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 <asm/byteorder.h>
28 #include <uapi/linux/filter.h>
29 #include <uapi/linux/bpf.h>
37 struct sock_reuseport;
39 struct ctl_table_header;
41 /* ArgX, context and stack frame pointer register positions. Note,
42 * Arg1, Arg2, Arg3, etc are used as argument mappings of function
43 * calls in BPF_CALL instruction.
45 #define BPF_REG_ARG1 BPF_REG_1
46 #define BPF_REG_ARG2 BPF_REG_2
47 #define BPF_REG_ARG3 BPF_REG_3
48 #define BPF_REG_ARG4 BPF_REG_4
49 #define BPF_REG_ARG5 BPF_REG_5
50 #define BPF_REG_CTX BPF_REG_6
51 #define BPF_REG_FP BPF_REG_10
53 /* Additional register mappings for converted user programs. */
54 #define BPF_REG_A BPF_REG_0
55 #define BPF_REG_X BPF_REG_7
56 #define BPF_REG_TMP BPF_REG_2 /* scratch reg */
57 #define BPF_REG_D BPF_REG_8 /* data, callee-saved */
58 #define BPF_REG_H BPF_REG_9 /* hlen, callee-saved */
60 /* Kernel hidden auxiliary/helper register. */
61 #define BPF_REG_AX MAX_BPF_REG
62 #define MAX_BPF_EXT_REG (MAX_BPF_REG + 1)
63 #define MAX_BPF_JIT_REG MAX_BPF_EXT_REG
65 /* unused opcode to mark special call to bpf_tail_call() helper */
66 #define BPF_TAIL_CALL 0xf0
68 /* unused opcode to mark call to interpreter with arguments */
69 #define BPF_CALL_ARGS 0xe0
71 /* unused opcode to mark speculation barrier for mitigating
72 * Speculative Store Bypass
74 #define BPF_NOSPEC 0xc0
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, \
263 /* Atomic memory add, *(uint *)(dst_reg + off16) += src_reg */
265 #define BPF_STX_XADD(SIZE, DST, SRC, OFF) \
266 ((struct bpf_insn) { \
267 .code = BPF_STX | BPF_SIZE(SIZE) | BPF_XADD, \
273 /* Memory store, *(uint *) (dst_reg + off16) = imm32 */
275 #define BPF_ST_MEM(SIZE, DST, OFF, IMM) \
276 ((struct bpf_insn) { \
277 .code = BPF_ST | BPF_SIZE(SIZE) | BPF_MEM, \
283 /* Conditional jumps against registers, if (dst_reg 'op' src_reg) goto pc + off16 */
285 #define BPF_JMP_REG(OP, DST, SRC, OFF) \
286 ((struct bpf_insn) { \
287 .code = BPF_JMP | BPF_OP(OP) | BPF_X, \
293 /* Conditional jumps against immediates, if (dst_reg 'op' imm32) goto pc + off16 */
295 #define BPF_JMP_IMM(OP, DST, IMM, OFF) \
296 ((struct bpf_insn) { \
297 .code = BPF_JMP | BPF_OP(OP) | BPF_K, \
303 /* Like BPF_JMP_REG, but with 32-bit wide operands for comparison. */
305 #define BPF_JMP32_REG(OP, DST, SRC, OFF) \
306 ((struct bpf_insn) { \
307 .code = BPF_JMP32 | BPF_OP(OP) | BPF_X, \
313 /* Like BPF_JMP_IMM, but with 32-bit wide operands for comparison. */
315 #define BPF_JMP32_IMM(OP, DST, IMM, OFF) \
316 ((struct bpf_insn) { \
317 .code = BPF_JMP32 | BPF_OP(OP) | BPF_K, \
323 /* Unconditional jumps, goto pc + off16 */
325 #define BPF_JMP_A(OFF) \
326 ((struct bpf_insn) { \
327 .code = BPF_JMP | BPF_JA, \
335 #define BPF_CALL_REL(TGT) \
336 ((struct bpf_insn) { \
337 .code = BPF_JMP | BPF_CALL, \
339 .src_reg = BPF_PSEUDO_CALL, \
345 #define BPF_CAST_CALL(x) \
346 ((u64 (*)(u64, u64, u64, u64, u64))(x))
348 #define BPF_EMIT_CALL(FUNC) \
349 ((struct bpf_insn) { \
350 .code = BPF_JMP | BPF_CALL, \
354 .imm = ((FUNC) - __bpf_call_base) })
356 /* Raw code statement block */
358 #define BPF_RAW_INSN(CODE, DST, SRC, OFF, IMM) \
359 ((struct bpf_insn) { \
368 #define BPF_EXIT_INSN() \
