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 int ip6_frag_equal(struct inet_frag_queue *q1, struct inet_frag_queue *q2)
148 struct frag_queue *fq1, *fq2;
150 fq1 = container_of(q1, struct frag_queue, q);
151 fq2 = container_of(q2, struct frag_queue, q);
152 return (fq1->id == fq2->id &&
153 ipv6_addr_equal(&fq2->saddr, &fq1->saddr) &&
154 ipv6_addr_equal(&fq2->daddr, &fq1->daddr));
156 EXPORT_SYMBOL(ip6_frag_equal);
158 /* Memory Tracking Functions. */
159 static inline void frag_kfree_skb(struct sk_buff *skb, int *work)
162 *work -= skb->truesize;
163 atomic_sub(skb->truesize, &ip6_frags.mem);
167 void ip6_frag_init(struct inet_frag_queue *q, void *a)
169 struct frag_queue *fq = container_of(q, struct frag_queue, q);
170 struct ip6_create_arg *arg = a;
173 ipv6_addr_copy(&fq->saddr, arg->src);
174 ipv6_addr_copy(&fq->daddr, arg->dst);
176 EXPORT_SYMBOL(ip6_frag_init);
178 static void ip6_frag_free(struct inet_frag_queue *fq)
180 kfree(container_of(fq, struct frag_queue, q));
183 /* Destruction primitives. */
185 static __inline__ void fq_put(struct frag_queue *fq)
187 inet_frag_put(&fq->q, &ip6_frags);
190 /* Kill fq entry. It is not destroyed immediately,
191 * because caller (and someone more) holds reference count.
193 static __inline__ void fq_kill(struct frag_queue *fq)
195 inet_frag_kill(&fq->q, &ip6_frags);
198 static void ip6_evictor(struct inet6_dev *idev)
202 evicted = inet_frag_evictor(&ip6_frags);
204 IP6_ADD_STATS_BH(idev, IPSTATS_MIB_REASMFAILS, evicted);
207 static void ip6_frag_expire(unsigned long data)
209 struct frag_queue *fq;
210 struct net_device *dev = NULL;
212 fq = container_of((struct inet_frag_queue *)data, struct frag_queue, q);
214 spin_lock(&fq->q.lock);
216 if (fq->q.last_in & COMPLETE)
221 dev = dev_get_by_index(&init_net, fq->iif);
226 IP6_INC_STATS_BH(__in6_dev_get(dev), IPSTATS_MIB_REASMTIMEOUT);
227 IP6_INC_STATS_BH(__in6_dev_get(dev), IPSTATS_MIB_REASMFAILS);
230 /* Don't send error if the first segment did not arrive. */
231 if (!(fq->q.last_in&FIRST_IN) || !fq->q.fragments)
235 But use as source device on which LAST ARRIVED
236 segment was received. And do not use fq->dev
237 pointer directly, device might already disappeared.
239 fq->q.fragments->dev = dev;
240 icmpv6_send(fq->q.fragments, ICMPV6_TIME_EXCEED, ICMPV6_EXC_FRAGTIME, 0, dev);
244 spin_unlock(&fq->q.lock);
248 /* Creation primitives. */
250 static struct frag_queue *
251 ip6_frag_create(__be32 id, struct in6_addr *src, struct in6_addr *dst,
252 struct inet6_dev *idev, unsigned int hash)
254 struct inet_frag_queue *q;
255 struct ip6_create_arg arg;
261 q = inet_frag_create(&ip6_frags, &arg, hash);
265 return container_of(q, struct frag_queue, q);
268 IP6_INC_STATS_BH(idev, IPSTATS_MIB_REASMFAILS);
272 static __inline__ struct frag_queue *
273 fq_find(__be32 id, struct in6_addr *src, struct in6_addr *dst,
274 struct inet6_dev *idev)
276 struct frag_queue *fq;
277 struct hlist_node *n;
280 read_lock(&ip6_frags.lock);
281 hash = ip6qhashfn(id, src, dst);
282 hlist_for_each_entry(fq, n, &ip6_frags.hash[hash], q.list) {
284 ipv6_addr_equal(src, &fq->saddr) &&
285 ipv6_addr_equal(dst, &fq->daddr)) {
286 atomic_inc(&fq->q.refcnt);
287 read_unlock(&ip6_frags.lock);
291 read_unlock(&ip6_frags.lock);
293 return ip6_frag_create(id, src, dst, idev, hash);
297 static int ip6_frag_queue(struct frag_queue *fq, struct sk_buff *skb,
298 struct frag_hdr *fhdr, int nhoff)
300 struct sk_buff *prev, *next;
301 struct net_device *dev;
304 if (fq->q.last_in & COMPLETE)
307 offset = ntohs(fhdr->frag_off) & ~0x7;
308 end = offset + (ntohs(ipv6_hdr(skb)->payload_len) -
309 ((u8 *)(fhdr + 1) - (u8 *)(ipv6_hdr(skb) + 1)));
311 if ((unsigned int)end > IPV6_MAXPLEN) {
312 IP6_INC_STATS_BH(ip6_dst_idev(skb->dst),
313 IPSTATS_MIB_INHDRERRORS);
314 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
315 ((u8 *)&fhdr->frag_off -
316 skb_network_header(skb)));
320 if (skb->ip_summed == CHECKSUM_COMPLETE) {
321 const unsigned char *nh = skb_network_header(skb);
322 skb->csum = csum_sub(skb->csum,
323 csum_partial(nh, (u8 *)(fhdr + 1) - nh,
327 /* Is this the final fragment? */
328 if (!(fhdr->frag_off & htons(IP6_MF))) {
329 /* If we already have some bits beyond end
330 * or have different end, the segment is corrupted.
