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/sha.h>
26 #include <net/sch_generic.h>
28 #include <asm/byteorder.h>
29 #include <uapi/linux/filter.h>
30 #include <uapi/linux/bpf.h>
38 struct sock_reuseport;
40 struct ctl_table_header;
42 /* ArgX, context and stack frame pointer register positions. Note,
43 * Arg1, Arg2, Arg3, etc are used as argument mappings of function
44 * calls in BPF_CALL instruction.
46 #define BPF_REG_ARG1 BPF_REG_1
47 #define BPF_REG_ARG2 BPF_REG_2
48 #define BPF_REG_ARG3 BPF_REG_3
49 #define BPF_REG_ARG4 BPF_REG_4
50 #define BPF_REG_ARG5 BPF_REG_5
51 #define BPF_REG_CTX BPF_REG_6
52 #define BPF_REG_FP BPF_REG_10
54 /* Additional register mappings for converted user programs. */
55 #define BPF_REG_A BPF_REG_0
56 #define BPF_REG_X BPF_REG_7
57 #define BPF_REG_TMP BPF_REG_2 /* scratch reg */
58 #define BPF_REG_D BPF_REG_8 /* data, callee-saved */
59 #define BPF_REG_H BPF_REG_9 /* hlen, callee-saved */
61 /* Kernel hidden auxiliary/helper register. */
62 #define BPF_REG_AX MAX_BPF_REG
63 #define MAX_BPF_EXT_REG (MAX_BPF_REG + 1)
64 #define MAX_BPF_JIT_REG MAX_BPF_EXT_REG
66 /* unused opcode to mark special call to bpf_tail_call() helper */
67 #define BPF_TAIL_CALL 0xf0
69 /* unused opcode to mark special load instruction. Same as BPF_ABS */
70 #define BPF_PROBE_MEM 0x20
72 /* unused opcode to mark call to interpreter with arguments */
73 #define BPF_CALL_ARGS 0xe0
75 /* unused opcode to mark speculation barrier for mitigating
76 * Speculative Store Bypass
78 #define BPF_NOSPEC 0xc0
80 /* As per nm, we expose JITed images as text (code) section for
81 * kallsyms. That way, tools like perf can find it to match
84 #define BPF_SYM_ELF_TYPE 't'
86 /* BPF program can access up to 512 bytes of stack space. */
87 #define MAX_BPF_STACK 512
89 /* Helper macros for filter block array initializers. */
91 /* ALU ops on registers, bpf_add|sub|...: dst_reg += src_reg */
93 #define BPF_ALU64_REG(OP, DST, SRC) \
94 ((struct bpf_insn) { \
95 .code = BPF_ALU64 | BPF_OP(OP) | BPF_X, \
101 #define BPF_ALU32_REG(OP, DST, SRC) \
102 ((struct bpf_insn) { \
103 .code = BPF_ALU | BPF_OP(OP) | BPF_X, \
109 /* ALU ops on immediates, bpf_add|sub|...: dst_reg += imm32 */
111 #define BPF_ALU64_IMM(OP, DST, IMM) \
112 ((struct bpf_insn) { \
113 .code = BPF_ALU64 | BPF_OP(OP) | BPF_K, \
119 #define BPF_ALU32_IMM(OP, DST, IMM) \
120 ((struct bpf_insn) { \
121 .code = BPF_ALU | BPF_OP(OP) | BPF_K, \
127 /* Endianess conversion, cpu_to_{l,b}e(), {l,b}e_to_cpu() */
129 #define BPF_ENDIAN(TYPE, DST, LEN) \
130 ((struct bpf_insn) { \
131 .code = BPF_ALU | BPF_END | BPF_SRC(TYPE), \
137 /* Short form of mov, dst_reg = src_reg */
139 #define BPF_MOV64_REG(DST, SRC) \
140 ((struct bpf_insn) { \
141 .code = BPF_ALU64 | BPF_MOV | BPF_X, \
147 #define BPF_MOV32_REG(DST, SRC) \
148 ((struct bpf_insn) { \
149 .code = BPF_ALU | BPF_MOV | BPF_X, \
155 /* Short form of mov, dst_reg = imm32 */
157 #define BPF_MOV64_IMM(DST, IMM) \
158 ((struct bpf_insn) { \
159 .code = BPF_ALU64 | BPF_MOV | BPF_K, \
165 #define BPF_MOV32_IMM(DST, IMM) \
166 ((struct bpf_insn) { \
167 .code = BPF_ALU | BPF_MOV | BPF_K, \
173 /* Special form of mov32, used for doing explicit zero extension on dst. */
174 #define BPF_ZEXT_REG(DST) \
175 ((struct bpf_insn) { \
176 .code = BPF_ALU | BPF_MOV | BPF_X, \
182 static inline bool insn_is_zext(const struct bpf_insn *insn)
184 return insn->code == (BPF_ALU | BPF_MOV | BPF_X) && insn->imm == 1;
187 /* BPF_LD_IMM64 macro encodes single 'load 64-bit immediate' insn */
188 #define BPF_LD_IMM64(DST, IMM) \
189 BPF_LD_IMM64_RAW(DST, 0, IMM)
191 #define BPF_LD_IMM64_RAW(DST, SRC, IMM) \
192 ((struct bpf_insn) { \
193 .code = BPF_LD | BPF_DW | BPF_IMM, \
197 .imm = (__u32) (IMM) }), \
198 ((struct bpf_insn) { \
199 .code = 0, /* zero is reserved opcode */ \
203 .imm = ((__u64) (IMM)) >> 32 })
205 /* pseudo BPF_LD_IMM64 insn used to refer to process-local map_fd */
206 #define BPF_LD_MAP_FD(DST, MAP_FD) \
207 BPF_LD_IMM64_RAW(DST, BPF_PSEUDO_MAP_FD, MAP_FD)
