2 * IPv6 fragment reassembly
3 * Linux INET6 implementation
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 static u8 ip6_frag_ecn(const struct ipv6hdr *ipv6h)
67 return 1 << (ipv6_get_dsfield(ipv6h) & INET_ECN_MASK);
70 static struct inet_frags ip6_frags;
72 static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *prev,
73 struct net_device *dev);
75 void ip6_frag_init(struct inet_frag_queue *q, const void *a)
77 struct frag_queue *fq = container_of(q, struct frag_queue, q);
78 const struct frag_v6_compare_key *key = a;
83 EXPORT_SYMBOL(ip6_frag_init);
85 void ip6_expire_frag_queue(struct net *net, struct frag_queue *fq)
87 struct net_device *dev = NULL;
91 spin_lock(&fq->q.lock);
93 if (fq->q.flags & INET_FRAG_COMPLETE)
96 inet_frag_kill(&fq->q);
98 dev = dev_get_by_index_rcu(net, fq->iif);
102 __IP6_INC_STATS(net, __in6_dev_get(dev), IPSTATS_MIB_REASMFAILS);
103 __IP6_INC_STATS(net, __in6_dev_get(dev), IPSTATS_MIB_REASMTIMEOUT);
105 /* Don't send error if the first segment did not arrive. */
106 head = fq->q.fragments;
107 if (!(fq->q.flags & INET_FRAG_FIRST_IN) || !head)
110 /* But use as source device on which LAST ARRIVED
111 * segment was received. And do not use fq->dev
112 * pointer directly, device might already disappeared.
116 spin_unlock(&fq->q.lock);
118 icmpv6_send(head, ICMPV6_TIME_EXCEED, ICMPV6_EXC_FRAGTIME, 0);
123 spin_unlock(&fq->q.lock);
126 inet_frag_put(&fq->q);
128 EXPORT_SYMBOL(ip6_expire_frag_queue);
130 static void ip6_frag_expire(struct timer_list *t)
132 struct inet_frag_queue *frag = from_timer(frag, t, timer);
133 struct frag_queue *fq;
136 fq = container_of(frag, struct frag_queue, q);
137 net = container_of(fq->q.net, struct net, ipv6.frags);
139 ip6_expire_frag_queue(net, fq);
142 static struct frag_queue *
143 fq_find(struct net *net, __be32 id, const struct ipv6hdr *hdr, int iif)
145 struct frag_v6_compare_key key = {
149 .user = IP6_DEFRAG_LOCAL_DELIVER,
152 struct inet_frag_queue *q;
154 if (!(ipv6_addr_type(&hdr->daddr) & (IPV6_ADDR_MULTICAST |
155 IPV6_ADDR_LINKLOCAL)))
158 q = inet_frag_find(&net->ipv6.frags, &key);
162 return container_of(q, struct frag_queue, q);
165 static int ip6_frag_queue(struct frag_queue *fq, struct sk_buff *skb,
166 struct frag_hdr *fhdr, int nhoff,
169 struct sk_buff *prev, *next;
170 struct net_device *dev;
171 int offset, end, fragsize;
172 struct net *net = dev_net(skb_dst(skb)->dev);
175 if (fq->q.flags & INET_FRAG_COMPLETE)
178 offset = ntohs(fhdr->frag_off) & ~0x7;
179 end = offset + (ntohs(ipv6_hdr(skb)->payload_len) -
180 ((u8 *)(fhdr + 1) - (u8 *)(ipv6_hdr(skb) + 1)));
182 if ((unsigned int)end > IPV6_MAXPLEN) {
183 *prob_offset = (u8 *)&fhdr->frag_off - skb_network_header(skb);
187 ecn = ip6_frag_ecn(ipv6_hdr(skb));
189 if (skb->ip_summed == CHECKSUM_COMPLETE) {
190 const unsigned char *nh = skb_network_header(skb);
191 skb->csum = csum_sub(skb->csum,
192 csum_partial(nh, (u8 *)(fhdr + 1) - nh,
196 /* Is this the final fragment? */
197 if (!(fhdr->frag_off & htons(IP6_MF))) {
198 /* If we already have some bits beyond end
199 * or have different end, the segment is corrupted.
