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>
23 #include <linux/vmalloc.h>
25 #include <net/sch_generic.h>
27 #include <uapi/linux/filter.h>
28 #include <uapi/linux/bpf.h>
36 struct sock_reuseport;
38 struct ctl_table_header;
40 /* ArgX, context and stack frame pointer register positions. Note,
41 * Arg1, Arg2, Arg3, etc are used as argument mappings of function
42 * calls in BPF_CALL instruction.
44 #define BPF_REG_ARG1 BPF_REG_1
45 #define BPF_REG_ARG2 BPF_REG_2
46 #define BPF_REG_ARG3 BPF_REG_3
47 #define BPF_REG_ARG4 BPF_REG_4
48 #define BPF_REG_ARG5 BPF_REG_5
49 #define BPF_REG_CTX BPF_REG_6
50 #define BPF_REG_FP BPF_REG_10
52 /* Additional register mappings for converted user programs. */
53 #define BPF_REG_A BPF_REG_0
54 #define BPF_REG_X BPF_REG_7
55 #define BPF_REG_TMP BPF_REG_2 /* scratch reg */
56 #define BPF_REG_D BPF_REG_8 /* data, callee-saved */
57 #define BPF_REG_H BPF_REG_9 /* hlen, callee-saved */
59 /* Kernel hidden auxiliary/helper register. */
60 #define BPF_REG_AX MAX_BPF_REG
61 #define MAX_BPF_EXT_REG (MAX_BPF_REG + 1)
62 #define MAX_BPF_JIT_REG MAX_BPF_EXT_REG
64 /* unused opcode to mark special call to bpf_tail_call() helper */
65 #define BPF_TAIL_CALL 0xf0
67 /* unused opcode to mark call to interpreter with arguments */
68 #define BPF_CALL_ARGS 0xe0
70 /* As per nm, we expose JITed images as text (code) section for
71 * kallsyms. That way, tools like perf can find it to match
74 #define BPF_SYM_ELF_TYPE 't'
76 /* BPF program can access up to 512 bytes of stack space. */
77 #define MAX_BPF_STACK 512
79 /* Helper macros for filter block array initializers. */
81 /* ALU ops on registers, bpf_add|sub|...: dst_reg += src_reg */
83 #define BPF_ALU64_REG(OP, DST, SRC) \
84 ((struct bpf_insn) { \
85 .code = BPF_ALU64 | BPF_OP(OP) | BPF_X, \
91 #define BPF_ALU32_REG(OP, DST, SRC) \
92 ((struct bpf_insn) { \
93 .code = BPF_ALU | BPF_OP(OP) | BPF_X, \
99 /* ALU ops on immediates, bpf_add|sub|...: dst_reg += imm32 */
101 #define BPF_ALU64_IMM(OP, DST, IMM) \
102 ((struct bpf_insn) { \
103 .code = BPF_ALU64 | BPF_OP(OP) | BPF_K, \
109 #define BPF_ALU32_IMM(OP, DST, IMM) \
110 ((struct bpf_insn) { \
111 .code = BPF_ALU | BPF_OP(OP) | BPF_K, \
117 /* Endianess conversion, cpu_to_{l,b}e(), {l,b}e_to_cpu() */
119 #define BPF_ENDIAN(TYPE, DST, LEN) \
120 ((struct bpf_insn) { \
121 .code = BPF_ALU | BPF_END | BPF_SRC(TYPE), \
127 /* Short form of mov, dst_reg = src_reg */
129 #define BPF_MOV64_REG(DST, SRC) \
130 ((struct bpf_insn) { \
131 .code = BPF_ALU64 | BPF_MOV | BPF_X, \
137 #define BPF_MOV32_REG(DST, SRC) \
138 ((struct bpf_insn) { \
139 .code = BPF_ALU | BPF_MOV | BPF_X, \
145 /* Short form of mov, dst_reg = imm32 */
147 #define BPF_MOV64_IMM(DST, IMM) \
148 ((struct bpf_insn) { \
149 .code = BPF_ALU64 | BPF_MOV | BPF_K, \
155 #define BPF_MOV32_IMM(DST, IMM) \
156 ((struct bpf_insn) { \
157 .code = BPF_ALU | BPF_MOV | BPF_K, \
163 /* Special form of mov32, used for doing explicit zero extension on dst. */
164 #define BPF_ZEXT_REG(DST) \
165 ((struct bpf_insn) { \
166 .code = BPF_ALU | BPF_MOV | BPF_X, \
172 static inline bool insn_is_zext(const struct bpf_insn *insn)
174 return insn->code == (BPF_ALU | BPF_MOV | BPF_X) && insn->imm == 1;
177 /* BPF_LD_IMM64 macro encodes single 'load 64-bit immediate' insn */
