1 /* SPDX-License-Identifier: GPL-2.0 */
3 * Linux Socket Filter Data Structures
5 #ifndef __LINUX_FILTER_H__
6 #define __LINUX_FILTER_H__
10 #include <linux/atomic.h>
11 #include <linux/refcount.h>
12 #include <linux/compat.h>
13 #include <linux/skbuff.h>
14 #include <linux/linkage.h>
15 #include <linux/printk.h>
16 #include <linux/workqueue.h>
17 #include <linux/sched.h>
18 #include <linux/capability.h>
19 #include <linux/cryptohash.h>
20 #include <linux/set_memory.h>
21 #include <linux/kallsyms.h>
22 #include <linux/if_vlan.h>
23 #include <linux/vmalloc.h>
25 #include <net/sch_generic.h>
27 #include <asm/byteorder.h>
28 #include <uapi/linux/filter.h>
29 #include <uapi/linux/bpf.h>
37 struct sock_reuseport
;
39 struct ctl_table_header
;
41 /* ArgX, context and stack frame pointer register positions. Note,
42 * Arg1, Arg2, Arg3, etc are used as argument mappings of function
43 * calls in BPF_CALL instruction.
45 #define BPF_REG_ARG1 BPF_REG_1
46 #define BPF_REG_ARG2 BPF_REG_2
47 #define BPF_REG_ARG3 BPF_REG_3
48 #define BPF_REG_ARG4 BPF_REG_4
49 #define BPF_REG_ARG5 BPF_REG_5
50 #define BPF_REG_CTX BPF_REG_6
51 #define BPF_REG_FP BPF_REG_10
53 /* Additional register mappings for converted user programs. */
54 #define BPF_REG_A BPF_REG_0
55 #define BPF_REG_X BPF_REG_7
56 #define BPF_REG_TMP BPF_REG_2 /* scratch reg */
57 #define BPF_REG_D BPF_REG_8 /* data, callee-saved */
58 #define BPF_REG_H BPF_REG_9 /* hlen, callee-saved */
60 /* Kernel hidden auxiliary/helper register. */
61 #define BPF_REG_AX MAX_BPF_REG
62 #define MAX_BPF_EXT_REG (MAX_BPF_REG + 1)
63 #define MAX_BPF_JIT_REG MAX_BPF_EXT_REG
65 /* unused opcode to mark special call to bpf_tail_call() helper */
66 #define BPF_TAIL_CALL 0xf0
68 /* unused opcode to mark special load instruction. Same as BPF_ABS */
69 #define BPF_PROBE_MEM 0x20
71 /* unused opcode to mark call to interpreter with arguments */
72 #define BPF_CALL_ARGS 0xe0
74 /* As per nm, we expose JITed images as text (code) section for
75 * kallsyms. That way, tools like perf can find it to match
78 #define BPF_SYM_ELF_TYPE 't'
80 /* BPF program can access up to 512 bytes of stack space. */
81 #define MAX_BPF_STACK 512
83 /* Helper macros for filter block array initializers. */
85 /* ALU ops on registers, bpf_add|sub|...: dst_reg += src_reg */
87 #define BPF_ALU64_REG(OP, DST, SRC) \
88 ((struct bpf_insn) { \
89 .code = BPF_ALU64 | BPF_OP(OP) | BPF_X, \
95 #define BPF_ALU32_REG(OP, DST, SRC) \
96 ((struct bpf_insn) { \
97 .code = BPF_ALU | BPF_OP(OP) | BPF_X, \
103 /* ALU ops on immediates, bpf_add|sub|...: dst_reg += imm32 */
105 #define BPF_ALU64_IMM(OP, DST, IMM) \
106 ((struct bpf_insn) { \
107 .code = BPF_ALU64 | BPF_OP(OP) | BPF_K, \
113 #define BPF_ALU32_IMM(OP, DST, IMM) \
114 ((struct bpf_insn) { \
115 .code = BPF_ALU | BPF_OP(OP) | BPF_K, \
121 /* Endianess conversion, cpu_to_{l,b}e(), {l,b}e_to_cpu() */
123 #define BPF_ENDIAN(TYPE, DST, LEN) \
124 ((struct bpf_insn) { \
125 .code = BPF_ALU | BPF_END | BPF_SRC(TYPE), \
131 /* Short form of mov, dst_reg = src_reg */
133 #define BPF_MOV64_REG(DST, SRC) \
134 ((struct bpf_insn) { \
135 .code = BPF_ALU64 | BPF_MOV | BPF_X, \
141 #define BPF_MOV32_REG(DST, SRC) \
142 ((struct bpf_insn) { \
143 .code = BPF_ALU | BPF_MOV | BPF_X, \
149 /* Short form of mov, dst_reg = imm32 */
151 #define BPF_MOV64_IMM(DST, IMM) \
152 ((struct bpf_insn) { \
153 .code = BPF_ALU64 | BPF_MOV | BPF_K, \
159 #define BPF_MOV32_IMM(DST, IMM) \
160 ((struct bpf_insn) { \
161 .code = BPF_ALU | BPF_MOV | BPF_K, \
167 /* Special form of mov32, used for doing explicit zero extension on dst. */
168 #define BPF_ZEXT_REG(DST) \
169 ((struct bpf_insn) { \
