2 * IPv6 fragment reassembly
3 * Linux INET6 implementation
6 * Pedro Roque <roque@di.fc.ul.pt>
8 * Based on: net/ipv4/ip_fragment.c
10 * This program is free software; you can redistribute it and/or
11 * modify it under the terms of the GNU General Public License
12 * as published by the Free Software Foundation; either version
13 * 2 of the License, or (at your option) any later version.
18 * Andi Kleen Make it work with multiple hosts.
19 * More RFC compliance.
21 * Horst von Brand Add missing #include <linux/string.h>
22 * Alexey Kuznetsov SMP races, threading, cleanup.
23 * Patrick McHardy LRU queue of frag heads for evictor.
24 * Mitsuru KANDA @USAGI Register inet6_protocol{}.
26 * YOSHIFUJI,H. @USAGI Always remove fragment header to
27 * calculate ICV correctly.
30 #define pr_fmt(fmt) "IPv6: " fmt
32 #include <linux/errno.h>
33 #include <linux/types.h>
34 #include <linux/string.h>
35 #include <linux/socket.h>
36 #include <linux/sockios.h>
37 #include <linux/jiffies.h>
38 #include <linux/net.h>
39 #include <linux/list.h>
40 #include <linux/netdevice.h>
41 #include <linux/in6.h>
42 #include <linux/ipv6.h>
43 #include <linux/icmpv6.h>
44 #include <linux/random.h>
45 #include <linux/jhash.h>
46 #include <linux/skbuff.h>
47 #include <linux/slab.h>
48 #include <linux/export.h>
54 #include <net/ip6_route.h>
55 #include <net/protocol.h>
56 #include <net/transp_v6.h>
57 #include <net/rawv6.h>
58 #include <net/ndisc.h>
59 #include <net/addrconf.h>
60 #include <net/inet_frag.h>
61 #include <net/inet_ecn.h>
63 static const char ip6_frag_cache_name[] = "ip6-frags";
65 struct ip6frag_skb_cb {
66 struct inet6_skb_parm h;
70 #define FRAG6_CB(skb) ((struct ip6frag_skb_cb *)((skb)->cb))
72 static u8 ip6_frag_ecn(const struct ipv6hdr *ipv6h)
74 return 1 << (ipv6_get_dsfield(ipv6h) & INET_ECN_MASK);
77 static struct inet_frags ip6_frags;
79 static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *prev,
80 struct net_device *dev);
82 void ip6_frag_init(struct inet_frag_queue *q, const void *a)
84 struct frag_queue *fq = container_of(q, struct frag_queue, q);
85 const struct frag_v6_compare_key *key = a;
90 EXPORT_SYMBOL(ip6_frag_init);
92 void ip6_expire_frag_queue(struct net *net, struct frag_queue *fq)
94 struct net_device *dev = NULL;
98 spin_lock(&fq->q.lock);
100 if (fq->q.flags & INET_FRAG_COMPLETE)
103 inet_frag_kill(&fq->q);
105 dev = dev_get_by_index_rcu(net, fq->iif);
109 IP6_INC_STATS_BH(net, __in6_dev_get(dev), IPSTATS_MIB_REASMFAILS);
110 IP6_INC_STATS_BH(net, __in6_dev_get(dev), IPSTATS_MIB_REASMTIMEOUT);
112 /* Don't send error if the first segment did not arrive. */
113 head = fq->q.fragments;
114 if (!(fq->q.flags & INET_FRAG_FIRST_IN) || !head)
117 /* But use as source device on which LAST ARRIVED
118 * segment was received. And do not use fq->dev
119 * pointer directly, device might already disappeared.
