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);
83 * callers should be careful not to use the hash value outside the ipfrag_lock
84 * as doing so could race with ipfrag_hash_rnd being recalculated.
86 static unsigned int inet6_hash_frag(__be32 id, const struct in6_addr *saddr,
87 const struct in6_addr *daddr)
89 net_get_random_once(&ip6_frags.rnd, sizeof(ip6_frags.rnd));
90 return jhash_3words(ipv6_addr_hash(saddr), ipv6_addr_hash(daddr),
91 (__force u32)id, ip6_frags.rnd);
94 static unsigned int ip6_hashfn(const struct inet_frag_queue *q)
96 const struct frag_queue *fq;
98 fq = container_of(q, struct frag_queue, q);
99 return inet6_hash_frag(fq->id, &fq->saddr, &fq->daddr);
102 bool ip6_frag_match(const struct inet_frag_queue *q, const void *a)
104 const struct frag_queue *fq;
105 const struct ip6_create_arg *arg = a;
107 fq = container_of(q, struct frag_queue, q);
108 return fq->id == arg->id &&
109 fq->user == arg->user &&
110 ipv6_addr_equal(&fq->saddr, arg->src) &&
111 ipv6_addr_equal(&fq->daddr, arg->dst) &&
112 (arg->iif == fq->iif ||
113 !(ipv6_addr_type(arg->dst) & (IPV6_ADDR_MULTICAST |
114 IPV6_ADDR_LINKLOCAL)));
116 EXPORT_SYMBOL(ip6_frag_match);
118 void ip6_frag_init(struct inet_frag_queue *q, const void *a)
120 struct frag_queue *fq = container_of(q, struct frag_queue, q);
121 const struct ip6_create_arg *arg = a;
124 fq->user = arg->user;
125 fq->saddr = *arg->src;
126 fq->daddr = *arg->dst;
129 EXPORT_SYMBOL(ip6_frag_init);
131 void ip6_expire_frag_queue(struct net *net, struct frag_queue *fq,
132 struct inet_frags *frags)
134 struct net_device *dev = NULL;
136 spin_lock(&fq->q.lock);
138 if (fq->q.flags & INET_FRAG_COMPLETE)
141 inet_frag_kill(&fq->q, frags);
144 dev = dev_get_by_index_rcu(net, fq->iif);
148 IP6_INC_STATS_BH(net, __in6_dev_get(dev), IPSTATS_MIB_REASMFAILS);
150 if (inet_frag_evicting(&fq->q))
153 IP6_INC_STATS_BH(net, __in6_dev_get(dev), IPSTATS_MIB_REASMTIMEOUT);
155 /* Don't send error if the first segment did not arrive. */
156 if (!(fq->q.flags & INET_FRAG_FIRST_IN) || !fq->q.fragments)
159 /* But use as source device on which LAST ARRIVED
160 * segment was received. And do not use fq->dev
161 * pointer directly, device might already disappeared.