209 /* Short form of mov based on type, BPF_X: dst_reg = src_reg, BPF_K: dst_reg = imm32 */
211 #define BPF_MOV64_RAW(TYPE, DST, SRC, IMM) \
212 ((struct bpf_insn) { \
213 .code = BPF_ALU64 | BPF_MOV | BPF_SRC(TYPE), \
219 #define BPF_MOV32_RAW(TYPE, DST, SRC, IMM) \
220 ((struct bpf_insn) { \
221 .code = BPF_ALU | BPF_MOV | BPF_SRC(TYPE), \
227 /* Direct packet access, R0 = *(uint *) (skb->data + imm32) */
229 #define BPF_LD_ABS(SIZE, IMM) \
230 ((struct bpf_insn) { \
231 .code = BPF_LD | BPF_SIZE(SIZE) | BPF_ABS, \
237 /* Indirect packet access, R0 = *(uint *) (skb->data + src_reg + imm32) */
239 #define BPF_LD_IND(SIZE, SRC, IMM) \
240 ((struct bpf_insn) { \
241 .code = BPF_LD | BPF_SIZE(SIZE) | BPF_IND, \
247 /* Memory load, dst_reg = *(uint *) (src_reg + off16) */
249 #define BPF_LDX_MEM(SIZE, DST, SRC, OFF) \
250 ((struct bpf_insn) { \
251 .code = BPF_LDX | BPF_SIZE(SIZE) | BPF_MEM, \
257 /* Memory store, *(uint *) (dst_reg + off16) = src_reg */
259 #define BPF_STX_MEM(SIZE, DST, SRC, OFF) \
260 ((struct bpf_insn) { \
261 .code = BPF_STX | BPF_SIZE(SIZE) | BPF_MEM, \
267 /* Atomic memory add, *(uint *)(dst_reg + off16) += src_reg */
269 #define BPF_STX_XADD(SIZE, DST, SRC, OFF) \
270 ((struct bpf_insn) { \
271 .code = BPF_STX | BPF_SIZE(SIZE) | BPF_XADD, \
277 /* Memory store, *(uint *) (dst_reg + off16) = imm32 */
279 #define BPF_ST_MEM(SIZE, DST, OFF, IMM) \
280 ((struct bpf_insn) { \
281 .code = BPF_ST | BPF_SIZE(SIZE) | BPF_MEM, \
287 /* Conditional jumps against registers, if (dst_reg 'op' src_reg) goto pc + off16 */
289 #define BPF_JMP_REG(OP, DST, SRC, OFF) \
290 ((struct bpf_insn) { \
291 .code = BPF_JMP | BPF_OP(OP) | BPF_X, \
297 /* Conditional jumps against immediates, if (dst_reg 'op' imm32) goto pc + off16 */
299 #define BPF_JMP_IMM(OP, DST, IMM, OFF) \
300 ((struct bpf_insn) { \
301 .code = BPF_JMP | BPF_OP(OP) | BPF_K, \
307 /* Like BPF_JMP_REG, but with 32-bit wide operands for comparison. */
309 #define BPF_JMP32_REG(OP, DST, SRC, OFF) \
310 ((struct bpf_insn) { \
311 .code = BPF_JMP32 | BPF_OP(OP) | BPF_X, \
317 /* Like BPF_JMP_IMM, but with 32-bit wide operands for comparison. */
319 #define BPF_JMP32_IMM(OP, DST, IMM, OFF) \
320 ((struct bpf_insn) { \
321 .code = BPF_JMP32 | BPF_OP(OP) | BPF_K, \
327 /* Unconditional jumps, goto pc + off16 */
329 #define BPF_JMP_A(OFF) \
330 ((struct bpf_insn) { \
331 .code = BPF_JMP | BPF_JA, \
339 #define BPF_CALL_REL(TGT) \
340 ((struct bpf_insn) { \
341 .code = BPF_JMP | BPF_CALL, \
343 .src_reg = BPF_PSEUDO_CALL, \
349 #define BPF_CAST_CALL(x) \
350 ((u64 (*)(u64, u64, u64, u64, u64))(x))
352 #define BPF_EMIT_CALL(FUNC) \
353 ((struct bpf_insn) { \
354 .code = BPF_JMP | BPF_CALL, \
358 .imm = ((FUNC) - __bpf_call_base) })
360 /* Raw code statement block */
362 #define BPF_RAW_INSN(CODE, DST, SRC, OFF, IMM) \
363 ((struct bpf_insn) { \
372 #define BPF_EXIT_INSN() \
373 ((struct bpf_insn) { \
374 .code = BPF_JMP | BPF_EXIT, \
380 /* Speculation barrier */
382 #define BPF_ST_NOSPEC() \
383 ((struct bpf_insn) { \
384 .code = BPF_ST | BPF_NOSPEC, \
390 /* Internal classic blocks for direct assignment */
392 #define __BPF_STMT(CODE, K) \
393 ((struct sock_filter) BPF_STMT(CODE, K))
395 #define __BPF_JUMP(CODE, K, JT, JF) \
396 ((struct sock_filter) BPF_JUMP(CODE, K, JT, JF))
398 #define bytes_to_bpf_size(bytes) \
400 int bpf_size = -EINVAL; \
402 if (bytes == sizeof(u8)) \
404 else if (bytes == sizeof(u16)) \
406 else if (bytes == sizeof(u32)) \
408 else if (bytes == sizeof(u64)) \
414 #define bpf_size_to_bytes(bpf_size) \
416 int bytes = -EINVAL; \
418 if (bpf_size == BPF_B) \
419 bytes = sizeof(u8); \
420 else if (bpf_size == BPF_H) \
421 bytes = sizeof(u16); \
422 else if (bpf_size == BPF_W) \
423 bytes = sizeof(u32); \
424 else if (bpf_size == BPF_DW) \
425 bytes = sizeof(u64); \
430 #define BPF_SIZEOF(type) \
432 const int __size = bytes_to_bpf_size(sizeof(type)); \
433 BUILD_BUG_ON(__size < 0); \
437 #define BPF_FIELD_SIZEOF(type, field) \
439 const int __size = bytes_to_bpf_size(sizeof_field(type, field)); \
440 BUILD_BUG_ON(__size < 0); \
444 #define BPF_LDST_BYTES(insn) \
