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/ipv6_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 static void ip6_frag_expire(struct timer_list *t)
77 struct inet_frag_queue *frag = from_timer(frag, t, timer);
78 struct frag_queue *fq;
81 fq = container_of(frag, struct frag_queue, q);
82 net = container_of(fq->q.net, struct net, ipv6.frags);
84 ip6frag_expire_frag_queue(net, fq);
87 static struct frag_queue *
88 fq_find(struct net *net, __be32 id, const struct ipv6hdr *hdr, int iif)
90 struct frag_v6_compare_key key = {
94 .user = IP6_DEFRAG_LOCAL_DELIVER,
97 struct inet_frag_queue *q;
99 if (!(ipv6_addr_type(&hdr->daddr) & (IPV6_ADDR_MULTICAST |
100 IPV6_ADDR_LINKLOCAL)))
103 q = inet_frag_find(&net->ipv6.frags, &key);
107 return container_of(q, struct frag_queue, q);
110 static int ip6_frag_queue(struct frag_queue *fq, struct sk_buff *skb,
111 struct frag_hdr *fhdr, int nhoff,
114 struct sk_buff *prev, *next;
115 struct net_device *dev;
116 int offset, end, fragsize;
117 struct net *net = dev_net(skb_dst(skb)->dev);
120 if (fq->q.flags & INET_FRAG_COMPLETE)
123 offset = ntohs(fhdr->frag_off) & ~0x7;
124 end = offset + (ntohs(ipv6_hdr(skb)->payload_len) -
125 ((u8 *)(fhdr + 1) - (u8 *)(ipv6_hdr(skb) + 1)));
127 if ((unsigned int)end > IPV6_MAXPLEN) {
128 *prob_offset = (u8 *)&fhdr->frag_off - skb_network_header(skb);
132 ecn = ip6_frag_ecn(ipv6_hdr(skb));
134 if (skb->ip_summed == CHECKSUM_COMPLETE) {
135 const unsigned char *nh = skb_network_header(skb);
136 skb->csum = csum_sub(skb->csum,
137 csum_partial(nh, (u8 *)(fhdr + 1) - nh,
141 /* Is this the final fragment? */
142 if (!(fhdr->frag_off & htons(IP6_MF))) {
143 /* If we already have some bits beyond end
144 * or have different end, the segment is corrupted.
146 if (end < fq->q.len ||
147 ((fq->q.flags & INET_FRAG_LAST_IN) && end != fq->q.len))
149 fq->q.flags |= INET_FRAG_LAST_IN;
152 /* Check if the fragment is rounded to 8 bytes.
153 * Required by the RFC.
156 /* RFC2460 says always send parameter problem in
159 *prob_offset = offsetof(struct ipv6hdr, payload_len);
162 if (end > fq->q.len) {
163 /* Some bits beyond end -> corruption. */
164 if (fq->q.flags & INET_FRAG_LAST_IN)
173 /* Point into the IP datagram 'data' part. */
174 if (!pskb_pull(skb, (u8 *) (fhdr + 1) - skb->data))
177 if (pskb_trim_rcsum(skb, end - offset))
180 /* Find out which fragments are in front and at the back of us
181 * in the chain of fragments so far. We must know where to put
182 * this fragment, right?
184 prev = fq->q.fragments_tail;
185 if (!prev || prev->ip_defrag_offset < offset) {
190 for (next = fq->q.fragments; next != NULL; next = next->next) {
191 if (next->ip_defrag_offset >= offset)
197 /* RFC5722, Section 4, amended by Errata ID : 3089
198 * When reassembling an IPv6 datagram, if
199 * one or more its constituent fragments is determined to be an
200 * overlapping fragment, the entire datagram (and any constituent
201 * fragments) MUST be silently discarded.
204 /* Check for overlap with preceding fragment. */
206 (prev->ip_defrag_offset + prev->len) > offset)
209 /* Look for overlap with succeeding segment. */
210 if (next && next->ip_defrag_offset < end)
213 /* Note : skb->ip_defrag_offset and skb->dev share the same location */
216 fq->iif = dev->ifindex;
217 /* Makes sure compiler wont do silly aliasing games */
219 skb->ip_defrag_offset = offset;
221 /* Insert this fragment in the chain of fragments. */
224 fq->q.fragments_tail = skb;
228 fq->q.fragments = skb;
230 fq->q.stamp = skb->tstamp;
231 fq->q.meat += skb->len;
233 add_frag_mem_limit(fq->q.net, skb->truesize);
235 fragsize = -skb_network_offset(skb) + skb->len;
236 if (fragsize > fq->q.max_size)
237 fq->q.max_size = fragsize;
239 /* The first fragment.
