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
, filter
->len
);
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
)
116 u32 A
= 0; /* Accumulator */
117 u32 X
= 0; /* Index Register */
118 u32 mem
[BPF_MEMWORDS
]; /* Scratch Memory Store */
119 unsigned long memvalid
= 0;
124 BUILD_BUG_ON(BPF_MEMWORDS
> BITS_PER_LONG
);
126 * Process array of filter instructions.
128 for (pc
= 0; pc
< flen
; pc
++) {
129 const struct sock_filter
*fentry
= &filter
[pc
];
132 switch (fentry
->code
) {
133 case BPF_S_ALU_ADD_X
:
136 case BPF_S_ALU_ADD_K
:
139 case BPF_S_ALU_SUB_X
:
142 case BPF_S_ALU_SUB_K
:
145 case BPF_S_ALU_MUL_X
:
148 case BPF_S_ALU_MUL_K
:
151 case BPF_S_ALU_DIV_X
:
156 case BPF_S_ALU_DIV_K
:
159 case BPF_S_ALU_AND_X
:
162 case BPF_S_ALU_AND_K
:
171 case BPF_S_ALU_LSH_X
:
174 case BPF_S_ALU_LSH_K
:
177 case BPF_S_ALU_RSH_X
:
180 case BPF_S_ALU_RSH_K
:
189 case BPF_S_JMP_JGT_K
:
190 pc
+= (A
> f_k
) ? fentry
->jt
: fentry
->jf
;
192 case BPF_S_JMP_JGE_K
:
193 pc
+= (A
>= f_k
) ? fentry
->jt
: fentry
->jf
;
195 case BPF_S_JMP_JEQ_K
:
196 pc
+= (A
== f_k
) ? fentry
->jt
: fentry
->jf
;
198 case BPF_S_JMP_JSET_K
:
199 pc
+= (A
& f_k
) ? fentry
->jt
: fentry
->jf
;
201 case BPF_S_JMP_JGT_X
:
202 pc
+= (A
> X
) ? fentry
->jt
: fentry
->jf
;
204 case BPF_S_JMP_JGE_X
:
205 pc
+= (A
>= X
) ? fentry
->jt
: fentry
->jf
;
207 case BPF_S_JMP_JEQ_X
:
208 pc
+= (A
== X
) ? fentry
->jt
: fentry
->jf
;
210 case BPF_S_JMP_JSET_X
:
211 pc
+= (A
& X
) ? fentry
->jt
: fentry
->jf
;
216 ptr
= load_pointer(skb
, k
, 4, &tmp
);
218 A
= get_unaligned_be32(ptr
);
225 ptr
= load_pointer(skb
, k
, 2, &tmp
);
227 A
= get_unaligned_be16(ptr
);
234 ptr
= load_pointer(skb
, k
, 1, &tmp
);
243 case BPF_S_LDX_W_LEN
:
255 case BPF_S_LDX_B_MSH
:
256 ptr
= load_pointer(skb
, f_k
, 1, &tmp
);
258 X
= (*(u8
*)ptr
& 0xf) << 2;
269 A
= (memvalid
& (1UL << f_k
)) ?
273 X
= (memvalid
& (1UL << f_k
)) ?
287 memvalid
|= 1UL << f_k
;
291 memvalid
|= 1UL << f_k
;
300 * Handle ancillary data, which are impossible
301 * (or very difficult) to get parsing packet contents.
303 switch (k
-SKF_AD_OFF
) {
304 case SKF_AD_PROTOCOL
:
305 A
= ntohs(skb
->protocol
);
313 A
= skb
->dev
->ifindex
;
319 A
= skb
->queue_mapping
;
326 case SKF_AD_NLATTR
: {
329 if (skb_is_nonlinear(skb
))
331 if (A
> skb
->len
- sizeof(struct nlattr
))
334 nla
= nla_find((struct nlattr
*)&skb
->data
[A
],
337 A
= (void *)nla
- (void *)skb
->data
;
342 case SKF_AD_NLATTR_NEST
: {
345 if (skb_is_nonlinear(skb
))
347 if (A
> skb
->len
- sizeof(struct nlattr
))
350 nla
= (struct nlattr
*)&skb
->data
[A
];
351 if (nla
->nla_len
> A
- skb
->len
)
354 nla
= nla_find_nested(nla
, X
);
356 A
= (void *)nla
- (void *)skb
->data
;
368 EXPORT_SYMBOL(sk_run_filter
);
371 * sk_chk_filter - verify socket filter code
372 * @filter: filter to verify
373 * @flen: length of filter
375 * Check the user's filter code. If we let some ugly
376 * filter code slip through kaboom! The filter must contain
377 * no references or jumps that are out of range, no illegal
378 * instructions, and must end with a RET instruction.
380 * All jumps are forward as they are not signed.
382 * Returns 0 if the rule set is legal or -EINVAL if not.
