2 * Linux Socket Filter - Kernel level socket filtering
4 * Based on the design of the Berkeley Packet Filter. The new
5 * internal format has been designed by PLUMgrid:
7 * Copyright (c) 2011 - 2014 PLUMgrid, http://plumgrid.com
11 * Jay Schulist <jschlst@samba.org>
12 * Alexei Starovoitov <ast@plumgrid.com>
13 * Daniel Borkmann <dborkman@redhat.com>
15 * This program is free software; you can redistribute it and/or
16 * modify it under the terms of the GNU General Public License
17 * as published by the Free Software Foundation; either version
18 * 2 of the License, or (at your option) any later version.
20 * Andi Kleen - Fix a few bad bugs and races.
21 * Kris Katterjohn - Added many additional checks in bpf_check_classic()
24 #include <linux/module.h>
25 #include <linux/types.h>
27 #include <linux/fcntl.h>
28 #include <linux/socket.h>
30 #include <linux/inet.h>
31 #include <linux/netdevice.h>
32 #include <linux/if_packet.h>
33 #include <linux/gfp.h>
35 #include <net/protocol.h>
36 #include <net/netlink.h>
37 #include <linux/skbuff.h>
39 #include <net/flow_dissector.h>
40 #include <linux/errno.h>
41 #include <linux/timer.h>
42 #include <asm/uaccess.h>
43 #include <asm/unaligned.h>
44 #include <linux/filter.h>
45 #include <linux/ratelimit.h>
46 #include <linux/seccomp.h>
47 #include <linux/if_vlan.h>
48 #include <linux/bpf.h>
51 * sk_filter - run a packet through a socket filter
52 * @sk: sock associated with &sk_buff
53 * @skb: buffer to filter
55 * Run the filter code and then cut skb->data to correct size returned by
56 * SK_RUN_FILTER. If pkt_len is 0 we toss packet. If skb->len is smaller
57 * than pkt_len we keep whole skb->data. This is the socket level
58 * wrapper to SK_RUN_FILTER. It returns 0 if the packet should
59 * be accepted or -EPERM if the packet should be tossed.
62 int sk_filter(struct sock
*sk
, struct sk_buff
*skb
)
65 struct sk_filter
*filter
;
68 * If the skb was allocated from pfmemalloc reserves, only
69 * allow SOCK_MEMALLOC sockets to use it as this socket is
72 if (skb_pfmemalloc(skb
) && !sock_flag(sk
, SOCK_MEMALLOC
))
75 err
= security_sock_rcv_skb(sk
, skb
);
80 filter
= rcu_dereference(sk
->sk_filter
);
82 unsigned int pkt_len
= SK_RUN_FILTER(filter
, skb
);
84 err
= pkt_len
? pskb_trim(skb
, pkt_len
) : -EPERM
;
90 EXPORT_SYMBOL(sk_filter
);
92 static u64
__skb_get_pay_offset(u64 ctx
, u64 a
, u64 x
, u64 r4
, u64 r5
)
94 return skb_get_poff((struct sk_buff
*)(unsigned long) ctx
);
97 static u64
__skb_get_nlattr(u64 ctx
, u64 a
, u64 x
, u64 r4
, u64 r5
)
99 struct sk_buff
*skb
= (struct sk_buff
*)(unsigned long) ctx
;
102 if (skb_is_nonlinear(skb
))
105 if (skb
->len
< sizeof(struct nlattr
))
108 if (a
> skb
->len
- sizeof(struct nlattr
))
111 nla
= nla_find((struct nlattr
*) &skb
->data
[a
], skb
->len
- a
, x
);
113 return (void *) nla
- (void *) skb
->data
;
118 static u64
__skb_get_nlattr_nest(u64 ctx
, u64 a
, u64 x
, u64 r4
, u64 r5
)
120 struct sk_buff
*skb
= (struct sk_buff
*)(unsigned long) ctx
;
123 if (skb_is_nonlinear(skb
))
126 if (skb
->len
< sizeof(struct nlattr
))
129 if (a
> skb
->len
- sizeof(struct nlattr
))
132 nla
= (struct nlattr
*) &skb
->data
[a
];
133 if (nla
->nla_len
> skb
->len
- a
)
136 nla
= nla_find_nested(nla
, x
);
138 return (void *) nla
- (void *) skb
->data
;
143 static u64
__get_raw_cpu_id(u64 ctx
, u64 a
, u64 x
, u64 r4
, u64 r5
)
145 return raw_smp_processor_id();
148 /* note that this only generates 32-bit random numbers */
149 static u64
__get_random_u32(u64 ctx
, u64 a
, u64 x
, u64 r4
, u64 r5
)
151 return prandom_u32();
154 static u32
convert_skb_access(int skb_field
, int dst_reg
, int src_reg
,
155 struct bpf_insn
*insn_buf
)
157 struct bpf_insn
*insn
= insn_buf
;
161 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff
, mark
) != 4);
163 *insn
++ = BPF_LDX_MEM(BPF_W
, dst_reg
, src_reg
,
164 offsetof(struct sk_buff
, mark
));
168 *insn
++ = BPF_LDX_MEM(BPF_B
, dst_reg
, src_reg
, PKT_TYPE_OFFSET());
169 *insn
++ = BPF_ALU32_IMM(BPF_AND
, dst_reg
, PKT_TYPE_MAX
);
170 #ifdef __BIG_ENDIAN_BITFIELD
171 *insn
++ = BPF_ALU32_IMM(BPF_RSH
, dst_reg
, 5);
176 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff
, queue_mapping
) != 2);
178 *insn
++ = BPF_LDX_MEM(BPF_H
, dst_reg
, src_reg
,
179 offsetof(struct sk_buff
, queue_mapping
));
182 case SKF_AD_VLAN_TAG
:
183 case SKF_AD_VLAN_TAG_PRESENT
:
184 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff
, vlan_tci
) != 2);
185 BUILD_BUG_ON(VLAN_TAG_PRESENT
!= 0x1000);
187 /* dst_reg = *(u16 *) (src_reg + offsetof(vlan_tci)) */
188 *insn
++ = BPF_LDX_MEM(BPF_H
, dst_reg
, src_reg
,
189 offsetof(struct sk_buff
, vlan_tci
));
190 if (skb_field
== SKF_AD_VLAN_TAG
) {
191 *insn
++ = BPF_ALU32_IMM(BPF_AND
, dst_reg
,
195 *insn
++ = BPF_ALU32_IMM(BPF_RSH
, dst_reg
, 12);
197 *insn