163 fq->q.fragments->dev = dev;
164 icmpv6_send(fq->q.fragments, ICMPV6_TIME_EXCEED, ICMPV6_EXC_FRAGTIME, 0);
168 spin_unlock(&fq->q.lock);
169 inet_frag_put(&fq->q, frags);
171 EXPORT_SYMBOL(ip6_expire_frag_queue);
173 static void ip6_frag_expire(unsigned long data)
175 struct frag_queue *fq;
178 fq = container_of((struct inet_frag_queue *)data, struct frag_queue, q);
179 net = container_of(fq->q.net, struct net, ipv6.frags);
181 ip6_expire_frag_queue(net, fq, &ip6_frags);
184 static struct frag_queue *
185 fq_find(struct net *net, __be32 id, const struct in6_addr *src,
186 const struct in6_addr *dst, int iif, u8 ecn)
188 struct inet_frag_queue *q;
189 struct ip6_create_arg arg;
193 arg.user = IP6_DEFRAG_LOCAL_DELIVER;
199 hash = inet6_hash_frag(id, src, dst);
201 q = inet_frag_find(&net->ipv6.frags, &ip6_frags, &arg, hash);
202 if (IS_ERR_OR_NULL(q)) {
203 inet_frag_maybe_warn_overflow(q, pr_fmt());
206 return container_of(q, struct frag_queue, q);
209 static int ip6_frag_queue(struct frag_queue *fq, struct sk_buff *skb,
210 struct frag_hdr *fhdr, int nhoff)
212 struct sk_buff *prev, *next;
213 struct net_device *dev;
215 struct net *net = dev_net(skb_dst(skb)->dev);
218 if (fq->q.flags & INET_FRAG_COMPLETE)
221 offset = ntohs(fhdr->frag_off) & ~0x7;
222 end = offset + (ntohs(ipv6_hdr(skb)->payload_len) -
223 ((u8 *)(fhdr + 1) - (u8 *)(ipv6_hdr(skb) + 1)));
225 if ((unsigned int)end > IPV6_MAXPLEN) {
226 IP6_INC_STATS_BH(net, ip6_dst_idev(skb_dst(skb)),
227 IPSTATS_MIB_INHDRERRORS);
228 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
229 ((u8 *)&fhdr->frag_off -
230 skb_network_header(skb)));
234 ecn = ip6_frag_ecn(ipv6_hdr(skb));
236 if (skb->ip_summed == CHECKSUM_COMPLETE) {
237 const unsigned char *nh = skb_network_header(skb);
238 skb->csum = csum_sub(skb->csum,
239 csum_partial(nh, (u8 *)(fhdr + 1) - nh,
243 /* Is this the final fragment? */
244 if (!(fhdr->frag_off & htons(IP6_MF))) {
245 /* If we already have some bits beyond end
246 * or have different end, the segment is corrupted.
248 if (end < fq->q.len ||
249 ((fq->q.flags & INET_FRAG_LAST_IN) && end != fq->q.len))
251 fq->q.flags |= INET_FRAG_LAST_IN;
254 /* Check if the fragment is rounded to 8 bytes.
255 * Required by the RFC.
258 /* RFC2460 says always send parameter problem in
261 IP6_INC_STATS_BH(net, ip6_dst_idev(skb_dst(skb)),
262 IPSTATS_MIB_INHDRERRORS);
263 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
264 offsetof(struct ipv6hdr, payload_len));
267 if (end > fq->q.len) {
268 /* Some bits beyond end -> corruption. */
269 if (fq->q.flags & INET_FRAG_LAST_IN)
278 /* Point into the IP datagram 'data' part. */
279 if (!pskb_pull(skb, (u8 *) (fhdr + 1) - skb->data))
282 if (pskb_trim_rcsum(skb, end - offset))
285 /* Find out which fragments are in front and at the back of us
286 * in the chain of fragments so far. We must know where to put
287 * this fragment, right?
289 prev = fq->q.fragments_tail;
290 if (!prev || FRAG6_CB(prev)->offset < offset) {
295 for (next = fq->q.fragments; next != NULL; next = next->next) {
296 if (FRAG6_CB(next)->offset >= offset)
302 /* RFC5722, Section 4, amended by Errata ID : 3089
303 * When reassembling an IPv6 datagram, if
304 * one or more its constituent fragments is determined to be an
305 * overlapping fragment, the entire datagram (and any constituent
306 * fragments) MUST be silently discarded.
309 /* Check for overlap with preceding fragment. */
311 (FRAG6_CB(prev)->offset + prev->len) > offset)
314 /* Look for overlap with succeeding segment. */
315 if (next && FRAG6_CB(next)->offset < end)
318 FRAG6_CB(skb)->offset = offset;
320 /* Insert this fragment in the chain of fragments. */
323 fq->q.fragments_tail = skb;
327 fq->q.fragments = skb;
331 fq->iif = dev->ifindex;
334 fq->q.stamp = skb->tstamp;
335 fq->q.meat += skb->len;
337 add_frag_mem_limit(fq->q.net, skb->truesize);
339 /* The first fragment.
340 * nhoffset is obtained from the first fragment, of course.
