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>
36 /* ArgX, context and stack frame pointer register positions. Note,
37 * Arg1, Arg2, Arg3, etc are used as argument mappings of function
38 * calls in BPF_CALL instruction.
40 #define BPF_REG_ARG1 BPF_REG_1
41 #define BPF_REG_ARG2 BPF_REG_2
42 #define BPF_REG_ARG3 BPF_REG_3
43 #define BPF_REG_ARG4 BPF_REG_4
44 #define BPF_REG_ARG5 BPF_REG_5
45 #define BPF_REG_CTX BPF_REG_6
46 #define BPF_REG_FP BPF_REG_10
48 /* Additional register mappings for converted user programs. */
49 #define BPF_REG_A BPF_REG_0
50 #define BPF_REG_X BPF_REG_7
51 #define BPF_REG_TMP BPF_REG_2 /* scratch reg */
52 #define BPF_REG_D BPF_REG_8 /* data, callee-saved */
53 #define BPF_REG_H BPF_REG_9 /* hlen, callee-saved */
55 /* Kernel hidden auxiliary/helper register for hardening step.
56 * Only used by eBPF JITs. It's nothing more than a temporary
57 * register that JITs use internally, only that here it's part
58 * of eBPF instructions that have been rewritten for blinding
59 * constants. See JIT pre-step in bpf_jit_blind_constants().
61 #define BPF_REG_AX MAX_BPF_REG
62 #define MAX_BPF_JIT_REG (MAX_BPF_REG + 1)
64 /* unused opcode to mark special call to bpf_tail_call() helper */
65 #define BPF_TAIL_CALL 0xf0
67 /* unused opcode to mark call to interpreter with arguments */
68 #define BPF_CALL_ARGS 0xe0
70 /* As per nm, we expose JITed images as text (code) section for
71 * kallsyms. That way, tools like perf can find it to match
74 #define BPF_SYM_ELF_TYPE 't'
76 /* BPF program can access up to 512 bytes of stack space. */
77 #define MAX_BPF_STACK 512
79 /* Helper macros for filter block array initializers. */
81 /* ALU ops on registers, bpf_add|sub|...: dst_reg += src_reg */
83 #define BPF_ALU64_REG(OP, DST, SRC) \
84 ((struct bpf_insn) { \
85 .code = BPF_ALU64 | BPF_OP(OP) | BPF_X, \
91 #define BPF_ALU32_REG(OP, DST, SRC) \
92 ((struct bpf_insn) { \
93 .code = BPF_ALU | BPF_OP(OP) | BPF_X, \
99 /* ALU ops on immediates, bpf_add|sub|...: dst_reg += imm32 */
101 #define BPF_ALU64_IMM(OP, DST, IMM) \
102 ((struct bpf_insn) { \
103 .code = BPF_ALU64 | BPF_OP(OP) | BPF_K, \
109 #define BPF_ALU32_IMM(OP, DST, IMM) \
110 ((struct bpf_insn) { \
111 .code = BPF_ALU | BPF_OP(OP) | BPF_K, \
117 /* Endianess conversion, cpu_to_{l,b}e(), {l,b}e_to_cpu() */
119 #define BPF_ENDIAN(TYPE, DST, LEN) \
120 ((struct bpf_insn) { \
121 .code = BPF_ALU | BPF_END | BPF_SRC(TYPE), \
127 /* Short form of mov, dst_reg = src_reg */
129 #define BPF_MOV64_REG(DST, SRC) \
130 ((struct bpf_insn) { \
131 .code = BPF_ALU64 | BPF_MOV | BPF_X, \
137 #define BPF_MOV32_REG(DST, SRC) \
138 ((struct bpf_insn) { \
139 .code = BPF_ALU | BPF_MOV | BPF_X, \
145 /* Short form of mov, dst_reg = imm32 */
147 #define BPF_MOV64_IMM(DST, IMM) \
148 ((struct bpf_insn) { \
149 .code = BPF_ALU64 | BPF_MOV | BPF_K, \
155 #define BPF_MOV32_IMM(DST, IMM) \
156 ((struct bpf_insn) { \