446 const int __size = bpf_size_to_bytes(BPF_SIZE((insn)->code)); \
447 WARN_ON(__size < 0); \
451 #define __BPF_MAP_0(m, v, ...) v
452 #define __BPF_MAP_1(m, v, t, a, ...) m(t, a)
453 #define __BPF_MAP_2(m, v, t, a, ...) m(t, a), __BPF_MAP_1(m, v, __VA_ARGS__)
454 #define __BPF_MAP_3(m, v, t, a, ...) m(t, a), __BPF_MAP_2(m, v, __VA_ARGS__)
455 #define __BPF_MAP_4(m, v, t, a, ...) m(t, a), __BPF_MAP_3(m, v, __VA_ARGS__)
456 #define __BPF_MAP_5(m, v, t, a, ...) m(t, a), __BPF_MAP_4(m, v, __VA_ARGS__)
458 #define __BPF_REG_0(...) __BPF_PAD(5)
459 #define __BPF_REG_1(...) __BPF_MAP(1, __VA_ARGS__), __BPF_PAD(4)
460 #define __BPF_REG_2(...) __BPF_MAP(2, __VA_ARGS__), __BPF_PAD(3)
461 #define __BPF_REG_3(...) __BPF_MAP(3, __VA_ARGS__), __BPF_PAD(2)
462 #define __BPF_REG_4(...) __BPF_MAP(4, __VA_ARGS__), __BPF_PAD(1)
463 #define __BPF_REG_5(...) __BPF_MAP(5, __VA_ARGS__)
465 #define __BPF_MAP(n, ...) __BPF_MAP_##n(__VA_ARGS__)
466 #define __BPF_REG(n, ...) __BPF_REG_##n(__VA_ARGS__)
468 #define __BPF_CAST(t, a) \
471 typeof(__builtin_choose_expr(sizeof(t) == sizeof(unsigned long), \
472 (unsigned long)0, (t)0))) a
476 #define __BPF_DECL_ARGS(t, a) t a
477 #define __BPF_DECL_REGS(t, a) u64 a
479 #define __BPF_PAD(n) \
480 __BPF_MAP(n, __BPF_DECL_ARGS, __BPF_N, u64, __ur_1, u64, __ur_2, \
481 u64, __ur_3, u64, __ur_4, u64, __ur_5)
483 #define BPF_CALL_x(x, name, ...) \
484 static __always_inline \
485 u64 ____##name(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__)); \
486 typedef u64 (*btf_##name)(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__)); \
487 u64 name(__BPF_REG(x, __BPF_DECL_REGS, __BPF_N, __VA_ARGS__)); \
488 u64 name(__BPF_REG(x, __BPF_DECL_REGS, __BPF_N, __VA_ARGS__)) \
490 return ((btf_##name)____##name)(__BPF_MAP(x,__BPF_CAST,__BPF_N,__VA_ARGS__));\
492 static __always_inline \
493 u64 ____##name(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__))
495 #define BPF_CALL_0(name, ...) BPF_CALL_x(0, name, __VA_ARGS__)
496 #define BPF_CALL_1(name, ...) BPF_CALL_x(1, name, __VA_ARGS__)
497 #define BPF_CALL_2(name, ...) BPF_CALL_x(2, name, __VA_ARGS__)
498 #define BPF_CALL_3(name, ...) BPF_CALL_x(3, name, __VA_ARGS__)
499 #define BPF_CALL_4(name, ...) BPF_CALL_x(4, name, __VA_ARGS__)
500 #define BPF_CALL_5(name, ...) BPF_CALL_x(5, name, __VA_ARGS__)
502 #define bpf_ctx_range(TYPE, MEMBER) \
503 offsetof(TYPE, MEMBER) ... offsetofend(TYPE, MEMBER) - 1
504 #define bpf_ctx_range_till(TYPE, MEMBER1, MEMBER2) \
505 offsetof(TYPE, MEMBER1) ... offsetofend(TYPE, MEMBER2) - 1
506 #if BITS_PER_LONG == 64
507 # define bpf_ctx_range_ptr(TYPE, MEMBER) \
508 offsetof(TYPE, MEMBER) ... offsetofend(TYPE, MEMBER) - 1
510 # define bpf_ctx_range_ptr(TYPE, MEMBER) \
511 offsetof(TYPE, MEMBER) ... offsetof(TYPE, MEMBER) + 8 - 1
512 #endif /* BITS_PER_LONG == 64 */
514 #define bpf_target_off(TYPE, MEMBER, SIZE, PTR_SIZE) \
516 BUILD_BUG_ON(sizeof_field(TYPE, MEMBER) != (SIZE)); \
517 *(PTR_SIZE) = (SIZE); \
518 offsetof(TYPE, MEMBER); \
521 /* A struct sock_filter is architecture independent. */
522 struct compat_sock_fprog {
524 compat_uptr_t filter; /* struct sock_filter * */
527 struct sock_fprog_kern {
529 struct sock_filter *filter;
532 /* Some arches need doubleword alignment for their instructions and/or data */
533 #define BPF_IMAGE_ALIGNMENT 8
535 struct bpf_binary_header {
537 u8 image[] __aligned(BPF_IMAGE_ALIGNMENT);
541 u16 pages; /* Number of allocated pages */
542 u16 jited:1, /* Is our filter JIT'ed? */
543 jit_requested:1,/* archs need to JIT the prog */
544 gpl_compatible:1, /* Is filter GPL compatible? */
545 cb_access:1, /* Is control block accessed? */
546 dst_needed:1, /* Do we need dst entry? */
547 blinded:1, /* Was blinded */
548 is_func:1, /* program is a bpf function */
549 kprobe_override:1, /* Do we override a kprobe? */
550 has_callchain_buf:1, /* callchain buffer allocated? */
551 enforce_expected_attach_type:1, /* Enforce expected_attach_type checking at attach time */
552 call_get_stack:1; /* Do we call bpf_get_stack() or bpf_get_stackid() */
