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 struct inet_frags_ctl ip6_frags_ctl __read_mostly = {
86 .high_thresh = 256 * 1024,
87 .low_thresh = 192 * 1024,
88 .timeout = IPV6_FRAG_TIMEOUT,
89 .secret_interval = 10 * 60 * HZ,
92 static struct inet_frags ip6_frags;
94 int ip6_frag_nqueues(void)
96 return ip6_frags.nqueues;
99 int ip6_frag_mem(void)
101 return atomic_read(&ip6_frags.mem);
104 static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *prev,
105 struct net_device *dev);
108 * callers should be careful not to use the hash value outside the ipfrag_lock
109 * as doing so could race with ipfrag_hash_rnd being recalculated.
111 static unsigned int ip6qhashfn(__be32 id, struct in6_addr *saddr,
112 struct in6_addr *daddr)
116 a = (__force u32)saddr->s6_addr32[0];
117 b = (__force u32)saddr->s6_addr32[1];
118 c = (__force u32)saddr->s6_addr32[2];
120 a += JHASH_GOLDEN_RATIO;
121 b += JHASH_GOLDEN_RATIO;
123 __jhash_mix(a, b, c);
125 a += (__force u32)saddr->s6_addr32[3];
126 b += (__force u32)daddr->s6_addr32[0];
127 c += (__force u32)daddr->s6_addr32[1];
128 __jhash_mix(a, b, c);
130 a += (__force u32)daddr->s6_addr32[2];
131 b += (__force u32)daddr->s6_addr32[3];
132 c += (__force u32)id;
133 __jhash_mix(a, b, c);
135 return c & (INETFRAGS_HASHSZ - 1);
138 static unsigned int ip6_hashfn(struct inet_frag_queue *q)
140 struct frag_queue *fq;
142 fq = container_of(q, struct frag_queue, q);
143 return ip6qhashfn(fq->id, &fq->saddr, &fq->daddr);
146 /* Memory Tracking Functions. */
147 static inline void frag_kfree_skb(struct sk_buff *skb, int *work)
150 *work -= skb->truesize;
151 atomic_sub(skb->truesize, &ip6_frags.mem);
155 static void ip6_frag_free(struct inet_frag_queue *fq)
157 kfree(container_of(fq, struct frag_queue, q));
160 static inline struct frag_queue *frag_alloc_queue(void)
162 struct frag_queue *fq = kzalloc(sizeof(struct frag_queue), GFP_ATOMIC);
166 atomic_add(sizeof(struct frag_queue), &ip6_frags.mem);
170 /* Destruction primitives. */
172 static __inline__ void fq_put(struct frag_queue *fq)
174 inet_frag_put(&fq->q, &ip6_frags);
177 /* Kill fq entry. It is not destroyed immediately,
178 * because caller (and someone more) holds reference count.
180 static __inline__ void fq_kill(struct frag_queue *fq)
182 inet_frag_kill(&fq->q, &ip6_frags);
185 static void ip6_evictor(struct inet6_dev *idev)
189 evicted = inet_frag_evictor(&ip6_frags);
191 IP6_ADD_STATS_BH(idev, IPSTATS_MIB_REASMFAILS, evicted);
194 static void ip6_frag_expire(unsigned long data)
196 struct frag_queue *fq = (struct frag_queue *) data;
197 struct net_device *dev = NULL;
199 spin_lock(&fq->q.lock);
201 if (fq->q.last_in & COMPLETE)
206 dev = dev_get_by_index(&init_net, fq->iif);
211 IP6_INC_STATS_BH(__in6_dev_get(dev), IPSTATS_MIB_REASMTIMEOUT);
212 IP6_INC_STATS_BH(__in6_dev_get(dev), IPSTATS_MIB_REASMFAILS);
215 /* Don't send error if the first segment did not arrive. */
216 if (!(fq->q.last_in&FIRST_IN) || !fq->q.fragments)
220 But use as source device on which LAST ARRIVED
221 segment was received. And do not use fq->dev
222 pointer directly, device might already disappeared.
