in the current stack, since explored(safe) state
10377 for (i
= 0; i
< old
->allocated_stack
; i
++) {
10378 spi
= i
/ BPF_REG_SIZE
;
10380 if (!(old
->stack
[spi
].spilled_ptr
.live
& REG_LIVE_READ
)) {
10381 i
+= BPF_REG_SIZE
- 1;
10382 /* explored state didn't use this */
10386 if (old
->stack
[spi
].slot_type
[i
% BPF_REG_SIZE
] == STACK_INVALID
)
10389 /* explored stack has more populated slots than current stack
10390 * and these slots were used
10392 if (i
>= cur
->allocated_stack
)
10395 /* if old state was safe with misc data in the stack
10396 * it will be safe with zero-initialized stack.
10397 * The opposite is not true
10399 if (old
->stack
[spi
].slot_type
[i
% BPF_REG_SIZE
] == STACK_MISC
&&
10400 cur
->stack
[spi
].slot_type
[i
% BPF_REG_SIZE
] == STACK_ZERO
)
10402 if (old
->stack
[spi
].slot_type
[i
% BPF_REG_SIZE
] !=
10403 cur
->stack
[spi
].slot_type
[i
% BPF_REG_SIZE
])
10404 /* Ex: old explored (safe) state has STACK_SPILL in
10405 * this stack slot, but current has STACK_MISC ->
10406 * this verifier states are not equivalent,
10407 * return false to continue verification of this path
10410 if (i
% BPF_REG_SIZE
)
10412 if (old
->stack
[spi
].slot_type
[0] != STACK_SPILL
)
10414 if (!regsafe(env
, &old
->stack
[spi
].spilled_ptr
,
10415 &cur
->stack
[spi
].spilled_ptr
, idmap
))
10416 /* when explored and current stack slot are both storing
10417 * spilled registers, check that stored pointers types
10418 * are the same as well.
10419 * Ex: explored safe path could have stored
10420 * (bpf_reg_state) {.type = PTR_TO_STACK, .off = -8}
10421 * but current path has stored:
10422 * (bpf_reg_state) {.type = PTR_TO_STACK, .off = -16}
10423 * such verifier states are not equivalent.
10424 * return false to continue verification of this path
10431 static bool refsafe(struct bpf_func_state
*old
, struct bpf_func_state
*cur
)
10433 if (old
->acquired_refs
!= cur
->acquired_refs
)
10435 return !memcmp(old
->refs
, cur
->refs
,
10436 sizeof(*old
->refs
) * old
->acquired_refs
);
10439 /* compare two verifier states
10441 * all states stored in state_list are known to be valid, since
10442 * verifier reached 'bpf_exit' instruction through them
10444 * this function is called when verifier exploring different branches of
10445 * execution popped from the state stack. If it sees an old state that has
10446 * more strict register state and more strict stack state then this execution
10447 * branch doesn't need to be explored further, since verifier already
10448 * concluded that more strict state leads to valid finish.
10450 * Therefore two states are equivalent if register state is more conservative
10451 * and explored stack state is more conservative than the current one.
10454 * (slot1=INV slot2=MISC) == (slot1=MISC slot2=MISC)
10455 * (slot1=MISC slot2=MISC) != (slot1=INV slot2=MISC)
10457 * In other words if current stack state (one being explored) has more
10458 * valid slots than old one that already passed validation, it means
10459 * the verifier can stop exploring and conclude that current state is valid too
10461 * Similarly with registers. If explored state has register type as invalid
10462 * whereas register type in current state is meaningful, it means that
10463 * the current state will reach 'bpf_exit' instruction safely
10465 static bool func_states_equal(struct bpf_verifier_env
*env
, struct bpf_func_state
*old
,
10466 struct bpf_func_state
*cur
)
10470 memset(env
->idmap_scratch
, 0, sizeof(env
->idmap_scratch
));
10471 for (i
= 0; i
< MAX_BPF_REG
; i
++)
10472 if (!regsafe(env
, &old
->regs
[i
], &cur
->regs
[i
],
10473 env
->idmap_scratch
))
10476 if (!stacksafe(env
, old
, cur
, env
->idmap_scratch
))
10479 if (!refsafe(old
, cur
))
10485 static bool states_equal(struct bpf_verifier_env
*env
,
10486 struct bpf_verifier_state
*old
,
10487 struct bpf_verifier_state
*cur
)
10491 if (old
->curframe
!= cur
->curframe
)
10494 /* Verification state from speculative execution simulation
10495 * must never prune a non-speculative execution one.
10497 if (old
->speculative
&& !cur
->speculative
)
10500 if (old
->active_spin_lock
!= cur
->active_spin_lock
)
10503 /* for states to be equal callsites have to be the same
10504 * and all frame states need to be equivalent
10506 for (i
= 0; i
<= old
->curframe
; i
++) {
10507 if (old
->frame
[i
]->callsite
!= cur
->frame
[i
]->callsite
)
10509 if (!func_states_equal(env
, old
->frame
[i
], cur
->frame
[i
]))
10515 /* Return 0 if no propagation happened. Return negative error code if error
10516 * happened. Otherwise, return the propagated bit.
10518 static int propagate_liveness_reg(struct bpf_verifier_env
*env
,
10519 struct bpf_reg_state
*reg
,
10520 struct bpf_reg_state
*parent_reg
)
10522 u8 parent_flag
= parent_reg
->live
& REG_LIVE_READ
;
10523 u8 flag
= reg
->live
& REG_LIVE_READ
;
10526 /* When comes here, read flags of PARENT_REG or REG could be any of
10527 * REG_LIVE_READ64, REG_LIVE_READ32, REG_LIVE_NONE. There is no need
10528 * of propagation if PARENT_REG has strongest REG_LIVE_READ64.
10530 if (parent_flag
== REG_LIVE_READ64
||
10531 /* Or if there is no read flag from REG. */
10533 /* Or if the read flag from REG is the same as PARENT_REG. */
10534 parent_flag
== flag
)
10537 err
= mark_reg_read(env
, reg
, parent_reg
, flag
);
10544 /* A write screens off any subsequent reads; but write marks come from the
10545 * straight-line code between a state and its parent. When we arrive at an
10546 * equivalent state (jump target or such) we didn't arrive by the straight-line
10547 * code, so read marks in the state must propagate to the parent regardless
10548 * of the state's write marks. That's what 'parent == state->parent' comparison
10549 * in mark_reg_read() is for.
10551 static int propagate_liveness(struct bpf_verifier_env
*env
,
10552 const struct bpf_verifier_state
*vstate
,
10553 struct bpf_verifier_state
*vparent
)
10555 struct bpf_reg_state
*state_reg
, *parent_reg
;
10556 struct bpf_func_state
*state
, *parent
;
10557 int i
, frame
, err
= 0;
10559 if (vparent
->curframe
!= vstate
->curframe
) {
10560 WARN(1, "propagate_live: parent frame %d current frame %d\n",
10561 vparent
->curframe
, vstate
->curframe
);
10564 /* Propagate read liveness of registers... */
10565 BUILD_BUG_ON(BPF_REG_FP
+ 1 != MAX_BPF_REG
);
10566 for (frame
= 0; frame
<= vstate
->curframe
; frame
++) {
10567 parent
= vparent
->frame
[frame
];
10568 state
= vstate
->frame
[frame
];
10569 parent_reg
= parent
->regs
;
10570 state_reg
= state
->regs
;
10571 /* We don't need to worry about FP liveness, it's read-only */
10572 for (i
= frame
< vstate
->curframe
? BPF_REG_6
: 0; i
< BPF_REG_FP
; i
++) {
10573 err
= propagate_liveness_reg(env
, &state_reg
[i
],
10577 if (err
== REG_LIVE_READ64
)
10578 mark_insn_zext(env
, &parent_reg
[i
]);
10581 /* Propagate stack slots. */
10582 for (i
= 0; i
< state
->allocated_stack
/ BPF_REG_SIZE
&&
10583 i
< parent
->allocated_stack
/ BPF_REG_SIZE
; i
++) {
10584 parent_reg
= &parent
->stack
[i
].spilled_ptr
;
10585 state_reg
= &state
->stack
[i
].spilled_ptr
;
10586 err
= propagate_liveness_reg(env
, state_reg
,
10595 /* find precise scalars in the previous equivalent state and
10596 * propagate them into the current state
10598 static int propagate_precision(struct bpf_verifier_env
*env
,
10599 const struct bpf_verifier_state
*old
)
10601 struct bpf_reg_state
*state_reg
;
10602 struct bpf_func_state
*state
;
10605 state
= old
->frame
[old
->curframe
];
10606 state_reg
= state
->regs
;
10607 for (i
= 0; i
< BPF_REG_FP
; i
++, state_reg
++) {
10608 if (state_reg
->type
!= SCALAR_VALUE
||
10609 !state_reg
->precise
)
10611 if (env
->log
.level
& BPF_LOG_LEVEL2
)
10612 verbose(env
, "propagating r%d\n", i
);
10613 err
= mark_chain_precision(env
, i
);
10618 for (i
= 0; i
< state
->allocated_stack
/ BPF_REG_SIZE
; i
++) {
10619 if (state
->stack
[i
].slot_type
[0] != STACK_SPILL
)
10621 state_reg
= &state
->stack
[i
].spilled_ptr
;
10622 if (state_reg
->type
!= SCALAR_VALUE
||
10623 !state_reg
->precise
)
10625 if (env
->log
.level
& BPF_LOG_LEVEL2
)
10626 verbose(env
, "propagating fp%d\n",
10627 (-i
- 1) * BPF_REG_SIZE
);
10628 err
= mark_chain_precision_stack(env
, i
);
10635 static bool states_maybe_looping(struct bpf_verifier_state
*old
,
10636 struct bpf_verifier_state
*cur
)
10638 struct bpf_func_state
*fold
, *fcur
;
10639 int i
, fr
= cur
->curframe
;
10641 if (old
->curframe
!= fr
)
10644 fold
= old
->frame
[fr
];
10645 fcur
= cur
->frame
[fr
];
10646 for (i
= 0; i
< MAX_BPF_REG
; i
++)
10647 if (memcmp(&fold
->regs
[i
], &fcur
->regs
[i
],
10648 offsetof(struct bpf_reg_state
, parent
)))
10654 static int is_state_visited(struct bpf_verifier_env
*env
, int insn_idx
)
10656 struct bpf_verifier_state_list
*new_sl
;
10657 struct bpf_verifier_state_list
*sl
, **pprev
;
10658 struct bpf_verifier_state
*cur
= env
->cur_state
, *new;
10659 int i
, j
, err
, states_cnt
= 0;
10660 bool add_new_state
= env
->test_state_freq
? true : false;
10662 cur
->last_insn_idx
= env
->prev_insn_idx
;
10663 if (!env
->insn_aux_data
[insn_idx
].prune_point
)
10664 /* this 'insn_idx' instruction wasn't marked, so we will not
10665 * be doing state search here
10669 /* bpf progs typically have pruning point every 4 instructions
10670 * http://vger.kernel.org/bpfconf2019.html#session-1
10671 * Do not add new state for future pruning if the verifier hasn't seen
10672 * at least 2 jumps and at least 8 instructions.
10673 * This heuristics helps decrease 'total_states' and 'peak_states' metric.
10674 * In tests that amounts to up to 50% reduction into total verifier
10675 * memory consumption and 20% verifier time speedup.
10677 if (env
->jmps_processed
- env
->prev_jmps_processed
>= 2 &&
10678 env
->insn_processed
- env
->prev_insn_processed
>= 8)
10679 add_new_state
= true;
10681 pprev
= explored_state(env
, insn_idx
);
10684 clean_live_states(env
, insn_idx
, cur
);
10688 if (sl
->state
.insn_idx
!= insn_idx
)
10691 if (sl
->state
.branches
) {
10692 struct bpf_func_state
*frame
= sl
->state
.frame
[sl
->state
.curframe
];
10694 if (frame
->in_async_callback_fn
&&
10695 frame
->async_entry_cnt
!= cur
->frame
[cur
->curframe
]->async_entry_cnt
) {
10696 /* Different async_entry_cnt means that the verifier is
10697 * processing another entry into async callback.
10698 * Seeing the same state is not an indication of infinite
10699 * loop or infinite recursion.
10700 * But finding the same state doesn't mean that it's safe
10701 * to stop processing the current state. The previous state
10702 * hasn't yet reached bpf_exit, since state.branches > 0.
10703 * Checking in_async_callback_fn alone is not enough either.
10704 * Since the verifier still needs to catch infinite loops
10705 * inside async callbacks.
10707 } else if (states_maybe_looping(&sl
->state
, cur
) &&
10708 states_equal(env
, &sl
->state
, cur
)) {
10709 verbose_linfo(env
, insn_idx
, "; ");
10710 verbose(env
, "infinite loop detected at insn %d\n", insn_idx
);
10713 /* if the verifier is processing a loop, avoid adding new state
10714 * too often, since different loop iterations have distinct
10715 * states and may not help future pruning.
10716 * This threshold shouldn't be too low to make sure that
10717 * a loop with large bound will be rejected quickly.
10718 * The most abusive loop will be:
10720 * if r1 < 1000000 goto pc-2
10721 * 1M insn_procssed limit / 100 == 10k peak states.
10722 * This threshold shouldn't be too high either, since states
10723 * at the end of the loop are likely to be useful in pruning.
10725 if (env
->jmps_processed
- env
->prev_jmps_processed
< 20 &&
10726 env
->insn_processed
- env
->prev_insn_processed
< 100)
10727 add_new_state
= false;
10730 if (states_equal(env
, &sl
->state
, cur
)) {
10732 /* reached equivalent register/stack state,
10733 * prune the search.
10734 * Registers read by the continuation are read by us.
10735 * If we have any write marks in env->cur_state, they
10736 * will prevent corresponding reads in the continuation
10737 * from reaching our parent (an explored_state). Our
10738 * own state will get the read marks recorded, but
10739 * they'll be immediately forgotten as we're pruning
10740 * this state and will pop a new one.
10742 err
= propagate_liveness(env
, &sl
->state
, cur
);
10744 /* if previous state reached the exit with precision and
10745 * current state is equivalent to it (except precsion marks)
10746 * the precision needs to be propagated back in
10747 * the current state.
10749 err
= err
? : push_jmp_history(env
, cur
);
10750 err
= err
? : propagate_precision(env
, &sl
->state
);
10756 /* when new state is not going to be added do not increase miss count.
10757 * Otherwise several loop iterations will remove the state
10758 * recorded earlier. The goal of these heuristics is to have
10759 * states from some iterations of the loop (some in the beginning
10760 * and some at the end) to help pruning.
10764 /* heuristic to determine whether this state is beneficial
10765 * to keep checking from state equivalence point of view.
10766 * Higher numbers increase max_states_per_insn and verification time,
10767 * but do not meaningfully decrease insn_processed.
10769 if (sl
->miss_cnt
> sl
->hit_cnt
* 3 + 3) {
10770 /* the state is unlikely to be useful. Remove it to
10771 * speed up verification
10774 if (sl
->state
.frame
[0]->regs
[0].live
& REG_LIVE_DONE
) {
10775 u32 br
= sl
->state
.branches
;
10778 "BUG live_done but branches_to_explore %d\n",
10780 free_verifier_state(&sl
->state
, false);
10782 env
->peak_states
--;
10784 /* cannot free this state, since parentage chain may
10785 * walk it later. Add it for free_list instead to
10786 * be freed at the end of verification
10788 sl
->next
= env
->free_list
;
10789 env
->free_list
= sl
;
10799 if (env
->max_states_per_insn
< states_cnt
)
10800 env
->max_states_per_insn
= states_cnt
;
10802 if (!env
->bpf_capable
&& states_cnt
> BPF_COMPLEXITY_LIMIT_STATES
)
10803 return push_jmp_history(env
, cur
);
10805 if (!add_new_state
)
10806 return push_jmp_history(env
, cur
);
10808 /* There were no equivalent states, remember the current one.
