2 * IPv6 fragment reassembly for connection tracking
4 * Copyright (C)2004 USAGI/WIDE Project
7 * Yasuyuki Kozakai @USAGI <yasuyuki.kozakai@toshiba.co.jp>
9 * Based on: net/ipv6/reassembly.c
11 * This program is free software; you can redistribute it and/or
12 * modify it under the terms of the GNU General Public License
13 * as published by the Free Software Foundation; either version
14 * 2 of the License, or (at your option) any later version.
17 #include <linux/errno.h>
18 #include <linux/types.h>
19 #include <linux/string.h>
20 #include <linux/socket.h>
21 #include <linux/sockios.h>
22 #include <linux/jiffies.h>
23 #include <linux/net.h>
24 #include <linux/list.h>
25 #include <linux/netdevice.h>
26 #include <linux/in6.h>
27 #include <linux/ipv6.h>
28 #include <linux/icmpv6.h>
29 #include <linux/random.h>
33 #include <net/inet_frag.h>
36 #include <net/protocol.h>
37 #include <net/transp_v6.h>
38 #include <net/rawv6.h>
39 #include <net/ndisc.h>
40 #include <net/addrconf.h>
41 #include <net/netfilter/ipv6/nf_conntrack_ipv6.h>
42 #include <linux/sysctl.h>
43 #include <linux/netfilter.h>
44 #include <linux/netfilter_ipv6.h>
45 #include <linux/kernel.h>
46 #include <linux/module.h>
49 struct nf_ct_frag6_skb_cb
51 struct inet6_skb_parm h;
56 #define NFCT_FRAG6_CB(skb) ((struct nf_ct_frag6_skb_cb*)((skb)->cb))
58 struct nf_ct_frag6_queue
60 struct inet_frag_queue q;
62 __be32 id; /* fragment id */
63 struct in6_addr saddr;
64 struct in6_addr daddr;
70 static struct inet_frags nf_frags;
71 static struct netns_frags nf_init_frags;
74 struct ctl_table nf_ct_ipv6_sysctl_table[] = {
76 .procname = "nf_conntrack_frag6_timeout",
77 .data = &nf_init_frags.timeout,
78 .maxlen = sizeof(unsigned int),
80 .proc_handler = proc_dointvec_jiffies,
83 .procname = "nf_conntrack_frag6_low_thresh",
84 .data = &nf_init_frags.low_thresh,
85 .maxlen = sizeof(unsigned int),
87 .proc_handler = proc_dointvec,
90 .procname = "nf_conntrack_frag6_high_thresh",
91 .data = &nf_init_frags.high_thresh,
92 .maxlen = sizeof(unsigned int),
94 .proc_handler = proc_dointvec,
100 static unsigned int nf_hashfn(struct inet_frag_queue *q)
102 const struct nf_ct_frag6_queue *nq;
104 nq = container_of(q, struct nf_ct_frag6_queue, q);
105 return inet6_hash_frag(nq->id, &nq->saddr, &nq->daddr, nf_frags.rnd);
108 static void nf_skb_free(struct sk_buff *skb)
110 if (NFCT_FRAG6_CB(skb)->orig)
111 kfree_skb(NFCT_FRAG6_CB(skb)->orig);
114 /* Memory Tracking Functions. */
115 static inline void frag_kfree_skb(struct sk_buff *skb, unsigned int *work)
118 *work -= skb->truesize;
119 atomic_sub(skb->truesize, &nf_init_frags.mem);
124 /* Destruction primitives. */
126 static __inline__ void fq_put(struct nf_ct_frag6_queue *fq)
128 inet_frag_put(&fq->q, &nf_frags);
131 /* Kill fq entry. It is not destroyed immediately,
132 * because caller (and someone more) holds reference count.