369 ((struct bpf_insn) { \
370 .code = BPF_JMP | BPF_EXIT, \
376 /* Speculation barrier */
378 #define BPF_ST_NOSPEC() \
379 ((struct bpf_insn) { \
380 .code = BPF_ST | BPF_NOSPEC, \
386 /* Internal classic blocks for direct assignment */
388 #define __BPF_STMT(CODE, K) \
389 ((struct sock_filter) BPF_STMT(CODE, K))
391 #define __BPF_JUMP(CODE, K, JT, JF) \
392 ((struct sock_filter) BPF_JUMP(CODE, K, JT, JF))
394 #define bytes_to_bpf_size(bytes) \
396 int bpf_size = -EINVAL; \
398 if (bytes == sizeof(u8)) \
400 else if (bytes == sizeof(u16)) \
402 else if (bytes == sizeof(u32)) \
404 else if (bytes == sizeof(u64)) \
410 #define bpf_size_to_bytes(bpf_size) \
412 int bytes = -EINVAL; \
414 if (bpf_size == BPF_B) \
415 bytes = sizeof(u8); \
416 else if (bpf_size == BPF_H) \
417 bytes = sizeof(u16); \
418 else if (bpf_size == BPF_W) \
419 bytes = sizeof(u32); \
420 else if (bpf_size == BPF_DW) \
421 bytes = sizeof(u64); \
426 #define BPF_SIZEOF(type) \
428 const int __size = bytes_to_bpf_size(sizeof(type)); \
429 BUILD_BUG_ON(__size < 0); \
433 #define BPF_FIELD_SIZEOF(type, field) \
435 const int __size = bytes_to_bpf_size(FIELD_SIZEOF(type, field)); \
436 BUILD_BUG_ON(__size < 0); \
440 #define BPF_LDST_BYTES(insn) \
442 const int __size = bpf_size_to_bytes(BPF_SIZE((insn)->code)); \
443 WARN_ON(__size < 0); \
447 #define __BPF_MAP_0(m, v, ...) v
448 #define __BPF_MAP_1(m, v, t, a, ...) m(t, a)
449 #define __BPF_MAP_2(m, v, t, a, ...) m(t, a), __BPF_MAP_1(m, v, __VA_ARGS__)
450 #define __BPF_MAP_3(m, v, t, a, ...) m(t, a), __BPF_MAP_2(m, v, __VA_ARGS__)
451 #define __BPF_MAP_4(m, v, t, a, ...) m(t, a), __BPF_MAP_3(m, v, __VA_ARGS__)
452 #define __BPF_MAP_5(m, v, t, a, ...) m(t, a), __BPF_MAP_4(m, v, __VA_ARGS__)
454 #define __BPF_REG_0(...) __BPF_PAD(5)
455 #define __BPF_REG_1(...) __BPF_MAP(1, __VA_ARGS__), __BPF_PAD(4)
456 #define __BPF_REG_2(...) __BPF_MAP(2, __VA_ARGS__), __BPF_PAD(3)
457 #define __BPF_REG_3(...) __BPF_MAP(3, __VA_ARGS__), __BPF_PAD(2)
458 #define __BPF_REG_4(...) __BPF_MAP(4, __VA_ARGS__), __BPF_PAD(1)
459 #define __BPF_REG_5(...) __BPF_MAP(5, __VA_ARGS__)
461 #define __BPF_MAP(n, ...) __BPF_MAP_##n(__VA_ARGS__)
462 #define __BPF_REG(n, ...) __BPF_REG_##n(__VA_ARGS__)
464 #define __BPF_CAST(t, a) \
467 typeof(__builtin_choose_expr(sizeof(t) == sizeof(unsigned long), \
468 (unsigned long)0, (t)0))) a
472 #define __BPF_DECL_ARGS(t, a) t a
473 #define __BPF_DECL_REGS(t, a) u64 a
475 #define __BPF_PAD(n) \
476 __BPF_MAP(n, __BPF_DECL_ARGS, __BPF_N, u64, __ur_1, u64, __ur_2, \
477 u64, __ur_3, u64, __ur_4, u64, __ur_5)
479 #define BPF_CALL_x(x, name, ...) \
480 static __always_inline \
481 u64 ____##name(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__)); \
482 u64 name(__BPF_REG(x, __BPF_DECL_REGS, __BPF_N, __VA_ARGS__)); \
483 u64 name(__BPF_REG(x, __BPF_DECL_REGS, __BPF_N, __VA_ARGS__)) \
485 return ____##name(__BPF_MAP(x,__BPF_CAST,__BPF_N,__VA_ARGS__));\
487 static __always_inline \
488 u64 ____##name(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__))
490 #define BPF_CALL_0(name, ...) BPF_CALL_x(0, name, __VA_ARGS__)
491 #define BPF_CALL_1(name, ...) BPF_CALL_x(1, name, __VA_ARGS__)
492 #define BPF_CALL_2(name, ...) BPF_CALL_x(2, name, __VA_ARGS__)
