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 #define pr_fmt(fmt) "IPv6-nf: " fmt
19 #include <linux/errno.h>
20 #include <linux/types.h>
21 #include <linux/string.h>
22 #include <linux/socket.h>
23 #include <linux/sockios.h>
24 #include <linux/jiffies.h>
25 #include <linux/net.h>
26 #include <linux/list.h>
27 #include <linux/netdevice.h>
28 #include <linux/in6.h>
29 #include <linux/ipv6.h>
30 #include <linux/icmpv6.h>
31 #include <linux/random.h>
32 #include <linux/slab.h>
36 #include <net/inet_frag.h>
39 #include <net/protocol.h>
40 #include <net/transp_v6.h>
41 #include <net/rawv6.h>
42 #include <net/ndisc.h>
43 #include <net/addrconf.h>
44 #include <net/inet_ecn.h>
45 #include <net/netfilter/ipv6/nf_conntrack_ipv6.h>
46 #include <linux/sysctl.h>
47 #include <linux/netfilter.h>
48 #include <linux/netfilter_ipv6.h>
49 #include <linux/kernel.h>
50 #include <linux/module.h>
51 #include <net/netfilter/ipv6/nf_defrag_ipv6.h>
53 static const char nf_frags_cache_name[] = "nf-frags";
55 struct nf_ct_frag6_skb_cb
57 struct inet6_skb_parm h;
61 #define NFCT_FRAG6_CB(skb) ((struct nf_ct_frag6_skb_cb *)((skb)->cb))
63 static struct inet_frags nf_frags;
68 static struct ctl_table nf_ct_frag6_sysctl_table[] = {
70 .procname = "nf_conntrack_frag6_timeout",
71 .data = &init_net.nf_frag.frags.timeout,
72 .maxlen = sizeof(unsigned int),
74 .proc_handler = proc_dointvec_jiffies,
77 .procname = "nf_conntrack_frag6_low_thresh",
78 .data = &init_net.nf_frag.frags.low_thresh,
79 .maxlen = sizeof(unsigned int),
81 .proc_handler = proc_dointvec_minmax,
83 .extra2 = &init_net.nf_frag.frags.high_thresh
86 .procname = "nf_conntrack_frag6_high_thresh",
87 .data = &init_net.nf_frag.frags.high_thresh,
88 .maxlen = sizeof(unsigned int),
90 .proc_handler = proc_dointvec_minmax,
91 .extra1 = &init_net.nf_frag.frags.low_thresh
96 static int nf_ct_frag6_sysctl_register(struct net *net)
98 struct ctl_table *table;
99 struct ctl_table_header *hdr;
101 table = nf_ct_frag6_sysctl_table;
102 if (!net_eq(net, &init_net)) {
103 table = kmemdup(table, sizeof(nf_ct_frag6_sysctl_table),
108 table[0].data = &net->nf_frag.frags.timeout;
109 table[1].data = &net->nf_frag.frags.low_thresh;
110 table[1].extra2 = &net->nf_frag.frags.high_thresh;
111 table[2].data = &net->nf_frag.frags.high_thresh;
112 table[2].extra1 = &net->nf_frag.frags.low_thresh;
113 table[2].extra2 = &init_net.nf_frag.frags.high_thresh;
116 hdr = register_net_sysctl(net, "net/netfilter", table);
120 net->nf_frag.sysctl.frags_hdr = hdr;
124 if (!net_eq(net, &init_net))
130 static void __net_exit nf_ct_frags6_sysctl_unregister(struct net *net)
132 struct ctl_table *table;
134 table = net->nf_frag.sysctl.frags_hdr->ctl_table_arg;
135 unregister_net_sysctl_table(net->nf_frag.sysctl.frags_hdr);
136 if (!net_eq(net, &init_net))
141 static int nf_ct_frag6_sysctl_register(struct net *net)
145 static void __net_exit nf_ct_frags6_sysctl_unregister(struct net *net)
150 static inline u8 ip6_frag_ecn(const struct ipv6hdr *ipv6h)
152 return 1 << (ipv6_get_dsfield(ipv6h) & INET_ECN_MASK);
155 static unsigned int nf_hash_frag(__be32 id, const struct in6_addr *saddr,
156 const struct in6_addr *daddr)
158 net_get_random_once(&nf_frags.rnd, sizeof(nf_frags.rnd));
159 return jhash_3words(ipv6_addr_hash(saddr), ipv6_addr_hash(daddr),
160 (__force u32)id, nf_frags.rnd);
164 static unsigned int nf_hashfn(const struct inet_frag_queue *q)
166 const struct frag_queue *nq;
168 nq = container_of(q, struct frag_queue, q);
169 return nf_hash_frag(nq->id, &nq->saddr, &nq->daddr);
172 static void nf_ct_frag6_expire(struct timer_list *t)
174 struct inet_frag_queue *frag = from_timer(frag, t, timer);
