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 inline void frag_free_queue(struct frag_queue *fq, int *work)
158 *work -= sizeof(struct frag_queue);
159 atomic_sub(sizeof(struct frag_queue), &ip6_frags.mem);
163 static inline struct frag_queue *frag_alloc_queue(void)
165 struct frag_queue *fq = kzalloc(sizeof(struct frag_queue), GFP_ATOMIC);
169 atomic_add(sizeof(struct frag_queue), &ip6_frags.mem);
173 /* Destruction primitives. */
175 /* Complete destruction of fq. */
176 static void ip6_frag_destroy(struct frag_queue *fq, int *work)
180 BUG_TRAP(fq->q.last_in&COMPLETE);
181 BUG_TRAP(del_timer(&fq->q.timer) == 0);
183 /* Release all fragment data. */
184 fp = fq->q.fragments;
186 struct sk_buff *xp = fp->next;
188 frag_kfree_skb(fp, work);
192 frag_free_queue(fq, work);
195 static __inline__ void fq_put(struct frag_queue *fq, int *work)
197 if (atomic_dec_and_test(&fq->q.refcnt))
198 ip6_frag_destroy(fq, work);
201 /* Kill fq entry. It is not destroyed immediately,
202 * because caller (and someone more) holds reference count.
204 static __inline__ void fq_kill(struct frag_queue *fq)
206 inet_frag_kill(&fq->q, &ip6_frags);
209 static void ip6_evictor(struct inet6_dev *idev)
211 struct frag_queue *fq;
212 struct list_head *tmp;
215 work = atomic_read(&ip6_frags.mem) - ip6_frags_ctl.low_thresh;
220 read_lock(&ip6_frags.lock);
221 if (list_empty(&ip6_frags.lru_list)) {
222 read_unlock(&ip6_frags.lock);
225 tmp = ip6_frags.lru_list.next;
226 fq = list_entry(tmp, struct frag_queue, q.lru_list);
227 atomic_inc(&fq->q.refcnt);
228 read_unlock(&ip6_frags.lock);
230 spin_lock(&fq->q.lock);
231 if (!(fq->q.last_in&COMPLETE))
233 spin_unlock(&fq->q.lock);
236 IP6_INC_STATS_BH(idev, IPSTATS_MIB_REASMFAILS);
240 static void ip6_frag_expire(unsigned long data)
242 struct frag_queue *fq = (struct frag_queue *) data;
243 struct net_device *dev = NULL;
245 spin_lock(&fq->q.lock);
247 if (fq->q.last_in & COMPLETE)
252 dev = dev_get_by_index(&init_net, fq->iif);
257 IP6_INC_STATS_BH(__in6_dev_get(dev), IPSTATS_MIB_REASMTIMEOUT);
258 IP6_INC_STATS_BH(__in6_dev_get(dev), IPSTATS_MIB_REASMFAILS);
261 /* Don't send error if the first segment did not arrive. */
262 if (!(fq->q.last_in&FIRST_IN) || !fq->q.fragments)
266 But use as source device on which LAST ARRIVED
267 segment was received. And do not use fq->dev
268 pointer directly, device might already disappeared.