170 .code = BPF_ALU | BPF_MOV | BPF_X, \
176 static inline bool insn_is_zext(const struct bpf_insn
*insn
)
178 return insn
->code
== (BPF_ALU
| BPF_MOV
| BPF_X
) && insn
->imm
== 1;
181 /* BPF_LD_IMM64 macro encodes single 'load 64-bit immediate' insn */
182 #define BPF_LD_IMM64(DST, IMM) \
183 BPF_LD_IMM64_RAW(DST, 0, IMM)
185 #define BPF_LD_IMM64_RAW(DST, SRC, IMM) \
186 ((struct bpf_insn) { \
187 .code = BPF_LD | BPF_DW | BPF_IMM, \
191 .imm = (__u32) (IMM) }), \
192 ((struct bpf_insn) { \
193 .code = 0, /* zero is reserved opcode */ \
197 .imm = ((__u64) (IMM)) >> 32 })
199 /* pseudo BPF_LD_IMM64 insn used to refer to process-local map_fd */
200 #define BPF_LD_MAP_FD(DST, MAP_FD) \
201 BPF_LD_IMM64_RAW(DST, BPF_PSEUDO_MAP_FD, MAP_FD)
203 /* Short form of mov based on type, BPF_X: dst_reg = src_reg, BPF_K: dst_reg = imm32 */
205 #define BPF_MOV64_RAW(TYPE, DST, SRC, IMM) \
206 ((struct bpf_insn) { \
207 .code = BPF_ALU64 | BPF_MOV | BPF_SRC(TYPE), \
213 #define BPF_MOV32_RAW(TYPE, DST, SRC, IMM) \
214 ((struct bpf_insn) { \
215 .code = BPF_ALU | BPF_MOV | BPF_SRC(TYPE), \
221 /* Direct packet access, R0 = *(uint *) (skb->data + imm32) */
223 #define BPF_LD_ABS(SIZE, IMM) \
224 ((struct bpf_insn) { \
225 .code = BPF_LD | BPF_SIZE(SIZE) | BPF_ABS, \
231 /* Indirect packet access, R0 = *(uint *) (skb->data + src_reg + imm32) */
233 #define BPF_LD_IND(SIZE, SRC, IMM) \
234 ((struct bpf_insn) { \
235 .code = BPF_LD | BPF_SIZE(SIZE) | BPF_IND, \
241 /* Memory load, dst_reg = *(uint *) (src_reg + off16) */
243 #define BPF_LDX_MEM(SIZE, DST, SRC, OFF) \
244 ((struct bpf_insn) { \
245 .code = BPF_LDX | BPF_SIZE(SIZE) | BPF_MEM, \
251 /* Memory store, *(uint *) (dst_reg + off16) = src_reg */
253 #define BPF_STX_MEM(SIZE, DST, SRC, OFF) \
254 ((struct bpf_insn) { \
255 .code = BPF_STX | BPF_SIZE(SIZE) | BPF_MEM, \
261 /* Atomic memory add, *(uint *)(dst_reg + off16) += src_reg */
263 #define BPF_STX_XADD(SIZE, DST, SRC, OFF) \
264 ((struct bpf_insn) { \
265 .code = BPF_STX | BPF_SIZE(SIZE) | BPF_XADD, \
271 /* Memory store, *(uint *) (dst_reg + off16) = imm32 */
273 #define BPF_ST_MEM(SIZE, DST, OFF, IMM) \
274 ((struct bpf_insn) { \
275 .code = BPF_ST | BPF_SIZE(SIZE) | BPF_MEM, \
281 /* Conditional jumps against registers, if (dst_reg 'op' src_reg) goto pc + off16 */
283 #define BPF_JMP_REG(OP, DST, SRC, OFF) \
284 ((struct bpf_insn) { \
285 .code = BPF_JMP | BPF_OP(OP) | BPF_X, \
291 /* Conditional jumps against immediates, if (dst_reg 'op' imm32) goto pc + off16 */
293 #define BPF_JMP_IMM(OP, DST, IMM, OFF) \
294 ((struct bpf_insn) { \
295 .code = BPF_JMP | BPF_OP(OP) | BPF_K, \
301 /* Like BPF_JMP_REG, but with 32-bit wide operands for comparison. */
303 #define BPF_JMP32_REG(OP, DST, SRC, OFF) \
304 ((struct bpf_insn) { \
305 .code = BPF_JMP32 | BPF_OP(OP) | BPF_X, \
311 /* Like BPF_JMP_IMM, but with 32-bit wide operands for comparison. */
313 #define BPF_JMP32_IMM(OP, DST, IMM, OFF) \
314 ((struct bpf_insn) { \
315 .code = BPF_JMP32 | BPF_OP(OP) | BPF_K, \
321 /* Unconditional jumps, goto pc + off16 */
323 #define BPF_JMP_A(OFF) \
324 ((struct bpf_insn) { \
325 .code = BPF_JMP | BPF_JA, \
333 #define BPF_CALL_REL(TGT) \
334 ((struct bpf_insn) { \
335 .code = BPF_JMP | BPF_CALL, \
337 .src_reg = BPF_PSEUDO_CALL, \
343 #define BPF_CAST_CALL(x) \
344 ((u64 (*)(u64, u64, u64, u64, u64))(x))
346 #define BPF_EMIT_CALL(FUNC) \
347 ((struct bpf_insn) { \
348 .code = BPF_JMP | BPF_CALL, \
352 .imm = ((FUNC) - __bpf_call_base) })
354 /* Raw code statement block */
356 #define BPF_RAW_INSN(CODE, DST, SRC, OFF, IMM) \
357 ((struct bpf_insn) { \
366 #define BPF_EXIT_INSN() \
367 ((struct bpf_insn) { \
