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>
30 #include <linux/slab.h>
34 #include <net/inet_frag.h>
37 #include <net/protocol.h>
38 #include <net/transp_v6.h>
39 #include <net/rawv6.h>
40 #include <net/ndisc.h>
41 #include <net/addrconf.h>
42 #include <net/netfilter/ipv6/nf_conntrack_ipv6.h>
43 #include <linux/sysctl.h>
44 #include <linux/netfilter.h>
45 #include <linux/netfilter_ipv6.h>
46 #include <linux/kernel.h>
47 #include <linux/module.h>
48 #include <net/netfilter/ipv6/nf_defrag_ipv6.h>
51 struct nf_ct_frag6_skb_cb
53 struct inet6_skb_parm h;
58 #define NFCT_FRAG6_CB(skb) ((struct nf_ct_frag6_skb_cb*)((skb)->cb))
60 static struct inet_frags nf_frags;
63 static struct ctl_table nf_ct_frag6_sysctl_table[] = {
65 .procname = "nf_conntrack_frag6_timeout",
66 .data = &init_net.nf_frag.frags.timeout,
67 .maxlen = sizeof(unsigned int),
69 .proc_handler = proc_dointvec_jiffies,
72 .procname = "nf_conntrack_frag6_low_thresh",
73 .data = &init_net.nf_frag.frags.low_thresh,
74 .maxlen = sizeof(unsigned int),
76 .proc_handler = proc_dointvec,
79 .procname = "nf_conntrack_frag6_high_thresh",
80 .data = &init_net.nf_frag.frags.high_thresh,
81 .maxlen = sizeof(unsigned int),
83 .proc_handler = proc_dointvec,
88 static int nf_ct_frag6_sysctl_register(struct net *net)
90 struct ctl_table *table;
91 struct ctl_table_header *hdr;
93 table = nf_ct_frag6_sysctl_table;
94 if (!net_eq(net, &init_net)) {
95 table = kmemdup(table, sizeof(nf_ct_frag6_sysctl_table),
100 table[0].data = &net->ipv6.frags.high_thresh;
101 table[1].data = &net->ipv6.frags.low_thresh;
102 table[2].data = &net->ipv6.frags.timeout;
105 hdr = register_net_sysctl(net, "net/netfilter", table);
109 net->nf_frag.sysctl.frags_hdr = hdr;
113 if (!net_eq(net, &init_net))
119 static void __net_exit nf_ct_frags6_sysctl_unregister(struct net *net)
121 struct ctl_table *table;
123 table = net->nf_frag.sysctl.frags_hdr->ctl_table_arg;
124 unregister_net_sysctl_table(net->nf_frag.sysctl.frags_hdr);
125 if (!net_eq(net, &init_net))
130 static int nf_ct_frag6_sysctl_register(struct net *net)
134 static void __net_exit nf_ct_frags6_sysctl_unregister(struct net *net)
139 static unsigned int nf_hashfn(struct inet_frag_queue *q)
141 const struct frag_queue *nq;
143 nq = container_of(q, struct frag_queue, q);
144 return inet6_hash_frag(nq->id, &nq->saddr, &nq->daddr, nf_frags.rnd);
147 static void nf_skb_free(struct sk_buff *skb)
149 if (NFCT_FRAG6_CB(skb)->orig)
150 kfree_skb(NFCT_FRAG6_CB(skb)->orig);
153 static void nf_ct_frag6_expire(unsigned long data)
155 struct frag_queue *fq;
158 fq = container_of((struct inet_frag_queue *)data, struct frag_queue, q);
159 net = container_of(fq->q.net, struct net, nf_frag.frags);
161 ip6_expire_frag_queue(net, fq, &nf_frags);
164 /* Creation primitives. */
165 static inline struct frag_queue *fq_find(struct net *net, __be32 id,
166 u32 user, struct in6_addr *src,
167 struct in6_addr *dst)
169 struct inet_frag_queue *q;
170 struct ip6_create_arg arg;
178 read_lock_bh(&nf_frags.lock);
179 hash = inet6_hash_frag(id, src, dst, nf_frags.rnd);
181 q = inet_frag_find(&net->nf_frag.frags, &nf_frags, &arg, hash);
186 return container_of(q, struct frag_queue, q);
193 static int nf_ct_frag6_queue(struct frag_queue *fq, struct sk_buff *skb,
194 const struct frag_hdr *fhdr, int nhoff)
196 struct sk_buff *prev, *next;
197 unsigned int payload_len;
200 if (fq->q.last_in & INET_FRAG_COMPLETE) {
201 pr_debug("Already completed\n");
205 payload_len = ntohs(ipv6_hdr(skb)->payload_len);
207 offset = ntohs(fhdr->frag_off) & ~0x7;
208 end = offset + (payload_len -
209 ((u8 *)(fhdr + 1) - (u8 *)(ipv6_hdr(skb) + 1)));
211 if ((unsigned int)end > IPV6_MAXPLEN) {
212 pr_debug("offset is too large.\n");
216 if (skb->ip_summed == CHECKSUM_COMPLETE) {
217 const unsigned char *nh = skb_network_header(skb);
218 skb->csum = csum_sub(skb->csum,
219 csum_partial(nh, (u8 *)(fhdr + 1) - nh,
223 /* Is this the final fragment? */
224 if (!(fhdr->frag_off & htons(IP6_MF))) {
225 /* If we already have some bits beyond end
226 * or have different end, the segment is corrupted.