240 * nhoffset is obtained from the first fragment, of course.
243 fq->nhoffset = nhoff;
244 fq->q.flags |= INET_FRAG_FIRST_IN;
247 if (fq->q.flags == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
248 fq->q.meat == fq->q.len) {
250 unsigned long orefdst = skb->_skb_refdst;
252 skb->_skb_refdst = 0UL;
253 res = ip6_frag_reasm(fq, prev, dev);
254 skb->_skb_refdst = orefdst;
262 inet_frag_kill(&fq->q);
264 __IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
265 IPSTATS_MIB_REASMFAILS);
271 * Check if this packet is complete.
272 * Returns NULL on failure by any reason, and pointer
273 * to current nexthdr field in reassembled frame.
275 * It is called with locked fq, and caller must check that
276 * queue is eligible for reassembly i.e. it is not COMPLETE,
277 * the last and the first frames arrived and all the bits are here.
279 static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *prev,
280 struct net_device *dev)
282 struct net *net = container_of(fq->q.net, struct net, ipv6.frags);
283 struct sk_buff *fp, *head = fq->q.fragments;
289 inet_frag_kill(&fq->q);
291 ecn = ip_frag_ecn_table[fq->ecn];
292 if (unlikely(ecn == 0xff))
295 /* Make the one we just received the head. */
298 fp = skb_clone(head, GFP_ATOMIC);
303 fp->next = head->next;
305 fq->q.fragments_tail = fp;
308 skb_morph(head, fq->q.fragments);
309 head->next = fq->q.fragments->next;
311 consume_skb(fq->q.fragments);
312 fq->q.fragments = head;
315 WARN_ON(head == NULL);
316 WARN_ON(head->ip_defrag_offset != 0);
318 /* Unfragmented part is taken from the first segment. */
319 payload_len = ((head->data - skb_network_header(head)) -
320 sizeof(struct ipv6hdr) + fq->q.len -
321 sizeof(struct frag_hdr));
322 if (payload_len > IPV6_MAXPLEN)
325 /* Head of list must not be cloned. */
326 if (skb_unclone(head, GFP_ATOMIC))
329 /* If the first fragment is fragmented itself, we split
330 * it to two chunks: the first with data and paged part
331 * and the second, holding only fragments. */
332 if (skb_has_frag_list(head)) {
333 struct sk_buff *clone;
336 clone = alloc_skb(0, GFP_ATOMIC);
339 clone->next = head->next;
341 skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
342 skb_frag_list_init(head);
343 for (i = 0; i < skb_shinfo(head)->nr_frags; i++)
344 plen += skb_frag_size(&skb_shinfo(head)->frags[i]);
345 clone->len = clone->data_len = head->data_len - plen;
346 head->data_len -= clone->len;
347 head->len -= clone->len;
349 clone->ip_summed = head->ip_summed;
350 add_frag_mem_limit(fq->q.net, clone->truesize);
353 /* We have to remove fragment header from datagram and to relocate
354 * header in order to calculate ICV correctly. */
355 nhoff = fq->nhoffset;
356 skb_network_header(head)[nhoff] = skb_transport_header(head)[0];
357 memmove(head->head + sizeof(struct frag_hdr), head->head,
358 (head->data - head->head) - sizeof(struct frag_hdr));
359 if (skb_mac_header_was_set(head))
360 head->mac_header += sizeof(struct frag_hdr);
361 head->network_header += sizeof(struct frag_hdr);
363 skb_reset_transport_header(head);
364 skb_push(head, head->data - skb_network_header(head));
366 sum_truesize = head->truesize;
367 for (fp = head->next; fp;) {
370 struct sk_buff *next = fp->next;
372 sum_truesize += fp->truesize;
373 if (head->ip_summed != fp->ip_summed)
374 head->ip_summed = CHECKSUM_NONE;
375 else if (head->ip_summed == CHECKSUM_COMPLETE)
376 head->csum = csum_add(head->csum, fp->csum);
378 if (skb_try_coalesce(head, fp, &headstolen, &delta)) {