157 .code = BPF_ALU | BPF_MOV | BPF_K, \
163 /* BPF_LD_IMM64 macro encodes single 'load 64-bit immediate' insn */
164 #define BPF_LD_IMM64(DST, IMM) \
165 BPF_LD_IMM64_RAW(DST, 0, IMM)
167 #define BPF_LD_IMM64_RAW(DST, SRC, IMM) \
168 ((struct bpf_insn) { \
169 .code = BPF_LD | BPF_DW | BPF_IMM, \
173 .imm = (__u32) (IMM) }), \
174 ((struct bpf_insn) { \
175 .code = 0, /* zero is reserved opcode */ \
179 .imm = ((__u64) (IMM)) >> 32 })
181 /* pseudo BPF_LD_IMM64 insn used to refer to process-local map_fd */
182 #define BPF_LD_MAP_FD(DST, MAP_FD) \
183 BPF_LD_IMM64_RAW(DST, BPF_PSEUDO_MAP_FD, MAP_FD)
185 /* Short form of mov based on type, BPF_X: dst_reg = src_reg, BPF_K: dst_reg = imm32 */
187 #define BPF_MOV64_RAW(TYPE, DST, SRC, IMM) \
188 ((struct bpf_insn) { \
189 .code = BPF_ALU64 | BPF_MOV | BPF_SRC(TYPE), \
195 #define BPF_MOV32_RAW(TYPE, DST, SRC, IMM) \
196 ((struct bpf_insn) { \
197 .code = BPF_ALU | BPF_MOV | BPF_SRC(TYPE), \
203 /* Direct packet access, R0 = *(uint *) (skb->data + imm32) */
205 #define BPF_LD_ABS(SIZE, IMM) \
206 ((struct bpf_insn) { \
207 .code = BPF_LD | BPF_SIZE(SIZE) | BPF_ABS, \
213 /* Indirect packet access, R0 = *(uint *) (skb->data + src_reg + imm32) */
215 #define BPF_LD_IND(SIZE, SRC, IMM) \
216 ((struct bpf_insn) { \
217 .code = BPF_LD | BPF_SIZE(SIZE) | BPF_IND, \
223 /* Memory load, dst_reg = *(uint *) (src_reg + off16) */
225 #define BPF_LDX_MEM(SIZE, DST, SRC, OFF) \
226 ((struct bpf_insn) { \
227 .code = BPF_LDX | BPF_SIZE(SIZE) | BPF_MEM, \
233 /* Memory store, *(uint *) (dst_reg + off16) = src_reg */
235 #define BPF_STX_MEM(SIZE, DST, SRC, OFF) \
236 ((struct bpf_insn) { \
237 .code = BPF_STX | BPF_SIZE(SIZE) | BPF_MEM, \
243 /* Atomic memory add, *(uint *)(dst_reg + off16) += src_reg */
245 #define BPF_STX_XADD(SIZE, DST, SRC, OFF) \
246 ((struct bpf_insn) { \
247 .code = BPF_STX | BPF_SIZE(SIZE) | BPF_XADD, \
253 /* Memory store, *(uint *) (dst_reg + off16) = imm32 */
255 #define BPF_ST_MEM(SIZE, DST, OFF, IMM) \
256 ((struct bpf_insn) { \
257 .code = BPF_ST | BPF_SIZE(SIZE) | BPF_MEM, \
263 /* Conditional jumps against registers, if (dst_reg 'op' src_reg) goto pc + off16 */
265 #define BPF_JMP_REG(OP, DST, SRC, OFF) \
266 ((struct bpf_insn) { \
267 .code = BPF_JMP | BPF_OP(OP) | BPF_X, \
273 /* Conditional jumps against immediates, if (dst_reg 'op' imm32) goto pc + off16 */
275 #define BPF_JMP_IMM(OP, DST, IMM, OFF) \
276 ((struct bpf_insn) { \
277 .code = BPF_JMP | BPF_OP(OP) | BPF_K, \
283 /* Unconditional jumps, goto pc + off16 */
285 #define BPF_JMP_A(OFF) \
286 ((struct bpf_insn) { \
287 .code = BPF_JMP | BPF_JA, \
295 #define BPF_CALL_REL(TGT) \
296 ((struct bpf_insn) { \
297 .code = BPF_JMP | BPF_CALL, \
299 .src_reg = BPF_PSEUDO_CALL, \
305 #define BPF_CAST_CALL(x) \
306 ((u64 (*)(u64, u64, u64, u64, u64))(x))
308 #define BPF_EMIT_CALL(FUNC) \
309 ((struct bpf_insn) { \
310 .code = BPF_JMP | BPF_CALL, \
314 .imm = ((FUNC) - __bpf_call_base) })
