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 <net/sch_generic.h>
18 #include <asm/cacheflush.h>
20 #include <uapi/linux/filter.h>
21 #include <uapi/linux/bpf.h>
28 /* ArgX, context and stack frame pointer register positions. Note,
29 * Arg1, Arg2, Arg3, etc are used as argument mappings of function
30 * calls in BPF_CALL instruction.
32 #define BPF_REG_ARG1 BPF_REG_1
33 #define BPF_REG_ARG2 BPF_REG_2
34 #define BPF_REG_ARG3 BPF_REG_3
35 #define BPF_REG_ARG4 BPF_REG_4
36 #define BPF_REG_ARG5 BPF_REG_5
37 #define BPF_REG_CTX BPF_REG_6
38 #define BPF_REG_FP BPF_REG_10
40 /* Additional register mappings for converted user programs. */
41 #define BPF_REG_A BPF_REG_0
42 #define BPF_REG_X BPF_REG_7
43 #define BPF_REG_TMP BPF_REG_8
45 /* BPF program can access up to 512 bytes of stack space. */
46 #define MAX_BPF_STACK 512
48 /* Helper macros for filter block array initializers. */
50 /* ALU ops on registers, bpf_add|sub|...: dst_reg += src_reg */
52 #define BPF_ALU64_REG(OP, DST, SRC) \
53 ((struct bpf_insn) { \
54 .code = BPF_ALU64 | BPF_OP(OP) | BPF_X, \
60 #define BPF_ALU32_REG(OP, DST, SRC) \
61 ((struct bpf_insn) { \
62 .code = BPF_ALU | BPF_OP(OP) | BPF_X, \
68 /* ALU ops on immediates, bpf_add|sub|...: dst_reg += imm32 */
70 #define BPF_ALU64_IMM(OP, DST, IMM) \
71 ((struct bpf_insn) { \
72 .code = BPF_ALU64 | BPF_OP(OP) | BPF_K, \
78 #define BPF_ALU32_IMM(OP, DST, IMM) \
79 ((struct bpf_insn) { \
80 .code = BPF_ALU | BPF_OP(OP) | BPF_K, \
86 /* Endianess conversion, cpu_to_{l,b}e(), {l,b}e_to_cpu() */
88 #define BPF_ENDIAN(TYPE, DST, LEN) \
89 ((struct bpf_insn) { \
90 .code = BPF_ALU | BPF_END | BPF_SRC(TYPE), \
96 /* Short form of mov, dst_reg = src_reg */
98 #define BPF_MOV64_REG(DST, SRC) \
99 ((struct bpf_insn) { \
100 .code = BPF_ALU64 | BPF_MOV | BPF_X, \
106 #define BPF_MOV32_REG(DST, SRC) \
107 ((struct bpf_insn) { \
108 .code = BPF_ALU | BPF_MOV | BPF_X, \
114 /* Short form of mov, dst_reg = imm32 */
116 #define BPF_MOV64_IMM(DST, IMM) \
117 ((struct bpf_insn) { \
118 .code = BPF_ALU64 | BPF_MOV | BPF_K, \
124 #define BPF_MOV32_IMM(DST, IMM) \
125 ((struct bpf_insn) { \
126 .code = BPF_ALU | BPF_MOV | BPF_K, \
132 /* BPF_LD_IMM64 macro encodes single 'load 64-bit immediate' insn */
133 #define BPF_LD_IMM64(DST, IMM) \
134 BPF_LD_IMM64_RAW(DST, 0, IMM)
136 #define BPF_LD_IMM64_RAW(DST, SRC, IMM) \
137 ((struct bpf_insn) { \
138 .code = BPF_LD | BPF_DW | BPF_IMM, \
142 .imm = (__u32) (IMM) }), \
143 ((struct bpf_insn) { \
144 .code = 0, /* zero is reserved opcode */ \
148 .imm = ((__u64) (IMM)) >> 32 })
150 /* pseudo BPF_LD_IMM64 insn used to refer to process-local map_fd */
151 #define BPF_LD_MAP_FD(DST, MAP_FD) \
152 BPF_LD_IMM64_RAW(DST, BPF_PSEUDO_MAP_FD, MAP_FD)
154 /* Short form of mov based on type, BPF_X: dst_reg = src_reg, BPF_K: dst_reg = imm32 */
156 #define BPF_MOV64_RAW(TYPE, DST, SRC, IMM) \
157 ((struct bpf_insn) { \
158 .code = BPF_ALU64 | BPF_MOV | BPF_SRC(TYPE), \