493 #define BPF_CALL_3(name, ...) BPF_CALL_x(3, name, __VA_ARGS__)
494 #define BPF_CALL_4(name, ...) BPF_CALL_x(4, name, __VA_ARGS__)
495 #define BPF_CALL_5(name, ...) BPF_CALL_x(5, name, __VA_ARGS__)
497 #define bpf_ctx_range(TYPE, MEMBER) \
498 offsetof(TYPE, MEMBER) ... offsetofend(TYPE, MEMBER) - 1
499 #define bpf_ctx_range_till(TYPE, MEMBER1, MEMBER2) \
500 offsetof(TYPE, MEMBER1) ... offsetofend(TYPE, MEMBER2) - 1
501 #if BITS_PER_LONG == 64
502 # define bpf_ctx_range_ptr(TYPE, MEMBER) \
503 offsetof(TYPE, MEMBER) ... offsetofend(TYPE, MEMBER) - 1
505 # define bpf_ctx_range_ptr(TYPE, MEMBER) \
506 offsetof(TYPE, MEMBER) ... offsetof(TYPE, MEMBER) + 8 - 1
507 #endif /* BITS_PER_LONG == 64 */
509 #define bpf_target_off(TYPE, MEMBER, SIZE, PTR_SIZE) \
511 BUILD_BUG_ON(FIELD_SIZEOF(TYPE, MEMBER) != (SIZE)); \
512 *(PTR_SIZE) = (SIZE); \
513 offsetof(TYPE, MEMBER); \
517 /* A struct sock_filter is architecture independent. */
518 struct compat_sock_fprog {
520 compat_uptr_t filter; /* struct sock_filter * */
524 struct sock_fprog_kern {
526 struct sock_filter *filter;
529 struct bpf_binary_header {
531 /* Some arches need word alignment for their instructions */
532 u8 image[] __aligned(4);
536 u16 pages; /* Number of allocated pages */
537 u16 jited:1, /* Is our filter JIT'ed? */
538 jit_requested:1,/* archs need to JIT the prog */
539 gpl_compatible:1, /* Is filter GPL compatible? */
540 cb_access:1, /* Is control block accessed? */
541 dst_needed:1, /* Do we need dst entry? */
542 blinded:1, /* Was blinded */
543 is_func:1, /* program is a bpf function */
544 kprobe_override:1, /* Do we override a kprobe? */
545 has_callchain_buf:1, /* callchain buffer allocated? */
546 enforce_expected_attach_type:1; /* Enforce expected_attach_type checking at attach time */
547 enum bpf_prog_type type; /* Type of BPF program */
548 enum bpf_attach_type expected_attach_type; /* For some prog types */
549 u32 len; /* Number of filter blocks */
550 u32 jited_len; /* Size of jited insns in bytes */
551 u8 tag[BPF_TAG_SIZE];
552 struct bpf_prog_aux *aux; /* Auxiliary fields */
553 struct sock_fprog_kern *orig_prog; /* Original BPF program */
554 unsigned int (*bpf_func)(const void *ctx,
555 const struct bpf_insn *insn);
556 /* Instructions for interpreter */
558 struct sock_filter insns[0];
559 struct bpf_insn insnsi[0];
566 struct bpf_prog *prog;
569 DECLARE_STATIC_KEY_FALSE(bpf_stats_enabled_key);
571 #define BPF_PROG_RUN(prog, ctx) ({ \
574 if (static_branch_unlikely(&bpf_stats_enabled_key)) { \
575 struct bpf_prog_stats *stats; \
576 u64 start = sched_clock(); \
577 ret = (*(prog)->bpf_func)(ctx, (prog)->insnsi); \
578 stats = this_cpu_ptr(prog->aux->stats); \
579 u64_stats_update_begin(&stats->syncp); \
581 stats->nsecs += sched_clock() - start; \
582 u64_stats_update_end(&stats->syncp); \
584 ret = (*(prog)->bpf_func)(ctx, (prog)->insnsi); \
588 #define BPF_SKB_CB_LEN QDISC_CB_PRIV_LEN
590 struct bpf_skb_data_end {
591 struct qdisc_skb_cb qdisc_cb;
596 struct bpf_redirect_info {
601 struct bpf_map *map_to_flush;
605 DECLARE_PER_CPU(struct bpf_redirect_info, bpf_redirect_info);
607 /* flags for bpf_redirect_info kern_flags */