134 static __inline__ void fq_kill(struct nf_ct_frag6_queue *fq)
136 inet_frag_kill(&fq->q, &nf_frags);
139 static void nf_ct_frag6_evictor(void)
142 inet_frag_evictor(&nf_init_frags, &nf_frags);
146 static void nf_ct_frag6_expire(unsigned long data)
148 struct nf_ct_frag6_queue *fq;
150 fq = container_of((struct inet_frag_queue *)data,
151 struct nf_ct_frag6_queue, q);
153 spin_lock(&fq->q.lock);
155 if (fq->q.last_in & INET_FRAG_COMPLETE)
161 spin_unlock(&fq->q.lock);
165 /* Creation primitives. */
167 static __inline__ struct nf_ct_frag6_queue *
168 fq_find(__be32 id, u32 user, struct in6_addr *src, struct in6_addr *dst)
170 struct inet_frag_queue *q;
171 struct ip6_create_arg arg;
179 read_lock_bh(&nf_frags.lock);
180 hash = inet6_hash_frag(id, src, dst, nf_frags.rnd);
182 q = inet_frag_find(&nf_init_frags, &nf_frags, &arg, hash);
187 return container_of(q, struct nf_ct_frag6_queue, q);
190 pr_debug("Can't alloc new queue\n");
195 static int nf_ct_frag6_queue(struct nf_ct_frag6_queue *fq, struct sk_buff *skb,
196 const struct frag_hdr *fhdr, int nhoff)
198 struct sk_buff *prev, *next;
201 if (fq->q.last_in & INET_FRAG_COMPLETE) {
202 pr_debug("Allready completed\n");
206 offset = ntohs(fhdr->frag_off) & ~0x7;
207 end = offset + (ntohs(ipv6_hdr(skb)->payload_len) -
208 ((u8 *)(fhdr + 1) - (u8 *)(ipv6_hdr(skb) + 1)));
210 if ((unsigned int)end > IPV6_MAXPLEN) {
211 pr_debug("offset is too large.\n");
215 if (skb->ip_summed == CHECKSUM_COMPLETE) {
216 const unsigned char *nh = skb_network_header(skb);
217 skb->csum = csum_sub(skb->csum,
218 csum_partial(nh, (u8 *)(fhdr + 1) - nh,
222 /* Is this the final fragment? */
223 if (!(fhdr->frag_off & htons(IP6_MF))) {
224 /* If we already have some bits beyond end
225 * or have different end, the segment is corrupted.
227 if (end < fq->q.len ||
228 ((fq->q.last_in & INET_FRAG_LAST_IN) && end != fq->q.len)) {
229 pr_debug("already received last fragment\n");
232 fq->q.last_in |= INET_FRAG_LAST_IN;
235 /* Check if the fragment is rounded to 8 bytes.
236 * Required by the RFC.
239 /* RFC2460 says always send parameter problem in
242 pr_debug("end of fragment not rounded to 8 bytes.\n");
245 if (end > fq->q.len) {
246 /* Some bits beyond end -> corruption. */
247 if (fq->q.last_in & INET_FRAG_LAST_IN) {
248 pr_debug("last packet already reached.\n");
258 /* Point into the IP datagram 'data' part. */
259 if (!pskb_pull(skb, (u8 *) (fhdr + 1) - skb->data)) {
260 pr_debug("queue: message is too short.\n");
263 if (pskb_trim_rcsum(skb, end - offset)) {
264 pr_debug("Can't trim\n");
268 /* Find out which fragments are in front and at the back of us
269 * in the chain of fragments so far. We must know where to put
270 * this fragment, right?
273 for (next = fq->q.fragments; next != NULL; next = next->next) {
274 if (NFCT_FRAG6_CB(next)->offset >= offset)
279 /* We found where to put this one. Check for overlap with
280 * preceding fragment, and, if needed, align things so that
281 * any overlaps are eliminated.