368 .code = BPF_JMP | BPF_EXIT, \
374 /* Internal classic blocks for direct assignment */
376 #define __BPF_STMT(CODE, K) \
377 ((struct sock_filter) BPF_STMT(CODE, K))
379 #define __BPF_JUMP(CODE, K, JT, JF) \
380 ((struct sock_filter) BPF_JUMP(CODE, K, JT, JF))
382 #define bytes_to_bpf_size(bytes) \
384 int bpf_size = -EINVAL; \
386 if (bytes == sizeof(u8)) \
388 else if (bytes == sizeof(u16)) \
390 else if (bytes == sizeof(u32)) \
392 else if (bytes == sizeof(u64)) \
398 #define bpf_size_to_bytes(bpf_size) \
400 int bytes = -EINVAL; \
402 if (bpf_size == BPF_B) \
403 bytes = sizeof(u8); \
404 else if (bpf_size == BPF_H) \
405 bytes = sizeof(u16); \
406 else if (bpf_size == BPF_W) \
407 bytes = sizeof(u32); \
408 else if (bpf_size == BPF_DW) \
409 bytes = sizeof(u64); \
414 #define BPF_SIZEOF(type) \
416 const int __size = bytes_to_bpf_size(sizeof(type)); \
417 BUILD_BUG_ON(__size < 0); \
421 #define BPF_FIELD_SIZEOF(type, field) \
423 const int __size = bytes_to_bpf_size(sizeof_field(type, field)); \
424 BUILD_BUG_ON(__size < 0); \
428 #define BPF_LDST_BYTES(insn) \
430 const int __size = bpf_size_to_bytes(BPF_SIZE((insn)->code)); \
431 WARN_ON(__size < 0); \
435 #define __BPF_MAP_0(m, v, ...) v
436 #define __BPF_MAP_1(m, v, t, a, ...) m(t, a)
437 #define __BPF_MAP_2(m, v, t, a, ...) m(t, a), __BPF_MAP_1(m, v, __VA_ARGS__)
438 #define __BPF_MAP_3(m, v, t, a, ...) m(t, a), __BPF_MAP_2(m, v, __VA_ARGS__)
439 #define __BPF_MAP_4(m, v, t, a, ...) m(t, a), __BPF_MAP_3(m, v, __VA_ARGS__)
440 #define __BPF_MAP_5(m, v, t, a, ...) m(t, a), __BPF_MAP_4(m, v, __VA_ARGS__)
442 #define __BPF_REG_0(...) __BPF_PAD(5)
443 #define __BPF_REG_1(...) __BPF_MAP(1, __VA_ARGS__), __BPF_PAD(4)
444 #define __BPF_REG_2(...) __BPF_MAP(2, __VA_ARGS__), __BPF_PAD(3)
445 #define __BPF_REG_3(...) __BPF_MAP(3, __VA_ARGS__), __BPF_PAD(2)
446 #define __BPF_REG_4(...) __BPF_MAP(4, __VA_ARGS__), __BPF_PAD(1)
447 #define __BPF_REG_5(...) __BPF_MAP(5, __VA_ARGS__)
449 #define __BPF_MAP(n, ...) __BPF_MAP_##n(__VA_ARGS__)
450 #define __BPF_REG(n, ...) __BPF_REG_##n(__VA_ARGS__)
452 #define __BPF_CAST(t, a) \
455 typeof(__builtin_choose_expr(sizeof(t) == sizeof(unsigned long), \
456 (unsigned long)0, (t)0))) a
460 #define __BPF_DECL_ARGS(t, a) t a
461 #define __BPF_DECL_REGS(t, a) u64 a
463 #define __BPF_PAD(n) \
464 __BPF_MAP(n, __BPF_DECL_ARGS, __BPF_N, u64, __ur_1, u64, __ur_2, \
465 u64, __ur_3, u64, __ur_4, u64, __ur_5)
467 #define BPF_CALL_x(x, name, ...) \
468 static __always_inline \
469 u64 ____##name(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__)); \
470 typedef u64 (*btf_##name)(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__)); \
471 u64 name(__BPF_REG(x, __BPF_DECL_REGS, __BPF_N, __VA_ARGS__)); \
472 u64 name(__BPF_REG(x, __BPF_DECL_REGS, __BPF_N, __VA_ARGS__)) \
474 return ((btf_##name)____##name)(__BPF_MAP(x,__BPF_CAST,__BPF_N,__VA_ARGS__));\
476 static __always_inline \
477 u64 ____##name(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__))
479 #define BPF_CALL_0(name, ...) BPF_CALL_x(0, name, __VA_ARGS__)
480 #define BPF_CALL_1(name, ...) BPF_CALL_x(1, name, __VA_ARGS__)
481 #define BPF_CALL_2(name, ...) BPF_CALL_x(2, name, __VA_ARGS__)
482 #define BPF_CALL_3(name, ...) BPF_CALL_x(3, name, __VA_ARGS__)
483 #define BPF_CALL_4(name, ...) BPF_CALL_x(4, name, __VA_ARGS__)
484 #define BPF_CALL_5(name, ...) BPF_CALL_x(5, name, __VA_ARGS__)
486 #define bpf_ctx_range(TYPE, MEMBER) \
487 offsetof(TYPE, MEMBER) ... offsetofend(TYPE, MEMBER) - 1
488 #define bpf_ctx_range_till(TYPE, MEMBER1, MEMBER2) \
489 offsetof(TYPE, MEMBER1) ... offsetofend(TYPE, MEMBER2) - 1
490 #if BITS_PER_LONG == 64