608 #define BPF_RI_F_RF_NO_DIRECT BIT(0) /* no napi_direct on return_frame */
610 /* Compute the linear packet data range [data, data_end) which
611 * will be accessed by various program types (cls_bpf, act_bpf,
612 * lwt, ...). Subsystems allowing direct data access must (!)
613 * ensure that cb[] area can be written to when BPF program is
614 * invoked (otherwise cb[] save/restore is necessary).
616 static inline void bpf_compute_data_pointers(struct sk_buff *skb)
618 struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb;
620 BUILD_BUG_ON(sizeof(*cb) > FIELD_SIZEOF(struct sk_buff, cb));
621 cb->data_meta = skb->data - skb_metadata_len(skb);
622 cb->data_end = skb->data + skb_headlen(skb);
625 /* Similar to bpf_compute_data_pointers(), except that save orginal
626 * data in cb->data and cb->meta_data for restore.
628 static inline void bpf_compute_and_save_data_end(
629 struct sk_buff *skb, void **saved_data_end)
631 struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb;
633 *saved_data_end = cb->data_end;
634 cb->data_end = skb->data + skb_headlen(skb);
637 /* Restore data saved by bpf_compute_data_pointers(). */
638 static inline void bpf_restore_data_end(
639 struct sk_buff *skb, void *saved_data_end)
641 struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb;
643 cb->data_end = saved_data_end;
646 static inline u8 *bpf_skb_cb(struct sk_buff *skb)
648 /* eBPF programs may read/write skb->cb[] area to transfer meta
649 * data between tail calls. Since this also needs to work with
650 * tc, that scratch memory is mapped to qdisc_skb_cb's data area.
652 * In some socket filter cases, the cb unfortunately needs to be
653 * saved/restored so that protocol specific skb->cb[] data won't
654 * be lost. In any case, due to unpriviledged eBPF programs
655 * attached to sockets, we need to clear the bpf_skb_cb() area
656 * to not leak previous contents to user space.
658 BUILD_BUG_ON(FIELD_SIZEOF(struct __sk_buff, cb) != BPF_SKB_CB_LEN);
659 BUILD_BUG_ON(FIELD_SIZEOF(struct __sk_buff, cb) !=
660 FIELD_SIZEOF(struct qdisc_skb_cb, data));
662 return qdisc_skb_cb(skb)->data;
665 static inline u32 __bpf_prog_run_save_cb(const struct bpf_prog *prog,
668 u8 *cb_data = bpf_skb_cb(skb);
669 u8 cb_saved[BPF_SKB_CB_LEN];
672 if (unlikely(prog->cb_access)) {
673 memcpy(cb_saved, cb_data, sizeof(cb_saved));
674 memset(cb_data, 0, sizeof(cb_saved));
677 res = BPF_PROG_RUN(prog, skb);
679 if (unlikely(prog->cb_access))
680 memcpy(cb_data, cb_saved, sizeof(cb_saved));
685 static inline u32 bpf_prog_run_save_cb(const struct bpf_prog *prog,
691 res = __bpf_prog_run_save_cb(prog, skb);
696 static inline u32 bpf_prog_run_clear_cb(const struct bpf_prog *prog,
699 u8 *cb_data = bpf_skb_cb(skb);
702 if (unlikely(prog->cb_access))
703 memset(cb_data, 0, BPF_SKB_CB_LEN);
706 res = BPF_PROG_RUN(prog, skb);
711 static __always_inline u32 bpf_prog_run_xdp(const struct bpf_prog *prog,
712 struct xdp_buff *xdp)
714 /* Caller needs to hold rcu_read_lock() (!), otherwise program
715 * can be released while still running, or map elements could be
716 * freed early while still having concurrent users. XDP fastpath
717 * already takes rcu_read_lock() when fetching the program, so
718 * it's not necessary here anymore.