175 struct frag_queue *fq;
178 fq = container_of(frag, struct frag_queue, q);
179 net = container_of(fq->q.net, struct net, nf_frag.frags);
181 ip6_expire_frag_queue(net, fq, &nf_frags);
184 /* Creation primitives. */
185 static inline struct frag_queue *fq_find(struct net *net, __be32 id,
186 u32 user, struct in6_addr *src,
187 struct in6_addr *dst, int iif, u8 ecn)
189 struct inet_frag_queue *q;
190 struct ip6_create_arg arg;
201 hash = nf_hash_frag(id, src, dst);
203 q = inet_frag_find(&net->nf_frag.frags, &nf_frags, &arg, hash);
205 if (IS_ERR_OR_NULL(q)) {
206 inet_frag_maybe_warn_overflow(q, pr_fmt());
209 return container_of(q, struct frag_queue, q);
213 static int nf_ct_frag6_queue(struct frag_queue *fq, struct sk_buff *skb,
214 const struct frag_hdr *fhdr, int nhoff)
216 struct sk_buff *prev, *next;
217 unsigned int payload_len;
221 if (fq->q.flags & INET_FRAG_COMPLETE) {
222 pr_debug("Already completed\n");
226 payload_len = ntohs(ipv6_hdr(skb)->payload_len);
228 offset = ntohs(fhdr->frag_off) & ~0x7;
229 end = offset + (payload_len -
230 ((u8 *)(fhdr + 1) - (u8 *)(ipv6_hdr(skb) + 1)));
232 if ((unsigned int)end > IPV6_MAXPLEN) {
233 pr_debug("offset is too large.\n");
237 ecn = ip6_frag_ecn(ipv6_hdr(skb));
239 if (skb->ip_summed == CHECKSUM_COMPLETE) {
240 const unsigned char *nh = skb_network_header(skb);
241 skb->csum = csum_sub(skb->csum,
242 csum_partial(nh, (u8 *)(fhdr + 1) - nh,
246 /* Is this the final fragment? */
247 if (!(fhdr->frag_off & htons(IP6_MF))) {
248 /* If we already have some bits beyond end
249 * or have different end, the segment is corrupted.
251 if (end < fq->q.len ||
252 ((fq->q.flags & INET_FRAG_LAST_IN) && end != fq->q.len)) {
253 pr_debug("already received last fragment\n");
256 fq->q.flags |= INET_FRAG_LAST_IN;
259 /* Check if the fragment is rounded to 8 bytes.
260 * Required by the RFC.
263 /* RFC2460 says always send parameter problem in
266 pr_debug("end of fragment not rounded to 8 bytes.\n");
269 if (end > fq->q.len) {
270 /* Some bits beyond end -> corruption. */
271 if (fq->q.flags & INET_FRAG_LAST_IN) {
272 pr_debug("last packet already reached.\n");
282 /* Point into the IP datagram 'data' part. */
283 if (!pskb_pull(skb, (u8 *) (fhdr + 1) - skb->data)) {
284 pr_debug("queue: message is too short.\n");
287 if (pskb_trim_rcsum(skb, end - offset)) {
288 pr_debug("Can't trim\n");
292 /* Find out which fragments are in front and at the back of us
293 * in the chain of fragments so far. We must know where to put
294 * this fragment, right?
296 prev = fq->q.fragments_tail;
297 if (!prev || NFCT_FRAG6_CB(prev)->offset < offset) {
302 for (next = fq->q.fragments; next != NULL; next = next->next) {
303 if (NFCT_FRAG6_CB(next)->offset >= offset)
309 /* RFC5722, Section 4:
310 * When reassembling an IPv6 datagram, if
311 * one or more its constituent fragments is determined to be an
312 * overlapping fragment, the entire datagram (and any constituent
313 * fragments, including those not yet received) MUST be silently
317 /* Check for overlap with preceding fragment. */
319 (NFCT_FRAG6_CB(prev)->offset + prev->len) > offset)
322 /* Look for overlap with succeeding segment. */
323 if (next && NFCT_FRAG6_CB(next)->offset < end)
326 NFCT_FRAG6_CB(skb)->offset = offset;
328 /* Insert this fragment in the chain of fragments. */
331 fq->q.fragments_tail = skb;
335 fq->q.fragments = skb;
338 fq->iif = skb->dev->ifindex;
341 fq->q.stamp = skb->tstamp;
342 fq->q.meat += skb->len;
344 if (payload_len > fq->q.max_size)
345 fq->q.max_size = payload_len;
346 add_frag_mem_limit(fq->q.net, skb->truesize);
348 /* The first fragment.