201 if (end < fq->q.len ||
202 ((fq->q.flags & INET_FRAG_LAST_IN) && end != fq->q.len))
204 fq->q.flags |= INET_FRAG_LAST_IN;
207 /* Check if the fragment is rounded to 8 bytes.
208 * Required by the RFC.
211 /* RFC2460 says always send parameter problem in
214 *prob_offset = offsetof(struct ipv6hdr, payload_len);
217 if (end > fq->q.len) {
218 /* Some bits beyond end -> corruption. */
219 if (fq->q.flags & INET_FRAG_LAST_IN)
228 /* Point into the IP datagram 'data' part. */
229 if (!pskb_pull(skb, (u8 *) (fhdr + 1) - skb->data))
232 if (pskb_trim_rcsum(skb, end - offset))
235 /* Find out which fragments are in front and at the back of us
236 * in the chain of fragments so far. We must know where to put
237 * this fragment, right?
239 prev = fq->q.fragments_tail;
240 if (!prev || prev->ip_defrag_offset < offset) {
245 for (next = fq->q.fragments; next != NULL; next = next->next) {
246 if (next->ip_defrag_offset >= offset)
252 /* RFC5722, Section 4, amended by Errata ID : 3089
253 * When reassembling an IPv6 datagram, if
254 * one or more its constituent fragments is determined to be an
255 * overlapping fragment, the entire datagram (and any constituent
256 * fragments) MUST be silently discarded.
259 /* Check for overlap with preceding fragment. */
261 (prev->ip_defrag_offset + prev->len) > offset)
264 /* Look for overlap with succeeding segment. */
265 if (next && next->ip_defrag_offset < end)
268 /* Note : skb->ip_defrag_offset and skb->dev share the same location */
271 fq->iif = dev->ifindex;
272 /* Makes sure compiler wont do silly aliasing games */
274 skb->ip_defrag_offset = offset;
276 /* Insert this fragment in the chain of fragments. */
279 fq->q.fragments_tail = skb;
283 fq->q.fragments = skb;
285 fq->q.stamp = skb->tstamp;
286 fq->q.meat += skb->len;
288 add_frag_mem_limit(fq->q.net, skb->truesize);
290 fragsize = -skb_network_offset(skb) + skb->len;
291 if (fragsize > fq->q.max_size)
292 fq->q.max_size = fragsize;
294 /* The first fragment.
295 * nhoffset is obtained from the first fragment, of course.
298 fq->nhoffset = nhoff;
299 fq->q.flags |= INET_FRAG_FIRST_IN;
302 if (fq->q.flags == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
303 fq->q.meat == fq->q.len) {
305 unsigned long orefdst = skb->_skb_refdst;
307 skb->_skb_refdst = 0UL;
308 res = ip6_frag_reasm(fq, prev, dev);
309 skb->_skb_refdst = orefdst;
317 inet_frag_kill(&fq->q);
319 __IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
320 IPSTATS_MIB_REASMFAILS);
326 * Check if this packet is complete.
327 * Returns NULL on failure by any reason, and pointer
328 * to current nexthdr field in reassembled frame.
330 * It is called with locked fq, and caller must check that
331 * queue is eligible for reassembly i.e. it is not COMPLETE,
332 * the last and the first frames arrived and all the bits are here.
