1 /* SPDX-License-Identifier: GPL-2.0 */
3 * Linux Socket Filter Data Structures
5 #ifndef __LINUX_FILTER_H__
6 #define __LINUX_FILTER_H__
8 #include <linux/atomic.h>
10 #include <linux/refcount.h>
11 #include <linux/compat.h>
12 #include <linux/skbuff.h>
13 #include <linux/linkage.h>
14 #include <linux/printk.h>
15 #include <linux/workqueue.h>
16 #include <linux/sched.h>
17 #include <linux/sched/clock.h>
18 #include <linux/capability.h>
19 #include <linux/set_memory.h>
20 #include <linux/kallsyms.h>
21 #include <linux/if_vlan.h>
22 #include <linux/vmalloc.h>
23 #include <linux/sockptr.h>
24 #include <crypto/sha1.h>
25 #include <linux/u64_stats_sync.h>
27 #include <net/sch_generic.h>
29 #include <asm/byteorder.h>
30 #include <uapi/linux/filter.h>
38 struct sock_reuseport
;
40 struct ctl_table_header
;
42 /* ArgX, context and stack frame pointer register positions. Note,
43 * Arg1, Arg2, Arg3, etc are used as argument mappings of function
44 * calls in BPF_CALL instruction.
46 #define BPF_REG_ARG1 BPF_REG_1
47 #define BPF_REG_ARG2 BPF_REG_2
48 #define BPF_REG_ARG3 BPF_REG_3
49 #define BPF_REG_ARG4 BPF_REG_4
50 #define BPF_REG_ARG5 BPF_REG_5
51 #define BPF_REG_CTX BPF_REG_6
52 #define BPF_REG_FP BPF_REG_10
54 /* Additional register mappings for converted user programs. */
55 #define BPF_REG_A BPF_REG_0
56 #define BPF_REG_X BPF_REG_7
57 #define BPF_REG_TMP BPF_REG_2 /* scratch reg */
58 #define BPF_REG_D BPF_REG_8 /* data, callee-saved */
59 #define BPF_REG_H BPF_REG_9 /* hlen, callee-saved */
61 /* Kernel hidden auxiliary/helper register. */
62 #define BPF_REG_AX MAX_BPF_REG
63 #define MAX_BPF_EXT_REG (MAX_BPF_REG + 1)
64 #define MAX_BPF_JIT_REG MAX_BPF_EXT_REG
66 /* unused opcode to mark special call to bpf_tail_call() helper */
67 #define BPF_TAIL_CALL 0xf0
69 /* unused opcode to mark special load instruction. Same as BPF_ABS */
70 #define BPF_PROBE_MEM 0x20
72 /* unused opcode to mark special ldsx instruction. Same as BPF_IND */
73 #define BPF_PROBE_MEMSX 0x40
75 /* unused opcode to mark special load instruction. Same as BPF_MSH */
76 #define BPF_PROBE_MEM32 0xa0
78 /* unused opcode to mark special atomic instruction */
79 #define BPF_PROBE_ATOMIC 0xe0
81 /* unused opcode to mark call to interpreter with arguments */
82 #define BPF_CALL_ARGS 0xe0
84 /* unused opcode to mark speculation barrier for mitigating
85 * Speculative Store Bypass
87 #define BPF_NOSPEC 0xc0
89 /* As per nm, we expose JITed images as text (code) section for
90 * kallsyms. That way, tools like perf can find it to match
93 #define BPF_SYM_ELF_TYPE 't'
95 /* BPF program can access up to 512 bytes of stack space. */
96 #define MAX_BPF_STACK 512
98 /* Helper macros for filter block array initializers. */
100 /* ALU ops on registers, bpf_add|sub|...: dst_reg += src_reg */
102 #define BPF_ALU64_REG_OFF(OP, DST, SRC, OFF) \
103 ((struct bpf_insn) { \
104 .code = BPF_ALU64 | BPF_OP(OP) | BPF_X, \
110 #define BPF_ALU64_REG(OP, DST, SRC) \
111 BPF_ALU64_REG_OFF(OP, DST, SRC, 0)
113 #define BPF_ALU32_REG_OFF(OP, DST, SRC, OFF) \
114 ((struct bpf_insn) { \
115 .code = BPF_ALU | BPF_OP(OP) | BPF_X, \
121 #define BPF_ALU32_REG(OP, DST, SRC) \
122 BPF_ALU32_REG_OFF(OP, DST, SRC, 0)
124 /* ALU ops on immediates, bpf_add|sub|...: dst_reg += imm32 */
126 #define BPF_ALU64_IMM_OFF(OP, DST, IMM, OFF) \
127 ((struct bpf_insn) { \
128 .code = BPF_ALU64 | BPF_OP(OP) | BPF_K, \
133 #define BPF_ALU64_IMM(OP, DST, IMM) \
134 BPF_ALU64_IMM_OFF(OP, DST, IMM, 0)
136 #define BPF_ALU32_IMM_OFF(OP, DST, IMM, OFF) \
137 ((struct bpf_insn) { \
138 .code = BPF_ALU | BPF_OP(OP) | BPF_K, \
143 #define BPF_ALU32_IMM(OP, DST, IMM) \
144 BPF_ALU32_IMM_OFF(OP, DST, IMM, 0)
146 /* Endianess conversion, cpu_to_{l,b}e(), {l,b}e_to_cpu() */
148 #define BPF_ENDIAN(TYPE, DST, LEN) \
149 ((struct bpf_insn) { \
150 .code = BPF_ALU | BPF_END | BPF_SRC(TYPE), \
156 /* Byte Swap, bswap16/32/64 */
158 #define BPF_BSWAP(DST, LEN) \
159 ((struct bpf_insn) { \
160 .code = BPF_ALU64 | BPF_END | BPF_SRC(BPF_TO_LE), \
166 /* Short form of mov, dst_reg = src_reg */
168 #define BPF_MOV64_REG(DST, SRC) \
169 ((struct bpf_insn) { \
170 .code = BPF_ALU64 | BPF_MOV | BPF_X, \
176 #define BPF_MOV32_REG(DST, SRC) \
177 ((struct bpf_insn) { \
178 .code = BPF_ALU | BPF_MOV | BPF_X, \
184 /* Special (internal-only) form of mov, used to resolve per-CPU addrs:
185 * dst_reg = src_reg + <percpu_base_off>
186 * BPF_ADDR_PERCPU is used as a special insn->off value.
188 #define BPF_ADDR_PERCPU (-1)
190 #define BPF_MOV64_PERCPU_REG(DST, SRC) \
191 ((struct bpf_insn) { \
192 .code = BPF_ALU64 | BPF_MOV | BPF_X, \
195 .off = BPF_ADDR_PERCPU, \
198 static inline bool insn_is_mov_percpu_addr(const struct bpf_insn
*insn
)
200 return insn
->code
== (BPF_ALU64
| BPF_MOV
| BPF_X
) && insn
->off
== BPF_ADDR_PERCPU
;
203 /* Short form of mov, dst_reg = imm32 */
205 #define BPF_MOV64_IMM(DST, IMM) \
206 ((struct bpf_insn) { \
207 .code = BPF_ALU64 | BPF_MOV | BPF_K, \
213 #define BPF_MOV32_IMM(DST, IMM) \
214 ((struct bpf_insn) { \
215 .code = BPF_ALU | BPF_MOV | BPF_K, \
221 /* Short form of movsx, dst_reg = (s8,s16,s32)src_reg */
223 #define BPF_MOVSX64_REG(DST, SRC, OFF) \
224 ((struct bpf_insn) { \
225 .code = BPF_ALU64 | BPF_MOV | BPF_X, \
231 #define BPF_MOVSX32_REG(DST, SRC, OFF) \
232 ((struct bpf_insn) { \
233 .code = BPF_ALU | BPF_MOV | BPF_X, \
239 /* Special form of mov32, used for doing explicit zero extension on dst. */
240 #define BPF_ZEXT_REG(DST) \
241 ((struct bpf_insn) { \
242 .code = BPF_ALU | BPF_MOV | BPF_X, \
248 static inline bool insn_is_zext(const struct bpf_insn
*insn
)
250 return insn
->code
== (BPF_ALU
| BPF_MOV
| BPF_X
) && insn
->imm
== 1;
253 /* addr_space_cast from as(0) to as(1) is for converting bpf arena pointers
254 * to pointers in user vma.