270 fq->q.fragments->dev = dev;
271 icmpv6_send(fq->q.fragments, ICMPV6_TIME_EXCEED, ICMPV6_EXC_FRAGTIME, 0, dev);
275 spin_unlock(&fq->q.lock);
279 /* Creation primitives. */
282 static struct frag_queue *ip6_frag_intern(struct frag_queue *fq_in)
284 struct frag_queue *fq;
287 struct hlist_node *n;
290 write_lock(&ip6_frags.lock);
291 hash = ip6qhashfn(fq_in->id, &fq_in->saddr, &fq_in->daddr);
293 hlist_for_each_entry(fq, n, &ip6_frags.hash[hash], q.list) {
294 if (fq->id == fq_in->id &&
295 ipv6_addr_equal(&fq_in->saddr, &fq->saddr) &&
296 ipv6_addr_equal(&fq_in->daddr, &fq->daddr)) {
297 atomic_inc(&fq->q.refcnt);
298 write_unlock(&ip6_frags.lock);
299 fq_in->q.last_in |= COMPLETE;
307 if (!mod_timer(&fq->q.timer, jiffies + ip6_frags_ctl.timeout))
308 atomic_inc(&fq->q.refcnt);
310 atomic_inc(&fq->q.refcnt);
311 hlist_add_head(&fq->q.list, &ip6_frags.hash[hash]);
312 INIT_LIST_HEAD(&fq->q.lru_list);
313 list_add_tail(&fq->q.lru_list, &ip6_frags.lru_list);
315 write_unlock(&ip6_frags.lock);
320 static struct frag_queue *
321 ip6_frag_create(__be32 id, struct in6_addr *src, struct in6_addr *dst,
322 struct inet6_dev *idev)
324 struct frag_queue *fq;
326 if ((fq = frag_alloc_queue()) == NULL)
330 ipv6_addr_copy(&fq->saddr, src);
331 ipv6_addr_copy(&fq->daddr, dst);
333 init_timer(&fq->q.timer);
334 fq->q.timer.function = ip6_frag_expire;
335 fq->q.timer.data = (long) fq;
336 spin_lock_init(&fq->q.lock);
337 atomic_set(&fq->q.refcnt, 1);
339 return ip6_frag_intern(fq);
342 IP6_INC_STATS_BH(idev, IPSTATS_MIB_REASMFAILS);
346 static __inline__ struct frag_queue *
347 fq_find(__be32 id, struct in6_addr *src, struct in6_addr *dst,
348 struct inet6_dev *idev)
350 struct frag_queue *fq;
351 struct hlist_node *n;
354 read_lock(&ip6_frags.lock);
355 hash = ip6qhashfn(id, src, dst);
356 hlist_for_each_entry(fq, n, &ip6_frags.hash[hash], q.list) {
358 ipv6_addr_equal(src, &fq->saddr) &&
359 ipv6_addr_equal(dst, &fq->daddr)) {
360 atomic_inc(&fq->q.refcnt);
361 read_unlock(&ip6_frags.lock);
365 read_unlock(&ip6_frags.lock);
367 return ip6_frag_create(id, src, dst, idev);
371 static int ip6_frag_queue(struct frag_queue *fq, struct sk_buff *skb,
372 struct frag_hdr *fhdr, int nhoff)
374 struct sk_buff *prev, *next;
375 struct net_device *dev;
378 if (fq->q.last_in & COMPLETE)
381 offset = ntohs(fhdr->frag_off) & ~0x7;
382 end = offset + (ntohs(ipv6_hdr(skb)->payload_len) -
383 ((u8 *)(fhdr + 1) - (u8 *)(ipv6_hdr(skb) + 1)));
385 if ((unsigned int)end > IPV6_MAXPLEN) {
386 IP6_INC_STATS_BH(ip6_dst_idev(skb->dst),
387 IPSTATS_MIB_INHDRERRORS);
388 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
389 ((u8 *)&fhdr->frag_off -
390 skb_network_header(skb)));
394 if (skb->ip_summed == CHECKSUM_COMPLETE) {
395 const unsigned char *nh = skb_network_header(skb);
396 skb->csum = csum_sub(skb->csum,
397 csum_partial(nh, (u8 *)(fhdr + 1) - nh,
401 /* Is this the final fragment? */
402 if (!(fhdr->frag_off & htons(IP6_MF))) {
403 /* If we already have some bits beyond end
404 * or have different end, the segment is corrupted.
406 if (end < fq->q.len ||
407 ((fq->q.last_in & LAST_IN) && end != fq->q.len))
409 fq->q.last_in |= LAST_IN;
412 /* Check if the fragment is rounded to 8 bytes.
413 * Required by the RFC.
416 /* RFC2460 says always send parameter problem in
419 IP6_INC_STATS_BH(ip6_dst_idev(skb->dst),
420 IPSTATS_MIB_INHDRERRORS);
421 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
422 offsetof(struct ipv6hdr, payload_len));
425 if (end > fq->q.len) {
426 /* Some bits beyond end -> corruption. */
427 if (fq->q.last_in & LAST_IN)
436 /* Point into the IP datagram 'data' part. */
437 if (!pskb_pull(skb, (u8 *) (fhdr + 1) - skb->data))
440 if (pskb_trim_rcsum(skb, end - offset))
443 /* Find out which fragments are in front and at the back of us
444 * in the chain of fragments so far. We must know where to put
445 * this fragment, right?
448 for(next = fq->q.fragments; next != NULL; next = next->next) {
449 if (FRAG6_CB(next)->offset >= offset)
454 /* We found where to put this one. Check for overlap with
455 * preceding fragment, and, if needed, align things so that
456 * any overlaps are eliminated.
