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.
29 #include <linux/errno.h>
30 #include <linux/types.h>
31 #include <linux/string.h>
32 #include <linux/socket.h>
33 #include <linux/sockios.h>
34 #include <linux/jiffies.h>
35 #include <linux/net.h>
36 #include <linux/list.h>
37 #include <linux/netdevice.h>
38 #include <linux/in6.h>
39 #include <linux/ipv6.h>
40 #include <linux/icmpv6.h>
41 #include <linux/random.h>
42 #include <linux/jhash.h>
43 #include <linux/skbuff.h>
44 #include <linux/slab.h>
45 #include <linux/export.h>
51 #include <net/ip6_route.h>
52 #include <net/protocol.h>
53 #include <net/transp_v6.h>
54 #include <net/rawv6.h>
55 #include <net/ndisc.h>
56 #include <net/addrconf.h>
57 #include <net/inet_frag.h>
61 struct inet6_skb_parm h;
65 #define FRAG6_CB(skb) ((struct ip6frag_skb_cb*)((skb)->cb))
68 static struct inet_frags ip6_frags;
70 static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *prev,
71 struct net_device *dev);
74 * callers should be careful not to use the hash value outside the ipfrag_lock
75 * as doing so could race with ipfrag_hash_rnd being recalculated.
77 unsigned int inet6_hash_frag(__be32 id, const struct in6_addr *saddr,
78 const struct in6_addr *daddr, u32 rnd)
82 c = jhash_3words(ipv6_addr_hash(saddr), ipv6_addr_hash(daddr), id, rnd);
84 return c & (INETFRAGS_HASHSZ - 1);
86 EXPORT_SYMBOL_GPL(inet6_hash_frag);
88 static unsigned int ip6_hashfn(struct inet_frag_queue *q)
90 struct frag_queue *fq;
92 fq = container_of(q, struct frag_queue, q);
93 return inet6_hash_frag(fq->id, &fq->saddr, &fq->daddr, ip6_frags.rnd);
96 bool ip6_frag_match(struct inet_frag_queue *q, void *a)
98 struct frag_queue *fq;
99 struct ip6_create_arg *arg = a;
101 fq = container_of(q, struct frag_queue, q);
102 return fq->id == arg->id &&
103 fq->user == arg->user &&
104 ipv6_addr_equal(&fq->saddr, arg->src) &&
105 ipv6_addr_equal(&fq->daddr, arg->dst);
107 EXPORT_SYMBOL(ip6_frag_match);
109 void ip6_frag_init(struct inet_frag_queue *q, void *a)
111 struct frag_queue *fq = container_of(q, struct frag_queue, q);
112 struct ip6_create_arg *arg = a;
115 fq->user = arg->user;
116 fq->saddr = *arg->src;
117 fq->daddr = *arg->dst;
119 EXPORT_SYMBOL(ip6_frag_init);
121 void ip6_expire_frag_queue(struct net *net, struct frag_queue *fq,
122 struct inet_frags *frags)
124 struct net_device *dev = NULL;
126 spin_lock(&fq->q.lock);
128 if (fq->q.last_in & INET_FRAG_COMPLETE)
131 inet_frag_kill(&fq->q, frags);
134 dev = dev_get_by_index_rcu(net, fq->iif);
138 IP6_INC_STATS_BH(net, __in6_dev_get(dev), IPSTATS_MIB_REASMTIMEOUT);
139 IP6_INC_STATS_BH(net, __in6_dev_get(dev), IPSTATS_MIB_REASMFAILS);
141 /* Don't send error if the first segment did not arrive. */
142 if (!(fq->q.last_in & INET_FRAG_FIRST_IN) || !fq->q.fragments)
146 But use as source device on which LAST ARRIVED
147 segment was received. And do not use fq->dev
148 pointer directly, device might already disappeared.
