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 <linux/skbuff.h>
32 #include <linux/errno.h>
33 #include <linux/timer.h>
34 #include <asm/system.h>
35 #include <asm/uaccess.h>
36 #include <asm/unaligned.h>
37 #include <linux/filter.h>
39 /* No hurry in this branch */
40 static void *__load_pointer(struct sk_buff
*skb
, int k
)
45 ptr
= skb_network_header(skb
) + k
- SKF_NET_OFF
;
46 else if (k
>= SKF_LL_OFF
)
47 ptr
= skb_mac_header(skb
) + k
- SKF_LL_OFF
;
49 if (ptr
>= skb
->head
&& ptr
< skb_tail_pointer(skb
))
54 static inline void *load_pointer(struct sk_buff
*skb
, int k
,
55 unsigned int size
, void *buffer
)
58 return skb_header_pointer(skb
, k
, size
, buffer
);
62 return __load_pointer(skb
, k
);
67 * sk_run_filter - run a filter on a socket
68 * @skb: buffer to run the filter on
69 * @filter: filter to apply
70 * @flen: length of filter
72 * Decode and apply filter instructions to the skb->data.
73 * Return length to keep, 0 for none. skb is the data we are
74 * filtering, filter is the array of filter instructions, and
75 * len is the number of filter blocks in the array.
77 unsigned int sk_run_filter(struct sk_buff
*skb
, struct sock_filter
*filter
, int flen
)
79 struct sock_filter
*fentry
; /* We walk down these */
81 u32 A
= 0; /* Accumulator */
82 u32 X
= 0; /* Index Register */
83 u32 mem
[BPF_MEMWORDS
]; /* Scratch Memory Store */
89 * Process array of filter instructions.
91 for (pc
= 0; pc
< flen
; pc
++) {
94 switch (fentry
->code
) {
101 case BPF_S_ALU_SUB_X
:
104 case BPF_S_ALU_SUB_K
:
107 case BPF_S_ALU_MUL_X
:
110 case BPF_S_ALU_MUL_K
:
113 case BPF_S_ALU_DIV_X
:
118 case BPF_S_ALU_DIV_K
:
121 case BPF_S_ALU_AND_X
:
124 case BPF_S_ALU_AND_K
:
133 case BPF_S_ALU_LSH_X
:
136 case BPF_S_ALU_LSH_K
:
139 case BPF_S_ALU_RSH_X
:
142 case BPF_S_ALU_RSH_K
:
151 case BPF_S_JMP_JGT_K
:
152 pc
+= (A
> fentry
->k
) ? fentry
->jt
: fentry
->jf
;
154 case BPF_S_JMP_JGE_K
:
155 pc
+= (A
>= fentry
->k
) ? fentry
->jt
: fentry
->jf
;
157 case BPF_S_JMP_JEQ_K
:
158 pc
+= (A
== fentry
->k
) ? fentry
->jt
: fentry
->jf
;
160 case BPF_S_JMP_JSET_K
:
161 pc
+= (A
& fentry
->k
) ? fentry
->jt
: fentry
->jf
;
163 case BPF_S_JMP_JGT_X
:
164 pc
+= (A
> X
) ? fentry
->jt
: fentry
->jf
;
166 case BPF_S_JMP_JGE_X
:
167 pc
+= (A
>= X
) ? fentry
->jt
: fentry
->jf
;
169 case BPF_S_JMP_JEQ_X
:
170 pc
+= (A
== X
) ? fentry
->jt
: fentry
->jf
;
172 case BPF_S_JMP_JSET_X
:
173 pc
+= (A
& X
) ? fentry
->jt
: fentry
->jf
;
178 ptr
= load_pointer(skb
, k
, 4, &tmp
);
180 A
= ntohl(get_unaligned((__be32
*)ptr
));
187 ptr
= load_pointer(skb
, k
, 2, &tmp
);
189 A
= ntohs(get_unaligned((__be16
*)ptr
));
196 ptr
= load_pointer(skb
, k
, 1, &tmp
);
205 case BPF_S_LDX_W_LEN
:
217 case BPF_S_LDX_B_MSH
:
218 ptr
= load_pointer(skb
, fentry
->k
, 1, &tmp
);
220 X
= (*(u8
*)ptr
& 0xf) << 2;
258 * Handle ancillary data, which are impossible
259 * (or very difficult) to get parsing packet contents.
261 switch (k
-SKF_AD_OFF
) {
262 case SKF_AD_PROTOCOL
:
263 A
= ntohs(skb
->protocol
);
269 A
= skb
->dev
->ifindex
;
280 * sk_chk_filter - verify socket filter code
281 * @filter: filter to verify
282 * @flen: length of filter
284 * Check the user's filter code. If we let some ugly
285 * filter code slip through kaboom! The filter must contain
286 * no references or jumps that are out of range, no illegal
287 * instructions, and must end with a RET instruction.