316 /* Raw code statement block */
318 #define BPF_RAW_INSN(CODE, DST, SRC, OFF, IMM) \
319 ((struct bpf_insn) { \
328 #define BPF_EXIT_INSN() \
329 ((struct bpf_insn) { \
330 .code = BPF_JMP | BPF_EXIT, \
336 /* Internal classic blocks for direct assignment */
338 #define __BPF_STMT(CODE, K) \
339 ((struct sock_filter) BPF_STMT(CODE, K))
341 #define __BPF_JUMP(CODE, K, JT, JF) \
342 ((struct sock_filter) BPF_JUMP(CODE, K, JT, JF))
344 #define bytes_to_bpf_size(bytes) \
346 int bpf_size = -EINVAL; \
348 if (bytes == sizeof(u8)) \
350 else if (bytes == sizeof(u16)) \
352 else if (bytes == sizeof(u32)) \
354 else if (bytes == sizeof(u64)) \
360 #define bpf_size_to_bytes(bpf_size) \
362 int bytes = -EINVAL; \
364 if (bpf_size == BPF_B) \
365 bytes = sizeof(u8); \
366 else if (bpf_size == BPF_H) \
367 bytes = sizeof(u16); \
368 else if (bpf_size == BPF_W) \
369 bytes = sizeof(u32); \
370 else if (bpf_size == BPF_DW) \
371 bytes = sizeof(u64); \
376 #define BPF_SIZEOF(type) \
378 const int __size = bytes_to_bpf_size(sizeof(type)); \
379 BUILD_BUG_ON(__size < 0); \
383 #define BPF_FIELD_SIZEOF(type, field) \
385 const int __size = bytes_to_bpf_size(FIELD_SIZEOF(type, field)); \
386 BUILD_BUG_ON(__size < 0); \
390 #define BPF_LDST_BYTES(insn) \
392 const int __size = bpf_size_to_bytes(BPF_SIZE((insn)->code)); \
393 WARN_ON(__size < 0); \
397 #define __BPF_MAP_0(m, v, ...) v
398 #define __BPF_MAP_1(m, v, t, a, ...) m(t, a)
399 #define __BPF_MAP_2(m, v, t, a, ...) m(t, a), __BPF_MAP_1(m, v, __VA_ARGS__)
400 #define __BPF_MAP_3(m, v, t, a, ...) m(t, a), __BPF_MAP_2(m, v, __VA_ARGS__)
401 #define __BPF_MAP_4(m, v, t, a, ...) m(t, a), __BPF_MAP_3(m, v, __VA_ARGS__)
402 #define __BPF_MAP_5(m, v, t, a, ...) m(t, a), __BPF_MAP_4(m, v, __VA_ARGS__)
404 #define __BPF_REG_0(...) __BPF_PAD(5)
405 #define __BPF_REG_1(...) __BPF_MAP(1, __VA_ARGS__), __BPF_PAD(4)
406 #define __BPF_REG_2(...) __BPF_MAP(2, __VA_ARGS__), __BPF_PAD(3)
407 #define __BPF_REG_3(...) __BPF_MAP(3, __VA_ARGS__), __BPF_PAD(2)
408 #define __BPF_REG_4(...) __BPF_MAP(4, __VA_ARGS__), __BPF_PAD(1)
409 #define __BPF_REG_5(...) __BPF_MAP(5, __VA_ARGS__)
411 #define __BPF_MAP(n, ...) __BPF_MAP_##n(__VA_ARGS__)
412 #define __BPF_REG(n, ...) __BPF_REG_##n(__VA_ARGS__)
414 #define __BPF_CAST(t, a) \
417 typeof(__builtin_choose_expr(sizeof(t) == sizeof(unsigned long), \
418 (unsigned long)0, (t)0))) a
422 #define __BPF_DECL_ARGS(t, a) t a
423 #define __BPF_DECL_REGS(t, a) u64 a
425 #define __BPF_PAD(n) \
426 __BPF_MAP(n, __BPF_DECL_ARGS, __BPF_N, u64, __ur_1, u64, __ur_2, \
427 u64, __ur_3, u64, __ur_4, u64, __ur_5)
429 #define BPF_CALL_x(x, name, ...) \
430 static __always_inline \
431 u64 ____##name(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__)); \
432 u64 name(__BPF_REG(x, __BPF_DECL_REGS, __BPF_N, __VA_ARGS__)); \
433 u64 name(__BPF_REG(x, __BPF_DECL_REGS, __BPF_N, __VA_ARGS__)) \