10809 * Technically the current state is not proven to be safe yet,
10810 * but it will either reach outer most bpf_exit (which means it's safe)
10811 * or it will be rejected. When there are no loops the verifier won't be
10812 * seeing this tuple (frame[0].callsite, frame[1].callsite, .. insn_idx)
10813 * again on the way to bpf_exit.
10814 * When looping the sl->state.branches will be > 0 and this state
10815 * will not be considered for equivalence until branches == 0.
10817 new_sl
= kzalloc(sizeof(struct bpf_verifier_state_list
), GFP_KERNEL
);
10820 env
->total_states
++;
10821 env
->peak_states
++;
10822 env
->prev_jmps_processed
= env
->jmps_processed
;
10823 env
->prev_insn_processed
= env
->insn_processed
;
10825 /* add new state to the head of linked list */
10826 new = &new_sl
->state
;
10827 err
= copy_verifier_state(new, cur
);
10829 free_verifier_state(new, false);
10833 new->insn_idx
= insn_idx
;
10834 WARN_ONCE(new->branches
!= 1,
10835 "BUG is_state_visited:branches_to_explore=%d insn %d\n", new->branches
, insn_idx
);
10838 cur
->first_insn_idx
= insn_idx
;
10839 clear_jmp_history(cur
);
10840 new_sl
->next
= *explored_state(env
, insn_idx
);
10841 *explored_state(env
, insn_idx
) = new_sl
;
10842 /* connect new state to parentage chain. Current frame needs all
10843 * registers connected. Only r6 - r9 of the callers are alive (pushed
10844 * to the stack implicitly by JITs) so in callers' frames connect just
10845 * r6 - r9 as an optimization. Callers will have r1 - r5 connected to
10846 * the state of the call instruction (with WRITTEN set), and r0 comes
10847 * from callee with its full parentage chain, anyway.
10849 /* clear write marks in current state: the writes we did are not writes
10850 * our child did, so they don't screen off its reads from us.
10851 * (There are no read marks in current state, because reads always mark
10852 * their parent and current state never has children yet. Only
10853 * explored_states can get read marks.)
10855 for (j
= 0; j
<= cur
->curframe
; j
++) {
10856 for (i
= j
< cur
->curframe
? BPF_REG_6
: 0; i
< BPF_REG_FP
; i
++)
10857 cur
->frame
[j
]->regs
[i
].parent
= &new->frame
[j
]->regs
[i
];
10858 for (i
= 0; i
< BPF_REG_FP
; i
++)
10859 cur
->frame
[j
]->regs
[i
].live
= REG_LIVE_NONE
;
10862 /* all stack frames are accessible from callee, clear them all */
10863 for (j
= 0; j
<= cur
->curframe
; j
++) {
10864 struct bpf_func_state
*frame
= cur
->frame
[j
];
10865 struct bpf_func_state
*newframe
= new->frame
[j
];
10867 for (i
= 0; i
< frame
->allocated_stack
/ BPF_REG_SIZE
; i
++) {
10868 frame
->stack
[i
].spilled_ptr
.live
= REG_LIVE_NONE
;
10869 frame
->stack
[i
].spilled_ptr
.parent
=
10870 &newframe
->stack
[i
].spilled_ptr
;
10876 /* Return true if it's OK to have the same insn return a different type. */
10877 static bool reg_type_mismatch_ok(enum bpf_reg_type type
)
10881 case PTR_TO_SOCKET
:
10882 case PTR_TO_SOCKET_OR_NULL
:
10883 case PTR_TO_SOCK_COMMON
:
10884 case PTR_TO_SOCK_COMMON_OR_NULL
:
10885 case PTR_TO_TCP_SOCK
:
10886 case PTR_TO_TCP_SOCK_OR_NULL
:
10887 case PTR_TO_XDP_SOCK
:
10888 case PTR_TO_BTF_ID
:
10889 case PTR_TO_BTF_ID_OR_NULL
:
10896 /* If an instruction was previously used with particular pointer types, then we
10897 * need to be careful to avoid cases such as the below, where it may be ok
10898 * for one branch accessing the pointer, but not ok for the other branch:
10903 * R1 = some_other_valid_ptr;
10906 * R2 = *(u32 *)(R1 + 0);
10908 static bool reg_type_mismatch(enum bpf_reg_type src
, enum bpf_reg_type prev
)
10910 return src
!= prev
&& (!reg_type_mismatch_ok(src
) ||
10911 !reg_type_mismatch_ok(prev
));
10914 static int do_check(struct bpf_verifier_env
*env
)
10916 bool pop_log
= !(env
->log
.level
& BPF_LOG_LEVEL2
);
10917 struct bpf_verifier_state
*state
= env
->cur_state
;
10918 struct bpf_insn
*insns
= env
->prog
->insnsi
;
10919 struct bpf_reg_state
*regs
;
10920 int insn_cnt
= env
->prog
->len
;
10921 bool do_print_state
= false;
10922 int prev_insn_idx
= -1;
10925 struct bpf_insn
*insn
;
10929 env
->prev_insn_idx
= prev_insn_idx
;
10930 if (env
->insn_idx
>= insn_cnt
) {
10931 verbose(env
, "invalid insn idx %d insn_cnt %d\n",
10932 env
->insn_idx
, insn_cnt
);
10936 insn
= &insns
[env
->insn_idx
];
10937 class = BPF_CLASS(insn
->code
);
10939 if (++env
->insn_processed
> BPF_COMPLEXITY_LIMIT_INSNS
) {
10941 "BPF program is too large. Processed %d insn\n",
10942 env
->insn_processed
);
10946 err
= is_state_visited(env
, env
->insn_idx
);
10950 /* found equivalent state, can prune the search */
10951 if (env
->log
.level
& BPF_LOG_LEVEL
) {
10952 if (do_print_state
)
10953 verbose(env
, "\nfrom %d to %d%s: safe\n",
10954 env
->prev_insn_idx
, env
->insn_idx
,
10955 env
->cur_state
->speculative
?
10956 " (speculative execution)" : "");
10958 verbose(env
, "%d: safe\n", env
->insn_idx
);
10960 goto process_bpf_exit
;
10963 if (signal_pending(current
))
10966 if (need_resched())
10969 if (env
->log
.level
& BPF_LOG_LEVEL2
||
10970 (env
->log
.level
& BPF_LOG_LEVEL
&& do_print_state
)) {
10971 if (env
->log
.level
& BPF_LOG_LEVEL2
)
10972 verbose(env
, "%d:", env
->insn_idx
);
10974 verbose(env
, "\nfrom %d to %d%s:",
10975 env
->prev_insn_idx
, env
->insn_idx
,
10976 env
->cur_state
->speculative
?
10977 " (speculative execution)" : "");
10978 print_verifier_state(env
, state
->frame
[state
->curframe
]);
10979 do_print_state
= false;
10982 if (env
->log
.level
& BPF_LOG_LEVEL
) {
10983 const struct bpf_insn_cbs cbs
= {
10984 .cb_call
= disasm_kfunc_name
,
10985 .cb_print
= verbose
,
10986 .private_data
= env
,
10989 verbose_linfo(env
, env
->insn_idx
, "; ");
10990 verbose(env
, "%d: ", env
->insn_idx
);
10991 print_bpf_insn(&cbs
, insn
, env
->allow_ptr_leaks
);
10994 if (bpf_prog_is_dev_bound(env
->prog
->aux
)) {
10995 err
= bpf_prog_offload_verify_insn(env
, env
->insn_idx
,
10996 env
->prev_insn_idx
);
11001 regs
= cur_regs(env
);
11002 sanitize_mark_insn_seen(env
);
11003 prev_insn_idx
= env
->insn_idx
;
11005 if (class == BPF_ALU
|| class == BPF_ALU64
) {
11006 err
= check_alu_op(env
, insn
);
11010 } else if (class == BPF_LDX
) {
11011 enum bpf_reg_type
*prev_src_type
, src_reg_type
;
11013 /* check for reserved fields is already done */
11015 /* check src operand */
11016 err
= check_reg_arg(env
, insn
->src_reg
, SRC_OP
);
11020 err
= check_reg_arg(env
, insn
->dst_reg
, DST_OP_NO_MARK
);
11024 src_reg_type
= regs
[insn
->src_reg
].type
;
11026 /* check that memory (src_reg + off) is readable,
11027 * the state of dst_reg will be updated by this func
11029 err
= check_mem_access(env
, env
->insn_idx
, insn
->src_reg
,
11030 insn
->off
, BPF_SIZE(insn
->code
),
11031 BPF_READ
, insn
->dst_reg
, false);
11035 prev_src_type
= &env
->insn_aux_data
[env
->insn_idx
].ptr_type
;
11037 if (*prev_src_type
== NOT_INIT
) {
11038 /* saw a valid insn
11039 * dst_reg = *(u32 *)(src_reg + off)
11040 * save type to validate intersecting paths
11042 *prev_src_type
= src_reg_type
;
11044 } else if (reg_type_mismatch(src_reg_type
, *prev_src_type
)) {
11045 /* ABuser program is trying to use the same insn
11046 * dst_reg = *(u32*) (src_reg + off)
11047 * with different pointer types:
11048 * src_reg == ctx in one branch and
11049 * src_reg == stack|map in some other branch.
11052 verbose(env
, "same insn cannot be used with different pointers\n");
11056 } else if (class == BPF_STX
) {
11057 enum bpf_reg_type
*prev_dst_type
, dst_reg_type
;
11059 if (BPF_MODE(insn
->code
) == BPF_ATOMIC
) {
11060 err
= check_atomic(env
, env
->insn_idx
, insn
);
11067 if (BPF_MODE(insn
->code
) != BPF_MEM
|| insn
->imm
!= 0) {
11068 verbose(env
, "BPF_STX uses reserved fields\n");
11072 /* check src1 operand */
11073 err
= check_reg_arg(env
, insn
->src_reg
, SRC_OP
);
11076 /* check src2 operand */
11077 err
= check_reg_arg(env
, insn
->dst_reg
, SRC_OP
);
11081 dst_reg_type
= regs
[insn
->dst_reg
].type
;
11083 /* check that memory (dst_reg + off) is writeable */
11084 err
= check_mem_access(env
, env
->insn_idx
, insn
->dst_reg
,
11085 insn
->off
, BPF_SIZE(insn
->code
),
11086 BPF_WRITE
, insn
->src_reg
, false);
11090 prev_dst_type
= &env
->insn_aux_data
[env
->insn_idx
].ptr_type
;
11092 if (*prev_dst_type
== NOT_INIT
) {
11093 *prev_dst_type
= dst_reg_type
;
11094 } else if (reg_type_mismatch(dst_reg_type
, *prev_dst_type
)) {
11095 verbose(env
, "same insn cannot be used with different pointers\n");
11099 } else if (class == BPF_ST
) {
11100 if (BPF_MODE(insn
->code
) != BPF_MEM
||
11101 insn
->src_reg
!= BPF_REG_0
) {
11102 verbose(env
, "BPF_ST uses reserved fields\n");
11105 /* check src operand */
11106 err
= check_reg_arg(env
, insn
->dst_reg
, SRC_OP
);
11110 if (is_ctx_reg(env
, insn
->dst_reg
)) {
11111 verbose(env
, "BPF_ST stores into R%d %s is not allowed\n",
11113 reg_type_str
[reg_state(env
, insn
->dst_reg
)->type
]);
11117 /* check that memory (dst_reg + off) is writeable */
11118 err
= check_mem_access(env
, env
->insn_idx
, insn
->dst_reg
,
11119 insn
->off
, BPF_SIZE(insn
->code
),
11120 BPF_WRITE
, -1, false);
11124 } else if (class == BPF_JMP
|| class == BPF_JMP32
) {
11125 u8 opcode
= BPF_OP(insn
->code
);
11127 env
->jmps_processed
++;
11128 if (opcode
== BPF_CALL
) {
11129 if (BPF_SRC(insn
->code
) != BPF_K
||
11131 (insn
->src_reg
!= BPF_REG_0
&&
11132 insn
->src_reg
!= BPF_PSEUDO_CALL
&&
11133 insn
->src_reg
!= BPF_PSEUDO_KFUNC_CALL
) ||
11134 insn
->dst_reg
!= BPF_REG_0
||
11135 class == BPF_JMP32
) {
11136 verbose(env
, "BPF_CALL uses reserved fields\n");
11140 if (env
->cur_state
->active_spin_lock
&&
11141 (insn
->src_reg
== BPF_PSEUDO_CALL
||
11142 insn
->imm
!= BPF_FUNC_spin_unlock
)) {
11143 verbose(env
, "function calls are not allowed while holding a lock\n");
11146 if (insn
->src_reg
== BPF_PSEUDO_CALL
)
11147 err
= check_func_call(env
, insn
, &env
->insn_idx
);
11148 else if (insn
->src_reg
== BPF_PSEUDO_KFUNC_CALL
)
11149 err
= check_kfunc_call(env
, insn
);
11151 err
= check_helper_call(env
, insn
, &env
->insn_idx
);
11154 } else if (opcode
== BPF_JA
) {
11155 if (BPF_SRC(insn
->code
) != BPF_K
||
11157 insn
->src_reg
!= BPF_REG_0
||
11158 insn
->dst_reg
!= BPF_REG_0
||
11159 class == BPF_JMP32
) {
11160 verbose(env
, "BPF_JA uses reserved fields\n");
11164 env
->insn_idx
+= insn
->off
+ 1;
11167 } else if (opcode
== BPF_EXIT
) {
11168 if (BPF_SRC(insn
->code
) != BPF_K
||
11170 insn
->src_reg
!= BPF_REG_0
||
11171 insn
->dst_reg
!= BPF_REG_0
||
11172 class == BPF_JMP32
) {
11173 verbose(env
, "BPF_EXIT uses reserved fields\n");
11177 if (env
->cur_state
->active_spin_lock
) {
11178 verbose(env
, "bpf_spin_unlock is missing\n");
11182 if (state
->curframe
) {
11183 /* exit from nested function */
11184 err
= prepare_func_exit(env
, &env
->insn_idx
);
11187 do_print_state
= true;
11191 err
= check_reference_leak(env
);
11195 err
= check_return_code(env
);
11199 update_branch_counts(env
, env
->cur_state
);
11200 err
= pop_stack(env
, &prev_insn_idx
,
11201 &env
->insn_idx
, pop_log
);
11203 if (err
!= -ENOENT
)
11207 do_print_state
= true;
11211 err
= check_cond_jmp_op(env
, insn
, &env
->insn_idx
);
11215 } else if (class == BPF_LD
) {
11216 u8 mode
= BPF_MODE(insn
->code
);
11218 if (mode
== BPF_ABS
|| mode
== BPF_IND
) {
11219 err
= check_ld_abs(env
, insn
);
11223 } else if (mode
== BPF_IMM
) {
11224 err
= check_ld_imm(env
, insn
);
11229 sanitize_mark_insn_seen(env
);
11231 verbose(env
, "invalid BPF_LD mode\n");
11235 verbose(env
, "unknown insn class %d\n", class);
11245 static int find_btf_percpu_datasec(struct btf
*btf
)
11247 const struct btf_type
*t
;
11252 * Both vmlinux and module each have their own ".data..percpu"
11253 * DATASECs in BTF. So for module's case, we need to skip vmlinux BTF
11254 * types to look at only module's own BTF types.
