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 struct ctl_table_header;
39 /* ArgX, context and stack frame pointer register positions. Note,
40 * Arg1, Arg2, Arg3, etc are used as argument mappings of function
41 * calls in BPF_CALL instruction.
43 #define BPF_REG_ARG1 BPF_REG_1
44 #define BPF_REG_ARG2 BPF_REG_2
45 #define BPF_REG_ARG3 BPF_REG_3
46 #define BPF_REG_ARG4 BPF_REG_4
47 #define BPF_REG_ARG5 BPF_REG_5
48 #define BPF_REG_CTX BPF_REG_6
49 #define BPF_REG_FP BPF_REG_10
51 /* Additional register mappings for converted user programs. */
52 #define BPF_REG_A BPF_REG_0
53 #define BPF_REG_X BPF_REG_7
54 #define BPF_REG_TMP BPF_REG_2 /* scratch reg */
55 #define BPF_REG_D BPF_REG_8 /* data, callee-saved */
56 #define BPF_REG_H BPF_REG_9 /* hlen, callee-saved */
58 /* Kernel hidden auxiliary/helper register. */
59 #define BPF_REG_AX MAX_BPF_REG
60 #define MAX_BPF_EXT_REG (MAX_BPF_REG + 1)
61 #define MAX_BPF_JIT_REG MAX_BPF_EXT_REG
63 /* unused opcode to mark special call to bpf_tail_call() helper */
64 #define BPF_TAIL_CALL 0xf0
66 /* unused opcode to mark call to interpreter with arguments */
67 #define BPF_CALL_ARGS 0xe0
69 /* As per nm, we expose JITed images as text (code) section for
70 * kallsyms. That way, tools like perf can find it to match
73 #define BPF_SYM_ELF_TYPE 't'
75 /* BPF program can access up to 512 bytes of stack space. */
76 #define MAX_BPF_STACK 512
78 /* Helper macros for filter block array initializers. */
80 /* ALU ops on registers, bpf_add|sub|...: dst_reg += src_reg */
82 #define BPF_ALU64_REG(OP, DST, SRC) \
83 ((struct bpf_insn) { \
84 .code = BPF_ALU64 | BPF_OP(OP) | BPF_X, \
90 #define BPF_ALU32_REG(OP, DST, SRC) \
91 ((struct bpf_insn) { \
92 .code = BPF_ALU | BPF_OP(OP) | BPF_X, \
98 /* ALU ops on immediates, bpf_add|sub|...: dst_reg += imm32 */
100 #define BPF_ALU64_IMM(OP, DST, IMM) \
101 ((struct bpf_insn) { \
102 .code = BPF_ALU64 | BPF_OP(OP) | BPF_K, \
108 #define BPF_ALU32_IMM(OP, DST, IMM) \
109 ((struct bpf_insn) { \
110 .code = BPF_ALU | BPF_OP(OP) | BPF_K, \
116 /* Endianess conversion, cpu_to_{l,b}e(), {l,b}e_to_cpu() */
118 #define BPF_ENDIAN(TYPE, DST, LEN) \
119 ((struct bpf_insn) { \
120 .code = BPF_ALU | BPF_END | BPF_SRC(TYPE), \
126 /* Short form of mov, dst_reg = src_reg */
128 #define BPF_MOV64_REG(DST, SRC) \
129 ((struct bpf_insn) { \
130 .code = BPF_ALU64 | BPF_MOV | BPF_X, \
136 #define BPF_MOV32_REG(DST, SRC) \
137 ((struct bpf_insn) { \
138 .code = BPF_ALU | BPF_MOV | BPF_X, \
144 /* Short form of mov, dst_reg = imm32 */
146 #define BPF_MOV64_IMM(DST, IMM) \
147 ((struct bpf_insn) { \
148 .code = BPF_ALU64 | BPF_MOV | BPF_K, \
154 #define BPF_MOV32_IMM(DST, IMM) \
155 ((struct bpf_insn) { \
156 .code = BPF_ALU | BPF_MOV | BPF_K, \
162 /* BPF_LD_IMM64 macro encodes single 'load 64-bit immediate' insn */
163 #define BPF_LD_IMM64(DST, IMM) \
164 BPF_LD_IMM64_RAW(DST, 0, IMM)
166 #define BPF_LD_IMM64_RAW(DST, SRC, IMM) \
167 ((struct bpf_insn) { \
168 .code = BPF_LD | BPF_DW | BPF_IMM, \
172 .imm = (__u32) (IMM) }), \
173 ((struct bpf_insn) { \
174 .code = 0, /* zero is reserved opcode */ \
178 .imm = ((__u64) (IMM)) >> 32 })
180 /* pseudo BPF_LD_IMM64 insn used to refer to process-local map_fd */