150 fq->q.fragments->dev = dev;
151 icmpv6_send(fq->q.fragments, ICMPV6_TIME_EXCEED, ICMPV6_EXC_FRAGTIME, 0);
155 spin_unlock(&fq->q.lock);
156 inet_frag_put(&fq->q, frags);
158 EXPORT_SYMBOL(ip6_expire_frag_queue);
160 static void ip6_frag_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, ipv6.frags);
168 ip6_expire_frag_queue(net, fq, &ip6_frags);
171 static __inline__ struct frag_queue *
172 fq_find(struct net *net, __be32 id, const struct in6_addr *src, const struct in6_addr *dst)
174 struct inet_frag_queue *q;
175 struct ip6_create_arg arg;
179 arg.user = IP6_DEFRAG_LOCAL_DELIVER;
183 read_lock(&ip6_frags.lock);
184 hash = inet6_hash_frag(id, src, dst, ip6_frags.rnd);
186 q = inet_frag_find(&net->ipv6.frags, &ip6_frags, &arg, hash);
190 return container_of(q, struct frag_queue, q);
193 static int ip6_frag_queue(struct frag_queue *fq, struct sk_buff *skb,
194 struct frag_hdr *fhdr, int nhoff)
196 struct sk_buff *prev, *next;
197 struct net_device *dev;
199 struct net *net = dev_net(skb_dst(skb)->dev);
201 if (fq->q.last_in & INET_FRAG_COMPLETE)
204 offset = ntohs(fhdr->frag_off) & ~0x7;
205 end = offset + (ntohs(ipv6_hdr(skb)->payload_len) -
206 ((u8 *)(fhdr + 1) - (u8 *)(ipv6_hdr(skb) + 1)));
208 if ((unsigned int)end > IPV6_MAXPLEN) {
209 IP6_INC_STATS_BH(net, ip6_dst_idev(skb_dst(skb)),
210 IPSTATS_MIB_INHDRERRORS);
211 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
212 ((u8 *)&fhdr->frag_off -
213 skb_network_header(skb)));
217 if (skb->ip_summed == CHECKSUM_COMPLETE) {
218 const unsigned char *nh = skb_network_header(skb);
219 skb->csum = csum_sub(skb->csum,
220 csum_partial(nh, (u8 *)(fhdr + 1) - nh,
224 /* Is this the final fragment? */
225 if (!(fhdr->frag_off & htons(IP6_MF))) {
226 /* If we already have some bits beyond end
227 * or have different end, the segment is corrupted.
229 if (end < fq->q.len ||
230 ((fq->q.last_in & INET_FRAG_LAST_IN) && end != fq->q.len))
232 fq->q.last_in |= INET_FRAG_LAST_IN;
235 /* Check if the fragment is rounded to 8 bytes.
236 * Required by the RFC.
239 /* RFC2460 says always send parameter problem in
242 IP6_INC_STATS_BH(net, ip6_dst_idev(skb_dst(skb)),
243 IPSTATS_MIB_INHDRERRORS);
244 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
245 offsetof(struct ipv6hdr, payload_len));
248 if (end > fq->q.len) {
249 /* Some bits beyond end -> corruption. */
250 if (fq->q.last_in & INET_FRAG_LAST_IN)
259 /* Point into the IP datagram 'data' part. */
260 if (!pskb_pull(skb, (u8 *) (fhdr + 1) - skb->data))
263 if (pskb_trim_rcsum(skb, end - offset))
266 /* Find out which fragments are in front and at the back of us
267 * in the chain of fragments so far. We must know where to put
268 * this fragment, right?
270 prev = fq->q.fragments_tail;
271 if (!prev || FRAG6_CB(prev)->offset < offset) {
276 for(next = fq->q.fragments; next != NULL; next = next->next) {
277 if (FRAG6_CB(next)->offset >= offset)
283 /* RFC5722, Section 4, amended by Errata ID : 3089
284 * When reassembling an IPv6 datagram, if
285 * one or more its constituent fragments is determined to be an
286 * overlapping fragment, the entire datagram (and any constituent
287 * fragments) MUST be silently discarded.
