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>
24 #include <net/sch_generic.h>
26 #include <uapi/linux/filter.h>
27 #include <uapi/linux/bpf.h>
35 struct sock_reuseport;
37 /* ArgX, context and stack frame pointer register positions. Note,
38 * Arg1, Arg2, Arg3, etc are used as argument mappings of function
39 * calls in BPF_CALL instruction.
41 #define BPF_REG_ARG1 BPF_REG_1
42 #define BPF_REG_ARG2 BPF_REG_2
43 #define BPF_REG_ARG3 BPF_REG_3
44 #define BPF_REG_ARG4 BPF_REG_4
45 #define BPF_REG_ARG5 BPF_REG_5
46 #define BPF_REG_CTX BPF_REG_6
47 #define BPF_REG_FP BPF_REG_10
49 /* Additional register mappings for converted user programs. */
50 #define BPF_REG_A BPF_REG_0
51 #define BPF_REG_X BPF_REG_7
52 #define BPF_REG_TMP BPF_REG_2 /* scratch reg */
53 #define BPF_REG_D BPF_REG_8 /* data, callee-saved */
54 #define BPF_REG_H BPF_REG_9 /* hlen, callee-saved */
56 /* Kernel hidden auxiliary/helper register for hardening step.
57 * Only used by eBPF JITs. It's nothing more than a temporary
58 * register that JITs use internally, only that here it's part
59 * of eBPF instructions that have been rewritten for blinding
60 * constants. See JIT pre-step in bpf_jit_blind_constants().
62 #define BPF_REG_AX MAX_BPF_REG
63 #define MAX_BPF_JIT_REG (MAX_BPF_REG + 1)
65 /* unused opcode to mark special call to bpf_tail_call() helper */
66 #define BPF_TAIL_CALL 0xf0
68 /* unused opcode to mark call to interpreter with arguments */
69 #define BPF_CALL_ARGS 0xe0
71 /* As per nm, we expose JITed images as text (code) section for
72 * kallsyms. That way, tools like perf can find it to match
75 #define BPF_SYM_ELF_TYPE 't'
77 /* BPF program can access up to 512 bytes of stack space. */
78 #define MAX_BPF_STACK 512
80 /* Helper macros for filter block array initializers. */
82 /* ALU ops on registers, bpf_add|sub|...: dst_reg += src_reg */
84 #define BPF_ALU64_REG(OP, DST, SRC) \
85 ((struct bpf_insn) { \
86 .code = BPF_ALU64 | BPF_OP(OP) | BPF_X, \
92 #define BPF_ALU32_REG(OP, DST, SRC) \
93 ((struct bpf_insn) { \
94 .code = BPF_ALU | BPF_OP(OP) | BPF_X, \
100 /* ALU ops on immediates, bpf_add|sub|...: dst_reg += imm32 */
102 #define BPF_ALU64_IMM(OP, DST, IMM) \
103 ((struct bpf_insn) { \
104 .code = BPF_ALU64 | BPF_OP(OP) | BPF_K, \
110 #define BPF_ALU32_IMM(OP, DST, IMM) \
111 ((struct bpf_insn) { \
112 .code = BPF_ALU | BPF_OP(OP) | BPF_K, \
118 /* Endianess conversion, cpu_to_{l,b}e(), {l,b}e_to_cpu() */
120 #define BPF_ENDIAN(TYPE, DST, LEN) \
121 ((struct bpf_insn) { \
122 .code = BPF_ALU | BPF_END | BPF_SRC(TYPE), \
128 /* Short form of mov, dst_reg = src_reg */
130 #define BPF_MOV64_REG(DST, SRC) \
131 ((struct bpf_insn) { \
132 .code = BPF_ALU64 | BPF_MOV | BPF_X, \
138 #define BPF_MOV32_REG(DST, SRC) \
139 ((struct bpf_insn) { \
140 .code = BPF_ALU | BPF_MOV | BPF_X, \
146 /* Short form of mov, dst_reg = imm32 */
148 #define BPF_MOV64_IMM(DST, IMM) \
149 ((struct bpf_insn) { \
150 .code = BPF_ALU64 | BPF_MOV | BPF_K, \
156 #define BPF_MOV32_IMM(DST, IMM) \
157 ((struct bpf_insn) { \
158 .code = BPF_ALU | BPF_MOV | BPF_K, \