379 kfree_skb_partial(fp, headstolen);
381 if (!skb_shinfo(head)->frag_list)
382 skb_shinfo(head)->frag_list = fp;
383 head->data_len += fp->len;
384 head->len += fp->len;
385 head->truesize += fp->truesize;
389 sub_frag_mem_limit(fq->q.net, sum_truesize);
391 skb_mark_not_on_list(head);
393 head->tstamp = fq->q.stamp;
394 ipv6_hdr(head)->payload_len = htons(payload_len);
395 ipv6_change_dsfield(ipv6_hdr(head), 0xff, ecn);
396 IP6CB(head)->nhoff = nhoff;
397 IP6CB(head)->flags |= IP6SKB_FRAGMENTED;
398 IP6CB(head)->frag_max_size = fq->q.max_size;
400 /* Yes, and fold redundant checksum back. 8) */
401 skb_postpush_rcsum(head, skb_network_header(head),
402 skb_network_header_len(head));
405 __IP6_INC_STATS(net, __in6_dev_get(dev), IPSTATS_MIB_REASMOKS);
407 fq->q.fragments = NULL;
408 fq->q.rb_fragments = RB_ROOT;
409 fq->q.fragments_tail = NULL;
413 net_dbg_ratelimited("ip6_frag_reasm: payload len = %d\n", payload_len);
416 net_dbg_ratelimited("ip6_frag_reasm: no memory for reassembly\n");
419 __IP6_INC_STATS(net, __in6_dev_get(dev), IPSTATS_MIB_REASMFAILS);
421 inet_frag_kill(&fq->q);
425 static int ipv6_frag_rcv(struct sk_buff *skb)
427 struct frag_hdr *fhdr;
428 struct frag_queue *fq;
429 const struct ipv6hdr *hdr = ipv6_hdr(skb);
430 struct net *net = dev_net(skb_dst(skb)->dev);
433 if (IP6CB(skb)->flags & IP6SKB_FRAGMENTED)
436 __IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMREQDS);
438 /* Jumbo payload inhibits frag. header */
439 if (hdr->payload_len == 0)
442 if (!pskb_may_pull(skb, (skb_transport_offset(skb) +
443 sizeof(struct frag_hdr))))
447 fhdr = (struct frag_hdr *)skb_transport_header(skb);
449 if (!(fhdr->frag_off & htons(0xFFF9))) {
450 /* It is not a fragmented frame */
451 skb->transport_header += sizeof(struct frag_hdr);
453 ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMOKS);
455 IP6CB(skb)->nhoff = (u8 *)fhdr - skb_network_header(skb);
456 IP6CB(skb)->flags |= IP6SKB_FRAGMENTED;
460 if (skb->len - skb_network_offset(skb) < IPV6_MIN_MTU &&
461 fhdr->frag_off & htons(IP6_MF))
464 iif = skb->dev ? skb->dev->ifindex : 0;
465 fq = fq_find(net, fhdr->identification, hdr, iif);
470 spin_lock(&fq->q.lock);
473 ret = ip6_frag_queue(fq, skb, fhdr, IP6CB(skb)->nhoff,
476 spin_unlock(&fq->q.lock);
477 inet_frag_put(&fq->q);
479 __IP6_INC_STATS(net, __in6_dev_get_safely(skb->dev),
480 IPSTATS_MIB_INHDRERRORS);
481 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD, prob_offset);
486 __IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMFAILS);
491 __IP6_INC_STATS(net, __in6_dev_get_safely(skb->dev),
492 IPSTATS_MIB_INHDRERRORS);
493 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD, skb_network_header_len(skb));
497 static const struct inet6_protocol frag_protocol = {
498 .handler = ipv6_frag_rcv,
499 .flags = INET6_PROTO_NOPOLICY,
504 static struct ctl_table ip6_frags_ns_ctl_table[] = {
506 .procname = "ip6frag_high_thresh",
507 .data = &init_net.ipv6.frags.high_thresh,
508 .maxlen = sizeof(unsigned long),
510 .proc_handler = proc_doulongvec_minmax,
511 .extra1 = &init_net.ipv6.frags.low_thresh
514 .procname = "ip6frag_low_thresh",
515 .data = &init_net.ipv6.frags.low_thresh,
516 .maxlen = sizeof(unsigned long),
518 .proc_handler = proc_doulongvec_minmax,
519 .extra2 = &init_net.ipv6.frags.high_thresh
522 .procname = "ip6frag_time",
523 .data = &init_net.ipv6.frags.timeout,
524 .maxlen = sizeof(int),
526 .proc_handler = proc_dointvec_jiffies,
531 /* secret interval has been deprecated */
