2 * IPv6 fragment reassembly
3 * Linux INET6 implementation
6 * Pedro Roque <roque@di.fc.ul.pt>
8 * $Id: reassembly.c,v 1.26 2001/03/07 22:00:57 davem Exp $
10 * Based on: net/ipv4/ip_fragment.c
12 * This program is free software; you can redistribute it and/or
13 * modify it under the terms of the GNU General Public License
14 * as published by the Free Software Foundation; either version
15 * 2 of the License, or (at your option) any later version.
20 * Andi Kleen Make it work with multiple hosts.
21 * More RFC compliance.
23 * Horst von Brand Add missing #include <linux/string.h>
24 * Alexey Kuznetsov SMP races, threading, cleanup.
25 * Patrick McHardy LRU queue of frag heads for evictor.
26 * Mitsuru KANDA @USAGI Register inet6_protocol{}.
28 * YOSHIFUJI,H. @USAGI Always remove fragment header to
29 * calculate ICV correctly.
31 #include <linux/errno.h>
32 #include <linux/types.h>
33 #include <linux/string.h>
34 #include <linux/socket.h>
35 #include <linux/sockios.h>
36 #include <linux/jiffies.h>
37 #include <linux/net.h>
38 #include <linux/list.h>
39 #include <linux/netdevice.h>
40 #include <linux/in6.h>
41 #include <linux/ipv6.h>
42 #include <linux/icmpv6.h>
43 #include <linux/random.h>
44 #include <linux/jhash.h>
45 #include <linux/skbuff.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))
69 * Equivalent of ipv4 struct ipq
74 struct inet_frag_queue q;
76 __be32 id; /* fragment id */
77 struct in6_addr saddr;
78 struct in6_addr daddr;
85 static struct inet_frags ip6_frags;
87 int ip6_frag_nqueues(struct net *net)
89 return net->ipv6.frags.nqueues;
92 int ip6_frag_mem(void)
94 return atomic_read(&ip6_frags.mem);
97 static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *prev,
98 struct net_device *dev);
101 * callers should be careful not to use the hash value outside the ipfrag_lock
102 * as doing so could race with ipfrag_hash_rnd being recalculated.
104 static unsigned int ip6qhashfn(__be32 id, struct in6_addr *saddr,
105 struct in6_addr *daddr)
109 a = (__force u32)saddr->s6_addr32[0];
110 b = (__force u32)saddr->s6_addr32[1];
111 c = (__force u32)saddr->s6_addr32[2];
113 a += JHASH_GOLDEN_RATIO;
114 b += JHASH_GOLDEN_RATIO;
116 __jhash_mix(a, b, c);
118 a += (__force u32)saddr->s6_addr32[3];
119 b += (__force u32)daddr->s6_addr32[0];
120 c += (__force u32)daddr->s6_addr32[1];
121 __jhash_mix(a, b, c);
123 a += (__force u32)daddr->s6_addr32[2];
124 b += (__force u32)daddr->s6_addr32[3];
125 c += (__force u32)id;
126 __jhash_mix(a, b, c);
128 return c & (INETFRAGS_HASHSZ - 1);
131 static unsigned int ip6_hashfn(struct inet_frag_queue *q)
133 struct frag_queue *fq;
135 fq = container_of(q, struct frag_queue, q);
136 return ip6qhashfn(fq->id, &fq->saddr, &fq->daddr);
139 int ip6_frag_match(struct inet_frag_queue *q, void *a)
141 struct frag_queue *fq;
142 struct ip6_create_arg *arg = a;
144 fq = container_of(q, struct frag_queue, q);
145 return (fq->id == arg->id &&
146 ipv6_addr_equal(&fq->saddr, arg->src) &&
147 ipv6_addr_equal(&fq->daddr, arg->dst));
149 EXPORT_SYMBOL(ip6_frag_match);
151 /* Memory Tracking Functions. */
152 static inline void frag_kfree_skb(struct sk_buff *skb, int *work)
155 *work -= skb->truesize;
156 atomic_sub(skb->truesize, &ip6_frags.mem);
160 void ip6_frag_init(struct inet_frag_queue *q, void *a)
162 struct frag_queue *fq = container_of(q, struct frag_queue, q);
163 struct ip6_create_arg *arg = a;
166 ipv6_addr_copy(&fq->saddr, arg->src);
167 ipv6_addr_copy(&fq->daddr, arg->dst);
169 EXPORT_SYMBOL(ip6_frag_init);
171 /* Destruction primitives. */
173 static __inline__ void fq_put(struct frag_queue *fq)
175 inet_frag_put(&fq->q, &ip6_frags);
178 /* Kill fq entry. It is not destroyed immediately,
179 * because caller (and someone more) holds reference count.