164 #define BPF_MOV32_RAW(TYPE, DST, SRC, IMM) \
165 ((struct bpf_insn) { \
166 .code = BPF_ALU | BPF_MOV | BPF_SRC(TYPE), \
172 /* Direct packet access, R0 = *(uint *) (skb->data + imm32) */
174 #define BPF_LD_ABS(SIZE, IMM) \
175 ((struct bpf_insn) { \
176 .code = BPF_LD | BPF_SIZE(SIZE) | BPF_ABS, \
182 /* Indirect packet access, R0 = *(uint *) (skb->data + src_reg + imm32) */
184 #define BPF_LD_IND(SIZE, SRC, IMM) \
185 ((struct bpf_insn) { \
186 .code = BPF_LD | BPF_SIZE(SIZE) | BPF_IND, \
192 /* Memory load, dst_reg = *(uint *) (src_reg + off16) */
194 #define BPF_LDX_MEM(SIZE, DST, SRC, OFF) \
195 ((struct bpf_insn) { \
196 .code = BPF_LDX | BPF_SIZE(SIZE) | BPF_MEM, \
202 /* Memory store, *(uint *) (dst_reg + off16) = src_reg */
204 #define BPF_STX_MEM(SIZE, DST, SRC, OFF) \
205 ((struct bpf_insn) { \
206 .code = BPF_STX | BPF_SIZE(SIZE) | BPF_MEM, \
212 /* Atomic memory add, *(uint *)(dst_reg + off16) += src_reg */
214 #define BPF_STX_XADD(SIZE, DST, SRC, OFF) \
215 ((struct bpf_insn) { \
216 .code = BPF_STX | BPF_SIZE(SIZE) | BPF_XADD, \
222 /* Memory store, *(uint *) (dst_reg + off16) = imm32 */
224 #define BPF_ST_MEM(SIZE, DST, OFF, IMM) \
225 ((struct bpf_insn) { \
226 .code = BPF_ST | BPF_SIZE(SIZE) | BPF_MEM, \
232 /* Conditional jumps against registers, if (dst_reg 'op' src_reg) goto pc + off16 */
234 #define BPF_JMP_REG(OP, DST, SRC, OFF) \
235 ((struct bpf_insn) { \
236 .code = BPF_JMP | BPF_OP(OP) | BPF_X, \
242 /* Conditional jumps against immediates, if (dst_reg 'op' imm32) goto pc + off16 */
244 #define BPF_JMP_IMM(OP, DST, IMM, OFF) \
245 ((struct bpf_insn) { \
246 .code = BPF_JMP | BPF_OP(OP) | BPF_K, \
254 #define BPF_EMIT_CALL(FUNC) \
255 ((struct bpf_insn) { \
256 .code = BPF_JMP | BPF_CALL, \
260 .imm = ((FUNC) - __bpf_call_base) })
262 /* Raw code statement block */
264 #define BPF_RAW_INSN(CODE, DST, SRC, OFF, IMM) \
265 ((struct bpf_insn) { \
274 #define BPF_EXIT_INSN() \
275 ((struct bpf_insn) { \
276 .code = BPF_JMP | BPF_EXIT, \
282 /* Internal classic blocks for direct assignment */
284 #define __BPF_STMT(CODE, K) \
285 ((struct sock_filter) BPF_STMT(CODE, K))
287 #define __BPF_JUMP(CODE, K, JT, JF) \
288 ((struct sock_filter) BPF_JUMP(CODE, K, JT, JF))
290 #define bytes_to_bpf_size(bytes) \
292 int bpf_size = -EINVAL; \
294 if (bytes == sizeof(u8)) \
296 else if (bytes == sizeof(u16)) \
298 else if (bytes == sizeof(u32)) \
300 else if (bytes == sizeof(u64)) \
307 /* A struct sock_filter is architecture independent. */
308 struct compat_sock_fprog
{
310 compat_uptr_t filter
; /* struct sock_filter * */
314 struct sock_fprog_kern
{
316 struct sock_filter
*filter
;
319 struct bpf_binary_header
{
325 u16 pages
; /* Number of allocated pages */
326 kmemcheck_bitfield_begin(meta
);
327 u16 jited
:1, /* Is our filter JIT'ed? */
328 gpl_compatible
:1, /* Is filter GPL compatible? */
329 cb_access
:1, /* Is control block accessed? */
330 dst_needed
:1; /* Do we need dst entry? */
