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;
62 #define NFCT_FRAG6_CB(skb) ((struct nf_ct_frag6_skb_cb *)((skb)->cb))
64 static struct inet_frags nf_frags;
69 static struct ctl_table nf_ct_frag6_sysctl_table[] = {
71 .procname = "nf_conntrack_frag6_timeout",
72 .data = &init_net.nf_frag.frags.timeout,
73 .maxlen = sizeof(unsigned int),
75 .proc_handler = proc_dointvec_jiffies,
78 .procname = "nf_conntrack_frag6_low_thresh",
79 .data = &init_net.nf_frag.frags.low_thresh,
80 .maxlen = sizeof(unsigned int),
82 .proc_handler = proc_dointvec_minmax,
84 .extra2 = &init_net.nf_frag.frags.high_thresh
87 .procname = "nf_conntrack_frag6_high_thresh",
88 .data = &init_net.nf_frag.frags.high_thresh,
89 .maxlen = sizeof(unsigned int),
91 .proc_handler = proc_dointvec_minmax,
92 .extra1 = &init_net.nf_frag.frags.low_thresh
97 static int nf_ct_frag6_sysctl_register(struct net *net)
99 struct ctl_table *table;
100 struct ctl_table_header *hdr;
102 table = nf_ct_frag6_sysctl_table;
103 if (!net_eq(net, &init_net)) {
104 table = kmemdup(table, sizeof(nf_ct_frag6_sysctl_table),
109 table[0].data = &net->nf_frag.frags.timeout;
110 table[1].data = &net->nf_frag.frags.low_thresh;
111 table[1].extra2 = &net->nf_frag.frags.high_thresh;
112 table[2].data = &net->nf_frag.frags.high_thresh;
113 table[2].extra1 = &net->nf_frag.frags.low_thresh;
114 table[2].extra2 = &init_net.nf_frag.frags.high_thresh;
117 hdr = register_net_sysctl(net, "net/netfilter", table);
121 net->nf_frag.sysctl.frags_hdr = hdr;
125 if (!net_eq(net, &init_net))
131 static void __net_exit nf_ct_frags6_sysctl_unregister(struct net *net)
133 struct ctl_table *table;
135 table = net->nf_frag.sysctl.frags_hdr->ctl_table_arg;
136 unregister_net_sysctl_table(net->nf_frag.sysctl.frags_hdr);
137 if (!net_eq(net, &init_net))
142 static int nf_ct_frag6_sysctl_register(struct net *net)
146 static void __net_exit nf_ct_frags6_sysctl_unregister(struct net *net)
151 static inline u8 ip6_frag_ecn(const struct ipv6hdr *ipv6h)
153 return 1 << (ipv6_get_dsfield(ipv6h) & INET_ECN_MASK);
156 static unsigned int nf_hash_frag(__be32 id, const struct in6_addr *saddr,
157 const struct in6_addr *daddr)
159 net_get_random_once(&nf_frags.rnd, sizeof(nf_frags.rnd));
160 return jhash_3words(ipv6_addr_hash(saddr), ipv6_addr_hash(daddr),
161 (__force u32)id, nf_frags.rnd);
165 static unsigned int nf_hashfn(const struct inet_frag_queue *q)
167 const struct frag_queue *nq;
169 nq = container_of(q, struct frag_queue, q);
170 return nf_hash_frag(nq->id, &nq->saddr, &nq->daddr);
173 static void nf_skb_free(struct sk_buff *skb)
175 if (NFCT_FRAG6_CB(skb)->orig)
176 kfree_skb(NFCT_FRAG6_CB(skb)->orig);
179 static void nf_ct_frag6_expire(unsigned long data)
181 struct frag_queue *fq;
184 fq = container_of((struct inet_frag_queue *)data, struct frag_queue, q);
185 net = container_of(fq->q.net, struct net, nf_frag.frags);
187 ip6_expire_frag_queue(net, fq, &nf_frags);
190 /* Creation primitives. */
191 static inline struct frag_queue *fq_find(struct net *net, __be32 id,
192 u32 user, struct in6_addr *src,
193 struct in6_addr *dst, int iif, u8 ecn)
195 struct inet_frag_queue *q;
196 struct ip6_create_arg arg;
207 hash = nf_hash_frag(id, src, dst);
209 q = inet_frag_find(&net->nf_frag.frags, &nf_frags, &arg, hash);
211 if (IS_ERR_OR_NULL(q)) {
212 inet_frag_maybe_warn_overflow(q, pr_fmt());
215 return container_of(q, struct frag_queue, q);
219 static int nf_ct_frag6_queue(struct frag_queue *fq, struct sk_buff *skb,
220 const struct frag_hdr *fhdr, int nhoff)
222 struct sk_buff *prev, *next;
223 unsigned int payload_len;
227 if (fq->q.flags & INET_FRAG_COMPLETE) {
228 pr_debug("Already completed\n");
232 payload_len = ntohs(ipv6_hdr(skb)->payload_len);
234 offset = ntohs(fhdr->frag_off) & ~0x7;
235 end = offset + (payload_len -
236 ((u8 *)(fhdr + 1) - (u8 *)(ipv6_hdr(skb) + 1)));
238 if ((unsigned int)end > IPV6_MAXPLEN) {
239 pr_debug("offset is too large.\n");
243 ecn = ip6_frag_ecn(ipv6_hdr(skb));
245 if (skb->ip_summed == CHECKSUM_COMPLETE) {
246 const unsigned char *nh = skb_network_header(skb);
247 skb->csum = csum_sub(skb->csum,
248 csum_partial(nh, (u8 *)(fhdr + 1) - nh,
252 /* Is this the final fragment? */
253 if (!(fhdr->frag_off & htons(IP6_MF))) {
254 /* If we already have some bits beyond end
255 * or have different end, the segment is corrupted.