384 int sk_chk_filter(struct sock_filter
*filter
, int flen
)
387 * Valid instructions are initialized to non-0.
388 * Invalid instructions are initialized to 0.
390 static const u8 codes
[] = {
391 [BPF_ALU
|BPF_ADD
|BPF_K
] = BPF_S_ALU_ADD_K
+ 1,
392 [BPF_ALU
|BPF_ADD
|BPF_X
] = BPF_S_ALU_ADD_X
+ 1,
393 [BPF_ALU
|BPF_SUB
|BPF_K
] = BPF_S_ALU_SUB_K
+ 1,
394 [BPF_ALU
|BPF_SUB
|BPF_X
] = BPF_S_ALU_SUB_X
+ 1,
395 [BPF_ALU
|BPF_MUL
|BPF_K
] = BPF_S_ALU_MUL_K
+ 1,
396 [BPF_ALU
|BPF_MUL
|BPF_X
] = BPF_S_ALU_MUL_X
+ 1,
397 [BPF_ALU
|BPF_DIV
|BPF_X
] = BPF_S_ALU_DIV_X
+ 1,
398 [BPF_ALU
|BPF_AND
|BPF_K
] = BPF_S_ALU_AND_K
+ 1,
399 [BPF_ALU
|BPF_AND
|BPF_X
] = BPF_S_ALU_AND_X
+ 1,
400 [BPF_ALU
|BPF_OR
|BPF_K
] = BPF_S_ALU_OR_K
+ 1,
401 [BPF_ALU
|BPF_OR
|BPF_X
] = BPF_S_ALU_OR_X
+ 1,
402 [BPF_ALU
|BPF_LSH
|BPF_K
] = BPF_S_ALU_LSH_K
+ 1,
403 [BPF_ALU
|BPF_LSH
|BPF_X
] = BPF_S_ALU_LSH_X
+ 1,
404 [BPF_ALU
|BPF_RSH
|BPF_K
] = BPF_S_ALU_RSH_K
+ 1,
405 [BPF_ALU
|BPF_RSH
|BPF_X
] = BPF_S_ALU_RSH_X
+ 1,
406 [BPF_ALU
|BPF_NEG
] = BPF_S_ALU_NEG
+ 1,
407 [BPF_LD
|BPF_W
|BPF_ABS
] = BPF_S_LD_W_ABS
+ 1,
408 [BPF_LD
|BPF_H
|BPF_ABS
] = BPF_S_LD_H_ABS
+ 1,
409 [BPF_LD
|BPF_B
|BPF_ABS
] = BPF_S_LD_B_ABS
+ 1,
410 [BPF_LD
|BPF_W
|BPF_LEN
] = BPF_S_LD_W_LEN
+ 1,
411 [BPF_LD
|BPF_W
|BPF_IND
] = BPF_S_LD_W_IND
+ 1,
412 [BPF_LD
|BPF_H
|BPF_IND
] = BPF_S_LD_H_IND
+ 1,
413 [BPF_LD
|BPF_B
|BPF_IND
] = BPF_S_LD_B_IND
+ 1,
414 [BPF_LD
|BPF_IMM
] = BPF_S_LD_IMM
+ 1,
415 [BPF_LDX
|BPF_W
|BPF_LEN
] = BPF_S_LDX_W_LEN
+ 1,
416 [BPF_LDX
|BPF_B
|BPF_MSH
] = BPF_S_LDX_B_MSH
+ 1,
417 [BPF_LDX
|BPF_IMM
] = BPF_S_LDX_IMM
+ 1,
418 [BPF_MISC
|BPF_TAX
] = BPF_S_MISC_TAX
+ 1,
419 [BPF_MISC
|BPF_TXA
] = BPF_S_MISC_TXA
+ 1,
420 [BPF_RET
|BPF_K
] = BPF_S_RET_K
+ 1,
421 [BPF_RET
|BPF_A
] = BPF_S_RET_A
+ 1,
422 [BPF_ALU
|BPF_DIV
|BPF_K
] = BPF_S_ALU_DIV_K
+ 1,
423 [BPF_LD
|BPF_MEM
] = BPF_S_LD_MEM
+ 1,
424 [BPF_LDX
|BPF_MEM
] = BPF_S_LDX_MEM
+ 1,
425 [BPF_ST
] = BPF_S_ST
+ 1,
426 [BPF_STX
] = BPF_S_STX
+ 1,
427 [BPF_JMP
|BPF_JA
] = BPF_S_JMP_JA
+ 1,
428 [BPF_JMP
|BPF_JEQ
|BPF_K
] = BPF_S_JMP_JEQ_K
+ 1,
429 [BPF_JMP
|BPF_JEQ
|BPF_X
] = BPF_S_JMP_JEQ_X
+ 1,
430 [BPF_JMP
|BPF_JGE
|BPF_K
] = BPF_S_JMP_JGE_K
+ 1,
431 [BPF_JMP
|BPF_JGE
|BPF_X
] = BPF_S_JMP_JGE_X
+ 1,
432 [BPF_JMP
|BPF_JGT
|BPF_K
] = BPF_S_JMP_JGT_K
+ 1,
433 [BPF_JMP
|BPF_JGT
|BPF_X
] = BPF_S_JMP_JGT_X
+ 1,
434 [BPF_JMP
|BPF_JSET
|BPF_K
] = BPF_S_JMP_JSET_K
+ 1,
435 [BPF_JMP
|BPF_JSET
|BPF_X
] = BPF_S_JMP_JSET_X
+ 1,
439 if (flen
== 0 || flen
> BPF_MAXINSNS
)
442 /* check the filter code now */
443 for (pc
= 0; pc
< flen
; pc
++) {
444 struct sock_filter
*ftest
= &filter
[pc
];
445 u16 code
= ftest
->code
;
447 if (code
>= ARRAY_SIZE(codes
))
450 /* Undo the '+ 1' in codes[] after validation. */
453 /* Some instructions need special checks */
455 case BPF_S_ALU_DIV_K
:
456 /* check for division by zero */
464 /* check for invalid memory addresses */
465 if (ftest
->k
>= BPF_MEMWORDS
)
470 * Note, the large ftest->k might cause loops.