256 static inline bool insn_is_cast_user(const struct bpf_insn
*insn
)
258 return insn
->code
== (BPF_ALU64
| BPF_MOV
| BPF_X
) &&
259 insn
->off
== BPF_ADDR_SPACE_CAST
&&
260 insn
->imm
== 1U << 16;
263 /* BPF_LD_IMM64 macro encodes single 'load 64-bit immediate' insn */
264 #define BPF_LD_IMM64(DST, IMM) \
265 BPF_LD_IMM64_RAW(DST, 0, IMM)
267 #define BPF_LD_IMM64_RAW(DST, SRC, IMM) \
268 ((struct bpf_insn) { \
269 .code = BPF_LD | BPF_DW | BPF_IMM, \
273 .imm = (__u32) (IMM) }), \
274 ((struct bpf_insn) { \
275 .code = 0, /* zero is reserved opcode */ \
279 .imm = ((__u64) (IMM)) >> 32 })
281 /* pseudo BPF_LD_IMM64 insn used to refer to process-local map_fd */
282 #define BPF_LD_MAP_FD(DST, MAP_FD) \
283 BPF_LD_IMM64_RAW(DST, BPF_PSEUDO_MAP_FD, MAP_FD)
285 /* Short form of mov based on type, BPF_X: dst_reg = src_reg, BPF_K: dst_reg = imm32 */
287 #define BPF_MOV64_RAW(TYPE, DST, SRC, IMM) \
288 ((struct bpf_insn) { \
289 .code = BPF_ALU64 | BPF_MOV | BPF_SRC(TYPE), \
295 #define BPF_MOV32_RAW(TYPE, DST, SRC, IMM) \
296 ((struct bpf_insn) { \
297 .code = BPF_ALU | BPF_MOV | BPF_SRC(TYPE), \
303 /* Direct packet access, R0 = *(uint *) (skb->data + imm32) */
305 #define BPF_LD_ABS(SIZE, IMM) \
306 ((struct bpf_insn) { \
307 .code = BPF_LD | BPF_SIZE(SIZE) | BPF_ABS, \
313 /* Indirect packet access, R0 = *(uint *) (skb->data + src_reg + imm32) */
315 #define BPF_LD_IND(SIZE, SRC, IMM) \
316 ((struct bpf_insn) { \
317 .code = BPF_LD | BPF_SIZE(SIZE) | BPF_IND, \
323 /* Memory load, dst_reg = *(uint *) (src_reg + off16) */
325 #define BPF_LDX_MEM(SIZE, DST, SRC, OFF) \
326 ((struct bpf_insn) { \
327 .code = BPF_LDX | BPF_SIZE(SIZE) | BPF_MEM, \
333 /* Memory load, dst_reg = *(signed size *) (src_reg + off16) */
335 #define BPF_LDX_MEMSX(SIZE, DST, SRC, OFF) \
336 ((struct bpf_insn) { \
337 .code = BPF_LDX | BPF_SIZE(SIZE) | BPF_MEMSX, \
343 /* Memory store, *(uint *) (dst_reg + off16) = src_reg */
345 #define BPF_STX_MEM(SIZE, DST, SRC, OFF) \
346 ((struct bpf_insn) { \
347 .code = BPF_STX | BPF_SIZE(SIZE) | BPF_MEM, \
357 * BPF_ADD *(uint *) (dst_reg + off16) += src_reg
358 * BPF_AND *(uint *) (dst_reg + off16) &= src_reg
359 * BPF_OR *(uint *) (dst_reg + off16) |= src_reg
360 * BPF_XOR *(uint *) (dst_reg + off16) ^= src_reg
361 * BPF_ADD | BPF_FETCH src_reg = atomic_fetch_add(dst_reg + off16, src_reg);
362 * BPF_AND | BPF_FETCH src_reg = atomic_fetch_and(dst_reg + off16, src_reg);
363 * BPF_OR | BPF_FETCH src_reg = atomic_fetch_or(dst_reg + off16, src_reg);
364 * BPF_XOR | BPF_FETCH src_reg = atomic_fetch_xor(dst_reg + off16, src_reg);
365 * BPF_XCHG src_reg = atomic_xchg(dst_reg + off16, src_reg)
366 * BPF_CMPXCHG r0 = atomic_cmpxchg(dst_reg + off16, r0, src_reg)
369 #define BPF_ATOMIC_OP(SIZE, OP, DST, SRC, OFF) \
370 ((struct bpf_insn) { \
371 .code = BPF_STX | BPF_SIZE(SIZE) | BPF_ATOMIC, \
378 #define BPF_STX_XADD(SIZE, DST, SRC, OFF) BPF_ATOMIC_OP(SIZE, BPF_ADD, DST, SRC, OFF)
380 /* Memory store, *(uint *) (dst_reg + off16) = imm32 */
382 #define BPF_ST_MEM(SIZE, DST, OFF, IMM) \
383 ((struct bpf_insn) { \
384 .code = BPF_ST | BPF_SIZE(SIZE) | BPF_MEM, \
390 /* Conditional jumps against registers, if (dst_reg 'op' src_reg) goto pc + off16 */
392 #define BPF_JMP_REG(OP, DST, SRC, OFF) \
393 ((struct bpf_insn) { \
394 .code = BPF_JMP | BPF_OP(OP) | BPF_X, \
400 /* Conditional jumps against immediates, if (dst_reg 'op' imm32) goto pc + off16 */
402 #define BPF_JMP_IMM(OP, DST, IMM, OFF) \
403 ((struct bpf_insn) { \
404 .code = BPF_JMP | BPF_OP(OP) | BPF_K, \
410 /* Like BPF_JMP_REG, but with 32-bit wide operands for comparison. */
412 #define BPF_JMP32_REG(OP, DST, SRC, OFF) \
413 ((struct bpf_insn) { \
414 .code = BPF_JMP32 | BPF_OP(OP) | BPF_X, \
420 /* Like BPF_JMP_IMM, but with 32-bit wide operands for comparison. */
422 #define BPF_JMP32_IMM(OP, DST, IMM, OFF) \
423 ((struct bpf_insn) { \
424 .code = BPF_JMP32 | BPF_OP(OP) | BPF_K, \
430 /* Unconditional jumps, goto pc + off16 */
432 #define BPF_JMP_A(OFF) \
433 ((struct bpf_insn) { \
434 .code = BPF_JMP | BPF_JA, \
440 /* Unconditional jumps, gotol pc + imm32 */
442 #define BPF_JMP32_A(IMM) \
443 ((struct bpf_insn) { \
444 .code = BPF_JMP32 | BPF_JA, \
452 #define BPF_CALL_REL(TGT) \
453 ((struct bpf_insn) { \
454 .code = BPF_JMP | BPF_CALL, \
456 .src_reg = BPF_PSEUDO_CALL, \
460 /* Convert function address to BPF immediate */
462 #define BPF_CALL_IMM(x) ((void *)(x) - (void *)__bpf_call_base)
464 #define BPF_EMIT_CALL(FUNC) \
465 ((struct bpf_insn) { \
466 .code = BPF_JMP | BPF_CALL, \
470 .imm = BPF_CALL_IMM(FUNC) })
472 /* Raw code statement block */
474 #define BPF_RAW_INSN(CODE, DST, SRC, OFF, IMM) \
475 ((struct bpf_insn) { \
484 #define BPF_EXIT_INSN() \
485 ((struct bpf_insn) { \
486 .code = BPF_JMP | BPF_EXIT, \
492 /* Speculation barrier */
494 #define BPF_ST_NOSPEC() \
495 ((struct bpf_insn) { \
496 .code = BPF_ST | BPF_NOSPEC, \
502 /* Internal classic blocks for direct assignment */
504 #define __BPF_STMT(CODE, K) \
505 ((struct sock_filter) BPF_STMT(CODE, K))
507 #define __BPF_JUMP(CODE, K, JT, JF) \
508 ((struct sock_filter) BPF_JUMP(CODE, K, JT, JF))
510 #define bytes_to_bpf_size(bytes) \
512 int bpf_size = -EINVAL; \
514 if (bytes == sizeof(u8)) \
516 else if (bytes == sizeof(u16)) \
518 else if (bytes == sizeof(u32)) \
520 else if (bytes == sizeof(u64)) \
526 #define bpf_size_to_bytes(bpf_size) \
528 int bytes = -EINVAL; \
530 if (bpf_size == BPF_B) \
531 bytes = sizeof(u8); \