228 if (end < fq->q.len ||
229 ((fq->q.last_in & INET_FRAG_LAST_IN) && end != fq->q.len)) {
230 pr_debug("already received last fragment\n");
233 fq->q.last_in |= INET_FRAG_LAST_IN;
236 /* Check if the fragment is rounded to 8 bytes.
237 * Required by the RFC.
240 /* RFC2460 says always send parameter problem in
243 pr_debug("end of fragment not rounded to 8 bytes.\n");
246 if (end > fq->q.len) {
247 /* Some bits beyond end -> corruption. */
248 if (fq->q.last_in & INET_FRAG_LAST_IN) {
249 pr_debug("last packet already reached.\n");
259 /* Point into the IP datagram 'data' part. */
260 if (!pskb_pull(skb, (u8 *) (fhdr + 1) - skb->data)) {
261 pr_debug("queue: message is too short.\n");
264 if (pskb_trim_rcsum(skb, end - offset)) {
265 pr_debug("Can't trim\n");
269 /* Find out which fragments are in front and at the back of us
270 * in the chain of fragments so far. We must know where to put
271 * this fragment, right?
273 prev = fq->q.fragments_tail;
274 if (!prev || NFCT_FRAG6_CB(prev)->offset < offset) {
279 for (next = fq->q.fragments; next != NULL; next = next->next) {
280 if (NFCT_FRAG6_CB(next)->offset >= offset)
286 /* RFC5722, Section 4:
287 * When reassembling an IPv6 datagram, if
288 * one or more its constituent fragments is determined to be an
289 * overlapping fragment, the entire datagram (and any constituent
290 * fragments, including those not yet received) MUST be silently
294 /* Check for overlap with preceding fragment. */
296 (NFCT_FRAG6_CB(prev)->offset + prev->len) > offset)
299 /* Look for overlap with succeeding segment. */
300 if (next && NFCT_FRAG6_CB(next)->offset < end)
303 NFCT_FRAG6_CB(skb)->offset = offset;
305 /* Insert this fragment in the chain of fragments. */
308 fq->q.fragments_tail = skb;
312 fq->q.fragments = skb;
315 fq->q.stamp = skb->tstamp;
316 fq->q.meat += skb->len;
317 if (payload_len > fq->q.max_size)
318 fq->q.max_size = payload_len;
319 atomic_add(skb->truesize, &fq->q.net->mem);
321 /* The first fragment.
322 * nhoffset is obtained from the first fragment, of course.
325 fq->nhoffset = nhoff;
326 fq->q.last_in |= INET_FRAG_FIRST_IN;
328 write_lock(&nf_frags.lock);
329 list_move_tail(&fq->q.lru_list, &fq->q.net->lru_list);
330 write_unlock(&nf_frags.lock);
334 inet_frag_kill(&fq->q, &nf_frags);
340 * Check if this packet is complete.
341 * Returns NULL on failure by any reason, and pointer
342 * to current nexthdr field in reassembled frame.
344 * It is called with locked fq, and caller must check that
345 * queue is eligible for reassembly i.e. it is not COMPLETE,
346 * the last and the first frames arrived and all the bits are here.
