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>
58 int sysctl_ip6frag_high_thresh __read_mostly = 256*1024;
59 int sysctl_ip6frag_low_thresh __read_mostly = 192*1024;
61 int sysctl_ip6frag_time __read_mostly = IPV6_FRAG_TIMEOUT;
65 struct inet6_skb_parm h;
69 #define FRAG6_CB(skb) ((struct ip6frag_skb_cb*)((skb)->cb))
73 * Equivalent of ipv4 struct ipq
78 struct hlist_node list;
79 struct list_head lru_list; /* lru list member */
81 __be32 id; /* fragment id */
82 struct in6_addr saddr;
83 struct in6_addr daddr;
87 struct timer_list timer; /* expire timer */
88 struct sk_buff *fragments;
94 __u8 last_in; /* has first/last segment arrived? */
103 #define IP6Q_HASHSZ 64
105 static struct hlist_head ip6_frag_hash[IP6Q_HASHSZ];
106 static DEFINE_RWLOCK(ip6_frag_lock);
107 static u32 ip6_frag_hash_rnd;
108 static LIST_HEAD(ip6_frag_lru_list);
109 int ip6_frag_nqueues = 0;
111 static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *prev,
112 struct net_device *dev);
114 static __inline__ void __fq_unlink(struct frag_queue *fq)
116 hlist_del(&fq->list);
117 list_del(&fq->lru_list);
121 static __inline__ void fq_unlink(struct frag_queue *fq)
123 write_lock(&ip6_frag_lock);
125 write_unlock(&ip6_frag_lock);
129 * callers should be careful not to use the hash value outside the ipfrag_lock
130 * as doing so could race with ipfrag_hash_rnd being recalculated.
132 static unsigned int ip6qhashfn(__be32 id, struct in6_addr *saddr,
133 struct in6_addr *daddr)
137 a = (__force u32)saddr->s6_addr32[0];
138 b = (__force u32)saddr->s6_addr32[1];
139 c = (__force u32)saddr->s6_addr32[2];
141 a += JHASH_GOLDEN_RATIO;
142 b += JHASH_GOLDEN_RATIO;
143 c += ip6_frag_hash_rnd;
144 __jhash_mix(a, b, c);
146 a += (__force u32)saddr->s6_addr32[3];
147 b += (__force u32)daddr->s6_addr32[0];
148 c += (__force u32)daddr->s6_addr32[1];
149 __jhash_mix(a, b, c);
151 a += (__force u32)daddr->s6_addr32[2];
152 b += (__force u32)daddr->s6_addr32[3];
153 c += (__force u32)id;
154 __jhash_mix(a, b, c);
156 return c & (IP6Q_HASHSZ - 1);
159 static struct timer_list ip6_frag_secret_timer;
160 int sysctl_ip6frag_secret_interval __read_mostly = 10 * 60 * HZ;
162 static void ip6_frag_secret_rebuild(unsigned long dummy)
164 unsigned long now = jiffies;
167 write_lock(&ip6_frag_lock);
168 get_random_bytes(&ip6_frag_hash_rnd, sizeof(u32));
169 for (i = 0; i < IP6Q_HASHSZ; i++) {
170 struct frag_queue *q;
171 struct hlist_node *p, *n;
173 hlist_for_each_entry_safe(q, p, n, &ip6_frag_hash[i], list) {
174 unsigned int hval = ip6qhashfn(q->id,
181 /* Relink to new hash chain. */
182 hlist_add_head(&q->list,
183 &ip6_frag_hash[hval]);
188 write_unlock(&ip6_frag_lock);
190 mod_timer(&ip6_frag_secret_timer, now + sysctl_ip6frag_secret_interval);
193 atomic_t ip6_frag_mem = ATOMIC_INIT(0);
195 /* Memory Tracking Functions. */
196 static inline void frag_kfree_skb(struct sk_buff *skb, int *work)
199 *work -= skb->truesize;
200 atomic_sub(skb->truesize, &ip6_frag_mem);
204 static inline void frag_free_queue(struct frag_queue *fq, int *work)
207 *work -= sizeof(struct frag_queue);
208 atomic_sub(sizeof(struct frag_queue), &ip6_frag_mem);
212 static inline struct frag_queue *frag_alloc_queue(void)
214 struct frag_queue *fq = kzalloc(sizeof(struct frag_queue), GFP_ATOMIC);
218 atomic_add(sizeof(struct frag_queue), &ip6_frag_mem);
222 /* Destruction primitives. */
224 /* Complete destruction of fq. */
225 static void ip6_frag_destroy(struct frag_queue *fq, int *work)
229 BUG_TRAP(fq->last_in&COMPLETE);
230 BUG_TRAP(del_timer(&fq->timer) == 0);
232 /* Release all fragment data. */
235 struct sk_buff *xp = fp->next;
237 frag_kfree_skb(fp, work);
241 frag_free_queue(fq, work);
244 static __inline__ void fq_put(struct frag_queue *fq, int *work)
246 if (atomic_dec_and_test(&fq->refcnt))
247 ip6_frag_destroy(fq, work);
250 /* Kill fq entry. It is not destroyed immediately,
251 * because caller (and someone more) holds reference count.