224 fq->q.fragments->dev = dev;
225 icmpv6_send(fq->q.fragments, ICMPV6_TIME_EXCEED, ICMPV6_EXC_FRAGTIME, 0, dev);
229 spin_unlock(&fq->q.lock);
233 /* Creation primitives. */
236 static struct frag_queue *ip6_frag_intern(struct frag_queue *fq_in)
238 struct frag_queue *fq;
241 struct hlist_node *n;
244 write_lock(&ip6_frags.lock);
245 hash = ip6qhashfn(fq_in->id, &fq_in->saddr, &fq_in->daddr);
247 hlist_for_each_entry(fq, n, &ip6_frags.hash[hash], q.list) {
248 if (fq->id == fq_in->id &&
249 ipv6_addr_equal(&fq_in->saddr, &fq->saddr) &&
250 ipv6_addr_equal(&fq_in->daddr, &fq->daddr)) {
251 atomic_inc(&fq->q.refcnt);
252 write_unlock(&ip6_frags.lock);
253 fq_in->q.last_in |= COMPLETE;
261 if (!mod_timer(&fq->q.timer, jiffies + ip6_frags_ctl.timeout))
262 atomic_inc(&fq->q.refcnt);
264 atomic_inc(&fq->q.refcnt);
265 hlist_add_head(&fq->q.list, &ip6_frags.hash[hash]);
266 INIT_LIST_HEAD(&fq->q.lru_list);
267 list_add_tail(&fq->q.lru_list, &ip6_frags.lru_list);
269 write_unlock(&ip6_frags.lock);
274 static struct frag_queue *
275 ip6_frag_create(__be32 id, struct in6_addr *src, struct in6_addr *dst,
276 struct inet6_dev *idev)
278 struct frag_queue *fq;
280 if ((fq = frag_alloc_queue()) == NULL)
284 ipv6_addr_copy(&fq->saddr, src);
285 ipv6_addr_copy(&fq->daddr, dst);
287 init_timer(&fq->q.timer);
288 fq->q.timer.function = ip6_frag_expire;
289 fq->q.timer.data = (long) fq;
290 spin_lock_init(&fq->q.lock);
291 atomic_set(&fq->q.refcnt, 1);
293 return ip6_frag_intern(fq);
296 IP6_INC_STATS_BH(idev, IPSTATS_MIB_REASMFAILS);
300 static __inline__ struct frag_queue *
301 fq_find(__be32 id, struct in6_addr *src, struct in6_addr *dst,
302 struct inet6_dev *idev)
304 struct frag_queue *fq;
305 struct hlist_node *n;
308 read_lock(&ip6_frags.lock);
309 hash = ip6qhashfn(id, src, dst);
310 hlist_for_each_entry(fq, n, &ip6_frags.hash[hash], q.list) {
312 ipv6_addr_equal(src, &fq->saddr) &&
313 ipv6_addr_equal(dst, &fq->daddr)) {
314 atomic_inc(&fq->q.refcnt);
315 read_unlock(&ip6_frags.lock);
319 read_unlock(&ip6_frags.lock);
321 return ip6_frag_create(id, src, dst, idev);
325 static int ip6_frag_queue(struct frag_queue *fq, struct sk_buff *skb,
326 struct frag_hdr *fhdr, int nhoff)
328 struct sk_buff *prev, *next;
329 struct net_device *dev;
332 if (fq->q.last_in & COMPLETE)
335 offset = ntohs(fhdr->frag_off) & ~0x7;
336 end = offset + (ntohs(ipv6_hdr(skb)->payload_len) -
337 ((u8 *)(fhdr + 1) - (u8 *)(ipv6_hdr(skb) + 1)));
339 if ((unsigned int)end > IPV6_MAXPLEN) {
340 IP6_INC_STATS_BH(ip6_dst_idev(skb->dst),
341 IPSTATS_MIB_INHDRERRORS);
342 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
343 ((u8 *)&fhdr->frag_off -
344 skb_network_header(skb)));
348 if (skb->ip_summed == CHECKSUM_COMPLETE) {
349 const unsigned char *nh = skb_network_header(skb);
350 skb->csum = csum_sub(skb->csum,
351 csum_partial(nh, (u8 *)(fhdr + 1) - nh,
355 /* Is this the final fragment? */
356 if (!(fhdr->frag_off & htons(IP6_MF))) {
357 /* If we already have some bits beyond end
358 * or have different end, the segment is corrupted.
360 if (end < fq->q.len ||
361 ((fq->q.last_in & LAST_IN) && end != fq->q.len))
363 fq->q.last_in |= LAST_IN;
366 /* Check if the fragment is rounded to 8 bytes.
367 * Required by the RFC.
370 /* RFC2460 says always send parameter problem in
373 IP6_INC_STATS_BH(ip6_dst_idev(skb->dst),
374 IPSTATS_MIB_INHDRERRORS);
375 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
376 offsetof(struct ipv6hdr, payload_len));
379 if (end > fq->q.len) {
380 /* Some bits beyond end -> corruption. */
381 if (fq->q.last_in & LAST_IN)
390 /* Point into the IP datagram 'data' part. */
391 if (!pskb_pull(skb, (u8 *) (fhdr + 1) - skb->data))
394 if (pskb_trim_rcsum(skb, end - offset))
397 /* Find out which fragments are in front and at the back of us
398 * in the chain of fragments so far. We must know where to put
399 * this fragment, right?