123 spin_unlock(&fq->q.lock);
125 icmpv6_send(head, ICMPV6_TIME_EXCEED, ICMPV6_EXC_FRAGTIME, 0);
130 spin_unlock(&fq->q.lock);
133 inet_frag_put(&fq->q);
135 EXPORT_SYMBOL(ip6_expire_frag_queue);
137 static void ip6_frag_expire(unsigned long data)
139 struct frag_queue *fq;
142 fq = container_of((struct inet_frag_queue *)data, struct frag_queue, q);
143 net = container_of(fq->q.net, struct net, ipv6.frags);
145 ip6_expire_frag_queue(net, fq);
148 static struct frag_queue *
149 fq_find(struct net *net, __be32 id, const struct ipv6hdr *hdr, int iif)
151 struct frag_v6_compare_key key = {
155 .user = IP6_DEFRAG_LOCAL_DELIVER,
158 struct inet_frag_queue *q;
160 if (!(ipv6_addr_type(&hdr->daddr) & (IPV6_ADDR_MULTICAST |
161 IPV6_ADDR_LINKLOCAL)))
164 q = inet_frag_find(&net->ipv6.frags, &key);
168 return container_of(q, struct frag_queue, q);
171 static int ip6_frag_queue(struct frag_queue *fq, struct sk_buff *skb,
172 struct frag_hdr *fhdr, int nhoff)
174 struct sk_buff *prev, *next;
175 struct net_device *dev;
177 struct net *net = dev_net(skb_dst(skb)->dev);
180 if (fq->q.flags & INET_FRAG_COMPLETE)
183 offset = ntohs(fhdr->frag_off) & ~0x7;
184 end = offset + (ntohs(ipv6_hdr(skb)->payload_len) -
185 ((u8 *)(fhdr + 1) - (u8 *)(ipv6_hdr(skb) + 1)));
187 if ((unsigned int)end > IPV6_MAXPLEN) {
188 IP6_INC_STATS_BH(net, ip6_dst_idev(skb_dst(skb)),
189 IPSTATS_MIB_INHDRERRORS);
190 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
191 ((u8 *)&fhdr->frag_off -
192 skb_network_header(skb)));
196 ecn = ip6_frag_ecn(ipv6_hdr(skb));
198 if (skb->ip_summed == CHECKSUM_COMPLETE) {
199 const unsigned char *nh = skb_network_header(skb);
200 skb->csum = csum_sub(skb->csum,
201 csum_partial(nh, (u8 *)(fhdr + 1) - nh,
205 /* Is this the final fragment? */
206 if (!(fhdr->frag_off & htons(IP6_MF))) {
207 /* If we already have some bits beyond end
208 * or have different end, the segment is corrupted.
210 if (end < fq->q.len ||
211 ((fq->q.flags & INET_FRAG_LAST_IN) && end != fq->q.len))
213 fq->q.flags |= INET_FRAG_LAST_IN;
216 /* Check if the fragment is rounded to 8 bytes.
217 * Required by the RFC.
220 /* RFC2460 says always send parameter problem in
223 IP6_INC_STATS_BH(net, ip6_dst_idev(skb_dst(skb)),
224 IPSTATS_MIB_INHDRERRORS);
225 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
226 offsetof(struct ipv6hdr, payload_len));
229 if (end > fq->q.len) {
230 /* Some bits beyond end -> corruption. */
231 if (fq->q.flags & INET_FRAG_LAST_IN)
240 /* Point into the IP datagram 'data' part. */
241 if (!pskb_pull(skb, (u8 *) (fhdr + 1) - skb->data))
244 if (pskb_trim_rcsum(skb, end - offset))
247 /* Find out which fragments are in front and at the back of us
248 * in the chain of fragments so far. We must know where to put
249 * this fragment, right?
251 prev = fq->q.fragments_tail;
252 if (!prev || FRAG6_CB(prev)->offset < offset) {
257 for (next = fq->q.fragments; next != NULL; next = next->next) {
258 if (FRAG6_CB(next)->offset >= offset)
264 /* RFC5722, Section 4, amended by Errata ID : 3089
265 * When reassembling an IPv6 datagram, if
266 * one or more its constituent fragments is determined to be an
267 * overlapping fragment, the entire datagram (and any constituent
268 * fragments) MUST be silently discarded.
271 /* Check for overlap with preceding fragment. */
273 (FRAG6_CB(prev)->offset + prev->len) > offset)
276 /* Look for overlap with succeeding segment. */
277 if (next && FRAG6_CB(next)->offset < end)
280 FRAG6_CB(skb)->offset = offset;
282 /* Insert this fragment in the chain of fragments. */
285 fq->q.fragments_tail = skb;
289 fq->q.fragments = skb;
293 fq->iif = dev->ifindex;
296 fq->q.stamp = skb->tstamp;
297 fq->q.meat += skb->len;
299 add_frag_mem_limit(fq->q.net, skb->truesize);
301 /* The first fragment.
302 * nhoffset is obtained from the first fragment, of course.
305 fq->nhoffset = nhoff;
306 fq->q.flags |= INET_FRAG_FIRST_IN;
309 if (fq->q.flags == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
310 fq->q.meat == fq->q.len) {
312 unsigned long orefdst = skb->_skb_refdst;
314 skb->_skb_refdst = 0UL;
315 res = ip6_frag_reasm(fq, prev, dev);
316 skb->_skb_refdst = orefdst;
324 inet_frag_kill(&fq->q);
326 IP6_INC_STATS_BH(net, ip6_dst_idev(skb_dst(skb)),
327 IPSTATS_MIB_REASMFAILS);
333 * Check if this packet is complete.