459 int i = (FRAG6_CB(prev)->offset + prev->len) - offset;
465 if (!pskb_pull(skb, i))
467 if (skb->ip_summed != CHECKSUM_UNNECESSARY)
468 skb->ip_summed = CHECKSUM_NONE;
472 /* Look for overlap with succeeding segments.
473 * If we can merge fragments, do it.
475 while (next && FRAG6_CB(next)->offset < end) {
476 int i = end - FRAG6_CB(next)->offset; /* overlap is 'i' bytes */
479 /* Eat head of the next overlapped fragment
480 * and leave the loop. The next ones cannot overlap.
482 if (!pskb_pull(next, i))
484 FRAG6_CB(next)->offset += i; /* next fragment */
486 if (next->ip_summed != CHECKSUM_UNNECESSARY)
487 next->ip_summed = CHECKSUM_NONE;
490 struct sk_buff *free_it = next;
492 /* Old fragment is completely overridden with
500 fq->q.fragments = next;
502 fq->q.meat -= free_it->len;
503 frag_kfree_skb(free_it, NULL);
507 FRAG6_CB(skb)->offset = offset;
509 /* Insert this fragment in the chain of fragments. */
514 fq->q.fragments = skb;
518 fq->iif = dev->ifindex;
521 fq->q.stamp = skb->tstamp;
522 fq->q.meat += skb->len;
523 atomic_add(skb->truesize, &ip6_frags.mem);
525 /* The first fragment.
526 * nhoffset is obtained from the first fragment, of course.
529 fq->nhoffset = nhoff;
530 fq->q.last_in |= FIRST_IN;
533 if (fq->q.last_in == (FIRST_IN | LAST_IN) && fq->q.meat == fq->q.len)
534 return ip6_frag_reasm(fq, prev, dev);
536 write_lock(&ip6_frags.lock);
537 list_move_tail(&fq->q.lru_list, &ip6_frags.lru_list);
538 write_unlock(&ip6_frags.lock);
542 IP6_INC_STATS(ip6_dst_idev(skb->dst), IPSTATS_MIB_REASMFAILS);
548 * Check if this packet is complete.
549 * Returns NULL on failure by any reason, and pointer
550 * to current nexthdr field in reassembled frame.
552 * It is called with locked fq, and caller must check that
553 * queue is eligible for reassembly i.e. it is not COMPLETE,
554 * the last and the first frames arrived and all the bits are here.
556 static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *prev,
557 struct net_device *dev)
559 struct sk_buff *fp, *head = fq->q.fragments;
565 /* Make the one we just received the head. */
568 fp = skb_clone(head, GFP_ATOMIC);
573 fp->next = head->next;
576 skb_morph(head, fq->q.fragments);
577 head->next = fq->q.fragments->next;
579 kfree_skb(fq->q.fragments);
580 fq->q.fragments = head;
583 BUG_TRAP(head != NULL);
584 BUG_TRAP(FRAG6_CB(head)->offset == 0);
586 /* Unfragmented part is taken from the first segment. */
587 payload_len = ((head->data - skb_network_header(head)) -
588 sizeof(struct ipv6hdr) + fq->q.len -
589 sizeof(struct frag_hdr));
590 if (payload_len > IPV6_MAXPLEN)
593 /* Head of list must not be cloned. */
594 if (skb_cloned(head) && pskb_expand_head(head, 0, 0, GFP_ATOMIC))
597 /* If the first fragment is fragmented itself, we split
598 * it to two chunks: the first with data and paged part
599 * and the second, holding only fragments. */
600 if (skb_shinfo(head)->frag_list) {
601 struct sk_buff *clone;
604 if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL)
606 clone->next = head->next;
608 skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
609 skb_shinfo(head)->frag_list = NULL;
610 for (i=0; i<skb_shinfo(head)->nr_frags; i++)
611 plen += skb_shinfo(head)->frags[i].size;
612 clone->len = clone->data_len = head->data_len - plen;
613 head->data_len -= clone->len;
614 head->len -= clone->len;
616 clone->ip_summed = head->ip_summed;
617 atomic_add(clone->truesize, &ip6_frags.mem);