290 /* Check for overlap with preceding fragment. */
292 (FRAG6_CB(prev)->offset + prev->len) > offset)
295 /* Look for overlap with succeeding segment. */
296 if (next && FRAG6_CB(next)->offset < end)
299 FRAG6_CB(skb)->offset = offset;
301 /* Insert this fragment in the chain of fragments. */
304 fq->q.fragments_tail = skb;
308 fq->q.fragments = skb;
312 fq->iif = dev->ifindex;
315 fq->q.stamp = skb->tstamp;
316 fq->q.meat += skb->len;
317 add_frag_mem_limit(&fq->q, skb->truesize);
319 /* The first fragment.
320 * nhoffset is obtained from the first fragment, of course.
323 fq->nhoffset = nhoff;
324 fq->q.last_in |= INET_FRAG_FIRST_IN;
327 if (fq->q.last_in == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
328 fq->q.meat == fq->q.len)
329 return ip6_frag_reasm(fq, prev, dev);
331 inet_frag_lru_move(&fq->q);
335 inet_frag_kill(&fq->q, &ip6_frags);
337 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
338 IPSTATS_MIB_REASMFAILS);
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 int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *prev,
353 struct net_device *dev)
355 struct net *net = container_of(fq->q.net, struct net, ipv6.frags);
356 struct sk_buff *fp, *head = fq->q.fragments;
361 inet_frag_kill(&fq->q, &ip6_frags);
363 /* Make the one we just received the head. */
366 fp = skb_clone(head, GFP_ATOMIC);
371 fp->next = head->next;
373 fq->q.fragments_tail = fp;
376 skb_morph(head, fq->q.fragments);
377 head->next = fq->q.fragments->next;
379 consume_skb(fq->q.fragments);
380 fq->q.fragments = head;
383 WARN_ON(head == NULL);
384 WARN_ON(FRAG6_CB(head)->offset != 0);
386 /* Unfragmented part is taken from the first segment. */
387 payload_len = ((head->data - skb_network_header(head)) -
388 sizeof(struct ipv6hdr) + fq->q.len -
389 sizeof(struct frag_hdr));
390 if (payload_len > IPV6_MAXPLEN)
393 /* Head of list must not be cloned. */
394 if (skb_unclone(head, GFP_ATOMIC))
397 /* If the first fragment is fragmented itself, we split
398 * it to two chunks: the first with data and paged part
399 * and the second, holding only fragments. */
400 if (skb_has_frag_list(head)) {
401 struct sk_buff *clone;
404 if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL)
406 clone->next = head->next;
408 skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
409 skb_frag_list_init(head);
410 for (i = 0; i < skb_shinfo(head)->nr_frags; i++)
411 plen += skb_frag_size(&skb_shinfo(head)->frags[i]);
412 clone->len = clone->data_len = head->data_len - plen;
413 head->data_len -= clone->len;
414 head->len -= clone->len;
416 clone->ip_summed = head->ip_summed;
417 add_frag_mem_limit(&fq->q, clone->truesize);
420 /* We have to remove fragment header from datagram and to relocate
421 * header in order to calculate ICV correctly. */
422 nhoff = fq->nhoffset;
423 skb_network_header(head)[nhoff] = skb_transport_header(head)[0];
424 memmove(head->head + sizeof(struct frag_hdr), head->head,
425 (head->data - head->head) - sizeof(struct frag_hdr));
426 head->mac_header += sizeof(struct frag_hdr);
427 head->network_header += sizeof(struct frag_hdr);
429 skb_reset_transport_header(head);
430 skb_push(head, head->data - skb_network_header(head));
432 sum_truesize = head->truesize;