257 if (end < fq->q.len ||
258 ((fq->q.flags & INET_FRAG_LAST_IN) && end != fq->q.len)) {
259 pr_debug("already received last fragment\n");
262 fq->q.flags |= INET_FRAG_LAST_IN;
265 /* Check if the fragment is rounded to 8 bytes.
266 * Required by the RFC.
269 /* RFC2460 says always send parameter problem in
272 pr_debug("end of fragment not rounded to 8 bytes.\n");
275 if (end > fq->q.len) {
276 /* Some bits beyond end -> corruption. */
277 if (fq->q.flags & INET_FRAG_LAST_IN) {
278 pr_debug("last packet already reached.\n");
288 /* Point into the IP datagram 'data' part. */
289 if (!pskb_pull(skb, (u8 *) (fhdr + 1) - skb->data)) {
290 pr_debug("queue: message is too short.\n");
293 if (pskb_trim_rcsum(skb, end - offset)) {
294 pr_debug("Can't trim\n");
298 /* Find out which fragments are in front and at the back of us
299 * in the chain of fragments so far. We must know where to put
300 * this fragment, right?
302 prev = fq->q.fragments_tail;
303 if (!prev || NFCT_FRAG6_CB(prev)->offset < offset) {
308 for (next = fq->q.fragments; next != NULL; next = next->next) {
309 if (NFCT_FRAG6_CB(next)->offset >= offset)
315 /* RFC5722, Section 4:
316 * When reassembling an IPv6 datagram, if
317 * one or more its constituent fragments is determined to be an
318 * overlapping fragment, the entire datagram (and any constituent
319 * fragments, including those not yet received) MUST be silently
323 /* Check for overlap with preceding fragment. */
325 (NFCT_FRAG6_CB(prev)->offset + prev->len) > offset)
328 /* Look for overlap with succeeding segment. */
329 if (next && NFCT_FRAG6_CB(next)->offset < end)
332 NFCT_FRAG6_CB(skb)->offset = offset;
334 /* Insert this fragment in the chain of fragments. */
337 fq->q.fragments_tail = skb;
341 fq->q.fragments = skb;
344 fq->iif = skb->dev->ifindex;
347 fq->q.stamp = skb->tstamp;
348 fq->q.meat += skb->len;
350 if (payload_len > fq->q.max_size)
351 fq->q.max_size = payload_len;
352 add_frag_mem_limit(fq->q.net, skb->truesize);
354 /* The first fragment.
355 * nhoffset is obtained from the first fragment, of course.