348 static struct sk_buff *
349 nf_ct_frag6_reasm(struct frag_queue *fq, struct net_device *dev)
351 struct sk_buff *fp, *op, *head = fq->q.fragments;
354 inet_frag_kill(&fq->q, &nf_frags);
356 WARN_ON(head == NULL);
357 WARN_ON(NFCT_FRAG6_CB(head)->offset != 0);
359 /* Unfragmented part is taken from the first segment. */
360 payload_len = ((head->data - skb_network_header(head)) -
361 sizeof(struct ipv6hdr) + fq->q.len -
362 sizeof(struct frag_hdr));
363 if (payload_len > IPV6_MAXPLEN) {
364 pr_debug("payload len is too large.\n");
368 /* Head of list must not be cloned. */
369 if (skb_cloned(head) && pskb_expand_head(head, 0, 0, GFP_ATOMIC)) {
370 pr_debug("skb is cloned but can't expand head");
374 /* If the first fragment is fragmented itself, we split
375 * it to two chunks: the first with data and paged part
376 * and the second, holding only fragments. */
377 if (skb_has_frag_list(head)) {
378 struct sk_buff *clone;
381 clone = alloc_skb(0, GFP_ATOMIC);
385 clone->next = head->next;
387 skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
388 skb_frag_list_init(head);
389 for (i = 0; i < skb_shinfo(head)->nr_frags; i++)
390 plen += skb_frag_size(&skb_shinfo(head)->frags[i]);
391 clone->len = clone->data_len = head->data_len - plen;
392 head->data_len -= clone->len;
393 head->len -= clone->len;
395 clone->ip_summed = head->ip_summed;
397 NFCT_FRAG6_CB(clone)->orig = NULL;
398 atomic_add(clone->truesize, &fq->q.net->mem);
401 /* We have to remove fragment header from datagram and to relocate
402 * header in order to calculate ICV correctly. */
403 skb_network_header(head)[fq->nhoffset] = skb_transport_header(head)[0];
404 memmove(head->head + sizeof(struct frag_hdr), head->head,
405 (head->data - head->head) - sizeof(struct frag_hdr));
406 head->mac_header += sizeof(struct frag_hdr);
407 head->network_header += sizeof(struct frag_hdr);
409 skb_shinfo(head)->frag_list = head->next;
410 skb_reset_transport_header(head);
411 skb_push(head, head->data - skb_network_header(head));
413 for (fp=head->next; fp; fp = fp->next) {
414 head->data_len += fp->len;
415 head->len += fp->len;
416 if (head->ip_summed != fp->ip_summed)
417 head->ip_summed = CHECKSUM_NONE;
418 else if (head->ip_summed == CHECKSUM_COMPLETE)
419 head->csum = csum_add(head->csum, fp->csum);
420 head->truesize += fp->truesize;
422 atomic_sub(head->truesize, &fq->q.net->mem);
427 head->tstamp = fq->q.stamp;
428 ipv6_hdr(head)->payload_len = htons(payload_len);
429 IP6CB(head)->frag_max_size = sizeof(struct ipv6hdr) + fq->q.max_size;
431 /* Yes, and fold redundant checksum back. 8) */
432 if (head->ip_summed == CHECKSUM_COMPLETE)
433 head->csum = csum_partial(skb_network_header(head),
434 skb_network_header_len(head),
437 fq->q.fragments = NULL;
438 fq->q.fragments_tail = NULL;
440 /* all original skbs are linked into the NFCT_FRAG6_CB(head).orig */
441 fp = skb_shinfo(head)->frag_list;
442 if (fp && NFCT_FRAG6_CB(fp)->orig == NULL)
443 /* at above code, head skb is divided into two skbs. */
446 op = NFCT_FRAG6_CB(head)->orig;
447 for (; fp; fp = fp->next) {
448 struct sk_buff *orig = NFCT_FRAG6_CB(fp)->orig;
452 NFCT_FRAG6_CB(fp)->orig = NULL;
458 net_dbg_ratelimited("nf_ct_frag6_reasm: payload len = %d\n",
462 net_dbg_ratelimited("nf_ct_frag6_reasm: no memory for reassembly\n");
468 * find the header just before Fragment Header.
470 * if success return 0 and set ...
471 * (*prevhdrp): the value of "Next Header Field" in the header
472 * just before Fragment Header.
473 * (*prevhoff): the offset of "Next Header Field" in the header
474 * just before Fragment Header.
475 * (*fhoff) : the offset of Fragment Header.