164 /* BPF_LD_IMM64 macro encodes single 'load 64-bit immediate' insn */
165 #define BPF_LD_IMM64(DST, IMM) \
166 BPF_LD_IMM64_RAW(DST, 0, IMM)
168 #define BPF_LD_IMM64_RAW(DST, SRC, IMM) \
169 ((struct bpf_insn) { \
170 .code = BPF_LD | BPF_DW | BPF_IMM, \
174 .imm = (__u32) (IMM) }), \
175 ((struct bpf_insn) { \
176 .code = 0, /* zero is reserved opcode */ \
180 .imm = ((__u64) (IMM)) >> 32 })
182 /* pseudo BPF_LD_IMM64 insn used to refer to process-local map_fd */
183 #define BPF_LD_MAP_FD(DST, MAP_FD) \
184 BPF_LD_IMM64_RAW(DST, BPF_PSEUDO_MAP_FD, MAP_FD)
186 /* Short form of mov based on type, BPF_X: dst_reg = src_reg, BPF_K: dst_reg = imm32 */
188 #define BPF_MOV64_RAW(TYPE, DST, SRC, IMM) \
189 ((struct bpf_insn) { \
190 .code = BPF_ALU64 | BPF_MOV | BPF_SRC(TYPE), \
196 #define BPF_MOV32_RAW(TYPE, DST, SRC, IMM) \
197 ((struct bpf_insn) { \
198 .code = BPF_ALU | BPF_MOV | BPF_SRC(TYPE), \
204 /* Direct packet access, R0 = *(uint *) (skb->data + imm32) */
206 #define BPF_LD_ABS(SIZE, IMM) \
207 ((struct bpf_insn) { \
208 .code = BPF_LD | BPF_SIZE(SIZE) | BPF_ABS, \
214 /* Indirect packet access, R0 = *(uint *) (skb->data + src_reg + imm32) */
216 #define BPF_LD_IND(SIZE, SRC, IMM) \
217 ((struct bpf_insn) { \
218 .code = BPF_LD | BPF_SIZE(SIZE) | BPF_IND, \
224 /* Memory load, dst_reg = *(uint *) (src_reg + off16) */
226 #define BPF_LDX_MEM(SIZE, DST, SRC, OFF) \
227 ((struct bpf_insn) { \
228 .code = BPF_LDX | BPF_SIZE(SIZE) | BPF_MEM, \
234 /* Memory store, *(uint *) (dst_reg + off16) = src_reg */
236 #define BPF_STX_MEM(SIZE, DST, SRC, OFF) \
237 ((struct bpf_insn) { \
238 .code = BPF_STX | BPF_SIZE(SIZE) | BPF_MEM, \
244 /* Atomic memory add, *(uint *)(dst_reg + off16) += src_reg */
246 #define BPF_STX_XADD(SIZE, DST, SRC, OFF) \
247 ((struct bpf_insn) { \
248 .code = BPF_STX | BPF_SIZE(SIZE) | BPF_XADD, \
254 /* Memory store, *(uint *) (dst_reg + off16) = imm32 */
256 #define BPF_ST_MEM(SIZE, DST, OFF, IMM) \
257 ((struct bpf_insn) { \
258 .code = BPF_ST | BPF_SIZE(SIZE) | BPF_MEM, \
264 /* Conditional jumps against registers, if (dst_reg 'op' src_reg) goto pc + off16 */
266 #define BPF_JMP_REG(OP, DST, SRC, OFF) \
267 ((struct bpf_insn) { \
268 .code = BPF_JMP | BPF_OP(OP) | BPF_X, \
274 /* Conditional jumps against immediates, if (dst_reg 'op' imm32) goto pc + off16 */
276 #define BPF_JMP_IMM(OP, DST, IMM, OFF) \
277 ((struct bpf_insn) { \
278 .code = BPF_JMP | BPF_OP(OP) | BPF_K, \
284 /* Unconditional jumps, goto pc + off16 */
286 #define BPF_JMP_A(OFF) \
287 ((struct bpf_insn) { \
288 .code = BPF_JMP | BPF_JA, \
296 #define BPF_CALL_REL(TGT) \
297 ((struct bpf_insn) { \
298 .code = BPF_JMP | BPF_CALL, \
300 .src_reg = BPF_PSEUDO_CALL, \
306 #define BPF_CAST_CALL(x) \
307 ((u64 (*)(u64, u64, u64, u64, u64))(x))
309 #define BPF_EMIT_CALL(FUNC) \
310 ((struct bpf_insn) { \
311 .code = BPF_JMP | BPF_CALL, \
315 .imm = ((FUNC) - __bpf_call_base) })
317 /* Raw code statement block */
319 #define BPF_RAW_INSN(CODE, DST, SRC, OFF, IMM) \
320 ((struct bpf_insn) { \
329 #define BPF_EXIT_INSN() \