532 else if (bpf_size == BPF_H) \
533 bytes = sizeof(u16); \
534 else if (bpf_size == BPF_W) \
535 bytes = sizeof(u32); \
536 else if (bpf_size == BPF_DW) \
537 bytes = sizeof(u64); \
542 #define BPF_SIZEOF(type) \
544 const int __size = bytes_to_bpf_size(sizeof(type)); \
545 BUILD_BUG_ON(__size < 0); \
549 #define BPF_FIELD_SIZEOF(type, field) \
551 const int __size = bytes_to_bpf_size(sizeof_field(type, field)); \
552 BUILD_BUG_ON(__size < 0); \
556 #define BPF_LDST_BYTES(insn) \
558 const int __size = bpf_size_to_bytes(BPF_SIZE((insn)->code)); \
559 WARN_ON(__size < 0); \
563 #define __BPF_MAP_0(m, v, ...) v
564 #define __BPF_MAP_1(m, v, t, a, ...) m(t, a)
565 #define __BPF_MAP_2(m, v, t, a, ...) m(t, a), __BPF_MAP_1(m, v, __VA_ARGS__)
566 #define __BPF_MAP_3(m, v, t, a, ...) m(t, a), __BPF_MAP_2(m, v, __VA_ARGS__)
567 #define __BPF_MAP_4(m, v, t, a, ...) m(t, a), __BPF_MAP_3(m, v, __VA_ARGS__)
568 #define __BPF_MAP_5(m, v, t, a, ...) m(t, a), __BPF_MAP_4(m, v, __VA_ARGS__)
570 #define __BPF_REG_0(...) __BPF_PAD(5)
571 #define __BPF_REG_1(...) __BPF_MAP(1, __VA_ARGS__), __BPF_PAD(4)
572 #define __BPF_REG_2(...) __BPF_MAP(2, __VA_ARGS__), __BPF_PAD(3)
573 #define __BPF_REG_3(...) __BPF_MAP(3, __VA_ARGS__), __BPF_PAD(2)
574 #define __BPF_REG_4(...) __BPF_MAP(4, __VA_ARGS__), __BPF_PAD(1)
575 #define __BPF_REG_5(...) __BPF_MAP(5, __VA_ARGS__)
577 #define __BPF_MAP(n, ...) __BPF_MAP_##n(__VA_ARGS__)
578 #define __BPF_REG(n, ...) __BPF_REG_##n(__VA_ARGS__)
580 #define __BPF_CAST(t, a) \
583 typeof(__builtin_choose_expr(sizeof(t) == sizeof(unsigned long), \
584 (unsigned long)0, (t)0))) a
588 #define __BPF_DECL_ARGS(t, a) t a
589 #define __BPF_DECL_REGS(t, a) u64 a
591 #define __BPF_PAD(n) \
592 __BPF_MAP(n, __BPF_DECL_ARGS, __BPF_N, u64, __ur_1, u64, __ur_2, \
593 u64, __ur_3, u64, __ur_4, u64, __ur_5)
595 #define BPF_CALL_x(x, attr, name, ...) \
596 static __always_inline \
597 u64 ____##name(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__)); \
598 typedef u64 (*btf_##name)(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__)); \
599 attr u64 name(__BPF_REG(x, __BPF_DECL_REGS, __BPF_N, __VA_ARGS__)); \
600 attr u64 name(__BPF_REG(x, __BPF_DECL_REGS, __BPF_N, __VA_ARGS__)) \
602 return ((btf_##name)____##name)(__BPF_MAP(x,__BPF_CAST,__BPF_N,__VA_ARGS__));\
604 static __always_inline \
605 u64 ____##name(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__))
608 #define BPF_CALL_0(name, ...) BPF_CALL_x(0, __NOATTR, name, __VA_ARGS__)
609 #define BPF_CALL_1(name, ...) BPF_CALL_x(1, __NOATTR, name, __VA_ARGS__)
610 #define BPF_CALL_2(name, ...) BPF_CALL_x(2, __NOATTR, name, __VA_ARGS__)
611 #define BPF_CALL_3(name, ...) BPF_CALL_x(3, __NOATTR, name, __VA_ARGS__)
612 #define BPF_CALL_4(name, ...) BPF_CALL_x(4, __NOATTR, name, __VA_ARGS__)
613 #define BPF_CALL_5(name, ...) BPF_CALL_x(5, __NOATTR, name, __VA_ARGS__)
615 #define NOTRACE_BPF_CALL_1(name, ...) BPF_CALL_x(1, notrace, name, __VA_ARGS__)
617 #define bpf_ctx_range(TYPE, MEMBER) \
618 offsetof(TYPE, MEMBER) ... offsetofend(TYPE, MEMBER) - 1
619 #define bpf_ctx_range_till(TYPE, MEMBER1, MEMBER2) \
620 offsetof(TYPE, MEMBER1) ... offsetofend(TYPE, MEMBER2) - 1
621 #if BITS_PER_LONG == 64
622 # define bpf_ctx_range_ptr(TYPE, MEMBER) \
623 offsetof(TYPE, MEMBER) ... offsetofend(TYPE, MEMBER) - 1
625 # define bpf_ctx_range_ptr(TYPE, MEMBER) \
626 offsetof(TYPE, MEMBER) ... offsetof(TYPE, MEMBER) + 8 - 1
627 #endif /* BITS_PER_LONG == 64 */
629 #define bpf_target_off(TYPE, MEMBER, SIZE, PTR_SIZE) \
631 BUILD_BUG_ON(sizeof_field(TYPE, MEMBER) != (SIZE)); \
632 *(PTR_SIZE) = (SIZE); \
633 offsetof(TYPE, MEMBER); \
636 /* A struct sock_filter is architecture independent. */
637 struct compat_sock_fprog
{
639 compat_uptr_t filter
; /* struct sock_filter * */
642 struct sock_fprog_kern
{
644 struct sock_filter
*filter
;
647 /* Some arches need doubleword alignment for their instructions and/or data */
648 #define BPF_IMAGE_ALIGNMENT 8
650 struct bpf_binary_header
{
652 u8 image
[] __aligned(BPF_IMAGE_ALIGNMENT
);
655 struct bpf_prog_stats
{
659 struct u64_stats_sync syncp
;
660 } __aligned(2 * sizeof(u64
));
665 struct bpf_prog
*prog
;
668 DECLARE_STATIC_KEY_FALSE(bpf_stats_enabled_key
);
670 extern struct mutex nf_conn_btf_access_lock
;
671 extern int (*nfct_btf_struct_access
)(struct bpf_verifier_log
*log
,
672 const struct bpf_reg_state
*reg
,
675 typedef unsigned int (*bpf_dispatcher_fn
)(const void *ctx
,
676 const struct bpf_insn
*insnsi
,
677 unsigned int (*bpf_func
)(const void *,
678 const struct bpf_insn
*));
680 static __always_inline u32
__bpf_prog_run(const struct bpf_prog
*prog
,
682 bpf_dispatcher_fn dfunc
)
687 if (static_branch_unlikely(&bpf_stats_enabled_key
)) {
688 struct bpf_prog_stats
*stats
;
689 u64 duration
, start
= sched_clock();
692 ret
= dfunc(ctx
, prog
->insnsi
, prog
->bpf_func
);
694 duration
= sched_clock() - start
;
695 stats
= this_cpu_ptr(prog
->stats
);
696 flags
= u64_stats_update_begin_irqsave(&stats
->syncp
);
697 u64_stats_inc(&stats
->cnt
);
698 u64_stats_add(&stats
->nsecs
, duration
);
699 u64_stats_update_end_irqrestore(&stats
->syncp
, flags
);
701 ret
= dfunc(ctx
, prog
->insnsi
, prog
->bpf_func
);
706 static __always_inline u32
bpf_prog_run(const struct bpf_prog
*prog
, const void *ctx
)
708 return __bpf_prog_run(prog
, ctx
, bpf_dispatcher_nop_func
);
712 * Use in preemptible and therefore migratable context to make sure that
713 * the execution of the BPF program runs on one CPU.
