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>
29 #include <net/protocol.h>
30 #include <net/netlink.h>
31 #include <linux/skbuff.h>
33 #include <linux/errno.h>
34 #include <linux/timer.h>
35 #include <asm/system.h>
36 #include <asm/uaccess.h>
37 #include <asm/unaligned.h>
38 #include <linux/filter.h>
40 /* No hurry in this branch */
41 static void *__load_pointer(struct sk_buff
*skb
, int k
)
46 ptr
= skb_network_header(skb
) + k
- SKF_NET_OFF
;
47 else if (k
>= SKF_LL_OFF
)
48 ptr
= skb_mac_header(skb
) + k
- SKF_LL_OFF
;
50 if (ptr
>= skb
->head
&& ptr
< skb_tail_pointer(skb
))
55 static inline void *load_pointer(struct sk_buff
*skb
, int k
,
56 unsigned int size
, void *buffer
)
59 return skb_header_pointer(skb
, k
, size
, buffer
);
63 return __load_pointer(skb
, k
);
68 * sk_filter - run a packet through a socket filter
69 * @sk: sock associated with &sk_buff
70 * @skb: buffer to filter
72 * Run the filter code and then cut skb->data to correct size returned by
73 * sk_run_filter. If pkt_len is 0 we toss packet. If skb->len is smaller
74 * than pkt_len we keep whole skb->data. This is the socket level
75 * wrapper to sk_run_filter. It returns 0 if the packet should
76 * be accepted or -EPERM if the packet should be tossed.
79 int sk_filter(struct sock
*sk
, struct sk_buff
*skb
)
82 struct sk_filter
*filter
;
84 err
= security_sock_rcv_skb(sk
, skb
);
89 filter
= rcu_dereference(sk
->sk_filter
);
91 unsigned int pkt_len
= sk_run_filter(skb
, filter
->insns
,
93 err
= pkt_len
? pskb_trim(skb
, pkt_len
) : -EPERM
;
99 EXPORT_SYMBOL(sk_filter
);
102 * sk_run_filter - run a filter on a socket
103 * @skb: buffer to run the filter on
104 * @filter: filter to apply
105 * @flen: length of filter
107 * Decode and apply filter instructions to the skb->data.
108 * Return length to keep, 0 for none. skb is the data we are
109 * filtering, filter is the array of filter instructions, and
110 * len is the number of filter blocks in the array.
112 unsigned int sk_run_filter(struct sk_buff
*skb
, struct sock_filter
*filter
, int flen
)
114 struct sock_filter
*fentry
; /* We walk down these */
116 u32 A
= 0; /* Accumulator */
117 u32 X
= 0; /* Index Register */
118 u32 mem
[BPF_MEMWORDS
]; /* Scratch Memory Store */
124 * Process array of filter instructions.
126 for (pc
= 0; pc
< flen
; pc
++) {
127 fentry
= &filter
[pc
];
129 switch (fentry
->code
) {
130 case BPF_ALU
|BPF_ADD
|BPF_X
:
133 case BPF_ALU
|BPF_ADD
|BPF_K
:
136 case BPF_ALU
|BPF_SUB
|BPF_X
:
139 case BPF_ALU
|BPF_SUB
|BPF_K
:
142 case BPF_ALU
|BPF_MUL
|BPF_X
:
145 case BPF_ALU
|BPF_MUL
|BPF_K
:
148 case BPF_ALU
|BPF_DIV
|BPF_X
:
153 case BPF_ALU
|BPF_DIV
|BPF_K
:
156 case BPF_ALU
|BPF_AND
|BPF_X
:
159 case BPF_ALU
|BPF_AND
|BPF_K
:
162 case BPF_ALU
|BPF_OR
|BPF_X
:
165 case BPF_ALU
|BPF_OR
|BPF_K
:
168 case BPF_ALU
|BPF_LSH
|BPF_X
:
171 case BPF_ALU
|BPF_LSH
|BPF_K
:
174 case BPF_ALU
|BPF_RSH
|BPF_X
:
177 case BPF_ALU
|BPF_RSH
|BPF_K
:
180 case BPF_ALU
|BPF_NEG
:
186 case BPF_JMP
|BPF_JGT
|BPF_K
:
187 pc
+= (A
> fentry
->k
) ? fentry
->jt
: fentry
->jf
;
189 case BPF_JMP
|BPF_JGE
|BPF_K
:
190 pc
+= (A
>= fentry
->k
) ? fentry
->jt
: fentry
->jf
;
192 case BPF_JMP
|BPF_JEQ
|BPF_K
:
193 pc
+= (A
== fentry
->k
) ? fentry
->jt
: fentry
->jf
;
195 case BPF_JMP
|BPF_JSET
|BPF_K
:
196 pc
+= (A
& fentry
->k
) ? fentry
->jt
: fentry
->jf
;
198 case BPF_JMP
|BPF_JGT
|BPF_X
:
199 pc
+= (A
> X
) ? fentry
->jt
: fentry
->jf
;
201 case BPF_JMP
|BPF_JGE
|BPF_X
:
202 pc
+= (A
>= X
) ? fentry
->jt
: fentry
->jf
;
204 case BPF_JMP
|BPF_JEQ
|BPF_X
:
205 pc
+= (A