181 #define BPF_LD_MAP_FD(DST, MAP_FD) \
182 BPF_LD_IMM64_RAW(DST, BPF_PSEUDO_MAP_FD, MAP_FD)
184 /* Short form of mov based on type, BPF_X: dst_reg = src_reg, BPF_K: dst_reg = imm32 */
186 #define BPF_MOV64_RAW(TYPE, DST, SRC, IMM) \
187 ((struct bpf_insn) { \
188 .code = BPF_ALU64 | BPF_MOV | BPF_SRC(TYPE), \
194 #define BPF_MOV32_RAW(TYPE, DST, SRC, IMM) \
195 ((struct bpf_insn) { \
196 .code = BPF_ALU | BPF_MOV | BPF_SRC(TYPE), \
202 /* Direct packet access, R0 = *(uint *) (skb->data + imm32) */
204 #define BPF_LD_ABS(SIZE, IMM) \
205 ((struct bpf_insn) { \
206 .code = BPF_LD | BPF_SIZE(SIZE) | BPF_ABS, \
212 /* Indirect packet access, R0 = *(uint *) (skb->data + src_reg + imm32) */
214 #define BPF_LD_IND(SIZE, SRC, IMM) \
215 ((struct bpf_insn) { \
216 .code = BPF_LD | BPF_SIZE(SIZE) | BPF_IND, \
222 /* Memory load, dst_reg = *(uint *) (src_reg + off16) */
224 #define BPF_LDX_MEM(SIZE, DST, SRC, OFF) \
225 ((struct bpf_insn) { \
226 .code = BPF_LDX | BPF_SIZE(SIZE) | BPF_MEM, \
232 /* Memory store, *(uint *) (dst_reg + off16) = src_reg */
234 #define BPF_STX_MEM(SIZE, DST, SRC, OFF) \
235 ((struct bpf_insn) { \
236 .code = BPF_STX | BPF_SIZE(SIZE) | BPF_MEM, \
242 /* Atomic memory add, *(uint *)(dst_reg + off16) += src_reg */
244 #define BPF_STX_XADD(SIZE, DST, SRC, OFF) \
245 ((struct bpf_insn) { \
246 .code = BPF_STX | BPF_SIZE(SIZE) | BPF_XADD, \
252 /* Memory store, *(uint *) (dst_reg + off16) = imm32 */
254 #define BPF_ST_MEM(SIZE, DST, OFF, IMM) \
255 ((struct bpf_insn) { \
256 .code = BPF_ST | BPF_SIZE(SIZE) | BPF_MEM, \
262 /* Conditional jumps against registers, if (dst_reg 'op' src_reg) goto pc + off16 */
264 #define BPF_JMP_REG(OP, DST, SRC, OFF) \
265 ((struct bpf_insn) { \
266 .code = BPF_JMP | BPF_OP(OP) | BPF_X, \
272 /* Conditional jumps against immediates, if (dst_reg 'op' imm32) goto pc + off16 */
274 #define BPF_JMP_IMM(OP, DST, IMM, OFF) \
275 ((struct bpf_insn) { \
276 .code = BPF_JMP | BPF_OP(OP) | BPF_K, \
282 /* Like BPF_JMP_REG, but with 32-bit wide operands for comparison. */
284 #define BPF_JMP32_REG(OP, DST, SRC, OFF) \
285 ((struct bpf_insn) { \
286 .code = BPF_JMP32 | BPF_OP(OP) | BPF_X, \
292 /* Like BPF_JMP_IMM, but with 32-bit wide operands for comparison. */
294 #define BPF_JMP32_IMM(OP, DST, IMM, OFF) \
295 ((struct bpf_insn) { \
296 .code = BPF_JMP32 | BPF_OP(OP) | BPF_K, \
302 /* Unconditional jumps, goto pc + off16 */
304 #define BPF_JMP_A(OFF) \
305 ((struct bpf_insn) { \
306 .code = BPF_JMP | BPF_JA, \
314 #define BPF_CALL_REL(TGT) \
315 ((struct bpf_insn) { \
316 .code = BPF_JMP | BPF_CALL, \
318 .src_reg = BPF_PSEUDO_CALL, \
324 #define BPF_CAST_CALL(x) \
325 ((u64 (*)(u64, u64, u64, u64, u64))(x))
327 #define BPF_EMIT_CALL(FUNC) \
328 ((struct bpf_insn) { \
329 .code = BPF_JMP | BPF_CALL, \
333 .imm = ((FUNC) - __bpf_call_base) })
335 /* Raw code statement block */
337 #define BPF_RAW_INSN(CODE, DST, SRC, OFF, IMM) \
338 ((struct bpf_insn) { \
347 #define BPF_EXIT_INSN() \
348 ((struct bpf_insn) { \
349 .code = BPF_JMP | BPF_EXIT, \
355 /* Internal classic blocks for direct assignment */
357 #define __BPF_STMT(CODE, K) \
358 ((struct sock_filter) BPF_STMT(CODE, K))
360 #define __BPF_JUMP(CODE, K, JT, JF) \