178 #define BPF_LD_IMM64(DST, IMM) \
179 BPF_LD_IMM64_RAW(DST, 0, IMM)
181 #define BPF_LD_IMM64_RAW(DST, SRC, IMM) \
182 ((struct bpf_insn) { \
183 .code = BPF_LD | BPF_DW | BPF_IMM, \
187 .imm = (__u32) (IMM) }), \
188 ((struct bpf_insn) { \
189 .code = 0, /* zero is reserved opcode */ \
193 .imm = ((__u64) (IMM)) >> 32 })
195 /* pseudo BPF_LD_IMM64 insn used to refer to process-local map_fd */
196 #define BPF_LD_MAP_FD(DST, MAP_FD) \
197 BPF_LD_IMM64_RAW(DST, BPF_PSEUDO_MAP_FD, MAP_FD)
199 /* Short form of mov based on type, BPF_X: dst_reg = src_reg, BPF_K: dst_reg = imm32 */
201 #define BPF_MOV64_RAW(TYPE, DST, SRC, IMM) \
202 ((struct bpf_insn) { \
203 .code = BPF_ALU64 | BPF_MOV | BPF_SRC(TYPE), \
209 #define BPF_MOV32_RAW(TYPE, DST, SRC, IMM) \
210 ((struct bpf_insn) { \
211 .code = BPF_ALU | BPF_MOV | BPF_SRC(TYPE), \
217 /* Direct packet access, R0 = *(uint *) (skb->data + imm32) */
219 #define BPF_LD_ABS(SIZE, IMM) \
220 ((struct bpf_insn) { \
221 .code = BPF_LD | BPF_SIZE(SIZE) | BPF_ABS, \
227 /* Indirect packet access, R0 = *(uint *) (skb->data + src_reg + imm32) */
229 #define BPF_LD_IND(SIZE, SRC, IMM) \
230 ((struct bpf_insn) { \
231 .code = BPF_LD | BPF_SIZE(SIZE) | BPF_IND, \
237 /* Memory load, dst_reg = *(uint *) (src_reg + off16) */
239 #define BPF_LDX_MEM(SIZE, DST, SRC, OFF) \
240 ((struct bpf_insn) { \
241 .code = BPF_LDX | BPF_SIZE(SIZE) | BPF_MEM, \
247 /* Memory store, *(uint *) (dst_reg + off16) = src_reg */
249 #define BPF_STX_MEM(SIZE, DST, SRC, OFF) \
250 ((struct bpf_insn) { \
251 .code = BPF_STX | BPF_SIZE(SIZE) | BPF_MEM, \
257 /* Atomic memory add, *(uint *)(dst_reg + off16) += src_reg */
259 #define BPF_STX_XADD(SIZE, DST, SRC, OFF) \
260 ((struct bpf_insn) { \
261 .code = BPF_STX | BPF_SIZE(SIZE) | BPF_XADD, \
267 /* Memory store, *(uint *) (dst_reg + off16) = imm32 */
269 #define BPF_ST_MEM(SIZE, DST, OFF, IMM) \
270 ((struct bpf_insn) { \
271 .code = BPF_ST | BPF_SIZE(SIZE) | BPF_MEM, \
277 /* Conditional jumps against registers, if (dst_reg 'op' src_reg) goto pc + off16 */
279 #define BPF_JMP_REG(OP, DST, SRC, OFF) \
280 ((struct bpf_insn) { \
281 .code = BPF_JMP | BPF_OP(OP) | BPF_X, \
287 /* Conditional jumps against immediates, if (dst_reg 'op' imm32) goto pc + off16 */
289 #define BPF_JMP_IMM(OP, DST, IMM, OFF) \
290 ((struct bpf_insn) { \
291 .code = BPF_JMP | BPF_OP(OP) | BPF_K, \
297 /* Like BPF_JMP_REG, but with 32-bit wide operands for comparison. */
299 #define BPF_JMP32_REG(OP, DST, SRC, OFF) \
300 ((struct bpf_insn) { \
301 .code = BPF_JMP32 | BPF_OP(OP) | BPF_X, \
307 /* Like BPF_JMP_IMM, but with 32-bit wide operands for comparison. */
309 #define BPF_JMP32_IMM(OP, DST, IMM, OFF) \
310 ((struct bpf_insn) { \
311 .code = BPF_JMP32 | BPF_OP(OP) | BPF_K, \
317 /* Unconditional jumps, goto pc + off16 */
319 #define BPF_JMP_A(OFF) \
320 ((struct bpf_insn) { \
321 .code = BPF_JMP | BPF_JA, \
329 #define BPF_CALL_REL(TGT) \
330 ((struct bpf_insn) { \
331 .code = BPF_JMP | BPF_CALL, \
333 .src_reg = BPF_PSEUDO_CALL, \
339 #define BPF_CAST_CALL(x) \
340 ((u64 (*)(u64, u64, u64, u64, u64))(x))
342 #define BPF_EMIT_CALL(FUNC) \
343 ((struct bpf_insn) { \
344 .code = BPF_JMP | BPF_CALL, \
348 .imm = ((FUNC) - __bpf_call_base) })
350 /* Raw code statement block */
