2 * Linux Socket Filter Data Structures
4 #ifndef __LINUX_FILTER_H__
5 #define __LINUX_FILTER_H__
9 #include <linux/atomic.h>
10 #include <linux/compat.h>
11 #include <linux/skbuff.h>
12 #include <linux/linkage.h>
13 #include <linux/printk.h>
14 #include <linux/workqueue.h>
15 #include <linux/sched.h>
16 #include <linux/capability.h>
18 #include <net/sch_generic.h>
20 #include <asm/cacheflush.h>
22 #include <uapi/linux/filter.h>
23 #include <uapi/linux/bpf.h>
30 /* ArgX, context and stack frame pointer register positions. Note,
31 * Arg1, Arg2, Arg3, etc are used as argument mappings of function
32 * calls in BPF_CALL instruction.
34 #define BPF_REG_ARG1 BPF_REG_1
35 #define BPF_REG_ARG2 BPF_REG_2
36 #define BPF_REG_ARG3 BPF_REG_3
37 #define BPF_REG_ARG4 BPF_REG_4
38 #define BPF_REG_ARG5 BPF_REG_5
39 #define BPF_REG_CTX BPF_REG_6
40 #define BPF_REG_FP BPF_REG_10
42 /* Additional register mappings for converted user programs. */
43 #define BPF_REG_A BPF_REG_0
44 #define BPF_REG_X BPF_REG_7
45 #define BPF_REG_TMP BPF_REG_8
47 /* Kernel hidden auxiliary/helper register for hardening step.
48 * Only used by eBPF JITs. It's nothing more than a temporary
49 * register that JITs use internally, only that here it's part
50 * of eBPF instructions that have been rewritten for blinding
51 * constants. See JIT pre-step in bpf_jit_blind_constants().
53 #define BPF_REG_AX MAX_BPF_REG
54 #define MAX_BPF_JIT_REG (MAX_BPF_REG + 1)
56 /* BPF program can access up to 512 bytes of stack space. */
57 #define MAX_BPF_STACK 512
59 /* Helper macros for filter block array initializers. */
61 /* ALU ops on registers, bpf_add|sub|...: dst_reg += src_reg */
63 #define BPF_ALU64_REG(OP, DST, SRC) \
64 ((struct bpf_insn) { \
65 .code = BPF_ALU64 | BPF_OP(OP) | BPF_X, \
71 #define BPF_ALU32_REG(OP, DST, SRC) \
72 ((struct bpf_insn) { \
73 .code = BPF_ALU | BPF_OP(OP) | BPF_X, \
79 /* ALU ops on immediates, bpf_add|sub|...: dst_reg += imm32 */
81 #define BPF_ALU64_IMM(OP, DST, IMM) \
82 ((struct bpf_insn) { \
83 .code = BPF_ALU64 | BPF_OP(OP) | BPF_K, \
89 #define BPF_ALU32_IMM(OP, DST, IMM) \
90 ((struct bpf_insn) { \
91 .code = BPF_ALU | BPF_OP(OP) | BPF_K, \
97 /* Endianess conversion, cpu_to_{l,b}e(), {l,b}e_to_cpu() */
99 #define BPF_ENDIAN(TYPE, DST, LEN) \
100 ((struct bpf_insn) { \
101 .code = BPF_ALU | BPF_END | BPF_SRC(TYPE), \
107 /* Short form of mov, dst_reg = src_reg */
109 #define BPF_MOV64_REG(DST, SRC) \
110 ((struct bpf_insn) { \
111 .code = BPF_ALU64 | BPF_MOV | BPF_X, \
117 #define BPF_MOV32_REG(DST, SRC) \
118 ((struct bpf_insn) { \
119 .code = BPF_ALU | BPF_MOV | BPF_X, \
125 /* Short form of mov, dst_reg = imm32 */
127 #define BPF_MOV64_IMM(DST, IMM) \
128 ((struct bpf_insn) { \
129 .code = BPF_ALU64 | BPF_MOV | BPF_K, \
135 #define BPF_MOV32_IMM(DST, IMM) \
136 ((struct bpf_insn) { \
137 .code = BPF_ALU | BPF_MOV | BPF_K, \
143 /* BPF_LD_IMM64 macro encodes single 'load 64-bit immediate' insn */
144 #define BPF_LD_IMM64(DST, IMM) \
145 BPF_LD_IMM64_RAW(DST, 0, IMM)