361 ((struct sock_filter) BPF_JUMP(CODE, K, JT, JF))
363 #define bytes_to_bpf_size(bytes) \
365 int bpf_size = -EINVAL; \
367 if (bytes == sizeof(u8)) \
369 else if (bytes == sizeof(u16)) \
371 else if (bytes == sizeof(u32)) \
373 else if (bytes == sizeof(u64)) \
379 #define bpf_size_to_bytes(bpf_size) \
381 int bytes = -EINVAL; \
383 if (bpf_size == BPF_B) \
384 bytes = sizeof(u8); \
385 else if (bpf_size == BPF_H) \
386 bytes = sizeof(u16); \
387 else if (bpf_size == BPF_W) \
388 bytes = sizeof(u32); \
389 else if (bpf_size == BPF_DW) \
390 bytes = sizeof(u64); \
395 #define BPF_SIZEOF(type) \
397 const int __size = bytes_to_bpf_size(sizeof(type)); \
398 BUILD_BUG_ON(__size < 0); \
402 #define BPF_FIELD_SIZEOF(type, field) \
404 const int __size = bytes_to_bpf_size(FIELD_SIZEOF(type, field)); \
405 BUILD_BUG_ON(__size < 0); \
409 #define BPF_LDST_BYTES(insn) \
411 const int __size = bpf_size_to_bytes(BPF_SIZE((insn)->code)); \
412 WARN_ON(__size < 0); \
416 #define __BPF_MAP_0(m, v, ...) v
417 #define __BPF_MAP_1(m, v, t, a, ...) m(t, a)
418 #define __BPF_MAP_2(m, v, t, a, ...) m(t, a), __BPF_MAP_1(m, v, __VA_ARGS__)
419 #define __BPF_MAP_3(m, v, t, a, ...) m(t, a), __BPF_MAP_2(m, v, __VA_ARGS__)
420 #define __BPF_MAP_4(m, v, t, a, ...) m(t, a), __BPF_MAP_3(m, v, __VA_ARGS__)
421 #define __BPF_MAP_5(m, v, t, a, ...) m(t, a), __BPF_MAP_4(m, v, __VA_ARGS__)
423 #define __BPF_REG_0(...) __BPF_PAD(5)
424 #define __BPF_REG_1(...) __BPF_MAP(1, __VA_ARGS__), __BPF_PAD(4)
425 #define __BPF_REG_2(...) __BPF_MAP(2, __VA_ARGS__), __BPF_PAD(3)
426 #define __BPF_REG_3(...) __BPF_MAP(3, __VA_ARGS__), __BPF_PAD(2)
427 #define __BPF_REG_4(...) __BPF_MAP(4, __VA_ARGS__), __BPF_PAD(1)
428 #define __BPF_REG_5(...) __BPF_MAP(5, __VA_ARGS__)
430 #define __BPF_MAP(n, ...) __BPF_MAP_##n(__VA_ARGS__)
431 #define __BPF_REG(n, ...) __BPF_REG_##n(__VA_ARGS__)
433 #define __BPF_CAST(t, a) \
436 typeof(__builtin_choose_expr(sizeof(t) == sizeof(unsigned long), \
437 (unsigned long)0, (t)0))) a
441 #define __BPF_DECL_ARGS(t, a) t a
442 #define __BPF_DECL_REGS(t, a) u64 a
444 #define __BPF_PAD(n) \
445 __BPF_MAP(n, __BPF_DECL_ARGS, __BPF_N, u64, __ur_1, u64, __ur_2, \
446 u64, __ur_3, u64, __ur_4, u64, __ur_5)
448 #define BPF_CALL_x(x, name, ...) \
449 static __always_inline \
450 u64 ____##name(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__)); \
451 u64 name(__BPF_REG(x, __BPF_DECL_REGS, __BPF_N, __VA_ARGS__)); \
452 u64 name(__BPF_REG(x, __BPF_DECL_REGS, __BPF_N, __VA_ARGS__)) \
454 return ____##name(__BPF_MAP(x,__BPF_CAST,__BPF_N,__VA_ARGS__));\
456 static __always_inline \
457 u64 ____##name(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__))
459 #define BPF_CALL_0(name, ...) BPF_CALL_x(0, name, __VA_ARGS__)
460 #define BPF_CALL_1(name, ...) BPF_CALL_x(1, name, __VA_ARGS__)
461 #define BPF_CALL_2(name, ...) BPF_CALL_x(2, name, __VA_ARGS__)
462 #define BPF_CALL_3(name, ...) BPF_CALL_x(3, name, __VA_ARGS__)
463 #define BPF_CALL_4(name, ...) BPF_CALL_x(4, name, __VA_ARGS__)
464 #define BPF_CALL_5(name, ...) BPF_CALL_x(5, name, __VA_ARGS__)
466 #define bpf_ctx_range(TYPE, MEMBER) \
467 offsetof(TYPE, MEMBER) ... offsetofend(TYPE, MEMBER) - 1