435 return ____##name(__BPF_MAP(x,__BPF_CAST,__BPF_N,__VA_ARGS__));\
437 static __always_inline \
438 u64 ____##name(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__))
440 #define BPF_CALL_0(name, ...) BPF_CALL_x(0, name, __VA_ARGS__)
441 #define BPF_CALL_1(name, ...) BPF_CALL_x(1, name, __VA_ARGS__)
442 #define BPF_CALL_2(name, ...) BPF_CALL_x(2, name, __VA_ARGS__)
443 #define BPF_CALL_3(name, ...) BPF_CALL_x(3, name, __VA_ARGS__)
444 #define BPF_CALL_4(name, ...) BPF_CALL_x(4, name, __VA_ARGS__)
445 #define BPF_CALL_5(name, ...) BPF_CALL_x(5, name, __VA_ARGS__)
447 #define bpf_ctx_range(TYPE, MEMBER) \
448 offsetof(TYPE, MEMBER) ... offsetofend(TYPE, MEMBER) - 1
449 #define bpf_ctx_range_till(TYPE, MEMBER1, MEMBER2) \
450 offsetof(TYPE, MEMBER1) ... offsetofend(TYPE, MEMBER2) - 1
452 #define bpf_target_off(TYPE, MEMBER, SIZE, PTR_SIZE) \
454 BUILD_BUG_ON(FIELD_SIZEOF(TYPE, MEMBER) != (SIZE)); \
455 *(PTR_SIZE) = (SIZE); \
456 offsetof(TYPE, MEMBER); \
460 /* A struct sock_filter is architecture independent. */
461 struct compat_sock_fprog {
463 compat_uptr_t filter; /* struct sock_filter * */
467 struct sock_fprog_kern {
469 struct sock_filter *filter;
472 struct bpf_binary_header {
474 /* Some arches need word alignment for their instructions */
475 u8 image[] __aligned(4);
479 u16 pages; /* Number of allocated pages */
480 u16 jited:1, /* Is our filter JIT'ed? */
481 jit_requested:1,/* archs need to JIT the prog */
482 undo_set_mem:1, /* Passed set_memory_ro() checkpoint */
483 gpl_compatible:1, /* Is filter GPL compatible? */
484 cb_access:1, /* Is control block accessed? */
485 dst_needed:1, /* Do we need dst entry? */
486 blinded:1, /* Was blinded */
487 is_func:1, /* program is a bpf function */
488 kprobe_override:1, /* Do we override a kprobe? */
489 has_callchain_buf:1; /* callchain buffer allocated? */
490 enum bpf_prog_type type; /* Type of BPF program */
491 enum bpf_attach_type expected_attach_type; /* For some prog types */
492 u32 len; /* Number of filter blocks */
493 u32 jited_len; /* Size of jited insns in bytes */
494 u8 tag[BPF_TAG_SIZE];
495 struct bpf_prog_aux *aux; /* Auxiliary fields */
496 struct sock_fprog_kern *orig_prog; /* Original BPF program */
497 unsigned int (*bpf_func)(const void *ctx,
498 const struct bpf_insn *insn);
499 /* Instructions for interpreter */
501 struct sock_filter insns[0];
502 struct bpf_insn insnsi[0];
509 struct bpf_prog *prog;
512 #define BPF_PROG_RUN(filter, ctx) (*(filter)->bpf_func)(ctx, (filter)->insnsi)
514 #define BPF_SKB_CB_LEN QDISC_CB_PRIV_LEN
516 struct bpf_skb_data_end {
517 struct qdisc_skb_cb qdisc_cb;
531 struct scatterlist sg_data[MAX_SKB_FRAGS];
532 bool sg_copy[MAX_SKB_FRAGS];
534 struct sock *sk_redir;
537 struct list_head list;
540 /* Compute the linear packet data range [data, data_end) which
541 * will be accessed by various program types (cls_bpf, act_bpf,
542 * lwt, ...). Subsystems allowing direct data access must (!)