++ = BPF_ALU32_IMM(BPF_AND
, dst_reg
, 1);
202 return insn
- insn_buf
;
205 static bool convert_bpf_extensions(struct sock_filter
*fp
,
206 struct bpf_insn
**insnp
)
208 struct bpf_insn
*insn
= *insnp
;
212 case SKF_AD_OFF
+ SKF_AD_PROTOCOL
:
213 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff
, protocol
) != 2);
215 /* A = *(u16 *) (CTX + offsetof(protocol)) */
216 *insn
++ = BPF_LDX_MEM(BPF_H
, BPF_REG_A
, BPF_REG_CTX
,
217 offsetof(struct sk_buff
, protocol
));
218 /* A = ntohs(A) [emitting a nop or swap16] */
219 *insn
= BPF_ENDIAN(BPF_FROM_BE
, BPF_REG_A
, 16);
222 case SKF_AD_OFF
+ SKF_AD_PKTTYPE
:
223 cnt
= convert_skb_access(SKF_AD_PKTTYPE
, BPF_REG_A
, BPF_REG_CTX
, insn
);
227 case SKF_AD_OFF
+ SKF_AD_IFINDEX
:
228 case SKF_AD_OFF
+ SKF_AD_HATYPE
:
229 BUILD_BUG_ON(FIELD_SIZEOF(struct net_device
, ifindex
) != 4);
230 BUILD_BUG_ON(FIELD_SIZEOF(struct net_device
, type
) != 2);
231 BUILD_BUG_ON(bytes_to_bpf_size(FIELD_SIZEOF(struct sk_buff
, dev
)) < 0);
233 *insn
++ = BPF_LDX_MEM(bytes_to_bpf_size(FIELD_SIZEOF(struct sk_buff
, dev
)),
234 BPF_REG_TMP
, BPF_REG_CTX
,
235 offsetof(struct sk_buff
, dev
));
236 /* if (tmp != 0) goto pc + 1 */
237 *insn
++ = BPF_JMP_IMM(BPF_JNE
, BPF_REG_TMP
, 0, 1);
238 *insn
++ = BPF_EXIT_INSN();
239 if (fp
->k
== SKF_AD_OFF
+ SKF_AD_IFINDEX
)
240 *insn
= BPF_LDX_MEM(BPF_W
, BPF_REG_A
, BPF_REG_TMP
,
241 offsetof(struct net_device
, ifindex
));
243 *insn
= BPF_LDX_MEM(BPF_H
, BPF_REG_A
, BPF_REG_TMP
,
244 offsetof(struct net_device
, type
));
247 case SKF_AD_OFF
+ SKF_AD_MARK
:
248 cnt
= convert_skb_access(SKF_AD_MARK
, BPF_REG_A
, BPF_REG_CTX
, insn
);
252 case SKF_AD_OFF
+ SKF_AD_RXHASH
:
253 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff
, hash
) != 4);
255 *insn
= BPF_LDX_MEM(BPF_W
, BPF_REG_A
, BPF_REG_CTX
,
256 offsetof(struct sk_buff
, hash
));
259 case SKF_AD_OFF
+ SKF_AD_QUEUE
:
260 cnt
= convert_skb_access(SKF_AD_QUEUE
, BPF_REG_A
, BPF_REG_CTX
, insn
);
264 case SKF_AD_OFF
+ SKF_AD_VLAN_TAG
:
265 cnt
= convert_skb_access(SKF_AD_VLAN_TAG
,
266 BPF_REG_A
, BPF_REG_CTX
, insn
);
270 case SKF_AD_OFF
+ SKF_AD_VLAN_TAG_PRESENT
:
271 cnt
= convert_skb_access(SKF_AD_VLAN_TAG_PRESENT
,
272 BPF_REG_A
, BPF_REG_CTX
, insn
);
276 case SKF_AD_OFF
+ SKF_AD_VLAN_TPID
:
277 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff
, vlan_proto
) != 2);
279 /* A = *(u16 *) (CTX + offsetof(vlan_proto)) */
280 *insn
++ = BPF_LDX_MEM(BPF_H
, BPF_REG_A
, BPF_REG_CTX
,
281 offsetof(struct sk_buff
, vlan_proto
));
282 /* A = ntohs(A) [emitting a nop or swap16] */
283 *insn
= BPF_ENDIAN(BPF_FROM_BE
, BPF_REG_A
, 16);
286 case SKF_AD_OFF
+ SKF_AD_PAY_OFFSET
:
287 case SKF_AD_OFF
+ SKF_AD_NLATTR
:
288 case SKF_AD_OFF
+ SKF_AD_NLATTR_NEST
:
289 case SKF_AD_OFF
+ SKF_AD_CPU
:
290 case SKF_AD_OFF
+ SKF_AD_RANDOM
:
292 *insn
++ = BPF_MOV64_REG(BPF_REG_ARG1
, BPF_REG_CTX
);
294 *insn
++ = BPF_MOV64_REG(BPF_REG_ARG2
, BPF_REG_A
);
296 *insn
++ = BPF_MOV64_REG(BPF_REG_ARG3
, BPF_REG_X
);
297 /* Emit call(arg1=CTX, arg2=A, arg3=X) */
299 case SKF_AD_OFF
+ SKF_AD_PAY_OFFSET
:
300 *insn
= BPF_EMIT_CALL(__skb_get_pay_offset
);
302 case SKF_AD_OFF
+ SKF_AD_NLATTR
:
303 *insn
= BPF_EMIT_CALL(__skb_get_nlattr
);
305 case SKF_AD_OFF
+ SKF_AD_NLATTR_NEST
:
306 *insn
= BPF_EMIT_CALL(__skb_get_nlattr_nest
);
308 case SKF_AD_OFF
+ SKF_AD_CPU
:
309 *insn
= BPF_EMIT_CALL(__get_raw_cpu_id
);
311 case SKF_AD_OFF
+ SKF_AD_RANDOM
:
312 *insn
= BPF_EMIT_CALL(__get_random_u32
);
317 case SKF_AD_OFF
+ SKF_AD_ALU_XOR_X
:
319 *insn
= BPF_ALU32_REG(BPF_XOR
, BPF_REG_A
, BPF_REG_X
);
323 /* This is just a dummy call to avoid letting the compiler
324 * evict __bpf_call_base() as an optimization. Placed here
325 * where no-one bothers.
327 BUG_ON(__bpf_call_base(0, 0, 0, 0, 0) != 0);
336 * bpf_convert_filter - convert filter program
337 * @prog: the user passed filter program
338 * @len: the length of the user passed filter program
339 * @new_prog: buffer where converted program will be stored
340 * @new_len: pointer to store length of converted program
342 * Remap 'sock_filter' style BPF instruction set to 'sock_filter_ext' style.
343 * Conversion workflow:
345 * 1) First pass for calculating the new program length:
346 * bpf_convert_filter(old_prog, old_len, NULL, &new_len)
348 * 2) 2nd pass to remap in two passes: 1st pass finds new
349 * jump offsets, 2nd pass remapping:
350 * new_prog = kmalloc(sizeof(struct bpf_insn) * new_len);
351 * bpf_convert_filter(old_prog, old_len, new_prog, &new_len);
353 * User BPF's register A is mapped to our BPF register 6, user BPF
354 * register X is mapped to BPF register 7; frame pointer is always
355 * register 10; Context 'void *ctx' is stored in register 1, that is,
356 * for socket filters: ctx == 'struct sk_buff *', for seccomp:
357 * ctx == 'struct seccomp_data *'.