289 * All jumps are forward as they are not signed.
291 * Returns 0 if the rule set is legal or -EINVAL if not.
293 int sk_chk_filter(struct sock_filter
*filter
, int flen
)
295 struct sock_filter
*ftest
;
298 if (flen
== 0 || flen
> BPF_MAXINSNS
)
301 /* check the filter code now */
302 for (pc
= 0; pc
< flen
; pc
++) {
305 /* Only allow valid instructions */
306 switch (ftest
->code
) {
307 case BPF_ALU
|BPF_ADD
|BPF_K
:
308 ftest
->code
= BPF_S_ALU_ADD_K
;
310 case BPF_ALU
|BPF_ADD
|BPF_X
:
311 ftest
->code
= BPF_S_ALU_ADD_X
;
313 case BPF_ALU
|BPF_SUB
|BPF_K
:
314 ftest
->code
= BPF_S_ALU_SUB_K
;
316 case BPF_ALU
|BPF_SUB
|BPF_X
:
317 ftest
->code
= BPF_S_ALU_SUB_X
;
319 case BPF_ALU
|BPF_MUL
|BPF_K
:
320 ftest
->code
= BPF_S_ALU_MUL_K
;
322 case BPF_ALU
|BPF_MUL
|BPF_X
:
323 ftest
->code
= BPF_S_ALU_MUL_X
;
325 case BPF_ALU
|BPF_DIV
|BPF_X
:
326 ftest
->code
= BPF_S_ALU_DIV_X
;
328 case BPF_ALU
|BPF_AND
|BPF_K
:
329 ftest
->code
= BPF_S_ALU_AND_K
;
331 case BPF_ALU
|BPF_AND
|BPF_X
:
332 ftest
->code
= BPF_S_ALU_AND_X
;
334 case BPF_ALU
|BPF_OR
|BPF_K
:
335 ftest
->code
= BPF_S_ALU_OR_K
;
337 case BPF_ALU
|BPF_OR
|BPF_X
:
338 ftest
->code
= BPF_S_ALU_OR_X
;
340 case BPF_ALU
|BPF_LSH
|BPF_K
:
341 ftest
->code
= BPF_S_ALU_LSH_K
;
343 case BPF_ALU
|BPF_LSH
|BPF_X
:
344 ftest
->code
= BPF_S_ALU_LSH_X
;
346 case BPF_ALU
|BPF_RSH
|BPF_K
:
347 ftest
->code
= BPF_S_ALU_RSH_K
;
349 case BPF_ALU
|BPF_RSH
|BPF_X
:
350 ftest
->code
= BPF_S_ALU_RSH_X
;
352 case BPF_ALU
|BPF_NEG
:
353 ftest
->code
= BPF_S_ALU_NEG
;
355 case BPF_LD
|BPF_W
|BPF_ABS
:
356 ftest
->code
= BPF_S_LD_W_ABS
;
358 case BPF_LD
|BPF_H
|BPF_ABS
:
359 ftest
->code
= BPF_S_LD_H_ABS
;
361 case BPF_LD
|BPF_B
|BPF_ABS
:
362 ftest
->code
= BPF_S_LD_B_ABS
;
364 case BPF_LD
|BPF_W
|BPF_LEN
:
365 ftest
->code
= BPF_S_LD_W_LEN
;
367 case BPF_LD
|BPF_W
|BPF_IND
:
368 ftest
->code
= BPF_S_LD_W_IND
;
370 case BPF_LD
|BPF_H
|BPF_IND
:
371 ftest
->code
= BPF_S_LD_H_IND
;
373 case BPF_LD
|BPF_B
|BPF_IND
:
374 ftest
->code
= BPF_S_LD_B_IND
;
377 ftest
->code
= BPF_S_LD_IMM
;
379 case BPF_LDX
|BPF_W
|BPF_LEN
:
380 ftest
->code
= BPF_S_LDX_W_LEN
;
382 case BPF_LDX
|BPF_B
|BPF_MSH
:
383 ftest
->code
= BPF_S_LDX_B_MSH
;
385 case BPF_LDX
|BPF_IMM
:
386 ftest
->code
= BPF_S_LDX_IMM
;
388 case BPF_MISC
|BPF_TAX
:
389 ftest
->code
= BPF_S_MISC_TAX
;
391 case BPF_MISC
|BPF_TXA
:
392 ftest
->code
= BPF_S_MISC_TXA
;
395 ftest
->code
= BPF_S_RET_K
;
398 ftest
->code
= BPF_S_RET_A
;
401 /* Some instructions need special checks */
403 /* check for division by zero */
404 case BPF_ALU
|BPF_DIV
|BPF_K
:
407 ftest
->code
= BPF_S_ALU_DIV_K
;
410 /* check for invalid memory addresses */
412 if (ftest
->k
>= BPF_MEMWORDS
)
414 ftest
->code
= BPF_S_LD_MEM
;
416 case BPF_LDX
|BPF_MEM
:
417 if (ftest
->k
>= BPF_MEMWORDS
)
419 ftest
->code
= BPF_S_LDX_MEM
;
422 if (ftest
->k
>= BPF_MEMWORDS
)
424 ftest
->code
= BPF_S_ST
;
427 if (ftest
->k
>= BPF_MEMWORDS
)
429 ftest
->code
= BPF_S_STX
;
434 * Note, the large ftest->k might cause loops.