352 #define BPF_RAW_INSN(CODE, DST, SRC, OFF, IMM) \
353 ((struct bpf_insn) { \
362 #define BPF_EXIT_INSN() \
363 ((struct bpf_insn) { \
364 .code = BPF_JMP | BPF_EXIT, \
370 /* Internal classic blocks for direct assignment */
372 #define __BPF_STMT(CODE, K) \
373 ((struct sock_filter) BPF_STMT(CODE, K))
375 #define __BPF_JUMP(CODE, K, JT, JF) \
376 ((struct sock_filter) BPF_JUMP(CODE, K, JT, JF))
378 #define bytes_to_bpf_size(bytes) \
380 int bpf_size = -EINVAL; \
382 if (bytes == sizeof(u8)) \
384 else if (bytes == sizeof(u16)) \
386 else if (bytes == sizeof(u32)) \
388 else if (bytes == sizeof(u64)) \
394 #define bpf_size_to_bytes(bpf_size) \
396 int bytes = -EINVAL; \
398 if (bpf_size == BPF_B) \
399 bytes = sizeof(u8); \
400 else if (bpf_size == BPF_H) \
401 bytes = sizeof(u16); \
402 else if (bpf_size == BPF_W) \
403 bytes = sizeof(u32); \
404 else if (bpf_size == BPF_DW) \
405 bytes = sizeof(u64); \
410 #define BPF_SIZEOF(type) \
412 const int __size = bytes_to_bpf_size(sizeof(type)); \
413 BUILD_BUG_ON(__size < 0); \
417 #define BPF_FIELD_SIZEOF(type, field) \
419 const int __size = bytes_to_bpf_size(FIELD_SIZEOF(type, field)); \
420 BUILD_BUG_ON(__size < 0); \
424 #define BPF_LDST_BYTES(insn) \
426 const int __size = bpf_size_to_bytes(BPF_SIZE((insn)->code)); \
427 WARN_ON(__size < 0); \
431 #define __BPF_MAP_0(m, v, ...) v
432 #define __BPF_MAP_1(m, v, t, a, ...) m(t, a)
433 #define __BPF_MAP_2(m, v, t, a, ...) m(t, a), __BPF_MAP_1(m, v, __VA_ARGS__)
434 #define __BPF_MAP_3(m, v, t, a, ...) m(t, a), __BPF_MAP_2(m, v, __VA_ARGS__)
435 #define __BPF_MAP_4(m, v, t, a, ...) m(t, a), __BPF_MAP_3(m, v, __VA_ARGS__)
436 #define __BPF_MAP_5(m, v, t, a, ...) m(t, a), __BPF_MAP_4(m, v, __VA_ARGS__)
438 #define __BPF_REG_0(...) __BPF_PAD(5)
439 #define __BPF_REG_1(...) __BPF_MAP(1, __VA_ARGS__), __BPF_PAD(4)
440 #define __BPF_REG_2(...) __BPF_MAP(2, __VA_ARGS__), __BPF_PAD(3)
441 #define __BPF_REG_3(...) __BPF_MAP(3, __VA_ARGS__), __BPF_PAD(2)
442 #define __BPF_REG_4(...) __BPF_MAP(4, __VA_ARGS__), __BPF_PAD(1)
443 #define __BPF_REG_5(...) __BPF_MAP(5, __VA_ARGS__)
445 #define __BPF_MAP(n, ...) __BPF_MAP_##n(__VA_ARGS__)
446 #define __BPF_REG(n, ...) __BPF_REG_##n(__VA_ARGS__)
448 #define __BPF_CAST(t, a) \
451 typeof(__builtin_choose_expr(sizeof(t) == sizeof(unsigned long), \
452 (unsigned long)0, (t)0))) a
456 #define __BPF_DECL_ARGS(t, a) t a
457 #define __BPF_DECL_REGS(t, a) u64 a
459 #define __BPF_PAD(n) \
460 __BPF_MAP(n, __BPF_DECL_ARGS, __BPF_N, u64, __ur_1, u64, __ur_2, \
461 u64, __ur_3, u64, __ur_4, u64, __ur_5)
463 #define BPF_CALL_x(x, name, ...) \
464 static __always_inline \
465 u64 ____##name(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__)); \
466 u64 name(__BPF_REG(x, __BPF_DECL_REGS, __BPF_N, __VA_ARGS__)); \
467 u64 name(__BPF_REG(x, __BPF_DECL_REGS, __BPF_N, __VA_ARGS__)) \
469 return ____##name(__BPF_MAP(x,__BPF_CAST,__BPF_N,__VA_ARGS__));\
471 static __always_inline \
472 u64 ____##name(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__))
474 #define BPF_CALL_0(name, ...) BPF_CALL_x(0, name, __VA_ARGS__)
475 #define BPF_CALL_1(name, ...) BPF_CALL_x(1, name, __VA_ARGS__)
476 #define BPF_CALL_2(name, ...) BPF_CALL_x(2, name, __VA_ARGS__)
477 #define BPF_CALL_3(name, ...) BPF_CALL_x(3, name, __VA_ARGS__)