334 static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *prev,
335 struct net_device *dev)
337 struct net *net = container_of(fq->q.net, struct net, ipv6.frags);
338 struct sk_buff *fp, *head = fq->q.fragments;
344 inet_frag_kill(&fq->q);
346 ecn = ip_frag_ecn_table[fq->ecn];
347 if (unlikely(ecn == 0xff))
350 /* Make the one we just received the head. */
353 fp = skb_clone(head, GFP_ATOMIC);
358 fp->next = head->next;
360 fq->q.fragments_tail = fp;
363 skb_morph(head, fq->q.fragments);
364 head->next = fq->q.fragments->next;
366 consume_skb(fq->q.fragments);
367 fq->q.fragments = head;
370 WARN_ON(head == NULL);
371 WARN_ON(head->ip_defrag_offset != 0);
373 /* Unfragmented part is taken from the first segment. */
374 payload_len = ((head->data - skb_network_header(head)) -
375 sizeof(struct ipv6hdr) + fq->q.len -
376 sizeof(struct frag_hdr));
377 if (payload_len > IPV6_MAXPLEN)
380 /* Head of list must not be cloned. */
381 if (skb_unclone(head, GFP_ATOMIC))
384 /* If the first fragment is fragmented itself, we split
385 * it to two chunks: the first with data and paged part
386 * and the second, holding only fragments. */
387 if (skb_has_frag_list(head)) {
388 struct sk_buff *clone;
391 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;
405 add_frag_mem_limit(fq->q.net, clone->truesize);
408 /* We have to remove fragment header from datagram and to relocate
409 * header in order to calculate ICV correctly. */
410 nhoff = fq->nhoffset;
411 skb_network_header(head)[nhoff] = skb_transport_header(head)[0];
412 memmove(head->head + sizeof(struct frag_hdr), head->head,
413 (head->data - head->head) - sizeof(struct frag_hdr));
414 if (skb_mac_header_was_set(head))
415 head->mac_header += sizeof(struct frag_hdr);
416 head->network_header += sizeof(struct frag_hdr);
418 skb_reset_transport_header(head);
419 skb_push(head, head->data - skb_network_header(head));
421 sum_truesize = head->truesize;
422 for (fp = head->next; fp;) {
425 struct sk_buff *next = fp->next;
427 sum_truesize += fp->truesize;
428 if (head->ip_summed != fp->ip_summed)
429 head->ip_summed = CHECKSUM_NONE;
430 else if (head->ip_summed == CHECKSUM_COMPLETE)
431 head->csum = csum_add(head->csum, fp->csum);
433 if (skb_try_coalesce(head, fp, &headstolen, &delta)) {
434 kfree_skb_partial(fp, headstolen);
436 if (!skb_shinfo(head)->frag_list)
437 skb_shinfo(head)->frag_list = fp;
438 head->data_len += fp->len;
439 head->len += fp->len;
440 head->truesize += fp->truesize;
444 sub_frag_mem_limit(fq->q.net, sum_truesize);
448 head->tstamp = fq->q.stamp;
449 ipv6_hdr(head)->payload_len = htons(payload_len);
450 ipv6_change_dsfield(ipv6_hdr(head), 0xff, ecn);
451 IP6CB(head)->nhoff = nhoff;
452 IP6CB(head)->flags |= IP6SKB_FRAGMENTED;
453 IP6CB(head)->frag_max_size = fq->q.max_size;
455 /* Yes, and fold redundant checksum back. 8) */
456 skb_postpush_rcsum(head, skb_network_header(head),
457 skb_network_header_len(head));
460 __IP6_INC_STATS(net, __in6_dev_get(dev), IPSTATS_MIB_REASMOKS);
462 fq->q.fragments = NULL;
463 fq->q.fragments_tail = NULL;
467 net_dbg_ratelimited("ip6_frag_reasm: payload len = %d\n", payload_len);
470 net_dbg_ratelimited("ip6_frag_reasm: no memory for reassembly\n");