181 static __inline__ void fq_kill(struct frag_queue *fq)
183 inet_frag_kill(&fq->q, &ip6_frags);
186 static void ip6_evictor(struct inet6_dev *idev)
190 evicted = inet_frag_evictor(&ip6_frags);
192 IP6_ADD_STATS_BH(idev, IPSTATS_MIB_REASMFAILS, evicted);
195 static void ip6_frag_expire(unsigned long data)
197 struct frag_queue *fq;
198 struct net_device *dev = NULL;
200 fq = container_of((struct inet_frag_queue *)data, struct frag_queue, q);
202 spin_lock(&fq->q.lock);
204 if (fq->q.last_in & COMPLETE)
209 dev = dev_get_by_index(&init_net, fq->iif);
214 IP6_INC_STATS_BH(__in6_dev_get(dev), IPSTATS_MIB_REASMTIMEOUT);
215 IP6_INC_STATS_BH(__in6_dev_get(dev), IPSTATS_MIB_REASMFAILS);
218 /* Don't send error if the first segment did not arrive. */
219 if (!(fq->q.last_in&FIRST_IN) || !fq->q.fragments)
223 But use as source device on which LAST ARRIVED
224 segment was received. And do not use fq->dev
225 pointer directly, device might already disappeared.
227 fq->q.fragments->dev = dev;
228 icmpv6_send(fq->q.fragments, ICMPV6_TIME_EXCEED, ICMPV6_EXC_FRAGTIME, 0, dev);
232 spin_unlock(&fq->q.lock);
236 static __inline__ struct frag_queue *
237 fq_find(struct net *net, __be32 id, struct in6_addr *src, struct in6_addr *dst,
238 struct inet6_dev *idev)
240 struct inet_frag_queue *q;
241 struct ip6_create_arg arg;
247 hash = ip6qhashfn(id, src, dst);
249 q = inet_frag_find(&net->ipv6.frags, &ip6_frags, &arg, hash);
253 return container_of(q, struct frag_queue, q);
256 IP6_INC_STATS_BH(idev, IPSTATS_MIB_REASMFAILS);
260 static int ip6_frag_queue(struct frag_queue *fq, struct sk_buff *skb,
261 struct frag_hdr *fhdr, int nhoff)
263 struct sk_buff *prev, *next;
264 struct net_device *dev;
267 if (fq->q.last_in & COMPLETE)
270 offset = ntohs(fhdr->frag_off) & ~0x7;
271 end = offset + (ntohs(ipv6_hdr(skb)->payload_len) -
272 ((u8 *)(fhdr + 1) - (u8 *)(ipv6_hdr(skb) + 1)));
274 if ((unsigned int)end > IPV6_MAXPLEN) {
275 IP6_INC_STATS_BH(ip6_dst_idev(skb->dst),
276 IPSTATS_MIB_INHDRERRORS);
277 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
278 ((u8 *)&fhdr->frag_off -
279 skb_network_header(skb)));
283 if (skb->ip_summed == CHECKSUM_COMPLETE) {
284 const unsigned char *nh = skb_network_header(skb);
285 skb->csum = csum_sub(skb->csum,
286 csum_partial(nh, (u8 *)(fhdr + 1) - nh,
290 /* Is this the final fragment? */
291 if (!(fhdr->frag_off & htons(IP6_MF))) {
292 /* If we already have some bits beyond end
293 * or have different end, the segment is corrupted.
295 if (end < fq->q.len ||
296 ((fq->q.last_in & LAST_IN) && end != fq->q.len))
298 fq->q.last_in |= LAST_IN;
301 /* Check if the fragment is rounded to 8 bytes.
302 * Required by the RFC.