253 static __inline__ void fq_kill(struct frag_queue *fq)
255 if (del_timer(&fq->timer))
256 atomic_dec(&fq->refcnt);
258 if (!(fq->last_in & COMPLETE)) {
260 atomic_dec(&fq->refcnt);
261 fq->last_in |= COMPLETE;
265 static void ip6_evictor(struct inet6_dev *idev)
267 struct frag_queue *fq;
268 struct list_head *tmp;
271 work = atomic_read(&ip6_frag_mem) - sysctl_ip6frag_low_thresh;
276 read_lock(&ip6_frag_lock);
277 if (list_empty(&ip6_frag_lru_list)) {
278 read_unlock(&ip6_frag_lock);
281 tmp = ip6_frag_lru_list.next;
282 fq = list_entry(tmp, struct frag_queue, lru_list);
283 atomic_inc(&fq->refcnt);
284 read_unlock(&ip6_frag_lock);
286 spin_lock(&fq->lock);
287 if (!(fq->last_in&COMPLETE))
289 spin_unlock(&fq->lock);
292 IP6_INC_STATS_BH(idev, IPSTATS_MIB_REASMFAILS);
296 static void ip6_frag_expire(unsigned long data)
298 struct frag_queue *fq = (struct frag_queue *) data;
299 struct net_device *dev = NULL;
301 spin_lock(&fq->lock);
303 if (fq->last_in & COMPLETE)
308 dev = dev_get_by_index(&init_net, fq->iif);
313 IP6_INC_STATS_BH(__in6_dev_get(dev), IPSTATS_MIB_REASMTIMEOUT);
314 IP6_INC_STATS_BH(__in6_dev_get(dev), IPSTATS_MIB_REASMFAILS);
317 /* Don't send error if the first segment did not arrive. */
318 if (!(fq->last_in&FIRST_IN) || !fq->fragments)
322 But use as source device on which LAST ARRIVED
323 segment was received. And do not use fq->dev
324 pointer directly, device might already disappeared.
326 fq->fragments->dev = dev;
327 icmpv6_send(fq->fragments, ICMPV6_TIME_EXCEED, ICMPV6_EXC_FRAGTIME, 0, dev);
331 spin_unlock(&fq->lock);
335 /* Creation primitives. */
338 static struct frag_queue *ip6_frag_intern(struct frag_queue *fq_in)
340 struct frag_queue *fq;
343 struct hlist_node *n;
346 write_lock(&ip6_frag_lock);
347 hash = ip6qhashfn(fq_in->id, &fq_in->saddr, &fq_in->daddr);
349 hlist_for_each_entry(fq, n, &ip6_frag_hash[hash], list) {
350 if (fq->id == fq_in->id &&
351 ipv6_addr_equal(&fq_in->saddr, &fq->saddr) &&
352 ipv6_addr_equal(&fq_in->daddr, &fq->daddr)) {
353 atomic_inc(&fq->refcnt);
354 write_unlock(&ip6_frag_lock);
355 fq_in->last_in |= COMPLETE;
363 if (!mod_timer(&fq->timer, jiffies + sysctl_ip6frag_time))
364 atomic_inc(&fq->refcnt);
366 atomic_inc(&fq->refcnt);
367 hlist_add_head(&fq->list, &ip6_frag_hash[hash]);
368 INIT_LIST_HEAD(&fq->lru_list);
369 list_add_tail(&fq->lru_list, &ip6_frag_lru_list);
371 write_unlock(&ip6_frag_lock);
376 static struct frag_queue *
377 ip6_frag_create(__be32 id, struct in6_addr *src, struct in6_addr *dst,
378 struct inet6_dev *idev)
380 struct frag_queue *fq;
382 if ((fq = frag_alloc_queue()) == NULL)
386 ipv6_addr_copy(&fq->saddr, src);
387 ipv6_addr_copy(&fq->daddr, dst);
389 init_timer(&fq->timer);
390 fq->timer.function = ip6_frag_expire;
391 fq->timer.data = (long) fq;
392 spin_lock_init(&fq->lock);
393 atomic_set(&fq->refcnt, 1);
395 return ip6_frag_intern(fq);