334 * Returns NULL on failure by any reason, and pointer
335 * to current nexthdr field in reassembled frame.
337 * It is called with locked fq, and caller must check that
338 * queue is eligible for reassembly i.e. it is not COMPLETE,
339 * the last and the first frames arrived and all the bits are here.
341 static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *prev,
342 struct net_device *dev)
344 struct net *net = container_of(fq->q.net, struct net, ipv6.frags);
345 struct sk_buff *fp, *head = fq->q.fragments;
351 inet_frag_kill(&fq->q);
353 ecn = ip_frag_ecn_table[fq->ecn];
354 if (unlikely(ecn == 0xff))
357 /* Make the one we just received the head. */
360 fp = skb_clone(head, GFP_ATOMIC);
365 fp->next = head->next;
367 fq->q.fragments_tail = fp;
370 skb_morph(head, fq->q.fragments);
371 head->next = fq->q.fragments->next;
373 consume_skb(fq->q.fragments);
374 fq->q.fragments = head;
377 WARN_ON(head == NULL);
378 WARN_ON(FRAG6_CB(head)->offset != 0);
380 /* Unfragmented part is taken from the first segment. */
381 payload_len = ((head->data - skb_network_header(head)) -
382 sizeof(struct ipv6hdr) + fq->q.len -
383 sizeof(struct frag_hdr));
384 if (payload_len > IPV6_MAXPLEN)
387 /* Head of list must not be cloned. */
388 if (skb_unclone(head, GFP_ATOMIC))
391 /* If the first fragment is fragmented itself, we split
392 * it to two chunks: the first with data and paged part
393 * and the second, holding only fragments. */
394 if (skb_has_frag_list(head)) {
395 struct sk_buff *clone;
398 clone = alloc_skb(0, GFP_ATOMIC);
401 clone->next = head->next;
403 skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
404 skb_frag_list_init(head);
405 for (i = 0; i < skb_shinfo(head)->nr_frags; i++)
406 plen += skb_frag_size(&skb_shinfo(head)->frags[i]);
407 clone->len = clone->data_len = head->data_len - plen;
408 head->data_len -= clone->len;
409 head->len -= clone->len;
411 clone->ip_summed = head->ip_summed;
412 add_frag_mem_limit(fq->q.net, clone->truesize);
415 /* We have to remove fragment header from datagram and to relocate
416 * header in order to calculate ICV correctly. */
417 nhoff = fq->nhoffset;
418 skb_network_header(head)[nhoff] = skb_transport_header(head)[0];
419 memmove(head->head + sizeof(struct frag_hdr), head->head,
420 (head->data - head->head) - sizeof(struct frag_hdr));
421 head->mac_header += sizeof(struct frag_hdr);
422 head->network_header += sizeof(struct frag_hdr);
424 skb_reset_transport_header(head);
425 skb_push(head, head->data - skb_network_header(head));
427 sum_truesize = head->truesize;
428 for (fp = head->next; fp;) {
431 struct sk_buff *next = fp->next;
433 sum_truesize += fp->truesize;
434 if (head->ip_summed != fp->ip_summed)
435 head->ip_summed = CHECKSUM_NONE;
436 else if (head->ip_summed == CHECKSUM_COMPLETE)
437 head->csum = csum_add(head->csum, fp->csum);
439 if (skb_try_coalesce(head, fp, &headstolen, &delta)) {
440 kfree_skb_partial(fp, headstolen);
442 if (!skb_shinfo(head)->frag_list)
443 skb_shinfo(head)->frag_list = fp;
444 head->data_len += fp->len;
445 head->len += fp->len;
446 head->truesize += fp->truesize;
450 sub_frag_mem_limit(fq->q.net, sum_truesize);
454 head->tstamp = fq->q.stamp;
455 ipv6_hdr(head)->payload_len = htons(payload_len);
456 ipv6_change_dsfield(ipv6_hdr(head), 0xff, ecn);
457 IP6CB(head)->nhoff = nhoff;
458 IP6CB(head)->flags |= IP6SKB_FRAGMENTED;
460 /* Yes, and fold redundant checksum back. 8) */
461 skb_postpush_rcsum(head, skb_network_header(head),
