2 * Linux Socket Filter - Kernel level socket filtering
5 * Jay Schulist <jschlst@samba.org>
7 * Based on the design of:
8 * - The Berkeley Packet Filter
10 * This program is free software; you can redistribute it and/or
11 * modify it under the terms of the GNU General Public License
12 * as published by the Free Software Foundation; either version
13 * 2 of the License, or (at your option) any later version.
15 * Andi Kleen - Fix a few bad bugs and races.
16 * Kris Katterjohn - Added many additional checks in sk_chk_filter()
19 #include <linux/module.h>
20 #include <linux/types.h>
22 #include <linux/fcntl.h>
23 #include <linux/socket.h>
25 #include <linux/inet.h>
26 #include <linux/netdevice.h>
27 #include <linux/if_packet.h>
28 #include <linux/gfp.h>
30 #include <net/protocol.h>
31 #include <net/netlink.h>
32 #include <linux/skbuff.h>
34 #include <linux/errno.h>
35 #include <linux/timer.h>
36 #include <asm/system.h>
37 #include <asm/uaccess.h>
38 #include <asm/unaligned.h>
39 #include <linux/filter.h>
41 /* No hurry in this branch */
42 static void *__load_pointer(struct sk_buff
*skb
, int k
)
47 ptr
= skb_network_header(skb
) + k
- SKF_NET_OFF
;
48 else if (k
>= SKF_LL_OFF
)
49 ptr
= skb_mac_header(skb
) + k
- SKF_LL_OFF
;
51 if (ptr
>= skb
->head
&& ptr
< skb_tail_pointer(skb
))
56 static inline void *load_pointer(struct sk_buff
*skb
, int k
,
57 unsigned int size
, void *buffer
)
60 return skb_header_pointer(skb
, k
, size
, buffer
);
64 return __load_pointer(skb
, k
);
69 * sk_filter - run a packet through a socket filter
70 * @sk: sock associated with &sk_buff
71 * @skb: buffer to filter
73 * Run the filter code and then cut skb->data to correct size returned by
74 * sk_run_filter. If pkt_len is 0 we toss packet. If skb->len is smaller
75 * than pkt_len we keep whole skb->data. This is the socket level
76 * wrapper to sk_run_filter. It returns 0 if the packet should
77 * be accepted or -EPERM if the packet should be tossed.
80 int sk_filter(struct sock
*sk
, struct sk_buff
*skb
)
83 struct sk_filter
*filter
;
85 err
= security_sock_rcv_skb(sk
, skb
);
90 filter
= rcu_dereference_bh(sk
->sk_filter
);
92 unsigned int pkt_len
= sk_run_filter(skb
, filter
->insns
,
94 err
= pkt_len
? pskb_trim(skb
, pkt_len
) : -EPERM
;
100 EXPORT_SYMBOL(sk_filter
);
103 * sk_run_filter - run a filter on a socket
104 * @skb: buffer to run the filter on
105 * @filter: filter to apply
106 * @flen: length of filter
108 * Decode and apply filter instructions to the skb->data.
109 * Return length to keep, 0 for none. skb is the data we are
110 * filtering, filter is the array of filter instructions, and
111 * len is the number of filter blocks in the array.
113 unsigned int sk_run_filter(struct sk_buff
*skb
, struct sock_filter
*filter
, int flen
)
115 struct sock_filter
*fentry
; /* We walk down these */
117 u32 A
= 0; /* Accumulator */
118 u32 X
= 0; /* Index Register */
119 u32 mem
[BPF_MEMWORDS
]; /* Scratch Memory Store */
125 * Process array of filter instructions.