715 * This uses migrate_disable/enable() explicitly to document that the
716 * invocation of a BPF program does not require reentrancy protection
717 * against a BPF program which is invoked from a preempting task.
719 static inline u32
bpf_prog_run_pin_on_cpu(const struct bpf_prog
*prog
,
725 ret
= bpf_prog_run(prog
, ctx
);
730 #define BPF_SKB_CB_LEN QDISC_CB_PRIV_LEN
732 struct bpf_skb_data_end
{
733 struct qdisc_skb_cb qdisc_cb
;
738 struct bpf_nh_params
{
742 struct in6_addr ipv6_nh
;
746 /* flags for bpf_redirect_info kern_flags */
747 #define BPF_RI_F_RF_NO_DIRECT BIT(0) /* no napi_direct on return_frame */
748 #define BPF_RI_F_RI_INIT BIT(1)
749 #define BPF_RI_F_CPU_MAP_INIT BIT(2)
750 #define BPF_RI_F_DEV_MAP_INIT BIT(3)
751 #define BPF_RI_F_XSK_MAP_INIT BIT(4)
753 struct bpf_redirect_info
{
759 enum bpf_map_type map_type
;
760 struct bpf_nh_params nh
;
764 struct bpf_net_context
{
765 struct bpf_redirect_info ri
;
766 struct list_head cpu_map_flush_list
;
767 struct list_head dev_map_flush_list
;
768 struct list_head xskmap_map_flush_list
;
771 static inline struct bpf_net_context
*bpf_net_ctx_set(struct bpf_net_context
*bpf_net_ctx
)
773 struct task_struct
*tsk
= current
;
775 if (tsk
->bpf_net_context
!= NULL
)
777 bpf_net_ctx
->ri
.kern_flags
= 0;
779 tsk
->bpf_net_context
= bpf_net_ctx
;
783 static inline void bpf_net_ctx_clear(struct bpf_net_context
*bpf_net_ctx
)
786 current
->bpf_net_context
= NULL
;
789 static inline struct bpf_net_context
*bpf_net_ctx_get(void)
791 return current
->bpf_net_context
;
794 static inline struct bpf_redirect_info
*bpf_net_ctx_get_ri(void)
796 struct bpf_net_context
*bpf_net_ctx
= bpf_net_ctx_get();
798 if (!(bpf_net_ctx
->ri
.kern_flags
& BPF_RI_F_RI_INIT
)) {
799 memset(&bpf_net_ctx
->ri
, 0, offsetof(struct bpf_net_context
, ri
.nh
));
800 bpf_net_ctx
->ri
.kern_flags
|= BPF_RI_F_RI_INIT
;
803 return &bpf_net_ctx
->ri
;
806 static inline struct list_head
*bpf_net_ctx_get_cpu_map_flush_list(void)
808 struct bpf_net_context
*bpf_net_ctx
= bpf_net_ctx_get();
810 if (!(bpf_net_ctx
->ri
.kern_flags
& BPF_RI_F_CPU_MAP_INIT
)) {
811 INIT_LIST_HEAD(&bpf_net_ctx
->cpu_map_flush_list
);
812 bpf_net_ctx
->ri
.kern_flags
|= BPF_RI_F_CPU_MAP_INIT
;
815 return &bpf_net_ctx
->cpu_map_flush_list
;
818 static inline struct list_head
*bpf_net_ctx_get_dev_flush_list(void)
820 struct bpf_net_context
*bpf_net_ctx
= bpf_net_ctx_get();
822 if (!(bpf_net_ctx
->ri
.kern_flags
& BPF_RI_F_DEV_MAP_INIT
)) {
823 INIT_LIST_HEAD(&bpf_net_ctx
->dev_map_flush_list
);
824 bpf_net_ctx
->ri
.kern_flags
|= BPF_RI_F_DEV_MAP_INIT
;
827 return &bpf_net_ctx
->dev_map_flush_list
;
830 static inline struct list_head
*bpf_net_ctx_get_xskmap_flush_list(void)
832 struct bpf_net_context
*bpf_net_ctx
= bpf_net_ctx_get();
834 if (!(bpf_net_ctx
->ri
.kern_flags
& BPF_RI_F_XSK_MAP_INIT
)) {
835 INIT_LIST_HEAD(&bpf_net_ctx
->xskmap_map_flush_list
);
836 bpf_net_ctx
->ri
.kern_flags
|= BPF_RI_F_XSK_MAP_INIT
;
839 return &bpf_net_ctx
->xskmap_map_flush_list
;
842 static inline void bpf_net_ctx_get_all_used_flush_lists(struct list_head
**lh_map
,
843 struct list_head
**lh_dev
,
844 struct list_head
**lh_xsk
)
846 struct bpf_net_context
*bpf_net_ctx
= bpf_net_ctx_get();
847 u32 kern_flags
= bpf_net_ctx
->ri
.kern_flags
;
848 struct list_head
*lh
;
850 *lh_map
= *lh_dev
= *lh_xsk
= NULL
;
852 if (!IS_ENABLED(CONFIG_BPF_SYSCALL
))
855 lh
= &bpf_net_ctx
->dev_map_flush_list
;
856 if (kern_flags
& BPF_RI_F_DEV_MAP_INIT
&& !list_empty(lh
))
859 lh
= &bpf_net_ctx
->cpu_map_flush_list
;
860 if (kern_flags
& BPF_RI_F_CPU_MAP_INIT
&& !list_empty(lh
))
863 lh
= &bpf_net_ctx
->xskmap_map_flush_list
;
864 if (IS_ENABLED(CONFIG_XDP_SOCKETS
) &&
865 kern_flags
& BPF_RI_F_XSK_MAP_INIT
&& !list_empty(lh
))
869 /* Compute the linear packet data range [data, data_end) which
870 * will be accessed by various program types (cls_bpf, act_bpf,
871 * lwt, ...). Subsystems allowing direct data access must (!)
872 * ensure that cb[] area can be written to when BPF program is
873 * invoked (otherwise cb[] save/restore is necessary).
875 static inline void bpf_compute_data_pointers(struct sk_buff
*skb
)
877 struct bpf_skb_data_end
*cb
= (struct bpf_skb_data_end
*)skb
->cb
;
879 BUILD_BUG_ON(sizeof(*cb
) > sizeof_field(struct sk_buff
, cb
));
880 cb
->data_meta
= skb
->data
- skb_metadata_len(skb
);
881 cb
->data_end
= skb
->data
+ skb_headlen(skb
);
884 /* Similar to bpf_compute_data_pointers(), except that save orginal
885 * data in cb->data and cb->meta_data for restore.