398 IP6_INC_STATS_BH(idev, IPSTATS_MIB_REASMFAILS);
402 static __inline__ struct frag_queue *
403 fq_find(__be32 id, struct in6_addr *src, struct in6_addr *dst,
404 struct inet6_dev *idev)
406 struct frag_queue *fq;
407 struct hlist_node *n;
410 read_lock(&ip6_frag_lock);
411 hash = ip6qhashfn(id, src, dst);
412 hlist_for_each_entry(fq, n, &ip6_frag_hash[hash], list) {
414 ipv6_addr_equal(src, &fq->saddr) &&
415 ipv6_addr_equal(dst, &fq->daddr)) {
416 atomic_inc(&fq->refcnt);
417 read_unlock(&ip6_frag_lock);
421 read_unlock(&ip6_frag_lock);
423 return ip6_frag_create(id, src, dst, idev);
427 static int ip6_frag_queue(struct frag_queue *fq, struct sk_buff *skb,
428 struct frag_hdr *fhdr, int nhoff)
430 struct sk_buff *prev, *next;
431 struct net_device *dev;
434 if (fq->last_in & COMPLETE)
437 offset = ntohs(fhdr->frag_off) & ~0x7;
438 end = offset + (ntohs(ipv6_hdr(skb)->payload_len) -
439 ((u8 *)(fhdr + 1) - (u8 *)(ipv6_hdr(skb) + 1)));
441 if ((unsigned int)end > IPV6_MAXPLEN) {
442 IP6_INC_STATS_BH(ip6_dst_idev(skb->dst),
443 IPSTATS_MIB_INHDRERRORS);
444 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
445 ((u8 *)&fhdr->frag_off -
446 skb_network_header(skb)));
450 if (skb->ip_summed == CHECKSUM_COMPLETE) {
451 const unsigned char *nh = skb_network_header(skb);
452 skb->csum = csum_sub(skb->csum,
453 csum_partial(nh, (u8 *)(fhdr + 1) - nh,
457 /* Is this the final fragment? */
458 if (!(fhdr->frag_off & htons(IP6_MF))) {
459 /* If we already have some bits beyond end
460 * or have different end, the segment is corrupted.
463 ((fq->last_in & LAST_IN) && end != fq->len))
465 fq->last_in |= LAST_IN;
468 /* Check if the fragment is rounded to 8 bytes.
469 * Required by the RFC.
472 /* RFC2460 says always send parameter problem in
475 IP6_INC_STATS_BH(ip6_dst_idev(skb->dst),
476 IPSTATS_MIB_INHDRERRORS);
477 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
478 offsetof(struct ipv6hdr, payload_len));
482 /* Some bits beyond end -> corruption. */
483 if (fq->last_in & LAST_IN)
492 /* Point into the IP datagram 'data' part. */
493 if (!pskb_pull(skb, (u8 *) (fhdr + 1) - skb->data))
496 if (pskb_trim_rcsum(skb, end - offset))
499 /* Find out which fragments are in front and at the back of us
500 * in the chain of fragments so far. We must know where to put
501 * this fragment, right?
504 for(next = fq->fragments; next != NULL; next = next->next) {
505 if (FRAG6_CB(next)->offset >= offset)
510 /* We found where to put this one. Check for overlap with
511 * preceding fragment, and, if needed, align things so that
512 * any overlaps are eliminated.
515 int i = (FRAG6_CB(prev)->offset + prev->len) - offset;
521 if (!pskb_pull(skb, i))
523 if (skb->ip_summed != CHECKSUM_UNNECESSARY)
524 skb->ip_summed = CHECKSUM_NONE;
528 /* Look for overlap with succeeding segments.
529 * If we can merge fragments, do it.