330 ((struct bpf_insn) { \
331 .code = BPF_JMP | BPF_EXIT, \
337 /* Internal classic blocks for direct assignment */
339 #define __BPF_STMT(CODE, K) \
340 ((struct sock_filter) BPF_STMT(CODE, K))
342 #define __BPF_JUMP(CODE, K, JT, JF) \
343 ((struct sock_filter) BPF_JUMP(CODE, K, JT, JF))
345 #define bytes_to_bpf_size(bytes) \
347 int bpf_size = -EINVAL; \
349 if (bytes == sizeof(u8)) \
351 else if (bytes == sizeof(u16)) \
353 else if (bytes == sizeof(u32)) \
355 else if (bytes == sizeof(u64)) \
361 #define bpf_size_to_bytes(bpf_size) \
363 int bytes = -EINVAL; \
365 if (bpf_size == BPF_B) \
366 bytes = sizeof(u8); \
367 else if (bpf_size == BPF_H) \
368 bytes = sizeof(u16); \
369 else if (bpf_size == BPF_W) \
370 bytes = sizeof(u32); \
371 else if (bpf_size == BPF_DW) \
372 bytes = sizeof(u64); \
377 #define BPF_SIZEOF(type) \
379 const int __size = bytes_to_bpf_size(sizeof(type)); \
380 BUILD_BUG_ON(__size < 0); \
384 #define BPF_FIELD_SIZEOF(type, field) \
386 const int __size = bytes_to_bpf_size(FIELD_SIZEOF(type, field)); \
387 BUILD_BUG_ON(__size < 0); \
391 #define BPF_LDST_BYTES(insn) \
393 const int __size = bpf_size_to_bytes(BPF_SIZE((insn)->code)); \
394 WARN_ON(__size < 0); \
398 #define __BPF_MAP_0(m, v, ...) v
399 #define __BPF_MAP_1(m, v, t, a, ...) m(t, a)
400 #define __BPF_MAP_2(m, v, t, a, ...) m(t, a), __BPF_MAP_1(m, v, __VA_ARGS__)
401 #define __BPF_MAP_3(m, v, t, a, ...) m(t, a), __BPF_MAP_2(m, v, __VA_ARGS__)
402 #define __BPF_MAP_4(m, v, t, a, ...) m(t, a), __BPF_MAP_3(m, v, __VA_ARGS__)
403 #define __BPF_MAP_5(m, v, t, a, ...) m(t, a), __BPF_MAP_4(m, v, __VA_ARGS__)
405 #define __BPF_REG_0(...) __BPF_PAD(5)
406 #define __BPF_REG_1(...) __BPF_MAP(1, __VA_ARGS__), __BPF_PAD(4)
407 #define __BPF_REG_2(...) __BPF_MAP(2, __VA_ARGS__), __BPF_PAD(3)
408 #define __BPF_REG_3(...) __BPF_MAP(3, __VA_ARGS__), __BPF_PAD(2)
409 #define __BPF_REG_4(...) __BPF_MAP(4, __VA_ARGS__), __BPF_PAD(1)
410 #define __BPF_REG_5(...) __BPF_MAP(5, __VA_ARGS__)
412 #define __BPF_MAP(n, ...) __BPF_MAP_##n(__VA_ARGS__)
413 #define __BPF_REG(n, ...) __BPF_REG_##n(__VA_ARGS__)
415 #define __BPF_CAST(t, a) \
418 typeof(__builtin_choose_expr(sizeof(t) == sizeof(unsigned long), \
419 (unsigned long)0, (t)0))) a
423 #define __BPF_DECL_ARGS(t, a) t a
424 #define __BPF_DECL_REGS(t, a) u64 a
426 #define __BPF_PAD(n) \
427 __BPF_MAP(n, __BPF_DECL_ARGS, __BPF_N, u64, __ur_1, u64, __ur_2, \
428 u64, __ur_3, u64, __ur_4, u64, __ur_5)
430 #define BPF_CALL_x(x, name, ...) \
431 static __always_inline \
432 u64 ____##name(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__)); \
433 u64 name(__BPF_REG(x, __BPF_DECL_REGS, __BPF_N, __VA_ARGS__)); \
434 u64 name(__BPF_REG(x, __BPF_DECL_REGS, __BPF_N, __VA_ARGS__)) \
436 return ____##name(__BPF_MAP(x,__BPF_CAST,__BPF_N,__VA_ARGS__));\
438 static __always_inline \
439 u64 ____##name(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__))
441 #define BPF_CALL_0(name, ...) BPF_CALL_x(0, name, __VA_ARGS__)
442 #define BPF_CALL_1(name, ...) BPF_CALL_x(1, name, __VA_ARGS__)