478 #define BPF_CALL_4(name, ...) BPF_CALL_x(4, name, __VA_ARGS__)
479 #define BPF_CALL_5(name, ...) BPF_CALL_x(5, name, __VA_ARGS__)
481 #define bpf_ctx_range(TYPE, MEMBER) \
482 offsetof(TYPE, MEMBER) ... offsetofend(TYPE, MEMBER) - 1
483 #define bpf_ctx_range_till(TYPE, MEMBER1, MEMBER2) \
484 offsetof(TYPE, MEMBER1) ... offsetofend(TYPE, MEMBER2) - 1
485 #if BITS_PER_LONG == 64
486 # define bpf_ctx_range_ptr(TYPE, MEMBER) \
487 offsetof(TYPE, MEMBER) ... offsetofend(TYPE, MEMBER) - 1
489 # define bpf_ctx_range_ptr(TYPE, MEMBER) \
490 offsetof(TYPE, MEMBER) ... offsetof(TYPE, MEMBER) + 8 - 1
491 #endif /* BITS_PER_LONG == 64 */
493 #define bpf_target_off(TYPE, MEMBER, SIZE, PTR_SIZE) \
495 BUILD_BUG_ON(FIELD_SIZEOF(TYPE, MEMBER) != (SIZE)); \
496 *(PTR_SIZE) = (SIZE); \
497 offsetof(TYPE, MEMBER); \
501 /* A struct sock_filter is architecture independent. */
502 struct compat_sock_fprog {
504 compat_uptr_t filter; /* struct sock_filter * */
508 struct sock_fprog_kern {
510 struct sock_filter *filter;
513 struct bpf_binary_header {
515 /* Some arches need word alignment for their instructions */
516 u8 image[] __aligned(4);
520 u16 pages; /* Number of allocated pages */
521 u16 jited:1, /* Is our filter JIT'ed? */
522 jit_requested:1,/* archs need to JIT the prog */
523 gpl_compatible:1, /* Is filter GPL compatible? */
524 cb_access:1, /* Is control block accessed? */
525 dst_needed:1, /* Do we need dst entry? */
526 blinded:1, /* Was blinded */
527 is_func:1, /* program is a bpf function */
528 kprobe_override:1, /* Do we override a kprobe? */
529 has_callchain_buf:1; /* callchain buffer allocated? */
530 enum bpf_prog_type type; /* Type of BPF program */
531 enum bpf_attach_type expected_attach_type; /* For some prog types */
532 u32 len; /* Number of filter blocks */
533 u32 jited_len; /* Size of jited insns in bytes */
534 u8 tag[BPF_TAG_SIZE];
535 struct bpf_prog_aux *aux; /* Auxiliary fields */
536 struct sock_fprog_kern *orig_prog; /* Original BPF program */
537 unsigned int (*bpf_func)(const void *ctx,
538 const struct bpf_insn *insn);
539 /* Instructions for interpreter */
541 struct sock_filter insns[0];
542 struct bpf_insn insnsi[0];
549 struct bpf_prog *prog;
552 DECLARE_STATIC_KEY_FALSE(bpf_stats_enabled_key);
554 #define BPF_PROG_RUN(prog, ctx) ({ \
557 if (static_branch_unlikely(&bpf_stats_enabled_key)) { \
558 struct bpf_prog_stats *stats; \
559 u64 start = sched_clock(); \
560 ret = (*(prog)->bpf_func)(ctx, (prog)->insnsi); \
561 stats = this_cpu_ptr(prog->aux->stats); \
562 u64_stats_update_begin(&stats->syncp); \
564 stats->nsecs += sched_clock() - start; \
565 u64_stats_update_end(&stats->syncp); \
567 ret = (*(prog)->bpf_func)(ctx, (prog)->insnsi); \
571 #define BPF_SKB_CB_LEN QDISC_CB_PRIV_LEN
573 struct bpf_skb_data_end {
574 struct qdisc_skb_cb qdisc_cb;
579 struct bpf_redirect_info {
583 struct bpf_map *map_to_flush;
587 DECLARE_PER_CPU(struct bpf_redirect_info, bpf_redirect_info);
589 /* flags for bpf_redirect_info kern_flags */
590 #define BPF_RI_F_RF_NO_DIRECT BIT(0) /* no napi_direct on return_frame */
592 /* Compute the linear packet data range [data, data_end) which
593 * will be accessed by various program types (cls_bpf, act_bpf,
594 * lwt, ...). Subsystems allowing direct data access must (!)