== X
) ? fentry
->jt
: fentry
->jf
;
207 case BPF_JMP
|BPF_JSET
|BPF_X
:
208 pc
+= (A
& X
) ? fentry
->jt
: fentry
->jf
;
210 case BPF_LD
|BPF_W
|BPF_ABS
:
213 ptr
= load_pointer(skb
, k
, 4, &tmp
);
215 A
= get_unaligned_be32(ptr
);
219 case BPF_LD
|BPF_H
|BPF_ABS
:
222 ptr
= load_pointer(skb
, k
, 2, &tmp
);
224 A
= get_unaligned_be16(ptr
);
228 case BPF_LD
|BPF_B
|BPF_ABS
:
231 ptr
= load_pointer(skb
, k
, 1, &tmp
);
237 case BPF_LD
|BPF_W
|BPF_LEN
:
240 case BPF_LDX
|BPF_W
|BPF_LEN
:
243 case BPF_LD
|BPF_W
|BPF_IND
:
246 case BPF_LD
|BPF_H
|BPF_IND
:
249 case BPF_LD
|BPF_B
|BPF_IND
:
252 case BPF_LDX
|BPF_B
|BPF_MSH
:
253 ptr
= load_pointer(skb
, fentry
->k
, 1, &tmp
);
255 X
= (*(u8
*)ptr
& 0xf) << 2;
262 case BPF_LDX
|BPF_IMM
:
268 case BPF_LDX
|BPF_MEM
:
271 case BPF_MISC
|BPF_TAX
:
274 case BPF_MISC
|BPF_TXA
:
293 * Handle ancillary data, which are impossible
294 * (or very difficult) to get parsing packet contents.
296 switch (k
-SKF_AD_OFF
) {
297 case SKF_AD_PROTOCOL
:
298 A
= ntohs(skb
->protocol
);
304 A
= skb
->dev
->ifindex
;
310 A
= skb
->queue_mapping
;
312 case SKF_AD_NLATTR
: {
315 if (skb_is_nonlinear(skb
))
317 if (A
> skb
->len
- sizeof(struct nlattr
))
320 nla
= nla_find((struct nlattr
*)&skb
->data
[A
],
323 A
= (void *)nla
- (void *)skb
->data
;
328 case SKF_AD_NLATTR_NEST
: {
331 if (skb_is_nonlinear(skb
))
333 if (A
> skb
->len
- sizeof(struct nlattr
))
336 nla
= (struct nlattr
*)&skb
->data
[A
];
337 if (nla
->nla_len
> A
- skb
->len
)
340 nla
= nla_find_nested(nla
, X
);
342 A
= (void *)nla
- (void *)skb
->data
;
354 EXPORT_SYMBOL(sk_run_filter
);
357 * sk_chk_filter - verify socket filter code
358 * @filter: filter to verify
359 * @flen: length of filter
361 * Check the user's filter code. If we let some ugly
362 * filter code slip through kaboom! The filter must contain
363 * no references or jumps that are out of range, no illegal
364 * instructions, and must end with a RET instruction.
366 * All jumps are forward as they are not signed.
368 * Returns 0 if the rule set is legal or -EINVAL if not.
370 int sk_chk_filter(struct sock_filter
*filter
, int flen
)
372 struct sock_filter
*ftest
;
375 if (flen
== 0 || flen
> BPF_MAXINSNS
)
378 /* check the filter code now */
379 for (pc
= 0; pc
< flen
; pc
++) {
382 /* Only allow valid instructions */
383 switch (ftest
->code
) {
384 case BPF_ALU
|BPF_ADD
|BPF_K
:
385 case BPF_ALU
|BPF_ADD
|BPF_X
:
386 case BPF_ALU
|BPF_SUB
|BPF_K
:
387 case BPF_ALU
|BPF_SUB
|BPF_X
:
388 case BPF_ALU
|BPF_MUL
|BPF_K
:
389 case BPF_ALU
|BPF_MUL
|BPF_X
:
390 case BPF_ALU
|BPF_DIV
|BPF_X
:
391 case BPF_ALU
|BPF_AND
|BPF_K
:
392 case BPF_ALU
|BPF_AND
|BPF_X
:
393 case BPF_ALU
|BPF_OR
|BPF_K
:
394 case BPF_ALU
|BPF_OR
|BPF_X
:
395 case BPF_ALU
|BPF_LSH
|BPF_K
:
396 case BPF_ALU
|BPF_LSH
|BPF_X
:
397 case BPF_ALU
|BPF_RSH
|BPF_K
:
398 case BPF_ALU
|BPF_RSH
|BPF_X
:
399 case BPF_ALU
|BPF_NEG
:
400 case BPF_LD
|BPF_W
|BPF_ABS
:
401 case BPF_LD
|BPF_H
|BPF_ABS
:
402 case BPF_LD
|BPF_B
|BPF_ABS
:
403 case BPF_LD
|BPF_W
|BPF_LEN
:
404 case BPF_LD
|BPF_W
|BPF_IND
:
405 case BPF_LD
|BPF_H
|BPF_IND
:
406 case BPF_LD
|BPF_B
|BPF_IND
:
408 case BPF_LDX
|BPF_W
|BPF_LEN
:
409 case BPF_LDX
|BPF_B
|BPF_MSH
:
410 case BPF_LDX
|BPF_IMM
:
411 case BPF_MISC
|BPF_TAX
:
412 case BPF_MISC
|BPF_TXA
:
417 /* Some instructions need special checks */
419 case BPF_ALU
|BPF_DIV
|BPF_K
:
420 /* check for division by zero */
426 case BPF_LDX
|BPF_MEM
:
429 /* check for invalid memory addresses */
430 if (ftest
->k
>= BPF_MEMWORDS
)
436 * Note, the large ftest->k might cause loops.