443 #define BPF_CALL_2(name, ...) BPF_CALL_x(2, name, __VA_ARGS__)
444 #define BPF_CALL_3(name, ...) BPF_CALL_x(3, name, __VA_ARGS__)
445 #define BPF_CALL_4(name, ...) BPF_CALL_x(4, name, __VA_ARGS__)
446 #define BPF_CALL_5(name, ...) BPF_CALL_x(5, name, __VA_ARGS__)
448 #define bpf_ctx_range(TYPE, MEMBER) \
449 offsetof(TYPE, MEMBER) ... offsetofend(TYPE, MEMBER) - 1
450 #define bpf_ctx_range_till(TYPE, MEMBER1, MEMBER2) \
451 offsetof(TYPE, MEMBER1) ... offsetofend(TYPE, MEMBER2) - 1
452 #if BITS_PER_LONG == 64
453 # define bpf_ctx_range_ptr(TYPE, MEMBER) \
454 offsetof(TYPE, MEMBER) ... offsetofend(TYPE, MEMBER) - 1
456 # define bpf_ctx_range_ptr(TYPE, MEMBER) \
457 offsetof(TYPE, MEMBER) ... offsetof(TYPE, MEMBER) + 8 - 1
458 #endif /* BITS_PER_LONG == 64 */
460 #define bpf_target_off(TYPE, MEMBER, SIZE, PTR_SIZE) \
462 BUILD_BUG_ON(FIELD_SIZEOF(TYPE, MEMBER) != (SIZE)); \
463 *(PTR_SIZE) = (SIZE); \
464 offsetof(TYPE, MEMBER); \
468 /* A struct sock_filter is architecture independent. */
469 struct compat_sock_fprog {
471 compat_uptr_t filter; /* struct sock_filter * */
475 struct sock_fprog_kern {
477 struct sock_filter *filter;
480 struct bpf_binary_header {
482 /* Some arches need word alignment for their instructions */
483 u8 image[] __aligned(4);
487 u16 pages; /* Number of allocated pages */
488 u16 jited:1, /* Is our filter JIT'ed? */
489 jit_requested:1,/* archs need to JIT the prog */
490 undo_set_mem:1, /* Passed set_memory_ro() checkpoint */
491 gpl_compatible:1, /* Is filter GPL compatible? */
492 cb_access:1, /* Is control block accessed? */
493 dst_needed:1, /* Do we need dst entry? */
494 blinded:1, /* Was blinded */
495 is_func:1, /* program is a bpf function */
496 kprobe_override:1, /* Do we override a kprobe? */
497 has_callchain_buf:1; /* callchain buffer allocated? */
498 enum bpf_prog_type type; /* Type of BPF program */
499 enum bpf_attach_type expected_attach_type; /* For some prog types */
500 u32 len; /* Number of filter blocks */
501 u32 jited_len; /* Size of jited insns in bytes */
502 u8 tag[BPF_TAG_SIZE];
503 struct bpf_prog_aux *aux; /* Auxiliary fields */
504 struct sock_fprog_kern *orig_prog; /* Original BPF program */
505 unsigned int (*bpf_func)(const void *ctx,
506 const struct bpf_insn *insn);
507 /* Instructions for interpreter */
509 struct sock_filter insns[0];
510 struct bpf_insn insnsi[0];
517 struct bpf_prog *prog;
520 #define BPF_PROG_RUN(filter, ctx) (*(filter)->bpf_func)(ctx, (filter)->insnsi)
522 #define BPF_SKB_CB_LEN QDISC_CB_PRIV_LEN
524 struct bpf_skb_data_end {
525 struct qdisc_skb_cb qdisc_cb;
530 struct bpf_redirect_info {
534 struct bpf_map *map_to_flush;
538 DECLARE_PER_CPU(struct bpf_redirect_info, bpf_redirect_info);
540 /* flags for bpf_redirect_info kern_flags */
541 #define BPF_RI_F_RF_NO_DIRECT BIT(0) /* no napi_direct on return_frame */
543 /* Compute the linear packet data range [data, data_end) which
544 * will be accessed by various program types (cls_bpf, act_bpf,
545 * lwt, ...). Subsystems allowing direct data access must (!)
