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>
40 #include <linux/reciprocal_div.h>
90 /* No hurry in this branch */
91 static void *__load_pointer(struct sk_buff *skb, int k)
96 ptr = skb_network_header(skb) + k - SKF_NET_OFF;
97 else if (k >= SKF_LL_OFF)
98 ptr = skb_mac_header(skb) + k - SKF_LL_OFF;
100 if (ptr >= skb->head && ptr < skb_tail_pointer(skb))
105 static inline void *load_pointer(struct sk_buff *skb, int k,
106 unsigned int size, void *buffer)
109 return skb_header_pointer(skb, k, size, buffer);
113 return __load_pointer(skb, k);
118 * sk_filter - run a packet through a socket filter
119 * @sk: sock associated with &sk_buff
120 * @skb: buffer to filter
122 * Run the filter code and then cut skb->data to correct size returned by
123 * sk_run_filter. If pkt_len is 0 we toss packet. If skb->len is smaller
124 * than pkt_len we keep whole skb->data. This is the socket level
125 * wrapper to sk_run_filter. It returns 0 if the packet should
126 * be accepted or -EPERM if the packet should be tossed.
129 int sk_filter(struct sock *sk, struct sk_buff *skb)
132 struct sk_filter *filter;
134 err = security_sock_rcv_skb(sk, skb);
139 filter = rcu_dereference_bh(sk->sk_filter);
141 unsigned int pkt_len = sk_run_filter(skb, filter->insns);
143 err = pkt_len ? pskb_trim(skb, pkt_len) : -EPERM;
145 rcu_read_unlock_bh();
149 EXPORT_SYMBOL(sk_filter);
152 * sk_run_filter - run a filter on a socket
153 * @skb: buffer to run the filter on
154 * @filter: filter to apply
156 * Decode and apply filter instructions to the skb->data.
157 * Return length to keep, 0 for none. @skb is the data we are
158 * filtering, @filter is the array of filter instructions.
159 * Because all jumps are guaranteed to be before last instruction,
160 * and last instruction guaranteed to be a RET, we dont need to check
161 * flen. (We used to pass to this function the length of filter)
163 unsigned int sk_run_filter(struct sk_buff *skb, const struct sock_filter *fentry)
166 u32 A = 0; /* Accumulator */
167 u32 X = 0; /* Index Register */
168 u32 mem[BPF_MEMWORDS]; /* Scratch Memory Store */
169 unsigned long memvalid = 0;
173 BUILD_BUG_ON(BPF_MEMWORDS > BITS_PER_LONG);
175 * Process array of filter instructions.
178 #if defined(CONFIG_X86_32)
179 #define K (fentry->k)
181 const u32 K = fentry->k;
184 switch (fentry->code) {
185 case BPF_S_ALU_ADD_X:
188 case BPF_S_ALU_ADD_K:
191 case BPF_S_ALU_SUB_X:
194 case BPF_S_ALU_SUB_K:
197 case BPF_S_ALU_MUL_X:
200 case BPF_S_ALU_MUL_K:
203 case BPF_S_ALU_DIV_X:
208 case BPF_S_ALU_DIV_K:
209 A = reciprocal_divide(A, K);
211 case BPF_S_ALU_AND_X:
214 case BPF_S_ALU_AND_K:
223 case BPF_S_ALU_LSH_X:
226 case BPF_S_ALU_LSH_K:
229 case BPF_S_ALU_RSH_X:
232 case BPF_S_ALU_RSH_K:
241 case BPF_S_JMP_JGT_K:
242 fentry += (A > K) ? fentry->jt : fentry->jf;
244 case BPF_S_JMP_JGE_K:
245 fentry += (A >= K) ? fentry->jt : fentry->jf;
247 case BPF_S_JMP_JEQ_K:
248 fentry += (A == K) ? fentry->jt : fentry->jf;
250 case BPF_S_JMP_JSET_K:
251 fentry += (A & K) ? fentry->jt : fentry->jf;
253 case BPF_S_JMP_JGT_X:
254 fentry += (A > X) ? fentry->jt : fentry->jf;
256 case BPF_S_JMP_JGE_X:
257 fentry += (A >= X) ? fentry->jt : fentry->jf;
259 case BPF_S_JMP_JEQ_X:
260 fentry += (A == X) ? fentry->jt : fentry->jf;
262 case BPF_S_JMP_JSET_X:
263 fentry += (A & X) ? fentry->jt : fentry->jf;
268 ptr = load_pointer(skb, k, 4, &tmp);
270 A = get_unaligned_be32(ptr);
277 ptr = load_pointer(skb, k, 2, &tmp);
279 A = get_unaligned_be16(ptr);
286 ptr = load_pointer(skb, k, 1, &tmp);
295 case BPF_S_LDX_W_LEN:
307 case BPF_S_LDX_B_MSH:
308 ptr = load_pointer(skb, K, 1, &tmp);
310 X = (*(u8 *)ptr & 0xf) << 2;
321 A = (memvalid & (1UL << K)) ?
325 X = (memvalid & (1UL << K)) ?
339 memvalid |= 1UL << K;
343 memvalid |= 1UL << K;
352 * Handle ancillary data, which are impossible
353 * (or very difficult) to get parsing packet contents.