720 return BPF_PROG_RUN(prog, xdp);
723 static inline u32 bpf_prog_insn_size(const struct bpf_prog *prog)
725 return prog->len * sizeof(struct bpf_insn);
728 static inline u32 bpf_prog_tag_scratch_size(const struct bpf_prog *prog)
730 return round_up(bpf_prog_insn_size(prog) +
731 sizeof(__be64) + 1, SHA_MESSAGE_BYTES);
734 static inline unsigned int bpf_prog_size(unsigned int proglen)
736 return max(sizeof(struct bpf_prog),
737 offsetof(struct bpf_prog, insns[proglen]));
740 static inline bool bpf_prog_was_classic(const struct bpf_prog *prog)
742 /* When classic BPF programs have been loaded and the arch
743 * does not have a classic BPF JIT (anymore), they have been
744 * converted via bpf_migrate_filter() to eBPF and thus always
745 * have an unspec program type.
747 return prog->type == BPF_PROG_TYPE_UNSPEC;
750 static inline u32 bpf_ctx_off_adjust_machine(u32 size)
752 const u32 size_machine = sizeof(unsigned long);
754 if (size > size_machine && size % size_machine == 0)
761 bpf_ctx_narrow_access_ok(u32 off, u32 size, u32 size_default)
763 return size <= size_default && (size & (size - 1)) == 0;
767 bpf_ctx_narrow_access_offset(u32 off, u32 size, u32 size_default)
769 u8 access_off = off & (size_default - 1);
771 #ifdef __LITTLE_ENDIAN
774 return size_default - (access_off + size);
778 #define bpf_ctx_wide_access_ok(off, size, type, field) \
779 (size == sizeof(__u64) && \
780 off >= offsetof(type, field) && \
781 off + sizeof(__u64) <= offsetofend(type, field) && \
782 off % sizeof(__u64) == 0)
784 #define bpf_classic_proglen(fprog) (fprog->len * sizeof(fprog->filter[0]))
786 static inline void bpf_prog_lock_ro(struct bpf_prog *fp)
788 #ifndef CONFIG_BPF_JIT_ALWAYS_ON
790 set_vm_flush_reset_perms(fp);
791 set_memory_ro((unsigned long)fp, fp->pages);
796 static inline void bpf_jit_binary_lock_ro(struct bpf_binary_header *hdr)
798 set_vm_flush_reset_perms(hdr);
799 set_memory_ro((unsigned long)hdr, hdr->pages);
800 set_memory_x((unsigned long)hdr, hdr->pages);
803 static inline struct bpf_binary_header *
804 bpf_jit_binary_hdr(const struct bpf_prog *fp)
806 unsigned long real_start = (unsigned long)fp->bpf_func;
807 unsigned long addr = real_start & PAGE_MASK;
812 int sk_filter_trim_cap(struct sock *sk, struct sk_buff *skb, unsigned int cap);
813 static inline int sk_filter(struct sock *sk, struct sk_buff *skb)
815 return sk_filter_trim_cap(sk, skb, 1);
818 struct bpf_prog *bpf_prog_select_runtime(struct bpf_prog *fp, int *err);
819 void bpf_prog_free(struct bpf_prog *fp);
821 bool bpf_opcode_in_insntable(u8 code);
823 void bpf_prog_free_linfo(struct bpf_prog *prog);
824 void bpf_prog_fill_jited_linfo(struct bpf_prog *prog,
825 const u32 *insn_to_jit_off);
826 int bpf_prog_alloc_jited_linfo(struct bpf_prog *prog);
827 void bpf_prog_free_jited_linfo(struct bpf_prog *prog);
828 void bpf_prog_free_unused_jited_linfo(struct bpf_prog *prog);
830 struct bpf_prog *bpf_prog_alloc(unsigned int size, gfp_t gfp_extra_flags);
831 struct bpf_prog *bpf_prog_alloc_no_stats(unsigned int size, gfp_t gfp_extra_flags);