147 #define BPF_LD_IMM64_RAW(DST, SRC, IMM) \
148 ((struct bpf_insn) { \
149 .code = BPF_LD | BPF_DW | BPF_IMM, \
153 .imm = (__u32) (IMM) }), \
154 ((struct bpf_insn) { \
155 .code = 0, /* zero is reserved opcode */ \
159 .imm = ((__u64) (IMM)) >> 32 })
161 /* pseudo BPF_LD_IMM64 insn used to refer to process-local map_fd */
162 #define BPF_LD_MAP_FD(DST, MAP_FD) \
163 BPF_LD_IMM64_RAW(DST, BPF_PSEUDO_MAP_FD, MAP_FD)
165 /* Short form of mov based on type, BPF_X: dst_reg = src_reg, BPF_K: dst_reg = imm32 */
167 #define BPF_MOV64_RAW(TYPE, DST, SRC, IMM) \
168 ((struct bpf_insn) { \
169 .code = BPF_ALU64 | BPF_MOV | BPF_SRC(TYPE), \
175 #define BPF_MOV32_RAW(TYPE, DST, SRC, IMM) \
176 ((struct bpf_insn) { \
177 .code = BPF_ALU | BPF_MOV | BPF_SRC(TYPE), \
183 /* Direct packet access, R0 = *(uint *) (skb->data + imm32) */
185 #define BPF_LD_ABS(SIZE, IMM) \
186 ((struct bpf_insn) { \
187 .code = BPF_LD | BPF_SIZE(SIZE) | BPF_ABS, \
193 /* Indirect packet access, R0 = *(uint *) (skb->data + src_reg + imm32) */
195 #define BPF_LD_IND(SIZE, SRC, IMM) \
196 ((struct bpf_insn) { \
197 .code = BPF_LD | BPF_SIZE(SIZE) | BPF_IND, \
203 /* Memory load, dst_reg = *(uint *) (src_reg + off16) */
205 #define BPF_LDX_MEM(SIZE, DST, SRC, OFF) \
206 ((struct bpf_insn) { \
207 .code = BPF_LDX | BPF_SIZE(SIZE) | BPF_MEM, \
213 /* Memory store, *(uint *) (dst_reg + off16) = src_reg */
215 #define BPF_STX_MEM(SIZE, DST, SRC, OFF) \
216 ((struct bpf_insn) { \
217 .code = BPF_STX | BPF_SIZE(SIZE) | BPF_MEM, \
223 /* Atomic memory add, *(uint *)(dst_reg + off16) += src_reg */
225 #define BPF_STX_XADD(SIZE, DST, SRC, OFF) \
226 ((struct bpf_insn) { \
227 .code = BPF_STX | BPF_SIZE(SIZE) | BPF_XADD, \
233 /* Memory store, *(uint *) (dst_reg + off16) = imm32 */
235 #define BPF_ST_MEM(SIZE, DST, OFF, IMM) \
236 ((struct bpf_insn) { \
237 .code = BPF_ST | BPF_SIZE(SIZE) | BPF_MEM, \
243 /* Conditional jumps against registers, if (dst_reg 'op' src_reg) goto pc + off16 */
245 #define BPF_JMP_REG(OP, DST, SRC, OFF) \
246 ((struct bpf_insn) { \
247 .code = BPF_JMP | BPF_OP(OP) | BPF_X, \
253 /* Conditional jumps against immediates, if (dst_reg 'op' imm32) goto pc + off16 */
255 #define BPF_JMP_IMM(OP, DST, IMM, OFF) \
256 ((struct bpf_insn) { \
257 .code = BPF_JMP | BPF_OP(OP) | BPF_K, \
265 #define BPF_EMIT_CALL(FUNC) \
266 ((struct bpf_insn) { \
267 .code = BPF_JMP | BPF_CALL, \
271 .imm = ((FUNC) - __bpf_call_base) })
273 /* Raw code statement block */
275 #define BPF_RAW_INSN(CODE, DST, SRC, OFF, IMM) \
276 ((struct bpf_insn) { \
285 #define BPF_EXIT_INSN() \
286 ((struct bpf_insn) { \
287 .code = BPF_JMP | BPF_EXIT, \
293 /* Internal classic blocks for direct assignment */
295 #define __BPF_STMT(CODE, K) \
296 ((struct sock_filter) BPF_STMT(CODE, K))
298 #define __BPF_JUMP(CODE, K, JT, JF) \
299 ((struct sock_filter) BPF_JUMP(CODE, K, JT, JF))
301 #define bytes_to_bpf_size(bytes) \
303 int bpf_size = -EINVAL; \
305 if (bytes == sizeof(u8)) \