11256 n
= btf_nr_types(btf
);
11257 if (btf_is_module(btf
))
11258 i
= btf_nr_types(btf_vmlinux
);
11262 for(; i
< n
; i
++) {
11263 t
= btf_type_by_id(btf
, i
);
11264 if (BTF_INFO_KIND(t
->info
) != BTF_KIND_DATASEC
)
11267 tname
= btf_name_by_offset(btf
, t
->name_off
);
11268 if (!strcmp(tname
, ".data..percpu"))
11275 /* replace pseudo btf_id with kernel symbol address */
11276 static int check_pseudo_btf_id(struct bpf_verifier_env
*env
,
11277 struct bpf_insn
*insn
,
11278 struct bpf_insn_aux_data
*aux
)
11280 const struct btf_var_secinfo
*vsi
;
11281 const struct btf_type
*datasec
;
11282 struct btf_mod_pair
*btf_mod
;
11283 const struct btf_type
*t
;
11284 const char *sym_name
;
11285 bool percpu
= false;
11286 u32 type
, id
= insn
->imm
;
11290 int i
, btf_fd
, err
;
11292 btf_fd
= insn
[1].imm
;
11294 btf
= btf_get_by_fd(btf_fd
);
11296 verbose(env
, "invalid module BTF object FD specified.\n");
11300 if (!btf_vmlinux
) {
11301 verbose(env
, "kernel is missing BTF, make sure CONFIG_DEBUG_INFO_BTF=y is specified in Kconfig.\n");
11308 t
= btf_type_by_id(btf
, id
);
11310 verbose(env
, "ldimm64 insn specifies invalid btf_id %d.\n", id
);
11315 if (!btf_type_is_var(t
)) {
11316 verbose(env
, "pseudo btf_id %d in ldimm64 isn't KIND_VAR.\n", id
);
11321 sym_name
= btf_name_by_offset(btf
, t
->name_off
);
11322 addr
= kallsyms_lookup_name(sym_name
);
11324 verbose(env
, "ldimm64 failed to find the address for kernel symbol '%s'.\n",
11330 datasec_id
= find_btf_percpu_datasec(btf
);
11331 if (datasec_id
> 0) {
11332 datasec
= btf_type_by_id(btf
, datasec_id
);
11333 for_each_vsi(i
, datasec
, vsi
) {
11334 if (vsi
->type
== id
) {
11341 insn
[0].imm
= (u32
)addr
;
11342 insn
[1].imm
= addr
>> 32;
11345 t
= btf_type_skip_modifiers(btf
, type
, NULL
);
11347 aux
->btf_var
.reg_type
= PTR_TO_PERCPU_BTF_ID
;
11348 aux
->btf_var
.btf
= btf
;
11349 aux
->btf_var
.btf_id
= type
;
11350 } else if (!btf_type_is_struct(t
)) {
11351 const struct btf_type
*ret
;
11355 /* resolve the type size of ksym. */
11356 ret
= btf_resolve_size(btf
, t
, &tsize
);
11358 tname
= btf_name_by_offset(btf
, t
->name_off
);
11359 verbose(env
, "ldimm64 unable to resolve the size of type '%s': %ld\n",
11360 tname
, PTR_ERR(ret
));
11364 aux
->btf_var
.reg_type
= PTR_TO_MEM
;
11365 aux
->btf_var
.mem_size
= tsize
;
11367 aux
->btf_var
.reg_type
= PTR_TO_BTF_ID
;
11368 aux
->btf_var
.btf
= btf
;
11369 aux
->btf_var
.btf_id
= type
;
11372 /* check whether we recorded this BTF (and maybe module) already */
11373 for (i
= 0; i
< env
->used_btf_cnt
; i
++) {
11374 if (env
->used_btfs
[i
].btf
== btf
) {
11380 if (env
->used_btf_cnt
>= MAX_USED_BTFS
) {
11385 btf_mod
= &env
->used_btfs
[env
->used_btf_cnt
];
11386 btf_mod
->btf
= btf
;
11387 btf_mod
->module
= NULL
;
11389 /* if we reference variables from kernel module, bump its refcount */
11390 if (btf_is_module(btf
)) {
11391 btf_mod
->module
= btf_try_get_module(btf
);
11392 if (!btf_mod
->module
) {
11398 env
->used_btf_cnt
++;
11406 static int check_map_prealloc(struct bpf_map
*map
)
11408 return (map
->map_type
!= BPF_MAP_TYPE_HASH
&&
11409 map
->map_type
!= BPF_MAP_TYPE_PERCPU_HASH
&&
11410 map
->map_type
!= BPF_MAP_TYPE_HASH_OF_MAPS
) ||
11411 !(map
->map_flags
& BPF_F_NO_PREALLOC
);
11414 static bool is_tracing_prog_type(enum bpf_prog_type type
)
11417 case BPF_PROG_TYPE_KPROBE
:
11418 case BPF_PROG_TYPE_TRACEPOINT
:
11419 case BPF_PROG_TYPE_PERF_EVENT
:
11420 case BPF_PROG_TYPE_RAW_TRACEPOINT
:
11427 static bool is_preallocated_map(struct bpf_map
*map
)
11429 if (!check_map_prealloc(map
))
11431 if (map
->inner_map_meta
&& !check_map_prealloc(map
->inner_map_meta
))
11436 static int check_map_prog_compatibility(struct bpf_verifier_env
*env
,
11437 struct bpf_map
*map
,
11438 struct bpf_prog
*prog
)
11441 enum bpf_prog_type prog_type
= resolve_prog_type(prog
);
11443 * Validate that trace type programs use preallocated hash maps.
11445 * For programs attached to PERF events this is mandatory as the
11446 * perf NMI can hit any arbitrary code sequence.
11448 * All other trace types using preallocated hash maps are unsafe as
11449 * well because tracepoint or kprobes can be inside locked regions
11450 * of the memory allocator or at a place where a recursion into the
11451 * memory allocator would see inconsistent state.
11453 * On RT enabled kernels run-time allocation of all trace type
11454 * programs is strictly prohibited due to lock type constraints. On
11455 * !RT kernels it is allowed for backwards compatibility reasons for
11456 * now, but warnings are emitted so developers are made aware of
11457 * the unsafety and can fix their programs before this is enforced.
11459 if (is_tracing_prog_type(prog_type
) && !is_preallocated_map(map
)) {
11460 if (prog_type
== BPF_PROG_TYPE_PERF_EVENT
) {
11461 verbose(env
, "perf_event programs can only use preallocated hash map\n");
11464 if (IS_ENABLED(CONFIG_PREEMPT_RT
)) {
11465 verbose(env
, "trace type programs can only use preallocated hash map\n");
11468 WARN_ONCE(1, "trace type BPF program uses run-time allocation\n");
11469 verbose(env
, "trace type programs with run-time allocated hash maps are unsafe. Switch to preallocated hash maps.\n");
11472 if (map_value_has_spin_lock(map
)) {
11473 if (prog_type
== BPF_PROG_TYPE_SOCKET_FILTER
) {
11474 verbose(env
, "socket filter progs cannot use bpf_spin_lock yet\n");
11478 if (is_tracing_prog_type(prog_type
)) {
11479 verbose(env
, "tracing progs cannot use bpf_spin_lock yet\n");
11483 if (prog
->aux
->sleepable
) {
11484 verbose(env
, "sleepable progs cannot use bpf_spin_lock yet\n");
11489 if (map_value_has_timer(map
)) {
11490 if (is_tracing_prog_type(prog_type
)) {
11491 verbose(env
, "tracing progs cannot use bpf_timer yet\n");
11496 if ((bpf_prog_is_dev_bound(prog
->aux
) || bpf_map_is_dev_bound(map
)) &&
11497 !bpf_offload_prog_map_match(prog
, map
)) {
11498 verbose(env
, "offload device mismatch between prog and map\n");
11502 if (map
->map_type
== BPF_MAP_TYPE_STRUCT_OPS
) {
11503 verbose(env
, "bpf_struct_ops map cannot be used in prog\n");
11507 if (prog
->aux
->sleepable
)
11508 switch (map
->map_type
) {
11509 case BPF_MAP_TYPE_HASH
:
11510 case BPF_MAP_TYPE_LRU_HASH
:
11511 case BPF_MAP_TYPE_ARRAY
:
11512 case BPF_MAP_TYPE_PERCPU_HASH
:
11513 case BPF_MAP_TYPE_PERCPU_ARRAY
:
11514 case BPF_MAP_TYPE_LRU_PERCPU_HASH
:
11515 case BPF_MAP_TYPE_ARRAY_OF_MAPS
:
11516 case BPF_MAP_TYPE_HASH_OF_MAPS
:
11517 if (!is_preallocated_map(map
)) {
11519 "Sleepable programs can only use preallocated maps\n");
11523 case BPF_MAP_TYPE_RINGBUF
:
11527 "Sleepable programs can only use array, hash, and ringbuf maps\n");
11534 static bool bpf_map_is_cgroup_storage(struct bpf_map
*map
)
11536 return (map
->map_type
== BPF_MAP_TYPE_CGROUP_STORAGE
||
11537 map
->map_type
== BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE
);
11540 /* find and rewrite pseudo imm in ld_imm64 instructions:
11542 * 1. if it accesses map FD, replace it with actual map pointer.
11543 * 2. if it accesses btf_id of a VAR, replace it with pointer to the var.
11545 * NOTE: btf_vmlinux is required for converting pseudo btf_id.
11547 static int resolve_pseudo_ldimm64(struct bpf_verifier_env
*env
)
11549 struct bpf_insn
*insn
= env
->prog
->insnsi
;
11550 int insn_cnt
= env
->prog
->len
;
11553 err
= bpf_prog_calc_tag(env
->prog
);
11557 for (i
= 0; i
< insn_cnt
; i
++, insn
++) {
11558 if (BPF_CLASS(insn
->code
) == BPF_LDX
&&
11559 (BPF_MODE(insn
->code
) != BPF_MEM
|| insn
->imm
!= 0)) {
11560 verbose(env
, "BPF_LDX uses reserved fields\n");
11564 if (insn
[0].code
== (BPF_LD
| BPF_IMM
| BPF_DW
)) {
11565 struct bpf_insn_aux_data
*aux
;
11566 struct bpf_map
*map
;
11571 if (i
== insn_cnt
- 1 || insn
[1].code
!= 0 ||
11572 insn
[1].dst_reg
!= 0 || insn
[1].src_reg
!= 0 ||
11573 insn
[1].off
!= 0) {
11574 verbose(env
, "invalid bpf_ld_imm64 insn\n");
11578 if (insn
[0].src_reg
== 0)
11579 /* valid generic load 64-bit imm */
11582 if (insn
[0].src_reg
== BPF_PSEUDO_BTF_ID
) {
11583 aux
= &env
->insn_aux_data
[i
];
11584 err
= check_pseudo_btf_id(env
, insn
, aux
);
11590 if (insn
[0].src_reg
== BPF_PSEUDO_FUNC
) {
11591 aux
= &env
->insn_aux_data
[i
];
11592 aux
->ptr_type
= PTR_TO_FUNC
;
11596 /* In final convert_pseudo_ld_imm64() step, this is
11597 * converted into regular 64-bit imm load insn.
11599 switch (insn
[0].src_reg
) {
11600 case BPF_PSEUDO_MAP_VALUE
:
11601 case BPF_PSEUDO_MAP_IDX_VALUE
:
11603 case BPF_PSEUDO_MAP_FD
:
11604 case BPF_PSEUDO_MAP_IDX
:
11605 if (insn
[1].imm
== 0)
11609 verbose(env
, "unrecognized bpf_ld_imm64 insn\n");
11613 switch (insn
[0].src_reg
) {
11614 case BPF_PSEUDO_MAP_IDX_VALUE
:
11615 case BPF_PSEUDO_MAP_IDX
:
11616 if (bpfptr_is_null(env
->fd_array
)) {
11617 verbose(env
, "fd_idx without fd_array is invalid\n");
11620 if (copy_from_bpfptr_offset(&fd
, env
->fd_array
,
11621 insn
[0].imm
* sizeof(fd
),
11631 map
= __bpf_map_get(f
);
11633 verbose(env
, "fd %d is not pointing to valid bpf_map\n",
11635 return PTR_ERR(map
);
11638 err
= check_map_prog_compatibility(env
, map
, env
->prog
);
11644 aux
= &env
->insn_aux_data
[i
];
11645 if (insn
[0].src_reg
== BPF_PSEUDO_MAP_FD
||
11646 insn
[0].src_reg
== BPF_PSEUDO_MAP_IDX
) {
11647 addr
= (unsigned long)map
;
11649 u32 off
= insn
[1].imm
;
11651 if (off
>= BPF_MAX_VAR_OFF
) {
11652 verbose(env
, "direct value offset of %u is not allowed\n", off
);
11657 if (!map
->ops
->map_direct_value_addr
) {
11658 verbose(env
, "no direct value access support for this map type\n");
11663 err
= map
->ops
->map_direct_value_addr(map
, &addr
, off
);
11665 verbose(env
, "invalid access to map value pointer, value_size=%u off=%u\n",
11666 map
->value_size
, off
);
11671 aux
->map_off
= off
;
11675 insn
[0].imm
= (u32
)addr
;
11676 insn
[1].imm
= addr
>> 32;
11678 /* check whether we recorded this map already */
11679 for (j
= 0; j
< env
->used_map_cnt
; j
++) {
11680 if (env
->used_maps
[j
] == map
) {
11681 aux
->map_index
= j
;
11687 if (env
->used_map_cnt
>= MAX_USED_MAPS
) {
11692 /* hold the map. If the program is rejected by verifier,
11693 * the map will be released by release_maps() or it
11694 * will be used by the valid program until it's unloaded
11695 * and all maps are released in free_used_maps()
11699 aux
->map_index
= env
->used_map_cnt
;
11700 env
->used_maps
[env
->used_map_cnt
++] = map
;
11702 if (bpf_map_is_cgroup_storage(map
) &&
11703 bpf_cgroup_storage_assign(env
->prog
->aux
, map
)) {
11704 verbose(env
, "only one cgroup storage of each type is allowed\n");
11716 /* Basic sanity check before we invest more work here. */
11717 if (!bpf_opcode_in_insntable(insn
->code
)) {
11718 verbose(env
, "unknown opcode %02x\n", insn
->code
);
11723 /* now all pseudo BPF_LD_IMM64 instructions load valid
11724 * 'struct bpf_map *' into a register instead of user map_fd.
11725 * These pointers will be used later by verifier to validate map access.
11730 /* drop refcnt of maps used by the rejected program */
11731 static void release_maps(struct bpf_verifier_env
*env
)
11733 __bpf_free_used_maps(env
->prog
->aux
, env
->used_maps
,
11734 env
->used_map_cnt
);
11737 /* drop refcnt of maps used by the rejected program */
11738 static void release_btfs(struct bpf_verifier_env
*env
)
11740 __bpf_free_used_btfs(env
->prog
->aux
, env
->used_btfs
,
11741 env
->used_btf_cnt
);
11744 /* convert pseudo BPF_LD_IMM64 into generic BPF_LD_IMM64 */
11745 static void convert_pseudo_ld_imm64(struct bpf_verifier_env
*env
)
11747 struct bpf_insn
*insn
= env
->prog
->insnsi
;
11748 int insn_cnt
= env
->prog
->len
;
11751 for (i
= 0; i
< insn_cnt
; i
++, insn
++) {
11752 if (insn
->code
!= (BPF_LD
| BPF_IMM
| BPF_DW
))
11754 if (insn
->src_reg
== BPF_PSEUDO_FUNC
)
11760 /* single env->prog->insni[off] instruction was replaced with the range
11761 * insni[off, off + cnt). Adjust corresponding insn_aux_data by copying
11762 * [0, off) and [off, end) to new locations, so the patched range stays zero
11764 static void adjust_insn_aux_data(struct bpf_verifier_env
*env
,
11765 struct bpf_insn_aux_data
*new_data
,
11766 struct bpf_prog
*new_prog
, u32 off
, u32 cnt
)
11768 struct bpf_insn_aux_data
*old_data
= env
->insn_aux_data
;
11769 struct bpf_insn
*insn
= new_prog
->insnsi
;
11770 u32 old_seen
= old_data
[off
].seen
;
11774 /* aux info at OFF always needs adjustment, no matter fast path
11775 * (cnt == 1) is taken or not. There is no guarantee INSN at OFF is the
11776 * original insn at old prog.