832 struct bpf_prog *bpf_prog_realloc(struct bpf_prog *fp_old, unsigned int size,
833 gfp_t gfp_extra_flags);
834 void __bpf_prog_free(struct bpf_prog *fp);
836 static inline void bpf_prog_unlock_free(struct bpf_prog *fp)
841 typedef int (*bpf_aux_classic_check_t)(struct sock_filter *filter,
844 int bpf_prog_create(struct bpf_prog **pfp, struct sock_fprog_kern *fprog);
845 int bpf_prog_create_from_user(struct bpf_prog **pfp, struct sock_fprog *fprog,
846 bpf_aux_classic_check_t trans, bool save_orig);
847 void bpf_prog_destroy(struct bpf_prog *fp);
849 int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk);
850 int sk_attach_bpf(u32 ufd, struct sock *sk);
851 int sk_reuseport_attach_filter(struct sock_fprog *fprog, struct sock *sk);
852 int sk_reuseport_attach_bpf(u32 ufd, struct sock *sk);
853 void sk_reuseport_prog_free(struct bpf_prog *prog);
854 int sk_detach_filter(struct sock *sk);
855 int sk_get_filter(struct sock *sk, struct sock_filter __user *filter,
858 bool sk_filter_charge(struct sock *sk, struct sk_filter *fp);
859 void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp);
861 u64 __bpf_call_base(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5);
862 #define __bpf_call_base_args \
863 ((u64 (*)(u64, u64, u64, u64, u64, const struct bpf_insn *)) \
864 (void *)__bpf_call_base)
866 struct bpf_prog *bpf_int_jit_compile(struct bpf_prog *prog);
867 void bpf_jit_compile(struct bpf_prog *prog);
868 bool bpf_jit_needs_zext(void);
869 bool bpf_helper_changes_pkt_data(void *func);
871 static inline bool bpf_dump_raw_ok(const struct cred *cred)
873 /* Reconstruction of call-sites is dependent on kallsyms,
874 * thus make dump the same restriction.
876 return kallsyms_show_value(cred);
879 struct bpf_prog *bpf_patch_insn_single(struct bpf_prog *prog, u32 off,
880 const struct bpf_insn *patch, u32 len);
881 int bpf_remove_insns(struct bpf_prog *prog, u32 off, u32 cnt);
883 void bpf_clear_redirect_map(struct bpf_map *map);
885 static inline bool xdp_return_frame_no_direct(void)
887 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
889 return ri->kern_flags & BPF_RI_F_RF_NO_DIRECT;
892 static inline void xdp_set_return_frame_no_direct(void)
894 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
896 ri->kern_flags |= BPF_RI_F_RF_NO_DIRECT;
899 static inline void xdp_clear_return_frame_no_direct(void)
901 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
903 ri->kern_flags &= ~BPF_RI_F_RF_NO_DIRECT;
906 static inline int xdp_ok_fwd_dev(const struct net_device *fwd,
911 if (unlikely(!(fwd->flags & IFF_UP)))
914 len = fwd->mtu + fwd->hard_header_len + VLAN_HLEN;
921 /* The pair of xdp_do_redirect and xdp_do_flush_map MUST be called in the
922 * same cpu context. Further for best results no more than a single map
923 * for the do_redirect/do_flush pair should be used. This limitation is
924 * because we only track one map and force a flush when the map changes.
925 * This does not appear to be a real limitation for existing software.