307 else if (bytes == sizeof(u16)) \
309 else if (bytes == sizeof(u32)) \
311 else if (bytes == sizeof(u64)) \
317 #define BPF_SIZEOF(type) \
319 const int __size = bytes_to_bpf_size(sizeof(type)); \
320 BUILD_BUG_ON(__size < 0); \
324 #define BPF_FIELD_SIZEOF(type, field) \
326 const int __size = bytes_to_bpf_size(FIELD_SIZEOF(type, field)); \
327 BUILD_BUG_ON(__size < 0); \
331 #define __BPF_MAP_0(m, v, ...) v
332 #define __BPF_MAP_1(m, v, t, a, ...) m(t, a)
333 #define __BPF_MAP_2(m, v, t, a, ...) m(t, a), __BPF_MAP_1(m, v, __VA_ARGS__)
334 #define __BPF_MAP_3(m, v, t, a, ...) m(t, a), __BPF_MAP_2(m, v, __VA_ARGS__)
335 #define __BPF_MAP_4(m, v, t, a, ...) m(t, a), __BPF_MAP_3(m, v, __VA_ARGS__)
336 #define __BPF_MAP_5(m, v, t, a, ...) m(t, a), __BPF_MAP_4(m, v, __VA_ARGS__)
338 #define __BPF_REG_0(...) __BPF_PAD(5)
339 #define __BPF_REG_1(...) __BPF_MAP(1, __VA_ARGS__), __BPF_PAD(4)
340 #define __BPF_REG_2(...) __BPF_MAP(2, __VA_ARGS__), __BPF_PAD(3)
341 #define __BPF_REG_3(...) __BPF_MAP(3, __VA_ARGS__), __BPF_PAD(2)
342 #define __BPF_REG_4(...) __BPF_MAP(4, __VA_ARGS__), __BPF_PAD(1)
343 #define __BPF_REG_5(...) __BPF_MAP(5, __VA_ARGS__)
345 #define __BPF_MAP(n, ...) __BPF_MAP_##n(__VA_ARGS__)
346 #define __BPF_REG(n, ...) __BPF_REG_##n(__VA_ARGS__)
348 #define __BPF_CAST(t, a) \
351 typeof(__builtin_choose_expr(sizeof(t) == sizeof(unsigned long), \
352 (unsigned long)0, (t)0))) a
356 #define __BPF_DECL_ARGS(t, a) t a
357 #define __BPF_DECL_REGS(t, a) u64 a
359 #define __BPF_PAD(n) \
360 __BPF_MAP(n, __BPF_DECL_ARGS, __BPF_N, u64, __ur_1, u64, __ur_2, \
361 u64, __ur_3, u64, __ur_4, u64, __ur_5)
363 #define BPF_CALL_x(x, name, ...) \
364 static __always_inline \
365 u64 ____##name(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__)); \
366 u64 name(__BPF_REG(x, __BPF_DECL_REGS, __BPF_N, __VA_ARGS__)); \
367 u64 name(__BPF_REG(x, __BPF_DECL_REGS, __BPF_N, __VA_ARGS__)) \
369 return ____##name(__BPF_MAP(x,__BPF_CAST,__BPF_N,__VA_ARGS__));\
371 static __always_inline \
372 u64 ____##name(__BPF_MAP(x, __BPF_DECL_ARGS, __BPF_V, __VA_ARGS__))
374 #define BPF_CALL_0(name, ...) BPF_CALL_x(0, name, __VA_ARGS__)
375 #define BPF_CALL_1(name, ...) BPF_CALL_x(1, name, __VA_ARGS__)
376 #define BPF_CALL_2(name, ...) BPF_CALL_x(2, name, __VA_ARGS__)
377 #define BPF_CALL_3(name, ...) BPF_CALL_x(3, name, __VA_ARGS__)
378 #define BPF_CALL_4(name, ...) BPF_CALL_x(4, name, __VA_ARGS__)
379 #define BPF_CALL_5(name, ...) BPF_CALL_x(5, name, __VA_ARGS__)
382 /* A struct sock_filter is architecture independent. */
383 struct compat_sock_fprog {
385 compat_uptr_t filter; /* struct sock_filter * */
389 struct sock_fprog_kern {
391 struct sock_filter *filter;
394 struct bpf_binary_header {
400 u16 pages; /* Number of allocated pages */
401 kmemcheck_bitfield_begin(meta);
402 u16 jited:1, /* Is our filter JIT'ed? */
403 gpl_compatible:1, /* Is filter GPL compatible? */
404 cb_access:1, /* Is control block accessed? */
405 dst_needed:1; /* Do we need dst entry? */