491 # define bpf_ctx_range_ptr(TYPE, MEMBER) \
492 offsetof(TYPE, MEMBER) ... offsetofend(TYPE, MEMBER) - 1
494 # define bpf_ctx_range_ptr(TYPE, MEMBER) \
495 offsetof(TYPE, MEMBER) ... offsetof(TYPE, MEMBER) + 8 - 1
496 #endif /* BITS_PER_LONG == 64 */
498 #define bpf_target_off(TYPE, MEMBER, SIZE, PTR_SIZE) \
500 BUILD_BUG_ON(sizeof_field(TYPE, MEMBER) != (SIZE)); \
501 *(PTR_SIZE) = (SIZE); \
502 offsetof(TYPE, MEMBER); \
506 /* A struct sock_filter is architecture independent. */
507 struct compat_sock_fprog
{
509 compat_uptr_t filter
; /* struct sock_filter * */
513 struct sock_fprog_kern
{
515 struct sock_filter
*filter
;
518 /* Some arches need doubleword alignment for their instructions and/or data */
519 #define BPF_IMAGE_ALIGNMENT 8
521 struct bpf_binary_header
{
523 u8 image
[] __aligned(BPF_IMAGE_ALIGNMENT
);
527 u16 pages
; /* Number of allocated pages */
528 u16 jited
:1, /* Is our filter JIT'ed? */
529 jit_requested
:1,/* archs need to JIT the prog */
530 gpl_compatible
:1, /* Is filter GPL compatible? */
531 cb_access
:1, /* Is control block accessed? */
532 dst_needed
:1, /* Do we need dst entry? */
533 blinded
:1, /* Was blinded */
534 is_func
:1, /* program is a bpf function */
535 kprobe_override
:1, /* Do we override a kprobe? */
536 has_callchain_buf
:1, /* callchain buffer allocated? */
537 enforce_expected_attach_type
:1; /* Enforce expected_attach_type checking at attach time */
538 enum bpf_prog_type type
; /* Type of BPF program */
539 enum bpf_attach_type expected_attach_type
; /* For some prog types */
540 u32 len
; /* Number of filter blocks */
541 u32 jited_len
; /* Size of jited insns in bytes */
542 u8 tag
[BPF_TAG_SIZE
];
543 struct bpf_prog_aux
*aux
; /* Auxiliary fields */
544 struct sock_fprog_kern
*orig_prog
; /* Original BPF program */
545 unsigned int (*bpf_func
)(const void *ctx
,
546 const struct bpf_insn
*insn
);
547 /* Instructions for interpreter */
549 struct sock_filter insns
[0];
550 struct bpf_insn insnsi
[0];
557 struct bpf_prog
*prog
;
560 DECLARE_STATIC_KEY_FALSE(bpf_stats_enabled_key
);
562 #define BPF_PROG_RUN(prog, ctx) ({ \
565 if (static_branch_unlikely(&bpf_stats_enabled_key)) { \
566 struct bpf_prog_stats *stats; \
567 u64 start = sched_clock(); \
568 ret = (*(prog)->bpf_func)(ctx, (prog)->insnsi); \
569 stats = this_cpu_ptr(prog->aux->stats); \
570 u64_stats_update_begin(&stats->syncp); \
572 stats->nsecs += sched_clock() - start; \
573 u64_stats_update_end(&stats->syncp); \
575 ret = (*(prog)->bpf_func)(ctx, (prog)->insnsi); \
579 #define BPF_SKB_CB_LEN QDISC_CB_PRIV_LEN
581 struct bpf_skb_data_end
{
582 struct qdisc_skb_cb qdisc_cb
;
587 struct bpf_redirect_info
{
592 struct bpf_map
*map_to_flush
;
596 DECLARE_PER_CPU(struct bpf_redirect_info
, bpf_redirect_info
);
598 /* flags for bpf_redirect_info kern_flags */
599 #define BPF_RI_F_RF_NO_DIRECT BIT(0) /* no napi_direct on return_frame */
601 /* Compute the linear packet data range [data, data_end) which
602 * will be accessed by various program types (cls_bpf, act_bpf,
603 * lwt, ...). Subsystems allowing direct data access must (!)
604 * ensure that cb[] area can be written to when BPF program is
605 * invoked (otherwise cb[] save/restore is necessary).
607 static inline void bpf_compute_data_pointers(struct sk_buff
*skb
)
609 struct bpf_skb_data_end
*cb
= (struct bpf_skb_data_end
*)skb
->cb
;
611 BUILD_BUG_ON(sizeof(*cb
) > sizeof_field(struct sk_buff
, cb
));
612 cb
->data_meta
= skb
->data
- skb_metadata_len(skb
);
613 cb
->data_end
= skb
->data
+ skb_headlen(skb
);
616 /* Similar to bpf_compute_data_pointers(), except that save orginal
617 * data in cb->data and cb->meta_data for restore.