553 enum bpf_prog_type type; /* Type of BPF program */
554 enum bpf_attach_type expected_attach_type; /* For some prog types */
555 u32 len; /* Number of filter blocks */
556 u32 jited_len; /* Size of jited insns in bytes */
557 u8 tag[BPF_TAG_SIZE];
558 struct bpf_prog_aux *aux; /* Auxiliary fields */
559 struct sock_fprog_kern *orig_prog; /* Original BPF program */
560 unsigned int (*bpf_func)(const void *ctx,
561 const struct bpf_insn *insn);
562 /* Instructions for interpreter */
563 struct sock_filter insns[0];
564 struct bpf_insn insnsi[];
570 struct bpf_prog *prog;
573 DECLARE_STATIC_KEY_FALSE(bpf_stats_enabled_key);
575 #define __BPF_PROG_RUN(prog, ctx, dfunc) ({ \
578 if (static_branch_unlikely(&bpf_stats_enabled_key)) { \
579 struct bpf_prog_stats *__stats; \
580 u64 __start = sched_clock(); \
581 __ret = dfunc(ctx, (prog)->insnsi, (prog)->bpf_func); \
582 __stats = this_cpu_ptr(prog->aux->stats); \
583 u64_stats_update_begin(&__stats->syncp); \
585 __stats->nsecs += sched_clock() - __start; \
586 u64_stats_update_end(&__stats->syncp); \
588 __ret = dfunc(ctx, (prog)->insnsi, (prog)->bpf_func); \
592 #define BPF_PROG_RUN(prog, ctx) \
593 __BPF_PROG_RUN(prog, ctx, bpf_dispatcher_nop_func)
596 * Use in preemptible and therefore migratable context to make sure that
597 * the execution of the BPF program runs on one CPU.
599 * This uses migrate_disable/enable() explicitly to document that the
600 * invocation of a BPF program does not require reentrancy protection
601 * against a BPF program which is invoked from a preempting task.
603 * For non RT enabled kernels migrate_disable/enable() maps to
604 * preempt_disable/enable(), i.e. it disables also preemption.
606 static inline u32 bpf_prog_run_pin_on_cpu(const struct bpf_prog *prog,
612 ret = __BPF_PROG_RUN(prog, ctx, bpf_dispatcher_nop_func);
617 #define BPF_SKB_CB_LEN QDISC_CB_PRIV_LEN
619 struct bpf_skb_data_end {
620 struct qdisc_skb_cb qdisc_cb;
625 struct bpf_nh_params {
629 struct in6_addr ipv6_nh;
633 struct bpf_redirect_info {
639 struct bpf_nh_params nh;
642 DECLARE_PER_CPU(struct bpf_redirect_info, bpf_redirect_info);
644 /* flags for bpf_redirect_info kern_flags */
645 #define BPF_RI_F_RF_NO_DIRECT BIT(0) /* no napi_direct on return_frame */
647 /* Compute the linear packet data range [data, data_end) which
648 * will be accessed by various program types (cls_bpf, act_bpf,
649 * lwt, ...). Subsystems allowing direct data access must (!)
650 * ensure that cb[] area can be written to when BPF program is
651 * invoked (otherwise cb[] save/restore is necessary).
653 static inline void bpf_compute_data_pointers(struct sk_buff *skb)
655 struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb;
657 BUILD_BUG_ON(sizeof(*cb) > sizeof_field(struct sk_buff, cb));
658 cb->data_meta = skb->data - skb_metadata_len(skb);
659 cb->data_end = skb->data + skb_headlen(skb);
662 /* Similar to bpf_compute_data_pointers(), except that save orginal
663 * data in cb->data and cb->meta_data for restore.
665 static inline void bpf_compute_and_save_data_end(
666 struct sk_buff *skb, void **saved_data_end)
668 struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb;
670 *saved_data_end = cb->data_end;
671 cb->data_end = skb->data + skb_headlen(skb);
674 /* Restore data saved by bpf_compute_data_pointers(). */
675 static inline void bpf_restore_data_end(
676 struct sk_buff *skb, void *saved_data_end)
678 struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb;
680 cb->data_end = saved_data_end;
683 static inline u8 *bpf_skb_cb(struct sk_buff *skb)
685 /* eBPF programs may read/write skb->cb[] area to transfer meta
686 * data between tail calls. Since this also needs to work with
687 * tc, that scratch memory is mapped to qdisc_skb_cb's data area.
689 * In some socket filter cases, the cb unfortunately needs to be
690 * saved/restored so that protocol specific skb->cb[] data won't
691 * be lost. In any case, due to unpriviledged eBPF programs
692 * attached to sockets, we need to clear the bpf_skb_cb() area
693 * to not leak previous contents to user space.