331 kmemcheck_bitfield_end(meta
);
332 u32 len
; /* Number of filter blocks */
333 enum bpf_prog_type type
; /* Type of BPF program */
334 struct bpf_prog_aux
*aux
; /* Auxiliary fields */
335 struct sock_fprog_kern
*orig_prog
; /* Original BPF program */
336 unsigned int (*bpf_func
)(const struct sk_buff
*skb
,
337 const struct bpf_insn
*filter
);
338 /* Instructions for interpreter */
340 struct sock_filter insns
[0];
341 struct bpf_insn insnsi
[0];
348 struct bpf_prog
*prog
;
351 #define BPF_PROG_RUN(filter, ctx) (*filter->bpf_func)(ctx, filter->insnsi)
353 #define BPF_SKB_CB_LEN QDISC_CB_PRIV_LEN
355 struct bpf_skb_data_end
{
356 struct qdisc_skb_cb qdisc_cb
;
360 /* compute the linear packet data range [data, data_end) which
361 * will be accessed by cls_bpf and act_bpf programs
363 static inline void bpf_compute_data_end(struct sk_buff
*skb
)
365 struct bpf_skb_data_end
*cb
= (struct bpf_skb_data_end
*)skb
->cb
;
367 BUILD_BUG_ON(sizeof(*cb
) > FIELD_SIZEOF(struct sk_buff
, cb
));
368 cb
->data_end
= skb
->data
+ skb_headlen(skb
);
371 static inline u8
*bpf_skb_cb(struct sk_buff
*skb
)
373 /* eBPF programs may read/write skb->cb[] area to transfer meta
374 * data between tail calls. Since this also needs to work with
375 * tc, that scratch memory is mapped to qdisc_skb_cb's data area.
377 * In some socket filter cases, the cb unfortunately needs to be
378 * saved/restored so that protocol specific skb->cb[] data won't
379 * be lost. In any case, due to unpriviledged eBPF programs
380 * attached to sockets, we need to clear the bpf_skb_cb() area
381 * to not leak previous contents to user space.
383 BUILD_BUG_ON(FIELD_SIZEOF(struct __sk_buff
, cb
) != BPF_SKB_CB_LEN
);
384 BUILD_BUG_ON(FIELD_SIZEOF(struct __sk_buff
, cb
) !=
385 FIELD_SIZEOF(struct qdisc_skb_cb
, data
));
387 return qdisc_skb_cb(skb
)->data
;
390 static inline u32
bpf_prog_run_save_cb(const struct bpf_prog
*prog
,
393 u8
*cb_data
= bpf_skb_cb(skb
);
394 u8 cb_saved
[BPF_SKB_CB_LEN
];
397 if (unlikely(prog
->cb_access
)) {
398 memcpy(cb_saved
, cb_data
, sizeof(cb_saved
));
399 memset(cb_data
, 0, sizeof(cb_saved
));
402 res
= BPF_PROG_RUN(prog
, skb
);
404 if (unlikely(prog
->cb_access
))
405 memcpy(cb_data
, cb_saved
, sizeof(cb_saved
));
410 static inline u32
bpf_prog_run_clear_cb(const struct bpf_prog
*prog
,
413 u8
*cb_data
= bpf_skb_cb(skb
);
415 if (unlikely(prog
->cb_access
))
416 memset(cb_data
, 0, BPF_SKB_CB_LEN
);
418 return BPF_PROG_RUN(prog
, skb
);
421 static inline unsigned int bpf_prog_size(unsigned int proglen
)
423 return max(sizeof(struct bpf_prog
),
424 offsetof(struct bpf_prog
, insns
[proglen
]));
427 static inline bool bpf_prog_was_classic(const struct bpf_prog
*prog
)
429 /* When classic BPF programs have been loaded and the arch
430 * does not have a classic BPF JIT (anymore), they have been
431 * converted via bpf_migrate_filter() to eBPF and thus always
432 * have an unspec program type.