546 * ensure that cb[] area can be written to when BPF program is
547 * invoked (otherwise cb[] save/restore is necessary).
549 static inline void bpf_compute_data_pointers(struct sk_buff *skb)
551 struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb;
553 BUILD_BUG_ON(sizeof(*cb) > FIELD_SIZEOF(struct sk_buff, cb));
554 cb->data_meta = skb->data - skb_metadata_len(skb);
555 cb->data_end = skb->data + skb_headlen(skb);
558 /* Similar to bpf_compute_data_pointers(), except that save orginal
559 * data in cb->data and cb->meta_data for restore.
561 static inline void bpf_compute_and_save_data_end(
562 struct sk_buff *skb, void **saved_data_end)
564 struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb;
566 *saved_data_end = cb->data_end;
567 cb->data_end = skb->data + skb_headlen(skb);
570 /* Restore data saved by bpf_compute_data_pointers(). */
571 static inline void bpf_restore_data_end(
572 struct sk_buff *skb, void *saved_data_end)
574 struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb;
576 cb->data_end = saved_data_end;
579 static inline u8 *bpf_skb_cb(struct sk_buff *skb)
581 /* eBPF programs may read/write skb->cb[] area to transfer meta
582 * data between tail calls. Since this also needs to work with
583 * tc, that scratch memory is mapped to qdisc_skb_cb's data area.
585 * In some socket filter cases, the cb unfortunately needs to be
586 * saved/restored so that protocol specific skb->cb[] data won't
587 * be lost. In any case, due to unpriviledged eBPF programs
588 * attached to sockets, we need to clear the bpf_skb_cb() area
589 * to not leak previous contents to user space.
591 BUILD_BUG_ON(FIELD_SIZEOF(struct __sk_buff, cb) != BPF_SKB_CB_LEN);
592 BUILD_BUG_ON(FIELD_SIZEOF(struct __sk_buff, cb) !=
593 FIELD_SIZEOF(struct qdisc_skb_cb, data));
595 return qdisc_skb_cb(skb)->data;
598 static inline u32 bpf_prog_run_save_cb(const struct bpf_prog *prog,
601 u8 *cb_data = bpf_skb_cb(skb);
602 u8 cb_saved[BPF_SKB_CB_LEN];
605 if (unlikely(prog->cb_access)) {
606 memcpy(cb_saved, cb_data, sizeof(cb_saved));
607 memset(cb_data, 0, sizeof(cb_saved));
610 res = BPF_PROG_RUN(prog, skb);
612 if (unlikely(prog->cb_access))
613 memcpy(cb_data, cb_saved, sizeof(cb_saved));
618 static inline u32 bpf_prog_run_clear_cb(const struct bpf_prog *prog,
621 u8 *cb_data = bpf_skb_cb(skb);
623 if (unlikely(prog->cb_access))
624 memset(cb_data, 0, BPF_SKB_CB_LEN);
626 return BPF_PROG_RUN(prog, skb);
629 static __always_inline u32 bpf_prog_run_xdp(const struct bpf_prog *prog,
630 struct xdp_buff *xdp)
632 /* Caller needs to hold rcu_read_lock() (!), otherwise program
633 * can be released while still running, or map elements could be
634 * freed early while still having concurrent users. XDP fastpath
635 * already takes rcu_read_lock() when fetching the program, so
636 * it's not necessary here anymore.
638 return BPF_PROG_RUN(prog, xdp);
641 static inline u32 bpf_prog_insn_size(const struct bpf_prog *prog)
643 return prog->len * sizeof(struct bpf_insn);
646 static inline u32 bpf_prog_tag_scratch_size(const struct bpf_prog *prog)
648 return round_up(bpf_prog_insn_size(prog) +
649 sizeof(__be64) + 1, SHA_MESSAGE_BYTES);
652 static inline unsigned int bpf_prog_size(unsigned int proglen)
654 return max(sizeof(struct bpf_prog),
655 offsetof(struct bpf_prog, insns[proglen]));
658 static inline bool bpf_prog_was_classic(const struct bpf_prog *prog)
660 /* When classic BPF programs have been loaded and the arch
661 * does not have a classic BPF JIT (anymore), they have been
662 * converted via bpf_migrate_filter() to eBPF and thus always
663 * have an unspec program type.