355 switch (k-SKF_AD_OFF) {
356 case SKF_AD_PROTOCOL:
357 A = ntohs(skb->protocol);
365 A = skb->dev->ifindex;
371 A = skb->queue_mapping;
378 case SKF_AD_NLATTR: {
381 if (skb_is_nonlinear(skb))
383 if (A > skb->len - sizeof(struct nlattr))
386 nla = nla_find((struct nlattr *)&skb->data[A],
389 A = (void *)nla - (void *)skb->data;
394 case SKF_AD_NLATTR_NEST: {
397 if (skb_is_nonlinear(skb))
399 if (A > skb->len - sizeof(struct nlattr))
402 nla = (struct nlattr *)&skb->data[A];
403 if (nla->nla_len > A - skb->len)
406 nla = nla_find_nested(nla, X);
408 A = (void *)nla - (void *)skb->data;
420 EXPORT_SYMBOL(sk_run_filter);
423 * sk_chk_filter - verify socket filter code
424 * @filter: filter to verify
425 * @flen: length of filter
427 * Check the user's filter code. If we let some ugly
428 * filter code slip through kaboom! The filter must contain
429 * no references or jumps that are out of range, no illegal
430 * instructions, and must end with a RET instruction.
432 * All jumps are forward as they are not signed.
434 * Returns 0 if the rule set is legal or -EINVAL if not.
436 int sk_chk_filter(struct sock_filter *filter, int flen)
439 * Valid instructions are initialized to non-0.
440 * Invalid instructions are initialized to 0.
442 static const u8 codes[] = {
443 [BPF_ALU|BPF_ADD|BPF_K] = BPF_S_ALU_ADD_K,
444 [BPF_ALU|BPF_ADD|BPF_X] = BPF_S_ALU_ADD_X,
445 [BPF_ALU|BPF_SUB|BPF_K] = BPF_S_ALU_SUB_K,
446 [BPF_ALU|BPF_SUB|BPF_X] = BPF_S_ALU_SUB_X,
447 [BPF_ALU|BPF_MUL|BPF_K] = BPF_S_ALU_MUL_K,
448 [BPF_ALU|BPF_MUL|BPF_X] = BPF_S_ALU_MUL_X,
449 [BPF_ALU|BPF_DIV|BPF_X] = BPF_S_ALU_DIV_X,
450 [BPF_ALU|BPF_AND|BPF_K] = BPF_S_ALU_AND_K,
451 [BPF_ALU|BPF_AND|BPF_X] = BPF_S_ALU_AND_X,
452 [BPF_ALU|BPF_OR|BPF_K] = BPF_S_ALU_OR_K,
453 [BPF_ALU|BPF_OR|BPF_X] = BPF_S_ALU_OR_X,
454 [BPF_ALU|BPF_LSH|BPF_K] = BPF_S_ALU_LSH_K,
455 [BPF_ALU|BPF_LSH|BPF_X] = BPF_S_ALU_LSH_X,
456 [BPF_ALU|BPF_RSH|BPF_K] = BPF_S_ALU_RSH_K,
457 [BPF_ALU|BPF_RSH|BPF_X] = BPF_S_ALU_RSH_X,
458 [BPF_ALU|BPF_NEG] = BPF_S_ALU_NEG,
459 [BPF_LD|BPF_W|BPF_ABS] = BPF_S_LD_W_ABS,
460 [BPF_LD|BPF_H|BPF_ABS] = BPF_S_LD_H_ABS,
461 [BPF_LD|BPF_B|BPF_ABS] = BPF_S_LD_B_ABS,
462 [BPF_LD|BPF_W|BPF_LEN] = BPF_S_LD_W_LEN,
463 [BPF_LD|BPF_W|BPF_IND] = BPF_S_LD_W_IND,
464 [BPF_LD|BPF_H|BPF_IND] = BPF_S_LD_H_IND,
465 [BPF_LD|BPF_B|BPF_IND] = BPF_S_LD_B_IND,
466 [BPF_LD|BPF_IMM] = BPF_S_LD_IMM,
467 [BPF_LDX|BPF_W|BPF_LEN] = BPF_S_LDX_W_LEN,
468 [BPF_LDX|BPF_B|BPF_MSH] = BPF_S_LDX_B_MSH,
469 [BPF_LDX|BPF_IMM] = BPF_S_LDX_IMM,
470 [BPF_MISC|BPF_TAX] = BPF_S_MISC_TAX,
471 [BPF_MISC|BPF_TXA] = BPF_S_MISC_TXA,
472 [BPF_RET|BPF_K] = BPF_S_RET_K,
473 [BPF_RET|BPF_A] = BPF_S_RET_A,
474 [BPF_ALU|BPF_DIV|BPF_K] = BPF_S_ALU_DIV_K,
475 [BPF_LD|BPF_MEM] = BPF_S_LD_MEM,
476 [BPF_LDX|BPF_MEM] = BPF_S_LDX_MEM,
478 [BPF_STX] = BPF_S_STX,
479 [BPF_JMP|BPF_JA] = BPF_S_JMP_JA,
480 [BPF_JMP|BPF_JEQ|BPF_K] = BPF_S_JMP_JEQ_K,
481 [BPF_JMP|BPF_JEQ|BPF_X] = BPF_S_JMP_JEQ_X,
482 [BPF_JMP|BPF_JGE|BPF_K] = BPF_S_JMP_JGE_K,
483 [BPF_JMP|BPF_JGE|BPF_X] = BPF_S_JMP_JGE_X,
484 [BPF_JMP|BPF_JGT|BPF_K] = BPF_S_JMP_JGT_K,
485 [BPF_JMP|BPF_JGT|BPF_X] = BPF_S_JMP_JGT_X,