359 static int bpf_convert_filter(struct sock_filter
*prog
, int len
,
360 struct bpf_insn
*new_prog
, int *new_len
)
362 int new_flen
= 0, pass
= 0, target
, i
;
363 struct bpf_insn
*new_insn
;
364 struct sock_filter
*fp
;
368 BUILD_BUG_ON(BPF_MEMWORDS
* sizeof(u32
) > MAX_BPF_STACK
);
369 BUILD_BUG_ON(BPF_REG_FP
+ 1 != MAX_BPF_REG
);
371 if (len
<= 0 || len
> BPF_MAXINSNS
)
375 addrs
= kcalloc(len
, sizeof(*addrs
),
376 GFP_KERNEL
| __GFP_NOWARN
);
386 *new_insn
= BPF_MOV64_REG(BPF_REG_CTX
, BPF_REG_ARG1
);
389 for (i
= 0; i
< len
; fp
++, i
++) {
390 struct bpf_insn tmp_insns
[6] = { };
391 struct bpf_insn
*insn
= tmp_insns
;
394 addrs
[i
] = new_insn
- new_prog
;
397 /* All arithmetic insns and skb loads map as-is. */
398 case BPF_ALU
| BPF_ADD
| BPF_X
:
399 case BPF_ALU
| BPF_ADD
| BPF_K
:
400 case BPF_ALU
| BPF_SUB
| BPF_X
:
401 case BPF_ALU
| BPF_SUB
| BPF_K
:
402 case BPF_ALU
| BPF_AND
| BPF_X
:
403 case BPF_ALU
| BPF_AND
| BPF_K
:
404 case BPF_ALU
| BPF_OR
| BPF_X
:
405 case BPF_ALU
| BPF_OR
| BPF_K
:
406 case BPF_ALU
| BPF_LSH
| BPF_X
:
407 case BPF_ALU
| BPF_LSH
| BPF_K
:
408 case BPF_ALU
| BPF_RSH
| BPF_X
:
409 case BPF_ALU
| BPF_RSH
| BPF_K
:
410 case BPF_ALU
| BPF_XOR
| BPF_X
:
411 case BPF_ALU
| BPF_XOR
| BPF_K
:
412 case BPF_ALU
| BPF_MUL
| BPF_X
:
413 case BPF_ALU
| BPF_MUL
| BPF_K
:
414 case BPF_ALU
| BPF_DIV
| BPF_X
:
415 case BPF_ALU
| BPF_DIV
| BPF_K
:
416 case BPF_ALU
| BPF_MOD
| BPF_X
:
417 case BPF_ALU
| BPF_MOD
| BPF_K
:
418 case BPF_ALU
| BPF_NEG
:
419 case BPF_LD
| BPF_ABS
| BPF_W
:
420 case BPF_LD
| BPF_ABS
| BPF_H
:
421 case BPF_LD
| BPF_ABS
| BPF_B
:
422 case BPF_LD
| BPF_IND
| BPF_W
:
423 case BPF_LD
| BPF_IND
| BPF_H
:
424 case BPF_LD
| BPF_IND
| BPF_B
:
425 /* Check for overloaded BPF extension and
426 * directly convert it if found, otherwise
427 * just move on with mapping.
429 if (BPF_CLASS(fp
->code
) == BPF_LD
&&
430 BPF_MODE(fp
->code
) == BPF_ABS
&&
431 convert_bpf_extensions(fp
, &insn
))
434 *insn
= BPF_RAW_INSN(fp
->code
, BPF_REG_A
, BPF_REG_X
, 0, fp
->k
);
437 /* Jump transformation cannot use BPF block macros
438 * everywhere as offset calculation and target updates
439 * require a bit more work than the rest, i.e. jump
440 * opcodes map as-is, but offsets need adjustment.
443 #define BPF_EMIT_JMP \
445 if (target >= len || target < 0) \
447 insn->off = addrs ? addrs[target] - addrs[i] - 1 : 0; \
448 /* Adjust pc relative offset for 2nd or 3rd insn. */ \
449 insn->off -= insn - tmp_insns; \
452 case BPF_JMP
| BPF_JA
:
453 target
= i
+ fp
->k
+ 1;
454 insn
->code
= fp
->code
;
458 case BPF_JMP
| BPF_JEQ
| BPF_K
:
459 case BPF_JMP
| BPF_JEQ
| BPF_X
:
460 case BPF_JMP
| BPF_JSET
| BPF_K
:
461 case BPF_JMP
| BPF_JSET
| BPF_X
:
462 case BPF_JMP
| BPF_JGT
| BPF_K
:
463 case BPF_JMP
| BPF_JGT
| BPF_X
:
464 case BPF_JMP
| BPF_JGE
| BPF_K
:
465 case BPF_JMP
| BPF_JGE
| BPF_X
:
466 if (BPF_SRC(fp
->code
) == BPF_K
&& (int) fp
->k
< 0) {
467 /* BPF immediates are signed, zero extend
468 * immediate into tmp register and use it
471 *insn
++ = BPF_MOV32_IMM(BPF_REG_TMP
, fp
->k
);
473 insn
->dst_reg
= BPF_REG_A
;
474 insn
->src_reg
= BPF_REG_TMP
;
477 insn
->dst_reg
= BPF_REG_A
;
478 insn
->src_reg
= BPF_REG_X
;
480 bpf_src
= BPF_SRC(fp
->code
);
483 /* Common case where 'jump_false' is next insn. */
485 insn
->code
= BPF_JMP
| BPF_OP(fp
->code
) | bpf_src
;
486 target
= i
+ fp
->jt
+ 1;
491 /* Convert JEQ into JNE when 'jump_true' is next insn. */
492 if (fp
->jt
== 0 && BPF_OP(fp
->code
) == BPF_JEQ
) {
493 insn
->code
= BPF_JMP
| BPF_JNE
| bpf_src
;
494 target
= i
+ fp
->jf
+ 1;
499 /* Other jumps are mapped into two insns: Jxx and JA. */
500 target
= i
+ fp
->jt
+ 1;
501 insn
->code
= BPF_JMP
| BPF_OP(fp
->code
) | bpf_src
;
505 insn
->code
= BPF_JMP
| BPF_JA
;
506 target
= i
+ fp
->jf
+ 1;
510 /* ldxb 4 * ([14] & 0xf) is remaped into 6 insns. */
511 case BPF_LDX
| BPF_MSH
| BPF_B
:
513 *insn
++ = BPF_MOV64_REG(BPF_REG_TMP
, BPF_REG_A
);
514 /* A = BPF_R0 = *(u8 *) (skb->data + K) */
515 *insn
++ = BPF_LD_ABS(BPF_B
, fp
->k
);
517 *insn
++ = BPF_ALU32_IMM(BPF_AND
, BPF_REG_A
, 0xf);
519 *insn
++ = BPF_ALU32_IMM(BPF_LSH
, BPF_REG_A
, 2);
521 *insn
++ = BPF_MOV64_REG(BPF_REG_X
, BPF_REG_A
);
523 *insn
= BPF_MOV64_REG(BPF_REG_A
, BPF_REG_TMP
);
526 /* RET_K, RET_A are remaped into 2 insns. */
527 case BPF_RET
| BPF_A
:
528 case BPF_RET
| BPF_K
:
529 *insn
++ = BPF_MOV32_RAW(BPF_RVAL(fp
->code
) == BPF_K
?