543 * ensure that cb[] area can be written to when BPF program is
544 * invoked (otherwise cb[] save/restore is necessary).
546 static inline void bpf_compute_data_pointers(struct sk_buff *skb)
548 struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb;
550 BUILD_BUG_ON(sizeof(*cb) > FIELD_SIZEOF(struct sk_buff, cb));
551 cb->data_meta = skb->data - skb_metadata_len(skb);
552 cb->data_end = skb->data + skb_headlen(skb);
555 static inline u8 *bpf_skb_cb(struct sk_buff *skb)
557 /* eBPF programs may read/write skb->cb[] area to transfer meta
558 * data between tail calls. Since this also needs to work with
559 * tc, that scratch memory is mapped to qdisc_skb_cb's data area.
561 * In some socket filter cases, the cb unfortunately needs to be
562 * saved/restored so that protocol specific skb->cb[] data won't
563 * be lost. In any case, due to unpriviledged eBPF programs
564 * attached to sockets, we need to clear the bpf_skb_cb() area
565 * to not leak previous contents to user space.
567 BUILD_BUG_ON(FIELD_SIZEOF(struct __sk_buff, cb) != BPF_SKB_CB_LEN);
568 BUILD_BUG_ON(FIELD_SIZEOF(struct __sk_buff, cb) !=
569 FIELD_SIZEOF(struct qdisc_skb_cb, data));
571 return qdisc_skb_cb(skb)->data;
574 static inline u32 bpf_prog_run_save_cb(const struct bpf_prog *prog,
577 u8 *cb_data = bpf_skb_cb(skb);
578 u8 cb_saved[BPF_SKB_CB_LEN];
581 if (unlikely(prog->cb_access)) {
582 memcpy(cb_saved, cb_data, sizeof(cb_saved));
583 memset(cb_data, 0, sizeof(cb_saved));
586 res = BPF_PROG_RUN(prog, skb);
588 if (unlikely(prog->cb_access))
589 memcpy(cb_data, cb_saved, sizeof(cb_saved));
594 static inline u32 bpf_prog_run_clear_cb(const struct bpf_prog *prog,
597 u8 *cb_data = bpf_skb_cb(skb);
599 if (unlikely(prog->cb_access))
600 memset(cb_data, 0, BPF_SKB_CB_LEN);
602 return BPF_PROG_RUN(prog, skb);
605 static __always_inline u32 bpf_prog_run_xdp(const struct bpf_prog *prog,
606 struct xdp_buff *xdp)
608 /* Caller needs to hold rcu_read_lock() (!), otherwise program
609 * can be released while still running, or map elements could be
610 * freed early while still having concurrent users. XDP fastpath
611 * already takes rcu_read_lock() when fetching the program, so
612 * it's not necessary here anymore.
614 return BPF_PROG_RUN(prog, xdp);
617 static inline u32 bpf_prog_insn_size(const struct bpf_prog *prog)
619 return prog->len * sizeof(struct bpf_insn);
622 static inline u32 bpf_prog_tag_scratch_size(const struct bpf_prog *prog)
624 return round_up(bpf_prog_insn_size(prog) +
625 sizeof(__be64) + 1, SHA_MESSAGE_BYTES);
628 static inline unsigned int bpf_prog_size(unsigned int proglen)
630 return max(sizeof(struct bpf_prog),
631 offsetof(struct bpf_prog, insns[proglen]));
634 static inline bool bpf_prog_was_classic(const struct bpf_prog *prog)
636 /* When classic BPF programs have been loaded and the arch
637 * does not have a classic BPF JIT (anymore), they have been
638 * converted via bpf_migrate_filter() to eBPF and thus always
639 * have an unspec program type.
