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/set_memory.h>
20 #include <linux/kallsyms.h>
21 #include <linux/if_vlan.h>
22 #include <linux/vmalloc.h>
23 #include <linux/sockptr.h>
24 #include <crypto/sha1.h>
25 #include <linux/u64_stats_sync.h>
27 #include <net/sch_generic.h>
29 #include <asm/byteorder.h>
30 #include <uapi/linux/filter.h>
31 #include <uapi/linux/bpf.h>
39 struct sock_reuseport;
41 struct ctl_table_header;
43 /* ArgX, context and stack frame pointer register positions. Note,
44 * Arg1, Arg2, Arg3, etc are used as argument mappings of function
45 * calls in BPF_CALL instruction.
47 #define BPF_REG_ARG1 BPF_REG_1
48 #define BPF_REG_ARG2 BPF_REG_2
49 #define BPF_REG_ARG3 BPF_REG_3
50 #define BPF_REG_ARG4 BPF_REG_4
51 #define BPF_REG_ARG5 BPF_REG_5
52 #define BPF_REG_CTX BPF_REG_6
53 #define BPF_REG_FP BPF_REG_10
55 /* Additional register mappings for converted user programs. */
56 #define BPF_REG_A BPF_REG_0
57 #define BPF_REG_X BPF_REG_7
58 #define BPF_REG_TMP BPF_REG_2 /* scratch reg */
59 #define BPF_REG_D BPF_REG_8 /* data, callee-saved */
60 #define BPF_REG_H BPF_REG_9 /* hlen, callee-saved */
62 /* Kernel hidden auxiliary/helper register. */
63 #define BPF_REG_AX MAX_BPF_REG
64 #define MAX_BPF_EXT_REG (MAX_BPF_REG + 1)
65 #define MAX_BPF_JIT_REG MAX_BPF_EXT_REG
67 /* unused opcode to mark special call to bpf_tail_call() helper */
68 #define BPF_TAIL_CALL 0xf0
70 /* unused opcode to mark special load instruction. Same as BPF_ABS */
71 #define BPF_PROBE_MEM 0x20
73 /* unused opcode to mark call to interpreter with arguments */
74 #define BPF_CALL_ARGS 0xe0
76 /* As per nm, we expose JITed images as text (code) section for
77 * kallsyms. That way, tools like perf can find it to match
80 #define BPF_SYM_ELF_TYPE 't'
82 /* BPF program can access up to 512 bytes of stack space. */
83 #define MAX_BPF_STACK 512
85 /* Helper macros for filter block array initializers. */
87 /* ALU ops on registers, bpf_add|sub|...: dst_reg += src_reg */
89 #define BPF_ALU64_REG(OP, DST, SRC) \
90 ((struct bpf_insn) { \
91 .code = BPF_ALU64 | BPF_OP(OP) | BPF_X, \
97 #define BPF_ALU32_REG(OP, DST, SRC) \
98 ((struct bpf_insn) { \
99 .code = BPF_ALU | BPF_OP(OP) | BPF_X, \
105 /* ALU ops on immediates, bpf_add|sub|...: dst_reg += imm32 */
107 #define BPF_ALU64_IMM(OP, DST, IMM) \
108 ((struct bpf_insn) { \
109 .code = BPF_ALU64 | BPF_OP(OP) | BPF_K, \
115 #define BPF_ALU32_IMM(OP, DST, IMM) \
116 ((struct bpf_insn) { \
117 .code = BPF_ALU | BPF_OP(OP) | BPF_K, \
123 /* Endianess conversion, cpu_to_{l,b}e(), {l,b}e_to_cpu() */
125 #define BPF_ENDIAN(TYPE, DST, LEN) \
126 ((struct bpf_insn) { \
127 .code = BPF_ALU | BPF_END | BPF_SRC(TYPE), \
133 /* Short form of mov, dst_reg = src_reg */
135 #define BPF_MOV64_REG(DST, SRC) \
136 ((struct bpf_insn) { \
137 .code = BPF_ALU64 | BPF_MOV | BPF_X, \
143 #define BPF_MOV32_REG(DST, SRC) \
144 ((struct bpf_insn) { \
145 .code = BPF_ALU | BPF_MOV | BPF_X, \
151 /* Short form of mov, dst_reg = imm32 */
153 #define BPF_MOV64_IMM(DST, IMM) \
154 ((struct bpf_insn) { \
155 .code = BPF_ALU64 | BPF_MOV | BPF_K, \
161 #define BPF_MOV32_IMM(DST, IMM) \
162 ((struct bpf_insn) { \
163 .code = BPF_ALU | BPF_MOV | BPF_K, \
169 /* Special form of mov32, used for doing explicit zero extension on dst. */
170 #define BPF_ZEXT_REG(DST) \
171 ((struct bpf_insn) { \
172 .code = BPF_ALU | BPF_MOV | BPF_X, \
178 static inline bool insn_is_zext(const struct bpf_insn *insn)
180 return insn->code == (BPF_ALU | BPF_MOV | BPF_X) && insn->imm == 1;
183 /* BPF_LD_IMM64 macro encodes single 'load 64-bit immediate' insn */
184 #define BPF_LD_IMM64(DST, IMM) \
185 BPF_LD_IMM64_RAW(DST, 0, IMM)
187 #define BPF_LD_IMM64_RAW(DST, SRC, IMM) \
188 ((struct bpf_insn) { \
189 .code = BPF_LD | BPF_DW | BPF_IMM, \
193 .imm = (__u32) (IMM) }), \
194 ((struct bpf_insn) { \
195 .code = 0, /* zero is reserved opcode */ \
199 .imm = ((__u64) (IMM)) >> 32 })
201 /* pseudo BPF_LD_IMM64 insn used to refer to process-local map_fd */
202 #define BPF_LD_MAP_FD(DST, MAP_FD) \
203 BPF_LD_IMM64_RAW(DST, BPF_PSEUDO_MAP_FD, MAP_FD)
205 /* Short form of mov based on type, BPF_X: dst_reg = src_reg, BPF_K: dst_reg = imm32 */
207 #define BPF_MOV64_RAW(TYPE, DST, SRC, IMM) \
208 ((struct bpf_insn) { \