530 BPF_K
: BPF_X
, BPF_REG_0
,
532 *insn
= BPF_EXIT_INSN();
535 /* Store to stack. */
538 *insn
= BPF_STX_MEM(BPF_W
, BPF_REG_FP
, BPF_CLASS(fp
->code
) ==
539 BPF_ST
? BPF_REG_A
: BPF_REG_X
,
540 -(BPF_MEMWORDS
- fp
->k
) * 4);
543 /* Load from stack. */
544 case BPF_LD
| BPF_MEM
:
545 case BPF_LDX
| BPF_MEM
:
546 *insn
= BPF_LDX_MEM(BPF_W
, BPF_CLASS(fp
->code
) == BPF_LD
?
547 BPF_REG_A
: BPF_REG_X
, BPF_REG_FP
,
548 -(BPF_MEMWORDS
- fp
->k
) * 4);
552 case BPF_LD
| BPF_IMM
:
553 case BPF_LDX
| BPF_IMM
:
554 *insn
= BPF_MOV32_IMM(BPF_CLASS(fp
->code
) == BPF_LD
?
555 BPF_REG_A
: BPF_REG_X
, fp
->k
);
559 case BPF_MISC
| BPF_TAX
:
560 *insn
= BPF_MOV64_REG(BPF_REG_X
, BPF_REG_A
);
564 case BPF_MISC
| BPF_TXA
:
565 *insn
= BPF_MOV64_REG(BPF_REG_A
, BPF_REG_X
);
568 /* A = skb->len or X = skb->len */
569 case BPF_LD
| BPF_W
| BPF_LEN
:
570 case BPF_LDX
| BPF_W
| BPF_LEN
:
571 *insn
= BPF_LDX_MEM(BPF_W
, BPF_CLASS(fp
->code
) == BPF_LD
?
572 BPF_REG_A
: BPF_REG_X
, BPF_REG_CTX
,
573 offsetof(struct sk_buff
, len
));
576 /* Access seccomp_data fields. */
577 case BPF_LDX
| BPF_ABS
| BPF_W
:
578 /* A = *(u32 *) (ctx + K) */
579 *insn
= BPF_LDX_MEM(BPF_W
, BPF_REG_A
, BPF_REG_CTX
, fp
->k
);
582 /* Unknown instruction. */
589 memcpy(new_insn
, tmp_insns
,
590 sizeof(*insn
) * (insn
- tmp_insns
));
591 new_insn
+= insn
- tmp_insns
;
595 /* Only calculating new length. */
596 *new_len
= new_insn
- new_prog
;
601 if (new_flen
!= new_insn
- new_prog
) {
602 new_flen
= new_insn
- new_prog
;
609 BUG_ON(*new_len
!= new_flen
);
618 * As we dont want to clear mem[] array for each packet going through
619 * __bpf_prog_run(), we check that filter loaded by user never try to read
620 * a cell if not previously written, and we check all branches to be sure
621 * a malicious user doesn't try to abuse us.
623 static int check_load_and_stores(const struct sock_filter
*filter
, int flen
)
625 u16
*masks
, memvalid
= 0; /* One bit per cell, 16 cells */
628 BUILD_BUG_ON(BPF_MEMWORDS
> 16);
630 masks
= kmalloc_array(flen
, sizeof(*masks
), GFP_KERNEL
);
634 memset(masks
, 0xff, flen
* sizeof(*masks
));
636 for (pc
= 0; pc
< flen
; pc
++) {
637 memvalid
&= masks
[pc
];
639 switch (filter
[pc
].code
) {
642 memvalid
|= (1 << filter
[pc
].k
);
644 case BPF_LD
| BPF_MEM
:
645 case BPF_LDX
| BPF_MEM
:
646 if (!(memvalid
& (1 << filter
[pc
].k
))) {
651 case BPF_JMP
| BPF_JA
:
652 /* A jump must set masks on target */
653 masks
[pc
+ 1 + filter
[pc
].k
] &= memvalid
;
656 case BPF_JMP
| BPF_JEQ
| BPF_K
:
657 case BPF_JMP
| BPF_JEQ
| BPF_X
:
658 case BPF_JMP
| BPF_JGE
| BPF_K
:
659 case BPF_JMP
| BPF_JGE
| BPF_X
:
660 case BPF_JMP
| BPF_JGT
| BPF_K
:
661 case BPF_JMP
| BPF_JGT
| BPF_X
:
662 case BPF_JMP
| BPF_JSET
| BPF_K
:
663 case BPF_JMP
| BPF_JSET
| BPF_X
:
664 /* A jump must set masks on targets */
665 masks
[pc
+ 1 + filter
[pc
].jt
] &= memvalid
;
666 masks
[pc
+ 1 + filter
[pc
].jf
] &= memvalid
;
676 static bool chk_code_allowed(u16 code_to_probe
)
678 static const bool codes
[] = {
679 /* 32 bit ALU operations */
680 [BPF_ALU
| BPF_ADD
| BPF_K
] = true,
681 [BPF_ALU
| BPF_ADD
| BPF_X
] = true,
682 [BPF_ALU
| BPF_SUB
| BPF_K
] = true,
683 [BPF_ALU
| BPF_SUB
| BPF_X
] = true,
684 [BPF_ALU
| BPF_MUL
| BPF_K
] = true,
685 [BPF_ALU
| BPF_MUL
| BPF_X
] = true,
686 [BPF_ALU
| BPF_DIV
| BPF_K
] = true,
687 [BPF_ALU
| BPF_DIV
| BPF_X
] = true,
688 [BPF_ALU
| BPF_MOD
| BPF_K
] = true,
689 [BPF_ALU
| BPF_MOD
| BPF_X
] = true,
690 [BPF_ALU
| BPF_AND
| BPF_K
] = true,
691 [BPF_ALU
| BPF_AND
| BPF_X
] = true,
692 [BPF_ALU
| BPF_OR
| BPF_K
] = true,
693 [BPF_ALU
| BPF_OR
| BPF_X
] = true,
694 [BPF_ALU
| BPF_XOR
| BPF_K
] = true,
695 [BPF_ALU
| BPF_XOR
| BPF_X
] = true,
696 [BPF_ALU
| BPF_LSH
| BPF_K
] = true,
697 [BPF_ALU
| BPF_LSH
| BPF_X
] = true,
698 [BPF_ALU
| BPF_RSH
| BPF_K
] = true,
699 [BPF_ALU
| BPF_RSH
| BPF_X
] = true,
700 [BPF_ALU
| BPF_NEG
] = true,
701 /* Load instructions */
702 [BPF_LD
| BPF_W
| BPF_ABS
] = true,
703 [BPF_LD
| BPF_H
| BPF_ABS
] = true,
704 [BPF_LD
| BPF_B
| BPF_ABS
] = true,
705 [BPF_LD
| BPF_W
| BPF_LEN
] = true,
706 [BPF_LD
| BPF_W
| BPF_IND
] = true,
707 [BPF_LD
| BPF_H
| BPF_IND
] = true,
708 [BPF_LD
| BPF_B
| BPF_IND
] = true,
709 [BPF_LD
| BPF_IMM
] = true,
710 [BPF_LD
| BPF_MEM
] = true,
711 [BPF_LDX
| BPF_W
| BPF_LEN
] = true,
712 [BPF_LDX
| BPF_B
| BPF_MSH
] = true,
713 [BPF_LDX
| BPF_IMM
] = true,
714 [BPF_LDX
| BPF_MEM
] = true,
715 /* Store instructions */
718 /* Misc instructions */
719 [BPF_MISC
| BPF_TAX
] = true,
720 [BPF_MISC
| BPF_TXA
] = true,
721 /* Return instructions */
722 [BPF_RET
| BPF_K
] = true,
723 [BPF_RET
| BPF_A
] = true,
724 /* Jump instructions */
725 [BPF_JMP
| BPF_JA
] = true,
726 [BPF_JMP
| BPF_JEQ
| BPF_K
] = true,
727 [BPF_JMP
| BPF_JEQ
| BPF_X
] = true,
728 [BPF_JMP
| BPF_JGE
| BPF_K
] = true,
729 [BPF_JMP
| BPF_JGE
| BPF_X
] = true,
730 [BPF_JMP
| BPF_JGT
| BPF_K
] = true,
731 [BPF_JMP
| BPF_JGT
| BPF_X
] = true,
732 [BPF_JMP
| BPF_JSET
| BPF_K
] = true,
733 [BPF_JMP
| BPF_JSET
| BPF_X
] = true,
736 if (code_to_probe
>= ARRAY_SIZE(codes
))
739 return codes
[code_to_probe
];
743 * bpf_check_classic - verify socket filter code
744 * @filter: filter to verify
745 * @flen: length of filter
747 * Check the user's filter code. If we let some ugly
748 * filter code slip through kaboom! The filter must contain
749 * no references or jumps that are out of range, no illegal
750 * instructions, and must end with a RET instruction.