641 return prog->type == BPF_PROG_TYPE_UNSPEC;
644 static inline u32 bpf_ctx_off_adjust_machine(u32 size)
646 const u32 size_machine = sizeof(unsigned long);
648 if (size > size_machine && size % size_machine == 0)
654 static inline bool bpf_ctx_narrow_align_ok(u32 off, u32 size_access,
657 size_default = bpf_ctx_off_adjust_machine(size_default);
658 size_access = bpf_ctx_off_adjust_machine(size_access);
660 #ifdef __LITTLE_ENDIAN
661 return (off & (size_default - 1)) == 0;
663 return (off & (size_default - 1)) + size_access == size_default;
668 bpf_ctx_narrow_access_ok(u32 off, u32 size, u32 size_default)
670 return bpf_ctx_narrow_align_ok(off, size, size_default) &&
671 size <= size_default && (size & (size - 1)) == 0;
674 #define bpf_classic_proglen(fprog) (fprog->len * sizeof(fprog->filter[0]))
676 static inline void bpf_prog_lock_ro(struct bpf_prog *fp)
678 fp->undo_set_mem = 1;
679 set_memory_ro((unsigned long)fp, fp->pages);
682 static inline void bpf_prog_unlock_ro(struct bpf_prog *fp)
684 if (fp->undo_set_mem)
685 set_memory_rw((unsigned long)fp, fp->pages);
688 static inline void bpf_jit_binary_lock_ro(struct bpf_binary_header *hdr)
690 set_memory_ro((unsigned long)hdr, hdr->pages);
693 static inline void bpf_jit_binary_unlock_ro(struct bpf_binary_header *hdr)
695 set_memory_rw((unsigned long)hdr, hdr->pages);
698 static inline struct bpf_binary_header *
699 bpf_jit_binary_hdr(const struct bpf_prog *fp)
701 unsigned long real_start = (unsigned long)fp->bpf_func;
702 unsigned long addr = real_start & PAGE_MASK;
707 int sk_filter_trim_cap(struct sock *sk, struct sk_buff *skb, unsigned int cap);
708 static inline int sk_filter(struct sock *sk, struct sk_buff *skb)
710 return sk_filter_trim_cap(sk, skb, 1);
713 struct bpf_prog *bpf_prog_select_runtime(struct bpf_prog *fp, int *err);
714 void bpf_prog_free(struct bpf_prog *fp);
716 bool bpf_opcode_in_insntable(u8 code);
718 struct bpf_prog *bpf_prog_alloc(unsigned int size, gfp_t gfp_extra_flags);
719 struct bpf_prog *bpf_prog_realloc(struct bpf_prog *fp_old, unsigned int size,
720 gfp_t gfp_extra_flags);
721 void __bpf_prog_free(struct bpf_prog *fp);
723 static inline void bpf_prog_unlock_free(struct bpf_prog *fp)
725 bpf_prog_unlock_ro(fp);
729 typedef int (*bpf_aux_classic_check_t)(struct sock_filter *filter,
732 int bpf_prog_create(struct bpf_prog **pfp, struct sock_fprog_kern *fprog);
733 int bpf_prog_create_from_user(struct bpf_prog **pfp, struct sock_fprog *fprog,
734 bpf_aux_classic_check_t trans, bool save_orig);
735 void bpf_prog_destroy(struct bpf_prog *fp);
737 int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk);
738 int sk_attach_bpf(u32 ufd, struct sock *sk);
739 int sk_reuseport_attach_filter(struct sock_fprog *fprog, struct sock *sk);
740 int sk_reuseport_attach_bpf(u32 ufd, struct sock *sk);
741 int sk_detach_filter(struct sock *sk);
742 int sk_get_filter(struct sock *sk, struct sock_filter __user *filter,
745 bool sk_filter_charge(struct sock *sk, struct sk_filter *fp);
746 void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp);
748 u64 __bpf_call_base(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5);
749 #define __bpf_call_base_args \
750 ((u64 (*)(u64, u64, u64, u64, u64, const struct bpf_insn *)) \
753 struct bpf_prog *bpf_int_jit_compile(struct bpf_prog *prog);
754 void bpf_jit_compile(struct bpf_prog *prog);
755 bool bpf_helper_changes_pkt_data(void *func);
757 static inline bool bpf_dump_raw_ok(void)
759 /* Reconstruction of call-sites is dependent on kallsyms,
760 * thus make dump the same restriction.