406 kmemcheck_bitfield_end(meta);
407 u32 len; /* Number of filter blocks */
408 enum bpf_prog_type type; /* Type of BPF program */
409 struct bpf_prog_aux *aux; /* Auxiliary fields */
410 struct sock_fprog_kern *orig_prog; /* Original BPF program */
411 unsigned int (*bpf_func)(const struct sk_buff *skb,
412 const struct bpf_insn *filter);
413 /* Instructions for interpreter */
415 struct sock_filter insns[0];
416 struct bpf_insn insnsi[0];
423 struct bpf_prog *prog;
426 #define BPF_PROG_RUN(filter, ctx) (*filter->bpf_func)(ctx, filter->insnsi)
428 #define BPF_SKB_CB_LEN QDISC_CB_PRIV_LEN
430 struct bpf_skb_data_end {
431 struct qdisc_skb_cb qdisc_cb;
440 /* compute the linear packet data range [data, data_end) which
441 * will be accessed by cls_bpf and act_bpf programs
443 static inline void bpf_compute_data_end(struct sk_buff *skb)
445 struct bpf_skb_data_end *cb = (struct bpf_skb_data_end *)skb->cb;
447 BUILD_BUG_ON(sizeof(*cb) > FIELD_SIZEOF(struct sk_buff, cb));
448 cb->data_end = skb->data + skb_headlen(skb);
451 static inline u8 *bpf_skb_cb(struct sk_buff *skb)
453 /* eBPF programs may read/write skb->cb[] area to transfer meta
454 * data between tail calls. Since this also needs to work with
455 * tc, that scratch memory is mapped to qdisc_skb_cb's data area.
457 * In some socket filter cases, the cb unfortunately needs to be
458 * saved/restored so that protocol specific skb->cb[] data won't
459 * be lost. In any case, due to unpriviledged eBPF programs
460 * attached to sockets, we need to clear the bpf_skb_cb() area
461 * to not leak previous contents to user space.
463 BUILD_BUG_ON(FIELD_SIZEOF(struct __sk_buff, cb) != BPF_SKB_CB_LEN);
464 BUILD_BUG_ON(FIELD_SIZEOF(struct __sk_buff, cb) !=
465 FIELD_SIZEOF(struct qdisc_skb_cb, data));
467 return qdisc_skb_cb(skb)->data;
470 static inline u32 bpf_prog_run_save_cb(const struct bpf_prog *prog,
473 u8 *cb_data = bpf_skb_cb(skb);
474 u8 cb_saved[BPF_SKB_CB_LEN];
477 if (unlikely(prog->cb_access)) {
478 memcpy(cb_saved, cb_data, sizeof(cb_saved));
479 memset(cb_data, 0, sizeof(cb_saved));
482 res = BPF_PROG_RUN(prog, skb);
484 if (unlikely(prog->cb_access))
485 memcpy(cb_data, cb_saved, sizeof(cb_saved));
490 static inline u32 bpf_prog_run_clear_cb(const struct bpf_prog *prog,
493 u8 *cb_data = bpf_skb_cb(skb);
495 if (unlikely(prog->cb_access))
496 memset(cb_data, 0, BPF_SKB_CB_LEN);
498 return BPF_PROG_RUN(prog, skb);
501 static inline u32 bpf_prog_run_xdp(const struct bpf_prog *prog,
502 struct xdp_buff *xdp)
507 ret = BPF_PROG_RUN(prog, (void *)xdp);
513 static inline unsigned int bpf_prog_size(unsigned int proglen)
515 return max(sizeof(struct bpf_prog),
516 offsetof(struct bpf_prog, insns[proglen]));
519 static inline bool bpf_prog_was_classic(const struct bpf_prog *prog)
521 /* When classic BPF programs have been loaded and the arch
522 * does not have a classic BPF JIT (anymore), they have been
523 * converted via bpf_migrate_filter() to eBPF and thus always
524 * have an unspec program type.