695 BUILD_BUG_ON(sizeof_field(struct __sk_buff, cb) != BPF_SKB_CB_LEN);
696 BUILD_BUG_ON(sizeof_field(struct __sk_buff, cb) !=
697 sizeof_field(struct qdisc_skb_cb, data));
699 return qdisc_skb_cb(skb)->data;
702 /* Must be invoked with migration disabled */
703 static inline u32 __bpf_prog_run_save_cb(const struct bpf_prog *prog,
706 u8 *cb_data = bpf_skb_cb(skb);
707 u8 cb_saved[BPF_SKB_CB_LEN];
710 if (unlikely(prog->cb_access)) {
711 memcpy(cb_saved, cb_data, sizeof(cb_saved));
712 memset(cb_data, 0, sizeof(cb_saved));
715 res = BPF_PROG_RUN(prog, skb);
717 if (unlikely(prog->cb_access))
718 memcpy(cb_data, cb_saved, sizeof(cb_saved));
723 static inline u32 bpf_prog_run_save_cb(const struct bpf_prog *prog,
729 res = __bpf_prog_run_save_cb(prog, skb);
734 static inline u32 bpf_prog_run_clear_cb(const struct bpf_prog *prog,
737 u8 *cb_data = bpf_skb_cb(skb);
740 if (unlikely(prog->cb_access))
741 memset(cb_data, 0, BPF_SKB_CB_LEN);
743 res = bpf_prog_run_pin_on_cpu(prog, skb);
747 DECLARE_BPF_DISPATCHER(xdp)
749 static __always_inline u32 bpf_prog_run_xdp(const struct bpf_prog *prog,
750 struct xdp_buff *xdp)
752 /* Caller needs to hold rcu_read_lock() (!), otherwise program
753 * can be released while still running, or map elements could be
754 * freed early while still having concurrent users. XDP fastpath
755 * already takes rcu_read_lock() when fetching the program, so
756 * it's not necessary here anymore.
758 return __BPF_PROG_RUN(prog, xdp, BPF_DISPATCHER_FUNC(xdp));
761 void bpf_prog_change_xdp(struct bpf_prog *prev_prog, struct bpf_prog *prog);
763 static inline u32 bpf_prog_insn_size(const struct bpf_prog *prog)
765 return prog->len * sizeof(struct bpf_insn);
768 static inline u32 bpf_prog_tag_scratch_size(const struct bpf_prog *prog)
770 return round_up(bpf_prog_insn_size(prog) +
771 sizeof(__be64) + 1, SHA1_BLOCK_SIZE);
774 static inline unsigned int bpf_prog_size(unsigned int proglen)
776 return max(sizeof(struct bpf_prog),
777 offsetof(struct bpf_prog, insns[proglen]));
780 static inline bool bpf_prog_was_classic(const struct bpf_prog *prog)
782 /* When classic BPF programs have been loaded and the arch
783 * does not have a classic BPF JIT (anymore), they have been
784 * converted via bpf_migrate_filter() to eBPF and thus always
785 * have an unspec program type.
787 return prog->type == BPF_PROG_TYPE_UNSPEC;
790 static inline u32 bpf_ctx_off_adjust_machine(u32 size)
792 const u32 size_machine = sizeof(unsigned long);
794 if (size > size_machine && size % size_machine == 0)
801 bpf_ctx_narrow_access_ok(u32 off, u32 size, u32 size_default)
803 return size <= size_default && (size & (size - 1)) == 0;
807 bpf_ctx_narrow_access_offset(u32 off, u32 size, u32 size_default)
809 u8 access_off = off & (size_default - 1);
811 #ifdef __LITTLE_ENDIAN
814 return size_default - (access_off + size);
818 #define bpf_ctx_wide_access_ok(off, size, type, field) \
819 (size == sizeof(__u64) && \
820 off >= offsetof(type, field) && \
821 off + sizeof(__u64) <= offsetofend(type, field) && \
822 off % sizeof(__u64) == 0)
824 #define bpf_classic_proglen(fprog) (fprog->len * sizeof(fprog->filter[0]))
826 static inline void bpf_prog_lock_ro(struct bpf_prog *fp)
828 #ifndef CONFIG_BPF_JIT_ALWAYS_ON
830 set_vm_flush_reset_perms(fp);
831 set_memory_ro((unsigned long)fp, fp->pages);
836 static inline void bpf_jit_binary_lock_ro(struct bpf_binary_header *hdr)
838 set_vm_flush_reset_perms(hdr);
839 set_memory_ro((unsigned long)hdr, hdr->pages);
840 set_memory_x((unsigned long)hdr, hdr->pages);
843 static inline struct bpf_binary_header *
844 bpf_jit_binary_hdr(const struct bpf_prog *fp)
846 unsigned long real_start = (unsigned long)fp->bpf_func;
847 unsigned long addr = real_start & PAGE_MASK;
852 int sk_filter_trim_cap(struct sock *sk, struct sk_buff *skb, unsigned int cap);
853 static inline int sk_filter(struct sock *sk, struct sk_buff *skb)
855 return sk_filter_trim_cap(sk, skb, 1);
858 struct bpf_prog *bpf_prog_select_runtime(struct bpf_prog *fp, int *err);
859 void bpf_prog_free(struct bpf_prog *fp);
861 bool bpf_opcode_in_insntable(u8 code);
863 void bpf_prog_free_linfo(struct bpf_prog *prog);
864 void bpf_prog_fill_jited_linfo(struct bpf_prog *prog,
865 const u32 *insn_to_jit_off);
866 int bpf_prog_alloc_jited_linfo(struct bpf_prog *prog);
867 void bpf_prog_free_jited_linfo(struct bpf_prog *prog);
868 void bpf_prog_free_unused_jited_linfo(struct bpf_prog *prog);
870 struct bpf_prog *bpf_prog_alloc(unsigned int size, gfp_t gfp_extra_flags);
871 struct bpf_prog *bpf_prog_alloc_no_stats(unsigned int size, gfp_t gfp_extra_flags);
872 struct bpf_prog *bpf_prog_realloc(struct bpf_prog *fp_old, unsigned int size,
873 gfp_t gfp_extra_flags);
874 void __bpf_prog_free(struct bpf_prog *fp);
876 static inline void bpf_prog_unlock_free(struct bpf_prog *fp)
881 typedef int (*bpf_aux_classic_check_t)(struct sock_filter *filter,
884 int bpf_prog_create(struct bpf_prog **pfp, struct sock_fprog_kern *fprog);
885 int bpf_prog_create_from_user(struct bpf_prog **pfp, struct sock_fprog *fprog,
886 bpf_aux_classic_check_t trans, bool save_orig);
887 void bpf_prog_destroy(struct bpf_prog *fp);
889 int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk);
890 int sk_attach_bpf(u32 ufd, struct sock *sk);
891 int sk_reuseport_attach_filter(struct sock_fprog *fprog, struct sock *sk);