619 static inline void bpf_compute_and_save_data_end(
620 struct sk_buff
*skb
, void **saved_data_end
)
622 struct bpf_skb_data_end
*cb
= (struct bpf_skb_data_end
*)skb
->cb
;
624 *saved_data_end
= cb
->data_end
;
625 cb
->data_end
= skb
->data
+ skb_headlen(skb
);
628 /* Restore data saved by bpf_compute_data_pointers(). */
629 static inline void bpf_restore_data_end(
630 struct sk_buff
*skb
, void *saved_data_end
)
632 struct bpf_skb_data_end
*cb
= (struct bpf_skb_data_end
*)skb
->cb
;
634 cb
->data_end
= saved_data_end
;
637 static inline u8
*bpf_skb_cb(struct sk_buff
*skb
)
639 /* eBPF programs may read/write skb->cb[] area to transfer meta
640 * data between tail calls. Since this also needs to work with
641 * tc, that scratch memory is mapped to qdisc_skb_cb's data area.
643 * In some socket filter cases, the cb unfortunately needs to be
644 * saved/restored so that protocol specific skb->cb[] data won't
645 * be lost. In any case, due to unpriviledged eBPF programs
646 * attached to sockets, we need to clear the bpf_skb_cb() area
647 * to not leak previous contents to user space.
649 BUILD_BUG_ON(sizeof_field(struct __sk_buff
, cb
) != BPF_SKB_CB_LEN
);
650 BUILD_BUG_ON(sizeof_field(struct __sk_buff
, cb
) !=
651 sizeof_field(struct qdisc_skb_cb
, data
));
653 return qdisc_skb_cb(skb
)->data
;
656 static inline u32
__bpf_prog_run_save_cb(const struct bpf_prog
*prog
,
659 u8
*cb_data
= bpf_skb_cb(skb
);
660 u8 cb_saved
[BPF_SKB_CB_LEN
];
663 if (unlikely(prog
->cb_access
)) {
664 memcpy(cb_saved
, cb_data
, sizeof(cb_saved
));
665 memset(cb_data
, 0, sizeof(cb_saved
));
668 res
= BPF_PROG_RUN(prog
, skb
);
670 if (unlikely(prog
->cb_access
))
671 memcpy(cb_data
, cb_saved
, sizeof(cb_saved
));
676 static inline u32
bpf_prog_run_save_cb(const struct bpf_prog
*prog
,
682 res
= __bpf_prog_run_save_cb(prog
, skb
);
687 static inline u32
bpf_prog_run_clear_cb(const struct bpf_prog
*prog
,
690 u8
*cb_data
= bpf_skb_cb(skb
);
693 if (unlikely(prog
->cb_access
))
694 memset(cb_data
, 0, BPF_SKB_CB_LEN
);
697 res
= BPF_PROG_RUN(prog
, skb
);
702 static __always_inline u32
bpf_prog_run_xdp(const struct bpf_prog
*prog
,
703 struct xdp_buff
*xdp
)
705 /* Caller needs to hold rcu_read_lock() (!), otherwise program
706 * can be released while still running, or map elements could be
707 * freed early while still having concurrent users. XDP fastpath
708 * already takes rcu_read_lock() when fetching the program, so
709 * it's not necessary here anymore.
711 return BPF_PROG_RUN(prog
, xdp
);
714 static inline u32
bpf_prog_insn_size(const struct bpf_prog
*prog
)
716 return prog
->len
* sizeof(struct bpf_insn
);
719 static inline u32
bpf_prog_tag_scratch_size(const struct bpf_prog
*prog
)
721 return round_up(bpf_prog_insn_size(prog
) +
722 sizeof(__be64
) + 1, SHA_MESSAGE_BYTES
);
725 static inline unsigned int bpf_prog_size(unsigned int proglen
)
727 return max(sizeof(struct bpf_prog
),
728 offsetof(struct bpf_prog
, insns
[proglen
]));
731 static inline bool bpf_prog_was_classic(const struct bpf_prog
*prog
)
733 /* When classic BPF programs have been loaded and the arch
734 * does not have a classic BPF JIT (anymore), they have been
735 * converted via bpf_migrate_filter() to eBPF and thus always
736 * have an unspec program type.