358 fq->nhoffset = nhoff;
359 fq->q.flags |= INET_FRAG_FIRST_IN;
365 inet_frag_kill(&fq->q, &nf_frags);
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 frag_queue *fq, struct net_device *dev)
382 struct sk_buff *fp, *op, *head = fq->q.fragments;
386 inet_frag_kill(&fq->q, &nf_frags);
388 WARN_ON(head == NULL);
389 WARN_ON(NFCT_FRAG6_CB(head)->offset != 0);
391 ecn = ip_frag_ecn_table[fq->ecn];
392 if (unlikely(ecn == 0xff))
395 /* Unfragmented part is taken from the first segment. */
396 payload_len = ((head->data - skb_network_header(head)) -
397 sizeof(struct ipv6hdr) + fq->q.len -
398 sizeof(struct frag_hdr));
399 if (payload_len > IPV6_MAXPLEN) {
400 pr_debug("payload len is too large.\n");
404 /* Head of list must not be cloned. */
405 if (skb_unclone(head, GFP_ATOMIC)) {
406 pr_debug("skb is cloned but can't expand head");
410 /* If the first fragment is fragmented itself, we split
411 * it to two chunks: the first with data and paged part
412 * and the second, holding only fragments. */
413 if (skb_has_frag_list(head)) {
414 struct sk_buff *clone;
417 clone = alloc_skb(0, GFP_ATOMIC);
421 clone->next = head->next;
423 skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
424 skb_frag_list_init(head);
425 for (i = 0; i < skb_shinfo(head)->nr_frags; i++)
426 plen += skb_frag_size(&skb_shinfo(head)->frags[i]);
427 clone->len = clone->data_len = head->data_len - plen;
428 head->data_len -= clone->len;
429 head->len -= clone->len;
431 clone->ip_summed = head->ip_summed;
433 NFCT_FRAG6_CB(clone)->orig = NULL;
434 add_frag_mem_limit(fq->q.net, clone->truesize);
437 /* We have to remove fragment header from datagram and to relocate
438 * header in order to calculate ICV correctly. */
439 skb_network_header(head)[fq->nhoffset] = skb_transport_header(head)[0];
440 memmove(head->head + sizeof(struct frag_hdr), head->head,
441 (head->data - head->head) - sizeof(struct frag_hdr));
442 head->mac_header += sizeof(struct frag_hdr);
443 head->network_header += sizeof(struct frag_hdr);
445 skb_shinfo(head)->frag_list = head->next;
446 skb_reset_transport_header(head);
447 skb_push(head, head->data - skb_network_header(head));
449 for (fp = head->next; fp; fp = fp->next) {
450 head->data_len += fp->len;
451 head->len += fp->len;
452 if (head->ip_summed != fp->ip_summed)
453 head->ip_summed = CHECKSUM_NONE;
454 else if (head->ip_summed == CHECKSUM_COMPLETE)
455 head->csum = csum_add(head->csum, fp->csum);
456 head->truesize += fp->truesize;
458 sub_frag_mem_limit(fq->q.net, head->truesize);
463 head->tstamp = fq->q.stamp;
464 ipv6_hdr(head)->payload_len = htons(payload_len);
465 ipv6_change_dsfield(ipv6_hdr(head), 0xff, ecn);
466 IP6CB(head)->frag_max_size = sizeof(struct ipv6hdr) + fq->q.max_size;
468 /* Yes, and fold redundant checksum back. 8) */
469 if (head->ip_summed == CHECKSUM_COMPLETE)
470 head->csum = csum_partial(skb_network_header(head),
471 skb_network_header_len(head),
474 fq->q.fragments = NULL;
475 fq->q.fragments_tail = NULL;
477 /* all original skbs are linked into the NFCT_FRAG6_CB(head).orig */
478 fp = skb_shinfo(head)->frag_list;
479 if (fp && NFCT_FRAG6_CB(fp)->orig == NULL)
480 /* at above code, head skb is divided into two skbs. */
483 op = NFCT_FRAG6_CB(head)->orig;
484 for (; fp; fp = fp->next) {
485 struct sk_buff *orig = NFCT_FRAG6_CB(fp)->orig;
489 NFCT_FRAG6_CB(fp)->orig = NULL;
495 net_dbg_ratelimited("nf_ct_frag6_reasm: payload len = %d\n",
499 net_dbg_ratelimited("nf_ct_frag6_reasm: no memory for reassembly\n");
505 * find the header just before Fragment Header.
507 * if success return 0 and set ...
508 * (*prevhdrp): the value of "Next Header Field" in the header
509 * just before Fragment Header.
510 * (*prevhoff): the offset of "Next Header Field" in the header
511 * just before Fragment Header.
512 * (*fhoff) : the offset of Fragment Header.