473 __IP6_INC_STATS(net, __in6_dev_get(dev), IPSTATS_MIB_REASMFAILS);
478 static int ipv6_frag_rcv(struct sk_buff *skb)
480 struct frag_hdr *fhdr;
481 struct frag_queue *fq;
482 const struct ipv6hdr *hdr = ipv6_hdr(skb);
483 struct net *net = dev_net(skb_dst(skb)->dev);
486 if (IP6CB(skb)->flags & IP6SKB_FRAGMENTED)
489 __IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMREQDS);
491 /* Jumbo payload inhibits frag. header */
492 if (hdr->payload_len == 0)
495 if (!pskb_may_pull(skb, (skb_transport_offset(skb) +
496 sizeof(struct frag_hdr))))
500 fhdr = (struct frag_hdr *)skb_transport_header(skb);
502 if (!(fhdr->frag_off & htons(0xFFF9))) {
503 /* It is not a fragmented frame */
504 skb->transport_header += sizeof(struct frag_hdr);
506 ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMOKS);
508 IP6CB(skb)->nhoff = (u8 *)fhdr - skb_network_header(skb);
509 IP6CB(skb)->flags |= IP6SKB_FRAGMENTED;
513 iif = skb->dev ? skb->dev->ifindex : 0;
514 fq = fq_find(net, fhdr->identification, hdr, iif);
519 spin_lock(&fq->q.lock);
522 ret = ip6_frag_queue(fq, skb, fhdr, IP6CB(skb)->nhoff,
525 spin_unlock(&fq->q.lock);
526 inet_frag_put(&fq->q);
528 __IP6_INC_STATS(net, __in6_dev_get_safely(skb->dev),
529 IPSTATS_MIB_INHDRERRORS);
530 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD, prob_offset);
535 __IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMFAILS);
540 __IP6_INC_STATS(net, __in6_dev_get_safely(skb->dev),
541 IPSTATS_MIB_INHDRERRORS);
542 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD, skb_network_header_len(skb));
546 static const struct inet6_protocol frag_protocol = {
547 .handler = ipv6_frag_rcv,
548 .flags = INET6_PROTO_NOPOLICY,
553 static struct ctl_table ip6_frags_ns_ctl_table[] = {
555 .procname = "ip6frag_high_thresh",
556 .data = &init_net.ipv6.frags.high_thresh,
557 .maxlen = sizeof(unsigned long),
559 .proc_handler = proc_doulongvec_minmax,
560 .extra1 = &init_net.ipv6.frags.low_thresh
563 .procname = "ip6frag_low_thresh",
564 .data = &init_net.ipv6.frags.low_thresh,
565 .maxlen = sizeof(unsigned long),
567 .proc_handler = proc_doulongvec_minmax,
568 .extra2 = &init_net.ipv6.frags.high_thresh
571 .procname = "ip6frag_time",
572 .data = &init_net.ipv6.frags.timeout,
573 .maxlen = sizeof(int),
575 .proc_handler = proc_dointvec_jiffies,
580 /* secret interval has been deprecated */
581 static int ip6_frags_secret_interval_unused;
582 static struct ctl_table ip6_frags_ctl_table[] = {
584 .procname = "ip6frag_secret_interval",
585 .data = &ip6_frags_secret_interval_unused,
586 .maxlen = sizeof(int),
588 .proc_handler = proc_dointvec_jiffies,
593 static int __net_init ip6_frags_ns_sysctl_register(struct net *net)
595 struct ctl_table *table;
596 struct ctl_table_header *hdr;
598 table = ip6_frags_ns_ctl_table;
599 if (!net_eq(net, &init_net)) {
600 table = kmemdup(table, sizeof(ip6_frags_ns_ctl_table), GFP_KERNEL);
604 table[0].data = &net->ipv6.frags.high_thresh;
605 table[0].extra1 = &net->ipv6.frags.low_thresh;
606 table[0].extra2 = &init_net.ipv6.frags.high_thresh;
607 table[1].data = &net->ipv6.frags.low_thresh;
608 table[1].extra2 = &net->ipv6.frags.high_thresh;
609 table[2].data = &net->ipv6.frags.timeout;
612 hdr = register_net_sysctl(net, "net/ipv6", table);