471 * Compare this with conditional jumps below,
472 * where offsets are limited. --ANK (981016)
474 if (ftest
->k
>= (unsigned)(flen
-pc
-1))
477 case BPF_S_JMP_JEQ_K
:
478 case BPF_S_JMP_JEQ_X
:
479 case BPF_S_JMP_JGE_K
:
480 case BPF_S_JMP_JGE_X
:
481 case BPF_S_JMP_JGT_K
:
482 case BPF_S_JMP_JGT_X
:
483 case BPF_S_JMP_JSET_X
:
484 case BPF_S_JMP_JSET_K
:
485 /* for conditionals both must be safe */
486 if (pc
+ ftest
->jt
+ 1 >= flen
||
487 pc
+ ftest
->jf
+ 1 >= flen
)
494 /* last instruction must be a RET code */
495 switch (filter
[flen
- 1].code
) {
502 EXPORT_SYMBOL(sk_chk_filter
);
505 * sk_filter_rcu_release: Release a socket filter by rcu_head
506 * @rcu: rcu_head that contains the sk_filter to free
508 static void sk_filter_rcu_release(struct rcu_head
*rcu
)
510 struct sk_filter
*fp
= container_of(rcu
, struct sk_filter
, rcu
);
512 sk_filter_release(fp
);
515 static void sk_filter_delayed_uncharge(struct sock
*sk
, struct sk_filter
*fp
)
517 unsigned int size
= sk_filter_len(fp
);
519 atomic_sub(size
, &sk
->sk_omem_alloc
);
520 call_rcu_bh(&fp
->rcu
, sk_filter_rcu_release
);
524 * sk_attach_filter - attach a socket filter
525 * @fprog: the filter program
526 * @sk: the socket to use
528 * Attach the user's filter code. We first run some sanity checks on
529 * it to make sure it does not explode on us later. If an error
530 * occurs or there is insufficient memory for the filter a negative
531 * errno code is returned. On success the return is zero.
533 int sk_attach_filter(struct sock_fprog
*fprog
, struct sock
*sk
)
535 struct sk_filter
*fp
, *old_fp
;
536 unsigned int fsize
= sizeof(struct sock_filter
) * fprog
->len
;
539 /* Make sure new filter is there and in the right amounts. */
540 if (fprog
->filter
== NULL
)
543 fp
= sock_kmalloc(sk
, fsize
+sizeof(*fp
), GFP_KERNEL
);
546 if (copy_from_user(fp
->insns
, fprog
->filter
, fsize
)) {
547 sock_kfree_s(sk
, fp
, fsize
+sizeof(*fp
));
551 atomic_set(&fp
->refcnt
, 1);
552 fp
->len
= fprog
->len
;
554 err
= sk_chk_filter(fp
->insns
, fp
->len
);
556 sk_filter_uncharge(sk
, fp
);
560 old_fp
= rcu_dereference_protected(sk
->sk_filter
,
561 sock_owned_by_user(sk
));
562 rcu_assign_pointer(sk
->sk_filter
, fp
);
565 sk_filter_delayed_uncharge(sk
, old_fp
);
568 EXPORT_SYMBOL_GPL(sk_attach_filter
);
570 int sk_detach_filter(struct sock
*sk
)
573 struct sk_filter
*filter
;
575 filter
= rcu_dereference_protected(sk
->sk_filter
,
576 sock_owned_by_user(sk
));
578 rcu_assign_pointer(sk
->sk_filter
, NULL
);
579 sk_filter_delayed_uncharge(sk
, filter
);
584 EXPORT_SYMBOL_GPL(sk_detach_filter
);