468 #define bpf_ctx_range_till(TYPE, MEMBER1, MEMBER2) \
469 offsetof(TYPE, MEMBER1) ... offsetofend(TYPE, MEMBER2) - 1
470 #if BITS_PER_LONG == 64
471 # define bpf_ctx_range_ptr(TYPE, MEMBER) \
472 offsetof(TYPE, MEMBER) ... offsetofend(TYPE, MEMBER) - 1
474 # define bpf_ctx_range_ptr(TYPE, MEMBER) \
475 offsetof(TYPE, MEMBER) ... offsetof(TYPE, MEMBER) + 8 - 1
476 #endif /* BITS_PER_LONG == 64 */
478 #define bpf_target_off(TYPE, MEMBER, SIZE, PTR_SIZE) \
480 BUILD_BUG_ON(FIELD_SIZEOF(TYPE, MEMBER) != (SIZE)); \
481 *(PTR_SIZE) = (SIZE); \
482 offsetof(TYPE, MEMBER); \
486 /* A struct sock_filter is architecture independent. */
487 struct compat_sock_fprog {
489 compat_uptr_t filter; /* struct sock_filter * */
493 struct sock_fprog_kern {
495 struct sock_filter *filter;
498 struct bpf_binary_header {
500 /* Some arches need word alignment for their instructions */
501 u8 image[] __aligned(4);
505 u16 pages; /* Number of allocated pages */
506 u16 jited:1, /* Is our filter JIT'ed? */
507 jit_requested:1,/* archs need to JIT the prog */
508 undo_set_mem:1, /* Passed set_memory_ro() checkpoint */
509 gpl_compatible:1, /* Is filter GPL compatible? */
510 cb_access:1, /* Is control block accessed? */
511 dst_needed:1, /* Do we need dst entry? */
512 blinded:1, /* Was blinded */
513 is_func:1, /* program is a bpf function */
514 kprobe_override:1, /* Do we override a kprobe? */
515 has_callchain_buf:1; /* callchain buffer allocated? */
516 enum bpf_prog_type type; /* Type of BPF program */
517 enum bpf_attach_type expected_attach_type; /* For some prog types */
518 u32 len; /* Number of filter blocks */
519 u32 jited_len; /* Size of jited insns in bytes */
520 u8 tag[BPF_TAG_SIZE];
521 struct bpf_prog_aux *aux; /* Auxiliary fields */
522 struct sock_fprog_kern *orig_prog; /* Original BPF program */
523 unsigned int (*bpf_func)(const void *ctx,
524 const struct bpf_insn *insn);
525 /* Instructions for interpreter */
527 struct sock_filter insns[0];
528 struct bpf_insn insnsi[0];
535 struct bpf_prog *prog;
538 DECLARE_STATIC_KEY_FALSE(bpf_stats_enabled_key);
540 #define BPF_PROG_RUN(prog, ctx) ({ \
543 if (static_branch_unlikely(&bpf_stats_enabled_key)) { \
544 struct bpf_prog_stats *stats; \
545 u64 start = sched_clock(); \
546 ret = (*(prog)->bpf_func)(ctx, (prog)->insnsi); \
547 stats = this_cpu_ptr(prog->aux->stats); \
548 u64_stats_update_begin(&stats->syncp); \
550 stats->nsecs += sched_clock() - start; \
551 u64_stats_update_end(&stats->syncp); \
553 ret = (*(prog)->bpf_func)(ctx, (prog)->insnsi); \
557 #define BPF_SKB_CB_LEN QDISC_CB_PRIV_LEN
559 struct bpf_skb_data_end {
560 struct qdisc_skb_cb qdisc_cb;
565 struct bpf_redirect_info {
569 struct bpf_map *map_to_flush;
573 DECLARE_PER_CPU(struct bpf_redirect_info, bpf_redirect_info);
575 /* flags for bpf_redirect_info kern_flags */
576 #define BPF_RI_F_RF_NO_DIRECT BIT(0) /* no napi_direct on return_frame */
578 /* Compute the linear packet data range [data, data_end) which
579 * will be accessed by various program types (cls_bpf, act_bpf,
580 * lwt, ...). Subsystems allowing direct data access must (!)
581 * ensure that cb[] area can be written to when BPF program is
582 * invoked (otherwise cb[] save/restore is necessary).