927 int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb,
928 struct xdp_buff *xdp, struct bpf_prog *prog);
929 int xdp_do_redirect(struct net_device *dev,
930 struct xdp_buff *xdp,
931 struct bpf_prog *prog);
932 void xdp_do_flush_map(void);
934 void bpf_warn_invalid_xdp_action(u32 act);
937 struct sock *bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk,
938 struct bpf_prog *prog, struct sk_buff *skb,
941 static inline struct sock *
942 bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk,
943 struct bpf_prog *prog, struct sk_buff *skb,
950 #ifdef CONFIG_BPF_JIT
951 extern int bpf_jit_enable;
952 extern int bpf_jit_harden;
953 extern int bpf_jit_kallsyms;
954 extern long bpf_jit_limit;
955 extern long bpf_jit_limit_max;
957 typedef void (*bpf_jit_fill_hole_t)(void *area, unsigned int size);
959 struct bpf_binary_header *
960 bpf_jit_binary_alloc(unsigned int proglen, u8 **image_ptr,
961 unsigned int alignment,
962 bpf_jit_fill_hole_t bpf_fill_ill_insns);
963 void bpf_jit_binary_free(struct bpf_binary_header *hdr);
964 u64 bpf_jit_alloc_exec_limit(void);
965 void *bpf_jit_alloc_exec(unsigned long size);
966 void bpf_jit_free_exec(void *addr);
967 void bpf_jit_free(struct bpf_prog *fp);
969 int bpf_jit_get_func_addr(const struct bpf_prog *prog,
970 const struct bpf_insn *insn, bool extra_pass,
971 u64 *func_addr, bool *func_addr_fixed);
973 struct bpf_prog *bpf_jit_blind_constants(struct bpf_prog *fp);
974 void bpf_jit_prog_release_other(struct bpf_prog *fp, struct bpf_prog *fp_other);
976 static inline void bpf_jit_dump(unsigned int flen, unsigned int proglen,
977 u32 pass, void *image)
979 pr_err("flen=%u proglen=%u pass=%u image=%pK from=%s pid=%d\n", flen,
980 proglen, pass, image, current->comm, task_pid_nr(current));
983 print_hex_dump(KERN_ERR, "JIT code: ", DUMP_PREFIX_OFFSET,
984 16, 1, image, proglen, false);
987 static inline bool bpf_jit_is_ebpf(void)
989 # ifdef CONFIG_HAVE_EBPF_JIT
996 static inline bool ebpf_jit_enabled(void)
998 return bpf_jit_enable && bpf_jit_is_ebpf();
1001 static inline bool bpf_prog_ebpf_jited(const struct bpf_prog *fp)
1003 return fp->jited && bpf_jit_is_ebpf();
1006 static inline bool bpf_jit_blinding_enabled(struct bpf_prog *prog)
1008 /* These are the prerequisites, should someone ever have the
1009 * idea to call blinding outside of them, we make sure to
1012 if (!bpf_jit_is_ebpf())
1014 if (!prog->jit_requested)
1016 if (!bpf_jit_harden)
1018 if (bpf_jit_harden == 1 && capable(CAP_SYS_ADMIN))
1024 static inline bool bpf_jit_kallsyms_enabled(void)
1026 /* There are a couple of corner cases where kallsyms should
1027 * not be enabled f.e. on hardening.
1031 if (!bpf_jit_kallsyms)
1033 if (bpf_jit_kallsyms == 1)
1039 const char *__bpf_address_lookup(unsigned long addr, unsigned long *size,
1040 unsigned long *off, char *sym);
1041 bool is_bpf_text_address(unsigned long addr);
1042 int bpf_get_kallsym(unsigned int symnum, unsigned long *value, char *type,
1045 static inline const char *
1046 bpf_address_lookup(unsigned long addr, unsigned long *size,
1047 unsigned long *off, char **modname, char *sym)
1049 const char *ret = __bpf_address_lookup(addr, size, off, sym);
1056 void bpf_prog_kallsyms_add(struct bpf_prog *fp);
1057 void bpf_prog_kallsyms_del(struct bpf_prog *fp);
1058 void bpf_get_prog_name(const struct bpf_prog *prog, char *sym);
1060 #else /* CONFIG_BPF_JIT */
1062 static inline bool ebpf_jit_enabled(void)