284 int i = (NFCT_FRAG6_CB(prev)->offset + prev->len) - offset;
289 pr_debug("overlap\n");
292 if (!pskb_pull(skb, i)) {
293 pr_debug("Can't pull\n");
296 if (skb->ip_summed != CHECKSUM_UNNECESSARY)
297 skb->ip_summed = CHECKSUM_NONE;
301 /* Look for overlap with succeeding segments.
302 * If we can merge fragments, do it.
304 while (next && NFCT_FRAG6_CB(next)->offset < end) {
305 /* overlap is 'i' bytes */
306 int i = end - NFCT_FRAG6_CB(next)->offset;
309 /* Eat head of the next overlapped fragment
310 * and leave the loop. The next ones cannot overlap.
312 pr_debug("Eat head of the overlapped parts.: %d", i);
313 if (!pskb_pull(next, i))
317 NFCT_FRAG6_CB(next)->offset += i;
319 if (next->ip_summed != CHECKSUM_UNNECESSARY)
320 next->ip_summed = CHECKSUM_NONE;
323 struct sk_buff *free_it = next;
325 /* Old fragmnet is completely overridden with
333 fq->q.fragments = next;
335 fq->q.meat -= free_it->len;
336 frag_kfree_skb(free_it, NULL);
340 NFCT_FRAG6_CB(skb)->offset = offset;
342 /* Insert this fragment in the chain of fragments. */
347 fq->q.fragments = skb;
350 fq->q.stamp = skb->tstamp;
351 fq->q.meat += skb->len;
352 atomic_add(skb->truesize, &nf_init_frags.mem);
354 /* The first fragment.
355 * nhoffset is obtained from the first fragment, of course.
358 fq->nhoffset = nhoff;
359 fq->q.last_in |= INET_FRAG_FIRST_IN;
361 write_lock(&nf_frags.lock);
362 list_move_tail(&fq->q.lru_list, &nf_init_frags.lru_list);
363 write_unlock(&nf_frags.lock);
371 * Check if this packet is complete.
372 * Returns NULL on failure by any reason, and pointer
373 * to current nexthdr field in reassembled frame.
375 * It is called with locked fq, and caller must check that
376 * queue is eligible for reassembly i.e. it is not COMPLETE,
377 * the last and the first frames arrived and all the bits are here.
379 static struct sk_buff *
380 nf_ct_frag6_reasm(struct nf_ct_frag6_queue *fq, struct net_device *dev)
382 struct sk_buff *fp, *op, *head = fq->q.fragments;
387 WARN_ON(head == NULL);
388 WARN_ON(NFCT_FRAG6_CB(head)->offset != 0);
390 /* Unfragmented part is taken from the first segment. */
391 payload_len = ((head->data - skb_network_header(head)) -
392 sizeof(struct ipv6hdr) + fq->q.len -
393 sizeof(struct frag_hdr));
394 if (payload_len > IPV6_MAXPLEN) {
395 pr_debug("payload len is too large.\n");
399 /* Head of list must not be cloned. */
400 if (skb_cloned(head) && pskb_expand_head(head, 0, 0, GFP_ATOMIC)) {
401 pr_debug("skb is cloned but can't expand head");
405 /* If the first fragment is fragmented itself, we split
406 * it to two chunks: the first with data and paged part
407 * and the second, holding only fragments. */
408 if (skb_has_frags(head)) {
409 struct sk_buff *clone;
412 if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL) {
413 pr_debug("Can't alloc skb\n");
416 clone->next = head->next;
418 skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
419 skb_frag_list_init(head);
420 for (i=0; i<skb_shinfo(head)->nr_frags; i++)