209 .code = BPF_ALU64 | BPF_MOV | BPF_SRC(TYPE), \
215 #define BPF_MOV32_RAW(TYPE, DST, SRC, IMM) \
216 ((struct bpf_insn) { \
217 .code = BPF_ALU | BPF_MOV | BPF_SRC(TYPE), \
223 /* Direct packet access, R0 = *(uint *) (skb->data + imm32) */
225 #define BPF_LD_ABS(SIZE, IMM) \
226 ((struct bpf_insn) { \
227 .code = BPF_LD | BPF_SIZE(SIZE) | BPF_ABS, \
233 /* Indirect packet access, R0 = *(uint *) (skb->data + src_reg + imm32) */
235 #define BPF_LD_IND(SIZE, SRC, IMM) \
236 ((struct bpf_insn) { \
237 .code = BPF_LD | BPF_SIZE(SIZE) | BPF_IND, \
243 /* Memory load, dst_reg = *(uint *) (src_reg + off16) */
245 #define BPF_LDX_MEM(SIZE, DST, SRC, OFF) \
246 ((struct bpf_insn) { \
247 .code = BPF_LDX | BPF_SIZE(SIZE) | BPF_MEM, \
253 /* Memory store, *(uint *) (dst_reg + off16) = src_reg */
255 #define BPF_STX_MEM(SIZE, DST, SRC, OFF) \
256 ((struct bpf_insn) { \
257 .code = BPF_STX | BPF_SIZE(SIZE) | BPF_MEM, \
267 * BPF_ADD *(uint *) (dst_reg + off16) += src_reg
268 * BPF_AND *(uint *) (dst_reg + off16) &= src_reg
269 * BPF_OR *(uint *) (dst_reg + off16) |= src_reg
270 * BPF_XOR *(uint *) (dst_reg + off16) ^= src_reg
271 * BPF_ADD | BPF_FETCH src_reg = atomic_fetch_add(dst_reg + off16, src_reg);
272 * BPF_AND | BPF_FETCH src_reg = atomic_fetch_and(dst_reg + off16, src_reg);
273 * BPF_OR | BPF_FETCH src_reg = atomic_fetch_or(dst_reg + off16, src_reg);
274 * BPF_XOR | BPF_FETCH src_reg = atomic_fetch_xor(dst_reg + off16, src_reg);
275 * BPF_XCHG src_reg = atomic_xchg(dst_reg + off16, src_reg)
276 * BPF_CMPXCHG r0 = atomic_cmpxchg(dst_reg + off16, r0, src_reg)
279 #define BPF_ATOMIC_OP(SIZE, OP, DST, SRC, OFF) \
280 ((struct bpf_insn) { \
281 .code = BPF_STX | BPF_SIZE(SIZE) | BPF_ATOMIC, \
288 #define BPF_STX_XADD(SIZE, DST, SRC, OFF) BPF_ATOMIC_OP(SIZE, BPF_ADD, DST, SRC, OFF)
290 /* Memory store, *(uint *) (dst_reg + off16) = imm32 */
292 #define BPF_ST_MEM(SIZE, DST, OFF, IMM) \
293 ((struct bpf_insn) { \
294 .code = BPF_ST | BPF_SIZE(SIZE) | BPF_MEM, \
300 /* Conditional jumps against registers, if (dst_reg 'op' src_reg) goto pc + off16 */
302 #define BPF_JMP_REG(OP, DST, SRC, OFF) \
303 ((struct bpf_insn) { \
304 .code = BPF_JMP | BPF_OP(OP) | BPF_X, \
310 /* Conditional jumps against immediates, if (dst_reg 'op' imm32) goto pc + off16 */
312 #define BPF_JMP_IMM(OP, DST, IMM, OFF) \
313 ((struct bpf_insn) { \
314 .code = BPF_JMP | BPF_OP(OP) | BPF_K, \
320 /* Like BPF_JMP_REG, but with 32-bit wide operands for comparison. */
322 #define BPF_JMP32_REG(OP, DST, SRC, OFF) \
323 ((struct bpf_insn) { \
324 .code = BPF_JMP32 | BPF_OP(OP) | BPF_X, \
330 /* Like BPF_JMP_IMM, but with 32-bit wide operands for comparison. */
332 #define BPF_JMP32_IMM(OP, DST, IMM, OFF) \
333 ((struct bpf_insn) { \
334 .code = BPF_JMP32 | BPF_OP(OP) | BPF_K, \
340 /* Unconditional jumps, goto pc + off16 */
342 #define BPF_JMP_A(OFF) \
343 ((struct bpf_insn) { \
344 .code = BPF_JMP | BPF_JA, \
352 #define BPF_CALL_REL(TGT) \
353 ((struct bpf_insn) { \
354 .code = BPF_JMP | BPF_CALL, \
356 .src_reg = BPF_PSEUDO_CALL, \
362 #define BPF_CAST_CALL(x) \
363 ((u64 (*)(u64, u64, u64, u64, u64))(x))
365 #define BPF_EMIT_CALL(FUNC) \
366 ((struct bpf_insn) { \
367 .code = BPF_JMP | BPF_CALL, \
371 .imm = ((FUNC) - __bpf_call_base) })
373 /* Raw code statement block */
375 #define BPF_RAW_INSN(CODE, DST, SRC, OFF, IMM) \
376 ((struct bpf_insn) { \
385 #define BPF_EXIT_INSN() \
386 ((struct bpf_insn) { \
387 .code = BPF_JMP | BPF_EXIT, \
393 /* Internal classic blocks for direct assignment */
395 #define __BPF_STMT(CODE, K) \
396 ((struct sock_filter) BPF_STMT(CODE, K))
398 #define __BPF_JUMP(CODE, K, JT, JF) \
399 ((struct sock_filter) BPF_JUMP(CODE, K, JT, JF))
401 #define bytes_to_bpf_size(bytes) \
403 int bpf_size = -EINVAL; \
405 if (bytes == sizeof(u8)) \
407 else if (bytes == sizeof(u16)) \
409 else if (bytes == sizeof(u32)) \
411 else if (bytes == sizeof(u64)) \
417 #define bpf_size_to_bytes(bpf_size) \
419 int bytes = -EINVAL; \
421 if (bpf_size == BPF_B) \
422 bytes = sizeof(u8); \
423 else if (bpf_size == BPF_H) \
424 bytes = sizeof(u16); \
425 else if (bpf_size == BPF_W) \
426 bytes = sizeof(u32); \
427 else if (bpf_size == BPF_DW) \
428 bytes = sizeof(u64); \
433 #define BPF_SIZEOF(type) \
435 const int __size = bytes_to_bpf_size(sizeof(type)); \