752 * All jumps are forward as they are not signed.
754 * Returns 0 if the rule set is legal or -EINVAL if not.
756 static int bpf_check_classic(const struct sock_filter
*filter
,
762 if (flen
== 0 || flen
> BPF_MAXINSNS
)
765 /* Check the filter code now */
766 for (pc
= 0; pc
< flen
; pc
++) {
767 const struct sock_filter
*ftest
= &filter
[pc
];
769 /* May we actually operate on this code? */
770 if (!chk_code_allowed(ftest
->code
))
773 /* Some instructions need special checks */
774 switch (ftest
->code
) {
775 case BPF_ALU
| BPF_DIV
| BPF_K
:
776 case BPF_ALU
| BPF_MOD
| BPF_K
:
777 /* Check for division by zero */
781 case BPF_LD
| BPF_MEM
:
782 case BPF_LDX
| BPF_MEM
:
785 /* Check for invalid memory addresses */
786 if (ftest
->k
>= BPF_MEMWORDS
)
789 case BPF_JMP
| BPF_JA
:
790 /* Note, the large ftest->k might cause loops.
791 * Compare this with conditional jumps below,
792 * where offsets are limited. --ANK (981016)
794 if (ftest
->k
>= (unsigned int)(flen
- pc
- 1))
797 case BPF_JMP
| BPF_JEQ
| BPF_K
:
798 case BPF_JMP
| BPF_JEQ
| BPF_X
:
799 case BPF_JMP
| BPF_JGE
| BPF_K
:
800 case BPF_JMP
| BPF_JGE
| BPF_X
:
801 case BPF_JMP
| BPF_JGT
| BPF_K
:
802 case BPF_JMP
| BPF_JGT
| BPF_X
:
803 case BPF_JMP
| BPF_JSET
| BPF_K
:
804 case BPF_JMP
| BPF_JSET
| BPF_X
:
805 /* Both conditionals must be safe */
806 if (pc
+ ftest
->jt
+ 1 >= flen
||
807 pc
+ ftest
->jf
+ 1 >= flen
)
810 case BPF_LD
| BPF_W
| BPF_ABS
:
811 case BPF_LD
| BPF_H
| BPF_ABS
:
812 case BPF_LD
| BPF_B
| BPF_ABS
:
814 if (bpf_anc_helper(ftest
) & BPF_ANC
)
816 /* Ancillary operation unknown or unsupported */
817 if (anc_found
== false && ftest
->k
>= SKF_AD_OFF
)
822 /* Last instruction must be a RET code */
823 switch (filter
[flen
- 1].code
) {
824 case BPF_RET
| BPF_K
:
825 case BPF_RET
| BPF_A
:
826 return check_load_and_stores(filter
, flen
);
832 static int bpf_prog_store_orig_filter(struct bpf_prog
*fp
,
833 const struct sock_fprog
*fprog
)
835 unsigned int fsize
= bpf_classic_proglen(fprog
);
836 struct sock_fprog_kern
*fkprog
;
838 fp
->orig_prog
= kmalloc(sizeof(*fkprog
), GFP_KERNEL
);
842 fkprog
= fp
->orig_prog
;
843 fkprog
->len
= fprog
->len
;
845 fkprog
->filter
= kmemdup(fp
->insns
, fsize
,
846 GFP_KERNEL
| __GFP_NOWARN
);
847 if (!fkprog
->filter
) {
848 kfree(fp
->orig_prog
);
855 static void bpf_release_orig_filter(struct bpf_prog
*fp
)
857 struct sock_fprog_kern
*fprog
= fp
->orig_prog
;
860 kfree(fprog
->filter
);
865 static void __bpf_prog_release(struct bpf_prog
*prog
)
867 if (prog
->type
== BPF_PROG_TYPE_SOCKET_FILTER
) {
870 bpf_release_orig_filter(prog
);
875 static void __sk_filter_release(struct sk_filter
*fp
)
877 __bpf_prog_release(fp
->prog
);
882 * sk_filter_release_rcu - Release a socket filter by rcu_head
883 * @rcu: rcu_head that contains the sk_filter to free
885 static void sk_filter_release_rcu(struct rcu_head
*rcu
)
887 struct sk_filter
*fp
= container_of(rcu
, struct sk_filter
, rcu
);
889 __sk_filter_release(fp
);
893 * sk_filter_release - release a socket filter
894 * @fp: filter to remove
896 * Remove a filter from a socket and release its resources.
898 static void sk_filter_release(struct sk_filter
*fp
)
900 if (atomic_dec_and_test(&fp
->refcnt
))
901 call_rcu(&fp
->rcu
, sk_filter_release_rcu
);
904 void sk_filter_uncharge(struct sock
*sk
, struct sk_filter
*fp
)
906 u32 filter_size
= bpf_prog_size(fp
->prog
->len
);
908 atomic_sub(filter_size
, &sk
->sk_omem_alloc
);
909 sk_filter_release(fp
);
912 /* try to charge the socket memory if there is space available
913 * return true on success
915 bool sk_filter_charge(struct sock
*sk
, struct sk_filter
*fp
)
917 u32 filter_size
= bpf_prog_size(fp
->prog
->len
);
919 /* same check as in sock_kmalloc() */
920 if (filter_size
<= sysctl_optmem_max
&&
921 atomic_read(&sk
->sk_omem_alloc
) + filter_size
< sysctl_optmem_max
) {
922 atomic_inc(&fp
->refcnt
);
923 atomic_add(filter_size
, &sk
->sk_omem_alloc
);
929 static struct bpf_prog
*bpf_migrate_filter(struct bpf_prog
*fp
)
931 struct sock_filter
*old_prog
;
932 struct bpf_prog
*old_fp
;
933 int err
, new_len
, old_len
= fp
->len
;
935 /* We are free to overwrite insns et al right here as it
936 * won't be used at this point in time anymore internally
937 * after the migration to the internal BPF instruction
940 BUILD_BUG_ON(sizeof(struct sock_filter
) !=
941 sizeof(struct bpf_insn
));
943 /* Conversion cannot happen on overlapping memory areas,
944 * so we need to keep the user BPF around until the 2nd
945 * pass. At this time, the user BPF is stored in fp->insns.