434 return prog
->type
== BPF_PROG_TYPE_UNSPEC
;
437 #define bpf_classic_proglen(fprog) (fprog->len * sizeof(fprog->filter[0]))
439 #ifdef CONFIG_DEBUG_SET_MODULE_RONX
440 static inline void bpf_prog_lock_ro(struct bpf_prog
*fp
)
442 set_memory_ro((unsigned long)fp
, fp
->pages
);
445 static inline void bpf_prog_unlock_ro(struct bpf_prog
*fp
)
447 set_memory_rw((unsigned long)fp
, fp
->pages
);
450 static inline void bpf_prog_lock_ro(struct bpf_prog
*fp
)
454 static inline void bpf_prog_unlock_ro(struct bpf_prog
*fp
)
457 #endif /* CONFIG_DEBUG_SET_MODULE_RONX */
459 int sk_filter(struct sock
*sk
, struct sk_buff
*skb
);
461 int bpf_prog_select_runtime(struct bpf_prog
*fp
);
462 void bpf_prog_free(struct bpf_prog
*fp
);
464 struct bpf_prog
*bpf_prog_alloc(unsigned int size
, gfp_t gfp_extra_flags
);
465 struct bpf_prog
*bpf_prog_realloc(struct bpf_prog
*fp_old
, unsigned int size
,
466 gfp_t gfp_extra_flags
);
467 void __bpf_prog_free(struct bpf_prog
*fp
);
469 static inline void bpf_prog_unlock_free(struct bpf_prog
*fp
)
471 bpf_prog_unlock_ro(fp
);
475 typedef int (*bpf_aux_classic_check_t
)(struct sock_filter
*filter
,
478 int bpf_prog_create(struct bpf_prog
**pfp
, struct sock_fprog_kern
*fprog
);
479 int bpf_prog_create_from_user(struct bpf_prog
**pfp
, struct sock_fprog
*fprog
,
480 bpf_aux_classic_check_t trans
, bool save_orig
);
481 void bpf_prog_destroy(struct bpf_prog
*fp
);
483 int sk_attach_filter(struct sock_fprog
*fprog
, struct sock
*sk
);
484 int sk_attach_bpf(u32 ufd
, struct sock
*sk
);
485 int sk_reuseport_attach_filter(struct sock_fprog
*fprog
, struct sock
*sk
);
486 int sk_reuseport_attach_bpf(u32 ufd
, struct sock
*sk
);
487 int sk_detach_filter(struct sock
*sk
);
488 int sk_get_filter(struct sock
*sk
, struct sock_filter __user
*filter
,
491 bool sk_filter_charge(struct sock
*sk
, struct sk_filter
*fp
);
492 void sk_filter_uncharge(struct sock
*sk
, struct sk_filter
*fp
);
494 u64
__bpf_call_base(u64 r1
, u64 r2
, u64 r3
, u64 r4
, u64 r5
);
495 void bpf_int_jit_compile(struct bpf_prog
*fp
);
496 bool bpf_helper_changes_skb_data(void *func
);
498 struct bpf_prog
*bpf_patch_insn_single(struct bpf_prog
*prog
, u32 off
,
499 const struct bpf_insn
*patch
, u32 len
);
501 #ifdef CONFIG_BPF_JIT
502 extern int bpf_jit_enable
;
504 typedef void (*bpf_jit_fill_hole_t
)(void *area
, unsigned int size
);
506 struct bpf_binary_header