436 BUILD_BUG_ON(__size < 0); \
440 #define BPF_FIELD_SIZEOF(type, field) \
442 const int __size = bytes_to_bpf_size(sizeof_field(type, field)); \
443 BUILD_BUG_ON(__size < 0); \
447 #define BPF_LDST_BYTES(insn) \
449 const int __size = bpf_size_to_bytes(BPF_SIZE((insn)->code)); \
450 WARN_ON(__size < 0); \
454 #define __BPF_MAP_0(m, v, ...) v
455 #define __BPF_MAP_1(m, v, t, a, ...) m(t, a)
456 #define __BPF_MAP_2(m, v, t, a, ...) m(t, a), __BPF_MAP_1(m, v, __VA_ARGS__)
457 #define __BPF_MAP_3(m, v, t, a, ...) m(t, a), __BPF_MAP_2(m, v, __VA_ARGS__)
458 #define __BPF_MAP_4(m, v, t, a, ...) m(t, a), __BPF_MAP_3(m, v, __VA_ARGS__)
459 #define __BPF_MAP_5(m, v, t, a, ...) m(t, a), __BPF_MAP_4(m, v, __VA_ARGS__)
461 #define __BPF_REG_0(...) __BPF_PAD(5)
462 #define __BPF_REG_1(...) __BPF_MAP(1, __VA_ARGS__), __BPF_PAD(4)
463 #define __BPF_REG_2(...) __BPF_MAP(2, __VA_ARGS__), __BPF_PAD(3)
464 #define __BPF_REG_3(...) __BPF_MAP(3, __VA_ARGS__), __BPF_PAD(2)
465 #define __BPF_REG_4(...) __BPF_MAP(4, __VA_ARGS__), __BPF_PAD(1)
466 #define __BPF_REG_5(...) __BPF_MAP(5, __VA_ARGS__)
468 #define __BPF_MAP(n, ...) __BPF_MAP_##n(__VA_ARGS__)
469 #define __BPF_REG(n, ...) __BPF_REG_##n(__VA_ARGS__)
471 #define __BPF_CAST(t, a) \
474 typeof(__builtin_choose_expr(sizeof(t) == sizeof(unsigned long), \
475 (unsigned long)0, (t)0))) a
479 #define __BPF_DECL_ARGS(t, a) t a
480 #define __BPF_DECL_REGS(t, a) u64 a
482 #define __BPF_PAD(n) \
483 __BPF_MAP(n, __BPF_DECL_ARGS, __BPF_N, u64, __ur_1, u64, __ur_2, \
484 u64, __ur_3, u64, __ur_4, u64, __ur_5)
486 #define BPF_CALL_x(x, name, ...) \
487 static __always_inline \
488 u64 ____##name(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__)); \
489 typedef u64 (*btf_##name)(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__)); \
490 u64 name(__BPF_REG(x, __BPF_DECL_REGS, __BPF_N, __VA_ARGS__)); \
491 u64 name(__BPF_REG(x, __BPF_DECL_REGS, __BPF_N, __VA_ARGS__)) \
493 return ((btf_##name)____##name)(__BPF_MAP(x,__BPF_CAST,__BPF_N,__VA_ARGS__));\
495 static __always_inline \
496 u64 ____##name(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__))
498 #define BPF_CALL_0(name, ...) BPF_CALL_x(0, name, __VA_ARGS__)
499 #define BPF_CALL_1(name, ...) BPF_CALL_x(1, name, __VA_ARGS__)
500 #define BPF_CALL_2(name, ...) BPF_CALL_x(2, name, __VA_ARGS__)
501 #define BPF_CALL_3(name, ...) BPF_CALL_x(3, name, __VA_ARGS__)
502 #define BPF_CALL_4(name, ...) BPF_CALL_x(4, name, __VA_ARGS__)
503 #define BPF_CALL_5(name, ...) BPF_CALL_x(5, name, __VA_ARGS__)
505 #define bpf_ctx_range(TYPE, MEMBER) \
506 offsetof(TYPE, MEMBER) ... offsetofend(TYPE, MEMBER) - 1
507 #define bpf_ctx_range_till(TYPE, MEMBER1, MEMBER2) \
508 offsetof(TYPE, MEMBER1) ... offsetofend(TYPE, MEMBER2) - 1
509 #if BITS_PER_LONG == 64
510 # define bpf_ctx_range_ptr(TYPE, MEMBER) \
511 offsetof(TYPE, MEMBER) ... offsetofend(TYPE, MEMBER) - 1
513 # define bpf_ctx_range_ptr(TYPE, MEMBER) \
514 offsetof(TYPE, MEMBER) ... offsetof(TYPE, MEMBER) + 8 - 1
515 #endif /* BITS_PER_LONG == 64 */
517 #define bpf_target_off(TYPE, MEMBER, SIZE, PTR_SIZE) \
519 BUILD_BUG_ON(sizeof_field(TYPE, MEMBER) != (SIZE)); \
520 *(PTR_SIZE) = (SIZE); \
521 offsetof(TYPE, MEMBER); \
524 /* A struct sock_filter is architecture independent. */
525 struct compat_sock_fprog {
527 compat_uptr_t filter; /* struct sock_filter * */
530 struct sock_fprog_kern {
532 struct sock_filter *filter;
535 /* Some arches need doubleword alignment for their instructions and/or data */
536 #define BPF_IMAGE_ALIGNMENT 8
538 struct bpf_binary_header {
540 u8 image[] __aligned(BPF_IMAGE_ALIGNMENT);
543 struct bpf_prog_stats {
547 struct u64_stats_sync syncp;
548 } __aligned(2 * sizeof(u64));
551 u16 pages; /* Number of allocated pages */
552 u16 jited:1, /* Is our filter JIT'ed? */
553 jit_requested:1,/* archs need to JIT the prog */
554 gpl_compatible:1, /* Is filter GPL compatible? */
555 cb_access:1, /* Is control block accessed? */
556 dst_needed:1, /* Do we need dst entry? */
557 blinded:1, /* Was blinded */
558 is_func:1, /* program is a bpf function */
559 kprobe_override:1, /* Do we override a kprobe? */
560 has_callchain_buf:1, /* callchain buffer allocated? */
561 enforce_expected_attach_type:1, /* Enforce expected_attach_type checking at attach time */