892 int sk_reuseport_attach_bpf(u32 ufd, struct sock *sk);
893 void sk_reuseport_prog_free(struct bpf_prog *prog);
894 int sk_detach_filter(struct sock *sk);
895 int sk_get_filter(struct sock *sk, struct sock_filter __user *filter,
898 bool sk_filter_charge(struct sock *sk, struct sk_filter *fp);
899 void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp);
901 u64 __bpf_call_base(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5);
902 #define __bpf_call_base_args \
903 ((u64 (*)(u64, u64, u64, u64, u64, const struct bpf_insn *)) \
904 (void *)__bpf_call_base)
906 struct bpf_prog *bpf_int_jit_compile(struct bpf_prog *prog);
907 void bpf_jit_compile(struct bpf_prog *prog);
908 bool bpf_jit_needs_zext(void);
909 bool bpf_helper_changes_pkt_data(void *func);
911 static inline bool bpf_dump_raw_ok(const struct cred *cred)
913 /* Reconstruction of call-sites is dependent on kallsyms,
914 * thus make dump the same restriction.
916 return kallsyms_show_value(cred);
919 struct bpf_prog *bpf_patch_insn_single(struct bpf_prog *prog, u32 off,
920 const struct bpf_insn *patch, u32 len);
921 int bpf_remove_insns(struct bpf_prog *prog, u32 off, u32 cnt);
923 void bpf_clear_redirect_map(struct bpf_map *map);
925 static inline bool xdp_return_frame_no_direct(void)
927 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
929 return ri->kern_flags & BPF_RI_F_RF_NO_DIRECT;
932 static inline void xdp_set_return_frame_no_direct(void)
934 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
936 ri->kern_flags |= BPF_RI_F_RF_NO_DIRECT;
939 static inline void xdp_clear_return_frame_no_direct(void)
941 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
943 ri->kern_flags &= ~BPF_RI_F_RF_NO_DIRECT;
946 static inline int xdp_ok_fwd_dev(const struct net_device *fwd,
951 if (unlikely(!(fwd->flags & IFF_UP)))
954 len = fwd->mtu + fwd->hard_header_len + VLAN_HLEN;
961 /* The pair of xdp_do_redirect and xdp_do_flush MUST be called in the
962 * same cpu context. Further for best results no more than a single map
963 * for the do_redirect/do_flush pair should be used. This limitation is
964 * because we only track one map and force a flush when the map changes.
965 * This does not appear to be a real limitation for existing software.
967 int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb,
968 struct xdp_buff *xdp, struct bpf_prog *prog);
969 int xdp_do_redirect(struct net_device *dev,
970 struct xdp_buff *xdp,
971 struct bpf_prog *prog);
972 void xdp_do_flush(void);
974 /* The xdp_do_flush_map() helper has been renamed to drop the _map suffix, as
975 * it is no longer only flushing maps. Keep this define for compatibility
976 * until all drivers are updated - do not use xdp_do_flush_map() in new code!
978 #define xdp_do_flush_map xdp_do_flush
980 void bpf_warn_invalid_xdp_action(u32 act);
983 struct sock *bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk,
984 struct bpf_prog *prog, struct sk_buff *skb,
987 static inline struct sock *
988 bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk,
989 struct bpf_prog *prog, struct sk_buff *skb,
996 #ifdef CONFIG_BPF_JIT
997 extern int bpf_jit_enable;
998 extern int bpf_jit_harden;
999 extern int bpf_jit_kallsyms;
1000 extern long bpf_jit_limit;
1001 extern long bpf_jit_limit_max;
1003 typedef void (*bpf_jit_fill_hole_t)(void *area, unsigned int size);
1005 struct bpf_binary_header *
1006 bpf_jit_binary_alloc(unsigned int proglen, u8 **image_ptr,
1007 unsigned int alignment,
1008 bpf_jit_fill_hole_t bpf_fill_ill_insns);
1009 void bpf_jit_binary_free(struct bpf_binary_header *hdr);
1010 u64 bpf_jit_alloc_exec_limit(void);
1011 void *bpf_jit_alloc_exec(unsigned long size);
1012 void bpf_jit_free_exec(void *addr);
1013 void bpf_jit_free(struct bpf_prog *fp);
1015 int bpf_jit_add_poke_descriptor(struct bpf_prog *prog,
1016 struct bpf_jit_poke_descriptor *poke);
1018 int bpf_jit_get_func_addr(const struct bpf_prog *prog,
1019 const struct bpf_insn *insn, bool extra_pass,
1020 u64 *func_addr, bool *func_addr_fixed);
1022 struct bpf_prog *bpf_jit_blind_constants(struct bpf_prog *fp);
1023 void bpf_jit_prog_release_other(struct bpf_prog *fp, struct bpf_prog *fp_other);
1025 static inline void bpf_jit_dump(unsigned int flen, unsigned int proglen,
1026 u32 pass, void *image)
1028 pr_err("flen=%u proglen=%u pass=%u image=%pK from=%s pid=%d\n", flen,
1029 proglen, pass, image, current->comm, task_pid_nr(current));
1032 print_hex_dump(KERN_ERR, "JIT code: ", DUMP_PREFIX_OFFSET,
1033 16, 1, image, proglen, false);
1036 static inline bool bpf_jit_is_ebpf(void)
1038 # ifdef CONFIG_HAVE_EBPF_JIT
1045 static inline bool ebpf_jit_enabled(void)
1047 return bpf_jit_enable && bpf_jit_is_ebpf();
1050 static inline bool bpf_prog_ebpf_jited(const struct bpf_prog *fp)
1052 return fp->jited && bpf_jit_is_ebpf();
1055 static inline bool bpf_jit_blinding_enabled(struct bpf_prog *prog)
1057 /* These are the prerequisites, should someone ever have the
1058 * idea to call blinding outside of them, we make sure to
1061 if (!bpf_jit_is_ebpf())
1063 if (!prog->jit_requested)
1065 if (!bpf_jit_harden)
1067 if (bpf_jit_harden == 1 && capable(CAP_SYS_ADMIN))
1073 static inline bool bpf_jit_kallsyms_enabled(void)
1075 /* There are a couple of corner cases where kallsyms should
1076 * not be enabled f.e. on hardening.