887 static inline void bpf_compute_and_save_data_end(
888 struct sk_buff
*skb
, void **saved_data_end
)
890 struct bpf_skb_data_end
*cb
= (struct bpf_skb_data_end
*)skb
->cb
;
892 *saved_data_end
= cb
->data_end
;
893 cb
->data_end
= skb
->data
+ skb_headlen(skb
);
896 /* Restore data saved by bpf_compute_and_save_data_end(). */
897 static inline void bpf_restore_data_end(
898 struct sk_buff
*skb
, void *saved_data_end
)
900 struct bpf_skb_data_end
*cb
= (struct bpf_skb_data_end
*)skb
->cb
;
902 cb
->data_end
= saved_data_end
;
905 static inline u8
*bpf_skb_cb(const struct sk_buff
*skb
)
907 /* eBPF programs may read/write skb->cb[] area to transfer meta
908 * data between tail calls. Since this also needs to work with
909 * tc, that scratch memory is mapped to qdisc_skb_cb's data area.
911 * In some socket filter cases, the cb unfortunately needs to be
912 * saved/restored so that protocol specific skb->cb[] data won't
913 * be lost. In any case, due to unpriviledged eBPF programs
914 * attached to sockets, we need to clear the bpf_skb_cb() area
915 * to not leak previous contents to user space.
917 BUILD_BUG_ON(sizeof_field(struct __sk_buff
, cb
) != BPF_SKB_CB_LEN
);
918 BUILD_BUG_ON(sizeof_field(struct __sk_buff
, cb
) !=
919 sizeof_field(struct qdisc_skb_cb
, data
));
921 return qdisc_skb_cb(skb
)->data
;
924 /* Must be invoked with migration disabled */
925 static inline u32
__bpf_prog_run_save_cb(const struct bpf_prog
*prog
,
928 const struct sk_buff
*skb
= ctx
;
929 u8
*cb_data
= bpf_skb_cb(skb
);
930 u8 cb_saved
[BPF_SKB_CB_LEN
];
933 if (unlikely(prog
->cb_access
)) {
934 memcpy(cb_saved
, cb_data
, sizeof(cb_saved
));
935 memset(cb_data
, 0, sizeof(cb_saved
));
938 res
= bpf_prog_run(prog
, skb
);
940 if (unlikely(prog
->cb_access
))
941 memcpy(cb_data
, cb_saved
, sizeof(cb_saved
));
946 static inline u32
bpf_prog_run_save_cb(const struct bpf_prog
*prog
,
952 res
= __bpf_prog_run_save_cb(prog
, skb
);
957 static inline u32
bpf_prog_run_clear_cb(const struct bpf_prog
*prog
,
960 u8
*cb_data
= bpf_skb_cb(skb
);
963 if (unlikely(prog
->cb_access
))
964 memset(cb_data
, 0, BPF_SKB_CB_LEN
);
966 res
= bpf_prog_run_pin_on_cpu(prog
, skb
);
970 DECLARE_BPF_DISPATCHER(xdp
)
972 DECLARE_STATIC_KEY_FALSE(bpf_master_redirect_enabled_key
);
974 u32
xdp_master_redirect(struct xdp_buff
*xdp
);
976 void bpf_prog_change_xdp(struct bpf_prog
*prev_prog
, struct bpf_prog
*prog
);
978 static inline u32
bpf_prog_insn_size(const struct bpf_prog
*prog
)
980 return prog
->len
* sizeof(struct bpf_insn
);
983 static inline u32
bpf_prog_tag_scratch_size(const struct bpf_prog
*prog
)
985 return round_up(bpf_prog_insn_size(prog
) +
986 sizeof(__be64
) + 1, SHA1_BLOCK_SIZE
);
989 static inline unsigned int bpf_prog_size(unsigned int proglen
)
991 return max(sizeof(struct bpf_prog
),
992 offsetof(struct bpf_prog
, insns
[proglen
]));
995 static inline bool bpf_prog_was_classic(const struct bpf_prog
*prog
)
997 /* When classic BPF programs have been loaded and the arch
998 * does not have a classic BPF JIT (anymore), they have been
999 * converted via bpf_migrate_filter() to eBPF and thus always
1000 * have an unspec program type.
1002 return prog
->type
== BPF_PROG_TYPE_UNSPEC
;
1005 static inline u32
bpf_ctx_off_adjust_machine(u32 size
)
1007 const u32 size_machine
= sizeof(unsigned long);
1009 if (size
> size_machine
&& size
% size_machine
== 0)
1010 size
= size_machine
;
1016 bpf_ctx_narrow_access_ok(u32 off
, u32 size
, u32 size_default
)
1018 return size
<= size_default
&& (size
& (size
- 1)) == 0;
1022 bpf_ctx_narrow_access_offset(u32 off
, u32 size
, u32 size_default
)
1024 u8 access_off
= off
& (size_default
- 1);
1026 #ifdef __LITTLE_ENDIAN
1029 return size_default
- (access_off
+ size
);
1033 #define bpf_ctx_wide_access_ok(off, size, type, field) \
1034 (size == sizeof(__u64) && \
1035 off >= offsetof(type, field) && \
1036 off + sizeof(__u64) <= offsetofend(type, field) && \
1037 off % sizeof(__u64) == 0)
1039 #define bpf_classic_proglen(fprog) (fprog->len * sizeof(fprog->filter[0]))
1041 static inline int __must_check
bpf_prog_lock_ro(struct bpf_prog
*fp
)
1043 #ifndef CONFIG_BPF_JIT_ALWAYS_ON
1045 set_vm_flush_reset_perms(fp
);
1046 return set_memory_ro((unsigned long)fp
, fp
->pages
);
1052 static inline int __must_check
1053 bpf_jit_binary_lock_ro(struct bpf_binary_header
*hdr
)
1055 set_vm_flush_reset_perms(hdr
);
1056 return set_memory_rox((unsigned long)hdr
, hdr
->size
>> PAGE_SHIFT
);
1059 int sk_filter_trim_cap(struct sock
*sk
, struct sk_buff
*skb
, unsigned int cap
);
1060 static inline int sk_filter(struct sock
*sk
, struct sk_buff
*skb
)
1062 return sk_filter_trim_cap(sk
, skb
, 1);
1065 struct bpf_prog
*bpf_prog_select_runtime(struct bpf_prog
*fp
, int *err
);
1066 void bpf_prog_free(struct bpf_prog
*fp
);
1068 bool bpf_opcode_in_insntable(u8 code
);
1070 void bpf_prog_fill_jited_linfo(struct bpf_prog
*prog
,
1071 const u32
*insn_to_jit_off
);
1072 int bpf_prog_alloc_jited_linfo(struct bpf_prog
*prog
);
1073 void bpf_prog_jit_attempt_done(struct bpf_prog
*prog
);
1075 struct bpf_prog
*bpf_prog_alloc(unsigned int size
, gfp_t gfp_extra_flags
);
1076 struct bpf_prog
*bpf_prog_alloc_no_stats(unsigned int size
, gfp_t gfp_extra_flags
);
1077 struct bpf_prog
*bpf_prog_realloc(struct bpf_prog
*fp_old
, unsigned int size
,
1078 gfp_t gfp_extra_flags
);
1079 void __bpf_prog_free(struct bpf_prog
*fp
);
1081 static inline void bpf_prog_unlock_free(struct bpf_prog
*fp
)
1083 __bpf_prog_free(fp
);
1086 typedef int (*bpf_aux_classic_check_t
)(struct sock_filter
*filter
,
1089 int bpf_prog_create(struct bpf_prog
**pfp
, struct sock_fprog_kern
*fprog
);
1090 int bpf_prog_create_from_user(struct bpf_prog
**pfp
, struct sock_fprog
*fprog
,
1091 bpf_aux_classic_check_t trans
, bool save_orig
);
1092 void bpf_prog_destroy(struct bpf_prog
*fp
);
1094 int sk_attach_filter(struct sock_fprog
*fprog
, struct sock
*sk
);
1095 int sk_attach_bpf(u32 ufd
, struct sock
*sk
);
1096 int sk_reuseport_attach_filter(struct sock_fprog
*fprog
, struct sock
*sk
);
1097 int sk_reuseport_attach_bpf(u32 ufd
, struct sock
*sk
);
1098 void sk_reuseport_prog_free(struct bpf_prog
*prog
);
1099 int sk_detach_filter(struct sock
*sk
);
1100 int sk_get_filter(struct sock
*sk
, sockptr_t optval
, unsigned int len
);
1102 bool sk_filter_charge(struct sock
*sk
, struct sk_filter
*fp
);
1103 void sk_filter_uncharge(struct sock
*sk
, struct sk_filter
*fp
);
1105 u64
__bpf_call_base(u64 r1
, u64 r2
, u64 r3
, u64 r4
, u64 r5
);
1106 #define __bpf_call_base_args \
1107 ((u64 (*)(u64, u64, u64, u64, u64, const struct bpf_insn *)) \
1108 (void *)__bpf_call_base)
1110 struct bpf_prog
*bpf_int_jit_compile(struct bpf_prog
*prog
);
1111 void bpf_jit_compile(struct bpf_prog
*prog
);
1112 bool bpf_jit_needs_zext(void);
1113 bool bpf_jit_inlines_helper_call(s32 imm
);
1114 bool bpf_jit_supports_subprog_tailcalls(void);
1115 bool bpf_jit_supports_percpu_insn(void);
1116 bool bpf_jit_supports_kfunc_call(void);
1117 bool bpf_jit_supports_far_kfunc_call(void);
1118 bool bpf_jit_supports_exceptions(void);
1119 bool bpf_jit_supports_ptr_xchg(void);
1120 bool bpf_jit_supports_arena(void);
1121 bool bpf_jit_supports_insn(struct bpf_insn
*insn
, bool in_arena
);
1122 bool bpf_jit_supports_private_stack(void);
1123 u64
bpf_arch_uaddress_limit(void);
1124 void arch_bpf_stack_walk(bool (*consume_fn
)(void *cookie
, u64 ip
, u64 sp
, u64 bp
), void *cookie
);
1125 bool bpf_helper_changes_pkt_data(void *func
);
1127 static inline bool bpf_dump_raw_ok(const struct cred
*cred
)
1129 /* Reconstruction of call-sites is dependent on kallsyms,
1130 * thus make dump the same restriction.