562 call_get_stack:1; /* Do we call bpf_get_stack() or bpf_get_stackid() */
563 enum bpf_prog_type type; /* Type of BPF program */
564 enum bpf_attach_type expected_attach_type; /* For some prog types */
565 u32 len; /* Number of filter blocks */
566 u32 jited_len; /* Size of jited insns in bytes */
567 u8 tag[BPF_TAG_SIZE];
568 struct bpf_prog_stats __percpu *stats;
569 int __percpu *active;
570 unsigned int (*bpf_func)(const void *ctx,
571 const struct bpf_insn *insn);
572 struct bpf_prog_aux *aux; /* Auxiliary fields */
573 struct sock_fprog_kern *orig_prog; /* Original BPF program */
574 /* Instructions for interpreter */
575 struct sock_filter insns[0];
576 struct bpf_insn insnsi[];
582 struct bpf_prog *prog;
585 DECLARE_STATIC_KEY_FALSE(bpf_stats_enabled_key);
587 #define __BPF_PROG_RUN(prog, ctx, dfunc) ({ \
590 if (static_branch_unlikely(&bpf_stats_enabled_key)) { \
591 struct bpf_prog_stats *__stats; \
592 u64 __start = sched_clock(); \
593 __ret = dfunc(ctx, (prog)->insnsi, (prog)->bpf_func); \
594 __stats = this_cpu_ptr(prog->stats); \
595 u64_stats_update_begin(&__stats->syncp); \
597 __stats->nsecs += sched_clock() - __start; \
598 u64_stats_update_end(&__stats->syncp); \
600 __ret = dfunc(ctx, (prog)->insnsi, (prog)->bpf_func); \
604 #define BPF_PROG_RUN(prog, ctx) \
605 __BPF_PROG_RUN(prog, ctx, bpf_dispatcher_nop_func)
608 * Use in preemptible and therefore migratable context to make sure that
609 * the execution of the BPF program runs on one CPU.
611 * This uses migrate_disable/enable() explicitly to document that the
612 * invocation of a BPF program does not require reentrancy protection
613 * against a BPF program which is invoked from a preempting task.
615 * For non RT enabled kernels migrate_disable/enable() maps to
616 * preempt_disable/enable(), i.e. it disables also preemption.
618 static inline u32 bpf_prog_run_pin_on_cpu(const struct bpf_prog *prog,
624 ret = __BPF_PROG_RUN(prog, ctx, bpf_dispatcher_nop_func);
629 #define BPF_SKB_CB_LEN QDISC_CB_PRIV_LEN
631 struct bpf_skb_data_end {
632 struct qdisc_skb_cb qdisc_cb;
637 struct bpf_nh_params {
641 struct in6_addr ipv6_nh;
645 struct bpf_redirect_info {
651 enum bpf_map_type map_type;
653 struct bpf_nh_params nh;
656 DECLARE_PER_CPU(struct bpf_redirect_info, bpf_redirect_info);
658 /* flags for bpf_redirect_info kern_flags */
659 #define BPF_RI_F_RF_NO_DIRECT BIT(0) /* no napi_direct on return_frame */
661 /* Compute the linear packet data range [data, data_end) which
662 * will be accessed by various program types (cls_bpf, act_bpf,
663 * lwt, ...). Subsystems allowing direct data access must (!)
664 * ensure that cb[] area can be written to when BPF program is
665 * invoked (otherwise cb[] save/restore is necessary).
667 static inline void bpf_compute_data_pointers(struct sk_buff *skb)
669 struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb;
671 BUILD_BUG_ON(sizeof(*cb) > sizeof_field(struct sk_buff, cb));
672 cb->data_meta = skb->data - skb_metadata_len(skb);
673 cb->data_end = skb->data + skb_headlen(skb);
676 /* Similar to bpf_compute_data_pointers(), except that save orginal
677 * data in cb->data and cb->meta_data for restore.
679 static inline void bpf_compute_and_save_data_end(
680 struct sk_buff *skb, void **saved_data_end)
682 struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb;
684 *saved_data_end = cb->data_end;
685 cb->data_end = skb->data + skb_headlen(skb);
688 /* Restore data saved by bpf_compute_data_pointers(). */
689 static inline void bpf_restore_data_end(
690 struct sk_buff *skb, void *saved_data_end)
692 struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb;
694 cb->data_end = saved_data_end;
697 static inline u8 *bpf_skb_cb(struct sk_buff *skb)
699 /* eBPF programs may read/write skb->cb[] area to transfer meta
700 * data between tail calls. Since this also needs to work with
701 * tc, that scratch memory is mapped to qdisc_skb_cb's data area.
703 * In some socket filter cases, the cb unfortunately needs to be
704 * saved/restored so that protocol specific skb->cb[] data won't
705 * be lost. In any case, due to unpriviledged eBPF programs
706 * attached to sockets, we need to clear the bpf_skb_cb() area
707 * to not leak previous contents to user space.