421 plen += skb_shinfo(head)->frags[i].size;
422 clone->len = clone->data_len = head->data_len - plen;
423 head->data_len -= clone->len;
424 head->len -= clone->len;
426 clone->ip_summed = head->ip_summed;
428 NFCT_FRAG6_CB(clone)->orig = NULL;
429 atomic_add(clone->truesize, &nf_init_frags.mem);
432 /* We have to remove fragment header from datagram and to relocate
433 * header in order to calculate ICV correctly. */
434 skb_network_header(head)[fq->nhoffset] = skb_transport_header(head)[0];
435 memmove(head->head + sizeof(struct frag_hdr), head->head,
436 (head->data - head->head) - sizeof(struct frag_hdr));
437 head->mac_header += sizeof(struct frag_hdr);
438 head->network_header += sizeof(struct frag_hdr);
440 skb_shinfo(head)->frag_list = head->next;
441 skb_reset_transport_header(head);
442 skb_push(head, head->data - skb_network_header(head));
443 atomic_sub(head->truesize, &nf_init_frags.mem);
445 for (fp=head->next; fp; fp = fp->next) {
446 head->data_len += fp->len;
447 head->len += fp->len;
448 if (head->ip_summed != fp->ip_summed)
449 head->ip_summed = CHECKSUM_NONE;
450 else if (head->ip_summed == CHECKSUM_COMPLETE)
451 head->csum = csum_add(head->csum, fp->csum);
452 head->truesize += fp->truesize;
453 atomic_sub(fp->truesize, &nf_init_frags.mem);
458 head->tstamp = fq->q.stamp;
459 ipv6_hdr(head)->payload_len = htons(payload_len);
461 /* Yes, and fold redundant checksum back. 8) */
462 if (head->ip_summed == CHECKSUM_COMPLETE)
463 head->csum = csum_partial(skb_network_header(head),
464 skb_network_header_len(head),
467 fq->q.fragments = NULL;
469 /* all original skbs are linked into the NFCT_FRAG6_CB(head).orig */
470 fp = skb_shinfo(head)->frag_list;
471 if (NFCT_FRAG6_CB(fp)->orig == NULL)
472 /* at above code, head skb is divided into two skbs. */
475 op = NFCT_FRAG6_CB(head)->orig;
476 for (; fp; fp = fp->next) {
477 struct sk_buff *orig = NFCT_FRAG6_CB(fp)->orig;
481 NFCT_FRAG6_CB(fp)->orig = NULL;
488 printk(KERN_DEBUG "nf_ct_frag6_reasm: payload len = %d\n", payload_len);
492 printk(KERN_DEBUG "nf_ct_frag6_reasm: no memory for reassembly\n");
498 * find the header just before Fragment Header.
500 * if success return 0 and set ...
501 * (*prevhdrp): the value of "Next Header Field" in the header
502 * just before Fragment Header.
503 * (*prevhoff): the offset of "Next Header Field" in the header
504 * just before Fragment Header.
505 * (*fhoff) : the offset of Fragment Header.
507 * Based on ipv6_skip_hdr() in net/ipv6/exthdr.c
511 find_prev_fhdr(struct sk_buff *skb, u8 *prevhdrp, int *prevhoff, int *fhoff)
513 u8 nexthdr = ipv6_hdr(skb)->nexthdr;
514 const int netoff = skb_network_offset(skb);
515 u8 prev_nhoff = netoff + offsetof(struct ipv6hdr, nexthdr);
516 int start = netoff + sizeof(struct ipv6hdr);
517 int len = skb->len - start;
518 u8 prevhdr = NEXTHDR_IPV6;
520 while (nexthdr != NEXTHDR_FRAGMENT) {
521 struct ipv6_opt_hdr hdr;
524 if (!ipv6_ext_hdr(nexthdr)) {
527 if (nexthdr == NEXTHDR_NONE) {