1132 return kallsyms_show_value(cred
);
1135 struct bpf_prog
*bpf_patch_insn_single(struct bpf_prog
*prog
, u32 off
,
1136 const struct bpf_insn
*patch
, u32 len
);
1137 int bpf_remove_insns(struct bpf_prog
*prog
, u32 off
, u32 cnt
);
1139 static inline bool xdp_return_frame_no_direct(void)
1141 struct bpf_redirect_info
*ri
= bpf_net_ctx_get_ri();
1143 return ri
->kern_flags
& BPF_RI_F_RF_NO_DIRECT
;
1146 static inline void xdp_set_return_frame_no_direct(void)
1148 struct bpf_redirect_info
*ri
= bpf_net_ctx_get_ri();
1150 ri
->kern_flags
|= BPF_RI_F_RF_NO_DIRECT
;
1153 static inline void xdp_clear_return_frame_no_direct(void)
1155 struct bpf_redirect_info
*ri
= bpf_net_ctx_get_ri();
1157 ri
->kern_flags
&= ~BPF_RI_F_RF_NO_DIRECT
;
1160 static inline int xdp_ok_fwd_dev(const struct net_device
*fwd
,
1161 unsigned int pktlen
)
1165 if (unlikely(!(fwd
->flags
& IFF_UP
)))
1168 len
= fwd
->mtu
+ fwd
->hard_header_len
+ VLAN_HLEN
;
1175 /* The pair of xdp_do_redirect and xdp_do_flush MUST be called in the
1176 * same cpu context. Further for best results no more than a single map
1177 * for the do_redirect/do_flush pair should be used. This limitation is
1178 * because we only track one map and force a flush when the map changes.
1179 * This does not appear to be a real limitation for existing software.
1181 int xdp_do_generic_redirect(struct net_device
*dev
, struct sk_buff
*skb
,
1182 struct xdp_buff
*xdp
, struct bpf_prog
*prog
);
1183 int xdp_do_redirect(struct net_device
*dev
,
1184 struct xdp_buff
*xdp
,
1185 struct bpf_prog
*prog
);
1186 int xdp_do_redirect_frame(struct net_device
*dev
,
1187 struct xdp_buff
*xdp
,
1188 struct xdp_frame
*xdpf
,
1189 struct bpf_prog
*prog
);
1190 void xdp_do_flush(void);
1192 void bpf_warn_invalid_xdp_action(struct net_device
*dev
, struct bpf_prog
*prog
, u32 act
);
1195 struct sock
*bpf_run_sk_reuseport(struct sock_reuseport
*reuse
, struct sock
*sk
,
1196 struct bpf_prog
*prog
, struct sk_buff
*skb
,
1197 struct sock
*migrating_sk
,
1200 static inline struct sock
*
1201 bpf_run_sk_reuseport(struct sock_reuseport
*reuse
, struct sock
*sk
,
1202 struct bpf_prog
*prog
, struct sk_buff
*skb
,
1203 struct sock
*migrating_sk
,
1210 #ifdef CONFIG_BPF_JIT
1211 extern int bpf_jit_enable
;
1212 extern int bpf_jit_harden
;
1213 extern int bpf_jit_kallsyms
;
1214 extern long bpf_jit_limit
;
1215 extern long bpf_jit_limit_max
;
1217 typedef void (*bpf_jit_fill_hole_t
)(void *area
, unsigned int size
);
1219 void bpf_jit_fill_hole_with_zero(void *area
, unsigned int size
);
1221 struct bpf_binary_header
*
1222 bpf_jit_binary_alloc(unsigned int proglen
, u8
**image_ptr
,
1223 unsigned int alignment
,
1224 bpf_jit_fill_hole_t bpf_fill_ill_insns
);
1225 void bpf_jit_binary_free(struct bpf_binary_header
*hdr
);
1226 u64
bpf_jit_alloc_exec_limit(void);
1227 void *bpf_jit_alloc_exec(unsigned long size
);
1228 void bpf_jit_free_exec(void *addr
);
1229 void bpf_jit_free(struct bpf_prog
*fp
);
1230 struct bpf_binary_header
*
1231 bpf_jit_binary_pack_hdr(const struct bpf_prog
*fp
);
1233 void *bpf_prog_pack_alloc(u32 size
, bpf_jit_fill_hole_t bpf_fill_ill_insns
);
1234 void bpf_prog_pack_free(void *ptr
, u32 size
);
1236 static inline bool bpf_prog_kallsyms_verify_off(const struct bpf_prog
*fp
)
1238 return list_empty(&fp
->aux
->ksym
.lnode
) ||
1239 fp
->aux
->ksym
.lnode
.prev
== LIST_POISON2
;
1242 struct bpf_binary_header
*
1243 bpf_jit_binary_pack_alloc(unsigned int proglen
, u8
**ro_image
,
1244 unsigned int alignment
,
1245 struct bpf_binary_header
**rw_hdr
,
1247 bpf_jit_fill_hole_t bpf_fill_ill_insns
);
1248 int bpf_jit_binary_pack_finalize(struct bpf_binary_header
*ro_header
,
1249 struct bpf_binary_header
*rw_header
);
1250 void bpf_jit_binary_pack_free(struct bpf_binary_header
*ro_header
,
1251 struct bpf_binary_header
*rw_header
);
1253 int bpf_jit_add_poke_descriptor(struct bpf_prog
*prog
,
1254 struct bpf_jit_poke_descriptor
*poke
);
1256 int bpf_jit_get_func_addr(const struct bpf_prog
*prog
,
1257 const struct bpf_insn
*insn
, bool extra_pass
,
1258 u64
*func_addr
, bool *func_addr_fixed
);
1260 struct bpf_prog
*bpf_jit_blind_constants(struct bpf_prog
*fp
);
1261 void bpf_jit_prog_release_other(struct bpf_prog
*fp
, struct bpf_prog
*fp_other
);
1263 static inline void bpf_jit_dump(unsigned int flen
, unsigned int proglen
,
1264 u32 pass
, void *image
)
1266 pr_err("flen=%u proglen=%u pass=%u image=%pK from=%s pid=%d\n", flen
,
1267 proglen
, pass
, image
, current
->comm
, task_pid_nr(current
));
1270 print_hex_dump(KERN_ERR
, "JIT code: ", DUMP_PREFIX_OFFSET
,
1271 16, 1, image
, proglen
, false);
1274 static inline bool bpf_jit_is_ebpf(void)
1276 # ifdef CONFIG_HAVE_EBPF_JIT
1283 static inline bool ebpf_jit_enabled(void)
1285 return bpf_jit_enable
&& bpf_jit_is_ebpf();
1288 static inline bool bpf_prog_ebpf_jited(const struct bpf_prog
*fp
)
1290 return fp
->jited
&& bpf_jit_is_ebpf();
1293 static inline bool bpf_jit_blinding_enabled(struct bpf_prog
*prog
)
1295 /* These are the prerequisites, should someone ever have the
1296 * idea to call blinding outside of them, we make sure to
1299 if (!bpf_jit_is_ebpf())
1301 if (!prog
->jit_requested
)
1303 if (!bpf_jit_harden
)
1305 if (bpf_jit_harden
== 1 && bpf_token_capable(prog
->aux
->token
, CAP_BPF
))
1311 static inline bool bpf_jit_kallsyms_enabled(void)
1313 /* There are a couple of corner cases where kallsyms should
1314 * not be enabled f.e. on hardening.