433 for (fp = head->next; fp;) {
436 struct sk_buff *next = fp->next;
438 sum_truesize += fp->truesize;
439 if (head->ip_summed != fp->ip_summed)
440 head->ip_summed = CHECKSUM_NONE;
441 else if (head->ip_summed == CHECKSUM_COMPLETE)
442 head->csum = csum_add(head->csum, fp->csum);
444 if (skb_try_coalesce(head, fp, &headstolen, &delta)) {
445 kfree_skb_partial(fp, headstolen);
447 if (!skb_shinfo(head)->frag_list)
448 skb_shinfo(head)->frag_list = fp;
449 head->data_len += fp->len;
450 head->len += fp->len;
451 head->truesize += fp->truesize;
455 sub_frag_mem_limit(&fq->q, sum_truesize);
459 head->tstamp = fq->q.stamp;
460 ipv6_hdr(head)->payload_len = htons(payload_len);
461 IP6CB(head)->nhoff = nhoff;
463 /* Yes, and fold redundant checksum back. 8) */
464 if (head->ip_summed == CHECKSUM_COMPLETE)
465 head->csum = csum_partial(skb_network_header(head),
466 skb_network_header_len(head),
470 IP6_INC_STATS_BH(net, __in6_dev_get(dev), IPSTATS_MIB_REASMOKS);
472 fq->q.fragments = NULL;
473 fq->q.fragments_tail = NULL;
477 net_dbg_ratelimited("ip6_frag_reasm: payload len = %d\n", payload_len);
480 net_dbg_ratelimited("ip6_frag_reasm: no memory for reassembly\n");
483 IP6_INC_STATS_BH(net, __in6_dev_get(dev), IPSTATS_MIB_REASMFAILS);
488 static int ipv6_frag_rcv(struct sk_buff *skb)
490 struct frag_hdr *fhdr;
491 struct frag_queue *fq;
492 const struct ipv6hdr *hdr = ipv6_hdr(skb);
493 struct net *net = dev_net(skb_dst(skb)->dev);
496 IP6_INC_STATS_BH(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMREQDS);
498 /* Jumbo payload inhibits frag. header */
499 if (hdr->payload_len==0)
502 if (!pskb_may_pull(skb, (skb_transport_offset(skb) +
503 sizeof(struct frag_hdr))))
507 fhdr = (struct frag_hdr *)skb_transport_header(skb);
509 if (!(fhdr->frag_off & htons(0xFFF9))) {
510 /* It is not a fragmented frame */
511 skb->transport_header += sizeof(struct frag_hdr);
512 IP6_INC_STATS_BH(net,
513 ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMOKS);
515 IP6CB(skb)->nhoff = (u8 *)fhdr - skb_network_header(skb);
519 evicted = inet_frag_evictor(&net->ipv6.frags, &ip6_frags, false);
521 IP6_ADD_STATS_BH(net, ip6_dst_idev(skb_dst(skb)),
522 IPSTATS_MIB_REASMFAILS, evicted);
524 fq = fq_find(net, fhdr->identification, &hdr->saddr, &hdr->daddr);
528 spin_lock(&fq->q.lock);
530 ret = ip6_frag_queue(fq, skb, fhdr, IP6CB(skb)->nhoff);
532 spin_unlock(&fq->q.lock);
533 inet_frag_put(&fq->q, &ip6_frags);
537 IP6_INC_STATS_BH(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMFAILS);
542 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_INHDRERRORS);
543 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD, skb_network_header_len(skb));
547 static const struct inet6_protocol frag_protocol =
549 .handler = ipv6_frag_rcv,
550 .flags = INET6_PROTO_NOPOLICY,
554 static struct ctl_table ip6_frags_ns_ctl_table[] = {
556 .procname = "ip6frag_high_thresh",
557 .data = &init_net.ipv6.frags.high_thresh,
558 .maxlen = sizeof(int),
560 .proc_handler = proc_dointvec
563 .procname = "ip6frag_low_thresh",
564 .data = &init_net.ipv6.frags.low_thresh,
565 .maxlen = sizeof(int),
567 .proc_handler = proc_dointvec
570 .procname = "ip6frag_time",
571 .data = &init_net.ipv6.frags.timeout,