526 return prog->type == BPF_PROG_TYPE_UNSPEC;
529 #define bpf_classic_proglen(fprog) (fprog->len * sizeof(fprog->filter[0]))
531 #ifdef CONFIG_DEBUG_SET_MODULE_RONX
532 static inline void bpf_prog_lock_ro(struct bpf_prog *fp)
534 set_memory_ro((unsigned long)fp, fp->pages);
537 static inline void bpf_prog_unlock_ro(struct bpf_prog *fp)
539 set_memory_rw((unsigned long)fp, fp->pages);
542 static inline void bpf_prog_lock_ro(struct bpf_prog *fp)
546 static inline void bpf_prog_unlock_ro(struct bpf_prog *fp)
549 #endif /* CONFIG_DEBUG_SET_MODULE_RONX */
551 int sk_filter_trim_cap(struct sock *sk, struct sk_buff *skb, unsigned int cap);
552 static inline int sk_filter(struct sock *sk, struct sk_buff *skb)
554 return sk_filter_trim_cap(sk, skb, 1);
557 struct bpf_prog *bpf_prog_select_runtime(struct bpf_prog *fp, int *err);
558 void bpf_prog_free(struct bpf_prog *fp);
560 struct bpf_prog *bpf_prog_alloc(unsigned int size, gfp_t gfp_extra_flags);
561 struct bpf_prog *bpf_prog_realloc(struct bpf_prog *fp_old, unsigned int size,
562 gfp_t gfp_extra_flags);
563 void __bpf_prog_free(struct bpf_prog *fp);
565 static inline void bpf_prog_unlock_free(struct bpf_prog *fp)
567 bpf_prog_unlock_ro(fp);
571 typedef int (*bpf_aux_classic_check_t)(struct sock_filter *filter,
574 int bpf_prog_create(struct bpf_prog **pfp, struct sock_fprog_kern *fprog);
575 int bpf_prog_create_from_user(struct bpf_prog **pfp, struct sock_fprog *fprog,
576 bpf_aux_classic_check_t trans, bool save_orig);
577 void bpf_prog_destroy(struct bpf_prog *fp);
579 int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk);
580 int sk_attach_bpf(u32 ufd, struct sock *sk);
581 int sk_reuseport_attach_filter(struct sock_fprog *fprog, struct sock *sk);
582 int sk_reuseport_attach_bpf(u32 ufd, struct sock *sk);
583 int sk_detach_filter(struct sock *sk);
584 int sk_get_filter(struct sock *sk, struct sock_filter __user *filter,
587 bool sk_filter_charge(struct sock *sk, struct sk_filter *fp);
588 void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp);
590 u64 __bpf_call_base(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5);
592 struct bpf_prog *bpf_int_jit_compile(struct bpf_prog *prog);
593 bool bpf_helper_changes_skb_data(void *func);
595 struct bpf_prog *bpf_patch_insn_single(struct bpf_prog *prog, u32 off,
596 const struct bpf_insn *patch, u32 len);
597 void bpf_warn_invalid_xdp_action(u32 act);
599 #ifdef CONFIG_BPF_JIT
600 extern int bpf_jit_enable;
601 extern int bpf_jit_harden;
603 typedef void (*bpf_jit_fill_hole_t)(void *area, unsigned int size);
605 struct bpf_binary_header *
606 bpf_jit_binary_alloc(unsigned int proglen, u8 **image_ptr,
607 unsigned int alignment,
