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;
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 long),
81 .proc_handler = proc_doulongvec_minmax,
82 .extra2 = &init_net.nf_frag.frags.high_thresh
85 .procname = "nf_conntrack_frag6_high_thresh",
86 .data = &init_net.nf_frag.frags.high_thresh,
87 .maxlen = sizeof(unsigned long),
89 .proc_handler = proc_doulongvec_minmax,
90 .extra1 = &init_net.nf_frag.frags.low_thresh
95 static int nf_ct_frag6_sysctl_register(struct net *net)
97 struct ctl_table *table;
98 struct ctl_table_header *hdr;
100 table = nf_ct_frag6_sysctl_table;
101 if (!net_eq(net, &init_net)) {
102 table = kmemdup(table, sizeof(nf_ct_frag6_sysctl_table),
107 table[0].data = &net->nf_frag.frags.timeout;
108 table[1].data = &net->nf_frag.frags.low_thresh;
109 table[1].extra2 = &net->nf_frag.frags.high_thresh;
110 table[2].data = &net->nf_frag.frags.high_thresh;
111 table[2].extra1 = &net->nf_frag.frags.low_thresh;
112 table[2].extra2 = &init_net.nf_frag.frags.high_thresh;
115 hdr = register_net_sysctl(net, "net/netfilter", table);
119 net->nf_frag_frags_hdr = hdr;
123 if (!net_eq(net, &init_net))
129 static void __net_exit nf_ct_frags6_sysctl_unregister(struct net *net)
131 struct ctl_table *table;
133 table = net->nf_frag_frags_hdr->ctl_table_arg;
134 unregister_net_sysctl_table(net->nf_frag_frags_hdr);
135 if (!net_eq(net, &init_net))
140 static int nf_ct_frag6_sysctl_register(struct net *net)
144 static void __net_exit nf_ct_frags6_sysctl_unregister(struct net *net)
149 static inline u8 ip6_frag_ecn(const struct ipv6hdr *ipv6h)
151 return 1 << (ipv6_get_dsfield(ipv6h) & INET_ECN_MASK);
154 static void nf_skb_free(struct sk_buff *skb)
156 if (NFCT_FRAG6_CB(skb)->orig)
157 kfree_skb(NFCT_FRAG6_CB(skb)->orig);
160 static void nf_ct_frag6_expire(unsigned long data)
162 struct frag_queue *fq;
165 fq = container_of((struct inet_frag_queue *)data, struct frag_queue, q);
166 net = container_of(fq->q.net, struct net, nf_frag.frags);
168 ip6_expire_frag_queue(net, fq);
171 /* Creation primitives. */
172 static struct frag_queue *fq_find(struct net *net, __be32 id, u32 user,
173 const struct ipv6hdr *hdr, int iif)
175 struct frag_v6_compare_key key = {
182 struct inet_frag_queue *q;
184 q = inet_frag_find(&net->nf_frag.frags, &key);
188 return container_of(q, struct frag_queue, q);
192 static int nf_ct_frag6_queue(struct frag_queue *fq, struct sk_buff *skb,
193 const struct frag_hdr *fhdr, int nhoff)
195 struct sk_buff *prev, *next;
196 unsigned int payload_len;
200 if (fq->q.flags & 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 ecn = ip6_frag_ecn(ipv6_hdr(skb));
218 if (skb->ip_summed == CHECKSUM_COMPLETE) {
219 const unsigned char *nh = skb_network_header(skb);
220 skb->csum = csum_sub(skb->csum,
221 csum_partial(nh, (u8 *)(fhdr + 1) - nh,
225 /* Is this the final fragment? */
226 if (!(fhdr->frag_off & htons(IP6_MF))) {
227 /* If we already have some bits beyond end
228 * or have different end, the segment is corrupted.
230 if (end < fq->q.len ||
231 ((fq->q.flags & INET_FRAG_LAST_IN) && end != fq->q.len)) {
232 pr_debug("already received last fragment\n");
235 fq->q.flags |= INET_FRAG_LAST_IN;
238 /* Check if the fragment is rounded to 8 bytes.
239 * Required by the RFC.