572 .maxlen = sizeof(int),
574 .proc_handler = proc_dointvec_jiffies,
579 static struct ctl_table ip6_frags_ctl_table[] = {
581 .procname = "ip6frag_secret_interval",
582 .data = &ip6_frags.secret_interval,
583 .maxlen = sizeof(int),
585 .proc_handler = proc_dointvec_jiffies,
590 static int __net_init ip6_frags_ns_sysctl_register(struct net *net)
592 struct ctl_table *table;
593 struct ctl_table_header *hdr;
595 table = ip6_frags_ns_ctl_table;
596 if (!net_eq(net, &init_net)) {
597 table = kmemdup(table, sizeof(ip6_frags_ns_ctl_table), GFP_KERNEL);
601 table[0].data = &net->ipv6.frags.high_thresh;
602 table[1].data = &net->ipv6.frags.low_thresh;
603 table[2].data = &net->ipv6.frags.timeout;
605 /* Don't export sysctls to unprivileged users */
606 if (net->user_ns != &init_user_ns)
607 table[0].procname = NULL;
610 hdr = register_net_sysctl(net, "net/ipv6", table);
614 net->ipv6.sysctl.frags_hdr = hdr;
618 if (!net_eq(net, &init_net))
624 static void __net_exit ip6_frags_ns_sysctl_unregister(struct net *net)
626 struct ctl_table *table;
628 table = net->ipv6.sysctl.frags_hdr->ctl_table_arg;
629 unregister_net_sysctl_table(net->ipv6.sysctl.frags_hdr);
630 if (!net_eq(net, &init_net))
634 static struct ctl_table_header *ip6_ctl_header;
636 static int ip6_frags_sysctl_register(void)
638 ip6_ctl_header = register_net_sysctl(&init_net, "net/ipv6",
639 ip6_frags_ctl_table);
640 return ip6_ctl_header == NULL ? -ENOMEM : 0;
643 static void ip6_frags_sysctl_unregister(void)
645 unregister_net_sysctl_table(ip6_ctl_header);
648 static inline int ip6_frags_ns_sysctl_register(struct net *net)
653 static inline void ip6_frags_ns_sysctl_unregister(struct net *net)
657 static inline int ip6_frags_sysctl_register(void)
662 static inline void ip6_frags_sysctl_unregister(void)
667 static int __net_init ipv6_frags_init_net(struct net *net)
669 net->ipv6.frags.high_thresh = IPV6_FRAG_HIGH_THRESH;
670 net->ipv6.frags.low_thresh = IPV6_FRAG_LOW_THRESH;
671 net->ipv6.frags.timeout = IPV6_FRAG_TIMEOUT;
673 inet_frags_init_net(&net->ipv6.frags);
675 return ip6_frags_ns_sysctl_register(net);
678 static void __net_exit ipv6_frags_exit_net(struct net *net)
680 ip6_frags_ns_sysctl_unregister(net);
681 inet_frags_exit_net(&net->ipv6.frags, &ip6_frags);
684 static struct pernet_operations ip6_frags_ops = {
685 .init = ipv6_frags_init_net,
686 .exit = ipv6_frags_exit_net,
689 int __init ipv6_frag_init(void)
693 ret = inet6_add_protocol(&frag_protocol, IPPROTO_FRAGMENT);
697 ret = ip6_frags_sysctl_register();
701 ret = register_pernet_subsys(&ip6_frags_ops);
705 ip6_frags.hashfn = ip6_hashfn;
706 ip6_frags.constructor = ip6_frag_init;
707 ip6_frags.destructor = NULL;
708 ip6_frags.skb_free = NULL;
709 ip6_frags.qsize = sizeof(struct frag_queue);
710 ip6_frags.match = ip6_frag_match;
711 ip6_frags.frag_expire = ip6_frag_expire;
712 ip6_frags.secret_interval = 10 * 60 * HZ;
713 inet_frags_init(&ip6_frags);
718 ip6_frags_sysctl_unregister();
720 inet6_del_protocol(&frag_protocol, IPPROTO_FRAGMENT);
724 void ipv6_frag_exit(void)
726 inet_frags_fini(&ip6_frags);
727 ip6_frags_sysctl_unregister();
728 unregister_pernet_subsys(&ip6_frags_ops);
729 inet6_del_protocol(&frag_protocol, IPPROTO_FRAGMENT);