947 old_prog
= kmemdup(fp
->insns
, old_len
* sizeof(struct sock_filter
),
948 GFP_KERNEL
| __GFP_NOWARN
);
954 /* 1st pass: calculate the new program length. */
955 err
= bpf_convert_filter(old_prog
, old_len
, NULL
, &new_len
);
959 /* Expand fp for appending the new filter representation. */
961 fp
= bpf_prog_realloc(old_fp
, bpf_prog_size(new_len
), 0);
963 /* The old_fp is still around in case we couldn't
964 * allocate new memory, so uncharge on that one.
973 /* 2nd pass: remap sock_filter insns into bpf_insn insns. */
974 err
= bpf_convert_filter(old_prog
, old_len
, fp
->insnsi
, &new_len
);
976 /* 2nd bpf_convert_filter() can fail only if it fails
977 * to allocate memory, remapping must succeed. Note,
978 * that at this time old_fp has already been released
983 bpf_prog_select_runtime(fp
);
991 __bpf_prog_release(fp
);
995 static struct bpf_prog
*bpf_prepare_filter(struct bpf_prog
*fp
,
996 bpf_aux_classic_check_t trans
)
1000 fp
->bpf_func
= NULL
;
1003 err
= bpf_check_classic(fp
->insns
, fp
->len
);
1005 __bpf_prog_release(fp
);
1006 return ERR_PTR(err
);
1009 /* There might be additional checks and transformations
1010 * needed on classic filters, f.e. in case of seccomp.
1013 err
= trans(fp
->insns
, fp
->len
);
1015 __bpf_prog_release(fp
);
1016 return ERR_PTR(err
);
1020 /* Probe if we can JIT compile the filter and if so, do
1021 * the compilation of the filter.
1023 bpf_jit_compile(fp
);
1025 /* JIT compiler couldn't process this filter, so do the
1026 * internal BPF translation for the optimized interpreter.
1029 fp
= bpf_migrate_filter(fp
);
1035 * bpf_prog_create - create an unattached filter
1036 * @pfp: the unattached filter that is created
1037 * @fprog: the filter program
1039 * Create a filter independent of any socket. We first run some
1040 * sanity checks on it to make sure it does not explode on us later.
1041 * If an error occurs or there is insufficient memory for the filter
1042 * a negative errno code is returned. On success the return is zero.
1044 int bpf_prog_create(struct bpf_prog
**pfp
, struct sock_fprog_kern
*fprog
)
1046 unsigned int fsize
= bpf_classic_proglen(fprog
);
1047 struct bpf_prog
*fp
;
1049 /* Make sure new filter is there and in the right amounts. */
1050 if (fprog
->filter
== NULL
)
1053 fp
= bpf_prog_alloc(bpf_prog_size(fprog
->len
), 0);
1057 memcpy(fp
->insns
, fprog
->filter
, fsize
);
1059 fp
->len
= fprog
->len
;
1060 /* Since unattached filters are not copied back to user
1061 * space through sk_get_filter(), we do not need to hold
1062 * a copy here, and can spare us the work.
1064 fp
->orig_prog
= NULL
;
1066 /* bpf_prepare_filter() already takes care of freeing
1067 * memory in case something goes wrong.
1069 fp
= bpf_prepare_filter(fp
, NULL
);
1076 EXPORT_SYMBOL_GPL(bpf_prog_create
);
1079 * bpf_prog_create_from_user - create an unattached filter from user buffer
1080 * @pfp: the unattached filter that is created
1081 * @fprog: the filter program
1082 * @trans: post-classic verifier transformation handler
1084 * This function effectively does the same as bpf_prog_create(), only
1085 * that it builds up its insns buffer from user space provided buffer.
1086 * It also allows for passing a bpf_aux_classic_check_t handler.
1088 int bpf_prog_create_from_user(struct bpf_prog
**pfp
, struct sock_fprog
*fprog
,
1089 bpf_aux_classic_check_t trans
)
1091 unsigned int fsize
= bpf_classic_proglen(fprog
);
1092 struct bpf_prog
*fp
;
1094 /* Make sure new filter is there and in the right amounts. */
1095 if (fprog
->filter
== NULL
)
1098 fp
= bpf_prog_alloc(bpf_prog_size(fprog
->len
), 0);
1102 if (copy_from_user(fp
->insns
, fprog
->filter
, fsize
)) {
1103 __bpf_prog_free(fp
);
1107 fp
->len
= fprog
->len
;
1108 /* Since unattached filters are not copied back to user
1109 * space through sk_get_filter(), we do not need to hold
1110 * a copy here, and can spare us the work.
1112 fp
->orig_prog
= NULL
;
1114 /* bpf_prepare_filter() already takes care of freeing
1115 * memory in case something goes wrong.
1117 fp
= bpf_prepare_filter(fp
, trans
);
1125 void bpf_prog_destroy(struct bpf_prog
*fp
)
1127 __bpf_prog_release(fp
);
1129 EXPORT_SYMBOL_GPL(bpf_prog_destroy
);
1131 static int __sk_attach_prog(struct bpf_prog
*prog
, struct sock
*sk
)
1133 struct sk_filter
*fp
, *old_fp
;
1135 fp
= kmalloc(sizeof(*fp
), GFP_KERNEL
);
1140 atomic_set(&fp
->refcnt
, 0);
1142 if (!sk_filter_charge(sk
, fp
)) {
1147 old_fp
= rcu_dereference_protected(sk
->sk_filter
,
1148 sock_owned_by_user(sk
));
1149 rcu_assign_pointer(sk
->sk_filter
, fp
);
1152 sk_filter_uncharge(sk
, old_fp
);
1158 * sk_attach_filter - attach a socket filter
1159 * @fprog: the filter program
1160 * @sk: the socket to use
1162 * Attach the user's filter code. We first run some sanity checks on
1163 * it to make sure it does not explode on us later. If an error
1164 * occurs or there is insufficient memory for the filter a negative
1165 * errno code is returned. On success the return is zero.