595 * ensure that cb[] area can be written to when BPF program is
596 * invoked (otherwise cb[] save/restore is necessary).
598 static inline void bpf_compute_data_pointers(struct sk_buff *skb)
600 struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb;
602 BUILD_BUG_ON(sizeof(*cb) > FIELD_SIZEOF(struct sk_buff, cb));
603 cb->data_meta = skb->data - skb_metadata_len(skb);
604 cb->data_end = skb->data + skb_headlen(skb);
607 /* Similar to bpf_compute_data_pointers(), except that save orginal
608 * data in cb->data and cb->meta_data for restore.
610 static inline void bpf_compute_and_save_data_end(
611 struct sk_buff *skb, void **saved_data_end)
613 struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb;
615 *saved_data_end = cb->data_end;
616 cb->data_end = skb->data + skb_headlen(skb);
619 /* Restore data saved by bpf_compute_data_pointers(). */
620 static inline void bpf_restore_data_end(
621 struct sk_buff *skb, void *saved_data_end)
623 struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb;
625 cb->data_end = saved_data_end;
628 static inline u8 *bpf_skb_cb(struct sk_buff *skb)
630 /* eBPF programs may read/write skb->cb[] area to transfer meta
631 * data between tail calls. Since this also needs to work with
632 * tc, that scratch memory is mapped to qdisc_skb_cb's data area.
634 * In some socket filter cases, the cb unfortunately needs to be
635 * saved/restored so that protocol specific skb->cb[] data won't
636 * be lost. In any case, due to unpriviledged eBPF programs
637 * attached to sockets, we need to clear the bpf_skb_cb() area
638 * to not leak previous contents to user space.
640 BUILD_BUG_ON(FIELD_SIZEOF(struct __sk_buff, cb) != BPF_SKB_CB_LEN);
641 BUILD_BUG_ON(FIELD_SIZEOF(struct __sk_buff, cb) !=
642 FIELD_SIZEOF(struct qdisc_skb_cb, data));
644 return qdisc_skb_cb(skb)->data;
647 static inline u32 __bpf_prog_run_save_cb(const struct bpf_prog *prog,
650 u8 *cb_data = bpf_skb_cb(skb);
651 u8 cb_saved[BPF_SKB_CB_LEN];
654 if (unlikely(prog->cb_access)) {
655 memcpy(cb_saved, cb_data, sizeof(cb_saved));
656 memset(cb_data, 0, sizeof(cb_saved));
659 res = BPF_PROG_RUN(prog, skb);
661 if (unlikely(prog->cb_access))
662 memcpy(cb_data, cb_saved, sizeof(cb_saved));
667 static inline u32 bpf_prog_run_save_cb(const struct bpf_prog *prog,
673 res = __bpf_prog_run_save_cb(prog, skb);
678 static inline u32 bpf_prog_run_clear_cb(const struct bpf_prog *prog,
681 u8 *cb_data = bpf_skb_cb(skb);
684 if (unlikely(prog->cb_access))
685 memset(cb_data, 0, BPF_SKB_CB_LEN);
688 res = BPF_PROG_RUN(prog, skb);
693 static __always_inline u32 bpf_prog_run_xdp(const struct bpf_prog *prog,
694 struct xdp_buff *xdp)
696 /* Caller needs to hold rcu_read_lock() (!), otherwise program
697 * can be released while still running, or map elements could be
698 * freed early while still having concurrent users. XDP fastpath
699 * already takes rcu_read_lock() when fetching the program, so
700 * it's not necessary here anymore.
702 return BPF_PROG_RUN(prog, xdp);
705 static inline u32 bpf_prog_insn_size(const struct bpf_prog *prog)
707 return prog->len * sizeof(struct bpf_insn);
710 static inline u32 bpf_prog_tag_scratch_size(const struct bpf_prog *prog)
712 return round_up(bpf_prog_insn_size(prog) +
713 sizeof(__be64) + 1, SHA_MESSAGE_BYTES);
716 static inline unsigned int bpf_prog_size(unsigned int proglen)
718 return max(sizeof(struct bpf_prog),
719 offsetof(struct bpf_prog, insns[proglen]));
722 static inline bool bpf_prog_was_classic(const struct bpf_prog *prog)
724 /* When classic BPF programs have been loaded and the arch
725 * does not have a classic BPF JIT (anymore), they have been
726 * converted via bpf_migrate_filter() to eBPF and thus always
727 * have an unspec program type.