1067 static inline bool bpf_prog_ebpf_jited(const struct bpf_prog *fp)
1072 static inline void bpf_jit_free(struct bpf_prog *fp)
1074 bpf_prog_unlock_free(fp);
1077 static inline bool bpf_jit_kallsyms_enabled(void)
1082 static inline const char *
1083 __bpf_address_lookup(unsigned long addr, unsigned long *size,
1084 unsigned long *off, char *sym)
1089 static inline bool is_bpf_text_address(unsigned long addr)
1094 static inline int bpf_get_kallsym(unsigned int symnum, unsigned long *value,
1095 char *type, char *sym)
1100 static inline const char *
1101 bpf_address_lookup(unsigned long addr, unsigned long *size,
1102 unsigned long *off, char **modname, char *sym)
1107 static inline void bpf_prog_kallsyms_add(struct bpf_prog *fp)
1111 static inline void bpf_prog_kallsyms_del(struct bpf_prog *fp)
1115 static inline void bpf_get_prog_name(const struct bpf_prog *prog, char *sym)
1120 #endif /* CONFIG_BPF_JIT */
1122 void bpf_prog_kallsyms_del_all(struct bpf_prog *fp);
1124 #define BPF_ANC BIT(15)
1126 static inline bool bpf_needs_clear_a(const struct sock_filter *first)
1128 switch (first->code) {
1129 case BPF_RET | BPF_K:
1130 case BPF_LD | BPF_W | BPF_LEN:
1133 case BPF_LD | BPF_W | BPF_ABS:
1134 case BPF_LD | BPF_H | BPF_ABS:
1135 case BPF_LD | BPF_B | BPF_ABS:
1136 if (first->k == SKF_AD_OFF + SKF_AD_ALU_XOR_X)
1145 static inline u16 bpf_anc_helper(const struct sock_filter *ftest)
1147 BUG_ON(ftest->code & BPF_ANC);
1149 switch (ftest->code) {
1150 case BPF_LD | BPF_W | BPF_ABS:
1151 case BPF_LD | BPF_H | BPF_ABS:
1152 case BPF_LD | BPF_B | BPF_ABS:
1153 #define BPF_ANCILLARY(CODE) case SKF_AD_OFF + SKF_AD_##CODE: \
1154 return BPF_ANC | SKF_AD_##CODE
1156 BPF_ANCILLARY(PROTOCOL);
1157 BPF_ANCILLARY(PKTTYPE);
1158 BPF_ANCILLARY(IFINDEX);
1159 BPF_ANCILLARY(NLATTR);
1160 BPF_ANCILLARY(NLATTR_NEST);
1161 BPF_ANCILLARY(MARK);
1162 BPF_ANCILLARY(QUEUE);
1163 BPF_ANCILLARY(HATYPE);
1164 BPF_ANCILLARY(RXHASH);
1166 BPF_ANCILLARY(ALU_XOR_X);
1167 BPF_ANCILLARY(VLAN_TAG);
1168 BPF_ANCILLARY(VLAN_TAG_PRESENT);
1169 BPF_ANCILLARY(PAY_OFFSET);
1170 BPF_ANCILLARY(RANDOM);
1171 BPF_ANCILLARY(VLAN_TPID);
1179 void *bpf_internal_load_pointer_neg_helper(const struct sk_buff *skb,
1180 int k, unsigned int size);
1182 static inline void *bpf_load_pointer(const struct sk_buff *skb, int k,
1183 unsigned int size, void *buffer)
1186 return skb_header_pointer(skb, k, size, buffer);
1188 return bpf_internal_load_pointer_neg_helper(skb, k, size);
1191 static inline int bpf_tell_extensions(void)
1196 struct bpf_sock_addr_kern {
1198 struct sockaddr *uaddr;
1199 /* Temporary "register" to make indirect stores to nested structures
1200 * defined above. We need three registers to make such a store, but
1201 * only two (src and dst) are available at convert_ctx_access time
1204 void *t_ctx; /* Attach type specific context. */
1207 struct bpf_sock_ops_kern {
1216 u64 temp; /* temp and everything after is not
1217 * initialized to 0 before calling
1218 * the BPF program. New fields that
1219 * should be initialized to 0 should
1220 * be inserted before temp.
1221 * temp is scratch storage used by
1222 * sock_ops_convert_ctx_access
1223 * as temporary storage of a register.
1227 struct bpf_sysctl_kern {
1228 struct ctl_table_header *head;
1229 struct ctl_table *table;
1237 /* Temporary "register" for indirect stores to ppos. */
1241 struct bpf_sockopt_kern {
1251 #endif /* __LINUX_FILTER_H__ */