305 /* RFC2460 says always send parameter problem in
308 IP6_INC_STATS_BH(ip6_dst_idev(skb->dst),
309 IPSTATS_MIB_INHDRERRORS);
310 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
311 offsetof(struct ipv6hdr, payload_len));
314 if (end > fq->q.len) {
315 /* Some bits beyond end -> corruption. */
316 if (fq->q.last_in & LAST_IN)
325 /* Point into the IP datagram 'data' part. */
326 if (!pskb_pull(skb, (u8 *) (fhdr + 1) - skb->data))
329 if (pskb_trim_rcsum(skb, end - offset))
332 /* Find out which fragments are in front and at the back of us
333 * in the chain of fragments so far. We must know where to put
334 * this fragment, right?
337 for(next = fq->q.fragments; next != NULL; next = next->next) {
338 if (FRAG6_CB(next)->offset >= offset)
343 /* We found where to put this one. Check for overlap with
344 * preceding fragment, and, if needed, align things so that
345 * any overlaps are eliminated.
348 int i = (FRAG6_CB(prev)->offset + prev->len) - offset;
354 if (!pskb_pull(skb, i))
356 if (skb->ip_summed != CHECKSUM_UNNECESSARY)
357 skb->ip_summed = CHECKSUM_NONE;
361 /* Look for overlap with succeeding segments.
362 * If we can merge fragments, do it.
364 while (next && FRAG6_CB(next)->offset < end) {
365 int i = end - FRAG6_CB(next)->offset; /* overlap is 'i' bytes */
368 /* Eat head of the next overlapped fragment
369 * and leave the loop. The next ones cannot overlap.
371 if (!pskb_pull(next, i))
373 FRAG6_CB(next)->offset += i; /* next fragment */
375 if (next->ip_summed != CHECKSUM_UNNECESSARY)
376 next->ip_summed = CHECKSUM_NONE;
379 struct sk_buff *free_it = next;
381 /* Old fragment is completely overridden with
389 fq->q.fragments = next;
391 fq->q.meat -= free_it->len;
392 frag_kfree_skb(free_it, NULL);
396 FRAG6_CB(skb)->offset = offset;
398 /* Insert this fragment in the chain of fragments. */
403 fq->q.fragments = skb;
407 fq->iif = dev->ifindex;
410 fq->q.stamp = skb->tstamp;
411 fq->q.meat += skb->len;
412 atomic_add(skb->truesize, &ip6_frags.mem);
414 /* The first fragment.
415 * nhoffset is obtained from the first fragment, of course.
418 fq->nhoffset = nhoff;
419 fq->q.last_in |= FIRST_IN;
422 if (fq->q.last_in == (FIRST_IN | LAST_IN) && fq->q.meat == fq->q.len)
423 return ip6_frag_reasm(fq, prev, dev);
425 write_lock(&ip6_frags.lock);
426 list_move_tail(&fq->q.lru_list, &ip6_frags.lru_list);
427 write_unlock(&ip6_frags.lock);
431 IP6_INC_STATS(ip6_dst_idev(skb->dst), IPSTATS_MIB_REASMFAILS);
437 * Check if this packet is complete.
438 * Returns NULL on failure by any reason, and pointer
439 * to current nexthdr field in reassembled frame.
441 * It is called with locked fq, and caller must check that
442 * queue is eligible for reassembly i.e. it is not COMPLETE,
443 * the last and the first frames arrived and all the bits are here.