127 for (pc
= 0; pc
< flen
; pc
++) {
128 fentry
= &filter
[pc
];
130 switch (fentry
->code
) {
131 case BPF_ALU
|BPF_ADD
|BPF_X
:
134 case BPF_ALU
|BPF_ADD
|BPF_K
:
137 case BPF_ALU
|BPF_SUB
|BPF_X
:
140 case BPF_ALU
|BPF_SUB
|BPF_K
:
143 case BPF_ALU
|BPF_MUL
|BPF_X
:
146 case BPF_ALU
|BPF_MUL
|BPF_K
:
149 case BPF_ALU
|BPF_DIV
|BPF_X
:
154 case BPF_ALU
|BPF_DIV
|BPF_K
:
157 case BPF_ALU
|BPF_AND
|BPF_X
:
160 case BPF_ALU
|BPF_AND
|BPF_K
:
163 case BPF_ALU
|BPF_OR
|BPF_X
:
166 case BPF_ALU
|BPF_OR
|BPF_K
:
169 case BPF_ALU
|BPF_LSH
|BPF_X
:
172 case BPF_ALU
|BPF_LSH
|BPF_K
:
175 case BPF_ALU
|BPF_RSH
|BPF_X
:
178 case BPF_ALU
|BPF_RSH
|BPF_K
:
181 case BPF_ALU
|BPF_NEG
:
187 case BPF_JMP
|BPF_JGT
|BPF_K
:
188 pc
+= (A
> fentry
->k
) ? fentry
->jt
: fentry
->jf
;
190 case BPF_JMP
|BPF_JGE
|BPF_K
:
191 pc
+= (A
>= fentry
->k
) ? fentry
->jt
: fentry
->jf
;
193 case BPF_JMP
|BPF_JEQ
|BPF_K
:
194 pc
+= (A
== fentry
->k
) ? fentry
->jt
: fentry
->jf
;
196 case BPF_JMP
|BPF_JSET
|BPF_K
:
197 pc
+= (A
& fentry
->k
) ? fentry
->jt
: fentry
->jf
;
199 case BPF_JMP
|BPF_JGT
|BPF_X
:
200 pc
+= (A
> X
) ? fentry
->jt
: fentry
->jf
;
202 case BPF_JMP
|BPF_JGE
|BPF_X
:
203 pc
+= (A
>= X
) ? fentry
->jt
: fentry
->jf
;
205 case BPF_JMP
|BPF_JEQ
|BPF_X
:
206 pc
+= (A
== X
) ? fentry
->jt
: fentry
->jf
;
208 case BPF_JMP
|BPF_JSET
|BPF_X
:
209 pc
+= (A
& X
) ? fentry
->jt
: fentry
->jf
;
211 case BPF_LD
|BPF_W
|BPF_ABS
:
214 ptr
= load_pointer(skb
, k
, 4, &tmp
);
216 A
= get_unaligned_be32(ptr
);
220 case BPF_LD
|BPF_H
|BPF_ABS
:
223 ptr
= load_pointer(skb
, k
, 2, &tmp
);
225 A
= get_unaligned_be16(ptr
);
229 case BPF_LD
|BPF_B
|BPF_ABS
:
232 ptr
= load_pointer(skb
, k
, 1, &tmp
);
238 case BPF_LD
|BPF_W
|BPF_LEN
:
241 case BPF_LDX
|BPF_W
|BPF_LEN
:
244 case BPF_LD
|BPF_W
|BPF_IND
:
247 case BPF_LD
|BPF_H
|BPF_IND
:
250 case BPF_LD
|BPF_B
|BPF_IND
:
253 case BPF_LDX
|BPF_B
|BPF_MSH
:
254 ptr
= load_pointer(skb
, fentry
->k
, 1, &tmp
);
256 X
= (*(u8
*)ptr
& 0xf) << 2;
263 case BPF_LDX
|BPF_IMM
:
269 case BPF_LDX
|BPF_MEM
:
272 case BPF_MISC
|BPF_TAX
:
275 case BPF_MISC
|BPF_TXA
:
294 * Handle ancillary data, which are impossible
295 * (or very difficult) to get parsing packet contents.