514 * Based on ipv6_skip_hdr() in net/ipv6/exthdr.c
518 find_prev_fhdr(struct sk_buff *skb, u8 *prevhdrp, int *prevhoff, int *fhoff)
520 u8 nexthdr = ipv6_hdr(skb)->nexthdr;
521 const int netoff = skb_network_offset(skb);
522 u8 prev_nhoff = netoff + offsetof(struct ipv6hdr, nexthdr);
523 int start = netoff + sizeof(struct ipv6hdr);
524 int len = skb->len - start;
525 u8 prevhdr = NEXTHDR_IPV6;
527 while (nexthdr != NEXTHDR_FRAGMENT) {
528 struct ipv6_opt_hdr hdr;
531 if (!ipv6_ext_hdr(nexthdr)) {
534 if (nexthdr == NEXTHDR_NONE) {
535 pr_debug("next header is none\n");
538 if (len < (int)sizeof(struct ipv6_opt_hdr)) {
539 pr_debug("too short\n");
542 if (skb_copy_bits(skb, start, &hdr, sizeof(hdr)))
544 if (nexthdr == NEXTHDR_AUTH)
545 hdrlen = (hdr.hdrlen+2)<<2;
547 hdrlen = ipv6_optlen(&hdr);
552 nexthdr = hdr.nexthdr;
561 *prevhoff = prev_nhoff;
567 struct sk_buff *nf_ct_frag6_gather(struct net *net, struct sk_buff *skb, u32 user)
569 struct sk_buff *clone;
570 struct net_device *dev = skb->dev;
571 struct frag_hdr *fhdr;
572 struct frag_queue *fq;
576 struct sk_buff *ret_skb = NULL;
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 clone = skb_clone(skb, GFP_ATOMIC);
589 pr_debug("Can't clone skb\n");
593 NFCT_FRAG6_CB(clone)->orig = skb;
595 if (!pskb_may_pull(clone, fhoff + sizeof(*fhdr))) {
596 pr_debug("message is too short.\n");
600 skb_set_transport_header(clone, fhoff);
601 hdr = ipv6_hdr(clone);
602 fhdr = (struct frag_hdr *)skb_transport_header(clone);
604 fq = fq_find(net, fhdr->identification, user, &hdr->saddr, &hdr->daddr,
605 skb->dev ? skb->dev->ifindex : 0, ip6_frag_ecn(hdr));
607 pr_debug("Can't find and can't create new queue\n");
611 spin_lock_bh(&fq->q.lock);
613 if (nf_ct_frag6_queue(fq, clone, fhdr, nhoff) < 0) {
614 spin_unlock_bh(&fq->q.lock);
615 pr_debug("Can't insert skb to queue\n");
616 inet_frag_put(&fq->q, &nf_frags);
620 if (fq->q.flags == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
621 fq->q.meat == fq->q.len) {
622 ret_skb = nf_ct_frag6_reasm(fq, dev);
624 pr_debug("Can't reassemble fragmented packets\n");
626 spin_unlock_bh(&fq->q.lock);
628 inet_frag_put(&fq->q, &nf_frags);
635 EXPORT_SYMBOL_GPL(nf_ct_frag6_gather);
637 void nf_ct_frag6_consume_orig(struct sk_buff *skb)
639 struct sk_buff *s, *s2;
641 for (s = NFCT_FRAG6_CB(skb)->orig; s;) {
648 EXPORT_SYMBOL_GPL(nf_ct_frag6_consume_orig);
650 static int nf_ct_net_init(struct net *net)
654 net->nf_frag.frags.high_thresh = IPV6_FRAG_HIGH_THRESH;
655 net->nf_frag.frags.low_thresh = IPV6_FRAG_LOW_THRESH;
656 net->nf_frag.frags.timeout = IPV6_FRAG_TIMEOUT;
657 res = inet_frags_init_net(&net->nf_frag.frags);
660 res = nf_ct_frag6_sysctl_register(net);
662 inet_frags_uninit_net(&net->nf_frag.frags);
666 static void nf_ct_net_exit(struct net *net)
668 nf_ct_frags6_sysctl_unregister(net);
669 inet_frags_exit_net(&net->nf_frag.frags, &nf_frags);
672 static struct pernet_operations nf_ct_net_ops = {
673 .init = nf_ct_net_init,
674 .exit = nf_ct_net_exit,
677 int nf_ct_frag6_init(void)
681 nf_frags.hashfn = nf_hashfn;
682 nf_frags.constructor = ip6_frag_init;
683 nf_frags.destructor = NULL;
684 nf_frags.skb_free = nf_skb_free;
685 nf_frags.qsize = sizeof(struct frag_queue);
686 nf_frags.match = ip6_frag_match;
687 nf_frags.frag_expire = nf_ct_frag6_expire;
688 nf_frags.frags_cache_name = nf_frags_cache_name;
689 ret = inet_frags_init(&nf_frags);
692 ret = register_pernet_subsys(&nf_ct_net_ops);
694 inet_frags_fini(&nf_frags);
700 void nf_ct_frag6_cleanup(void)
702 unregister_pernet_subsys(&nf_ct_net_ops);
703 inet_frags_fini(&nf_frags);