620 /* We have to remove fragment header from datagram and to relocate
621 * header in order to calculate ICV correctly. */
622 nhoff = fq->nhoffset;
623 skb_network_header(head)[nhoff] = skb_transport_header(head)[0];
624 memmove(head->head + sizeof(struct frag_hdr), head->head,
625 (head->data - head->head) - sizeof(struct frag_hdr));
626 head->mac_header += sizeof(struct frag_hdr);
627 head->network_header += sizeof(struct frag_hdr);
629 skb_shinfo(head)->frag_list = head->next;
630 skb_reset_transport_header(head);
631 skb_push(head, head->data - skb_network_header(head));
632 atomic_sub(head->truesize, &ip6_frags.mem);
634 for (fp=head->next; fp; fp = fp->next) {
635 head->data_len += fp->len;
636 head->len += fp->len;
637 if (head->ip_summed != fp->ip_summed)
638 head->ip_summed = CHECKSUM_NONE;
639 else if (head->ip_summed == CHECKSUM_COMPLETE)
640 head->csum = csum_add(head->csum, fp->csum);
641 head->truesize += fp->truesize;
642 atomic_sub(fp->truesize, &ip6_frags.mem);
647 head->tstamp = fq->q.stamp;
648 ipv6_hdr(head)->payload_len = htons(payload_len);
649 IP6CB(head)->nhoff = nhoff;
651 /* Yes, and fold redundant checksum back. 8) */
652 if (head->ip_summed == CHECKSUM_COMPLETE)
653 head->csum = csum_partial(skb_network_header(head),
654 skb_network_header_len(head),
658 IP6_INC_STATS_BH(__in6_dev_get(dev), IPSTATS_MIB_REASMOKS);
660 fq->q.fragments = NULL;
665 printk(KERN_DEBUG "ip6_frag_reasm: payload len = %d\n", payload_len);
669 printk(KERN_DEBUG "ip6_frag_reasm: no memory for reassembly\n");
672 IP6_INC_STATS_BH(__in6_dev_get(dev), IPSTATS_MIB_REASMFAILS);
677 static int ipv6_frag_rcv(struct sk_buff **skbp)
679 struct sk_buff *skb = *skbp;
680 struct frag_hdr *fhdr;
681 struct frag_queue *fq;
682 struct ipv6hdr *hdr = ipv6_hdr(skb);
684 IP6_INC_STATS_BH(ip6_dst_idev(skb->dst), IPSTATS_MIB_REASMREQDS);
686 /* Jumbo payload inhibits frag. header */
687 if (hdr->payload_len==0) {
688 IP6_INC_STATS(ip6_dst_idev(skb->dst), IPSTATS_MIB_INHDRERRORS);
689 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
690 skb_network_header_len(skb));
693 if (!pskb_may_pull(skb, (skb_transport_offset(skb) +
694 sizeof(struct frag_hdr)))) {
695 IP6_INC_STATS(ip6_dst_idev(skb->dst), IPSTATS_MIB_INHDRERRORS);
696 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
697 skb_network_header_len(skb));
702 fhdr = (struct frag_hdr *)skb_transport_header(skb);
704 if (!(fhdr->frag_off & htons(0xFFF9))) {
705 /* It is not a fragmented frame */
706 skb->transport_header += sizeof(struct frag_hdr);
707 IP6_INC_STATS_BH(ip6_dst_idev(skb->dst), IPSTATS_MIB_REASMOKS);
709 IP6CB(skb)->nhoff = (u8 *)fhdr - skb_network_header(skb);
713 if (atomic_read(&ip6_frags.mem) > ip6_frags_ctl.high_thresh)
714 ip6_evictor(ip6_dst_idev(skb->dst));
716 if ((fq = fq_find(fhdr->identification, &hdr->saddr, &hdr->daddr,
717 ip6_dst_idev(skb->dst))) != NULL) {
720 spin_lock(&fq->q.lock);
722 ret = ip6_frag_queue(fq, skb, fhdr, IP6CB(skb)->nhoff);
724 spin_unlock(&fq->q.lock);
729 IP6_INC_STATS_BH(ip6_dst_idev(skb->dst), IPSTATS_MIB_REASMFAILS);
734 static struct inet6_protocol frag_protocol =
736 .handler = ipv6_frag_rcv,
737 .flags = INET6_PROTO_NOPOLICY,
740 void __init ipv6_frag_init(void)
742 if (inet6_add_protocol(&frag_protocol, IPPROTO_FRAGMENT) < 0)
743 printk(KERN_ERR "ipv6_frag_init: Could not register protocol\n");
745 ip6_frags.ctl = &ip6_frags_ctl;
746 ip6_frags.hashfn = ip6_hashfn;
747 inet_frags_init(&ip6_frags);