665 return prog->type == BPF_PROG_TYPE_UNSPEC;
668 static inline u32 bpf_ctx_off_adjust_machine(u32 size)
670 const u32 size_machine = sizeof(unsigned long);
672 if (size > size_machine && size % size_machine == 0)
679 bpf_ctx_narrow_access_ok(u32 off, u32 size, u32 size_default)
681 return size <= size_default && (size & (size - 1)) == 0;
684 #define bpf_classic_proglen(fprog) (fprog->len * sizeof(fprog->filter[0]))
686 static inline void bpf_prog_lock_ro(struct bpf_prog *fp)
688 fp->undo_set_mem = 1;
689 set_memory_ro((unsigned long)fp, fp->pages);
692 static inline void bpf_prog_unlock_ro(struct bpf_prog *fp)
694 if (fp->undo_set_mem)
695 set_memory_rw((unsigned long)fp, fp->pages);
698 static inline void bpf_jit_binary_lock_ro(struct bpf_binary_header *hdr)
700 set_memory_ro((unsigned long)hdr, hdr->pages);
703 static inline void bpf_jit_binary_unlock_ro(struct bpf_binary_header *hdr)
705 set_memory_rw((unsigned long)hdr, hdr->pages);
708 static inline struct bpf_binary_header *
709 bpf_jit_binary_hdr(const struct bpf_prog *fp)
711 unsigned long real_start = (unsigned long)fp->bpf_func;
712 unsigned long addr = real_start & PAGE_MASK;
717 int sk_filter_trim_cap(struct sock *sk, struct sk_buff *skb, unsigned int cap);
718 static inline int sk_filter(struct sock *sk, struct sk_buff *skb)
720 return sk_filter_trim_cap(sk, skb, 1);
723 struct bpf_prog *bpf_prog_select_runtime(struct bpf_prog *fp, int *err);
724 void bpf_prog_free(struct bpf_prog *fp);
726 bool bpf_opcode_in_insntable(u8 code);
728 struct bpf_prog *bpf_prog_alloc(unsigned int size, gfp_t gfp_extra_flags);
729 struct bpf_prog *bpf_prog_realloc(struct bpf_prog *fp_old, unsigned int size,
730 gfp_t gfp_extra_flags);
731 void __bpf_prog_free(struct bpf_prog *fp);
733 static inline void bpf_prog_unlock_free(struct bpf_prog *fp)
735 bpf_prog_unlock_ro(fp);
739 typedef int (*bpf_aux_classic_check_t)(struct sock_filter *filter,
742 int bpf_prog_create(struct bpf_prog **pfp, struct sock_fprog_kern *fprog);
743 int bpf_prog_create_from_user(struct bpf_prog **pfp, struct sock_fprog *fprog,
744 bpf_aux_classic_check_t trans, bool save_orig);
745 void bpf_prog_destroy(struct bpf_prog *fp);
747 int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk);
748 int sk_attach_bpf(u32 ufd, struct sock *sk);
749 int sk_reuseport_attach_filter(struct sock_fprog *fprog, struct sock *sk);
750 int sk_reuseport_attach_bpf(u32 ufd, struct sock *sk);
751 void sk_reuseport_prog_free(struct bpf_prog *prog);
752 int sk_detach_filter(struct sock *sk);
753 int sk_get_filter(struct sock *sk, struct sock_filter __user *filter,
756 bool sk_filter_charge(struct sock *sk, struct sk_filter *fp);
757 void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp);
759 u64 __bpf_call_base(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5);
760 #define __bpf_call_base_args \
761 ((u64 (*)(u64, u64, u64, u64, u64, const struct bpf_insn *)) \
764 struct bpf_prog *bpf_int_jit_compile(struct bpf_prog *prog);
765 void bpf_jit_compile(struct bpf_prog *prog);
766 bool bpf_helper_changes_pkt_data(void *func);
768 static inline bool bpf_dump_raw_ok(void)
770 /* Reconstruction of call-sites is dependent on kallsyms,
771 * thus make dump the same restriction.