437 * Compare this with conditional jumps below,
438 * where offsets are limited. --ANK (981016)
440 if (ftest
->k
>= (unsigned)(flen
-pc
-1))
444 case BPF_JMP
|BPF_JEQ
|BPF_K
:
445 case BPF_JMP
|BPF_JEQ
|BPF_X
:
446 case BPF_JMP
|BPF_JGE
|BPF_K
:
447 case BPF_JMP
|BPF_JGE
|BPF_X
:
448 case BPF_JMP
|BPF_JGT
|BPF_K
:
449 case BPF_JMP
|BPF_JGT
|BPF_X
:
450 case BPF_JMP
|BPF_JSET
|BPF_K
:
451 case BPF_JMP
|BPF_JSET
|BPF_X
:
452 /* for conditionals both must be safe */
453 if (pc
+ ftest
->jt
+ 1 >= flen
||
454 pc
+ ftest
->jf
+ 1 >= flen
)
463 return (BPF_CLASS(filter
[flen
- 1].code
) == BPF_RET
) ? 0 : -EINVAL
;
465 EXPORT_SYMBOL(sk_chk_filter
);
468 * sk_filter_rcu_release: Release a socket filter by rcu_head
469 * @rcu: rcu_head that contains the sk_filter to free
471 static void sk_filter_rcu_release(struct rcu_head
*rcu
)
473 struct sk_filter
*fp
= container_of(rcu
, struct sk_filter
, rcu
);
475 sk_filter_release(fp
);
478 static void sk_filter_delayed_uncharge(struct sock
*sk
, struct sk_filter
*fp
)
480 unsigned int size
= sk_filter_len(fp
);
482 atomic_sub(size
, &sk
->sk_omem_alloc
);
483 call_rcu_bh(&fp
->rcu
, sk_filter_rcu_release
);
487 * sk_attach_filter - attach a socket filter
488 * @fprog: the filter program
489 * @sk: the socket to use
491 * Attach the user's filter code. We first run some sanity checks on
492 * it to make sure it does not explode on us later. If an error
493 * occurs or there is insufficient memory for the filter a negative
494 * errno code is returned. On success the return is zero.
496 int sk_attach_filter(struct sock_fprog
*fprog
, struct sock
*sk
)
498 struct sk_filter
*fp
, *old_fp
;
499 unsigned int fsize
= sizeof(struct sock_filter
) * fprog
->len
;
502 /* Make sure new filter is there and in the right amounts. */
503 if (fprog
->filter
== NULL
)
506 fp
= sock_kmalloc(sk
, fsize
+sizeof(*fp
), GFP_KERNEL
);
509 if (copy_from_user(fp
->insns
, fprog
->filter
, fsize
)) {
510 sock_kfree_s(sk
, fp
, fsize
+sizeof(*fp
));
514 atomic_set(&fp
->refcnt
, 1);
515 fp
->len
= fprog
->len
;
517 err
= sk_chk_filter(fp
->insns
, fp
->len
);
519 sk_filter_uncharge(sk
, fp
);
524 old_fp
= rcu_dereference(sk
->sk_filter
);
525 rcu_assign_pointer(sk
->sk_filter
, fp
);
526 rcu_read_unlock_bh();
529 sk_filter_delayed_uncharge(sk
, old_fp
);
533 int sk_detach_filter(struct sock
*sk
)
536 struct sk_filter
*filter
;
539 filter
= rcu_dereference(sk
->sk_filter
);
541 rcu_assign_pointer(sk
->sk_filter
, NULL
);
542 sk_filter_delayed_uncharge(sk
, filter
);
545 rcu_read_unlock_bh();