1080 if (!bpf_jit_kallsyms)
1082 if (bpf_jit_kallsyms == 1)
1088 const char *__bpf_address_lookup(unsigned long addr, unsigned long *size,
1089 unsigned long *off, char *sym);
1090 bool is_bpf_text_address(unsigned long addr);
1091 int bpf_get_kallsym(unsigned int symnum, unsigned long *value, char *type,
1094 static inline const char *
1095 bpf_address_lookup(unsigned long addr, unsigned long *size,
1096 unsigned long *off, char **modname, char *sym)
1098 const char *ret = __bpf_address_lookup(addr, size, off, sym);
1105 void bpf_prog_kallsyms_add(struct bpf_prog *fp);
1106 void bpf_prog_kallsyms_del(struct bpf_prog *fp);
1108 #else /* CONFIG_BPF_JIT */
1110 static inline bool ebpf_jit_enabled(void)
1115 static inline bool bpf_jit_blinding_enabled(struct bpf_prog *prog)
1120 static inline bool bpf_prog_ebpf_jited(const struct bpf_prog *fp)
1126 bpf_jit_add_poke_descriptor(struct bpf_prog *prog,
1127 struct bpf_jit_poke_descriptor *poke)
1132 static inline void bpf_jit_free(struct bpf_prog *fp)
1134 bpf_prog_unlock_free(fp);
1137 static inline bool bpf_jit_kallsyms_enabled(void)
1142 static inline const char *
1143 __bpf_address_lookup(unsigned long addr, unsigned long *size,
1144 unsigned long *off, char *sym)
1149 static inline bool is_bpf_text_address(unsigned long addr)
1154 static inline int bpf_get_kallsym(unsigned int symnum, unsigned long *value,
1155 char *type, char *sym)
1160 static inline const char *
1161 bpf_address_lookup(unsigned long addr, unsigned long *size,
1162 unsigned long *off, char **modname, char *sym)
1167 static inline void bpf_prog_kallsyms_add(struct bpf_prog *fp)
1171 static inline void bpf_prog_kallsyms_del(struct bpf_prog *fp)
1175 #endif /* CONFIG_BPF_JIT */
1177 void bpf_prog_kallsyms_del_all(struct bpf_prog *fp);
1179 #define BPF_ANC BIT(15)
1181 static inline bool bpf_needs_clear_a(const struct sock_filter *first)
1183 switch (first->code) {
1184 case BPF_RET | BPF_K:
1185 case BPF_LD | BPF_W | BPF_LEN:
1188 case BPF_LD | BPF_W | BPF_ABS:
1189 case BPF_LD | BPF_H | BPF_ABS:
1190 case BPF_LD | BPF_B | BPF_ABS:
1191 if (first->k == SKF_AD_OFF + SKF_AD_ALU_XOR_X)
1200 static inline u16 bpf_anc_helper(const struct sock_filter *ftest)
1202 BUG_ON(ftest->code & BPF_ANC);
1204 switch (ftest->code) {
1205 case BPF_LD | BPF_W | BPF_ABS:
1206 case BPF_LD | BPF_H | BPF_ABS:
1207 case BPF_LD | BPF_B | BPF_ABS:
1208 #define BPF_ANCILLARY(CODE) case SKF_AD_OFF + SKF_AD_##CODE: \
1209 return BPF_ANC | SKF_AD_##CODE
1211 BPF_ANCILLARY(PROTOCOL);
1212 BPF_ANCILLARY(PKTTYPE);
1213 BPF_ANCILLARY(IFINDEX);
1214 BPF_ANCILLARY(NLATTR);
1215 BPF_ANCILLARY(NLATTR_NEST);
1216 BPF_ANCILLARY(MARK);
1217 BPF_ANCILLARY(QUEUE);
1218 BPF_ANCILLARY(HATYPE);
1219 BPF_ANCILLARY(RXHASH);
1221 BPF_ANCILLARY(ALU_XOR_X);
1222 BPF_ANCILLARY(VLAN_TAG);
1223 BPF_ANCILLARY(VLAN_TAG_PRESENT);
1224 BPF_ANCILLARY(PAY_OFFSET);
1225 BPF_ANCILLARY(RANDOM);
1226 BPF_ANCILLARY(VLAN_TPID);
1234 void *bpf_internal_load_pointer_neg_helper(const struct sk_buff *skb,
1235 int k, unsigned int size);
1237 static inline void *bpf_load_pointer(const struct sk_buff *skb, int k,
1238 unsigned int size, void *buffer)
1241 return skb_header_pointer(skb, k, size, buffer);
1243 return bpf_internal_load_pointer_neg_helper(skb, k, size);
1246 static inline int bpf_tell_extensions(void)
1251 struct bpf_sock_addr_kern {
1253 struct sockaddr *uaddr;
1254 /* Temporary "register" to make indirect stores to nested structures
1255 * defined above. We need three registers to make such a store, but
1256 * only two (src and dst) are available at convert_ctx_access time
1259 void *t_ctx; /* Attach type specific context. */
1262 struct bpf_sock_ops_kern {
1269 struct sk_buff *syn_skb;
1270 struct sk_buff *skb;
1274 u8 remaining_opt_len;
1275 u64 temp; /* temp and everything after is not
1276 * initialized to 0 before calling
1277 * the BPF program. New fields that
1278 * should be initialized to 0 should
1279 * be inserted before temp.