584 static inline void bpf_compute_data_pointers(struct sk_buff *skb)
586 struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb;
588 BUILD_BUG_ON(sizeof(*cb) > FIELD_SIZEOF(struct sk_buff, cb));
589 cb->data_meta = skb->data - skb_metadata_len(skb);
590 cb->data_end = skb->data + skb_headlen(skb);
593 /* Similar to bpf_compute_data_pointers(), except that save orginal
594 * data in cb->data and cb->meta_data for restore.
596 static inline void bpf_compute_and_save_data_end(
597 struct sk_buff *skb, void **saved_data_end)
599 struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb;
601 *saved_data_end = cb->data_end;
602 cb->data_end = skb->data + skb_headlen(skb);
605 /* Restore data saved by bpf_compute_data_pointers(). */
606 static inline void bpf_restore_data_end(
607 struct sk_buff *skb, void *saved_data_end)
609 struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb;
611 cb->data_end = saved_data_end;
614 static inline u8 *bpf_skb_cb(struct sk_buff *skb)
616 /* eBPF programs may read/write skb->cb[] area to transfer meta
617 * data between tail calls. Since this also needs to work with
618 * tc, that scratch memory is mapped to qdisc_skb_cb's data area.
620 * In some socket filter cases, the cb unfortunately needs to be
621 * saved/restored so that protocol specific skb->cb[] data won't
622 * be lost. In any case, due to unpriviledged eBPF programs
623 * attached to sockets, we need to clear the bpf_skb_cb() area
624 * to not leak previous contents to user space.
626 BUILD_BUG_ON(FIELD_SIZEOF(struct __sk_buff, cb) != BPF_SKB_CB_LEN);
627 BUILD_BUG_ON(FIELD_SIZEOF(struct __sk_buff, cb) !=
628 FIELD_SIZEOF(struct qdisc_skb_cb, data));
630 return qdisc_skb_cb(skb)->data;
633 static inline u32 __bpf_prog_run_save_cb(const struct bpf_prog *prog,
636 u8 *cb_data = bpf_skb_cb(skb);
637 u8 cb_saved[BPF_SKB_CB_LEN];
640 if (unlikely(prog->cb_access)) {
641 memcpy(cb_saved, cb_data, sizeof(cb_saved));
642 memset(cb_data, 0, sizeof(cb_saved));
645 res = BPF_PROG_RUN(prog, skb);
647 if (unlikely(prog->cb_access))
648 memcpy(cb_data, cb_saved, sizeof(cb_saved));
653 static inline u32 bpf_prog_run_save_cb(const struct bpf_prog *prog,
659 res = __bpf_prog_run_save_cb(prog, skb);
664 static inline u32 bpf_prog_run_clear_cb(const struct bpf_prog *prog,
667 u8 *cb_data = bpf_skb_cb(skb);
670 if (unlikely(prog->cb_access))
671 memset(cb_data, 0, BPF_SKB_CB_LEN);
674 res = BPF_PROG_RUN(prog, skb);
679 static __always_inline u32 bpf_prog_run_xdp(const struct bpf_prog *prog,
680 struct xdp_buff *xdp)
682 /* Caller needs to hold rcu_read_lock() (!), otherwise program
683 * can be released while still running, or map elements could be
684 * freed early while still having concurrent users. XDP fastpath
685 * already takes rcu_read_lock() when fetching the program, so
686 * it's not necessary here anymore.
688 return BPF_PROG_RUN(prog, xdp);
691 static inline u32 bpf_prog_insn_size(const struct bpf_prog *prog)
693 return prog->len * sizeof(struct bpf_insn);
696 static inline u32 bpf_prog_tag_scratch_size(const struct bpf_prog *prog)
698 return round_up(bpf_prog_insn_size(prog) +
699 sizeof(__be64) + 1, SHA_MESSAGE_BYTES);
702 static inline unsigned int bpf_prog_size(unsigned int proglen)
704 return max(sizeof(struct bpf_prog),
705 offsetof(struct bpf_prog, insns[proglen]));
708 static inline bool bpf_prog_was_classic(const struct bpf_prog *prog)
710 /* When classic BPF programs have been loaded and the arch
711 * does not have a classic BPF JIT (anymore), they have been
712 * converted via bpf_migrate_filter() to eBPF and thus always
713 * have an unspec program type.