343 fq->nhoffset = nhoff;
344 fq->q.flags |= INET_FRAG_FIRST_IN;
347 if (fq->q.flags == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
348 fq->q.meat == fq->q.len) {
350 unsigned long orefdst = skb->_skb_refdst;
352 skb->_skb_refdst = 0UL;
353 res = ip6_frag_reasm(fq, prev, dev);
354 skb->_skb_refdst = orefdst;
362 inet_frag_kill(&fq->q, &ip6_frags);
364 IP6_INC_STATS_BH(net, ip6_dst_idev(skb_dst(skb)),
365 IPSTATS_MIB_REASMFAILS);
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 int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *prev,
380 struct net_device *dev)
382 struct net *net = container_of(fq->q.net, struct net, ipv6.frags);
383 struct sk_buff *fp, *head = fq->q.fragments;
389 inet_frag_kill(&fq->q, &ip6_frags);
391 ecn = ip_frag_ecn_table[fq->ecn];
392 if (unlikely(ecn == 0xff))
395 /* Make the one we just received the head. */
398 fp = skb_clone(head, GFP_ATOMIC);
403 fp->next = head->next;
405 fq->q.fragments_tail = fp;
408 skb_morph(head, fq->q.fragments);
409 head->next = fq->q.fragments->next;
411 consume_skb(fq->q.fragments);
412 fq->q.fragments = head;
415 WARN_ON(head == NULL);
416 WARN_ON(FRAG6_CB(head)->offset != 0);
418 /* Unfragmented part is taken from the first segment. */
419 payload_len = ((head->data - skb_network_header(head)) -
420 sizeof(struct ipv6hdr) + fq->q.len -
421 sizeof(struct frag_hdr));
422 if (payload_len > IPV6_MAXPLEN)
425 /* Head of list must not be cloned. */
426 if (skb_unclone(head, GFP_ATOMIC))
429 /* If the first fragment is fragmented itself, we split
430 * it to two chunks: the first with data and paged part
431 * and the second, holding only fragments. */
432 if (skb_has_frag_list(head)) {
433 struct sk_buff *clone;
436 clone = alloc_skb(0, GFP_ATOMIC);
439 clone->next = head->next;
441 skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
442 skb_frag_list_init(head);
443 for (i = 0; i < skb_shinfo(head)->nr_frags; i++)
444 plen += skb_frag_size(&skb_shinfo(head)->frags[i]);
445 clone->len = clone->data_len = head->data_len - plen;
446 head->data_len -= clone->len;
447 head->len -= clone->len;
449 clone->ip_summed = head->ip_summed;
450 add_frag_mem_limit(fq->q.net, clone->truesize);
453 /* We have to remove fragment header from datagram and to relocate
454 * header in order to calculate ICV correctly. */
455 nhoff = fq->nhoffset;
456 skb_network_header(head)[nhoff] = skb_transport_header(head)[0];
457 memmove(head->head + sizeof(struct frag_hdr), head->head,
458 (head->data - head->head) - sizeof(struct frag_hdr));
459 head->mac_header += sizeof(struct frag_hdr);
460 head->network_header += sizeof(struct frag_hdr);
462 skb_reset_transport_header(head);
463 skb_push(head, head->data - skb_network_header(head));
465 sum_truesize = head->truesize;
466 for (fp = head->next; fp;) {
469 struct sk_buff *next = fp->next;
471 sum_truesize += fp->truesize;
472 if (head->ip_summed != fp->ip_summed)
473 head->ip_summed = CHECKSUM_NONE;
474 else if (head->ip_summed == CHECKSUM_COMPLETE)
475 head->csum = csum_add(head->csum, fp->csum);
477 if (skb_try_coalesce(head, fp, &headstolen, &delta)) {
478 kfree_skb_partial(fp, headstolen);
480 if (!skb_shinfo(head)->frag_list)
481 skb_shinfo(head)->frag_list = fp;