1167 int sk_attach_filter(struct sock_fprog
*fprog
, struct sock
*sk
)
1169 unsigned int fsize
= bpf_classic_proglen(fprog
);
1170 unsigned int bpf_fsize
= bpf_prog_size(fprog
->len
);
1171 struct bpf_prog
*prog
;
1174 if (sock_flag(sk
, SOCK_FILTER_LOCKED
))
1177 /* Make sure new filter is there and in the right amounts. */
1178 if (fprog
->filter
== NULL
)
1181 prog
= bpf_prog_alloc(bpf_fsize
, 0);
1185 if (copy_from_user(prog
->insns
, fprog
->filter
, fsize
)) {
1186 __bpf_prog_free(prog
);
1190 prog
->len
= fprog
->len
;
1192 err
= bpf_prog_store_orig_filter(prog
, fprog
);
1194 __bpf_prog_free(prog
);
1198 /* bpf_prepare_filter() already takes care of freeing
1199 * memory in case something goes wrong.
1201 prog
= bpf_prepare_filter(prog
, NULL
);
1203 return PTR_ERR(prog
);
1205 err
= __sk_attach_prog(prog
, sk
);
1207 __bpf_prog_release(prog
);
1213 EXPORT_SYMBOL_GPL(sk_attach_filter
);
1215 int sk_attach_bpf(u32 ufd
, struct sock
*sk
)
1217 struct bpf_prog
*prog
;
1220 if (sock_flag(sk
, SOCK_FILTER_LOCKED
))
1223 prog
= bpf_prog_get(ufd
);
1225 return PTR_ERR(prog
);
1227 if (prog
->type
!= BPF_PROG_TYPE_SOCKET_FILTER
) {
1232 err
= __sk_attach_prog(prog
, sk
);
1242 * bpf_skb_clone_not_writable - is the header of a clone not writable
1243 * @skb: buffer to check
1244 * @len: length up to which to write, can be negative
1246 * Returns true if modifying the header part of the cloned buffer
1247 * does require the data to be copied. I.e. this version works with
1248 * negative lengths needed for eBPF case!
1250 static bool bpf_skb_clone_unwritable(const struct sk_buff
*skb
, int len
)
1252 return skb_header_cloned(skb
) ||
1253 (int) skb_headroom(skb
) + len
> skb
->hdr_len
;
1256 #define BPF_RECOMPUTE_CSUM(flags) ((flags) & 1)
1258 static u64
bpf_skb_store_bytes(u64 r1
, u64 r2
, u64 r3
, u64 r4
, u64 flags
)
1260 struct sk_buff
*skb
= (struct sk_buff
*) (long) r1
;
1261 int offset
= (int) r2
;
1262 void *from
= (void *) (long) r3
;
1263 unsigned int len
= (unsigned int) r4
;
1267 /* bpf verifier guarantees that:
1268 * 'from' pointer points to bpf program stack
1269 * 'len' bytes of it were initialized
1271 * 'skb' is a valid pointer to 'struct sk_buff'
1273 * so check for invalid 'offset' and too large 'len'
1275 if (unlikely((u32
) offset
> 0xffff || len
> sizeof(buf
)))
1278 offset
-= skb
->data
- skb_mac_header(skb
);
1279 if (unlikely(skb_cloned(skb
) &&
1280 bpf_skb_clone_unwritable(skb
, offset
+ len
)))
1283 ptr
= skb_header_pointer(skb
, offset
, len
, buf
);
1287 if (BPF_RECOMPUTE_CSUM(flags
))
1288 skb_postpull_rcsum(skb
, ptr
, len
);
1290 memcpy(ptr
, from
, len
);
1293 /* skb_store_bits cannot return -EFAULT here */
1294 skb_store_bits(skb
, offset
, ptr
, len
);
1296 if (BPF_RECOMPUTE_CSUM(flags
) && skb
->ip_summed
== CHECKSUM_COMPLETE
)
1297 skb
->csum
= csum_add(skb
->csum
, csum_partial(ptr
, len
, 0));
1301 const struct bpf_func_proto bpf_skb_store_bytes_proto
= {
1302 .func
= bpf_skb_store_bytes
,
1304 .ret_type
= RET_INTEGER
,
1305 .arg1_type
= ARG_PTR_TO_CTX
,
1306 .arg2_type
= ARG_ANYTHING
,
1307 .arg3_type
= ARG_PTR_TO_STACK
,
1308 .arg4_type
= ARG_CONST_STACK_SIZE
,
1309 .arg5_type
= ARG_ANYTHING
,
1312 #define BPF_HEADER_FIELD_SIZE(flags) ((flags) & 0x0f)
1313 #define BPF_IS_PSEUDO_HEADER(flags) ((flags) & 0x10)
1315 static u64
bpf_l3_csum_replace(u64 r1
, u64 r2
, u64 from
, u64 to
, u64 flags
)
1317 struct sk_buff
*skb
= (struct sk_buff
*) (long) r1
;
1318 int offset
= (int) r2
;
1321 if (unlikely((u32
) offset
> 0xffff))
1324 offset
-= skb
->data
- skb_mac_header(skb
);
1325 if (unlikely(skb_cloned(skb
) &&
1326 bpf_skb_clone_unwritable(skb
, offset
+ sizeof(sum
))))
1329 ptr
= skb_header_pointer(skb
, offset
, sizeof(sum
), &sum
);
1333 switch (BPF_HEADER_FIELD_SIZE(flags
)) {
1335 csum_replace2(ptr
, from
, to
);
1338 csum_replace4(ptr
, from
, to
);
1345 /* skb_store_bits guaranteed to not return -EFAULT here */
1346 skb_store_bits(skb
, offset
, ptr
, sizeof(sum
));
1351 const struct bpf_func_proto bpf_l3_csum_replace_proto
= {
1352 .func
= bpf_l3_csum_replace
,
1354 .ret_type
= RET_INTEGER
,
1355 .arg1_type
= ARG_PTR_TO_CTX
,
1356 .arg2_type
= ARG_ANYTHING
,
1357 .arg3_type
= ARG_ANYTHING
,
1358 .arg4_type
= ARG_ANYTHING
,
1359 .arg5_type
= ARG_ANYTHING
,
1362 static u64
bpf_l4_csum_replace(u64 r1
, u64 r2
, u64 from
, u64 to
, u64 flags
)
1364 struct sk_buff
*skb
= (struct sk_buff
*) (long) r1
;
1365 u32 is_pseudo
= BPF_IS_PSEUDO_HEADER(flags
);
1366 int offset
= (int) r2
;
1369 if (unlikely((u32
) offset
> 0xffff))
1372 offset
-= skb
->data
- skb_mac_header(skb
);
1373 if (unlikely(skb_cloned(skb
) &&
1374 bpf_skb_clone_unwritable(skb
, offset
+ sizeof(sum
))))
1377 ptr
= skb_header_pointer(skb
, offset
, sizeof(sum
), &sum
);
1381 switch (BPF_HEADER_FIELD_SIZE(flags
)) {
1383 inet_proto_csum_replace2(ptr
, skb
, from
, to
, is_pseudo
);
1386 inet_proto_csum_replace4(ptr
, skb
, from
, to
, is_pseudo
);
1393 /* skb_store_bits guaranteed to not return -EFAULT here */
1394 skb_store_bits(skb
, offset
, ptr
, sizeof(sum
));
1399 const struct bpf_func_proto bpf_l4_csum_replace_proto
= {
1400 .func
= bpf_l4_csum_replace
,
1402 .ret_type
= RET_INTEGER
,
1403 .arg1_type