11778 old_data
[off
].zext_dst
= insn_has_def32(env
, insn
+ off
+ cnt
- 1);
11782 prog_len
= new_prog
->len
;
11784 memcpy(new_data
, old_data
, sizeof(struct bpf_insn_aux_data
) * off
);
11785 memcpy(new_data
+ off
+ cnt
- 1, old_data
+ off
,
11786 sizeof(struct bpf_insn_aux_data
) * (prog_len
- off
- cnt
+ 1));
11787 for (i
= off
; i
< off
+ cnt
- 1; i
++) {
11788 /* Expand insni[off]'s seen count to the patched range. */
11789 new_data
[i
].seen
= old_seen
;
11790 new_data
[i
].zext_dst
= insn_has_def32(env
, insn
+ i
);
11792 env
->insn_aux_data
= new_data
;
11796 static void adjust_subprog_starts(struct bpf_verifier_env
*env
, u32 off
, u32 len
)
11802 /* NOTE: fake 'exit' subprog should be updated as well. */
11803 for (i
= 0; i
<= env
->subprog_cnt
; i
++) {
11804 if (env
->subprog_info
[i
].start
<= off
)
11806 env
->subprog_info
[i
].start
+= len
- 1;
11810 static void adjust_poke_descs(struct bpf_prog
*prog
, u32 off
, u32 len
)
11812 struct bpf_jit_poke_descriptor
*tab
= prog
->aux
->poke_tab
;
11813 int i
, sz
= prog
->aux
->size_poke_tab
;
11814 struct bpf_jit_poke_descriptor
*desc
;
11816 for (i
= 0; i
< sz
; i
++) {
11818 if (desc
->insn_idx
<= off
)
11820 desc
->insn_idx
+= len
- 1;
11824 static struct bpf_prog
*bpf_patch_insn_data(struct bpf_verifier_env
*env
, u32 off
,
11825 const struct bpf_insn
*patch
, u32 len
)
11827 struct bpf_prog
*new_prog
;
11828 struct bpf_insn_aux_data
*new_data
= NULL
;
11831 new_data
= vzalloc(array_size(env
->prog
->len
+ len
- 1,
11832 sizeof(struct bpf_insn_aux_data
)));
11837 new_prog
= bpf_patch_insn_single(env
->prog
, off
, patch
, len
);
11838 if (IS_ERR(new_prog
)) {
11839 if (PTR_ERR(new_prog
) == -ERANGE
)
11841 "insn %d cannot be patched due to 16-bit range\n",
11842 env
->insn_aux_data
[off
].orig_idx
);
11846 adjust_insn_aux_data(env
, new_data
, new_prog
, off
, len
);
11847 adjust_subprog_starts(env
, off
, len
);
11848 adjust_poke_descs(new_prog
, off
, len
);
11852 static int adjust_subprog_starts_after_remove(struct bpf_verifier_env
*env
,
11857 /* find first prog starting at or after off (first to remove) */
11858 for (i
= 0; i
< env
->subprog_cnt
; i
++)
11859 if (env
->subprog_info
[i
].start
>= off
)
11861 /* find first prog starting at or after off + cnt (first to stay) */
11862 for (j
= i
; j
< env
->subprog_cnt
; j
++)
11863 if (env
->subprog_info
[j
].start
>= off
+ cnt
)
11865 /* if j doesn't start exactly at off + cnt, we are just removing
11866 * the front of previous prog
11868 if (env
->subprog_info
[j
].start
!= off
+ cnt
)
11872 struct bpf_prog_aux
*aux
= env
->prog
->aux
;
11875 /* move fake 'exit' subprog as well */
11876 move
= env
->subprog_cnt
+ 1 - j
;
11878 memmove(env
->subprog_info
+ i
,
11879 env
->subprog_info
+ j
,
11880 sizeof(*env
->subprog_info
) * move
);
11881 env
->subprog_cnt
-= j
- i
;
11883 /* remove func_info */
11884 if (aux
->func_info
) {
11885 move
= aux
->func_info_cnt
- j
;
11887 memmove(aux
->func_info
+ i
,
11888 aux
->func_info
+ j
,
11889 sizeof(*aux
->func_info
) * move
);
11890 aux
->func_info_cnt
-= j
- i
;
11891 /* func_info->insn_off is set after all code rewrites,
11892 * in adjust_btf_func() - no need to adjust
11896 /* convert i from "first prog to remove" to "first to adjust" */
11897 if (env
->subprog_info
[i
].start
== off
)
11901 /* update fake 'exit' subprog as well */
11902 for (; i
<= env
->subprog_cnt
; i
++)
11903 env
->subprog_info
[i
].start
-= cnt
;
11908 static int bpf_adj_linfo_after_remove(struct bpf_verifier_env
*env
, u32 off
,
11911 struct bpf_prog
*prog
= env
->prog
;
11912 u32 i
, l_off
, l_cnt
, nr_linfo
;
11913 struct bpf_line_info
*linfo
;
11915 nr_linfo
= prog
->aux
->nr_linfo
;
11919 linfo
= prog
->aux
->linfo
;
11921 /* find first line info to remove, count lines to be removed */
11922 for (i
= 0; i
< nr_linfo
; i
++)
11923 if (linfo
[i
].insn_off
>= off
)
11928 for (; i
< nr_linfo
; i
++)
11929 if (linfo
[i
].insn_off
< off
+ cnt
)
11934 /* First live insn doesn't match first live linfo, it needs to "inherit"
11935 * last removed linfo. prog is already modified, so prog->len == off
11936 * means no live instructions after (tail of the program was removed).
11938 if (prog
->len
!= off
&& l_cnt
&&
11939 (i
== nr_linfo
|| linfo
[i
].insn_off
!= off
+ cnt
)) {
11941 linfo
[--i
].insn_off
= off
+ cnt
;
11944 /* remove the line info which refer to the removed instructions */
11946 memmove(linfo
+ l_off
, linfo
+ i
,
11947 sizeof(*linfo
) * (nr_linfo
- i
));
11949 prog
->aux
->nr_linfo
-= l_cnt
;
11950 nr_linfo
= prog
->aux
->nr_linfo
;
11953 /* pull all linfo[i].insn_off >= off + cnt in by cnt */
11954 for (i
= l_off
; i
< nr_linfo
; i
++)
11955 linfo
[i
].insn_off
-= cnt
;
11957 /* fix up all subprogs (incl. 'exit') which start >= off */
11958 for (i
= 0; i
<= env
->subprog_cnt
; i
++)
11959 if (env
->subprog_info
[i
].linfo_idx
> l_off
) {
11960 /* program may have started in the removed region but
11961 * may not be fully removed
11963 if (env
->subprog_info
[i
].linfo_idx
>= l_off
+ l_cnt
)
11964 env
->subprog_info
[i
].linfo_idx
-= l_cnt
;
11966 env
->subprog_info
[i
].linfo_idx
= l_off
;
11972 static int verifier_remove_insns(struct bpf_verifier_env
*env
, u32 off
, u32 cnt
)
11974 struct bpf_insn_aux_data
*aux_data
= env
->insn_aux_data
;
11975 unsigned int orig_prog_len
= env
->prog
->len
;
11978 if (bpf_prog_is_dev_bound(env
->prog
->aux
))
11979 bpf_prog_offload_remove_insns(env
, off
, cnt
);
11981 err
= bpf_remove_insns(env
->prog
, off
, cnt
);
11985 err
= adjust_subprog_starts_after_remove(env
, off
, cnt
);
11989 err
= bpf_adj_linfo_after_remove(env
, off
, cnt
);
11993 memmove(aux_data
+ off
, aux_data
+ off
+ cnt
,
11994 sizeof(*aux_data
) * (orig_prog_len
- off
- cnt
));
11999 /* The verifier does more data flow analysis than llvm and will not
12000 * explore branches that are dead at run time. Malicious programs can
12001 * have dead code too. Therefore replace all dead at-run-time code
12004 * Just nops are not optimal, e.g. if they would sit at the end of the
12005 * program and through another bug we would manage to jump there, then
12006 * we'd execute beyond program memory otherwise. Returning exception
12007 * code also wouldn't work since we can have subprogs where the dead
12008 * code could be located.
12010 static void sanitize_dead_code(struct bpf_verifier_env
*env
)
12012 struct bpf_insn_aux_data
*aux_data
= env
->insn_aux_data
;
12013 struct bpf_insn trap
= BPF_JMP_IMM(BPF_JA
, 0, 0, -1);
12014 struct bpf_insn
*insn
= env
->prog
->insnsi
;
12015 const int insn_cnt
= env
->prog
->len
;
12018 for (i
= 0; i
< insn_cnt
; i
++) {
12019 if (aux_data
[i
].seen
)
12021 memcpy(insn
+ i
, &trap
, sizeof(trap
));
12022 aux_data
[i
].zext_dst
= false;
12026 static bool insn_is_cond_jump(u8 code
)
12030 if (BPF_CLASS(code
) == BPF_JMP32
)
12033 if (BPF_CLASS(code
) != BPF_JMP
)
12037 return op
!= BPF_JA
&& op
!= BPF_EXIT
&& op
!= BPF_CALL
;
12040 static void opt_hard_wire_dead_code_branches(struct bpf_verifier_env
*env
)
12042 struct bpf_insn_aux_data
*aux_data
= env
->insn_aux_data
;
12043 struct bpf_insn ja
= BPF_JMP_IMM(BPF_JA
, 0, 0, 0);
12044 struct bpf_insn
*insn
= env
->prog
->insnsi
;
12045 const int insn_cnt
= env
->prog
->len
;
12048 for (i
= 0; i
< insn_cnt
; i
++, insn
++) {
12049 if (!insn_is_cond_jump(insn
->code
))
12052 if (!aux_data
[i
+ 1].seen
)
12053 ja
.off
= insn
->off
;
12054 else if (!aux_data
[i
+ 1 + insn
->off
].seen
)
12059 if (bpf_prog_is_dev_bound(env
->prog
->aux
))
12060 bpf_prog_offload_replace_insn(env
, i
, &ja
);
12062 memcpy(insn
, &ja
, sizeof(ja
));
12066 static int opt_remove_dead_code(struct bpf_verifier_env
*env
)
12068 struct bpf_insn_aux_data
*aux_data
= env
->insn_aux_data
;
12069 int insn_cnt
= env
->prog
->len
;
12072 for (i
= 0; i
< insn_cnt
; i
++) {
12076 while (i
+ j
< insn_cnt
&& !aux_data
[i
+ j
].seen
)
12081 err
= verifier_remove_insns(env
, i
, j
);
12084 insn_cnt
= env
->prog
->len
;
12090 static int opt_remove_nops(struct bpf_verifier_env
*env
)
12092 const struct bpf_insn ja
= BPF_JMP_IMM(BPF_JA
, 0, 0, 0);
12093 struct bpf_insn
*insn
= env
->prog
->insnsi
;
12094 int insn_cnt
= env
->prog
->len
;
12097 for (i
= 0; i
< insn_cnt
; i
++) {
12098 if (memcmp(&insn
[i
], &ja
, sizeof(ja
)))
12101 err
= verifier_remove_insns(env
, i
, 1);
12111 static int opt_subreg_zext_lo32_rnd_hi32(struct bpf_verifier_env
*env
,
12112 const union bpf_attr
*attr
)
12114 struct bpf_insn
*patch
, zext_patch
[2], rnd_hi32_patch
[4];
12115 struct bpf_insn_aux_data
*aux
= env
->insn_aux_data
;
12116 int i
, patch_len
, delta
= 0, len
= env
->prog
->len
;
12117 struct bpf_insn
*insns
= env
->prog
->insnsi
;
12118 struct bpf_prog
*new_prog
;
12121 rnd_hi32
= attr
->prog_flags
& BPF_F_TEST_RND_HI32
;
12122 zext_patch
[1] = BPF_ZEXT_REG(0);
12123 rnd_hi32_patch
[1] = BPF_ALU64_IMM(BPF_MOV
, BPF_REG_AX
, 0);
12124 rnd_hi32_patch
[2] = BPF_ALU64_IMM(BPF_LSH
, BPF_REG_AX
, 32);
12125 rnd_hi32_patch
[3] = BPF_ALU64_REG(BPF_OR
, 0, BPF_REG_AX
);
12126 for (i
= 0; i
< len
; i
++) {
12127 int adj_idx
= i
+ delta
;
12128 struct bpf_insn insn
;
12131 insn
= insns
[adj_idx
];
12132 load_reg
= insn_def_regno(&insn
);
12133 if (!aux
[adj_idx
].zext_dst
) {
12141 class = BPF_CLASS(code
);
12142 if (load_reg
== -1)
12145 /* NOTE: arg "reg" (the fourth one) is only used for
12146 * BPF_STX + SRC_OP, so it is safe to pass NULL
12149 if (is_reg64(env
, &insn
, load_reg
, NULL
, DST_OP
)) {
12150 if (class == BPF_LD
&&
12151 BPF_MODE(code
) == BPF_IMM
)
12156 /* ctx load could be transformed into wider load. */
12157 if (class == BPF_LDX
&&
12158 aux
[adj_idx
].ptr_type
== PTR_TO_CTX
)
12161 imm_rnd
= get_random_int();
12162 rnd_hi32_patch
[0] = insn
;
12163 rnd_hi32_patch
[1].imm
= imm_rnd
;
12164 rnd_hi32_patch
[3].dst_reg
= load_reg
;
12165 patch
= rnd_hi32_patch
;
12167 goto apply_patch_buffer
;
12170 /* Add in an zero-extend instruction if a) the JIT has requested
12171 * it or b) it's a CMPXCHG.
12173 * The latter is because: BPF_CMPXCHG always loads a value into
12174 * R0, therefore always zero-extends. However some archs'
12175 * equivalent instruction only does this load when the
12176 * comparison is successful. This detail of CMPXCHG is
12177 * orthogonal to the general zero-extension behaviour of the
12178 * CPU, so it's treated independently of bpf_jit_needs_zext.