773 return kallsyms_show_value() == 1;
776 struct bpf_prog *bpf_patch_insn_single(struct bpf_prog *prog, u32 off,
777 const struct bpf_insn *patch, u32 len);
779 void bpf_clear_redirect_map(struct bpf_map *map);
781 static inline bool xdp_return_frame_no_direct(void)
783 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
785 return ri->kern_flags & BPF_RI_F_RF_NO_DIRECT;
788 static inline void xdp_set_return_frame_no_direct(void)
790 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
792 ri->kern_flags |= BPF_RI_F_RF_NO_DIRECT;
795 static inline void xdp_clear_return_frame_no_direct(void)
797 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
799 ri->kern_flags &= ~BPF_RI_F_RF_NO_DIRECT;
802 static inline int xdp_ok_fwd_dev(const struct net_device *fwd,
807 if (unlikely(!(fwd->flags & IFF_UP)))
810 len = fwd->mtu + fwd->hard_header_len + VLAN_HLEN;
817 /* The pair of xdp_do_redirect and xdp_do_flush_map MUST be called in the
818 * same cpu context. Further for best results no more than a single map
819 * for the do_redirect/do_flush pair should be used. This limitation is
820 * because we only track one map and force a flush when the map changes.
821 * This does not appear to be a real limitation for existing software.
823 int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb,
824 struct xdp_buff *xdp, struct bpf_prog *prog);
825 int xdp_do_redirect(struct net_device *dev,
826 struct xdp_buff *xdp,
827 struct bpf_prog *prog);
828 void xdp_do_flush_map(void);
830 void bpf_warn_invalid_xdp_action(u32 act);
833 struct sock *bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk,
834 struct bpf_prog *prog, struct sk_buff *skb,
837 static inline struct sock *
838 bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk,
839 struct bpf_prog *prog, struct sk_buff *skb,
846 #ifdef CONFIG_BPF_JIT
847 extern int bpf_jit_enable;
848 extern int bpf_jit_harden;
849 extern int bpf_jit_kallsyms;
850 extern int bpf_jit_limit;
852 typedef void (*bpf_jit_fill_hole_t)(void *area, unsigned int size);
854 struct bpf_binary_header *
855 bpf_jit_binary_alloc(unsigned int proglen, u8 **image_ptr,
856 unsigned int alignment,
857 bpf_jit_fill_hole_t bpf_fill_ill_insns);
858 void bpf_jit_binary_free(struct bpf_binary_header *hdr);
860 void bpf_jit_free(struct bpf_prog *fp);
862 int bpf_jit_get_func_addr(const struct bpf_prog *prog,
863 const struct bpf_insn *insn, bool extra_pass,
864 u64 *func_addr, bool *func_addr_fixed);
866 struct bpf_prog *bpf_jit_blind_constants(struct bpf_prog *fp);
867 void bpf_jit_prog_release_other(struct bpf_prog *fp, struct bpf_prog *fp_other);
869 static inline void bpf_jit_dump(unsigned int flen, unsigned int proglen,
870 u32 pass, void *image)
872 pr_err("flen=%u proglen=%u pass=%u image=%pK from=%s pid=%d\n", flen,
873 proglen, pass, image, current->comm, task_pid_nr(current));
876 print_hex_dump(KERN_ERR, "JIT code: ", DUMP_PREFIX_OFFSET,
877 16, 1, image, proglen, false);
880 static inline bool bpf_jit_is_ebpf(void)
882 # ifdef CONFIG_HAVE_EBPF_JIT
889 static inline bool ebpf_jit_enabled(void)
891 return bpf_jit_enable && bpf_jit_is_ebpf();
894 static inline bool bpf_prog_ebpf_jited(const struct bpf_prog *fp)
896 return fp->jited && bpf_jit_is_ebpf();
899 static inline bool bpf_jit_blinding_enabled(struct bpf_prog *prog)
901 /* These are the prerequisites, should someone ever have the
902 * idea to call blinding outside of them, we make sure to
905 if (!bpf_jit_is_ebpf())
907 if (!prog->jit_requested)
911 if (bpf_jit_harden == 1 && capable(CAP_SYS_ADMIN))
917 static inline bool bpf_jit_kallsyms_enabled(void)
919 /* There are a couple of corner cases where kallsyms should
920 * not be enabled f.e. on hardening.