435 * Compare this with conditional jumps below,
436 * where offsets are limited. --ANK (981016)
438 if (ftest
->k
>= (unsigned)(flen
-pc
-1))
440 ftest
->code
= BPF_S_JMP_JA
;
443 case BPF_JMP
|BPF_JEQ
|BPF_K
:
444 ftest
->code
= BPF_S_JMP_JEQ_K
;
446 case BPF_JMP
|BPF_JEQ
|BPF_X
:
447 ftest
->code
= BPF_S_JMP_JEQ_X
;
449 case BPF_JMP
|BPF_JGE
|BPF_K
:
450 ftest
->code
= BPF_S_JMP_JGE_K
;
452 case BPF_JMP
|BPF_JGE
|BPF_X
:
453 ftest
->code
= BPF_S_JMP_JGE_X
;
455 case BPF_JMP
|BPF_JGT
|BPF_K
:
456 ftest
->code
= BPF_S_JMP_JGT_K
;
458 case BPF_JMP
|BPF_JGT
|BPF_X
:
459 ftest
->code
= BPF_S_JMP_JGT_X
;
461 case BPF_JMP
|BPF_JSET
|BPF_K
:
462 ftest
->code
= BPF_S_JMP_JSET_K
;
464 case BPF_JMP
|BPF_JSET
|BPF_X
:
465 ftest
->code
= BPF_S_JMP_JSET_X
;
472 /* for conditionals both must be safe */
473 switch (ftest
->code
) {
474 case BPF_S_JMP_JEQ_K
:
475 case BPF_S_JMP_JEQ_X
:
476 case BPF_S_JMP_JGE_K
:
477 case BPF_S_JMP_JGE_X
:
478 case BPF_S_JMP_JGT_K
:
479 case BPF_S_JMP_JGT_X
:
480 case BPF_S_JMP_JSET_X
:
481 case BPF_S_JMP_JSET_K
:
482 if (pc
+ ftest
->jt
+ 1 >= flen
||
483 pc
+ ftest
->jf
+ 1 >= flen
)
488 /* last instruction must be a RET code */
489 switch (filter
[flen
- 1].code
) {
500 * sk_attach_filter - attach a socket filter
501 * @fprog: the filter program
502 * @sk: the socket to use
504 * Attach the user's filter code. We first run some sanity checks on
505 * it to make sure it does not explode on us later. If an error
506 * occurs or there is insufficient memory for the filter a negative
507 * errno code is returned. On success the return is zero.
509 int sk_attach_filter(struct sock_fprog
*fprog
, struct sock
*sk
)
511 struct sk_filter
*fp
;
512 unsigned int fsize
= sizeof(struct sock_filter
) * fprog
->len
;
515 /* Make sure new filter is there and in the right amounts. */
516 if (fprog
->filter
== NULL
)
519 fp
= sock_kmalloc(sk
, fsize
+sizeof(*fp
), GFP_KERNEL
);
522 if (copy_from_user(fp
->insns
, fprog
->filter
, fsize
)) {
523 sock_kfree_s(sk
, fp
, fsize
+sizeof(*fp
));
527 atomic_set(&fp
->refcnt
, 1);
528 fp
->len
= fprog
->len
;
530 err
= sk_chk_filter(fp
->insns
, fp
->len
);
532 struct sk_filter
*old_fp
;
535 old_fp
= rcu_dereference(sk
->sk_filter
);
536 rcu_assign_pointer(sk
->sk_filter
, fp
);
537 rcu_read_unlock_bh();
542 sk_filter_release(sk
, fp
);
546 EXPORT_SYMBOL(sk_chk_filter
);
547 EXPORT_SYMBOL(sk_run_filter
);