297 switch (k
-SKF_AD_OFF
) {
298 case SKF_AD_PROTOCOL
:
299 A
= ntohs(skb
->protocol
);
307 A
= skb
->dev
->ifindex
;
313 A
= skb
->queue_mapping
;
320 case SKF_AD_NLATTR
: {
323 if (skb_is_nonlinear(skb
))
325 if (A
> skb
->len
- sizeof(struct nlattr
))
328 nla
= nla_find((struct nlattr
*)&skb
->data
[A
],
331 A
= (void *)nla
- (void *)skb
->data
;
336 case SKF_AD_NLATTR_NEST
: {
339 if (skb_is_nonlinear(skb
))
341 if (A
> skb
->len
- sizeof(struct nlattr
))
344 nla
= (struct nlattr
*)&skb
->data
[A
];
345 if (nla
->nla_len
> A
- skb
->len
)
348 nla
= nla_find_nested(nla
, X
);
350 A
= (void *)nla
- (void *)skb
->data
;
362 EXPORT_SYMBOL(sk_run_filter
);
365 * sk_chk_filter - verify socket filter code
366 * @filter: filter to verify
367 * @flen: length of filter
369 * Check the user's filter code. If we let some ugly
370 * filter code slip through kaboom! The filter must contain
371 * no references or jumps that are out of range, no illegal
372 * instructions, and must end with a RET instruction.
374 * All jumps are forward as they are not signed.
376 * Returns 0 if the rule set is legal or -EINVAL if not.
378 int sk_chk_filter(struct sock_filter
*filter
, int flen
)
380 struct sock_filter
*ftest
;
383 if (flen
== 0 || flen
> BPF_MAXINSNS
)
386 /* check the filter code now */
387 for (pc
= 0; pc
< flen
; pc
++) {
390 /* Only allow valid instructions */
391 switch (ftest
->code
) {
392 case BPF_ALU
|BPF_ADD
|BPF_K
:
393 case BPF_ALU
|BPF_ADD
|BPF_X
:
394 case BPF_ALU
|BPF_SUB
|BPF_K
:
395 case BPF_ALU
|BPF_SUB
|BPF_X
:
396 case BPF_ALU
|BPF_MUL
|BPF_K
:
397 case BPF_ALU
|BPF_MUL
|BPF_X
:
398 case BPF_ALU
|BPF_DIV
|BPF_X
:
399 case BPF_ALU
|BPF_AND
|BPF_K
:
400 case BPF_ALU
|BPF_AND
|BPF_X
:
401 case BPF_ALU
|BPF_OR
|BPF_K
:
402 case BPF_ALU
|BPF_OR
|BPF_X
:
403 case BPF_ALU
|BPF_LSH
|BPF_K
:
404 case BPF_ALU
|BPF_LSH
|BPF_X
:
405 case BPF_ALU
|BPF_RSH
|BPF_K
:
406 case BPF_ALU
|BPF_RSH
|BPF_X
:
407 case BPF_ALU
|BPF_NEG
:
408 case BPF_LD
|BPF_W
|BPF_ABS
:
409 case BPF_LD
|BPF_H
|BPF_ABS
:
410 case BPF_LD
|BPF_B
|BPF_ABS
:
411 case BPF_LD
|BPF_W
|BPF_LEN
:
412 case BPF_LD
|BPF_W
|BPF_IND
:
413 case BPF_LD
|BPF_H
|BPF_IND
:
414 case BPF_LD
|BPF_B
|BPF_IND
:
416 case BPF_LDX
|BPF_W
|BPF_LEN
:
417 case BPF_LDX
|BPF_B
|BPF_MSH
:
418 case BPF_LDX
|BPF_IMM
:
419 case BPF_MISC
|BPF_TAX
:
420 case BPF_MISC
|BPF_TXA
:
425 /* Some instructions need special checks */
427 case BPF_ALU
|BPF_DIV
|BPF_K
:
428 /* check for division by zero */
434 case BPF_LDX
|BPF_MEM
:
437 /* check for invalid memory addresses */
438 if (ftest
->k
>= BPF_MEMWORDS
)
444 * Note, the large ftest->k might cause loops.