715 return prog->type == BPF_PROG_TYPE_UNSPEC;
718 static inline u32 bpf_ctx_off_adjust_machine(u32 size)
720 const u32 size_machine = sizeof(unsigned long);
722 if (size > size_machine && size % size_machine == 0)
729 bpf_ctx_narrow_access_ok(u32 off, u32 size, u32 size_default)
731 return size <= size_default && (size & (size - 1)) == 0;
734 #define bpf_classic_proglen(fprog) (fprog->len * sizeof(fprog->filter[0]))
736 static inline void bpf_prog_lock_ro(struct bpf_prog *fp)
738 fp->undo_set_mem = 1;
739 set_memory_ro((unsigned long)fp, fp->pages);
742 static inline void bpf_prog_unlock_ro(struct bpf_prog *fp)
744 if (fp->undo_set_mem)
745 set_memory_rw((unsigned long)fp, fp->pages);
748 static inline void bpf_jit_binary_lock_ro(struct bpf_binary_header *hdr)
750 set_memory_ro((unsigned long)hdr, hdr->pages);
753 static inline void bpf_jit_binary_unlock_ro(struct bpf_binary_header *hdr)
755 set_memory_rw((unsigned long)hdr, hdr->pages);
758 static inline struct bpf_binary_header *
759 bpf_jit_binary_hdr(const struct bpf_prog *fp)
761 unsigned long real_start = (unsigned long)fp->bpf_func;
762 unsigned long addr = real_start & PAGE_MASK;
767 int sk_filter_trim_cap(struct sock *sk, struct sk_buff *skb, unsigned int cap);
768 static inline int sk_filter(struct sock *sk, struct sk_buff *skb)
770 return sk_filter_trim_cap(sk, skb, 1);
773 struct bpf_prog *bpf_prog_select_runtime(struct bpf_prog *fp, int *err);
774 void bpf_prog_free(struct bpf_prog *fp);
776 bool bpf_opcode_in_insntable(u8 code);
778 void bpf_prog_free_linfo(struct bpf_prog *prog);
779 void bpf_prog_fill_jited_linfo(struct bpf_prog *prog,
780 const u32 *insn_to_jit_off);
781 int bpf_prog_alloc_jited_linfo(struct bpf_prog *prog);
782 void bpf_prog_free_jited_linfo(struct bpf_prog *prog);
783 void bpf_prog_free_unused_jited_linfo(struct bpf_prog *prog);
785 struct bpf_prog *bpf_prog_alloc(unsigned int size, gfp_t gfp_extra_flags);
786 struct bpf_prog *bpf_prog_alloc_no_stats(unsigned int size, gfp_t gfp_extra_flags);
787 struct bpf_prog *bpf_prog_realloc(struct bpf_prog *fp_old, unsigned int size,
788 gfp_t gfp_extra_flags);
789 void __bpf_prog_free(struct bpf_prog *fp);
791 static inline void bpf_prog_unlock_free(struct bpf_prog *fp)
793 bpf_prog_unlock_ro(fp);
797 typedef int (*bpf_aux_classic_check_t)(struct sock_filter *filter,
800 int bpf_prog_create(struct bpf_prog **pfp, struct sock_fprog_kern *fprog);
801 int bpf_prog_create_from_user(struct bpf_prog **pfp, struct sock_fprog *fprog,
802 bpf_aux_classic_check_t trans, bool save_orig);
803 void bpf_prog_destroy(struct bpf_prog *fp);
805 int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk);
806 int sk_attach_bpf(u32 ufd, struct sock *sk);
807 int sk_reuseport_attach_filter(struct sock_fprog *fprog, struct sock *sk);
808 int sk_reuseport_attach_bpf(u32 ufd, struct sock *sk);
809 void sk_reuseport_prog_free(struct bpf_prog *prog);
810 int sk_detach_filter(struct sock *sk);
811 int sk_get_filter(struct sock *sk, struct sock_filter __user *filter,
814 bool sk_filter_charge(struct sock *sk, struct sk_filter *fp);
815 void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp);
817 u64 __bpf_call_base(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5);
818 #define __bpf_call_base_args \
819 ((u64 (*)(u64, u64, u64, u64, u64, const struct bpf_insn *)) \
822 struct bpf_prog *bpf_int_jit_compile(struct bpf_prog *prog);
823 void bpf_jit_compile(struct bpf_prog *prog);
824 bool bpf_helper_changes_pkt_data(void *func);
826 static inline bool bpf_dump_raw_ok(void)
828 /* Reconstruction of call-sites is dependent on kallsyms,
829 * thus make dump the same restriction.