1280 * temp is scratch storage used by
1281 * sock_ops_convert_ctx_access
1282 * as temporary storage of a register.
1286 struct bpf_sysctl_kern {
1287 struct ctl_table_header *head;
1288 struct ctl_table *table;
1296 /* Temporary "register" for indirect stores to ppos. */
1300 struct bpf_sockopt_kern {
1310 int copy_bpf_fprog_from_user(struct sock_fprog *dst, sockptr_t src, int len);
1312 struct bpf_sk_lookup_kern {
1322 const struct in6_addr *saddr;
1323 const struct in6_addr *daddr;
1325 struct sock *selected_sk;
1329 extern struct static_key_false bpf_sk_lookup_enabled;
1331 /* Runners for BPF_SK_LOOKUP programs to invoke on socket lookup.
1333 * Allowed return values for a BPF SK_LOOKUP program are SK_PASS and
1334 * SK_DROP. Their meaning is as follows:
1336 * SK_PASS && ctx.selected_sk != NULL: use selected_sk as lookup result
1337 * SK_PASS && ctx.selected_sk == NULL: continue to htable-based socket lookup
1338 * SK_DROP : terminate lookup with -ECONNREFUSED
1340 * This macro aggregates return values and selected sockets from
1341 * multiple BPF programs according to following rules in order:
1343 * 1. If any program returned SK_PASS and a non-NULL ctx.selected_sk,
1344 * macro result is SK_PASS and last ctx.selected_sk is used.
1345 * 2. If any program returned SK_DROP return value,
1346 * macro result is SK_DROP.
1347 * 3. Otherwise result is SK_PASS and ctx.selected_sk is NULL.
1349 * Caller must ensure that the prog array is non-NULL, and that the
1350 * array as well as the programs it contains remain valid.
1352 #define BPF_PROG_SK_LOOKUP_RUN_ARRAY(array, ctx, func) \
1354 struct bpf_sk_lookup_kern *_ctx = &(ctx); \
1355 struct bpf_prog_array_item *_item; \
1356 struct sock *_selected_sk = NULL; \
1357 bool _no_reuseport = false; \
1358 struct bpf_prog *_prog; \
1359 bool _all_pass = true; \
1362 migrate_disable(); \
1363 _item = &(array)->items[0]; \
1364 while ((_prog = READ_ONCE(_item->prog))) { \
1365 /* restore most recent selection */ \
1366 _ctx->selected_sk = _selected_sk; \
1367 _ctx->no_reuseport = _no_reuseport; \
1369 _ret = func(_prog, _ctx); \
1370 if (_ret == SK_PASS && _ctx->selected_sk) { \
1371 /* remember last non-NULL socket */ \
1372 _selected_sk = _ctx->selected_sk; \
1373 _no_reuseport = _ctx->no_reuseport; \
1374 } else if (_ret == SK_DROP && _all_pass) { \
1375 _all_pass = false; \
1379 _ctx->selected_sk = _selected_sk; \
1380 _ctx->no_reuseport = _no_reuseport; \
1382 _all_pass || _selected_sk ? SK_PASS : SK_DROP; \
1385 static inline bool bpf_sk_lookup_run_v4(struct net *net, int protocol,
1386 const __be32 saddr, const __be16 sport,
1387 const __be32 daddr, const u16 dport,
1390 struct bpf_prog_array *run_array;
1391 struct sock *selected_sk = NULL;
1392 bool no_reuseport = false;
1395 run_array = rcu_dereference(net->bpf.run_array[NETNS_BPF_SK_LOOKUP]);
1397 struct bpf_sk_lookup_kern ctx = {
1399 .protocol = protocol,
1407 act = BPF_PROG_SK_LOOKUP_RUN_ARRAY(run_array, ctx, BPF_PROG_RUN);
1408 if (act == SK_PASS) {
1409 selected_sk = ctx.selected_sk;
1410 no_reuseport = ctx.no_reuseport;
1412 selected_sk = ERR_PTR(-ECONNREFUSED);
1417 return no_reuseport;
1420 #if IS_ENABLED(CONFIG_IPV6)
1421 static inline bool bpf_sk_lookup_run_v6(struct net *net, int protocol,
1422 const struct in6_addr *saddr,
1424 const struct in6_addr *daddr,
1428 struct bpf_prog_array *run_array;
1429 struct sock *selected_sk = NULL;
1430 bool no_reuseport = false;
1433 run_array = rcu_dereference(net->bpf.run_array[NETNS_BPF_SK_LOOKUP]);
1435 struct bpf_sk_lookup_kern ctx = {
1437 .protocol = protocol,
1445 act = BPF_PROG_SK_LOOKUP_RUN_ARRAY(run_array, ctx, BPF_PROG_RUN);
1446 if (act == SK_PASS) {
1447 selected_sk = ctx.selected_sk;
1448 no_reuseport = ctx.no_reuseport;
1450 selected_sk = ERR_PTR(-ECONNREFUSED);
1455 return no_reuseport;
1457 #endif /* IS_ENABLED(CONFIG_IPV6) */
1459 #endif /* __LINUX_FILTER_H__ */