528 pr_debug("next header is none\n");
531 if (len < (int)sizeof(struct ipv6_opt_hdr)) {
532 pr_debug("too short\n");
535 if (skb_copy_bits(skb, start, &hdr, sizeof(hdr)))
537 if (nexthdr == NEXTHDR_AUTH)
538 hdrlen = (hdr.hdrlen+2)<<2;
540 hdrlen = ipv6_optlen(&hdr);
545 nexthdr = hdr.nexthdr;
554 *prevhoff = prev_nhoff;
560 struct sk_buff *nf_ct_frag6_gather(struct sk_buff *skb, u32 user)
562 struct sk_buff *clone;
563 struct net_device *dev = skb->dev;
564 struct frag_hdr *fhdr;
565 struct nf_ct_frag6_queue *fq;
569 struct sk_buff *ret_skb = NULL;
571 /* Jumbo payload inhibits frag. header */
572 if (ipv6_hdr(skb)->payload_len == 0) {
573 pr_debug("payload len = 0\n");
577 if (find_prev_fhdr(skb, &prevhdr, &nhoff, &fhoff) < 0)
580 clone = skb_clone(skb, GFP_ATOMIC);
582 pr_debug("Can't clone skb\n");
586 NFCT_FRAG6_CB(clone)->orig = skb;
588 if (!pskb_may_pull(clone, fhoff + sizeof(*fhdr))) {
589 pr_debug("message is too short.\n");
593 skb_set_transport_header(clone, fhoff);
594 hdr = ipv6_hdr(clone);
595 fhdr = (struct frag_hdr *)skb_transport_header(clone);
597 if (!(fhdr->frag_off & htons(0xFFF9))) {
598 pr_debug("Invalid fragment offset\n");
599 /* It is not a fragmented frame */
603 if (atomic_read(&nf_init_frags.mem) > nf_init_frags.high_thresh)
604 nf_ct_frag6_evictor();
606 fq = fq_find(fhdr->identification, user, &hdr->saddr, &hdr->daddr);
608 pr_debug("Can't find and can't create new queue\n");
612 spin_lock_bh(&fq->q.lock);
614 if (nf_ct_frag6_queue(fq, clone, fhdr, nhoff) < 0) {
615 spin_unlock_bh(&fq->q.lock);
616 pr_debug("Can't insert skb to queue\n");
621 if (fq->q.last_in == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
622 fq->q.meat == fq->q.len) {
623 ret_skb = nf_ct_frag6_reasm(fq, dev);
625 pr_debug("Can't reassemble fragmented packets\n");
627 spin_unlock_bh(&fq->q.lock);
637 void nf_ct_frag6_output(unsigned int hooknum, struct sk_buff *skb,
638 struct net_device *in, struct net_device *out,
639 int (*okfn)(struct sk_buff *))
641 struct sk_buff *s, *s2;
643 for (s = NFCT_FRAG6_CB(skb)->orig; s;) {
644 nf_conntrack_put_reasm(s->nfct_reasm);
645 nf_conntrack_get_reasm(skb);
651 NF_HOOK_THRESH(PF_INET6, hooknum, s, in, out, okfn,
652 NF_IP6_PRI_CONNTRACK_DEFRAG + 1);
655 nf_conntrack_put_reasm(skb);
658 int nf_ct_frag6_init(void)
660 nf_frags.hashfn = nf_hashfn;
661 nf_frags.constructor = ip6_frag_init;
662 nf_frags.destructor = NULL;
663 nf_frags.skb_free = nf_skb_free;
664 nf_frags.qsize = sizeof(struct nf_ct_frag6_queue);
665 nf_frags.match = ip6_frag_match;
666 nf_frags.frag_expire = nf_ct_frag6_expire;
667 nf_frags.secret_interval = 10 * 60 * HZ;
668 nf_init_frags.timeout = IPV6_FRAG_TIMEOUT;
669 nf_init_frags.high_thresh = IPV6_FRAG_HIGH_THRESH;
670 nf_init_frags.low_thresh = IPV6_FRAG_LOW_THRESH;
671 inet_frags_init_net(&nf_init_frags);
672 inet_frags_init(&nf_frags);
677 void nf_ct_frag6_cleanup(void)
679 inet_frags_fini(&nf_frags);
681 nf_init_frags.low_thresh = 0;
682 nf_ct_frag6_evictor();