242 /* RFC2460 says always send parameter problem in
245 pr_debug("end of fragment not rounded to 8 bytes.\n");
248 if (end > fq->q.len) {
249 /* Some bits beyond end -> corruption. */
250 if (fq->q.flags & INET_FRAG_LAST_IN) {
251 pr_debug("last packet already reached.\n");
261 /* Point into the IP datagram 'data' part. */
262 if (!pskb_pull(skb, (u8 *) (fhdr + 1) - skb->data)) {
263 pr_debug("queue: message is too short.\n");
266 if (pskb_trim_rcsum(skb, end - offset)) {
267 pr_debug("Can't trim\n");
271 /* Find out which fragments are in front and at the back of us
272 * in the chain of fragments so far. We must know where to put
273 * this fragment, right?
275 prev = fq->q.fragments_tail;
276 if (!prev || NFCT_FRAG6_CB(prev)->offset < offset) {
281 for (next = fq->q.fragments; next != NULL; next = next->next) {
282 if (NFCT_FRAG6_CB(next)->offset >= offset)
288 /* RFC5722, Section 4:
289 * When reassembling an IPv6 datagram, if
290 * one or more its constituent fragments is determined to be an
291 * overlapping fragment, the entire datagram (and any constituent
292 * fragments, including those not yet received) MUST be silently
296 /* Check for overlap with preceding fragment. */
298 (NFCT_FRAG6_CB(prev)->offset + prev->len) > offset)
301 /* Look for overlap with succeeding segment. */
302 if (next && NFCT_FRAG6_CB(next)->offset < end)
305 NFCT_FRAG6_CB(skb)->offset = offset;
307 /* Insert this fragment in the chain of fragments. */
310 fq->q.fragments_tail = skb;
314 fq->q.fragments = skb;
317 fq->iif = skb->dev->ifindex;
320 fq->q.stamp = skb->tstamp;
321 fq->q.meat += skb->len;
323 if (payload_len > fq->q.max_size)
324 fq->q.max_size = payload_len;
325 add_frag_mem_limit(fq->q.net, skb->truesize);
327 /* The first fragment.
328 * nhoffset is obtained from the first fragment, of course.
331 fq->nhoffset = nhoff;
332 fq->q.flags |= INET_FRAG_FIRST_IN;
338 inet_frag_kill(&fq->q);
344 * Check if this packet is complete.
345 * Returns NULL on failure by any reason, and pointer
346 * to current nexthdr field in reassembled frame.
348 * It is called with locked fq, and caller must check that
349 * queue is eligible for reassembly i.e. it is not COMPLETE,
350 * the last and the first frames arrived and all the bits are here.
352 static struct sk_buff *
353 nf_ct_frag6_reasm(struct frag_queue *fq, struct net_device *dev)
355 struct sk_buff *fp, *op, *head = fq->q.fragments;
359 inet_frag_kill(&fq->q);
361 WARN_ON(head == NULL);
362 WARN_ON(NFCT_FRAG6_CB(head)->offset != 0);
364 ecn = ip_frag_ecn_table[fq->ecn];
365 if (unlikely(ecn == 0xff))
368 /* Unfragmented part is taken from the first segment. */
369 payload_len = ((head->data - skb_network_header(head)) -
370 sizeof(struct ipv6hdr) + fq->q.len -
371 sizeof(struct frag_hdr));
372 if (payload_len > IPV6_MAXPLEN) {
373 pr_debug("payload len is too large.\n");
377 /* Head of list must not be cloned. */
378 if (skb_unclone(head, GFP_ATOMIC)) {
379 pr_debug("skb is cloned but can't expand head");