12180 if (!bpf_jit_needs_zext() && !is_cmpxchg_insn(&insn
))
12183 if (WARN_ON(load_reg
== -1)) {
12184 verbose(env
, "verifier bug. zext_dst is set, but no reg is defined\n");
12188 zext_patch
[0] = insn
;
12189 zext_patch
[1].dst_reg
= load_reg
;
12190 zext_patch
[1].src_reg
= load_reg
;
12191 patch
= zext_patch
;
12193 apply_patch_buffer
:
12194 new_prog
= bpf_patch_insn_data(env
, adj_idx
, patch
, patch_len
);
12197 env
->prog
= new_prog
;
12198 insns
= new_prog
->insnsi
;
12199 aux
= env
->insn_aux_data
;
12200 delta
+= patch_len
- 1;
12206 /* convert load instructions that access fields of a context type into a
12207 * sequence of instructions that access fields of the underlying structure:
12208 * struct __sk_buff -> struct sk_buff
12209 * struct bpf_sock_ops -> struct sock
12211 static int convert_ctx_accesses(struct bpf_verifier_env
*env
)
12213 const struct bpf_verifier_ops
*ops
= env
->ops
;
12214 int i
, cnt
, size
, ctx_field_size
, delta
= 0;
12215 const int insn_cnt
= env
->prog
->len
;
12216 struct bpf_insn insn_buf
[16], *insn
;
12217 u32 target_size
, size_default
, off
;
12218 struct bpf_prog
*new_prog
;
12219 enum bpf_access_type type
;
12220 bool is_narrower_load
;
12222 if (ops
->gen_prologue
|| env
->seen_direct_write
) {
12223 if (!ops
->gen_prologue
) {
12224 verbose(env
, "bpf verifier is misconfigured\n");
12227 cnt
= ops
->gen_prologue(insn_buf
, env
->seen_direct_write
,
12229 if (cnt
>= ARRAY_SIZE(insn_buf
)) {
12230 verbose(env
, "bpf verifier is misconfigured\n");
12233 new_prog
= bpf_patch_insn_data(env
, 0, insn_buf
, cnt
);
12237 env
->prog
= new_prog
;
12242 if (bpf_prog_is_dev_bound(env
->prog
->aux
))
12245 insn
= env
->prog
->insnsi
+ delta
;
12247 for (i
= 0; i
< insn_cnt
; i
++, insn
++) {
12248 bpf_convert_ctx_access_t convert_ctx_access
;
12251 if (insn
->code
== (BPF_LDX
| BPF_MEM
| BPF_B
) ||
12252 insn
->code
== (BPF_LDX
| BPF_MEM
| BPF_H
) ||
12253 insn
->code
== (BPF_LDX
| BPF_MEM
| BPF_W
) ||
12254 insn
->code
== (BPF_LDX
| BPF_MEM
| BPF_DW
)) {
12257 } else if (insn
->code
== (BPF_STX
| BPF_MEM
| BPF_B
) ||
12258 insn
->code
== (BPF_STX
| BPF_MEM
| BPF_H
) ||
12259 insn
->code
== (BPF_STX
| BPF_MEM
| BPF_W
) ||
12260 insn
->code
== (BPF_STX
| BPF_MEM
| BPF_DW
) ||
12261 insn
->code
== (BPF_ST
| BPF_MEM
| BPF_B
) ||
12262 insn
->code
== (BPF_ST
| BPF_MEM
| BPF_H
) ||
12263 insn
->code
== (BPF_ST
| BPF_MEM
| BPF_W
) ||
12264 insn
->code
== (BPF_ST
| BPF_MEM
| BPF_DW
)) {
12266 ctx_access
= BPF_CLASS(insn
->code
) == BPF_STX
;
12271 if (type
== BPF_WRITE
&&
12272 env
->insn_aux_data
[i
+ delta
].sanitize_stack_spill
) {
12273 struct bpf_insn patch
[] = {
12278 cnt
= ARRAY_SIZE(patch
);
12279 new_prog
= bpf_patch_insn_data(env
, i
+ delta
, patch
, cnt
);
12284 env
->prog
= new_prog
;
12285 insn
= new_prog
->insnsi
+ i
+ delta
;
12292 switch (env
->insn_aux_data
[i
+ delta
].ptr_type
) {
12294 if (!ops
->convert_ctx_access
)
12296 convert_ctx_access
= ops
->convert_ctx_access
;
12298 case PTR_TO_SOCKET
:
12299 case PTR_TO_SOCK_COMMON
:
12300 convert_ctx_access
= bpf_sock_convert_ctx_access
;
12302 case PTR_TO_TCP_SOCK
:
12303 convert_ctx_access
= bpf_tcp_sock_convert_ctx_access
;
12305 case PTR_TO_XDP_SOCK
:
12306 convert_ctx_access
= bpf_xdp_sock_convert_ctx_access
;
12308 case PTR_TO_BTF_ID
:
12309 if (type
== BPF_READ
) {
12310 insn
->code
= BPF_LDX
| BPF_PROBE_MEM
|
12311 BPF_SIZE((insn
)->code
);
12312 env
->prog
->aux
->num_exentries
++;
12313 } else if (resolve_prog_type(env
->prog
) != BPF_PROG_TYPE_STRUCT_OPS
) {
12314 verbose(env
, "Writes through BTF pointers are not allowed\n");
12322 ctx_field_size
= env
->insn_aux_data
[i
+ delta
].ctx_field_size
;
12323 size
= BPF_LDST_BYTES(insn
);
12325 /* If the read access is a narrower load of the field,
12326 * convert to a 4/8-byte load, to minimum program type specific
12327 * convert_ctx_access changes. If conversion is successful,
12328 * we will apply proper mask to the result.
12330 is_narrower_load
= size
< ctx_field_size
;
12331 size_default
= bpf_ctx_off_adjust_machine(ctx_field_size
);
12333 if (is_narrower_load
) {
12336 if (type
== BPF_WRITE
) {
12337 verbose(env
, "bpf verifier narrow ctx access misconfigured\n");
12342 if (ctx_field_size
== 4)
12344 else if (ctx_field_size
== 8)
12345 size_code
= BPF_DW
;
12347 insn
->off
= off
& ~(size_default
- 1);
12348 insn
->code
= BPF_LDX
| BPF_MEM
| size_code
;
12352 cnt
= convert_ctx_access(type
, insn
, insn_buf
, env
->prog
,
12354 if (cnt
== 0 || cnt
>= ARRAY_SIZE(insn_buf
) ||
12355 (ctx_field_size
&& !target_size
)) {
12356 verbose(env
, "bpf verifier is misconfigured\n");
12360 if (is_narrower_load
&& size
< target_size
) {
12361 u8 shift
= bpf_ctx_narrow_access_offset(
12362 off
, size
, size_default
) * 8;
12363 if (shift
&& cnt
+ 1 >= ARRAY_SIZE(insn_buf
)) {
12364 verbose(env
, "bpf verifier narrow ctx load misconfigured\n");
12367 if (ctx_field_size
<= 4) {
12369 insn_buf
[cnt
++] = BPF_ALU32_IMM(BPF_RSH
,
12372 insn_buf
[cnt
++] = BPF_ALU32_IMM(BPF_AND
, insn
->dst_reg
,
12373 (1 << size
* 8) - 1);
12376 insn_buf
[cnt
++] = BPF_ALU64_IMM(BPF_RSH
,
12379 insn_buf
[cnt
++] = BPF_ALU64_IMM(BPF_AND
, insn
->dst_reg
,
12380 (1ULL << size
* 8) - 1);
12384 new_prog
= bpf_patch_insn_data(env
, i
+ delta
, insn_buf
, cnt
);
12390 /* keep walking new program and skip insns we just inserted */
12391 env
->prog
= new_prog
;
12392 insn
= new_prog
->insnsi
+ i
+ delta
;
12398 static int jit_subprogs(struct bpf_verifier_env
*env
)
12400 struct bpf_prog
*prog
= env
->prog
, **func
, *tmp
;
12401 int i
, j
, subprog_start
, subprog_end
= 0, len
, subprog
;
12402 struct bpf_map
*map_ptr
;
12403 struct bpf_insn
*insn
;
12404 void *old_bpf_func
;
12405 int err
, num_exentries
;
12407 if (env
->subprog_cnt
<= 1)
12410 for (i
= 0, insn
= prog
->insnsi
; i
< prog
->len
; i
++, insn
++) {
12411 if (bpf_pseudo_func(insn
)) {
12412 env
->insn_aux_data
[i
].call_imm
= insn
->imm
;
12413 /* subprog is encoded in insn[1].imm */
12417 if (!bpf_pseudo_call(insn
))
12419 /* Upon error here we cannot fall back to interpreter but
12420 * need a hard reject of the program. Thus -EFAULT is
12421 * propagated in any case.
12423 subprog
= find_subprog(env
, i
+ insn
->imm
+ 1);
12425 WARN_ONCE(1, "verifier bug. No program starts at insn %d\n",
12426 i
+ insn
->imm
+ 1);
12429 /* temporarily remember subprog id inside insn instead of
12430 * aux_data, since next loop will split up all insns into funcs
12432 insn
->off
= subprog
;
12433 /* remember original imm in case JIT fails and fallback
12434 * to interpreter will be needed
12436 env
->insn_aux_data
[i
].call_imm
= insn
->imm
;
12437 /* point imm to __bpf_call_base+1 from JITs point of view */
12441 err
= bpf_prog_alloc_jited_linfo(prog
);
12443 goto out_undo_insn
;
12446 func
= kcalloc(env
->subprog_cnt
, sizeof(prog
), GFP_KERNEL
);
12448 goto out_undo_insn
;
12450 for (i
= 0; i
< env
->subprog_cnt
; i
++) {
12451 subprog_start
= subprog_end
;
12452 subprog_end
= env
->subprog_info
[i
+ 1].start
;
12454 len
= subprog_end
- subprog_start
;
12455 /* bpf_prog_run() doesn't call subprogs directly,
12456 * hence main prog stats include the runtime of subprogs.
12457 * subprogs don't have IDs and not reachable via prog_get_next_id
12458 * func[i]->stats will never be accessed and stays NULL
12460 func
[i
] = bpf_prog_alloc_no_stats(bpf_prog_size(len
), GFP_USER
);
12463 memcpy(func
[i
]->insnsi
, &prog
->insnsi
[subprog_start
],
12464 len
* sizeof(struct bpf_insn
));
12465 func
[i
]->type
= prog
->type
;
12466 func
[i
]->len
= len
;
12467 if (bpf_prog_calc_tag(func
[i
]))
12469 func
[i
]->is_func
= 1;
12470 func
[i
]->aux
->func_idx
= i
;
12471 /* Below members will be freed only at prog->aux */
12472 func
[i
]->aux
->btf
= prog
->aux
->btf
;
12473 func
[i
]->aux
->func_info
= prog
->aux
->func_info
;
12474 func
[i
]->aux
->poke_tab
= prog
->aux
->poke_tab
;
12475 func
[i
]->aux
->size_poke_tab
= prog
->aux
->size_poke_tab
;
12477 for (j
= 0; j
< prog
->aux
->size_poke_tab
; j
++) {
12478 struct bpf_jit_poke_descriptor
*poke
;
12480 poke
= &prog
->aux
->poke_tab
[j
];
12481 if (poke
->insn_idx
< subprog_end
&&
12482 poke
->insn_idx
>= subprog_start
)
12483 poke
->aux
= func
[i
]->aux
;
12486 /* Use bpf_prog_F_tag to indicate functions in stack traces.
12487 * Long term would need debug info to populate names
12489 func
[i
]->aux
->name
[0] = 'F';
12490 func
[i
]->aux
->stack_depth
= env
->subprog_info
[i
].stack_depth
;
12491 func
[i
]->jit_requested
= 1;
12492 func
[i
]->aux
->kfunc_tab
= prog
->aux
->kfunc_tab
;
12493 func
[i
]->aux
->linfo
= prog
->aux
->linfo
;
12494 func
[i
]->aux
->nr_linfo
= prog
->aux
->nr_linfo
;
12495 func
[i
]->aux
->jited_linfo
= prog
->aux
->jited_linfo
;
12496 func
[i
]->aux
->linfo_idx
= env
->subprog_info
[i
].linfo_idx
;
12498 insn
= func
[i
]->insnsi
;
12499 for (j
= 0; j
< func
[i
]->len
; j
++, insn
++) {
12500 if (BPF_CLASS(insn
->code
) == BPF_LDX
&&
12501 BPF_MODE(insn
->code
) == BPF_PROBE_MEM
)
12504 func
[i
]->aux
->num_exentries
= num_exentries
;
12505 func
[i
]->aux
->tail_call_reachable
= env
->subprog_info
[i
].tail_call_reachable
;
12506 func
[i
] = bpf_int_jit_compile(func
[i
]);
12507 if (!func
[i
]->jited
) {
12514 /* at this point all bpf functions were successfully JITed
12515 * now populate all bpf_calls with correct addresses and
12516 * run last pass of JIT
12518 for (i
= 0; i
< env
->subprog_cnt
; i
++) {
12519 insn
= func
[i
]->insnsi
;
12520 for (j
= 0; j
< func
[i
]->len
; j
++, insn
++) {
12521 if (bpf_pseudo_func(insn
)) {
12522 subprog
= insn
[1].imm
;
12523 insn
[0].imm
= (u32
)(long)func
[subprog
]->bpf_func
;
12524 insn
[1].imm
= ((u64
)(long)func
[subprog
]->bpf_func
) >> 32;
12527 if (!bpf_pseudo_call(insn
))
12529 subprog
= insn
->off
;
12530 insn
->imm
= BPF_CAST_CALL(func
[subprog
]->bpf_func
) -
12534 /* we use the aux data to keep a list of the start addresses
12535 * of the JITed images for each function in the program
12537 * for some architectures, such as powerpc64, the imm field
12538 * might not be large enough to hold the offset of the start
12539 * address of the callee's JITed image from __bpf_call_base
12541 * in such cases, we can lookup the start address of a callee
12542 * by using its subprog id, available from the off field of
12543 * the call instruction, as an index for this list
12545 func
[i
]->aux
->func
= func
;
12546 func
[i
]->aux
->func_cnt
= env
->subprog_cnt
;
12548 for (i
= 0; i
< env
->subprog_cnt
; i
++) {
12549 old_bpf_func
= func
[i
]->bpf_func
;
12550 tmp
= bpf_int_jit_compile(func
[i
]);
12551 if (tmp
!= func
[i
] || func
[i
]->bpf_func
!= old_bpf_func
) {
12552 verbose(env
, "JIT doesn't support bpf-to-bpf calls\n");
12559 /* finally lock prog and jit images for all functions and
12560 * populate kallsysm
12562 for (i
= 0; i
< env
->subprog_cnt
; i
++) {
12563 bpf_prog_lock_ro(func
[i
]);
12564 bpf_prog_kallsyms_add(func
[i
]);
12567 /* Last step: make now unused interpreter insns from main
12568 * prog consistent for later dump requests, so they can
12569 * later look the same as if they were interpreted only.
12571 for (i
= 0, insn
= prog
->insnsi
; i
< prog
->len
; i
++, insn
++) {
12572 if (bpf_pseudo_func(insn
)) {
12573 insn
[0].imm
= env
->insn_aux_data
[i
].call_imm
;
12574 insn
[1].imm
= find_subprog(env
, i
+ insn
[0].imm
+ 1);
12577 if (!bpf_pseudo_call(insn
))
12579 insn
->off
= env
->insn_aux_data
[i
].call_imm
;
12580 subprog
= find_subprog(env
, i
+ insn
->off
+ 1);
12581 insn
->imm
= subprog
;
12585 prog
->bpf_func
= func
[0]->bpf_func
;
12586 prog
->aux
->func
= func
;
12587 prog
->aux
->func_cnt
= env
->subprog_cnt
;
12588 bpf_prog_jit_attempt_done(prog
);
12591 /* We failed JIT'ing, so at this point we need to unregister poke
12592 * descriptors from subprogs, so that kernel is not attempting to
12593 * patch it anymore as we're freeing the subprog JIT memory.
12595 for (i
= 0; i
< prog
->aux
->size_poke_tab
; i
++) {
12596 map_ptr
= prog
->aux
->poke_tab
[i
].tail_call
.map
;
12597 map_ptr
->ops
->map_poke_untrack(map_ptr
, prog
->aux
);
12599 /* At this point we're guaranteed that poke descriptors are not
12600 * live anymore. We can just unlink its descriptor table as it's
12601 * released with the main prog.
12603 for (i
= 0; i
< env
->subprog_cnt
; i
++) {
12606 func
[i
]->aux
->poke_tab
= NULL
;
12607 bpf_jit_free(func
[i
]);
12611 /* cleanup main prog to be interpreted */
12612 prog
->jit_requested
= 0;
12613 for (i
= 0, insn
= prog
->insnsi
; i
< prog
->len
; i
++, insn
++) {
12614 if (!bpf_pseudo_call(insn
))
12617 insn
->imm
= env
->insn_aux_data
[i
].call_imm
;
12619 bpf_prog_jit_attempt_done(prog
);
12623 static int fixup_call_args(struct bpf_verifier_env
*env
)
12625 #ifndef CONFIG_BPF_JIT_ALWAYS_ON
12626 struct bpf_prog
*prog
= env
->prog
;
12627 struct bpf_insn
*insn
= prog
->insnsi
;
12628 bool has_kfunc_call
= bpf_prog_has_kfunc_call(prog
);
12633 if (env
->prog
->jit_requested
&&
12634 !bpf_prog_is_dev_bound(env
->prog
->aux
)) {
12635 err
= jit_subprogs(env
);
12638 if (err
== -EFAULT
)
12641 #ifndef CONFIG_BPF_JIT_ALWAYS_ON
12642 if (has_kfunc_call
) {
12643 verbose(env
, "calling kernel functions are not allowed in non-JITed programs\n");
12646 if (env
->subprog_cnt
> 1 && env
->prog
->aux
->tail_call_reachable
) {
12647 /* When JIT fails the progs with bpf2bpf calls and tail_calls
12648 * have to be rejected, since interpreter doesn't support them yet.