482 head->data_len += fp->len;
483 head->len += fp->len;
484 head->truesize += fp->truesize;
488 sub_frag_mem_limit(fq->q.net, sum_truesize);
492 head->tstamp = fq->q.stamp;
493 ipv6_hdr(head)->payload_len = htons(payload_len);
494 ipv6_change_dsfield(ipv6_hdr(head), 0xff, ecn);
495 IP6CB(head)->nhoff = nhoff;
496 IP6CB(head)->flags |= IP6SKB_FRAGMENTED;
498 /* Yes, and fold redundant checksum back. 8) */
499 skb_postpush_rcsum(head, skb_network_header(head),
500 skb_network_header_len(head));
503 IP6_INC_STATS_BH(net, __in6_dev_get(dev), IPSTATS_MIB_REASMOKS);
505 fq->q.fragments = NULL;
506 fq->q.fragments_tail = NULL;
510 net_dbg_ratelimited("ip6_frag_reasm: payload len = %d\n", payload_len);
513 net_dbg_ratelimited("ip6_frag_reasm: no memory for reassembly\n");
516 IP6_INC_STATS_BH(net, __in6_dev_get(dev), IPSTATS_MIB_REASMFAILS);
521 static int ipv6_frag_rcv(struct sk_buff *skb)
523 struct frag_hdr *fhdr;
524 struct frag_queue *fq;
525 const struct ipv6hdr *hdr = ipv6_hdr(skb);
526 struct net *net = dev_net(skb_dst(skb)->dev);
528 if (IP6CB(skb)->flags & IP6SKB_FRAGMENTED)
531 IP6_INC_STATS_BH(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMREQDS);
533 /* Jumbo payload inhibits frag. header */
534 if (hdr->payload_len == 0)
537 if (!pskb_may_pull(skb, (skb_transport_offset(skb) +
538 sizeof(struct frag_hdr))))
542 fhdr = (struct frag_hdr *)skb_transport_header(skb);
544 if (!(fhdr->frag_off & htons(0xFFF9))) {
545 /* It is not a fragmented frame */
546 skb->transport_header += sizeof(struct frag_hdr);
547 IP6_INC_STATS_BH(net,
548 ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMOKS);
550 IP6CB(skb)->nhoff = (u8 *)fhdr - skb_network_header(skb);
551 IP6CB(skb)->flags |= IP6SKB_FRAGMENTED;
555 fq = fq_find(net, fhdr->identification, &hdr->saddr, &hdr->daddr,
556 skb->dev ? skb->dev->ifindex : 0, ip6_frag_ecn(hdr));
560 spin_lock(&fq->q.lock);
562 ret = ip6_frag_queue(fq, skb, fhdr, IP6CB(skb)->nhoff);
564 spin_unlock(&fq->q.lock);
565 inet_frag_put(&fq->q, &ip6_frags);
569 IP6_INC_STATS_BH(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMFAILS);
574 IP6_INC_STATS_BH(net, ip6_dst_idev(skb_dst(skb)),
575 IPSTATS_MIB_INHDRERRORS);
576 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD, skb_network_header_len(skb));
580 static const struct inet6_protocol frag_protocol = {
581 .handler = ipv6_frag_rcv,
582 .flags = INET6_PROTO_NOPOLICY,
588 static struct ctl_table ip6_frags_ns_ctl_table[] = {
590 .procname = "ip6frag_high_thresh",
591 .data = &init_net.ipv6.frags.high_thresh,
592 .maxlen = sizeof(int),
594 .proc_handler = proc_dointvec_minmax,
595 .extra1 = &init_net.ipv6.frags.low_thresh
598 .procname = "ip6frag_low_thresh",
599 .data = &init_net.ipv6.frags.low_thresh,
600 .maxlen = sizeof(int),
602 .proc_handler = proc_dointvec_minmax,
604 .extra2 = &init_net.ipv6.frags.high_thresh
607 .procname = "ip6frag_time",
608 .data = &init_net.ipv6.frags.timeout,
609 .maxlen = sizeof(int),
611 .proc_handler = proc_dointvec_jiffies,
616 /* secret interval has been deprecated */
617 static int ip6_frags_secret_interval_unused;
618 static struct ctl_table ip6_frags_ctl_table[] = {
620 .procname = "ip6frag_secret_interval",
621 .data = &ip6_frags_secret_interval_unused,