402 for(next = fq->q.fragments; next != NULL; next = next->next) {
403 if (FRAG6_CB(next)->offset >= offset)
408 /* We found where to put this one. Check for overlap with
409 * preceding fragment, and, if needed, align things so that
410 * any overlaps are eliminated.
413 int i = (FRAG6_CB(prev)->offset + prev->len) - offset;
419 if (!pskb_pull(skb, i))
421 if (skb->ip_summed != CHECKSUM_UNNECESSARY)
422 skb->ip_summed = CHECKSUM_NONE;
426 /* Look for overlap with succeeding segments.
427 * If we can merge fragments, do it.
429 while (next && FRAG6_CB(next)->offset < end) {
430 int i = end - FRAG6_CB(next)->offset; /* overlap is 'i' bytes */
433 /* Eat head of the next overlapped fragment
434 * and leave the loop. The next ones cannot overlap.
436 if (!pskb_pull(next, i))
438 FRAG6_CB(next)->offset += i; /* next fragment */
440 if (next->ip_summed != CHECKSUM_UNNECESSARY)
441 next->ip_summed = CHECKSUM_NONE;
444 struct sk_buff *free_it = next;
446 /* Old fragment is completely overridden with
454 fq->q.fragments = next;
456 fq->q.meat -= free_it->len;
457 frag_kfree_skb(free_it, NULL);
461 FRAG6_CB(skb)->offset = offset;
463 /* Insert this fragment in the chain of fragments. */
468 fq->q.fragments = skb;
472 fq->iif = dev->ifindex;
475 fq->q.stamp = skb->tstamp;
476 fq->q.meat += skb->len;
477 atomic_add(skb->truesize, &ip6_frags.mem);
479 /* The first fragment.
480 * nhoffset is obtained from the first fragment, of course.
483 fq->nhoffset = nhoff;
484 fq->q.last_in |= FIRST_IN;
487 if (fq->q.last_in == (FIRST_IN | LAST_IN) && fq->q.meat == fq->q.len)
488 return ip6_frag_reasm(fq, prev, dev);
490 write_lock(&ip6_frags.lock);
491 list_move_tail(&fq->q.lru_list, &ip6_frags.lru_list);
492 write_unlock(&ip6_frags.lock);
496 IP6_INC_STATS(ip6_dst_idev(skb->dst), IPSTATS_MIB_REASMFAILS);
502 * Check if this packet is complete.
503 * Returns NULL on failure by any reason, and pointer
504 * to current nexthdr field in reassembled frame.
506 * It is called with locked fq, and caller must check that
507 * queue is eligible for reassembly i.e. it is not COMPLETE,
508 * the last and the first frames arrived and all the bits are here.
510 static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *prev,
511 struct net_device *dev)
513 struct sk_buff *fp, *head = fq->q.fragments;
519 /* Make the one we just received the head. */
522 fp = skb_clone(head, GFP_ATOMIC);
527 fp->next = head->next;
530 skb_morph(head, fq->q.fragments);
531 head->next = fq->q.fragments->next;
533 kfree_skb(fq->q.fragments);
534 fq->q.fragments = head;
537 BUG_TRAP(head != NULL);
538 BUG_TRAP(FRAG6_CB(head)->offset == 0);
540 /* Unfragmented part is taken from the first segment. */
541 payload_len = ((head->data - skb_network_header(head)) -
542 sizeof(struct ipv6hdr) + fq->q.len -
543 sizeof(struct frag_hdr));
544 if (payload_len > IPV6_MAXPLEN)
547 /* Head of list must not be cloned. */
548 if (skb_cloned(head) && pskb_expand_head(head, 0, 0, GFP_ATOMIC))
551 /* If the first fragment is fragmented itself, we split
552 * it to two chunks: the first with data and paged part
553 * and the second, holding only fragments. */
554 if (skb_shinfo(head)->frag_list) {
555 struct sk_buff *clone;
558 if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL)
560 clone->next = head->next;
562 skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
563 skb_shinfo(head)->frag_list = NULL;
564 for (i=0; i<skb_shinfo(head)->nr_frags; i++)
565 plen += skb_shinfo(head)->frags[i].size;
566 clone->len = clone->data_len = head->data_len - plen;
567 head->data_len -= clone->len;
568 head->len -= clone->len;