738 return prog
->type
== BPF_PROG_TYPE_UNSPEC
;
741 static inline u32
bpf_ctx_off_adjust_machine(u32 size
)
743 const u32 size_machine
= sizeof(unsigned long);
745 if (size
> size_machine
&& size
% size_machine
== 0)
752 bpf_ctx_narrow_access_ok(u32 off
, u32 size
, u32 size_default
)
754 return size
<= size_default
&& (size
& (size
- 1)) == 0;
758 bpf_ctx_narrow_access_offset(u32 off
, u32 size
, u32 size_default
)
760 u8 access_off
= off
& (size_default
- 1);
762 #ifdef __LITTLE_ENDIAN
765 return size_default
- (access_off
+ size
);
769 #define bpf_ctx_wide_access_ok(off, size, type, field) \
770 (size == sizeof(__u64) && \
771 off >= offsetof(type, field) && \
772 off + sizeof(__u64) <= offsetofend(type, field) && \
773 off % sizeof(__u64) == 0)
775 #define bpf_classic_proglen(fprog) (fprog->len * sizeof(fprog->filter[0]))
777 static inline void bpf_prog_lock_ro(struct bpf_prog
*fp
)
779 #ifndef CONFIG_BPF_JIT_ALWAYS_ON
781 set_vm_flush_reset_perms(fp
);
782 set_memory_ro((unsigned long)fp
, fp
->pages
);
787 static inline void bpf_jit_binary_lock_ro(struct bpf_binary_header
*hdr
)
789 set_vm_flush_reset_perms(hdr
);
790 set_memory_ro((unsigned long)hdr
, hdr
->pages
);
791 set_memory_x((unsigned long)hdr
, hdr
->pages
);
794 static inline struct bpf_binary_header
*
795 bpf_jit_binary_hdr(const struct bpf_prog
*fp
)
797 unsigned long real_start
= (unsigned long)fp
->bpf_func
;
798 unsigned long addr
= real_start
& PAGE_MASK
;
803 int sk_filter_trim_cap(struct sock
*sk
, struct sk_buff
*skb
, unsigned int cap
);
804 static inline int sk_filter(struct sock
*sk
, struct sk_buff
*skb
)
806 return sk_filter_trim_cap(sk
, skb
, 1);
809 struct bpf_prog
*bpf_prog_select_runtime(struct bpf_prog
*fp
, int *err
);
810 void bpf_prog_free(struct bpf_prog
*fp
);
812 bool bpf_opcode_in_insntable(u8 code
);
814 void bpf_prog_free_linfo(struct bpf_prog
*prog
);
815 void bpf_prog_fill_jited_linfo(struct bpf_prog
*prog
,
816 const u32
*insn_to_jit_off
);
817 int bpf_prog_alloc_jited_linfo(struct bpf_prog
*prog
);
818 void bpf_prog_free_jited_linfo(struct bpf_prog
*prog
);
819 void bpf_prog_free_unused_jited_linfo(struct bpf_prog
*prog
);
821 struct bpf_prog
*bpf_prog_alloc(unsigned int size
, gfp_t gfp_extra_flags
);
822 struct bpf_prog
*bpf_prog_alloc_no_stats(unsigned int size
, gfp_t gfp_extra_flags
);
823 struct bpf_prog
*bpf_prog_realloc(struct bpf_prog
*fp_old
, unsigned int size
,
824 gfp_t gfp_extra_flags
);
825 void __bpf_prog_free(struct bpf_prog
*fp
);
827 static inline void bpf_prog_unlock_free(struct bpf_prog
*fp
)
832 typedef int (*bpf_aux_classic_check_t
)(struct sock_filter
*filter
,
835 int bpf_prog_create(struct bpf_prog
**pfp
, struct sock_fprog_kern
*fprog
);
836 int bpf_prog_create_from_user(struct bpf_prog
**pfp
, struct sock_fprog
*fprog
,
837 bpf_aux_classic_check_t trans
, bool save_orig
);
838 void bpf_prog_destroy(struct bpf_prog
*fp
);
840 int sk_attach_filter(struct sock_fprog
*fprog
, struct sock
*sk
);
841 int sk_attach_bpf(u32 ufd
, struct sock
*sk
);
842 int sk_reuseport_attach_filter(struct sock_fprog
*fprog
, struct sock
*sk
);
843 int sk_reuseport_attach_bpf(u32 ufd
, struct sock
*sk
);
844 void sk_reuseport_prog_free(struct bpf_prog
*prog
);
845 int sk_detach_filter(struct sock
*sk
);
846 int sk_get_filter(struct sock
*sk
, struct sock_filter __user
*filter
,
849 bool sk_filter_charge(struct sock
*sk
, struct sk_filter
*fp
);
850 void sk_filter_uncharge(struct sock
*sk
, struct sk_filter
*fp
);
852 u64
__bpf_call_base(u64 r1
, u64 r2
, u64 r3
, u64 r4
, u64 r5
);
853 #define __bpf_call_base_args \
854 ((u64 (*)(u64, u64, u64, u64, u64, const struct bpf_insn *)) \
857 struct bpf_prog
*bpf_int_jit_compile(struct bpf_prog
*prog
);
858 void bpf_jit_compile(struct bpf_prog
*prog
);
859 bool bpf_jit_needs_zext(void);
860 bool bpf_helper_changes_pkt_data(void *func
);
862 static inline bool bpf_dump_raw_ok(void)
864 /* Reconstruction of call-sites is dependent on kallsyms,
865 * thus make dump the same restriction.
867 return kallsyms_show_value() == 1;
870 struct bpf_prog
*bpf_patch_insn_single(struct bpf_prog
*prog
, u32 off
,
871 const struct bpf_insn
*patch
, u32 len
);
872 int bpf_remove_insns(struct bpf_prog
*prog
, u32 off
, u32 cnt
);
874 void bpf_clear_redirect_map(struct bpf_map
*map
);
876 static inline bool xdp_return_frame_no_direct(void)
878 struct bpf_redirect_info
*ri
= this_cpu_ptr(&bpf_redirect_info
);
880 return ri
->kern_flags
& BPF_RI_F_RF_NO_DIRECT
;
883 static inline void xdp_set_return_frame_no_direct(void)
885 struct bpf_redirect_info
*ri
= this_cpu_ptr(&bpf_redirect_info
);
887 ri
->kern_flags
|= BPF_RI_F_RF_NO_DIRECT
;
890 static inline void xdp_clear_return_frame_no_direct(void)
892 struct bpf_redirect_info
*ri
= this_cpu_ptr(&bpf_redirect_info
);
894 ri
->kern_flags
&= ~BPF_RI_F_RF_NO_DIRECT
;
897 static inline int xdp_ok_fwd_dev(const struct net_device
*fwd
,
902 if (unlikely(!(fwd
->flags
& IFF_UP
)))
905 len
= fwd
->mtu
+ fwd
->hard_header_len
+ VLAN_HLEN
;
912 /* The pair of xdp_do_redirect and xdp_do_flush_map MUST be called in the
913 * same cpu context. Further for best results no more than a single map
914 * for the do_redirect/do_flush pair should be used. This limitation is
915 * because we only track one map and force a flush when the map changes.