831 return kallsyms_show_value() == 1;
834 struct bpf_prog *bpf_patch_insn_single(struct bpf_prog *prog, u32 off,
835 const struct bpf_insn *patch, u32 len);
836 int bpf_remove_insns(struct bpf_prog *prog, u32 off, u32 cnt);
838 void bpf_clear_redirect_map(struct bpf_map *map);
840 static inline bool xdp_return_frame_no_direct(void)
842 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
844 return ri->kern_flags & BPF_RI_F_RF_NO_DIRECT;
847 static inline void xdp_set_return_frame_no_direct(void)
849 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
851 ri->kern_flags |= BPF_RI_F_RF_NO_DIRECT;
854 static inline void xdp_clear_return_frame_no_direct(void)
856 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
858 ri->kern_flags &= ~BPF_RI_F_RF_NO_DIRECT;
861 static inline int xdp_ok_fwd_dev(const struct net_device *fwd,
866 if (unlikely(!(fwd->flags & IFF_UP)))
869 len = fwd->mtu + fwd->hard_header_len + VLAN_HLEN;
876 /* The pair of xdp_do_redirect and xdp_do_flush_map MUST be called in the
877 * same cpu context. Further for best results no more than a single map
878 * for the do_redirect/do_flush pair should be used. This limitation is
879 * because we only track one map and force a flush when the map changes.
880 * This does not appear to be a real limitation for existing software.
882 int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb,
883 struct xdp_buff *xdp, struct bpf_prog *prog);
884 int xdp_do_redirect(struct net_device *dev,
885 struct xdp_buff *xdp,
886 struct bpf_prog *prog);
887 void xdp_do_flush_map(void);
889 void bpf_warn_invalid_xdp_action(u32 act);
892 struct sock *bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk,
893 struct bpf_prog *prog, struct sk_buff *skb,
896 static inline struct sock *
897 bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk,
898 struct bpf_prog *prog, struct sk_buff *skb,
905 #ifdef CONFIG_BPF_JIT
906 extern int bpf_jit_enable;
907 extern int bpf_jit_harden;
908 extern int bpf_jit_kallsyms;
909 extern long bpf_jit_limit;
911 typedef void (*bpf_jit_fill_hole_t)(void *area, unsigned int size);
913 struct bpf_binary_header *
914 bpf_jit_binary_alloc(unsigned int proglen, u8 **image_ptr,
915 unsigned int alignment,
916 bpf_jit_fill_hole_t bpf_fill_ill_insns);
917 void bpf_jit_binary_free(struct bpf_binary_header *hdr);
918 u64 bpf_jit_alloc_exec_limit(void);
919 void *bpf_jit_alloc_exec(unsigned long size);
920 void bpf_jit_free_exec(void *addr);
921 void bpf_jit_free(struct bpf_prog *fp);
923 int bpf_jit_get_func_addr(const struct bpf_prog *prog,
924 const struct bpf_insn *insn, bool extra_pass,
925 u64 *func_addr, bool *func_addr_fixed);
927 struct bpf_prog *bpf_jit_blind_constants(struct bpf_prog *fp);
928 void bpf_jit_prog_release_other(struct bpf_prog *fp, struct bpf_prog *fp_other);
930 static inline void bpf_jit_dump(unsigned int flen, unsigned int proglen,
931 u32 pass, void *image)
933 pr_err("flen=%u proglen=%u pass=%u image=%pK from=%s pid=%d\n", flen,
934 proglen, pass, image, current->comm, task_pid_nr(current));
937 print_hex_dump(KERN_ERR, "JIT code: ", DUMP_PREFIX_OFFSET,
938 16, 1, image, proglen, false);
941 static inline bool bpf_jit_is_ebpf(void)
943 # ifdef CONFIG_HAVE_EBPF_JIT
950 static inline bool ebpf_jit_enabled(void)
952 return bpf_jit_enable && bpf_jit_is_ebpf();
955 static inline bool bpf_prog_ebpf_jited(const struct bpf_prog *fp)
957 return fp->jited && bpf_jit_is_ebpf();
960 static inline bool bpf_jit_blinding_enabled(struct bpf_prog *prog)
962 /* These are the prerequisites, should someone ever have the
963 * idea to call blinding outside of them, we make sure to
966 if (!bpf_jit_is_ebpf())
968 if (!prog->jit_requested)
972 if (bpf_jit_harden == 1 && capable(CAP_SYS_ADMIN))
978 static inline bool bpf_jit_kallsyms_enabled(void)
980 /* There are a couple of corner cases where kallsyms should
981 * not be enabled f.e. on hardening.