570 clone->ip_summed = head->ip_summed;
571 atomic_add(clone->truesize, &ip6_frags.mem);
574 /* We have to remove fragment header from datagram and to relocate
575 * header in order to calculate ICV correctly. */
576 nhoff = fq->nhoffset;
577 skb_network_header(head)[nhoff] = skb_transport_header(head)[0];
578 memmove(head->head + sizeof(struct frag_hdr), head->head,
579 (head->data - head->head) - sizeof(struct frag_hdr));
580 head->mac_header += sizeof(struct frag_hdr);
581 head->network_header += sizeof(struct frag_hdr);
583 skb_shinfo(head)->frag_list = head->next;
584 skb_reset_transport_header(head);
585 skb_push(head, head->data - skb_network_header(head));
586 atomic_sub(head->truesize, &ip6_frags.mem);
588 for (fp=head->next; fp; fp = fp->next) {
589 head->data_len += fp->len;
590 head->len += fp->len;
591 if (head->ip_summed != fp->ip_summed)
592 head->ip_summed = CHECKSUM_NONE;
593 else if (head->ip_summed == CHECKSUM_COMPLETE)
594 head->csum = csum_add(head->csum, fp->csum);
595 head->truesize += fp->truesize;
596 atomic_sub(fp->truesize, &ip6_frags.mem);
601 head->tstamp = fq->q.stamp;
602 ipv6_hdr(head)->payload_len = htons(payload_len);
603 IP6CB(head)->nhoff = nhoff;
605 /* Yes, and fold redundant checksum back. 8) */
606 if (head->ip_summed == CHECKSUM_COMPLETE)
607 head->csum = csum_partial(skb_network_header(head),
608 skb_network_header_len(head),
612 IP6_INC_STATS_BH(__in6_dev_get(dev), IPSTATS_MIB_REASMOKS);
614 fq->q.fragments = NULL;
619 printk(KERN_DEBUG "ip6_frag_reasm: payload len = %d\n", payload_len);
623 printk(KERN_DEBUG "ip6_frag_reasm: no memory for reassembly\n");
626 IP6_INC_STATS_BH(__in6_dev_get(dev), IPSTATS_MIB_REASMFAILS);
631 static int ipv6_frag_rcv(struct sk_buff *skb)
633 struct frag_hdr *fhdr;
634 struct frag_queue *fq;
635 struct ipv6hdr *hdr = ipv6_hdr(skb);
637 IP6_INC_STATS_BH(ip6_dst_idev(skb->dst), IPSTATS_MIB_REASMREQDS);
639 /* Jumbo payload inhibits frag. header */
640 if (hdr->payload_len==0) {
641 IP6_INC_STATS(ip6_dst_idev(skb->dst), IPSTATS_MIB_INHDRERRORS);
642 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
643 skb_network_header_len(skb));
646 if (!pskb_may_pull(skb, (skb_transport_offset(skb) +
647 sizeof(struct frag_hdr)))) {
648 IP6_INC_STATS(ip6_dst_idev(skb->dst), IPSTATS_MIB_INHDRERRORS);
649 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
650 skb_network_header_len(skb));
655 fhdr = (struct frag_hdr *)skb_transport_header(skb);
657 if (!(fhdr->frag_off & htons(0xFFF9))) {
658 /* It is not a fragmented frame */
659 skb->transport_header += sizeof(struct frag_hdr);
660 IP6_INC_STATS_BH(ip6_dst_idev(skb->dst), IPSTATS_MIB_REASMOKS);
662 IP6CB(skb)->nhoff = (u8 *)fhdr - skb_network_header(skb);
666 if (atomic_read(&ip6_frags.mem) > ip6_frags_ctl.high_thresh)
667 ip6_evictor(ip6_dst_idev(skb->dst));
669 if ((fq = fq_find(fhdr->identification, &hdr->saddr, &hdr->daddr,
670 ip6_dst_idev(skb->dst))) != NULL) {
673 spin_lock(&fq->q.lock);
675 ret = ip6_frag_queue(fq, skb, fhdr, IP6CB(skb)->nhoff);
677 spin_unlock(&fq->q.lock);
682 IP6_INC_STATS_BH(ip6_dst_idev(skb->dst), IPSTATS_MIB_REASMFAILS);
687 static struct inet6_protocol frag_protocol =
689 .handler = ipv6_frag_rcv,
690 .flags = INET6_PROTO_NOPOLICY,
693 void __init ipv6_frag_init(void)
695 if (inet6_add_protocol(&frag_protocol, IPPROTO_FRAGMENT) < 0)
696 printk(KERN_ERR "ipv6_frag_init: Could not register protocol\n");
698 ip6_frags.ctl = &ip6_frags_ctl;
699 ip6_frags.hashfn = ip6_hashfn;
700 ip6_frags.destructor = ip6_frag_free;
701 ip6_frags.skb_free = NULL;
702 ip6_frags.qsize = sizeof(struct frag_queue);
703 inet_frags_init(&ip6_frags);