12650 verbose(env
, "tail_calls are not allowed in non-JITed programs with bpf-to-bpf calls\n");
12653 for (i
= 0; i
< prog
->len
; i
++, insn
++) {
12654 if (bpf_pseudo_func(insn
)) {
12655 /* When JIT fails the progs with callback calls
12656 * have to be rejected, since interpreter doesn't support them yet.
12658 verbose(env
, "callbacks are not allowed in non-JITed programs\n");
12662 if (!bpf_pseudo_call(insn
))
12664 depth
= get_callee_stack_depth(env
, insn
, i
);
12667 bpf_patch_call_args(insn
, depth
);
12674 static int fixup_kfunc_call(struct bpf_verifier_env
*env
,
12675 struct bpf_insn
*insn
)
12677 const struct bpf_kfunc_desc
*desc
;
12679 /* insn->imm has the btf func_id. Replace it with
12680 * an address (relative to __bpf_base_call).
12682 desc
= find_kfunc_desc(env
->prog
, insn
->imm
);
12684 verbose(env
, "verifier internal error: kernel function descriptor not found for func_id %u\n",
12689 insn
->imm
= desc
->imm
;
12694 /* Do various post-verification rewrites in a single program pass.
12695 * These rewrites simplify JIT and interpreter implementations.
12697 static int do_misc_fixups(struct bpf_verifier_env
*env
)
12699 struct bpf_prog
*prog
= env
->prog
;
12700 bool expect_blinding
= bpf_jit_blinding_enabled(prog
);
12701 enum bpf_prog_type prog_type
= resolve_prog_type(prog
);
12702 struct bpf_insn
*insn
= prog
->insnsi
;
12703 const struct bpf_func_proto
*fn
;
12704 const int insn_cnt
= prog
->len
;
12705 const struct bpf_map_ops
*ops
;
12706 struct bpf_insn_aux_data
*aux
;
12707 struct bpf_insn insn_buf
[16];
12708 struct bpf_prog
*new_prog
;
12709 struct bpf_map
*map_ptr
;
12710 int i
, ret
, cnt
, delta
= 0;
12712 for (i
= 0; i
< insn_cnt
; i
++, insn
++) {
12713 /* Make divide-by-zero exceptions impossible. */
12714 if (insn
->code
== (BPF_ALU64
| BPF_MOD
| BPF_X
) ||
12715 insn
->code
== (BPF_ALU64
| BPF_DIV
| BPF_X
) ||
12716 insn
->code
== (BPF_ALU
| BPF_MOD
| BPF_X
) ||
12717 insn
->code
== (BPF_ALU
| BPF_DIV
| BPF_X
)) {
12718 bool is64
= BPF_CLASS(insn
->code
) == BPF_ALU64
;
12719 bool isdiv
= BPF_OP(insn
->code
) == BPF_DIV
;
12720 struct bpf_insn
*patchlet
;
12721 struct bpf_insn chk_and_div
[] = {
12722 /* [R,W]x div 0 -> 0 */
12723 BPF_RAW_INSN((is64
? BPF_JMP
: BPF_JMP32
) |
12724 BPF_JNE
| BPF_K
, insn
->src_reg
,
12726 BPF_ALU32_REG(BPF_XOR
, insn
->dst_reg
, insn
->dst_reg
),
12727 BPF_JMP_IMM(BPF_JA
, 0, 0, 1),
12730 struct bpf_insn chk_and_mod
[] = {
12731 /* [R,W]x mod 0 -> [R,W]x */
12732 BPF_RAW_INSN((is64
? BPF_JMP
: BPF_JMP32
) |
12733 BPF_JEQ
| BPF_K
, insn
->src_reg
,
12734 0, 1 + (is64
? 0 : 1), 0),
12736 BPF_JMP_IMM(BPF_JA
, 0, 0, 1),
12737 BPF_MOV32_REG(insn
->dst_reg
, insn
->dst_reg
),
12740 patchlet
= isdiv
? chk_and_div
: chk_and_mod
;
12741 cnt
= isdiv
? ARRAY_SIZE(chk_and_div
) :
12742 ARRAY_SIZE(chk_and_mod
) - (is64
? 2 : 0);
12744 new_prog
= bpf_patch_insn_data(env
, i
+ delta
, patchlet
, cnt
);
12749 env
->prog
= prog
= new_prog
;
12750 insn
= new_prog
->insnsi
+ i
+ delta
;
12754 /* Implement LD_ABS and LD_IND with a rewrite, if supported by the program type. */
12755 if (BPF_CLASS(insn
->code
) == BPF_LD
&&
12756 (BPF_MODE(insn
->code
) == BPF_ABS
||
12757 BPF_MODE(insn
->code
) == BPF_IND
)) {
12758 cnt
= env
->ops
->gen_ld_abs(insn
, insn_buf
);
12759 if (cnt
== 0 || cnt
>= ARRAY_SIZE(insn_buf
)) {
12760 verbose(env
, "bpf verifier is misconfigured\n");
12764 new_prog
= bpf_patch_insn_data(env
, i
+ delta
, insn_buf
, cnt
);
12769 env
->prog
= prog
= new_prog
;
12770 insn
= new_prog
->insnsi
+ i
+ delta
;
12774 /* Rewrite pointer arithmetic to mitigate speculation attacks. */
12775 if (insn
->code
== (BPF_ALU64
| BPF_ADD
| BPF_X
) ||
12776 insn
->code
== (BPF_ALU64
| BPF_SUB
| BPF_X
)) {
12777 const u8 code_add
= BPF_ALU64
| BPF_ADD
| BPF_X
;
12778 const u8 code_sub
= BPF_ALU64
| BPF_SUB
| BPF_X
;
12779 struct bpf_insn
*patch
= &insn_buf
[0];
12780 bool issrc
, isneg
, isimm
;
12783 aux
= &env
->insn_aux_data
[i
+ delta
];
12784 if (!aux
->alu_state
||
12785 aux
->alu_state
== BPF_ALU_NON_POINTER
)
12788 isneg
= aux
->alu_state
& BPF_ALU_NEG_VALUE
;
12789 issrc
= (aux
->alu_state
& BPF_ALU_SANITIZE
) ==
12790 BPF_ALU_SANITIZE_SRC
;
12791 isimm
= aux
->alu_state
& BPF_ALU_IMMEDIATE
;
12793 off_reg
= issrc
? insn
->src_reg
: insn
->dst_reg
;
12795 *patch
++ = BPF_MOV32_IMM(BPF_REG_AX
, aux
->alu_limit
);
12798 *patch
++ = BPF_ALU64_IMM(BPF_MUL
, off_reg
, -1);
12799 *patch
++ = BPF_MOV32_IMM(BPF_REG_AX
, aux
->alu_limit
);
12800 *patch
++ = BPF_ALU64_REG(BPF_SUB
, BPF_REG_AX
, off_reg
);
12801 *patch
++ = BPF_ALU64_REG(BPF_OR
, BPF_REG_AX
, off_reg
);
12802 *patch
++ = BPF_ALU64_IMM(BPF_NEG
, BPF_REG_AX
, 0);
12803 *patch
++ = BPF_ALU64_IMM(BPF_ARSH
, BPF_REG_AX
, 63);
12804 *patch
++ = BPF_ALU64_REG(BPF_AND
, BPF_REG_AX
, off_reg
);
12807 *patch
++ = BPF_MOV64_REG(insn
->dst_reg
, insn
->src_reg
);
12808 insn
->src_reg
= BPF_REG_AX
;
12810 insn
->code
= insn
->code
== code_add
?
12811 code_sub
: code_add
;
12813 if (issrc
&& isneg
&& !isimm
)
12814 *patch
++ = BPF_ALU64_IMM(BPF_MUL
, off_reg
, -1);
12815 cnt
= patch
- insn_buf
;
12817 new_prog
= bpf_patch_insn_data(env
, i
+ delta
, insn_buf
, cnt
);
12822 env
->prog
= prog
= new_prog
;
12823 insn
= new_prog
->insnsi
+ i
+ delta
;
12827 if (insn
->code
!= (BPF_JMP
| BPF_CALL
))
12829 if (insn
->src_reg
== BPF_PSEUDO_CALL
)
12831 if (insn
->src_reg
== BPF_PSEUDO_KFUNC_CALL
) {
12832 ret
= fixup_kfunc_call(env
, insn
);
12838 if (insn
->imm
== BPF_FUNC_get_route_realm
)
12839 prog
->dst_needed
= 1;
12840 if (insn
->imm
== BPF_FUNC_get_prandom_u32
)
12841 bpf_user_rnd_init_once();
12842 if (insn
->imm
== BPF_FUNC_override_return
)
12843 prog
->kprobe_override
= 1;
12844 if (insn
->imm
== BPF_FUNC_tail_call
) {
12845 /* If we tail call into other programs, we
12846 * cannot make any assumptions since they can
12847 * be replaced dynamically during runtime in
12848 * the program array.
12850 prog
->cb_access
= 1;
12851 if (!allow_tail_call_in_subprogs(env
))
12852 prog
->aux
->stack_depth
= MAX_BPF_STACK
;
12853 prog
->aux
->max_pkt_offset
= MAX_PACKET_OFF
;
12855 /* mark bpf_tail_call as different opcode to avoid
12856 * conditional branch in the interpreter for every normal
12857 * call and to prevent accidental JITing by JIT compiler
12858 * that doesn't support bpf_tail_call yet
12861 insn
->code
= BPF_JMP
| BPF_TAIL_CALL
;
12863 aux
= &env
->insn_aux_data
[i
+ delta
];
12864 if (env
->bpf_capable
&& !expect_blinding
&&
12865 prog
->jit_requested
&&
12866 !bpf_map_key_poisoned(aux
) &&
12867 !bpf_map_ptr_poisoned(aux
) &&
12868 !bpf_map_ptr_unpriv(aux
)) {
12869 struct bpf_jit_poke_descriptor desc
= {
12870 .reason
= BPF_POKE_REASON_TAIL_CALL
,
12871 .tail_call
.map
= BPF_MAP_PTR(aux
->map_ptr_state
),
12872 .tail_call
.key
= bpf_map_key_immediate(aux
),
12873 .insn_idx
= i
+ delta
,
12876 ret
= bpf_jit_add_poke_descriptor(prog
, &desc
);
12878 verbose(env
, "adding tail call poke descriptor failed\n");
12882 insn
->imm
= ret
+ 1;
12886 if (!bpf_map_ptr_unpriv(aux
))
12889 /* instead of changing every JIT dealing with tail_call
12890 * emit two extra insns:
12891 * if (index >= max_entries) goto out;
12892 * index &= array->index_mask;
12893 * to avoid out-of-bounds cpu speculation
12895 if (bpf_map_ptr_poisoned(aux
)) {
12896 verbose(env
, "tail_call abusing map_ptr\n");
12900 map_ptr
= BPF_MAP_PTR(aux
->map_ptr_state
);
12901 insn_buf
[0] = BPF_JMP_IMM(BPF_JGE
, BPF_REG_3
,
12902 map_ptr
->max_entries
, 2);
12903 insn_buf
[1] = BPF_ALU32_IMM(BPF_AND
, BPF_REG_3
,
12904 container_of(map_ptr
,
12907 insn_buf
[2] = *insn
;
12909 new_prog
= bpf_patch_insn_data(env
, i
+ delta
, insn_buf
, cnt
);
12914 env
->prog
= prog
= new_prog
;
12915 insn
= new_prog
->insnsi
+ i
+ delta
;
12919 if (insn
->imm
== BPF_FUNC_timer_set_callback
) {
12920 /* The verifier will process callback_fn as many times as necessary
12921 * with different maps and the register states prepared by
12922 * set_timer_callback_state will be accurate.
12924 * The following use case is valid:
12925 * map1 is shared by prog1, prog2, prog3.
12926 * prog1 calls bpf_timer_init for some map1 elements
12927 * prog2 calls bpf_timer_set_callback for some map1 elements.
12928 * Those that were not bpf_timer_init-ed will return -EINVAL.
12929 * prog3 calls bpf_timer_start for some map1 elements.
12930 * Those that were not both bpf_timer_init-ed and
12931 * bpf_timer_set_callback-ed will return -EINVAL.