916 * This does not appear to be a real limitation for existing software.
918 int xdp_do_generic_redirect(struct net_device
*dev
, struct sk_buff
*skb
,
919 struct xdp_buff
*xdp
, struct bpf_prog
*prog
);
920 int xdp_do_redirect(struct net_device
*dev
,
921 struct xdp_buff
*xdp
,
922 struct bpf_prog
*prog
);
923 void xdp_do_flush_map(void);
925 void bpf_warn_invalid_xdp_action(u32 act
);
928 struct sock
*bpf_run_sk_reuseport(struct sock_reuseport
*reuse
, struct sock
*sk
,
929 struct bpf_prog
*prog
, struct sk_buff
*skb
,
932 static inline struct sock
*
933 bpf_run_sk_reuseport(struct sock_reuseport
*reuse
, struct sock
*sk
,
934 struct bpf_prog
*prog
, struct sk_buff
*skb
,
941 #ifdef CONFIG_BPF_JIT
942 extern int bpf_jit_enable
;
943 extern int bpf_jit_harden
;
944 extern int bpf_jit_kallsyms
;
945 extern long bpf_jit_limit
;
947 typedef void (*bpf_jit_fill_hole_t
)(void *area
, unsigned int size
);
949 struct bpf_binary_header
*
950 bpf_jit_binary_alloc(unsigned int proglen
, u8
**image_ptr
,
951 unsigned int alignment
,
952 bpf_jit_fill_hole_t bpf_fill_ill_insns
);
953 void bpf_jit_binary_free(struct bpf_binary_header
*hdr
);
954 u64
bpf_jit_alloc_exec_limit(void);
955 void *bpf_jit_alloc_exec(unsigned long size
);
956 void bpf_jit_free_exec(void *addr
);
957 void bpf_jit_free(struct bpf_prog
*fp
);
959 int bpf_jit_add_poke_descriptor(struct bpf_prog
*prog
,
960 struct bpf_jit_poke_descriptor
*poke
);
962 int bpf_jit_get_func_addr(const struct bpf_prog
*prog
,
963 const struct bpf_insn
*insn
, bool extra_pass
,
964 u64
*func_addr
, bool *func_addr_fixed
);
966 struct bpf_prog
*bpf_jit_blind_constants(struct bpf_prog
*fp
);
967 void bpf_jit_prog_release_other(struct bpf_prog
*fp
, struct bpf_prog
*fp_other
);
969 static inline void bpf_jit_dump(unsigned int flen
, unsigned int proglen
,
970 u32 pass
, void *image
)
972 pr_err("flen=%u proglen=%u pass=%u image=%pK from=%s pid=%d\n", flen
,
973 proglen
, pass
, image
, current
->comm
, task_pid_nr(current
));
976 print_hex_dump(KERN_ERR
, "JIT code: ", DUMP_PREFIX_OFFSET
,
977 16, 1, image
, proglen
, false);
980 static inline bool bpf_jit_is_ebpf(void)
982 # ifdef CONFIG_HAVE_EBPF_JIT
989 static inline bool ebpf_jit_enabled(void)
991 return bpf_jit_enable
&& bpf_jit_is_ebpf();
994 static inline bool bpf_prog_ebpf_jited(const struct bpf_prog
*fp
)
996 return fp
->jited
&& bpf_jit_is_ebpf();
999 static inline bool bpf_jit_blinding_enabled(struct bpf_prog
*prog
)
1001 /* These are the prerequisites, should someone ever have the
1002 * idea to call blinding outside of them, we make sure to
1005 if (!bpf_jit_is_ebpf())
1007 if (!prog
->jit_requested
)
1009 if (!bpf_jit_harden
)
1011 if (bpf_jit_harden
== 1 && capable(CAP_SYS_ADMIN
))
1017 static inline bool bpf_jit_kallsyms_enabled(void)
1019 /* There are a couple of corner cases where kallsyms should
1020 * not be enabled f.e. on hardening.