532 static int ip6_frags_secret_interval_unused;
533 static struct ctl_table ip6_frags_ctl_table[] = {
535 .procname = "ip6frag_secret_interval",
536 .data = &ip6_frags_secret_interval_unused,
537 .maxlen = sizeof(int),
539 .proc_handler = proc_dointvec_jiffies,
544 static int __net_init ip6_frags_ns_sysctl_register(struct net *net)
546 struct ctl_table *table;
547 struct ctl_table_header *hdr;
549 table = ip6_frags_ns_ctl_table;
550 if (!net_eq(net, &init_net)) {
551 table = kmemdup(table, sizeof(ip6_frags_ns_ctl_table), GFP_KERNEL);
555 table[0].data = &net->ipv6.frags.high_thresh;
556 table[0].extra1 = &net->ipv6.frags.low_thresh;
557 table[1].data = &net->ipv6.frags.low_thresh;
558 table[1].extra2 = &net->ipv6.frags.high_thresh;
559 table[2].data = &net->ipv6.frags.timeout;
562 hdr = register_net_sysctl(net, "net/ipv6", table);
566 net->ipv6.sysctl.frags_hdr = hdr;
570 if (!net_eq(net, &init_net))
576 static void __net_exit ip6_frags_ns_sysctl_unregister(struct net *net)
578 struct ctl_table *table;
580 table = net->ipv6.sysctl.frags_hdr->ctl_table_arg;
581 unregister_net_sysctl_table(net->ipv6.sysctl.frags_hdr);
582 if (!net_eq(net, &init_net))
586 static struct ctl_table_header *ip6_ctl_header;
588 static int ip6_frags_sysctl_register(void)
590 ip6_ctl_header = register_net_sysctl(&init_net, "net/ipv6",
591 ip6_frags_ctl_table);
592 return ip6_ctl_header == NULL ? -ENOMEM : 0;
595 static void ip6_frags_sysctl_unregister(void)
597 unregister_net_sysctl_table(ip6_ctl_header);
600 static int ip6_frags_ns_sysctl_register(struct net *net)
605 static void ip6_frags_ns_sysctl_unregister(struct net *net)
609 static int ip6_frags_sysctl_register(void)
614 static void ip6_frags_sysctl_unregister(void)
619 static int __net_init ipv6_frags_init_net(struct net *net)
623 net->ipv6.frags.high_thresh = IPV6_FRAG_HIGH_THRESH;
624 net->ipv6.frags.low_thresh = IPV6_FRAG_LOW_THRESH;
625 net->ipv6.frags.timeout = IPV6_FRAG_TIMEOUT;
626 net->ipv6.frags.f = &ip6_frags;
628 res = inet_frags_init_net(&net->ipv6.frags);
632 res = ip6_frags_ns_sysctl_register(net);
634 inet_frags_exit_net(&net->ipv6.frags);
638 static void __net_exit ipv6_frags_exit_net(struct net *net)
640 ip6_frags_ns_sysctl_unregister(net);
641 inet_frags_exit_net(&net->ipv6.frags);
644 static struct pernet_operations ip6_frags_ops = {
645 .init = ipv6_frags_init_net,
646 .exit = ipv6_frags_exit_net,
649 static const struct rhashtable_params ip6_rhash_params = {
650 .head_offset = offsetof(struct inet_frag_queue, node),
651 .hashfn = ip6frag_key_hashfn,
652 .obj_hashfn = ip6frag_obj_hashfn,
653 .obj_cmpfn = ip6frag_obj_cmpfn,
654 .automatic_shrinking = true,
657 int __init ipv6_frag_init(void)
661 ip6_frags.constructor = ip6frag_init;
662 ip6_frags.destructor = NULL;
663 ip6_frags.qsize = sizeof(struct frag_queue);
664 ip6_frags.frag_expire = ip6_frag_expire;
665 ip6_frags.frags_cache_name = ip6_frag_cache_name;
666 ip6_frags.rhash_params = ip6_rhash_params;
667 ret = inet_frags_init(&ip6_frags);
671 ret = inet6_add_protocol(&frag_protocol, IPPROTO_FRAGMENT);
675 ret = ip6_frags_sysctl_register();
679 ret = register_pernet_subsys(&ip6_frags_ops);
687 ip6_frags_sysctl_unregister();
689 inet6_del_protocol(&frag_protocol, IPPROTO_FRAGMENT);
691 inet_frags_fini(&ip6_frags);
695 void ipv6_frag_exit(void)
697 inet_frags_fini(&ip6_frags);
698 ip6_frags_sysctl_unregister();
699 unregister_pernet_subsys(&ip6_frags_ops);
700 inet6_del_protocol(&frag_protocol, IPPROTO_FRAGMENT);