383 /* If the first fragment is fragmented itself, we split
384 * it to two chunks: the first with data and paged part
385 * and the second, holding only fragments. */
386 if (skb_has_frag_list(head)) {
387 struct sk_buff *clone;
390 clone = alloc_skb(0, GFP_ATOMIC);
394 clone->next = head->next;
396 skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
397 skb_frag_list_init(head);
398 for (i = 0; i < skb_shinfo(head)->nr_frags; i++)
399 plen += skb_frag_size(&skb_shinfo(head)->frags[i]);
400 clone->len = clone->data_len = head->data_len - plen;
401 head->data_len -= clone->len;
402 head->len -= clone->len;
404 clone->ip_summed = head->ip_summed;
406 NFCT_FRAG6_CB(clone)->orig = NULL;
407 add_frag_mem_limit(fq->q.net, clone->truesize);
410 /* We have to remove fragment header from datagram and to relocate
411 * header in order to calculate ICV correctly. */
412 skb_network_header(head)[fq->nhoffset] = skb_transport_header(head)[0];
413 memmove(head->head + sizeof(struct frag_hdr), head->head,
414 (head->data - head->head) - sizeof(struct frag_hdr));
415 head->mac_header += sizeof(struct frag_hdr);
416 head->network_header += sizeof(struct frag_hdr);
418 skb_shinfo(head)->frag_list = head->next;
419 skb_reset_transport_header(head);
420 skb_push(head, head->data - skb_network_header(head));
422 for (fp = head->next; fp; fp = fp->next) {
423 head->data_len += fp->len;
424 head->len += fp->len;
425 if (head->ip_summed != fp->ip_summed)
426 head->ip_summed = CHECKSUM_NONE;
427 else if (head->ip_summed == CHECKSUM_COMPLETE)
428 head->csum = csum_add(head->csum, fp->csum);
429 head->truesize += fp->truesize;
432 sub_frag_mem_limit(fq->q.net, head->truesize);
437 head->tstamp = fq->q.stamp;
438 ipv6_hdr(head)->payload_len = htons(payload_len);
439 ipv6_change_dsfield(ipv6_hdr(head), 0xff, ecn);
440 IP6CB(head)->frag_max_size = sizeof(struct ipv6hdr) + fq->q.max_size;
442 /* Yes, and fold redundant checksum back. 8) */
443 if (head->ip_summed == CHECKSUM_COMPLETE)
444 head->csum = csum_partial(skb_network_header(head),
445 skb_network_header_len(head),
448 fq->q.fragments = NULL;
449 fq->q.rb_fragments = RB_ROOT;
450 fq->q.fragments_tail = NULL;
452 /* all original skbs are linked into the NFCT_FRAG6_CB(head).orig */
453 fp = skb_shinfo(head)->frag_list;
454 if (fp && NFCT_FRAG6_CB(fp)->orig == NULL)
455 /* at above code, head skb is divided into two skbs. */
458 op = NFCT_FRAG6_CB(head)->orig;
459 for (; fp; fp = fp->next) {
460 struct sk_buff *orig = NFCT_FRAG6_CB(fp)->orig;
464 NFCT_FRAG6_CB(fp)->orig = NULL;
470 net_dbg_ratelimited("nf_ct_frag6_reasm: payload len = %d\n",
474 net_dbg_ratelimited("nf_ct_frag6_reasm: no memory for reassembly\n");
480 * find the header just before Fragment Header.
482 * if success return 0 and set ...
483 * (*prevhdrp): the value of "Next Header Field" in the header
484 * just before Fragment Header.
485 * (*prevhoff): the offset of "Next Header Field" in the header
486 * just before Fragment Header.
487 * (*fhoff) : the offset of Fragment Header.