= ARG_PTR_TO_CTX
,
1404 .arg2_type
= ARG_ANYTHING
,
1405 .arg3_type
= ARG_ANYTHING
,
1406 .arg4_type
= ARG_ANYTHING
,
1407 .arg5_type
= ARG_ANYTHING
,
1410 static const struct bpf_func_proto
*
1411 sk_filter_func_proto(enum bpf_func_id func_id
)
1414 case BPF_FUNC_map_lookup_elem
:
1415 return &bpf_map_lookup_elem_proto
;
1416 case BPF_FUNC_map_update_elem
:
1417 return &bpf_map_update_elem_proto
;
1418 case BPF_FUNC_map_delete_elem
:
1419 return &bpf_map_delete_elem_proto
;
1420 case BPF_FUNC_get_prandom_u32
:
1421 return &bpf_get_prandom_u32_proto
;
1422 case BPF_FUNC_get_smp_processor_id
:
1423 return &bpf_get_smp_processor_id_proto
;
1424 case BPF_FUNC_tail_call
:
1425 return &bpf_tail_call_proto
;
1431 static const struct bpf_func_proto
*
1432 tc_cls_act_func_proto(enum bpf_func_id func_id
)
1435 case BPF_FUNC_skb_store_bytes
:
1436 return &bpf_skb_store_bytes_proto
;
1437 case BPF_FUNC_l3_csum_replace
:
1438 return &bpf_l3_csum_replace_proto
;
1439 case BPF_FUNC_l4_csum_replace
:
1440 return &bpf_l4_csum_replace_proto
;
1442 return sk_filter_func_proto(func_id
);
1446 static bool sk_filter_is_valid_access(int off
, int size
,
1447 enum bpf_access_type type
)
1449 /* only read is allowed */
1450 if (type
!= BPF_READ
)
1454 if (off
< 0 || off
>= sizeof(struct __sk_buff
))
1457 /* disallow misaligned access */
1458 if (off
% size
!= 0)
1461 /* all __sk_buff fields are __u32 */
1468 static u32
sk_filter_convert_ctx_access(int dst_reg
, int src_reg
, int ctx_off
,
1469 struct bpf_insn
*insn_buf
)
1471 struct bpf_insn
*insn
= insn_buf
;
1474 case offsetof(struct __sk_buff
, len
):
1475 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff
, len
) != 4);
1477 *insn
++ = BPF_LDX_MEM(BPF_W
, dst_reg
, src_reg
,
1478 offsetof(struct sk_buff
, len
));
1481 case offsetof(struct __sk_buff
, protocol
):
1482 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff
, protocol
) != 2);
1484 *insn
++ = BPF_LDX_MEM(BPF_H
, dst_reg
, src_reg
,
1485 offsetof(struct sk_buff
, protocol
));
1488 case offsetof(struct __sk_buff
, vlan_proto
):
1489 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff
, vlan_proto
) != 2);
1491 *insn
++ = BPF_LDX_MEM(BPF_H
, dst_reg
, src_reg
,
1492 offsetof(struct sk_buff
, vlan_proto
));
1495 case offsetof(struct __sk_buff
, priority
):
1496 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff
, priority
) != 4);
1498 *insn
++ = BPF_LDX_MEM(BPF_W
, dst_reg
, src_reg
,
1499 offsetof(struct sk_buff
, priority
));
1502 case offsetof(struct __sk_buff
, mark
):
1503 return convert_skb_access(SKF_AD_MARK
, dst_reg
, src_reg
, insn
);
1505 case offsetof(struct __sk_buff
, pkt_type
):
1506 return convert_skb_access(SKF_AD_PKTTYPE
, dst_reg
, src_reg
, insn
);
1508 case offsetof(struct __sk_buff
, queue_mapping
):
1509 return convert_skb_access(SKF_AD_QUEUE
, dst_reg
, src_reg
, insn
);
1511 case offsetof(struct __sk_buff
, vlan_present
):
1512 return convert_skb_access(SKF_AD_VLAN_TAG_PRESENT
,
1513 dst_reg
, src_reg
, insn
);
1515 case offsetof(struct __sk_buff
, vlan_tci
):
1516 return convert_skb_access(SKF_AD_VLAN_TAG
,
1517 dst_reg
, src_reg
, insn
);
1520 return insn
- insn_buf
;
1523 static const struct bpf_verifier_ops sk_filter_ops
= {
1524 .get_func_proto
= sk_filter_func_proto
,
1525 .is_valid_access
= sk_filter_is_valid_access
,
1526 .convert_ctx_access
= sk_filter_convert_ctx_access
,
1529 static const struct bpf_verifier_ops tc_cls_act_ops
= {
1530 .get_func_proto
= tc_cls_act_func_proto
,
1531 .is_valid_access
= sk_filter_is_valid_access
,
1532 .convert_ctx_access
= sk_filter_convert_ctx_access
,
1535 static struct bpf_prog_type_list sk_filter_type __read_mostly
= {
1536 .ops
= &sk_filter_ops
,
1537 .type
= BPF_PROG_TYPE_SOCKET_FILTER
,
1540 static struct bpf_prog_type_list sched_cls_type __read_mostly
= {
1541 .ops
= &tc_cls_act_ops
,
1542 .type
= BPF_PROG_TYPE_SCHED_CLS
,
1545 static struct bpf_prog_type_list sched_act_type __read_mostly
= {
1546 .ops
= &tc_cls_act_ops
,
1547 .type
= BPF_PROG_TYPE_SCHED_ACT
,
1550 static int __init
register_sk_filter_ops(void)
1552 bpf_register_prog_type(&sk_filter_type
);
1553 bpf_register_prog_type(&sched_cls_type
);
1554 bpf_register_prog_type(&sched_act_type
);
1558 late_initcall(register_sk_filter_ops
);
1560 int sk_detach_filter(struct sock
*sk
)
1563 struct sk_filter
*filter
;
1565 if (sock_flag(sk
, SOCK_FILTER_LOCKED
))
1568 filter
= rcu_dereference_protected(sk
->sk_filter
,
1569 sock_owned_by_user(sk
));
1571 RCU_INIT_POINTER(sk
->sk_filter
, NULL
);
1572 sk_filter_uncharge(sk
, filter
);
1578 EXPORT_SYMBOL_GPL(sk_detach_filter
);
1580 int sk_get_filter(struct sock
*sk
, struct sock_filter __user
*ubuf
,
1583 struct sock_fprog_kern
*fprog
;
1584 struct sk_filter
*filter
;
1588 filter
= rcu_dereference_protected(sk
->sk_filter
,
1589 sock_owned_by_user(sk
));
1593 /* We're copying the filter that has been originally attached,
1594 * so no conversion/decode needed anymore.
1596 fprog
= filter
->prog
->orig_prog
;
1600 /* User space only enquires number of filter blocks. */
1604 if (len
< fprog
->len
)
1608 if (copy_to_user(ubuf
, fprog
->filter
, bpf_classic_proglen(fprog
)))
1611 /* Instead of bytes, the API requests to return the number