709 BUILD_BUG_ON(sizeof_field(struct __sk_buff, cb) != BPF_SKB_CB_LEN);
710 BUILD_BUG_ON(sizeof_field(struct __sk_buff, cb) !=
711 sizeof_field(struct qdisc_skb_cb, data));
713 return qdisc_skb_cb(skb)->data;
716 /* Must be invoked with migration disabled */
717 static inline u32 __bpf_prog_run_save_cb(const struct bpf_prog *prog,
720 u8 *cb_data = bpf_skb_cb(skb);
721 u8 cb_saved[BPF_SKB_CB_LEN];
724 if (unlikely(prog->cb_access)) {
725 memcpy(cb_saved, cb_data, sizeof(cb_saved));
726 memset(cb_data, 0, sizeof(cb_saved));
729 res = BPF_PROG_RUN(prog, skb);
731 if (unlikely(prog->cb_access))
732 memcpy(cb_data, cb_saved, sizeof(cb_saved));
737 static inline u32 bpf_prog_run_save_cb(const struct bpf_prog *prog,
743 res = __bpf_prog_run_save_cb(prog, skb);
748 static inline u32 bpf_prog_run_clear_cb(const struct bpf_prog *prog,
751 u8 *cb_data = bpf_skb_cb(skb);
754 if (unlikely(prog->cb_access))
755 memset(cb_data, 0, BPF_SKB_CB_LEN);
757 res = bpf_prog_run_pin_on_cpu(prog, skb);
761 DECLARE_BPF_DISPATCHER(xdp)
763 static __always_inline u32 bpf_prog_run_xdp(const struct bpf_prog *prog,
764 struct xdp_buff *xdp)
766 /* Caller needs to hold rcu_read_lock() (!), otherwise program
767 * can be released while still running, or map elements could be
768 * freed early while still having concurrent users. XDP fastpath
769 * already takes rcu_read_lock() when fetching the program, so
770 * it's not necessary here anymore.
772 return __BPF_PROG_RUN(prog, xdp, BPF_DISPATCHER_FUNC(xdp));
775 void bpf_prog_change_xdp(struct bpf_prog *prev_prog, struct bpf_prog *prog);
777 static inline u32 bpf_prog_insn_size(const struct bpf_prog *prog)
779 return prog->len * sizeof(struct bpf_insn);
782 static inline u32 bpf_prog_tag_scratch_size(const struct bpf_prog *prog)
784 return round_up(bpf_prog_insn_size(prog) +
785 sizeof(__be64) + 1, SHA1_BLOCK_SIZE);
788 static inline unsigned int bpf_prog_size(unsigned int proglen)
790 return max(sizeof(struct bpf_prog),
791 offsetof(struct bpf_prog, insns[proglen]));
794 static inline bool bpf_prog_was_classic(const struct bpf_prog *prog)
796 /* When classic BPF programs have been loaded and the arch
797 * does not have a classic BPF JIT (anymore), they have been
798 * converted via bpf_migrate_filter() to eBPF and thus always
799 * have an unspec program type.
801 return prog->type == BPF_PROG_TYPE_UNSPEC;
804 static inline u32 bpf_ctx_off_adjust_machine(u32 size)
806 const u32 size_machine = sizeof(unsigned long);
808 if (size > size_machine && size % size_machine == 0)
815 bpf_ctx_narrow_access_ok(u32 off, u32 size, u32 size_default)
817 return size <= size_default && (size & (size - 1)) == 0;
821 bpf_ctx_narrow_access_offset(u32 off, u32 size, u32 size_default)
823 u8 access_off = off & (size_default - 1);
825 #ifdef __LITTLE_ENDIAN
828 return size_default - (access_off + size);
832 #define bpf_ctx_wide_access_ok(off, size, type, field) \
833 (size == sizeof(__u64) && \
834 off >= offsetof(type, field) && \
835 off + sizeof(__u64) <= offsetofend(type, field) && \
836 off % sizeof(__u64) == 0)
838 #define bpf_classic_proglen(fprog) (fprog->len * sizeof(fprog->filter[0]))
840 static inline void bpf_prog_lock_ro(struct bpf_prog *fp)
842 #ifndef CONFIG_BPF_JIT_ALWAYS_ON
844 set_vm_flush_reset_perms(fp);
845 set_memory_ro((unsigned long)fp, fp->pages);
850 static inline void bpf_jit_binary_lock_ro(struct bpf_binary_header *hdr)
852 set_vm_flush_reset_perms(hdr);
853 set_memory_ro((unsigned long)hdr, hdr->pages);
854 set_memory_x((unsigned long)hdr, hdr->pages);
857 static inline struct bpf_binary_header *
858 bpf_jit_binary_hdr(const struct bpf_prog *fp)
860 unsigned long real_start = (unsigned long)fp->bpf_func;
861 unsigned long addr = real_start & PAGE_MASK;
866 int sk_filter_trim_cap(struct sock *sk, struct sk_buff *skb, unsigned int cap);
867 static inline int sk_filter(struct sock *sk, struct sk_buff *skb)
869 return sk_filter_trim_cap(sk, skb, 1);
872 struct bpf_prog *bpf_prog_select_runtime(struct bpf_prog *fp, int *err);
873 void bpf_prog_free(struct bpf_prog *fp);
875 bool bpf_opcode_in_insntable(u8 code);
877 void bpf_prog_free_linfo(struct bpf_prog *prog);
878 void bpf_prog_fill_jited_linfo(struct bpf_prog *prog,
879 const u32 *insn_to_jit_off);
880 int bpf_prog_alloc_jited_linfo(struct bpf_prog *prog);
881 void bpf_prog_jit_attempt_done(struct bpf_prog *prog);
883 struct bpf_prog *bpf_prog_alloc(unsigned int size, gfp_t gfp_extra_flags);
884 struct bpf_prog *bpf_prog_alloc_no_stats(unsigned int size, gfp_t gfp_extra_flags);
885 struct bpf_prog *bpf_prog_realloc(struct bpf_prog *fp_old, unsigned int size,
886 gfp_t gfp_extra_flags);
887 void __bpf_prog_free(struct bpf_prog *fp);
889 static inline void bpf_prog_unlock_free(struct bpf_prog *fp)
894 typedef int (*bpf_aux_classic_check_t)(struct sock_filter *filter,
897 int bpf_prog_create(struct bpf_prog **pfp, struct sock_fprog_kern *fprog);
898 int bpf_prog_create_from_user(struct bpf_prog **pfp, struct sock_fprog *fprog,
899 bpf_aux_classic_check_t trans, bool save_orig);
900 void bpf_prog_destroy(struct bpf_prog *fp);
902 int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk);
903 int sk_attach_bpf(u32 ufd, struct sock *sk);
904 int sk_reuseport_attach_filter(struct sock_fprog *fprog, struct sock *sk);
905 int sk_reuseport_attach_bpf(u32 ufd, struct sock *sk);
906 void sk_reuseport_prog_free(struct bpf_prog *prog);
907 int sk_detach_filter(struct sock *sk);
908 int sk_get_filter(struct sock *sk, struct sock_filter __user *filter,
911 bool sk_filter_charge(struct sock *sk, struct sk_filter *fp);
912 void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp);
914 u64 __bpf_call_base(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5);
915 #define __bpf_call_base_args \
916 ((u64 (*)(u64, u64, u64, u64, u64, const struct bpf_insn *)) \
917 (void *)__bpf_call_base)
919 struct bpf_prog *bpf_int_jit_compile(struct bpf_prog *prog);
920 void bpf_jit_compile(struct bpf_prog *prog);
921 bool bpf_jit_needs_zext(void);
922 bool bpf_jit_supports_kfunc_call(void);
923 bool bpf_helper_changes_pkt_data(void *func);
925 static inline bool bpf_dump_raw_ok(const struct cred *cred)
927 /* Reconstruction of call-sites is dependent on kallsyms,
928 * thus make dump the same restriction.