1318 if (!bpf_jit_kallsyms
)
1320 if (bpf_jit_kallsyms
== 1)
1326 int __bpf_address_lookup(unsigned long addr
, unsigned long *size
,
1327 unsigned long *off
, char *sym
);
1328 bool is_bpf_text_address(unsigned long addr
);
1329 int bpf_get_kallsym(unsigned int symnum
, unsigned long *value
, char *type
,
1331 struct bpf_prog
*bpf_prog_ksym_find(unsigned long addr
);
1334 bpf_address_lookup(unsigned long addr
, unsigned long *size
,
1335 unsigned long *off
, char **modname
, char *sym
)
1337 int ret
= __bpf_address_lookup(addr
, size
, off
, sym
);
1344 void bpf_prog_kallsyms_add(struct bpf_prog
*fp
);
1345 void bpf_prog_kallsyms_del(struct bpf_prog
*fp
);
1347 #else /* CONFIG_BPF_JIT */
1349 static inline bool ebpf_jit_enabled(void)
1354 static inline bool bpf_jit_blinding_enabled(struct bpf_prog
*prog
)
1359 static inline bool bpf_prog_ebpf_jited(const struct bpf_prog
*fp
)
1365 bpf_jit_add_poke_descriptor(struct bpf_prog
*prog
,
1366 struct bpf_jit_poke_descriptor
*poke
)
1371 static inline void bpf_jit_free(struct bpf_prog
*fp
)
1373 bpf_prog_unlock_free(fp
);
1376 static inline bool bpf_jit_kallsyms_enabled(void)
1382 __bpf_address_lookup(unsigned long addr
, unsigned long *size
,
1383 unsigned long *off
, char *sym
)
1388 static inline bool is_bpf_text_address(unsigned long addr
)
1393 static inline int bpf_get_kallsym(unsigned int symnum
, unsigned long *value
,
1394 char *type
, char *sym
)
1399 static inline struct bpf_prog
*bpf_prog_ksym_find(unsigned long addr
)
1405 bpf_address_lookup(unsigned long addr
, unsigned long *size
,
1406 unsigned long *off
, char **modname
, char *sym
)
1411 static inline void bpf_prog_kallsyms_add(struct bpf_prog
*fp
)
1415 static inline void bpf_prog_kallsyms_del(struct bpf_prog
*fp
)
1419 #endif /* CONFIG_BPF_JIT */
1421 void bpf_prog_kallsyms_del_all(struct bpf_prog
*fp
);
1423 #define BPF_ANC BIT(15)
1425 static inline bool bpf_needs_clear_a(const struct sock_filter
*first
)
1427 switch (first
->code
) {
1428 case BPF_RET
| BPF_K
:
1429 case BPF_LD
| BPF_W
| BPF_LEN
:
1432 case BPF_LD
| BPF_W
| BPF_ABS
:
1433 case BPF_LD
| BPF_H
| BPF_ABS
:
1434 case BPF_LD
| BPF_B
| BPF_ABS
:
1435 if (first
->k
== SKF_AD_OFF
+ SKF_AD_ALU_XOR_X
)
1444 static inline u16
bpf_anc_helper(const struct sock_filter
*ftest
)
1446 BUG_ON(ftest
->code
& BPF_ANC
);
1448 switch (ftest
->code
) {
1449 case BPF_LD
| BPF_W
| BPF_ABS
:
1450 case BPF_LD
| BPF_H
| BPF_ABS
:
1451 case BPF_LD
| BPF_B
| BPF_ABS
:
1452 #define BPF_ANCILLARY(CODE) case SKF_AD_OFF + SKF_AD_##CODE: \
1453 return BPF_ANC | SKF_AD_##CODE
1455 BPF_ANCILLARY(PROTOCOL
);
1456 BPF_ANCILLARY(PKTTYPE
);
1457 BPF_ANCILLARY(IFINDEX
);
1458 BPF_ANCILLARY(NLATTR
);
1459 BPF_ANCILLARY(NLATTR_NEST
);
1460 BPF_ANCILLARY(MARK
);
1461 BPF_ANCILLARY(QUEUE
);
1462 BPF_ANCILLARY(HATYPE
);
1463 BPF_ANCILLARY(RXHASH
);
1465 BPF_ANCILLARY(ALU_XOR_X
);
1466 BPF_ANCILLARY(VLAN_TAG
);
1467 BPF_ANCILLARY(VLAN_TAG_PRESENT
);
1468 BPF_ANCILLARY(PAY_OFFSET
);
1469 BPF_ANCILLARY(RANDOM
);
1470 BPF_ANCILLARY(VLAN_TPID
);
1478 void *bpf_internal_load_pointer_neg_helper(const struct sk_buff
*skb
,
1479 int k
, unsigned int size
);
1481 static inline int bpf_tell_extensions(void)
1486 struct bpf_sock_addr_kern
{
1488 struct sockaddr
*uaddr
;
1489 /* Temporary "register" to make indirect stores to nested structures
1490 * defined above. We need three registers to make such a store, but
1491 * only two (src and dst) are available at convert_ctx_access time
1494 void *t_ctx
; /* Attach type specific context. */
1498 struct bpf_sock_ops_kern
{
1505 struct sk_buff
*syn_skb
;
1506 struct sk_buff
*skb
;
1510 u8 remaining_opt_len
;
1511 u64 temp
; /* temp and everything after is not
1512 * initialized to 0 before calling
1513 * the BPF program. New fields that
1514 * should be initialized to 0 should
1515 * be inserted before temp.
1516 * temp is scratch storage used by
1517 * sock_ops_convert_ctx_access
1518 * as temporary storage of a register.
1522 struct bpf_sysctl_kern
{
1523 struct ctl_table_header
*head
;
1524 const struct ctl_table
*table
;
1532 /* Temporary "register" for indirect stores to ppos. */
1536 #define BPF_SOCKOPT_KERN_BUF_SIZE 32
1537 struct bpf_sockopt_buf
{
1538 u8 data
[BPF_SOCKOPT_KERN_BUF_SIZE
];
1541 struct bpf_sockopt_kern
{
1548 /* for retval in struct bpf_cg_run_ctx */
1549 struct task_struct
*current_task
;
1550 /* Temporary "register" for indirect stores to ppos. */
1554 int copy_bpf_fprog_from_user(struct sock_fprog
*dst
, sockptr_t src
, int len
);
1556 struct bpf_sk_lookup_kern
{
1566 const struct in6_addr
*saddr
;
1567 const struct in6_addr
*daddr
;
1569 struct sock
*selected_sk
;
1570 u32 ingress_ifindex
;
1574 extern struct static_key_false bpf_sk_lookup_enabled
;
1576 /* Runners for BPF_SK_LOOKUP programs to invoke on socket lookup.