729 return prog->type == BPF_PROG_TYPE_UNSPEC;
732 static inline u32 bpf_ctx_off_adjust_machine(u32 size)
734 const u32 size_machine = sizeof(unsigned long);
736 if (size > size_machine && size % size_machine == 0)
743 bpf_ctx_narrow_access_ok(u32 off, u32 size, u32 size_default)
745 return size <= size_default && (size & (size - 1)) == 0;
748 #define bpf_classic_proglen(fprog) (fprog->len * sizeof(fprog->filter[0]))
750 static inline void bpf_prog_lock_ro(struct bpf_prog *fp)
752 set_vm_flush_reset_perms(fp);
753 set_memory_ro((unsigned long)fp, fp->pages);
756 static inline void bpf_jit_binary_lock_ro(struct bpf_binary_header *hdr)
758 set_vm_flush_reset_perms(hdr);
759 set_memory_ro((unsigned long)hdr, hdr->pages);
760 set_memory_x((unsigned long)hdr, hdr->pages);
763 static inline struct bpf_binary_header *
764 bpf_jit_binary_hdr(const struct bpf_prog *fp)
766 unsigned long real_start = (unsigned long)fp->bpf_func;
767 unsigned long addr = real_start & PAGE_MASK;
772 int sk_filter_trim_cap(struct sock *sk, struct sk_buff *skb, unsigned int cap);
773 static inline int sk_filter(struct sock *sk, struct sk_buff *skb)
775 return sk_filter_trim_cap(sk, skb, 1);
778 struct bpf_prog *bpf_prog_select_runtime(struct bpf_prog *fp, int *err);
779 void bpf_prog_free(struct bpf_prog *fp);
781 bool bpf_opcode_in_insntable(u8 code);
783 void bpf_prog_free_linfo(struct bpf_prog *prog);
784 void bpf_prog_fill_jited_linfo(struct bpf_prog *prog,
785 const u32 *insn_to_jit_off);
786 int bpf_prog_alloc_jited_linfo(struct bpf_prog *prog);
787 void bpf_prog_free_jited_linfo(struct bpf_prog *prog);
788 void bpf_prog_free_unused_jited_linfo(struct bpf_prog *prog);
790 struct bpf_prog *bpf_prog_alloc(unsigned int size, gfp_t gfp_extra_flags);
791 struct bpf_prog *bpf_prog_alloc_no_stats(unsigned int size, gfp_t gfp_extra_flags);
792 struct bpf_prog *bpf_prog_realloc(struct bpf_prog *fp_old, unsigned int size,
793 gfp_t gfp_extra_flags);
794 void __bpf_prog_free(struct bpf_prog *fp);
796 static inline void bpf_prog_unlock_free(struct bpf_prog *fp)
801 typedef int (*bpf_aux_classic_check_t)(struct sock_filter *filter,
804 int bpf_prog_create(struct bpf_prog **pfp, struct sock_fprog_kern *fprog);
805 int bpf_prog_create_from_user(struct bpf_prog **pfp, struct sock_fprog *fprog,
806 bpf_aux_classic_check_t trans, bool save_orig);
807 void bpf_prog_destroy(struct bpf_prog *fp);
809 int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk);
810 int sk_attach_bpf(u32 ufd, struct sock *sk);
811 int sk_reuseport_attach_filter(struct sock_fprog *fprog, struct sock *sk);
812 int sk_reuseport_attach_bpf(u32 ufd, struct sock *sk);
813 void sk_reuseport_prog_free(struct bpf_prog *prog);
814 int sk_detach_filter(struct sock *sk);
815 int sk_get_filter(struct sock *sk, struct sock_filter __user *filter,
818 bool sk_filter_charge(struct sock *sk, struct sk_filter *fp);
819 void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp);
821 u64 __bpf_call_base(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5);
822 #define __bpf_call_base_args \
823 ((u64 (*)(u64, u64, u64, u64, u64, const struct bpf_insn *)) \
826 struct bpf_prog *bpf_int_jit_compile(struct bpf_prog *prog);
827 void bpf_jit_compile(struct bpf_prog *prog);
828 bool bpf_helper_changes_pkt_data(void *func);
830 static inline bool bpf_dump_raw_ok(void)
832 /* Reconstruction of call-sites is dependent on kallsyms,
833 * thus make dump the same restriction.