1024 if (!bpf_jit_kallsyms
)
1026 if (bpf_jit_kallsyms
== 1)
1032 const char *__bpf_address_lookup(unsigned long addr
, unsigned long *size
,
1033 unsigned long *off
, char *sym
);
1034 bool is_bpf_text_address(unsigned long addr
);
1035 int bpf_get_kallsym(unsigned int symnum
, unsigned long *value
, char *type
,
1038 static inline const char *
1039 bpf_address_lookup(unsigned long addr
, unsigned long *size
,
1040 unsigned long *off
, char **modname
, char *sym
)
1042 const char *ret
= __bpf_address_lookup(addr
, size
, off
, sym
);
1049 void bpf_prog_kallsyms_add(struct bpf_prog
*fp
);
1050 void bpf_prog_kallsyms_del(struct bpf_prog
*fp
);
1051 void bpf_get_prog_name(const struct bpf_prog
*prog
, char *sym
);
1053 #else /* CONFIG_BPF_JIT */
1055 static inline bool ebpf_jit_enabled(void)
1060 static inline bool bpf_jit_blinding_enabled(struct bpf_prog
*prog
)
1065 static inline bool bpf_prog_ebpf_jited(const struct bpf_prog
*fp
)
1071 bpf_jit_add_poke_descriptor(struct bpf_prog
*prog
,
1072 struct bpf_jit_poke_descriptor
*poke
)
1077 static inline void bpf_jit_free(struct bpf_prog
*fp
)
1079 bpf_prog_unlock_free(fp
);
1082 static inline bool bpf_jit_kallsyms_enabled(void)
1087 static inline const char *
1088 __bpf_address_lookup(unsigned long addr
, unsigned long *size
,
1089 unsigned long *off
, char *sym
)
1094 static inline bool is_bpf_text_address(unsigned long addr
)
1099 static inline int bpf_get_kallsym(unsigned int symnum
, unsigned long *value
,
1100 char *type
, char *sym
)
1105 static inline const char *
1106 bpf_address_lookup(unsigned long addr
, unsigned long *size
,
1107 unsigned long *off
, char **modname
, char *sym
)
1112 static inline void bpf_prog_kallsyms_add(struct bpf_prog
*fp
)
1116 static inline void bpf_prog_kallsyms_del(struct bpf_prog
*fp
)
1120 static inline void bpf_get_prog_name(const struct bpf_prog
*prog
, char *sym
)
1125 #endif /* CONFIG_BPF_JIT */
1127 void bpf_prog_kallsyms_del_all(struct bpf_prog
*fp
);
1129 #define BPF_ANC BIT(15)
1131 static inline bool bpf_needs_clear_a(const struct sock_filter
*first
)
1133 switch (first
->code
) {
1134 case BPF_RET
| BPF_K
:
1135 case BPF_LD
| BPF_W
| BPF_LEN
:
1138 case BPF_LD
| BPF_W
| BPF_ABS
:
1139 case BPF_LD
| BPF_H
| BPF_ABS
:
1140 case BPF_LD
| BPF_B
| BPF_ABS
:
1141 if (first
->k
== SKF_AD_OFF
+ SKF_AD_ALU_XOR_X
)
1150 static inline u16
bpf_anc_helper(const struct sock_filter
*ftest
)
1152 BUG_ON(ftest
->code
& BPF_ANC
);
1154 switch (ftest
->code
) {
1155 case BPF_LD
| BPF_W
| BPF_ABS
:
1156 case BPF_LD
| BPF_H
| BPF_ABS
:
1157 case BPF_LD
| BPF_B
| BPF_ABS
:
1158 #define BPF_ANCILLARY(CODE) case SKF_AD_OFF + SKF_AD_##CODE: \
1159 return BPF_ANC | SKF_AD_##CODE
1161 BPF_ANCILLARY(PROTOCOL
);
1162 BPF_ANCILLARY(PKTTYPE
);
1163 BPF_ANCILLARY(IFINDEX
);
1164 BPF_ANCILLARY(NLATTR
);
1165 BPF_ANCILLARY(NLATTR_NEST
);
1166 BPF_ANCILLARY(MARK
);
1167 BPF_ANCILLARY(QUEUE
);
1168 BPF_ANCILLARY(HATYPE
);
1169 BPF_ANCILLARY(RXHASH
);
1171 BPF_ANCILLARY(ALU_XOR_X
);
1172 BPF_ANCILLARY(VLAN_TAG
);
1173 BPF_ANCILLARY(VLAN_TAG_PRESENT
);
1174 BPF_ANCILLARY(PAY_OFFSET
);
1175 BPF_ANCILLARY(RANDOM
);
1176 BPF_ANCILLARY(VLAN_TPID
);
1184 void *bpf_internal_load_pointer_neg_helper(const struct sk_buff
*skb
,
1185 int k
, unsigned int size
);
1187 static inline void *bpf_load_pointer(const struct sk_buff
*skb
, int k
,
1188 unsigned int size
, void *buffer
)
1191 return skb_header_pointer(skb
, k
, size
, buffer
);
1193 return bpf_internal_load_pointer_neg_helper(skb
, k
, size
);
1196 static inline int bpf_tell_extensions(void)
1201 struct bpf_sock_addr_kern
{
1203 struct sockaddr
*uaddr
;
1204 /* Temporary "register" to make indirect stores to nested structures
1205 * defined above. We need three registers to make such a store, but
1206 * only two (src and dst) are available at convert_ctx_access time
1209 void *t_ctx
; /* Attach type specific context. */
1212 struct bpf_sock_ops_kern
{
1221 u64 temp
; /* temp and everything after is not
1222 * initialized to 0 before calling
1223 * the BPF program. New fields that
1224 * should be initialized to 0 should
1225 * be inserted before temp.
1226 * temp is scratch storage used by
1227 * sock_ops_convert_ctx_access
1228 * as temporary storage of a register.
1232 struct bpf_sysctl_kern
{
1233 struct ctl_table_header
*head
;
1234 struct ctl_table
*table
;
1242 /* Temporary "register" for indirect stores to ppos. */
1246 struct bpf_sockopt_kern
{
1256 #endif /* __LINUX_FILTER_H__ */