489 * Based on ipv6_skip_hdr() in net/ipv6/exthdr.c
493 find_prev_fhdr(struct sk_buff *skb, u8 *prevhdrp, int *prevhoff, int *fhoff)
495 u8 nexthdr = ipv6_hdr(skb)->nexthdr;
496 const int netoff = skb_network_offset(skb);
497 u8 prev_nhoff = netoff + offsetof(struct ipv6hdr, nexthdr);
498 int start = netoff + sizeof(struct ipv6hdr);
499 int len = skb->len - start;
500 u8 prevhdr = NEXTHDR_IPV6;
502 while (nexthdr != NEXTHDR_FRAGMENT) {
503 struct ipv6_opt_hdr hdr;
506 if (!ipv6_ext_hdr(nexthdr)) {
509 if (nexthdr == NEXTHDR_NONE) {
510 pr_debug("next header is none\n");
513 if (len < (int)sizeof(struct ipv6_opt_hdr)) {
514 pr_debug("too short\n");
517 if (skb_copy_bits(skb, start, &hdr, sizeof(hdr)))
519 if (nexthdr == NEXTHDR_AUTH)
520 hdrlen = (hdr.hdrlen+2)<<2;
522 hdrlen = ipv6_optlen(&hdr);
527 nexthdr = hdr.nexthdr;
536 *prevhoff = prev_nhoff;
542 struct sk_buff *nf_ct_frag6_gather(struct net *net, struct sk_buff *skb, u32 user)
544 struct sk_buff *clone;
545 struct net_device *dev = skb->dev;
546 struct frag_hdr *fhdr;
547 struct frag_queue *fq;
551 struct sk_buff *ret_skb = NULL;
553 /* Jumbo payload inhibits frag. header */
554 if (ipv6_hdr(skb)->payload_len == 0) {
555 pr_debug("payload len = 0\n");
559 if (find_prev_fhdr(skb, &prevhdr, &nhoff, &fhoff) < 0)
562 clone = skb_clone(skb, GFP_ATOMIC);
564 pr_debug("Can't clone skb\n");
568 NFCT_FRAG6_CB(clone)->orig = skb;
570 if (!pskb_may_pull(clone, fhoff + sizeof(*fhdr))) {
571 pr_debug("message is too short.\n");
575 skb_set_transport_header(clone, fhoff);
576 hdr = ipv6_hdr(clone);
577 fhdr = (struct frag_hdr *)skb_transport_header(clone);
579 if (clone->len - skb_network_offset(clone) < IPV6_MIN_MTU &&
580 fhdr->frag_off & htons(IP6_MF))
584 fq = fq_find(net, fhdr->identification, user, hdr,
585 skb->dev ? skb->dev->ifindex : 0);
587 pr_debug("Can't find and can't create new queue\n");
591 spin_lock_bh(&fq->q.lock);
593 if (nf_ct_frag6_queue(fq, clone, fhdr, nhoff) < 0) {
594 spin_unlock_bh(&fq->q.lock);
595 pr_debug("Can't insert skb to queue\n");
596 inet_frag_put(&fq->q);
600 if (fq->q.flags == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
601 fq->q.meat == fq->q.len) {
602 ret_skb = nf_ct_frag6_reasm(fq, dev);
604 pr_debug("Can't reassemble fragmented packets\n");
606 spin_unlock_bh(&fq->q.lock);
608 inet_frag_put(&fq->q);
615 EXPORT_SYMBOL_GPL(nf_ct_frag6_gather);
617 void nf_ct_frag6_consume_orig(struct sk_buff *skb)
619 struct sk_buff *s, *s2;
621 for (s = NFCT_FRAG6_CB(skb)->orig; s;) {
628 EXPORT_SYMBOL_GPL(nf_ct_frag6_consume_orig);
630 static int nf_ct_net_init(struct net *net)
634 net->nf_frag.frags.high_thresh = IPV6_FRAG_HIGH_THRESH;
635 net->nf_frag.frags.low_thresh = IPV6_FRAG_LOW_THRESH;
636 net->nf_frag.frags.timeout = IPV6_FRAG_TIMEOUT;
637 net->nf_frag.frags.f = &nf_frags;
639 res = inet_frags_init_net(&net->nf_frag.frags);
642 res = nf_ct_frag6_sysctl_register(net);
644 inet_frags_exit_net(&net->nf_frag.frags);
648 static void nf_ct_net_exit(struct net *net)
650 nf_ct_frags6_sysctl_unregister(net);
651 inet_frags_exit_net(&net->nf_frag.frags);
654 static struct pernet_operations nf_ct_net_ops = {
655 .init = nf_ct_net_init,
656 .exit = nf_ct_net_exit,
659 int nf_ct_frag6_init(void)
663 nf_frags.constructor = ip6_frag_init;
664 nf_frags.destructor = NULL;
665 nf_frags.skb_free = nf_skb_free;
666 nf_frags.qsize = sizeof(struct frag_queue);
667 nf_frags.frag_expire = nf_ct_frag6_expire;
668 nf_frags.frags_cache_name = nf_frags_cache_name;
669 nf_frags.rhash_params = ip6_rhash_params;
670 ret = inet_frags_init(&nf_frags);
673 ret = register_pernet_subsys(&nf_ct_net_ops);
675 inet_frags_fini(&nf_frags);
681 void nf_ct_frag6_cleanup(void)
683 unregister_pernet_subsys(&nf_ct_net_ops);
684 inet_frags_fini(&nf_frags);