462 skb_network_header_len(head));
465 IP6_INC_STATS_BH(net, __in6_dev_get(dev), IPSTATS_MIB_REASMOKS);
467 fq->q.fragments = NULL;
468 fq->q.rb_fragments = RB_ROOT;
469 fq->q.fragments_tail = NULL;
473 net_dbg_ratelimited("ip6_frag_reasm: payload len = %d\n", payload_len);
476 net_dbg_ratelimited("ip6_frag_reasm: no memory for reassembly\n");
479 IP6_INC_STATS_BH(net, __in6_dev_get(dev), IPSTATS_MIB_REASMFAILS);
484 static int ipv6_frag_rcv(struct sk_buff *skb)
486 struct frag_hdr *fhdr;
487 struct frag_queue *fq;
488 const struct ipv6hdr *hdr = ipv6_hdr(skb);
489 struct net *net = dev_net(skb_dst(skb)->dev);
492 if (IP6CB(skb)->flags & IP6SKB_FRAGMENTED)
495 IP6_INC_STATS_BH(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMREQDS);
497 /* Jumbo payload inhibits frag. header */
498 if (hdr->payload_len == 0)
501 if (!pskb_may_pull(skb, (skb_transport_offset(skb) +
502 sizeof(struct frag_hdr))))
506 fhdr = (struct frag_hdr *)skb_transport_header(skb);
508 if (!(fhdr->frag_off & htons(0xFFF9))) {
509 /* It is not a fragmented frame */
510 skb->transport_header += sizeof(struct frag_hdr);
511 IP6_INC_STATS_BH(net,
512 ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMOKS);
514 IP6CB(skb)->nhoff = (u8 *)fhdr - skb_network_header(skb);
515 IP6CB(skb)->flags |= IP6SKB_FRAGMENTED;
519 if (skb->len - skb_network_offset(skb) < IPV6_MIN_MTU &&
520 fhdr->frag_off & htons(IP6_MF))
523 iif = skb->dev ? skb->dev->ifindex : 0;
524 fq = fq_find(net, fhdr->identification, hdr, iif);
528 spin_lock(&fq->q.lock);
531 ret = ip6_frag_queue(fq, skb, fhdr, IP6CB(skb)->nhoff);
533 spin_unlock(&fq->q.lock);
534 inet_frag_put(&fq->q);
538 IP6_INC_STATS_BH(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMFAILS);
543 IP6_INC_STATS_BH(net, ip6_dst_idev(skb_dst(skb)),
544 IPSTATS_MIB_INHDRERRORS);
545 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD, skb_network_header_len(skb));
549 static const struct inet6_protocol frag_protocol = {
550 .handler = ipv6_frag_rcv,
551 .flags = INET6_PROTO_NOPOLICY,
556 static struct ctl_table ip6_frags_ns_ctl_table[] = {
558 .procname = "ip6frag_high_thresh",
559 .data = &init_net.ipv6.frags.high_thresh,
560 .maxlen = sizeof(unsigned long),
562 .proc_handler = proc_doulongvec_minmax,
563 .extra1 = &init_net.ipv6.frags.low_thresh
566 .procname = "ip6frag_low_thresh",
567 .data = &init_net.ipv6.frags.low_thresh,
568 .maxlen = sizeof(unsigned long),
570 .proc_handler = proc_doulongvec_minmax,
571 .extra2 = &init_net.ipv6.frags.high_thresh
574 .procname = "ip6frag_time",
575 .data = &init_net.ipv6.frags.timeout,
576 .maxlen = sizeof(int),
578 .proc_handler = proc_dointvec_jiffies,
583 /* secret interval has been deprecated */
584 static int ip6_frags_secret_interval_unused;
585 static struct ctl_table ip6_frags_ctl_table[] = {
587 .procname = "ip6frag_secret_interval",
588 .data = &ip6_frags_secret_interval_unused,
589 .maxlen = sizeof(int),
591 .proc_handler = proc_dointvec_jiffies,
596 static int __net_init ip6_frags_ns_sysctl_register(struct net *net)
598 struct ctl_table *table;
599 struct ctl_table_header *hdr;
601 table = ip6_frags_ns_ctl_table;
602 if (!net_eq(net, &init_net)) {
603 table = kmemdup(table, sizeof(ip6_frags_ns_ctl_table), GFP_KERNEL);
607 table[0].data = &net->ipv6.frags.high_thresh;
608 table[0].extra1 = &net->ipv6.frags.low_thresh;
609 table[0].extra2 = &init_net.ipv6.frags.high_thresh;
610 table[1].data = &net->ipv6.frags.low_thresh;