531 while (next && FRAG6_CB(next)->offset < end) {
532 int i = end - FRAG6_CB(next)->offset; /* overlap is 'i' bytes */
535 /* Eat head of the next overlapped fragment
536 * and leave the loop. The next ones cannot overlap.
538 if (!pskb_pull(next, i))
540 FRAG6_CB(next)->offset += i; /* next fragment */
542 if (next->ip_summed != CHECKSUM_UNNECESSARY)
543 next->ip_summed = CHECKSUM_NONE;
546 struct sk_buff *free_it = next;
548 /* Old fragment is completely overridden with
556 fq->fragments = next;
558 fq->meat -= free_it->len;
559 frag_kfree_skb(free_it, NULL);
563 FRAG6_CB(skb)->offset = offset;
565 /* Insert this fragment in the chain of fragments. */
574 fq->iif = dev->ifindex;
577 fq->stamp = skb->tstamp;
578 fq->meat += skb->len;
579 atomic_add(skb->truesize, &ip6_frag_mem);
581 /* The first fragment.
582 * nhoffset is obtained from the first fragment, of course.
585 fq->nhoffset = nhoff;
586 fq->last_in |= FIRST_IN;
589 if (fq->last_in == (FIRST_IN | LAST_IN) && fq->meat == fq->len)
590 return ip6_frag_reasm(fq, prev, dev);
592 write_lock(&ip6_frag_lock);
593 list_move_tail(&fq->lru_list, &ip6_frag_lru_list);
594 write_unlock(&ip6_frag_lock);
598 IP6_INC_STATS(ip6_dst_idev(skb->dst), IPSTATS_MIB_REASMFAILS);
604 * Check if this packet is complete.
605 * Returns NULL on failure by any reason, and pointer
606 * to current nexthdr field in reassembled frame.
608 * It is called with locked fq, and caller must check that
609 * queue is eligible for reassembly i.e. it is not COMPLETE,
610 * the last and the first frames arrived and all the bits are here.
612 static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *prev,
613 struct net_device *dev)
615 struct sk_buff *fp, *head = fq->fragments;
621 /* Make the one we just received the head. */
624 fp = skb_clone(head, GFP_ATOMIC);
629 fp->next = head->next;
632 skb_morph(head, fq->fragments);
633 head->next = fq->fragments->next;
635 kfree_skb(fq->fragments);
636 fq->fragments = head;
639 BUG_TRAP(head != NULL);
640 BUG_TRAP(FRAG6_CB(head)->offset == 0);
642 /* Unfragmented part is taken from the first segment. */
643 payload_len = ((head->data - skb_network_header(head)) -
644 sizeof(struct ipv6hdr) + fq->len -
645 sizeof(struct frag_hdr));
646 if (payload_len > IPV6_MAXPLEN)
649 /* Head of list must not be cloned. */
650 if (skb_cloned(head) && pskb_expand_head(head, 0, 0, GFP_ATOMIC))
653 /* If the first fragment is fragmented itself, we split
654 * it to two chunks: the first with data and paged part
655 * and the second, holding only fragments. */
656 if (skb_shinfo(head)->frag_list) {
657 struct sk_buff *clone;
660 if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL)
662 clone->next = head->next;
664 skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
665 skb_shinfo(head)->frag_list = NULL;
666 for (i=0; i<skb_shinfo(head)->nr_frags; i++)
667 plen += skb_shinfo(head)->frags[i].size;
668 clone->len = clone->data_len = head->data_len - plen;
669 head->data_len -= clone->len;
670 head->len -= clone->len;
672 clone->ip_summed = head->ip_summed;
673 atomic_add(clone->truesize, &ip6_frag_mem);
676 /* We have to remove fragment header from datagram and to relocate
677 * header in order to calculate ICV correctly. */
678 nhoff = fq->nhoffset;
679 skb_network_header(head)[nhoff] = skb_transport_header(head)[0];
680 memmove(head->head + sizeof(struct frag_hdr), head->head,