608 bpf_jit_fill_hole_t bpf_fill_ill_insns);
609 void bpf_jit_binary_free(struct bpf_binary_header *hdr);
611 void bpf_jit_compile(struct bpf_prog *fp);
612 void bpf_jit_free(struct bpf_prog *fp);
614 struct bpf_prog *bpf_jit_blind_constants(struct bpf_prog *fp);
615 void bpf_jit_prog_release_other(struct bpf_prog *fp, struct bpf_prog *fp_other);
617 static inline void bpf_jit_dump(unsigned int flen, unsigned int proglen,
618 u32 pass, void *image)
620 pr_err("flen=%u proglen=%u pass=%u image=%pK from=%s pid=%d\n", flen,
621 proglen, pass, image, current->comm, task_pid_nr(current));
624 print_hex_dump(KERN_ERR, "JIT code: ", DUMP_PREFIX_OFFSET,
625 16, 1, image, proglen, false);
628 static inline bool bpf_jit_is_ebpf(void)
630 # ifdef CONFIG_HAVE_EBPF_JIT
637 static inline bool bpf_jit_blinding_enabled(void)
639 /* These are the prerequisites, should someone ever have the
640 * idea to call blinding outside of them, we make sure to
643 if (!bpf_jit_is_ebpf())
649 if (bpf_jit_harden == 1 && capable(CAP_SYS_ADMIN))
655 static inline void bpf_jit_compile(struct bpf_prog *fp)
659 static inline void bpf_jit_free(struct bpf_prog *fp)
661 bpf_prog_unlock_free(fp);
663 #endif /* CONFIG_BPF_JIT */
665 #define BPF_ANC BIT(15)
667 static inline bool bpf_needs_clear_a(const struct sock_filter *first)
669 switch (first->code) {
670 case BPF_RET | BPF_K:
671 case BPF_LD | BPF_W | BPF_LEN:
674 case BPF_LD | BPF_W | BPF_ABS:
675 case BPF_LD | BPF_H | BPF_ABS:
676 case BPF_LD | BPF_B | BPF_ABS:
677 if (first->k == SKF_AD_OFF + SKF_AD_ALU_XOR_X)
686 static inline u16 bpf_anc_helper(const struct sock_filter *ftest)
688 BUG_ON(ftest->code & BPF_ANC);
690 switch (ftest->code) {
691 case BPF_LD | BPF_W | BPF_ABS:
692 case BPF_LD | BPF_H | BPF_ABS:
693 case BPF_LD | BPF_B | BPF_ABS:
694 #define BPF_ANCILLARY(CODE) case SKF_AD_OFF + SKF_AD_##CODE: \
695 return BPF_ANC | SKF_AD_##CODE
697 BPF_ANCILLARY(PROTOCOL);
698 BPF_ANCILLARY(PKTTYPE);
699 BPF_ANCILLARY(IFINDEX);
700 BPF_ANCILLARY(NLATTR);
701 BPF_ANCILLARY(NLATTR_NEST);
703 BPF_ANCILLARY(QUEUE);
704 BPF_ANCILLARY(HATYPE);
705 BPF_ANCILLARY(RXHASH);
707 BPF_ANCILLARY(ALU_XOR_X);
708 BPF_ANCILLARY(VLAN_TAG);
709 BPF_ANCILLARY(VLAN_TAG_PRESENT);
710 BPF_ANCILLARY(PAY_OFFSET);
711 BPF_ANCILLARY(RANDOM);
712 BPF_ANCILLARY(VLAN_TPID);
720 void *bpf_internal_load_pointer_neg_helper(const struct sk_buff *skb,
721 int k, unsigned int size);
723 static inline void *bpf_load_pointer(const struct sk_buff *skb, int k,
724 unsigned int size, void *buffer)
727 return skb_header_pointer(skb, k, size, buffer);
729 return bpf_internal_load_pointer_neg_helper(skb, k, size);
732 static inline int bpf_tell_extensions(void)
737 #endif /* __LINUX_FILTER_H__ */