445 static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *prev,
446 struct net_device *dev)
448 struct sk_buff *fp, *head = fq->q.fragments;
454 /* Make the one we just received the head. */
457 fp = skb_clone(head, GFP_ATOMIC);
462 fp->next = head->next;
465 skb_morph(head, fq->q.fragments);
466 head->next = fq->q.fragments->next;
468 kfree_skb(fq->q.fragments);
469 fq->q.fragments = head;
472 BUG_TRAP(head != NULL);
473 BUG_TRAP(FRAG6_CB(head)->offset == 0);
475 /* Unfragmented part is taken from the first segment. */
476 payload_len = ((head->data - skb_network_header(head)) -
477 sizeof(struct ipv6hdr) + fq->q.len -
478 sizeof(struct frag_hdr));
479 if (payload_len > IPV6_MAXPLEN)
482 /* Head of list must not be cloned. */
483 if (skb_cloned(head) && pskb_expand_head(head, 0, 0, GFP_ATOMIC))
486 /* If the first fragment is fragmented itself, we split
487 * it to two chunks: the first with data and paged part
488 * and the second, holding only fragments. */
489 if (skb_shinfo(head)->frag_list) {
490 struct sk_buff *clone;
493 if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL)
495 clone->next = head->next;
497 skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
498 skb_shinfo(head)->frag_list = NULL;
499 for (i=0; i<skb_shinfo(head)->nr_frags; i++)
500 plen += skb_shinfo(head)->frags[i].size;
501 clone->len = clone->data_len = head->data_len - plen;
502 head->data_len -= clone->len;
503 head->len -= clone->len;
505 clone->ip_summed = head->ip_summed;
506 atomic_add(clone->truesize, &ip6_frags.mem);
509 /* We have to remove fragment header from datagram and to relocate
510 * header in order to calculate ICV correctly. */
511 nhoff = fq->nhoffset;
512 skb_network_header(head)[nhoff] = skb_transport_header(head)[0];
513 memmove(head->head + sizeof(struct frag_hdr), head->head,
514 (head->data - head->head) - sizeof(struct frag_hdr));
515 head->mac_header += sizeof(struct frag_hdr);
516 head->network_header += sizeof(struct frag_hdr);
518 skb_shinfo(head)->frag_list = head->next;
519 skb_reset_transport_header(head);
520 skb_push(head, head->data - skb_network_header(head));
521 atomic_sub(head->truesize, &ip6_frags.mem);
523 for (fp=head->next; fp; fp = fp->next) {
524 head->data_len += fp->len;
525 head->len += fp->len;
526 if (head->ip_summed != fp->ip_summed)
527 head->ip_summed = CHECKSUM_NONE;
528 else if (head->ip_summed == CHECKSUM_COMPLETE)
529 head->csum = csum_add(head->csum, fp->csum);
530 head->truesize += fp->truesize;
531 atomic_sub(fp->truesize, &ip6_frags.mem);
536 head->tstamp = fq->q.stamp;
537 ipv6_hdr(head)->payload_len = htons(payload_len);
538 IP6CB(head)->nhoff = nhoff;
540 /* Yes, and fold redundant checksum back. 8) */
541 if (head->ip_summed == CHECKSUM_COMPLETE)
542 head->csum = csum_partial(skb_network_header(head),
543 skb_network_header_len(head),
547 IP6_INC_STATS_BH(__in6_dev_get(dev), IPSTATS_MIB_REASMOKS);
549 fq->q.fragments = NULL;
554 printk(KERN_DEBUG "ip6_frag_reasm: payload len = %d\n", payload_len);
558 printk(KERN_DEBUG "ip6_frag_reasm: no memory for reassembly\n");
561 IP6_INC_STATS_BH(__in6_dev_get(dev), IPSTATS_MIB_REASMFAILS);
566 static int ipv6_frag_rcv(struct sk_buff *skb)
568 struct frag_hdr *fhdr;
569 struct frag_queue *fq;
570 struct ipv6hdr *hdr = ipv6_hdr(skb);
573 IP6_INC_STATS_BH(ip6_dst_idev(skb->dst), IPSTATS_MIB_REASMREQDS);
575 /* Jumbo payload inhibits frag. header */
576 if (hdr->payload_len==0) {