930 return kallsyms_show_value(cred);
933 struct bpf_prog *bpf_patch_insn_single(struct bpf_prog *prog, u32 off,
934 const struct bpf_insn *patch, u32 len);
935 int bpf_remove_insns(struct bpf_prog *prog, u32 off, u32 cnt);
937 void bpf_clear_redirect_map(struct bpf_map *map);
939 static inline bool xdp_return_frame_no_direct(void)
941 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
943 return ri->kern_flags & BPF_RI_F_RF_NO_DIRECT;
946 static inline void xdp_set_return_frame_no_direct(void)
948 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
950 ri->kern_flags |= BPF_RI_F_RF_NO_DIRECT;
953 static inline void xdp_clear_return_frame_no_direct(void)
955 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
957 ri->kern_flags &= ~BPF_RI_F_RF_NO_DIRECT;
960 static inline int xdp_ok_fwd_dev(const struct net_device *fwd,
965 if (unlikely(!(fwd->flags & IFF_UP)))
968 len = fwd->mtu + fwd->hard_header_len + VLAN_HLEN;
975 /* The pair of xdp_do_redirect and xdp_do_flush MUST be called in the
976 * same cpu context. Further for best results no more than a single map
977 * for the do_redirect/do_flush pair should be used. This limitation is
978 * because we only track one map and force a flush when the map changes.
979 * This does not appear to be a real limitation for existing software.
981 int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb,
982 struct xdp_buff *xdp, struct bpf_prog *prog);
983 int xdp_do_redirect(struct net_device *dev,
984 struct xdp_buff *xdp,
985 struct bpf_prog *prog);
986 void xdp_do_flush(void);
988 /* The xdp_do_flush_map() helper has been renamed to drop the _map suffix, as
989 * it is no longer only flushing maps. Keep this define for compatibility
990 * until all drivers are updated - do not use xdp_do_flush_map() in new code!
992 #define xdp_do_flush_map xdp_do_flush
994 void bpf_warn_invalid_xdp_action(u32 act);
997 struct sock *bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk,
998 struct bpf_prog *prog, struct sk_buff *skb,
999 struct sock *migrating_sk,
1002 static inline struct sock *
1003 bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk,
1004 struct bpf_prog *prog, struct sk_buff *skb,
1005 struct sock *migrating_sk,
1012 #ifdef CONFIG_BPF_JIT
1013 extern int bpf_jit_enable;
1014 extern int bpf_jit_harden;
1015 extern int bpf_jit_kallsyms;
1016 extern long bpf_jit_limit;
1018 typedef void (*bpf_jit_fill_hole_t)(void *area, unsigned int size);
1020 struct bpf_binary_header *
1021 bpf_jit_binary_alloc(unsigned int proglen, u8 **image_ptr,
1022 unsigned int alignment,
1023 bpf_jit_fill_hole_t bpf_fill_ill_insns);
1024 void bpf_jit_binary_free(struct bpf_binary_header *hdr);
1025 u64 bpf_jit_alloc_exec_limit(void);
1026 void *bpf_jit_alloc_exec(unsigned long size);
1027 void bpf_jit_free_exec(void *addr);
1028 void bpf_jit_free(struct bpf_prog *fp);
1030 int bpf_jit_add_poke_descriptor(struct bpf_prog *prog,
1031 struct bpf_jit_poke_descriptor *poke);
1033 int bpf_jit_get_func_addr(const struct bpf_prog *prog,
1034 const struct bpf_insn *insn, bool extra_pass,
1035 u64 *func_addr, bool *func_addr_fixed);
1037 struct bpf_prog *bpf_jit_blind_constants(struct bpf_prog *fp);
1038 void bpf_jit_prog_release_other(struct bpf_prog *fp, struct bpf_prog *fp_other);
1040 static inline void bpf_jit_dump(unsigned int flen, unsigned int proglen,
1041 u32 pass, void *image)
1043 pr_err("flen=%u proglen=%u pass=%u image=%pK from=%s pid=%d\n", flen,
1044 proglen, pass, image, current->comm, task_pid_nr(current));
1047 print_hex_dump(KERN_ERR, "JIT code: ", DUMP_PREFIX_OFFSET,
1048 16, 1, image, proglen, false);
1051 static inline bool bpf_jit_is_ebpf(void)
1053 # ifdef CONFIG_HAVE_EBPF_JIT
1060 static inline bool ebpf_jit_enabled(void)
1062 return bpf_jit_enable && bpf_jit_is_ebpf();
1065 static inline bool bpf_prog_ebpf_jited(const struct bpf_prog *fp)
1067 return fp->jited && bpf_jit_is_ebpf();
1070 static inline bool bpf_jit_blinding_enabled(struct bpf_prog *prog)
1072 /* These are the prerequisites, should someone ever have the
1073 * idea to call blinding outside of them, we make sure to
1076 if (!bpf_jit_is_ebpf())
1078 if (!prog->jit_requested)
1080 if (!bpf_jit_harden)
1082 if (bpf_jit_harden == 1 && capable(CAP_SYS_ADMIN))
1088 static inline bool bpf_jit_kallsyms_enabled(void)
1090 /* There are a couple of corner cases where kallsyms should
1091 * not be enabled f.e. on hardening.