12933 struct bpf_insn ld_addrs
[2] = {
12934 BPF_LD_IMM64(BPF_REG_3
, (long)prog
->aux
),
12937 insn_buf
[0] = ld_addrs
[0];
12938 insn_buf
[1] = ld_addrs
[1];
12939 insn_buf
[2] = *insn
;
12942 new_prog
= bpf_patch_insn_data(env
, i
+ delta
, insn_buf
, cnt
);
12947 env
->prog
= prog
= new_prog
;
12948 insn
= new_prog
->insnsi
+ i
+ delta
;
12949 goto patch_call_imm
;
12952 /* BPF_EMIT_CALL() assumptions in some of the map_gen_lookup
12953 * and other inlining handlers are currently limited to 64 bit
12956 if (prog
->jit_requested
&& BITS_PER_LONG
== 64 &&
12957 (insn
->imm
== BPF_FUNC_map_lookup_elem
||
12958 insn
->imm
== BPF_FUNC_map_update_elem
||
12959 insn
->imm
== BPF_FUNC_map_delete_elem
||
12960 insn
->imm
== BPF_FUNC_map_push_elem
||
12961 insn
->imm
== BPF_FUNC_map_pop_elem
||
12962 insn
->imm
== BPF_FUNC_map_peek_elem
||
12963 insn
->imm
== BPF_FUNC_redirect_map
)) {
12964 aux
= &env
->insn_aux_data
[i
+ delta
];
12965 if (bpf_map_ptr_poisoned(aux
))
12966 goto patch_call_imm
;
12968 map_ptr
= BPF_MAP_PTR(aux
->map_ptr_state
);
12969 ops
= map_ptr
->ops
;
12970 if (insn
->imm
== BPF_FUNC_map_lookup_elem
&&
12971 ops
->map_gen_lookup
) {
12972 cnt
= ops
->map_gen_lookup(map_ptr
, insn_buf
);
12973 if (cnt
== -EOPNOTSUPP
)
12974 goto patch_map_ops_generic
;
12975 if (cnt
<= 0 || cnt
>= ARRAY_SIZE(insn_buf
)) {
12976 verbose(env
, "bpf verifier is misconfigured\n");
12980 new_prog
= bpf_patch_insn_data(env
, i
+ delta
,
12986 env
->prog
= prog
= new_prog
;
12987 insn
= new_prog
->insnsi
+ i
+ delta
;
12991 BUILD_BUG_ON(!__same_type(ops
->map_lookup_elem
,
12992 (void *(*)(struct bpf_map
*map
, void *key
))NULL
));
12993 BUILD_BUG_ON(!__same_type(ops
->map_delete_elem
,
12994 (int (*)(struct bpf_map
*map
, void *key
))NULL
));
12995 BUILD_BUG_ON(!__same_type(ops
->map_update_elem
,
12996 (int (*)(struct bpf_map
*map
, void *key
, void *value
,
12998 BUILD_BUG_ON(!__same_type(ops
->map_push_elem
,
12999 (int (*)(struct bpf_map
*map
, void *value
,
13001 BUILD_BUG_ON(!__same_type(ops
->map_pop_elem
,
13002 (int (*)(struct bpf_map
*map
, void *value
))NULL
));
13003 BUILD_BUG_ON(!__same_type(ops
->map_peek_elem
,
13004 (int (*)(struct bpf_map
*map
, void *value
))NULL
));
13005 BUILD_BUG_ON(!__same_type(ops
->map_redirect
,
13006 (int (*)(struct bpf_map
*map
, u32 ifindex
, u64 flags
))NULL
));
13008 patch_map_ops_generic
:
13009 switch (insn
->imm
) {
13010 case BPF_FUNC_map_lookup_elem
:
13011 insn
->imm
= BPF_CAST_CALL(ops
->map_lookup_elem
) -
13014 case BPF_FUNC_map_update_elem
:
13015 insn
->imm
= BPF_CAST_CALL(ops
->map_update_elem
) -
13018 case BPF_FUNC_map_delete_elem
:
13019 insn
->imm
= BPF_CAST_CALL(ops
->map_delete_elem
) -
13022 case BPF_FUNC_map_push_elem
:
13023 insn
->imm
= BPF_CAST_CALL(ops
->map_push_elem
) -
13026 case BPF_FUNC_map_pop_elem
:
13027 insn
->imm
= BPF_CAST_CALL(ops
->map_pop_elem
) -
13030 case BPF_FUNC_map_peek_elem
:
13031 insn
->imm
= BPF_CAST_CALL(ops
->map_peek_elem
) -
13034 case BPF_FUNC_redirect_map
:
13035 insn
->imm
= BPF_CAST_CALL(ops
->map_redirect
) -
13040 goto patch_call_imm
;
13043 /* Implement bpf_jiffies64 inline. */
13044 if (prog
->jit_requested
&& BITS_PER_LONG
== 64 &&
13045 insn
->imm
== BPF_FUNC_jiffies64
) {
13046 struct bpf_insn ld_jiffies_addr
[2] = {
13047 BPF_LD_IMM64(BPF_REG_0
,
13048 (unsigned long)&jiffies
),
13051 insn_buf
[0] = ld_jiffies_addr
[0];
13052 insn_buf
[1] = ld_jiffies_addr
[1];
13053 insn_buf
[2] = BPF_LDX_MEM(BPF_DW
, BPF_REG_0
,
13057 new_prog
= bpf_patch_insn_data(env
, i
+ delta
, insn_buf
,
13063 env
->prog
= prog
= new_prog
;
13064 insn
= new_prog
->insnsi
+ i
+ delta
;
13068 /* Implement bpf_get_func_ip inline. */
13069 if (prog_type
== BPF_PROG_TYPE_TRACING
&&
13070 insn
->imm
== BPF_FUNC_get_func_ip
) {
13071 /* Load IP address from ctx - 8 */
13072 insn_buf
[0] = BPF_LDX_MEM(BPF_DW
, BPF_REG_0
, BPF_REG_1
, -8);
13074 new_prog
= bpf_patch_insn_data(env
, i
+ delta
, insn_buf
, 1);
13078 env
->prog
= prog
= new_prog
;
13079 insn
= new_prog
->insnsi
+ i
+ delta
;
13084 fn
= env
->ops
->get_func_proto(insn
->imm
, env
->prog
);
13085 /* all functions that have prototype and verifier allowed
13086 * programs to call them, must be real in-kernel functions
13090 "kernel subsystem misconfigured func %s#%d\n",
13091 func_id_name(insn
->imm
), insn
->imm
);
13094 insn
->imm
= fn
->func
- __bpf_call_base
;
13097 /* Since poke tab is now finalized, publish aux to tracker. */
13098 for (i
= 0; i
< prog
->aux
->size_poke_tab
; i
++) {
13099 map_ptr
= prog
->aux
->poke_tab
[i
].tail_call
.map
;
13100 if (!map_ptr
->ops
->map_poke_track
||
13101 !map_ptr
->ops
->map_poke_untrack
||
13102 !map_ptr
->ops
->map_poke_run
) {
13103 verbose(env
, "bpf verifier is misconfigured\n");
13107 ret
= map_ptr
->ops
->map_poke_track(map_ptr
, prog
->aux
);
13109 verbose(env
, "tracking tail call prog failed\n");
13114 sort_kfunc_descs_by_imm(env
->prog
);
13119 static void free_states(struct bpf_verifier_env
*env
)
13121 struct bpf_verifier_state_list
*sl
, *sln
;
13124 sl
= env
->free_list
;
13127 free_verifier_state(&sl
->state
, false);
13131 env
->free_list
= NULL
;
13133 if (!env
->explored_states
)
13136 for (i
= 0; i
< state_htab_size(env
); i
++) {
13137 sl
= env
->explored_states
[i
];
13141 free_verifier_state(&sl
->state
, false);
13145 env
->explored_states
[i
] = NULL
;
13149 static int do_check_common(struct bpf_verifier_env
*env
, int subprog
)
13151 bool pop_log
= !(env
->log
.level
& BPF_LOG_LEVEL2
);
13152 struct bpf_verifier_state
*state
;
13153 struct bpf_reg_state
*regs
;
13156 env
->prev_linfo
= NULL
;
13159 state
= kzalloc(sizeof(struct bpf_verifier_state
), GFP_KERNEL
);
13162 state
->curframe
= 0;
13163 state
->speculative
= false;
13164 state
->branches
= 1;
13165 state
->frame
[0] = kzalloc(sizeof(struct bpf_func_state
), GFP_KERNEL
);
13166 if (!state
->frame
[0]) {
13170 env
->cur_state
= state
;
13171 init_func_state(env
, state
->frame
[0],
13172 BPF_MAIN_FUNC
/* callsite */,
13176 regs
= state
->frame
[state
->curframe
]->regs
;
13177 if (subprog
|| env
->prog
->type
== BPF_PROG_TYPE_EXT
) {
13178 ret
= btf_prepare_func_args(env
, subprog
, regs
);
13181 for (i
= BPF_REG_1
; i
<= BPF_REG_5
; i
++) {
13182 if (regs
[i
].type
== PTR_TO_CTX
)
13183 mark_reg_known_zero(env
, regs
, i
);
13184 else if (regs
[i
].type
== SCALAR_VALUE
)
13185 mark_reg_unknown(env
, regs
, i
);
13186 else if (regs
[i
].type
== PTR_TO_MEM_OR_NULL
) {
13187 const u32 mem_size
= regs
[i
].mem_size
;
13189 mark_reg_known_zero(env
, regs
, i
);
13190 regs
[i
].mem_size
= mem_size
;
13191 regs
[i
].id
= ++env
->id_gen
;
13195 /* 1st arg to a function */
13196 regs
[BPF_REG_1
].type
= PTR_TO_CTX
;
13197 mark_reg_known_zero(env
, regs
, BPF_REG_1
);
13198 ret
= btf_check_subprog_arg_match(env
, subprog
, regs
);
13199 if (ret
== -EFAULT
)
13200 /* unlikely verifier bug. abort.
13201 * ret == 0 and ret < 0 are sadly acceptable for
13202 * main() function due to backward compatibility.
13203 * Like socket filter program may be written as:
13204 * int bpf_prog(struct pt_regs *ctx)
13205 * and never dereference that ctx in the program.
13206 * 'struct pt_regs' is a type mismatch for socket
13207 * filter that should be using 'struct __sk_buff'.
13212 ret
= do_check(env
);
13214 /* check for NULL is necessary, since cur_state can be freed inside
13215 * do_check() under memory pressure.
13217 if (env
->cur_state
) {
13218 free_verifier_state(env
->cur_state
, true);
13219 env
->cur_state
= NULL
;
13221 while (!pop_stack(env
, NULL
, NULL
, false));
13222 if (!ret
&& pop_log
)
13223 bpf_vlog_reset(&env
->log
, 0);
13228 /* Verify all global functions in a BPF program one by one based on their BTF.
13229 * All global functions must pass verification. Otherwise the whole program is rejected.
13240 * foo() will be verified first for R1=any_scalar_value. During verification it
13241 * will be assumed that bar() already verified successfully and call to bar()
13242 * from foo() will be checked for type match only. Later bar() will be verified
13243 * independently to check that it's safe for R1=any_scalar_value.
13245 static int do_check_subprogs(struct bpf_verifier_env
*env
)
13247 struct bpf_prog_aux
*aux
= env
->prog
->aux
;
13250 if (!aux
->func_info
)
13253 for (i
= 1; i
< env
->subprog_cnt
; i
++) {
13254 if (aux
->func_info_aux
[i
].linkage
!= BTF_FUNC_GLOBAL
)
13256 env
->insn_idx
= env
->subprog_info
[i
].start
;
13257 WARN_ON_ONCE(env
->insn_idx
== 0);
13258 ret
= do_check_common(env
, i
);
13261 } else if (env
->log
.level
& BPF_LOG_LEVEL
) {
13263 "Func#%d is safe for any args that match its prototype\n",
13270 static int do_check_main(struct bpf_verifier_env
*env
)
13275 ret
= do_check_common(env
, 0);
13277 env
->prog
->aux
->stack_depth
= env
->subprog_info
[0].stack_depth
;
13282 static void print_verification_stats(struct bpf_verifier_env
*env
)
13286 if (env
->log
.level
& BPF_LOG_STATS
) {
13287 verbose(env
, "verification time %lld usec\n",
13288 div_u64(env
->verification_time
, 1000));
13289 verbose(env
, "stack depth ");
13290 for (i
= 0; i
< env
->subprog_cnt
; i
++) {
13291 u32 depth
= env
->subprog_info
[i
].stack_depth
;
13293 verbose(env
, "%d", depth
);
13294 if (i
+ 1 < env
->subprog_cnt
)
13297 verbose(env
, "\n");
13299 verbose(env
, "processed %d insns (limit %d) max_states_per_insn %d "
13300 "total_states %d peak_states %d mark_read %d\n",
13301 env
->insn_processed
, BPF_COMPLEXITY_LIMIT_INSNS
,
13302 env
->max_states_per_insn
, env
->total_states
,
13303 env
->peak_states
, env
->longest_mark_read_walk
);
13306 static int check_struct_ops_btf_id(struct bpf_verifier_env
*env
)
13308 const struct btf_type
*t
, *func_proto
;
13309 const struct bpf_struct_ops
*st_ops
;
13310 const struct btf_member
*member
;
13311 struct bpf_prog
*prog
= env
->prog
;
13312 u32 btf_id
, member_idx
;
13315 if (!prog
->gpl_compatible
) {
13316 verbose(env
, "struct ops programs must have a GPL compatible license\n");
13320 btf_id
= prog
->aux
->attach_btf_id
;
13321 st_ops
= bpf_struct_ops_find(btf_id
);
13323 verbose(env
, "attach_btf_id %u is not a supported struct\n",
13329 member_idx
= prog
->expected_attach_type
;
13330 if (member_idx
>= btf_type_vlen(t
)) {
13331 verbose(env
, "attach to invalid member idx %u of struct %s\n",
13332 member_idx
, st_ops
->name
);
13336 member
= &btf_type_member(t
)[member_idx
];
13337 mname
= btf_name_by_offset(btf_vmlinux
, member
->name_off
);
13338 func_proto
= btf_type_resolve_func_ptr(btf_vmlinux
, member
->type
,
13341 verbose(env
, "attach to invalid member %s(@idx %u) of struct %s\n",
13342 mname
, member_idx
, st_ops
->name
);
13346 if (st_ops
->check_member
) {
13347 int err
= st_ops
->check_member(t
, member
);
13350 verbose(env
, "attach to unsupported member %s of struct %s\n",
13351 mname
, st_ops
->name
);
13356 prog
->aux
->attach_func_proto
= func_proto
;
13357 prog
->aux
->attach_func_name
= mname
;
13358 env
->ops
= st_ops
->verifier_ops
;
13362 #define SECURITY_PREFIX "security_"
13364 static int check_attach_modify_return(unsigned long addr
, const char *func_name
)
13366 if (within_error_injection_list(addr
) ||
13367 !strncmp(SECURITY_PREFIX
, func_name
, sizeof(SECURITY_PREFIX
) - 1))
13373 /* list of non-sleepable functions that are otherwise on
13374 * ALLOW_ERROR_INJECTION list
13376 BTF_SET_START(btf_non_sleepable_error_inject
)
13377 /* Three functions below can be called from sleepable and non-sleepable context.
13378 * Assume non-sleepable from bpf safety point of view.
13380 BTF_ID(func
, __add_to_page_cache_locked
)
13381 BTF_ID(func
, should_fail_alloc_page
)
13382 BTF_ID(func
, should_failslab
)
13383 BTF_SET_END(btf_non_sleepable_error_inject
)
13385 static int check_non_sleepable_error_inject(u32 btf_id
)
13387 return btf_id_set_contains(&btf_non_sleepable_error_inject
, btf_id
);
13390 int bpf_check_attach_target(struct bpf_verifier_log
*log
,
13391 const struct bpf_prog
*prog
,
13392 const struct bpf_prog
*tgt_prog
,
13394 struct bpf_attach_target_info
*tgt_info
)
13396 bool prog_extension
= prog
->type
== BPF_PROG_TYPE_EXT
;
13397 const char prefix
[] = "btf_trace_";
13398 int ret
= 0, subprog
= -1, i
;
13399 const struct btf_type
*t
;
13400 bool conservative
= true;
13406 bpf_log(log
, "Tracing programs must provide btf_id\n");
13409 btf
= tgt_prog
? tgt_prog
->aux
->btf
: prog
->aux
->attach_btf
;
13412 "FENTRY/FEXIT program can only be attached to another program annotated with BTF\n");
13415 t
= btf_type_by_id(btf
, btf_id
);
13417 bpf_log(log
, "attach_btf_id %u is invalid\n", btf_id
);
13420 tname
= btf_name_by_offset(btf
, t
->name_off
);
13422 bpf_log(log
, "attach_btf_id %u doesn't have a name\n", btf_id
);
13426 struct bpf_prog_aux
*aux
= tgt_prog
->aux
;
13428 for (i
= 0; i
< aux
->func_info_cnt
; i
++)
13429 if (aux
->func_info
[i
].type_id
== btf_id
) {
13433 if (subprog
== -1) {
13434 bpf_log(log
, "Subprog %s doesn't exist\n", tname
);
13437 conservative
= aux
->func_info_aux
[subprog
].unreliable
;
13438 if (prog_extension
) {
13439 if (conservative
) {
13441 "Cannot replace static functions\n");
13444 if (!prog
->jit_requested
) {
13446 "Extension programs should be JITed\n");
13450 if (!tgt_prog
->jited
) {
13451 bpf_log(log
, "Can attach to only JITed progs\n");
13454 if (tgt_prog
->type
== prog
->type
) {
13455 /* Cannot fentry/fexit another fentry/fexit program.
13456 * Cannot attach program extension to another extension.
13457 * It's ok to attach fentry/fexit to extension program.
13459 bpf_log(log
, "Cannot recursively attach\n");
13462 if (tgt_prog
->type
== BPF_PROG_TYPE_TRACING
&&
13464 (tgt_prog
->expected_attach_type
== BPF_TRACE_FENTRY
||
13465 tgt_prog
->expected_attach_type
== BPF_TRACE_FEXIT
)) {
13466 /* Program extensions can extend all program types
13467 * except fentry/fexit. The reason is the following.