616 net->ipv6.sysctl.frags_hdr = hdr;
620 if (!net_eq(net, &init_net))
626 static void __net_exit ip6_frags_ns_sysctl_unregister(struct net *net)
628 struct ctl_table *table;
630 table = net->ipv6.sysctl.frags_hdr->ctl_table_arg;
631 unregister_net_sysctl_table(net->ipv6.sysctl.frags_hdr);
632 if (!net_eq(net, &init_net))
636 static struct ctl_table_header *ip6_ctl_header;
638 static int ip6_frags_sysctl_register(void)
640 ip6_ctl_header = register_net_sysctl(&init_net, "net/ipv6",
641 ip6_frags_ctl_table);
642 return ip6_ctl_header == NULL ? -ENOMEM : 0;
645 static void ip6_frags_sysctl_unregister(void)
647 unregister_net_sysctl_table(ip6_ctl_header);
650 static int ip6_frags_ns_sysctl_register(struct net *net)
655 static void ip6_frags_ns_sysctl_unregister(struct net *net)
659 static int ip6_frags_sysctl_register(void)
664 static void ip6_frags_sysctl_unregister(void)
669 static int __net_init ipv6_frags_init_net(struct net *net)
673 net->ipv6.frags.high_thresh = IPV6_FRAG_HIGH_THRESH;
674 net->ipv6.frags.low_thresh = IPV6_FRAG_LOW_THRESH;
675 net->ipv6.frags.timeout = IPV6_FRAG_TIMEOUT;
676 net->ipv6.frags.f = &ip6_frags;
678 res = inet_frags_init_net(&net->ipv6.frags);
682 res = ip6_frags_ns_sysctl_register(net);
684 inet_frags_exit_net(&net->ipv6.frags);
688 static void __net_exit ipv6_frags_exit_net(struct net *net)
690 ip6_frags_ns_sysctl_unregister(net);
691 inet_frags_exit_net(&net->ipv6.frags);
694 static struct pernet_operations ip6_frags_ops = {
695 .init = ipv6_frags_init_net,
696 .exit = ipv6_frags_exit_net,
699 static u32 ip6_key_hashfn(const void *data, u32 len, u32 seed)
702 sizeof(struct frag_v6_compare_key) / sizeof(u32), seed);
705 static u32 ip6_obj_hashfn(const void *data, u32 len, u32 seed)
707 const struct inet_frag_queue *fq = data;
709 return jhash2((const u32 *)&fq->key.v6,
710 sizeof(struct frag_v6_compare_key) / sizeof(u32), seed);
713 static int ip6_obj_cmpfn(struct rhashtable_compare_arg *arg, const void *ptr)
715 const struct frag_v6_compare_key *key = arg->key;
716 const struct inet_frag_queue *fq = ptr;
718 return !!memcmp(&fq->key, key, sizeof(*key));
721 const struct rhashtable_params ip6_rhash_params = {
722 .head_offset = offsetof(struct inet_frag_queue, node),
723 .hashfn = ip6_key_hashfn,
724 .obj_hashfn = ip6_obj_hashfn,
725 .obj_cmpfn = ip6_obj_cmpfn,
726 .automatic_shrinking = true,
728 EXPORT_SYMBOL(ip6_rhash_params);
730 int __init ipv6_frag_init(void)
734 ip6_frags.constructor = ip6_frag_init;
735 ip6_frags.destructor = NULL;
736 ip6_frags.qsize = sizeof(struct frag_queue);
737 ip6_frags.frag_expire = ip6_frag_expire;
738 ip6_frags.frags_cache_name = ip6_frag_cache_name;
739 ip6_frags.rhash_params = ip6_rhash_params;
740 ret = inet_frags_init(&ip6_frags);
744 ret = inet6_add_protocol(&frag_protocol, IPPROTO_FRAGMENT);
748 ret = ip6_frags_sysctl_register();
752 ret = register_pernet_subsys(&ip6_frags_ops);
760 ip6_frags_sysctl_unregister();
762 inet6_del_protocol(&frag_protocol, IPPROTO_FRAGMENT);
764 inet_frags_fini(&ip6_frags);
768 void ipv6_frag_exit(void)
770 inet_frags_fini(&ip6_frags);
771 ip6_frags_sysctl_unregister();
772 unregister_pernet_subsys(&ip6_frags_ops);
773 inet6_del_protocol(&frag_protocol, IPPROTO_FRAGMENT);