985 if (!bpf_jit_kallsyms)
987 if (bpf_jit_kallsyms == 1)
993 const char *__bpf_address_lookup(unsigned long addr, unsigned long *size,
994 unsigned long *off, char *sym);
995 bool is_bpf_text_address(unsigned long addr);
996 int bpf_get_kallsym(unsigned int symnum, unsigned long *value, char *type,
999 static inline const char *
1000 bpf_address_lookup(unsigned long addr, unsigned long *size,
1001 unsigned long *off, char **modname, char *sym)
1003 const char *ret = __bpf_address_lookup(addr, size, off, sym);
1010 void bpf_prog_kallsyms_add(struct bpf_prog *fp);
1011 void bpf_prog_kallsyms_del(struct bpf_prog *fp);
1012 void bpf_get_prog_name(const struct bpf_prog *prog, char *sym);
1014 #else /* CONFIG_BPF_JIT */
1016 static inline bool ebpf_jit_enabled(void)
1021 static inline bool bpf_prog_ebpf_jited(const struct bpf_prog *fp)
1026 static inline void bpf_jit_free(struct bpf_prog *fp)
1028 bpf_prog_unlock_free(fp);
1031 static inline bool bpf_jit_kallsyms_enabled(void)
1036 static inline const char *
1037 __bpf_address_lookup(unsigned long addr, unsigned long *size,
1038 unsigned long *off, char *sym)
1043 static inline bool is_bpf_text_address(unsigned long addr)
1048 static inline int bpf_get_kallsym(unsigned int symnum, unsigned long *value,
1049 char *type, char *sym)
1054 static inline const char *
1055 bpf_address_lookup(unsigned long addr, unsigned long *size,
1056 unsigned long *off, char **modname, char *sym)
1061 static inline void bpf_prog_kallsyms_add(struct bpf_prog *fp)
1065 static inline void bpf_prog_kallsyms_del(struct bpf_prog *fp)
1069 static inline void bpf_get_prog_name(const struct bpf_prog *prog, char *sym)
1074 #endif /* CONFIG_BPF_JIT */
1076 void bpf_prog_kallsyms_del_subprogs(struct bpf_prog *fp);
1077 void bpf_prog_kallsyms_del_all(struct bpf_prog *fp);
1079 #define BPF_ANC BIT(15)
1081 static inline bool bpf_needs_clear_a(const struct sock_filter *first)
1083 switch (first->code) {
1084 case BPF_RET | BPF_K:
1085 case BPF_LD | BPF_W | BPF_LEN:
1088 case BPF_LD | BPF_W | BPF_ABS:
1089 case BPF_LD | BPF_H | BPF_ABS:
1090 case BPF_LD | BPF_B | BPF_ABS:
1091 if (first->k == SKF_AD_OFF + SKF_AD_ALU_XOR_X)
1100 static inline u16 bpf_anc_helper(const struct sock_filter *ftest)
1102 BUG_ON(ftest->code & BPF_ANC);
1104 switch (ftest->code) {
1105 case BPF_LD | BPF_W | BPF_ABS:
1106 case BPF_LD | BPF_H | BPF_ABS:
1107 case BPF_LD | BPF_B | BPF_ABS:
1108 #define BPF_ANCILLARY(CODE) case SKF_AD_OFF + SKF_AD_##CODE: \
1109 return BPF_ANC | SKF_AD_##CODE
1111 BPF_ANCILLARY(PROTOCOL);
1112 BPF_ANCILLARY(PKTTYPE);
1113 BPF_ANCILLARY(IFINDEX);
1114 BPF_ANCILLARY(NLATTR);
1115 BPF_ANCILLARY(NLATTR_NEST);
1116 BPF_ANCILLARY(MARK);
1117 BPF_ANCILLARY(QUEUE);
1118 BPF_ANCILLARY(HATYPE);
1119 BPF_ANCILLARY(RXHASH);
1121 BPF_ANCILLARY(ALU_XOR_X);
1122 BPF_ANCILLARY(VLAN_TAG);
1123 BPF_ANCILLARY(VLAN_TAG_PRESENT);
1124 BPF_ANCILLARY(PAY_OFFSET);
1125 BPF_ANCILLARY(RANDOM);
1126 BPF_ANCILLARY(VLAN_TPID);
1134 void *bpf_internal_load_pointer_neg_helper(const struct sk_buff *skb,
1135 int k, unsigned int size);
1137 static inline void *bpf_load_pointer(const struct sk_buff *skb, int k,
1138 unsigned int size, void *buffer)
1141 return skb_header_pointer(skb, k, size, buffer);
1143 return bpf_internal_load_pointer_neg_helper(skb, k, size);
1146 static inline int bpf_tell_extensions(void)
1151 struct bpf_sock_addr_kern {
1153 struct sockaddr *uaddr;
1154 /* Temporary "register" to make indirect stores to nested structures
1155 * defined above. We need three registers to make such a store, but
1156 * only two (src and dst) are available at convert_ctx_access time
1159 void *t_ctx; /* Attach type specific context. */
1162 struct bpf_sock_ops_kern {
1171 u64 temp; /* temp and everything after is not
1172 * initialized to 0 before calling
1173 * the BPF program. New fields that
1174 * should be initialized to 0 should
1175 * be inserted before temp.
1176 * temp is scratch storage used by
1177 * sock_ops_convert_ctx_access
1178 * as temporary storage of a register.
1182 struct bpf_sysctl_kern {
1183 struct ctl_table_header *head;
1184 struct ctl_table *table;
1188 #endif /* __LINUX_FILTER_H__ */