1578 * Allowed return values for a BPF SK_LOOKUP program are SK_PASS and
1579 * SK_DROP. Their meaning is as follows:
1581 * SK_PASS && ctx.selected_sk != NULL: use selected_sk as lookup result
1582 * SK_PASS && ctx.selected_sk == NULL: continue to htable-based socket lookup
1583 * SK_DROP : terminate lookup with -ECONNREFUSED
1585 * This macro aggregates return values and selected sockets from
1586 * multiple BPF programs according to following rules in order:
1588 * 1. If any program returned SK_PASS and a non-NULL ctx.selected_sk,
1589 * macro result is SK_PASS and last ctx.selected_sk is used.
1590 * 2. If any program returned SK_DROP return value,
1591 * macro result is SK_DROP.
1592 * 3. Otherwise result is SK_PASS and ctx.selected_sk is NULL.
1594 * Caller must ensure that the prog array is non-NULL, and that the
1595 * array as well as the programs it contains remain valid.
1597 #define BPF_PROG_SK_LOOKUP_RUN_ARRAY(array, ctx, func) \
1599 struct bpf_sk_lookup_kern *_ctx = &(ctx); \
1600 struct bpf_prog_array_item *_item; \
1601 struct sock *_selected_sk = NULL; \
1602 bool _no_reuseport = false; \
1603 struct bpf_prog *_prog; \
1604 bool _all_pass = true; \
1607 migrate_disable(); \
1608 _item = &(array)->items[0]; \
1609 while ((_prog = READ_ONCE(_item->prog))) { \
1610 /* restore most recent selection */ \
1611 _ctx->selected_sk = _selected_sk; \
1612 _ctx->no_reuseport = _no_reuseport; \
1614 _ret = func(_prog, _ctx); \
1615 if (_ret == SK_PASS && _ctx->selected_sk) { \
1616 /* remember last non-NULL socket */ \
1617 _selected_sk = _ctx->selected_sk; \
1618 _no_reuseport = _ctx->no_reuseport; \
1619 } else if (_ret == SK_DROP && _all_pass) { \
1620 _all_pass = false; \
1624 _ctx->selected_sk = _selected_sk; \
1625 _ctx->no_reuseport = _no_reuseport; \
1627 _all_pass || _selected_sk ? SK_PASS : SK_DROP; \
1630 static inline bool bpf_sk_lookup_run_v4(const struct net
*net
, int protocol
,
1631 const __be32 saddr
, const __be16 sport
,
1632 const __be32 daddr
, const u16 dport
,
1633 const int ifindex
, struct sock
**psk
)
1635 struct bpf_prog_array
*run_array
;
1636 struct sock
*selected_sk
= NULL
;
1637 bool no_reuseport
= false;
1640 run_array
= rcu_dereference(net
->bpf
.run_array
[NETNS_BPF_SK_LOOKUP
]);
1642 struct bpf_sk_lookup_kern ctx
= {
1644 .protocol
= protocol
,
1649 .ingress_ifindex
= ifindex
,
1653 act
= BPF_PROG_SK_LOOKUP_RUN_ARRAY(run_array
, ctx
, bpf_prog_run
);
1654 if (act
== SK_PASS
) {
1655 selected_sk
= ctx
.selected_sk
;
1656 no_reuseport
= ctx
.no_reuseport
;
1658 selected_sk
= ERR_PTR(-ECONNREFUSED
);
1663 return no_reuseport
;
1666 #if IS_ENABLED(CONFIG_IPV6)
1667 static inline bool bpf_sk_lookup_run_v6(const struct net
*net
, int protocol
,
1668 const struct in6_addr
*saddr
,
1670 const struct in6_addr
*daddr
,
1672 const int ifindex
, struct sock
**psk
)
1674 struct bpf_prog_array
*run_array
;
1675 struct sock
*selected_sk
= NULL
;
1676 bool no_reuseport
= false;
1679 run_array
= rcu_dereference(net
->bpf
.run_array
[NETNS_BPF_SK_LOOKUP
]);
1681 struct bpf_sk_lookup_kern ctx
= {
1683 .protocol
= protocol
,
1688 .ingress_ifindex
= ifindex
,
1692 act
= BPF_PROG_SK_LOOKUP_RUN_ARRAY(run_array
, ctx
, bpf_prog_run
);
1693 if (act
== SK_PASS
) {
1694 selected_sk
= ctx
.selected_sk
;
1695 no_reuseport
= ctx
.no_reuseport
;
1697 selected_sk
= ERR_PTR(-ECONNREFUSED
);
1702 return no_reuseport
;
1704 #endif /* IS_ENABLED(CONFIG_IPV6) */
1706 static __always_inline
long __bpf_xdp_redirect_map(struct bpf_map
*map
, u64 index
,
1707 u64 flags
, const u64 flag_mask
,
1708 void *lookup_elem(struct bpf_map
*map
, u32 key
))
1710 struct bpf_redirect_info
*ri
= bpf_net_ctx_get_ri();
1711 const u64 action_mask
= XDP_ABORTED
| XDP_DROP
| XDP_PASS
| XDP_TX
;
1713 /* Lower bits of the flags are used as return code on lookup failure */
1714 if (unlikely(flags
& ~(action_mask
| flag_mask
)))
1717 ri
->tgt_value
= lookup_elem(map
, index
);
1718 if (unlikely(!ri
->tgt_value
) && !(flags
& BPF_F_BROADCAST
)) {
1719 /* If the lookup fails we want to clear out the state in the
1720 * redirect_info struct completely, so that if an eBPF program
1721 * performs multiple lookups, the last one always takes
1724 ri
->map_id
= INT_MAX
; /* Valid map id idr range: [1,INT_MAX[ */
1725 ri
->map_type
= BPF_MAP_TYPE_UNSPEC
;
1726 return flags
& action_mask
;
1729 ri
->tgt_index
= index
;
1730 ri
->map_id
= map
->id
;
1731 ri
->map_type
= map
->map_type
;
1733 if (flags
& BPF_F_BROADCAST
) {
1734 WRITE_ONCE(ri
->map
, map
);
1737 WRITE_ONCE(ri
->map
, NULL
);
1741 return XDP_REDIRECT
;
1745 int __bpf_skb_load_bytes(const struct sk_buff
*skb
, u32 offset
, void *to
, u32 len
);
1746 int __bpf_skb_store_bytes(struct sk_buff
*skb
, u32 offset
, const void *from
,
1747 u32 len
, u64 flags
);
1748 int __bpf_xdp_load_bytes(struct xdp_buff
*xdp
, u32 offset
, void *buf
, u32 len
);
1749 int __bpf_xdp_store_bytes(struct xdp_buff
*xdp
, u32 offset
, void *buf
, u32 len
);
1750 void *bpf_xdp_pointer(struct xdp_buff
*xdp
, u32 offset
, u32 len
);
1751 void bpf_xdp_copy_buf(struct xdp_buff
*xdp
, unsigned long off
,
1752 void *buf
, unsigned long len
, bool flush
);
1753 #else /* CONFIG_NET */
1754 static inline int __bpf_skb_load_bytes(const struct sk_buff
*skb
, u32 offset
,
1760 static inline int __bpf_skb_store_bytes(struct sk_buff
*skb
, u32 offset
,
1761 const void *from
, u32 len
, u64 flags
)
1766 static inline int __bpf_xdp_load_bytes(struct xdp_buff
*xdp
, u32 offset
,
1772 static inline int __bpf_xdp_store_bytes(struct xdp_buff
*xdp
, u32 offset
,
1778 static inline void *bpf_xdp_pointer(struct xdp_buff
*xdp
, u32 offset
, u32 len
)
1783 static inline void bpf_xdp_copy_buf(struct xdp_buff
*xdp
, unsigned long off
, void *buf
,
1784 unsigned long len
, bool flush
)
1787 #endif /* CONFIG_NET */
1789 #endif /* __LINUX_FILTER_H__ */