349 * nhoffset is obtained from the first fragment, of course.
352 fq->nhoffset = nhoff;
353 fq->q.flags |= INET_FRAG_FIRST_IN;
359 inet_frag_kill(&fq->q, &nf_frags);
365 * Check if this packet is complete.
367 * It is called with locked fq, and caller must check that
368 * queue is eligible for reassembly i.e. it is not COMPLETE,
369 * the last and the first frames arrived and all the bits are here.
371 * returns true if *prev skb has been transformed into the reassembled
372 * skb, false otherwise.
375 nf_ct_frag6_reasm(struct frag_queue *fq, struct sk_buff *prev, struct net_device *dev)
377 struct sk_buff *fp, *head = fq->q.fragments;
381 inet_frag_kill(&fq->q, &nf_frags);
383 WARN_ON(head == NULL);
384 WARN_ON(NFCT_FRAG6_CB(head)->offset != 0);
386 ecn = ip_frag_ecn_table[fq->ecn];
387 if (unlikely(ecn == 0xff))
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 net_dbg_ratelimited("nf_ct_frag6_reasm: payload len = %d\n",
400 /* Head of list must not be cloned. */
401 if (skb_unclone(head, GFP_ATOMIC))
404 /* If the first fragment is fragmented itself, we split
405 * it to two chunks: the first with data and paged part
406 * and the second, holding only fragments. */
407 if (skb_has_frag_list(head)) {
408 struct sk_buff *clone;
411 clone = alloc_skb(0, GFP_ATOMIC);
415 clone->next = head->next;
417 skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
418 skb_frag_list_init(head);
419 for (i = 0; i < skb_shinfo(head)->nr_frags; i++)
420 plen += skb_frag_size(&skb_shinfo(head)->frags[i]);
421 clone->len = clone->data_len = head->data_len - plen;
422 head->data_len -= clone->len;
423 head->len -= clone->len;
425 clone->ip_summed = head->ip_summed;
427 add_frag_mem_limit(fq->q.net, clone->truesize);
430 /* morph head into last received skb: prev.
432 * This allows callers of ipv6 conntrack defrag to continue
433 * to use the last skb(frag) passed into the reasm engine.
434 * The last skb frag 'silently' turns into the full reassembled skb.
436 * Since prev is also part of q->fragments we have to clone it first.
439 struct sk_buff *iter;
441 fp = skb_clone(prev, GFP_ATOMIC);
445 fp->next = prev->next;
449 if (iter->next == prev) {
456 skb_morph(prev, head);
457 prev->next = head->next;
462 /* We have to remove fragment header from datagram and to relocate
463 * header in order to calculate ICV correctly. */
464 skb_network_header(head)[fq->nhoffset] = skb_transport_header(head)[0];
465 memmove(head->head + sizeof(struct frag_hdr), head->head,
466 (head->data - head->head) - sizeof(struct frag_hdr));
467 head->mac_header += sizeof(struct frag_hdr);
468 head->network_header += sizeof(struct frag_hdr);
470 skb_shinfo(head)->frag_list = head->next;
471 skb_reset_transport_header(head);
472 skb_push(head, head->data - skb_network_header(head));
474 for (fp = head->next; fp; fp = fp->next) {
475 head->data_len += fp->len;
476 head->len += fp->len;
477 if (head->ip_summed != fp->ip_summed)
478 head->ip_summed = CHECKSUM_NONE;
479 else if (head->ip_summed == CHECKSUM_COMPLETE)
480 head->csum = csum_add(head->csum, fp->csum);
481 head->truesize += fp->truesize;
483 sub_frag_mem_limit(fq->q.net, head->truesize);
488 head->tstamp = fq->q.stamp;
489 ipv6_hdr(head)->payload_len = htons(payload_len);
490 ipv6_change_dsfield(ipv6_hdr(head), 0xff, ecn);
491 IP6CB(head)->frag_max_size = sizeof(struct ipv6hdr) + fq->q.max_size;
493 /* Yes, and fold redundant checksum back. 8) */
494 if (head->ip_summed == CHECKSUM_COMPLETE)
495 head->csum = csum_partial(skb_network_header(head),
496 skb_network_header_len(head),
499 fq->q.fragments = NULL;
500 fq->q.fragments_tail = NULL;
506 * find the header just before Fragment Header.
508 * if success return 0 and set ...
509 * (*prevhdrp): the value of "Next Header Field" in the header
510 * just before Fragment Header.
511 * (*prevhoff): the offset of "Next Header Field" in the header
512 * just before Fragment Header.
513 * (*fhoff) : the offset of Fragment Header.