486 [BPF_JMP|BPF_JSET|BPF_K] = BPF_S_JMP_JSET_K,
487 [BPF_JMP|BPF_JSET|BPF_X] = BPF_S_JMP_JSET_X,
491 if (flen == 0 || flen > BPF_MAXINSNS)
494 /* check the filter code now */
495 for (pc = 0; pc < flen; pc++) {
496 struct sock_filter *ftest = &filter[pc];
497 u16 code = ftest->code;
499 if (code >= ARRAY_SIZE(codes))
504 /* Some instructions need special checks */
506 case BPF_S_ALU_DIV_K:
507 /* check for division by zero */
510 ftest->k = reciprocal_value(ftest->k);
516 /* check for invalid memory addresses */
517 if (ftest->k >= BPF_MEMWORDS)
522 * Note, the large ftest->k might cause loops.
523 * Compare this with conditional jumps below,
524 * where offsets are limited. --ANK (981016)
526 if (ftest->k >= (unsigned)(flen-pc-1))
529 case BPF_S_JMP_JEQ_K:
530 case BPF_S_JMP_JEQ_X:
531 case BPF_S_JMP_JGE_K:
532 case BPF_S_JMP_JGE_X:
533 case BPF_S_JMP_JGT_K:
534 case BPF_S_JMP_JGT_X:
535 case BPF_S_JMP_JSET_X:
536 case BPF_S_JMP_JSET_K:
537 /* for conditionals both must be safe */
538 if (pc + ftest->jt + 1 >= flen ||
539 pc + ftest->jf + 1 >= flen)
546 /* last instruction must be a RET code */
547 switch (filter[flen - 1].code) {
554 EXPORT_SYMBOL(sk_chk_filter);
557 * sk_filter_rcu_release - Release a socket filter by rcu_head
558 * @rcu: rcu_head that contains the sk_filter to free
560 static void sk_filter_rcu_release(struct rcu_head *rcu)
562 struct sk_filter *fp = container_of(rcu, struct sk_filter, rcu);
564 sk_filter_release(fp);
567 static void sk_filter_delayed_uncharge(struct sock *sk, struct sk_filter *fp)
569 unsigned int size = sk_filter_len(fp);
571 atomic_sub(size, &sk->sk_omem_alloc);
572 call_rcu_bh(&fp->rcu, sk_filter_rcu_release);
576 * sk_attach_filter - attach a socket filter
577 * @fprog: the filter program
578 * @sk: the socket to use
580 * Attach the user's filter code. We first run some sanity checks on
581 * it to make sure it does not explode on us later. If an error
582 * occurs or there is insufficient memory for the filter a negative
583 * errno code is returned. On success the return is zero.
585 int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk)
587 struct sk_filter *fp, *old_fp;
588 unsigned int fsize = sizeof(struct sock_filter) * fprog->len;
591 /* Make sure new filter is there and in the right amounts. */
592 if (fprog->filter == NULL)
595 fp = sock_kmalloc(sk, fsize+sizeof(*fp), GFP_KERNEL);
598 if (copy_from_user(fp->insns, fprog->filter, fsize)) {
599 sock_kfree_s(sk, fp, fsize+sizeof(*fp));
603 atomic_set(&fp->refcnt, 1);
604 fp->len = fprog->len;
606 err = sk_chk_filter(fp->insns, fp->len);
608 sk_filter_uncharge(sk, fp);
612 old_fp = rcu_dereference_protected(sk->sk_filter,
613 sock_owned_by_user(sk));
614 rcu_assign_pointer(sk->sk_filter, fp);
617 sk_filter_delayed_uncharge(sk, old_fp);
620 EXPORT_SYMBOL_GPL(sk_attach_filter);
622 int sk_detach_filter(struct sock *sk)
625 struct sk_filter *filter;
627 filter = rcu_dereference_protected(sk->sk_filter,
628 sock_owned_by_user(sk));
630 rcu_assign_pointer(sk->sk_filter, NULL);
631 sk_filter_delayed_uncharge(sk, filter);
636 EXPORT_SYMBOL_GPL(sk_detach_filter);