577 IP6_INC_STATS(ip6_dst_idev(skb->dst), IPSTATS_MIB_INHDRERRORS);
578 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
579 skb_network_header_len(skb));
582 if (!pskb_may_pull(skb, (skb_transport_offset(skb) +
583 sizeof(struct frag_hdr)))) {
584 IP6_INC_STATS(ip6_dst_idev(skb->dst), IPSTATS_MIB_INHDRERRORS);
585 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
586 skb_network_header_len(skb));
591 fhdr = (struct frag_hdr *)skb_transport_header(skb);
593 if (!(fhdr->frag_off & htons(0xFFF9))) {
594 /* It is not a fragmented frame */
595 skb->transport_header += sizeof(struct frag_hdr);
596 IP6_INC_STATS_BH(ip6_dst_idev(skb->dst), IPSTATS_MIB_REASMOKS);
598 IP6CB(skb)->nhoff = (u8 *)fhdr - skb_network_header(skb);
602 net = skb->dev->nd_net;
603 if (atomic_read(&ip6_frags.mem) > init_net.ipv6.sysctl.frags.high_thresh)
604 ip6_evictor(ip6_dst_idev(skb->dst));
606 if ((fq = fq_find(net, fhdr->identification, &hdr->saddr, &hdr->daddr,
607 ip6_dst_idev(skb->dst))) != NULL) {
610 spin_lock(&fq->q.lock);
612 ret = ip6_frag_queue(fq, skb, fhdr, IP6CB(skb)->nhoff);
614 spin_unlock(&fq->q.lock);
619 IP6_INC_STATS_BH(ip6_dst_idev(skb->dst), IPSTATS_MIB_REASMFAILS);
624 static struct inet6_protocol frag_protocol =
626 .handler = ipv6_frag_rcv,
627 .flags = INET6_PROTO_NOPOLICY,
631 static struct ctl_table ip6_frags_ctl_table[] = {
633 .ctl_name = NET_IPV6_IP6FRAG_HIGH_THRESH,
634 .procname = "ip6frag_high_thresh",
635 .data = &init_net.ipv6.sysctl.frags.high_thresh,
636 .maxlen = sizeof(int),
638 .proc_handler = &proc_dointvec
641 .ctl_name = NET_IPV6_IP6FRAG_LOW_THRESH,
642 .procname = "ip6frag_low_thresh",
643 .data = &init_net.ipv6.sysctl.frags.low_thresh,
644 .maxlen = sizeof(int),
646 .proc_handler = &proc_dointvec
649 .ctl_name = NET_IPV6_IP6FRAG_TIME,
650 .procname = "ip6frag_time",
651 .data = &init_net.ipv6.sysctl.frags.timeout,
652 .maxlen = sizeof(int),
654 .proc_handler = &proc_dointvec_jiffies,
655 .strategy = &sysctl_jiffies,
658 .ctl_name = NET_IPV6_IP6FRAG_SECRET_INTERVAL,
659 .procname = "ip6frag_secret_interval",
660 .data = &init_net.ipv6.sysctl.frags.secret_interval,
661 .maxlen = sizeof(int),
663 .proc_handler = &proc_dointvec_jiffies,
664 .strategy = &sysctl_jiffies
669 static int ip6_frags_sysctl_register(struct net *net)
671 struct ctl_table_header *hdr;
673 hdr = register_net_sysctl_table(net, net_ipv6_ctl_path,
674 ip6_frags_ctl_table);
675 return hdr == NULL ? -ENOMEM : 0;
678 static inline int ip6_frags_sysctl_register(struct net *net)
684 static int ipv6_frags_init_net(struct net *net)
686 ip6_frags.ctl = &net->ipv6.sysctl.frags;
688 net->ipv6.sysctl.frags.high_thresh = 256 * 1024;
689 net->ipv6.sysctl.frags.low_thresh = 192 * 1024;
690 net->ipv6.sysctl.frags.timeout = IPV6_FRAG_TIMEOUT;
691 net->ipv6.sysctl.frags.secret_interval = 10 * 60 * HZ;
693 inet_frags_init_net(&net->ipv6.frags);
695 return ip6_frags_sysctl_register(net);
698 int __init ipv6_frag_init(void)
702 ret = inet6_add_protocol(&frag_protocol, IPPROTO_FRAGMENT);
706 ipv6_frags_init_net(&init_net);
708 ip6_frags.hashfn = ip6_hashfn;
709 ip6_frags.constructor = ip6_frag_init;
710 ip6_frags.destructor = NULL;
711 ip6_frags.skb_free = NULL;
712 ip6_frags.qsize = sizeof(struct frag_queue);
713 ip6_frags.match = ip6_frag_match;
714 ip6_frags.frag_expire = ip6_frag_expire;
715 inet_frags_init(&ip6_frags);
720 void ipv6_frag_exit(void)
722 inet_frags_fini(&ip6_frags);
723 inet6_del_protocol(&frag_protocol, IPPROTO_FRAGMENT);