835 return kallsyms_show_value() == 1;
838 struct bpf_prog *bpf_patch_insn_single(struct bpf_prog *prog, u32 off,
839 const struct bpf_insn *patch, u32 len);
840 int bpf_remove_insns(struct bpf_prog *prog, u32 off, u32 cnt);
842 void bpf_clear_redirect_map(struct bpf_map *map);
844 static inline bool xdp_return_frame_no_direct(void)
846 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
848 return ri->kern_flags & BPF_RI_F_RF_NO_DIRECT;
851 static inline void xdp_set_return_frame_no_direct(void)
853 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
855 ri->kern_flags |= BPF_RI_F_RF_NO_DIRECT;
858 static inline void xdp_clear_return_frame_no_direct(void)
860 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
862 ri->kern_flags &= ~BPF_RI_F_RF_NO_DIRECT;
865 static inline int xdp_ok_fwd_dev(const struct net_device *fwd,
870 if (unlikely(!(fwd->flags & IFF_UP)))
873 len = fwd->mtu + fwd->hard_header_len + VLAN_HLEN;
880 /* The pair of xdp_do_redirect and xdp_do_flush_map MUST be called in the
881 * same cpu context. Further for best results no more than a single map
882 * for the do_redirect/do_flush pair should be used. This limitation is
883 * because we only track one map and force a flush when the map changes.
884 * This does not appear to be a real limitation for existing software.
886 int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb,
887 struct xdp_buff *xdp, struct bpf_prog *prog);
888 int xdp_do_redirect(struct net_device *dev,
889 struct xdp_buff *xdp,
890 struct bpf_prog *prog);
891 void xdp_do_flush_map(void);
893 void bpf_warn_invalid_xdp_action(u32 act);
896 struct sock *bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk,
897 struct bpf_prog *prog, struct sk_buff *skb,
900 static inline struct sock *
901 bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk,
902 struct bpf_prog *prog, struct sk_buff *skb,
909 #ifdef CONFIG_BPF_JIT
910 extern int bpf_jit_enable;
911 extern int bpf_jit_harden;
912 extern int bpf_jit_kallsyms;
913 extern long bpf_jit_limit;
915 typedef void (*bpf_jit_fill_hole_t)(void *area, unsigned int size);
917 struct bpf_binary_header *
918 bpf_jit_binary_alloc(unsigned int proglen, u8 **image_ptr,
919 unsigned int alignment,
920 bpf_jit_fill_hole_t bpf_fill_ill_insns);
921 void bpf_jit_binary_free(struct bpf_binary_header *hdr);
922 u64 bpf_jit_alloc_exec_limit(void);
923 void *bpf_jit_alloc_exec(unsigned long size);
924 void bpf_jit_free_exec(void *addr);
925 void bpf_jit_free(struct bpf_prog *fp);
927 int bpf_jit_get_func_addr(const struct bpf_prog *prog,
928 const struct bpf_insn *insn, bool extra_pass,
929 u64 *func_addr, bool *func_addr_fixed);
931 struct bpf_prog *bpf_jit_blind_constants(struct bpf_prog *fp);
932 void bpf_jit_prog_release_other(struct bpf_prog *fp, struct bpf_prog *fp_other);
934 static inline void bpf_jit_dump(unsigned int flen, unsigned int proglen,
935 u32 pass, void *image)
937 pr_err("flen=%u proglen=%u pass=%u image=%pK from=%s pid=%d\n", flen,
938 proglen, pass, image, current->comm, task_pid_nr(current));
941 print_hex_dump(KERN_ERR, "JIT code: ", DUMP_PREFIX_OFFSET,
942 16, 1, image, proglen, false);
945 static inline bool bpf_jit_is_ebpf(void)
947 # ifdef CONFIG_HAVE_EBPF_JIT
954 static inline bool ebpf_jit_enabled(void)
956 return bpf_jit_enable && bpf_jit_is_ebpf();
959 static inline bool bpf_prog_ebpf_jited(const struct bpf_prog *fp)
961 return fp->jited && bpf_jit_is_ebpf();
964 static inline bool bpf_jit_blinding_enabled(struct bpf_prog *prog)
966 /* These are the prerequisites, should someone ever have the
967 * idea to call blinding outside of them, we make sure to
970 if (!bpf_jit_is_ebpf())
972 if (!prog->jit_requested)
976 if (bpf_jit_harden == 1 && capable(CAP_SYS_ADMIN))
982 static inline bool bpf_jit_kallsyms_enabled(void)
984 /* There are a couple of corner cases where kallsyms should
985 * not be enabled f.e. on hardening.