611 table[1].extra2 = &net->ipv6.frags.high_thresh;
612 table[2].data = &net->ipv6.frags.timeout;
614 /* Don't export sysctls to unprivileged users */
615 if (net->user_ns != &init_user_ns)
616 table[0].procname = NULL;
619 hdr = register_net_sysctl(net, "net/ipv6", table);
623 net->ipv6.sysctl.frags_hdr = hdr;
627 if (!net_eq(net, &init_net))
633 static void __net_exit ip6_frags_ns_sysctl_unregister(struct net *net)
635 struct ctl_table *table;
637 table = net->ipv6.sysctl.frags_hdr->ctl_table_arg;
638 unregister_net_sysctl_table(net->ipv6.sysctl.frags_hdr);
639 if (!net_eq(net, &init_net))
643 static struct ctl_table_header *ip6_ctl_header;
645 static int ip6_frags_sysctl_register(void)
647 ip6_ctl_header = register_net_sysctl(&init_net, "net/ipv6",
648 ip6_frags_ctl_table);
649 return ip6_ctl_header == NULL ? -ENOMEM : 0;
652 static void ip6_frags_sysctl_unregister(void)
654 unregister_net_sysctl_table(ip6_ctl_header);
657 static int ip6_frags_ns_sysctl_register(struct net *net)
662 static void ip6_frags_ns_sysctl_unregister(struct net *net)
666 static int ip6_frags_sysctl_register(void)
671 static void ip6_frags_sysctl_unregister(void)
676 static int __net_init ipv6_frags_init_net(struct net *net)
680 net->ipv6.frags.high_thresh = IPV6_FRAG_HIGH_THRESH;
681 net->ipv6.frags.low_thresh = IPV6_FRAG_LOW_THRESH;
682 net->ipv6.frags.timeout = IPV6_FRAG_TIMEOUT;
683 net->ipv6.frags.f = &ip6_frags;
685 res = inet_frags_init_net(&net->ipv6.frags);
689 res = ip6_frags_ns_sysctl_register(net);
691 inet_frags_exit_net(&net->ipv6.frags);
695 static void __net_exit ipv6_frags_exit_net(struct net *net)
697 ip6_frags_ns_sysctl_unregister(net);
698 inet_frags_exit_net(&net->ipv6.frags);
701 static struct pernet_operations ip6_frags_ops = {
702 .init = ipv6_frags_init_net,
703 .exit = ipv6_frags_exit_net,
706 static u32 ip6_key_hashfn(const void *data, u32 len, u32 seed)
709 sizeof(struct frag_v6_compare_key) / sizeof(u32), seed);
712 static u32 ip6_obj_hashfn(const void *data, u32 len, u32 seed)
714 const struct inet_frag_queue *fq = data;
716 return jhash2((const u32 *)&fq->key.v6,
717 sizeof(struct frag_v6_compare_key) / sizeof(u32), seed);
720 static int ip6_obj_cmpfn(struct rhashtable_compare_arg *arg, const void *ptr)
722 const struct frag_v6_compare_key *key = arg->key;
723 const struct inet_frag_queue *fq = ptr;
725 return !!memcmp(&fq->key, key, sizeof(*key));
728 const struct rhashtable_params ip6_rhash_params = {
729 .head_offset = offsetof(struct inet_frag_queue, node),
730 .hashfn = ip6_key_hashfn,
731 .obj_hashfn = ip6_obj_hashfn,
732 .obj_cmpfn = ip6_obj_cmpfn,
733 .automatic_shrinking = true,
735 EXPORT_SYMBOL(ip6_rhash_params);
737 int __init ipv6_frag_init(void)
741 ip6_frags.constructor = ip6_frag_init;
742 ip6_frags.destructor = NULL;
743 ip6_frags.qsize = sizeof(struct frag_queue);
744 ip6_frags.frag_expire = ip6_frag_expire;
745 ip6_frags.frags_cache_name = ip6_frag_cache_name;
746 ip6_frags.rhash_params = ip6_rhash_params;
747 ret = inet_frags_init(&ip6_frags);
751 ret = inet6_add_protocol(&frag_protocol, IPPROTO_FRAGMENT);
755 ret = ip6_frags_sysctl_register();
759 ret = register_pernet_subsys(&ip6_frags_ops);
767 ip6_frags_sysctl_unregister();
769 inet6_del_protocol(&frag_protocol, IPPROTO_FRAGMENT);
771 inet_frags_fini(&ip6_frags);
775 void ipv6_frag_exit(void)
777 ip6_frags_sysctl_unregister();
778 unregister_pernet_subsys(&ip6_frags_ops);
779 inet6_del_protocol(&frag_protocol, IPPROTO_FRAGMENT);
780 inet_frags_fini(&ip6_frags);