477 * Based on ipv6_skip_hdr() in net/ipv6/exthdr.c
481 find_prev_fhdr(struct sk_buff *skb, u8 *prevhdrp, int *prevhoff, int *fhoff)
483 u8 nexthdr = ipv6_hdr(skb)->nexthdr;
484 const int netoff = skb_network_offset(skb);
485 u8 prev_nhoff = netoff + offsetof(struct ipv6hdr, nexthdr);
486 int start = netoff + sizeof(struct ipv6hdr);
487 int len = skb->len - start;
488 u8 prevhdr = NEXTHDR_IPV6;
490 while (nexthdr != NEXTHDR_FRAGMENT) {
491 struct ipv6_opt_hdr hdr;
494 if (!ipv6_ext_hdr(nexthdr)) {
497 if (nexthdr == NEXTHDR_NONE) {
498 pr_debug("next header is none\n");
501 if (len < (int)sizeof(struct ipv6_opt_hdr)) {
502 pr_debug("too short\n");
505 if (skb_copy_bits(skb, start, &hdr, sizeof(hdr)))
507 if (nexthdr == NEXTHDR_AUTH)
508 hdrlen = (hdr.hdrlen+2)<<2;
510 hdrlen = ipv6_optlen(&hdr);
515 nexthdr = hdr.nexthdr;
524 *prevhoff = prev_nhoff;
530 struct sk_buff *nf_ct_frag6_gather(struct sk_buff *skb, u32 user)
532 struct sk_buff *clone;
533 struct net_device *dev = skb->dev;
534 struct net *net = skb_dst(skb) ? dev_net(skb_dst(skb)->dev)
536 struct frag_hdr *fhdr;
537 struct frag_queue *fq;
541 struct sk_buff *ret_skb = NULL;
543 /* Jumbo payload inhibits frag. header */
544 if (ipv6_hdr(skb)->payload_len == 0) {
545 pr_debug("payload len = 0\n");
549 if (find_prev_fhdr(skb, &prevhdr, &nhoff, &fhoff) < 0)
552 clone = skb_clone(skb, GFP_ATOMIC);
554 pr_debug("Can't clone skb\n");
558 NFCT_FRAG6_CB(clone)->orig = skb;
560 if (!pskb_may_pull(clone, fhoff + sizeof(*fhdr))) {
561 pr_debug("message is too short.\n");
565 skb_set_transport_header(clone, fhoff);
566 hdr = ipv6_hdr(clone);
567 fhdr = (struct frag_hdr *)skb_transport_header(clone);
570 inet_frag_evictor(&net->nf_frag.frags, &nf_frags, false);
573 fq = fq_find(net, fhdr->identification, user, &hdr->saddr, &hdr->daddr);
575 pr_debug("Can't find and can't create new queue\n");
579 spin_lock_bh(&fq->q.lock);
581 if (nf_ct_frag6_queue(fq, clone, fhdr, nhoff) < 0) {
582 spin_unlock_bh(&fq->q.lock);
583 pr_debug("Can't insert skb to queue\n");
584 inet_frag_put(&fq->q, &nf_frags);
588 if (fq->q.last_in == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
589 fq->q.meat == fq->q.len) {
590 ret_skb = nf_ct_frag6_reasm(fq, dev);
592 pr_debug("Can't reassemble fragmented packets\n");
594 spin_unlock_bh(&fq->q.lock);
596 inet_frag_put(&fq->q, &nf_frags);
604 void nf_ct_frag6_output(unsigned int hooknum, struct sk_buff *skb,
605 struct net_device *in, struct net_device *out,
606 int (*okfn)(struct sk_buff *))
608 struct sk_buff *s, *s2;
609 unsigned int ret = 0;
611 for (s = NFCT_FRAG6_CB(skb)->orig; s;) {
612 nf_conntrack_put_reasm(s->nfct_reasm);
613 nf_conntrack_get_reasm(skb);
619 if (ret != -ECANCELED)
620 ret = NF_HOOK_THRESH(NFPROTO_IPV6, hooknum, s,
622 NF_IP6_PRI_CONNTRACK_DEFRAG + 1);
628 nf_conntrack_put_reasm(skb);
631 static int nf_ct_net_init(struct net *net)
633 net->nf_frag.frags.high_thresh = IPV6_FRAG_HIGH_THRESH;
634 net->nf_frag.frags.low_thresh = IPV6_FRAG_LOW_THRESH;
635 net->nf_frag.frags.timeout = IPV6_FRAG_TIMEOUT;
636 inet_frags_init_net(&net->nf_frag.frags);
638 return nf_ct_frag6_sysctl_register(net);
641 static void nf_ct_net_exit(struct net *net)
643 nf_ct_frags6_sysctl_unregister(net);
644 inet_frags_exit_net(&net->nf_frag.frags, &nf_frags);
647 static struct pernet_operations nf_ct_net_ops = {
648 .init = nf_ct_net_init,
649 .exit = nf_ct_net_exit,
652 int nf_ct_frag6_init(void)
656 nf_frags.hashfn = nf_hashfn;
657 nf_frags.constructor = ip6_frag_init;
658 nf_frags.destructor = NULL;
659 nf_frags.skb_free = nf_skb_free;
660 nf_frags.qsize = sizeof(struct frag_queue);
661 nf_frags.match = ip6_frag_match;
662 nf_frags.frag_expire = nf_ct_frag6_expire;
663 nf_frags.secret_interval = 10 * 60 * HZ;
664 inet_frags_init(&nf_frags);
666 ret = register_pernet_subsys(&nf_ct_net_ops);
668 inet_frags_fini(&nf_frags);
673 void nf_ct_frag6_cleanup(void)
675 unregister_pernet_subsys(&nf_ct_net_ops);
676 inet_frags_fini(&nf_frags);