681 (head->data - head->head) - sizeof(struct frag_hdr));
682 head->mac_header += sizeof(struct frag_hdr);
683 head->network_header += sizeof(struct frag_hdr);
685 skb_shinfo(head)->frag_list = head->next;
686 skb_reset_transport_header(head);
687 skb_push(head, head->data - skb_network_header(head));
688 atomic_sub(head->truesize, &ip6_frag_mem);
690 for (fp=head->next; fp; fp = fp->next) {
691 head->data_len += fp->len;
692 head->len += fp->len;
693 if (head->ip_summed != fp->ip_summed)
694 head->ip_summed = CHECKSUM_NONE;
695 else if (head->ip_summed == CHECKSUM_COMPLETE)
696 head->csum = csum_add(head->csum, fp->csum);
697 head->truesize += fp->truesize;
698 atomic_sub(fp->truesize, &ip6_frag_mem);
703 head->tstamp = fq->stamp;
704 ipv6_hdr(head)->payload_len = htons(payload_len);
705 IP6CB(head)->nhoff = nhoff;
707 /* Yes, and fold redundant checksum back. 8) */
708 if (head->ip_summed == CHECKSUM_COMPLETE)
709 head->csum = csum_partial(skb_network_header(head),
710 skb_network_header_len(head),
714 IP6_INC_STATS_BH(__in6_dev_get(dev), IPSTATS_MIB_REASMOKS);
716 fq->fragments = NULL;
721 printk(KERN_DEBUG "ip6_frag_reasm: payload len = %d\n", payload_len);
725 printk(KERN_DEBUG "ip6_frag_reasm: no memory for reassembly\n");
728 IP6_INC_STATS_BH(__in6_dev_get(dev), IPSTATS_MIB_REASMFAILS);
733 static int ipv6_frag_rcv(struct sk_buff **skbp)
735 struct sk_buff *skb = *skbp;
736 struct frag_hdr *fhdr;
737 struct frag_queue *fq;
738 struct ipv6hdr *hdr = ipv6_hdr(skb);
740 IP6_INC_STATS_BH(ip6_dst_idev(skb->dst), IPSTATS_MIB_REASMREQDS);
742 /* Jumbo payload inhibits frag. header */
743 if (hdr->payload_len==0) {
744 IP6_INC_STATS(ip6_dst_idev(skb->dst), IPSTATS_MIB_INHDRERRORS);
745 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
746 skb_network_header_len(skb));
749 if (!pskb_may_pull(skb, (skb_transport_offset(skb) +
750 sizeof(struct frag_hdr)))) {
751 IP6_INC_STATS(ip6_dst_idev(skb->dst), IPSTATS_MIB_INHDRERRORS);
752 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
753 skb_network_header_len(skb));
758 fhdr = (struct frag_hdr *)skb_transport_header(skb);
760 if (!(fhdr->frag_off & htons(0xFFF9))) {
761 /* It is not a fragmented frame */
762 skb->transport_header += sizeof(struct frag_hdr);
763 IP6_INC_STATS_BH(ip6_dst_idev(skb->dst), IPSTATS_MIB_REASMOKS);
765 IP6CB(skb)->nhoff = (u8 *)fhdr - skb_network_header(skb);
769 if (atomic_read(&ip6_frag_mem) > sysctl_ip6frag_high_thresh)
770 ip6_evictor(ip6_dst_idev(skb->dst));
772 if ((fq = fq_find(fhdr->identification, &hdr->saddr, &hdr->daddr,
773 ip6_dst_idev(skb->dst))) != NULL) {
776 spin_lock(&fq->lock);
778 ret = ip6_frag_queue(fq, skb, fhdr, IP6CB(skb)->nhoff);
780 spin_unlock(&fq->lock);
785 IP6_INC_STATS_BH(ip6_dst_idev(skb->dst), IPSTATS_MIB_REASMFAILS);
790 static struct inet6_protocol frag_protocol =
792 .handler = ipv6_frag_rcv,
793 .flags = INET6_PROTO_NOPOLICY,
796 void __init ipv6_frag_init(void)
798 if (inet6_add_protocol(&frag_protocol, IPPROTO_FRAGMENT) < 0)
799 printk(KERN_ERR "ipv6_frag_init: Could not register protocol\n");
801 ip6_frag_hash_rnd = (u32) ((num_physpages ^ (num_physpages>>7)) ^
802 (jiffies ^ (jiffies >> 6)));
804 init_timer(&ip6_frag_secret_timer);
805 ip6_frag_secret_timer.function = ip6_frag_secret_rebuild;
806 ip6_frag_secret_timer.expires = jiffies + sysctl_ip6frag_secret_interval;
807 add_timer(&ip6_frag_secret_timer);