13468 * The fentry/fexit programs are used for performance
13469 * analysis, stats and can be attached to any program
13470 * type except themselves. When extension program is
13471 * replacing XDP function it is necessary to allow
13472 * performance analysis of all functions. Both original
13473 * XDP program and its program extension. Hence
13474 * attaching fentry/fexit to BPF_PROG_TYPE_EXT is
13475 * allowed. If extending of fentry/fexit was allowed it
13476 * would be possible to create long call chain
13477 * fentry->extension->fentry->extension beyond
13478 * reasonable stack size. Hence extending fentry is not
13481 bpf_log(log
, "Cannot extend fentry/fexit\n");
13485 if (prog_extension
) {
13486 bpf_log(log
, "Cannot replace kernel functions\n");
13491 switch (prog
->expected_attach_type
) {
13492 case BPF_TRACE_RAW_TP
:
13495 "Only FENTRY/FEXIT progs are attachable to another BPF prog\n");
13498 if (!btf_type_is_typedef(t
)) {
13499 bpf_log(log
, "attach_btf_id %u is not a typedef\n",
13503 if (strncmp(prefix
, tname
, sizeof(prefix
) - 1)) {
13504 bpf_log(log
, "attach_btf_id %u points to wrong type name %s\n",
13508 tname
+= sizeof(prefix
) - 1;
13509 t
= btf_type_by_id(btf
, t
->type
);
13510 if (!btf_type_is_ptr(t
))
13511 /* should never happen in valid vmlinux build */
13513 t
= btf_type_by_id(btf
, t
->type
);
13514 if (!btf_type_is_func_proto(t
))
13515 /* should never happen in valid vmlinux build */
13519 case BPF_TRACE_ITER
:
13520 if (!btf_type_is_func(t
)) {
13521 bpf_log(log
, "attach_btf_id %u is not a function\n",
13525 t
= btf_type_by_id(btf
, t
->type
);
13526 if (!btf_type_is_func_proto(t
))
13528 ret
= btf_distill_func_proto(log
, btf
, t
, tname
, &tgt_info
->fmodel
);
13533 if (!prog_extension
)
13536 case BPF_MODIFY_RETURN
:
13538 case BPF_TRACE_FENTRY
:
13539 case BPF_TRACE_FEXIT
:
13540 if (!btf_type_is_func(t
)) {
13541 bpf_log(log
, "attach_btf_id %u is not a function\n",
13545 if (prog_extension
&&
13546 btf_check_type_match(log
, prog
, btf
, t
))
13548 t
= btf_type_by_id(btf
, t
->type
);
13549 if (!btf_type_is_func_proto(t
))
13552 if ((prog
->aux
->saved_dst_prog_type
|| prog
->aux
->saved_dst_attach_type
) &&
13553 (!tgt_prog
|| prog
->aux
->saved_dst_prog_type
!= tgt_prog
->type
||
13554 prog
->aux
->saved_dst_attach_type
!= tgt_prog
->expected_attach_type
))
13557 if (tgt_prog
&& conservative
)
13560 ret
= btf_distill_func_proto(log
, btf
, t
, tname
, &tgt_info
->fmodel
);
13566 addr
= (long) tgt_prog
->bpf_func
;
13568 addr
= (long) tgt_prog
->aux
->func
[subprog
]->bpf_func
;
13570 addr
= kallsyms_lookup_name(tname
);
13573 "The address of function %s cannot be found\n",
13579 if (prog
->aux
->sleepable
) {
13581 switch (prog
->type
) {
13582 case BPF_PROG_TYPE_TRACING
:
13583 /* fentry/fexit/fmod_ret progs can be sleepable only if they are
13584 * attached to ALLOW_ERROR_INJECTION and are not in denylist.
13586 if (!check_non_sleepable_error_inject(btf_id
) &&
13587 within_error_injection_list(addr
))
13590 case BPF_PROG_TYPE_LSM
:
13591 /* LSM progs check that they are attached to bpf_lsm_*() funcs.
13592 * Only some of them are sleepable.
13594 if (bpf_lsm_is_sleepable_hook(btf_id
))
13601 bpf_log(log
, "%s is not sleepable\n", tname
);
13604 } else if (prog
->expected_attach_type
== BPF_MODIFY_RETURN
) {
13606 bpf_log(log
, "can't modify return codes of BPF programs\n");
13609 ret
= check_attach_modify_return(addr
, tname
);
13611 bpf_log(log
, "%s() is not modifiable\n", tname
);
13618 tgt_info
->tgt_addr
= addr
;
13619 tgt_info
->tgt_name
= tname
;
13620 tgt_info
->tgt_type
= t
;
13624 BTF_SET_START(btf_id_deny
)
13627 BTF_ID(func
, migrate_disable
)
13628 BTF_ID(func
, migrate_enable
)
13630 #if !defined CONFIG_PREEMPT_RCU && !defined CONFIG_TINY_RCU
13631 BTF_ID(func
, rcu_read_unlock_strict
)
13633 BTF_SET_END(btf_id_deny
)
13635 static int check_attach_btf_id(struct bpf_verifier_env
*env
)
13637 struct bpf_prog
*prog
= env
->prog
;
13638 struct bpf_prog
*tgt_prog
= prog
->aux
->dst_prog
;
13639 struct bpf_attach_target_info tgt_info
= {};
13640 u32 btf_id
= prog
->aux
->attach_btf_id
;
13641 struct bpf_trampoline
*tr
;
13645 if (prog
->type
== BPF_PROG_TYPE_SYSCALL
) {
13646 if (prog
->aux
->sleepable
)
13647 /* attach_btf_id checked to be zero already */
13649 verbose(env
, "Syscall programs can only be sleepable\n");
13653 if (prog
->aux
->sleepable
&& prog
->type
!= BPF_PROG_TYPE_TRACING
&&
13654 prog
->type
!= BPF_PROG_TYPE_LSM
) {
13655 verbose(env
, "Only fentry/fexit/fmod_ret and lsm programs can be sleepable\n");
13659 if (prog
->type
== BPF_PROG_TYPE_STRUCT_OPS
)
13660 return check_struct_ops_btf_id(env
);
13662 if (prog
->type
!= BPF_PROG_TYPE_TRACING
&&
13663 prog
->type
!= BPF_PROG_TYPE_LSM
&&
13664 prog
->type
!= BPF_PROG_TYPE_EXT
)
13667 ret
= bpf_check_attach_target(&env
->log
, prog
, tgt_prog
, btf_id
, &tgt_info
);
13671 if (tgt_prog
&& prog
->type
== BPF_PROG_TYPE_EXT
) {
13672 /* to make freplace equivalent to their targets, they need to
13673 * inherit env->ops and expected_attach_type for the rest of the
13676 env
->ops
= bpf_verifier_ops
[tgt_prog
->type
];
13677 prog
->expected_attach_type
= tgt_prog
->expected_attach_type
;
13680 /* store info about the attachment target that will be used later */
13681 prog
->aux
->attach_func_proto
= tgt_info
.tgt_type
;
13682 prog
->aux
->attach_func_name
= tgt_info
.tgt_name
;
13685 prog
->aux
->saved_dst_prog_type
= tgt_prog
->type
;
13686 prog
->aux
->saved_dst_attach_type
= tgt_prog
->expected_attach_type
;
13689 if (prog
->expected_attach_type
== BPF_TRACE_RAW_TP
) {
13690 prog
->aux
->attach_btf_trace
= true;
13692 } else if (prog
->expected_attach_type
== BPF_TRACE_ITER
) {
13693 if (!bpf_iter_prog_supported(prog
))
13698 if (prog
->type
== BPF_PROG_TYPE_LSM
) {
13699 ret
= bpf_lsm_verify_prog(&env
->log
, prog
);
13702 } else if (prog
->type
== BPF_PROG_TYPE_TRACING
&&
13703 btf_id_set_contains(&btf_id_deny
, btf_id
)) {
13707 key
= bpf_trampoline_compute_key(tgt_prog
, prog
->aux
->attach_btf
, btf_id
);
13708 tr
= bpf_trampoline_get(key
, &tgt_info
);
13712 prog
->aux
->dst_trampoline
= tr
;
13716 struct btf
*bpf_get_btf_vmlinux(void)
13718 if (!btf_vmlinux
&& IS_ENABLED(CONFIG_DEBUG_INFO_BTF
)) {
13719 mutex_lock(&bpf_verifier_lock
);
13721 btf_vmlinux
= btf_parse_vmlinux();
13722 mutex_unlock(&bpf_verifier_lock
);
13724 return btf_vmlinux
;
13727 int bpf_check(struct bpf_prog
**prog
, union bpf_attr
*attr
, bpfptr_t uattr
)
13729 u64 start_time
= ktime_get_ns();
13730 struct bpf_verifier_env
*env
;
13731 struct bpf_verifier_log
*log
;
13732 int i
, len
, ret
= -EINVAL
;
13735 /* no program is valid */
13736 if (ARRAY_SIZE(bpf_verifier_ops
) == 0)
13739 /* 'struct bpf_verifier_env' can be global, but since it's not small,
13740 * allocate/free it every time bpf_check() is called
13742 env
= kzalloc(sizeof(struct bpf_verifier_env
), GFP_KERNEL
);
13747 len
= (*prog
)->len
;
13748 env
->insn_aux_data
=
13749 vzalloc(array_size(sizeof(struct bpf_insn_aux_data
), len
));
13751 if (!env
->insn_aux_data
)
13753 for (i
= 0; i
< len
; i
++)
13754 env
->insn_aux_data
[i
].orig_idx
= i
;
13756 env
->ops
= bpf_verifier_ops
[env
->prog
->type
];
13757 env
->fd_array
= make_bpfptr(attr
->fd_array
, uattr
.is_kernel
);
13758 is_priv
= bpf_capable();
13760 bpf_get_btf_vmlinux();
13762 /* grab the mutex to protect few globals used by verifier */
13764 mutex_lock(&bpf_verifier_lock
);
13766 if (attr
->log_level
|| attr
->log_buf
|| attr
->log_size
) {
13767 /* user requested verbose verifier output
13768 * and supplied buffer to store the verification trace
13770 log
->level
= attr
->log_level
;
13771 log
->ubuf
= (char __user
*) (unsigned long) attr
->log_buf
;
13772 log
->len_total
= attr
->log_size
;
13774 /* log attributes have to be sane */
13775 if (!bpf_verifier_log_attr_valid(log
)) {
13781 if (IS_ERR(btf_vmlinux
)) {
13782 /* Either gcc or pahole or kernel are broken. */
13783 verbose(env
, "in-kernel BTF is malformed\n");
13784 ret
= PTR_ERR(btf_vmlinux
);
13785 goto skip_full_check
;
13788 env
->strict_alignment
= !!(attr
->prog_flags
& BPF_F_STRICT_ALIGNMENT
);
13789 if (!IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS
))
13790 env
->strict_alignment
= true;
13791 if (attr
->prog_flags
& BPF_F_ANY_ALIGNMENT
)
13792 env
->strict_alignment
= false;
13794 env
->allow_ptr_leaks
= bpf_allow_ptr_leaks();
13795 env
->allow_uninit_stack
= bpf_allow_uninit_stack();
13796 env
->allow_ptr_to_map_access
= bpf_allow_ptr_to_map_access();
13797 env
->bypass_spec_v1
= bpf_bypass_spec_v1();
13798 env
->bypass_spec_v4
= bpf_bypass_spec_v4();
13799 env
->bpf_capable
= bpf_capable();
13802 env
->test_state_freq
= attr
->prog_flags
& BPF_F_TEST_STATE_FREQ
;
13804 env
->explored_states
= kvcalloc(state_htab_size(env
),
13805 sizeof(struct bpf_verifier_state_list
*),
13808 if (!env
->explored_states
)
13809 goto skip_full_check
;
13811 ret
= add_subprog_and_kfunc(env
);
13813 goto skip_full_check
;
13815 ret
= check_subprogs(env
);
13817 goto skip_full_check
;
13819 ret
= check_btf_info(env
, attr
, uattr
);
13821 goto skip_full_check
;
13823 ret
= check_attach_btf_id(env
);
13825 goto skip_full_check
;
13827 ret
= resolve_pseudo_ldimm64(env
);
13829 goto skip_full_check
;
13831 if (bpf_prog_is_dev_bound(env
->prog
->aux
)) {
13832 ret
= bpf_prog_offload_verifier_prep(env
->prog
);
13834 goto skip_full_check
;
13837 ret
= check_cfg(env
);
13839 goto skip_full_check
;
13841 ret
= do_check_subprogs(env
);
13842 ret
= ret
?: do_check_main(env
);
13844 if (ret
== 0 && bpf_prog_is_dev_bound(env
->prog
->aux
))
13845 ret
= bpf_prog_offload_finalize(env
);
13848 kvfree(env
->explored_states
);
13851 ret
= check_max_stack_depth(env
);
13853 /* instruction rewrites happen after this point */
13856 opt_hard_wire_dead_code_branches(env
);
13858 ret
= opt_remove_dead_code(env
);
13860 ret
= opt_remove_nops(env
);
13863 sanitize_dead_code(env
);
13867 /* program is valid, convert *(u32*)(ctx + off) accesses */
13868 ret
= convert_ctx_accesses(env
);
13871 ret
= do_misc_fixups(env
);
13873 /* do 32-bit optimization after insn patching has done so those patched
13874 * insns could be handled correctly.
13876 if (ret
== 0 && !bpf_prog_is_dev_bound(env
->prog
->aux
)) {
13877 ret
= opt_subreg_zext_lo32_rnd_hi32(env
, attr
);
13878 env
->prog
->aux
->verifier_zext
= bpf_jit_needs_zext() ? !ret
13883 ret
= fixup_call_args(env
);
13885 env
->verification_time
= ktime_get_ns() - start_time
;
13886 print_verification_stats(env
);
13888 if (log
->level
&& bpf_verifier_log_full(log
))
13890 if (log
->level
&& !log
->ubuf
) {
13892 goto err_release_maps
;
13896 goto err_release_maps
;
13898 if (env
->used_map_cnt
) {
13899 /* if program passed verifier, update used_maps in bpf_prog_info */
13900 env
->prog
->aux
->used_maps
= kmalloc_array(env
->used_map_cnt
,
13901 sizeof(env
->used_maps
[0]),
13904 if (!env
->prog
->aux
->used_maps
) {
13906 goto err_release_maps
;
13909 memcpy(env
->prog
->aux
->used_maps
, env
->used_maps
,
13910 sizeof(env
->used_maps
[0]) * env
->used_map_cnt
);
13911 env
->prog
->aux
->used_map_cnt
= env
->used_map_cnt
;
13913 if (env
->used_btf_cnt
) {
13914 /* if program passed verifier, update used_btfs in bpf_prog_aux */
13915 env
->prog
->aux
->used_btfs
= kmalloc_array(env
->used_btf_cnt
,
13916 sizeof(env
->used_btfs
[0]),
13918 if (!env
->prog
->aux
->used_btfs
) {
13920 goto err_release_maps
;
13923 memcpy(env
->prog
->aux
->used_btfs
, env
->used_btfs
,
13924 sizeof(env
->used_btfs
[0]) * env
->used_btf_cnt
);
13925 env
->prog
->aux
->used_btf_cnt
= env
->used_btf_cnt
;
13927 if (env
->used_map_cnt
|| env
->used_btf_cnt
) {
13928 /* program is valid. Convert pseudo bpf_ld_imm64 into generic
13929 * bpf_ld_imm64 instructions
13931 convert_pseudo_ld_imm64(env
);
13934 adjust_btf_func(env
);
13937 if (!env
->prog
->aux
->used_maps
)
13938 /* if we didn't copy map pointers into bpf_prog_info, release
13939 * them now. Otherwise free_used_maps() will release them.
13942 if (!env
->prog
->aux
->used_btfs
)
13945 /* extension progs temporarily inherit the attach_type of their targets
13946 for verification purposes, so set it back to zero before returning
13948 if (env
->prog
->type
== BPF_PROG_TYPE_EXT
)
13949 env
->prog
->expected_attach_type
= 0;
13954 mutex_unlock(&bpf_verifier_lock
);
13955 vfree(env
->insn_aux_data
);