622 .maxlen = sizeof(int),
624 .proc_handler = proc_dointvec_jiffies,
629 static int __net_init ip6_frags_ns_sysctl_register(struct net *net)
631 struct ctl_table *table;
632 struct ctl_table_header *hdr;
634 table = ip6_frags_ns_ctl_table;
635 if (!net_eq(net, &init_net)) {
636 table = kmemdup(table, sizeof(ip6_frags_ns_ctl_table), GFP_KERNEL);
640 table[0].data = &net->ipv6.frags.high_thresh;
641 table[0].extra1 = &net->ipv6.frags.low_thresh;
642 table[0].extra2 = &init_net.ipv6.frags.high_thresh;
643 table[1].data = &net->ipv6.frags.low_thresh;
644 table[1].extra2 = &net->ipv6.frags.high_thresh;
645 table[2].data = &net->ipv6.frags.timeout;
647 /* Don't export sysctls to unprivileged users */
648 if (net->user_ns != &init_user_ns)
649 table[0].procname = NULL;
652 hdr = register_net_sysctl(net, "net/ipv6", table);
656 net->ipv6.sysctl.frags_hdr = hdr;
660 if (!net_eq(net, &init_net))
666 static void __net_exit ip6_frags_ns_sysctl_unregister(struct net *net)
668 struct ctl_table *table;
670 table = net->ipv6.sysctl.frags_hdr->ctl_table_arg;
671 unregister_net_sysctl_table(net->ipv6.sysctl.frags_hdr);
672 if (!net_eq(net, &init_net))
676 static struct ctl_table_header *ip6_ctl_header;
678 static int ip6_frags_sysctl_register(void)
680 ip6_ctl_header = register_net_sysctl(&init_net, "net/ipv6",
681 ip6_frags_ctl_table);
682 return ip6_ctl_header == NULL ? -ENOMEM : 0;
685 static void ip6_frags_sysctl_unregister(void)
687 unregister_net_sysctl_table(ip6_ctl_header);
690 static int ip6_frags_ns_sysctl_register(struct net *net)
695 static void ip6_frags_ns_sysctl_unregister(struct net *net)
699 static int ip6_frags_sysctl_register(void)
704 static void ip6_frags_sysctl_unregister(void)
709 static int __net_init ipv6_frags_init_net(struct net *net)
713 net->ipv6.frags.high_thresh = IPV6_FRAG_HIGH_THRESH;
714 net->ipv6.frags.low_thresh = IPV6_FRAG_LOW_THRESH;
715 net->ipv6.frags.timeout = IPV6_FRAG_TIMEOUT;
717 res = inet_frags_init_net(&net->ipv6.frags);
720 res = ip6_frags_ns_sysctl_register(net);
722 inet_frags_uninit_net(&net->ipv6.frags);
726 static void __net_exit ipv6_frags_exit_net(struct net *net)
728 ip6_frags_ns_sysctl_unregister(net);
729 inet_frags_exit_net(&net->ipv6.frags, &ip6_frags);
732 static struct pernet_operations ip6_frags_ops = {
733 .init = ipv6_frags_init_net,
734 .exit = ipv6_frags_exit_net,
737 int __init ipv6_frag_init(void)
741 ret = inet6_add_protocol(&frag_protocol, IPPROTO_FRAGMENT);
745 ret = ip6_frags_sysctl_register();
749 ret = register_pernet_subsys(&ip6_frags_ops);
753 ip6_frags.hashfn = ip6_hashfn;
754 ip6_frags.constructor = ip6_frag_init;
755 ip6_frags.destructor = NULL;
756 ip6_frags.skb_free = NULL;
757 ip6_frags.qsize = sizeof(struct frag_queue);
758 ip6_frags.match = ip6_frag_match;
759 ip6_frags.frag_expire = ip6_frag_expire;
760 ip6_frags.frags_cache_name = ip6_frag_cache_name;
761 ret = inet_frags_init(&ip6_frags);
768 ip6_frags_sysctl_unregister();
770 inet6_del_protocol(&frag_protocol, IPPROTO_FRAGMENT);
774 void ipv6_frag_exit(void)
776 inet_frags_fini(&ip6_frags);
777 ip6_frags_sysctl_unregister();
778 unregister_pernet_subsys(&ip6_frags_ops);
779 inet6_del_protocol(&frag_protocol, IPPROTO_FRAGMENT);