762 return kallsyms_show_value() == 1;
765 struct bpf_prog *bpf_patch_insn_single(struct bpf_prog *prog, u32 off,
766 const struct bpf_insn *patch, u32 len);
768 static inline int __xdp_generic_ok_fwd_dev(struct sk_buff *skb,
769 struct net_device *fwd)
773 if (unlikely(!(fwd->flags & IFF_UP)))
776 len = fwd->mtu + fwd->hard_header_len + VLAN_HLEN;
783 /* The pair of xdp_do_redirect and xdp_do_flush_map MUST be called in the
784 * same cpu context. Further for best results no more than a single map
785 * for the do_redirect/do_flush pair should be used. This limitation is
786 * because we only track one map and force a flush when the map changes.
787 * This does not appear to be a real limitation for existing software.
789 int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb,
790 struct xdp_buff *xdp, struct bpf_prog *prog);
791 int xdp_do_redirect(struct net_device *dev,
792 struct xdp_buff *xdp,
793 struct bpf_prog *prog);
794 void xdp_do_flush_map(void);
796 void bpf_warn_invalid_xdp_action(u32 act);
798 struct sock *do_sk_redirect_map(struct sk_buff *skb);
799 struct sock *do_msg_redirect_map(struct sk_msg_buff *md);
801 #ifdef CONFIG_BPF_JIT
802 extern int bpf_jit_enable;
803 extern int bpf_jit_harden;
804 extern int bpf_jit_kallsyms;
806 typedef void (*bpf_jit_fill_hole_t)(void *area, unsigned int size);
808 struct bpf_binary_header *
809 bpf_jit_binary_alloc(unsigned int proglen, u8 **image_ptr,
810 unsigned int alignment,
811 bpf_jit_fill_hole_t bpf_fill_ill_insns);
812 void bpf_jit_binary_free(struct bpf_binary_header *hdr);
814 void bpf_jit_free(struct bpf_prog *fp);
816 struct bpf_prog *bpf_jit_blind_constants(struct bpf_prog *fp);
817 void bpf_jit_prog_release_other(struct bpf_prog *fp, struct bpf_prog *fp_other);
819 static inline void bpf_jit_dump(unsigned int flen, unsigned int proglen,
820 u32 pass, void *image)
822 pr_err("flen=%u proglen=%u pass=%u image=%pK from=%s pid=%d\n", flen,
823 proglen, pass, image, current->comm, task_pid_nr(current));
826 print_hex_dump(KERN_ERR, "JIT code: ", DUMP_PREFIX_OFFSET,
827 16, 1, image, proglen, false);
830 static inline bool bpf_jit_is_ebpf(void)
832 # ifdef CONFIG_HAVE_EBPF_JIT
839 static inline bool ebpf_jit_enabled(void)
841 return bpf_jit_enable && bpf_jit_is_ebpf();
844 static inline bool bpf_prog_ebpf_jited(const struct bpf_prog *fp)
846 return fp->jited && bpf_jit_is_ebpf();
849 static inline bool bpf_jit_blinding_enabled(struct bpf_prog *prog)
851 /* These are the prerequisites, should someone ever have the
852 * idea to call blinding outside of them, we make sure to
855 if (!bpf_jit_is_ebpf())
857 if (!prog->jit_requested)
861 if (bpf_jit_harden == 1 && capable(CAP_SYS_ADMIN))
867 static inline bool bpf_jit_kallsyms_enabled(void)
869 /* There are a couple of corner cases where kallsyms should
870 * not be enabled f.e. on hardening.