924 if (!bpf_jit_kallsyms)
926 if (bpf_jit_kallsyms == 1)
932 const char *__bpf_address_lookup(unsigned long addr, unsigned long *size,
933 unsigned long *off, char *sym);
934 bool is_bpf_text_address(unsigned long addr);
935 int bpf_get_kallsym(unsigned int symnum, unsigned long *value, char *type,
938 static inline const char *
939 bpf_address_lookup(unsigned long addr, unsigned long *size,
940 unsigned long *off, char **modname, char *sym)
942 const char *ret = __bpf_address_lookup(addr, size, off, sym);
949 void bpf_prog_kallsyms_add(struct bpf_prog *fp);
950 void bpf_prog_kallsyms_del(struct bpf_prog *fp);
952 #else /* CONFIG_BPF_JIT */
954 static inline bool ebpf_jit_enabled(void)
959 static inline bool bpf_prog_ebpf_jited(const struct bpf_prog *fp)
964 static inline void bpf_jit_free(struct bpf_prog *fp)
966 bpf_prog_unlock_free(fp);
969 static inline bool bpf_jit_kallsyms_enabled(void)
974 static inline const char *
975 __bpf_address_lookup(unsigned long addr, unsigned long *size,
976 unsigned long *off, char *sym)
981 static inline bool is_bpf_text_address(unsigned long addr)
986 static inline int bpf_get_kallsym(unsigned int symnum, unsigned long *value,
987 char *type, char *sym)
992 static inline const char *
993 bpf_address_lookup(unsigned long addr, unsigned long *size,
994 unsigned long *off, char **modname, char *sym)
999 static inline void bpf_prog_kallsyms_add(struct bpf_prog *fp)
1003 static inline void bpf_prog_kallsyms_del(struct bpf_prog *fp)
1006 #endif /* CONFIG_BPF_JIT */
1008 void bpf_prog_kallsyms_del_subprogs(struct bpf_prog *fp);
1009 void bpf_prog_kallsyms_del_all(struct bpf_prog *fp);
1011 #define BPF_ANC BIT(15)
1013 static inline bool bpf_needs_clear_a(const struct sock_filter *first)
1015 switch (first->code) {
1016 case BPF_RET | BPF_K:
1017 case BPF_LD | BPF_W | BPF_LEN:
1020 case BPF_LD | BPF_W | BPF_ABS:
1021 case BPF_LD | BPF_H | BPF_ABS:
1022 case BPF_LD | BPF_B | BPF_ABS:
1023 if (first->k == SKF_AD_OFF + SKF_AD_ALU_XOR_X)
1032 static inline u16 bpf_anc_helper(const struct sock_filter *ftest)
1034 BUG_ON(ftest->code & BPF_ANC);
1036 switch (ftest->code) {
1037 case BPF_LD | BPF_W | BPF_ABS:
1038 case BPF_LD | BPF_H | BPF_ABS:
1039 case BPF_LD | BPF_B | BPF_ABS:
1040 #define BPF_ANCILLARY(CODE) case SKF_AD_OFF + SKF_AD_##CODE: \
1041 return BPF_ANC | SKF_AD_##CODE
1043 BPF_ANCILLARY(PROTOCOL);
1044 BPF_ANCILLARY(PKTTYPE);
1045 BPF_ANCILLARY(IFINDEX);
1046 BPF_ANCILLARY(NLATTR);
1047 BPF_ANCILLARY(NLATTR_NEST);
1048 BPF_ANCILLARY(MARK);
1049 BPF_ANCILLARY(QUEUE);
1050 BPF_ANCILLARY(HATYPE);
1051 BPF_ANCILLARY(RXHASH);
1053 BPF_ANCILLARY(ALU_XOR_X);
1054 BPF_ANCILLARY(VLAN_TAG);
1055 BPF_ANCILLARY(VLAN_TAG_PRESENT);
1056 BPF_ANCILLARY(PAY_OFFSET);
1057 BPF_ANCILLARY(RANDOM);
1058 BPF_ANCILLARY(VLAN_TPID);
1066 void *bpf_internal_load_pointer_neg_helper(const struct sk_buff *skb,
1067 int k, unsigned int size);
1069 static inline void *bpf_load_pointer(const struct sk_buff *skb, int k,
1070 unsigned int size, void *buffer)
1073 return skb_header_pointer(skb, k, size, buffer);
1075 return bpf_internal_load_pointer_neg_helper(skb, k, size);
1078 static inline int bpf_tell_extensions(void)
1083 struct bpf_sock_addr_kern {
1085 struct sockaddr *uaddr;
1086 /* Temporary "register" to make indirect stores to nested structures
1087 * defined above. We need three registers to make such a store, but
1088 * only two (src and dst) are available at convert_ctx_access time
1091 void *t_ctx; /* Attach type specific context. */
1094 struct bpf_sock_ops_kern {
1103 u64 temp; /* temp and everything after is not
1104 * initialized to 0 before calling
1105 * the BPF program. New fields that
1106 * should be initialized to 0 should
1107 * be inserted before temp.
1108 * temp is scratch storage used by
1109 * sock_ops_convert_ctx_access
1110 * as temporary storage of a register.
1114 #endif /* __LINUX_FILTER_H__ */