515 * Based on ipv6_skip_hdr() in net/ipv6/exthdr.c
519 find_prev_fhdr(struct sk_buff *skb, u8 *prevhdrp, int *prevhoff, int *fhoff)
521 u8 nexthdr = ipv6_hdr(skb)->nexthdr;
522 const int netoff = skb_network_offset(skb);
523 u8 prev_nhoff = netoff + offsetof(struct ipv6hdr, nexthdr);
524 int start = netoff + sizeof(struct ipv6hdr);
525 int len = skb->len - start;
526 u8 prevhdr = NEXTHDR_IPV6;
528 while (nexthdr != NEXTHDR_FRAGMENT) {
529 struct ipv6_opt_hdr hdr;
532 if (!ipv6_ext_hdr(nexthdr)) {
535 if (nexthdr == NEXTHDR_NONE) {
536 pr_debug("next header is none\n");
539 if (len < (int)sizeof(struct ipv6_opt_hdr)) {
540 pr_debug("too short\n");
543 if (skb_copy_bits(skb, start, &hdr, sizeof(hdr)))
545 if (nexthdr == NEXTHDR_AUTH)
546 hdrlen = (hdr.hdrlen+2)<<2;
548 hdrlen = ipv6_optlen(&hdr);
553 nexthdr = hdr.nexthdr;
562 *prevhoff = prev_nhoff;
568 int nf_ct_frag6_gather(struct net *net, struct sk_buff *skb, u32 user)
570 u16 savethdr = skb->transport_header;
571 struct net_device *dev = skb->dev;
572 int fhoff, nhoff, ret;
573 struct frag_hdr *fhdr;
574 struct frag_queue *fq;
578 /* Jumbo payload inhibits frag. header */
579 if (ipv6_hdr(skb)->payload_len == 0) {
580 pr_debug("payload len = 0\n");
584 if (find_prev_fhdr(skb, &prevhdr, &nhoff, &fhoff) < 0)
587 if (!pskb_may_pull(skb, fhoff + sizeof(*fhdr)))
590 skb_set_transport_header(skb, fhoff);
592 fhdr = (struct frag_hdr *)skb_transport_header(skb);
595 fq = fq_find(net, fhdr->identification, user, &hdr->saddr, &hdr->daddr,
596 skb->dev ? skb->dev->ifindex : 0, ip6_frag_ecn(hdr));
598 pr_debug("Can't find and can't create new queue\n");
602 spin_lock_bh(&fq->q.lock);
604 ret = nf_ct_frag6_queue(fq, skb, fhdr, nhoff);
606 if (ret == -EPROTO) {
607 skb->transport_header = savethdr;
613 /* after queue has assumed skb ownership, only 0 or -EINPROGRESS
617 if (fq->q.flags == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
618 fq->q.meat == fq->q.len &&
619 nf_ct_frag6_reasm(fq, skb, dev))
623 spin_unlock_bh(&fq->q.lock);
624 inet_frag_put(&fq->q, &nf_frags);
627 EXPORT_SYMBOL_GPL(nf_ct_frag6_gather);
629 static int nf_ct_net_init(struct net *net)
631 net->nf_frag.frags.high_thresh = IPV6_FRAG_HIGH_THRESH;
632 net->nf_frag.frags.low_thresh = IPV6_FRAG_LOW_THRESH;
633 net->nf_frag.frags.timeout = IPV6_FRAG_TIMEOUT;
634 inet_frags_init_net(&net->nf_frag.frags);
636 return nf_ct_frag6_sysctl_register(net);
639 static void nf_ct_net_exit(struct net *net)
641 nf_ct_frags6_sysctl_unregister(net);
642 inet_frags_exit_net(&net->nf_frag.frags, &nf_frags);
645 static struct pernet_operations nf_ct_net_ops = {
646 .init = nf_ct_net_init,
647 .exit = nf_ct_net_exit,
650 int nf_ct_frag6_init(void)
654 nf_frags.hashfn = nf_hashfn;
655 nf_frags.constructor = ip6_frag_init;
656 nf_frags.destructor = NULL;
657 nf_frags.qsize = sizeof(struct frag_queue);
658 nf_frags.match = ip6_frag_match;
659 nf_frags.frag_expire = nf_ct_frag6_expire;
660 nf_frags.frags_cache_name = nf_frags_cache_name;
661 ret = inet_frags_init(&nf_frags);
664 ret = register_pernet_subsys(&nf_ct_net_ops);
666 inet_frags_fini(&nf_frags);
672 void nf_ct_frag6_cleanup(void)
674 unregister_pernet_subsys(&nf_ct_net_ops);
675 inet_frags_fini(&nf_frags);