874 if (!bpf_jit_kallsyms)
876 if (bpf_jit_kallsyms == 1)
882 const char *__bpf_address_lookup(unsigned long addr, unsigned long *size,
883 unsigned long *off, char *sym);
884 bool is_bpf_text_address(unsigned long addr);
885 int bpf_get_kallsym(unsigned int symnum, unsigned long *value, char *type,
888 static inline const char *
889 bpf_address_lookup(unsigned long addr, unsigned long *size,
890 unsigned long *off, char **modname, char *sym)
892 const char *ret = __bpf_address_lookup(addr, size, off, sym);
899 void bpf_prog_kallsyms_add(struct bpf_prog *fp);
900 void bpf_prog_kallsyms_del(struct bpf_prog *fp);
902 #else /* CONFIG_BPF_JIT */
904 static inline bool ebpf_jit_enabled(void)
909 static inline bool bpf_prog_ebpf_jited(const struct bpf_prog *fp)
914 static inline void bpf_jit_free(struct bpf_prog *fp)
916 bpf_prog_unlock_free(fp);
919 static inline bool bpf_jit_kallsyms_enabled(void)
924 static inline const char *
925 __bpf_address_lookup(unsigned long addr, unsigned long *size,
926 unsigned long *off, char *sym)
931 static inline bool is_bpf_text_address(unsigned long addr)
936 static inline int bpf_get_kallsym(unsigned int symnum, unsigned long *value,
937 char *type, char *sym)
942 static inline const char *
943 bpf_address_lookup(unsigned long addr, unsigned long *size,
944 unsigned long *off, char **modname, char *sym)
949 static inline void bpf_prog_kallsyms_add(struct bpf_prog *fp)
953 static inline void bpf_prog_kallsyms_del(struct bpf_prog *fp)
956 #endif /* CONFIG_BPF_JIT */
958 void bpf_prog_kallsyms_del_subprogs(struct bpf_prog *fp);
959 void bpf_prog_kallsyms_del_all(struct bpf_prog *fp);
961 #define BPF_ANC BIT(15)
963 static inline bool bpf_needs_clear_a(const struct sock_filter *first)
965 switch (first->code) {
966 case BPF_RET | BPF_K:
967 case BPF_LD | BPF_W | BPF_LEN:
970 case BPF_LD | BPF_W | BPF_ABS:
971 case BPF_LD | BPF_H | BPF_ABS:
972 case BPF_LD | BPF_B | BPF_ABS:
973 if (first->k == SKF_AD_OFF + SKF_AD_ALU_XOR_X)
982 static inline u16 bpf_anc_helper(const struct sock_filter *ftest)
984 BUG_ON(ftest->code & BPF_ANC);
986 switch (ftest->code) {
987 case BPF_LD | BPF_W | BPF_ABS:
988 case BPF_LD | BPF_H | BPF_ABS:
989 case BPF_LD | BPF_B | BPF_ABS:
990 #define BPF_ANCILLARY(CODE) case SKF_AD_OFF + SKF_AD_##CODE: \
991 return BPF_ANC | SKF_AD_##CODE
993 BPF_ANCILLARY(PROTOCOL);
994 BPF_ANCILLARY(PKTTYPE);
995 BPF_ANCILLARY(IFINDEX);
996 BPF_ANCILLARY(NLATTR);
997 BPF_ANCILLARY(NLATTR_NEST);
999 BPF_ANCILLARY(QUEUE);
1000 BPF_ANCILLARY(HATYPE);
1001 BPF_ANCILLARY(RXHASH);
1003 BPF_ANCILLARY(ALU_XOR_X);
1004 BPF_ANCILLARY(VLAN_TAG);
1005 BPF_ANCILLARY(VLAN_TAG_PRESENT);
1006 BPF_ANCILLARY(PAY_OFFSET);
1007 BPF_ANCILLARY(RANDOM);
1008 BPF_ANCILLARY(VLAN_TPID);
1016 void *bpf_internal_load_pointer_neg_helper(const struct sk_buff *skb,
1017 int k, unsigned int size);
1019 static inline void *bpf_load_pointer(const struct sk_buff *skb, int k,
1020 unsigned int size, void *buffer)
1023 return skb_header_pointer(skb, k, size, buffer);
1025 return bpf_internal_load_pointer_neg_helper(skb, k, size);
1028 static inline int bpf_tell_extensions(void)
1033 struct bpf_sock_addr_kern {
1035 struct sockaddr *uaddr;
1036 /* Temporary "register" to make indirect stores to nested structures
1037 * defined above. We need three registers to make such a store, but
1038 * only two (src and dst) are available at convert_ctx_access time
1041 void *t_ctx; /* Attach type specific context. */
1044 struct bpf_sock_ops_kern {
1053 u64 temp; /* temp and everything after is not
1054 * initialized to 0 before calling
1055 * the BPF program. New fields that
1056 * should be initialized to 0 should
1057 * be inserted before temp.
1058 * temp is scratch storage used by
1059 * sock_ops_convert_ctx_access
1060 * as temporary storage of a register.
1064 #endif /* __LINUX_FILTER_H__ */