445 * Compare this with conditional jumps below,
446 * where offsets are limited. --ANK (981016)
448 if (ftest
->k
>= (unsigned)(flen
-pc
-1))
452 case BPF_JMP
|BPF_JEQ
|BPF_K
:
453 case BPF_JMP
|BPF_JEQ
|BPF_X
:
454 case BPF_JMP
|BPF_JGE
|BPF_K
:
455 case BPF_JMP
|BPF_JGE
|BPF_X
:
456 case BPF_JMP
|BPF_JGT
|BPF_K
:
457 case BPF_JMP
|BPF_JGT
|BPF_X
:
458 case BPF_JMP
|BPF_JSET
|BPF_K
:
459 case BPF_JMP
|BPF_JSET
|BPF_X
:
460 /* for conditionals both must be safe */
461 if (pc
+ ftest
->jt
+ 1 >= flen
||
462 pc
+ ftest
->jf
+ 1 >= flen
)
471 return (BPF_CLASS(filter
[flen
- 1].code
) == BPF_RET
) ? 0 : -EINVAL
;
473 EXPORT_SYMBOL(sk_chk_filter
);
476 * sk_filter_rcu_release: Release a socket filter by rcu_head
477 * @rcu: rcu_head that contains the sk_filter to free
479 static void sk_filter_rcu_release(struct rcu_head
*rcu
)
481 struct sk_filter
*fp
= container_of(rcu
, struct sk_filter
, rcu
);
483 sk_filter_release(fp
);
486 static void sk_filter_delayed_uncharge(struct sock
*sk
, struct sk_filter
*fp
)
488 unsigned int size
= sk_filter_len(fp
);
490 atomic_sub(size
, &sk
->sk_omem_alloc
);
491 call_rcu_bh(&fp
->rcu
, sk_filter_rcu_release
);
495 * sk_attach_filter - attach a socket filter
496 * @fprog: the filter program
497 * @sk: the socket to use
499 * Attach the user's filter code. We first run some sanity checks on
500 * it to make sure it does not explode on us later. If an error
501 * occurs or there is insufficient memory for the filter a negative
502 * errno code is returned. On success the return is zero.
504 int sk_attach_filter(struct sock_fprog
*fprog
, struct sock
*sk
)
506 struct sk_filter
*fp
, *old_fp
;
507 unsigned int fsize
= sizeof(struct sock_filter
) * fprog
->len
;
510 /* Make sure new filter is there and in the right amounts. */
511 if (fprog
->filter
== NULL
)
514 fp
= sock_kmalloc(sk
, fsize
+sizeof(*fp
), GFP_KERNEL
);
517 if (copy_from_user(fp
->insns
, fprog
->filter
, fsize
)) {
518 sock_kfree_s(sk
, fp
, fsize
+sizeof(*fp
));
522 atomic_set(&fp
->refcnt
, 1);
523 fp
->len
= fprog
->len
;
525 err
= sk_chk_filter(fp
->insns
, fp
->len
);
527 sk_filter_uncharge(sk
, fp
);
532 old_fp
= rcu_dereference_bh(sk
->sk_filter
);
533 rcu_assign_pointer(sk
->sk_filter
, fp
);
534 rcu_read_unlock_bh();
537 sk_filter_delayed_uncharge(sk
, old_fp
);
540 EXPORT_SYMBOL_GPL(sk_attach_filter
);
542 int sk_detach_filter(struct sock
*sk
)
545 struct sk_filter
*filter
;
548 filter
= rcu_dereference_bh(sk
->sk_filter
);
550 rcu_assign_pointer(sk
->sk_filter
, NULL
);
551 sk_filter_delayed_uncharge(sk
, filter
);
554 rcu_read_unlock_bh();
557 EXPORT_SYMBOL_GPL(sk_detach_filter
);