1095 if (!bpf_jit_kallsyms)
1097 if (bpf_jit_kallsyms == 1)
1103 const char *__bpf_address_lookup(unsigned long addr, unsigned long *size,
1104 unsigned long *off, char *sym);
1105 bool is_bpf_text_address(unsigned long addr);
1106 int bpf_get_kallsym(unsigned int symnum, unsigned long *value, char *type,
1109 static inline const char *
1110 bpf_address_lookup(unsigned long addr, unsigned long *size,
1111 unsigned long *off, char **modname, char *sym)
1113 const char *ret = __bpf_address_lookup(addr, size, off, sym);
1120 void bpf_prog_kallsyms_add(struct bpf_prog *fp);
1121 void bpf_prog_kallsyms_del(struct bpf_prog *fp);
1123 #else /* CONFIG_BPF_JIT */
1125 static inline bool ebpf_jit_enabled(void)
1130 static inline bool bpf_jit_blinding_enabled(struct bpf_prog *prog)
1135 static inline bool bpf_prog_ebpf_jited(const struct bpf_prog *fp)
1141 bpf_jit_add_poke_descriptor(struct bpf_prog *prog,
1142 struct bpf_jit_poke_descriptor *poke)
1147 static inline void bpf_jit_free(struct bpf_prog *fp)
1149 bpf_prog_unlock_free(fp);
1152 static inline bool bpf_jit_kallsyms_enabled(void)
1157 static inline const char *
1158 __bpf_address_lookup(unsigned long addr, unsigned long *size,
1159 unsigned long *off, char *sym)
1164 static inline bool is_bpf_text_address(unsigned long addr)
1169 static inline int bpf_get_kallsym(unsigned int symnum, unsigned long *value,
1170 char *type, char *sym)
1175 static inline const char *
1176 bpf_address_lookup(unsigned long addr, unsigned long *size,
1177 unsigned long *off, char **modname, char *sym)
1182 static inline void bpf_prog_kallsyms_add(struct bpf_prog *fp)
1186 static inline void bpf_prog_kallsyms_del(struct bpf_prog *fp)
1190 #endif /* CONFIG_BPF_JIT */
1192 void bpf_prog_kallsyms_del_all(struct bpf_prog *fp);
1194 #define BPF_ANC BIT(15)
1196 static inline bool bpf_needs_clear_a(const struct sock_filter *first)
1198 switch (first->code) {
1199 case BPF_RET | BPF_K:
1200 case BPF_LD | BPF_W | BPF_LEN:
1203 case BPF_LD | BPF_W | BPF_ABS:
1204 case BPF_LD | BPF_H | BPF_ABS:
1205 case BPF_LD | BPF_B | BPF_ABS:
1206 if (first->k == SKF_AD_OFF + SKF_AD_ALU_XOR_X)
1215 static inline u16 bpf_anc_helper(const struct sock_filter *ftest)
1217 BUG_ON(ftest->code & BPF_ANC);
1219 switch (ftest->code) {
1220 case BPF_LD | BPF_W | BPF_ABS:
1221 case BPF_LD | BPF_H | BPF_ABS:
1222 case BPF_LD | BPF_B | BPF_ABS:
1223 #define BPF_ANCILLARY(CODE) case SKF_AD_OFF + SKF_AD_##CODE: \
1224 return BPF_ANC | SKF_AD_##CODE
1226 BPF_ANCILLARY(PROTOCOL);
1227 BPF_ANCILLARY(PKTTYPE);
1228 BPF_ANCILLARY(IFINDEX);
1229 BPF_ANCILLARY(NLATTR);
1230 BPF_ANCILLARY(NLATTR_NEST);
1231 BPF_ANCILLARY(MARK);
1232 BPF_ANCILLARY(QUEUE);
1233 BPF_ANCILLARY(HATYPE);
1234 BPF_ANCILLARY(RXHASH);
1236 BPF_ANCILLARY(ALU_XOR_X);
1237 BPF_ANCILLARY(VLAN_TAG);
1238 BPF_ANCILLARY(VLAN_TAG_PRESENT);
1239 BPF_ANCILLARY(PAY_OFFSET);
1240 BPF_ANCILLARY(RANDOM);
1241 BPF_ANCILLARY(VLAN_TPID);
1249 void *bpf_internal_load_pointer_neg_helper(const struct sk_buff *skb,
1250 int k, unsigned int size);
1252 static inline int bpf_tell_extensions(void)
1257 struct bpf_sock_addr_kern {
1259 struct sockaddr *uaddr;
1260 /* Temporary "register" to make indirect stores to nested structures
1261 * defined above. We need three registers to make such a store, but
1262 * only two (src and dst) are available at convert_ctx_access time
1265 void *t_ctx; /* Attach type specific context. */
1268 struct bpf_sock_ops_kern {
1275 struct sk_buff *syn_skb;
1276 struct sk_buff *skb;
1280 u8 remaining_opt_len;
1281 u64 temp; /* temp and everything after is not
1282 * initialized to 0 before calling
1283 * the BPF program. New fields that
1284 * should be initialized to 0 should
1285 * be inserted before temp.
1286 * temp is scratch storage used by
1287 * sock_ops_convert_ctx_access
1288 * as temporary storage of a register.