*
507 bpf_jit_binary_alloc(unsigned int proglen
, u8
**image_ptr
,
508 unsigned int alignment
,
509 bpf_jit_fill_hole_t bpf_fill_ill_insns
);
510 void bpf_jit_binary_free(struct bpf_binary_header
*hdr
);
512 void bpf_jit_compile(struct bpf_prog
*fp
);
513 void bpf_jit_free(struct bpf_prog
*fp
);
515 static inline void bpf_jit_dump(unsigned int flen
, unsigned int proglen
,
516 u32 pass
, void *image
)
518 pr_err("flen=%u proglen=%u pass=%u image=%pK from=%s pid=%d\n", flen
,
519 proglen
, pass
, image
, current
->comm
, task_pid_nr(current
));
522 print_hex_dump(KERN_ERR
, "JIT code: ", DUMP_PREFIX_OFFSET
,
523 16, 1, image
, proglen
, false);
526 static inline void bpf_jit_compile(struct bpf_prog
*fp
)
530 static inline void bpf_jit_free(struct bpf_prog
*fp
)
532 bpf_prog_unlock_free(fp
);
534 #endif /* CONFIG_BPF_JIT */
536 #define BPF_ANC BIT(15)
538 static inline bool bpf_needs_clear_a(const struct sock_filter
*first
)
540 switch (first
->code
) {
541 case BPF_RET
| BPF_K
:
542 case BPF_LD
| BPF_W
| BPF_LEN
:
545 case BPF_LD
| BPF_W
| BPF_ABS
:
546 case BPF_LD
| BPF_H
| BPF_ABS
:
547 case BPF_LD
| BPF_B
| BPF_ABS
:
548 if (first
->k
== SKF_AD_OFF
+ SKF_AD_ALU_XOR_X
)
557 static inline u16
bpf_anc_helper(const struct sock_filter
*ftest
)
559 BUG_ON(ftest
->code
& BPF_ANC
);
561 switch (ftest
->code
) {
562 case BPF_LD
| BPF_W
| BPF_ABS
:
563 case BPF_LD
| BPF_H
| BPF_ABS
:
564 case BPF_LD
| BPF_B
| BPF_ABS
:
565 #define BPF_ANCILLARY(CODE) case SKF_AD_OFF + SKF_AD_##CODE: \
566 return BPF_ANC | SKF_AD_##CODE
568 BPF_ANCILLARY(PROTOCOL
);
569 BPF_ANCILLARY(PKTTYPE
);
570 BPF_ANCILLARY(IFINDEX
);
571 BPF_ANCILLARY(NLATTR
);
572 BPF_ANCILLARY(NLATTR_NEST
);
574 BPF_ANCILLARY(QUEUE
);
575 BPF_ANCILLARY(HATYPE
);
576 BPF_ANCILLARY(RXHASH
);
578 BPF_ANCILLARY(ALU_XOR_X
);
579 BPF_ANCILLARY(VLAN_TAG
);
580 BPF_ANCILLARY(VLAN_TAG_PRESENT
);
581 BPF_ANCILLARY(PAY_OFFSET
);
582 BPF_ANCILLARY(RANDOM
);
583 BPF_ANCILLARY(VLAN_TPID
);
591 void *bpf_internal_load_pointer_neg_helper(const struct sk_buff
*skb
,
592 int k
, unsigned int size
);
594 static inline void *bpf_load_pointer(const struct sk_buff
*skb
, int k
,
595 unsigned int size
, void *buffer
)
598 return skb_header_pointer(skb
, k
, size
, buffer
);
600 return bpf_internal_load_pointer_neg_helper(skb
, k
, size
);
603 static inline int bpf_tell_extensions(void)
608 #endif /* __LINUX_FILTER_H__ */