989 if (!bpf_jit_kallsyms)
991 if (bpf_jit_kallsyms == 1)
997 const char *__bpf_address_lookup(unsigned long addr, unsigned long *size,
998 unsigned long *off, char *sym);
999 bool is_bpf_text_address(unsigned long addr);
1000 int bpf_get_kallsym(unsigned int symnum, unsigned long *value, char *type,
1003 static inline const char *
1004 bpf_address_lookup(unsigned long addr, unsigned long *size,
1005 unsigned long *off, char **modname, char *sym)
1007 const char *ret = __bpf_address_lookup(addr, size, off, sym);
1014 void bpf_prog_kallsyms_add(struct bpf_prog *fp);
1015 void bpf_prog_kallsyms_del(struct bpf_prog *fp);
1016 void bpf_get_prog_name(const struct bpf_prog *prog, char *sym);
1018 #else /* CONFIG_BPF_JIT */
1020 static inline bool ebpf_jit_enabled(void)
1025 static inline bool bpf_prog_ebpf_jited(const struct bpf_prog *fp)
1030 static inline void bpf_jit_free(struct bpf_prog *fp)
1032 bpf_prog_unlock_free(fp);
1035 static inline bool bpf_jit_kallsyms_enabled(void)
1040 static inline const char *
1041 __bpf_address_lookup(unsigned long addr, unsigned long *size,
1042 unsigned long *off, char *sym)
1047 static inline bool is_bpf_text_address(unsigned long addr)
1052 static inline int bpf_get_kallsym(unsigned int symnum, unsigned long *value,
1053 char *type, char *sym)
1058 static inline const char *
1059 bpf_address_lookup(unsigned long addr, unsigned long *size,
1060 unsigned long *off, char **modname, char *sym)
1065 static inline void bpf_prog_kallsyms_add(struct bpf_prog *fp)
1069 static inline void bpf_prog_kallsyms_del(struct bpf_prog *fp)
1073 static inline void bpf_get_prog_name(const struct bpf_prog *prog, char *sym)
1078 #endif /* CONFIG_BPF_JIT */
1080 void bpf_prog_kallsyms_del_subprogs(struct bpf_prog *fp);
1081 void bpf_prog_kallsyms_del_all(struct bpf_prog *fp);
1083 #define BPF_ANC BIT(15)
1085 static inline bool bpf_needs_clear_a(const struct sock_filter *first)
1087 switch (first->code) {
1088 case BPF_RET | BPF_K:
1089 case BPF_LD | BPF_W | BPF_LEN:
1092 case BPF_LD | BPF_W | BPF_ABS:
1093 case BPF_LD | BPF_H | BPF_ABS:
1094 case BPF_LD | BPF_B | BPF_ABS:
1095 if (first->k == SKF_AD_OFF + SKF_AD_ALU_XOR_X)
1104 static inline u16 bpf_anc_helper(const struct sock_filter *ftest)
1106 BUG_ON(ftest->code & BPF_ANC);
1108 switch (ftest->code) {
1109 case BPF_LD | BPF_W | BPF_ABS:
1110 case BPF_LD | BPF_H | BPF_ABS:
1111 case BPF_LD | BPF_B | BPF_ABS:
1112 #define BPF_ANCILLARY(CODE) case SKF_AD_OFF + SKF_AD_##CODE: \
1113 return BPF_ANC | SKF_AD_##CODE
1115 BPF_ANCILLARY(PROTOCOL);
1116 BPF_ANCILLARY(PKTTYPE);
1117 BPF_ANCILLARY(IFINDEX);
1118 BPF_ANCILLARY(NLATTR);
1119 BPF_ANCILLARY(NLATTR_NEST);
1120 BPF_ANCILLARY(MARK);
1121 BPF_ANCILLARY(QUEUE);
1122 BPF_ANCILLARY(HATYPE);
1123 BPF_ANCILLARY(RXHASH);
1125 BPF_ANCILLARY(ALU_XOR_X);
1126 BPF_ANCILLARY(VLAN_TAG);
1127 BPF_ANCILLARY(VLAN_TAG_PRESENT);
1128 BPF_ANCILLARY(PAY_OFFSET);
1129 BPF_ANCILLARY(RANDOM);
1130 BPF_ANCILLARY(VLAN_TPID);
1138 void *bpf_internal_load_pointer_neg_helper(const struct sk_buff *skb,
1139 int k, unsigned int size);
1141 static inline void *bpf_load_pointer(const struct sk_buff *skb, int k,
1142 unsigned int size, void *buffer)
1145 return skb_header_pointer(skb, k, size, buffer);
1147 return bpf_internal_load_pointer_neg_helper(skb, k, size);
1150 static inline int bpf_tell_extensions(void)
1155 struct bpf_sock_addr_kern {
1157 struct sockaddr *uaddr;
1158 /* Temporary "register" to make indirect stores to nested structures
1159 * defined above. We need three registers to make such a store, but
1160 * only two (src and dst) are available at convert_ctx_access time
1163 void *t_ctx; /* Attach type specific context. */
1166 struct bpf_sock_ops_kern {
1175 u64 temp; /* temp and everything after is not
1176 * initialized to 0 before calling
1177 * the BPF program. New fields that
1178 * should be initialized to 0 should
1179 * be inserted before temp.
1180 * temp is scratch storage used by
1181 * sock_ops_convert_ctx_access
1182 * as temporary storage of a register.
1186 struct bpf_sysctl_kern {
1187 struct ctl_table_header *head;
1188 struct ctl_table *table;
1196 /* Temporary "register" for indirect stores to ppos. */
1200 #endif /* __LINUX_FILTER_H__ */