1292 struct bpf_sysctl_kern {
1293 struct ctl_table_header *head;
1294 struct ctl_table *table;
1302 /* Temporary "register" for indirect stores to ppos. */
1306 #define BPF_SOCKOPT_KERN_BUF_SIZE 32
1307 struct bpf_sockopt_buf {
1308 u8 data[BPF_SOCKOPT_KERN_BUF_SIZE];
1311 struct bpf_sockopt_kern {
1321 int copy_bpf_fprog_from_user(struct sock_fprog *dst, sockptr_t src, int len);
1323 struct bpf_sk_lookup_kern {
1333 const struct in6_addr *saddr;
1334 const struct in6_addr *daddr;
1336 struct sock *selected_sk;
1340 extern struct static_key_false bpf_sk_lookup_enabled;
1342 /* Runners for BPF_SK_LOOKUP programs to invoke on socket lookup.
1344 * Allowed return values for a BPF SK_LOOKUP program are SK_PASS and
1345 * SK_DROP. Their meaning is as follows:
1347 * SK_PASS && ctx.selected_sk != NULL: use selected_sk as lookup result
1348 * SK_PASS && ctx.selected_sk == NULL: continue to htable-based socket lookup
1349 * SK_DROP : terminate lookup with -ECONNREFUSED
1351 * This macro aggregates return values and selected sockets from
1352 * multiple BPF programs according to following rules in order:
1354 * 1. If any program returned SK_PASS and a non-NULL ctx.selected_sk,
1355 * macro result is SK_PASS and last ctx.selected_sk is used.
1356 * 2. If any program returned SK_DROP return value,
1357 * macro result is SK_DROP.
1358 * 3. Otherwise result is SK_PASS and ctx.selected_sk is NULL.
1360 * Caller must ensure that the prog array is non-NULL, and that the
1361 * array as well as the programs it contains remain valid.
1363 #define BPF_PROG_SK_LOOKUP_RUN_ARRAY(array, ctx, func) \
1365 struct bpf_sk_lookup_kern *_ctx = &(ctx); \
1366 struct bpf_prog_array_item *_item; \
1367 struct sock *_selected_sk = NULL; \
1368 bool _no_reuseport = false; \
1369 struct bpf_prog *_prog; \
1370 bool _all_pass = true; \
1373 migrate_disable(); \
1374 _item = &(array)->items[0]; \
1375 while ((_prog = READ_ONCE(_item->prog))) { \
1376 /* restore most recent selection */ \
1377 _ctx->selected_sk = _selected_sk; \
1378 _ctx->no_reuseport = _no_reuseport; \
1380 _ret = func(_prog, _ctx); \
1381 if (_ret == SK_PASS && _ctx->selected_sk) { \
1382 /* remember last non-NULL socket */ \
1383 _selected_sk = _ctx->selected_sk; \
1384 _no_reuseport = _ctx->no_reuseport; \
1385 } else if (_ret == SK_DROP && _all_pass) { \
1386 _all_pass = false; \
1390 _ctx->selected_sk = _selected_sk; \
1391 _ctx->no_reuseport = _no_reuseport; \
1393 _all_pass || _selected_sk ? SK_PASS : SK_DROP; \
1396 static inline bool bpf_sk_lookup_run_v4(struct net *net, int protocol,
1397 const __be32 saddr, const __be16 sport,
1398 const __be32 daddr, const u16 dport,
1401 struct bpf_prog_array *run_array;
1402 struct sock *selected_sk = NULL;
1403 bool no_reuseport = false;
1406 run_array = rcu_dereference(net->bpf.run_array[NETNS_BPF_SK_LOOKUP]);
1408 struct bpf_sk_lookup_kern ctx = {
1410 .protocol = protocol,
1418 act = BPF_PROG_SK_LOOKUP_RUN_ARRAY(run_array, ctx, BPF_PROG_RUN);
1419 if (act == SK_PASS) {
1420 selected_sk = ctx.selected_sk;
1421 no_reuseport = ctx.no_reuseport;
1423 selected_sk = ERR_PTR(-ECONNREFUSED);
1428 return no_reuseport;
1431 #if IS_ENABLED(CONFIG_IPV6)
1432 static inline bool bpf_sk_lookup_run_v6(struct net *net, int protocol,
1433 const struct in6_addr *saddr,
1435 const struct in6_addr *daddr,
1439 struct bpf_prog_array *run_array;
1440 struct sock *selected_sk = NULL;
1441 bool no_reuseport = false;
1444 run_array = rcu_dereference(net->bpf.run_array[NETNS_BPF_SK_LOOKUP]);
1446 struct bpf_sk_lookup_kern ctx = {
1448 .protocol = protocol,
1456 act = BPF_PROG_SK_LOOKUP_RUN_ARRAY(run_array, ctx, BPF_PROG_RUN);
1457 if (act == SK_PASS) {
1458 selected_sk = ctx.selected_sk;
1459 no_reuseport = ctx.no_reuseport;
1461 selected_sk = ERR_PTR(-ECONNREFUSED);
1466 return no_reuseport;
1468 #endif /* IS_ENABLED(CONFIG_IPV6) */
1470 static __always_inline int __bpf_xdp_redirect_map(struct bpf_map *map, u32 ifindex,
1471 u64 flags, const u64 flag_mask,
1472 void *lookup_elem(struct bpf_map *map, u32 key))
1474 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
1475 const u64 action_mask = XDP_ABORTED | XDP_DROP | XDP_PASS | XDP_TX;
1477 /* Lower bits of the flags are used as return code on lookup failure */
1478 if (unlikely(flags & ~(action_mask | flag_mask)))
1481 ri->tgt_value = lookup_elem(map, ifindex);
1482 if (unlikely(!ri->tgt_value) && !(flags & BPF_F_BROADCAST)) {
1483 /* If the lookup fails we want to clear out the state in the
1484 * redirect_info struct completely, so that if an eBPF program
1485 * performs multiple lookups, the last one always takes
1488 ri->map_id = INT_MAX; /* Valid map id idr range: [1,INT_MAX[ */
1489 ri->map_type = BPF_MAP_TYPE_UNSPEC;
1490 return flags & action_mask;
1493 ri->tgt_index = ifindex;
1494 ri->map_id = map->id;
1495 ri->map_type = map->map_type;
1497 if (flags & BPF_F_BROADCAST) {
1498 WRITE_ONCE(ri->map, map);
1501 WRITE_ONCE(ri->map, NULL);
1505 return XDP_REDIRECT;
1508 #endif /* __LINUX_FILTER_H__ */