332 if (end < fq->q.len ||
333 ((fq->q.last_in & LAST_IN) && end != fq->q.len))
335 fq->q.last_in |= LAST_IN;
338 /* Check if the fragment is rounded to 8 bytes.
339 * Required by the RFC.
342 /* RFC2460 says always send parameter problem in
345 IP6_INC_STATS_BH(ip6_dst_idev(skb->dst),
346 IPSTATS_MIB_INHDRERRORS);
347 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
348 offsetof(struct ipv6hdr, payload_len));
351 if (end > fq->q.len) {
352 /* Some bits beyond end -> corruption. */
353 if (fq->q.last_in & LAST_IN)
362 /* Point into the IP datagram 'data' part. */
363 if (!pskb_pull(skb, (u8 *) (fhdr + 1) - skb->data))
366 if (pskb_trim_rcsum(skb, end - offset))
369 /* Find out which fragments are in front and at the back of us
370 * in the chain of fragments so far. We must know where to put
371 * this fragment, right?
374 for(next = fq->q.fragments; next != NULL; next = next->next) {
375 if (FRAG6_CB(next)->offset >= offset)
380 /* We found where to put this one. Check for overlap with
381 * preceding fragment, and, if needed, align things so that
382 * any overlaps are eliminated.
385 int i = (FRAG6_CB(prev)->offset + prev->len) - offset;
391 if (!pskb_pull(skb, i))
393 if (skb->ip_summed != CHECKSUM_UNNECESSARY)
394 skb->ip_summed = CHECKSUM_NONE;
398 /* Look for overlap with succeeding segments.
399 * If we can merge fragments, do it.
401 while (next && FRAG6_CB(next)->offset < end) {
402 int i = end - FRAG6_CB(next)->offset; /* overlap is 'i' bytes */
405 /* Eat head of the next overlapped fragment
406 * and leave the loop. The next ones cannot overlap.
408 if (!pskb_pull(next, i))
410 FRAG6_CB(next)->offset += i; /* next fragment */
412 if (next->ip_summed != CHECKSUM_UNNECESSARY)
413 next->ip_summed = CHECKSUM_NONE;
416 struct sk_buff *free_it = next;
418 /* Old fragment is completely overridden with
426 fq->q.fragments = next;
428 fq->q.meat -= free_it->len;
429 frag_kfree_skb(free_it, NULL);
433 FRAG6_CB(skb)->offset = offset;
435 /* Insert this fragment in the chain of fragments. */
440 fq->q.fragments = skb;
444 fq->iif = dev->ifindex;
447 fq->q.stamp = skb->tstamp;
448 fq->q.meat += skb->len;
449 atomic_add(skb->truesize, &ip6_frags.mem);
451 /* The first fragment.
452 * nhoffset is obtained from the first fragment, of course.
455 fq->nhoffset = nhoff;
456 fq->q.last_in |= FIRST_IN;
459 if (fq->q.last_in == (FIRST_IN | LAST_IN) && fq->q.meat == fq->q.len)
460 return ip6_frag_reasm(fq, prev, dev);
462 write_lock(&ip6_frags.lock);
463 list_move_tail(&fq->q.lru_list, &ip6_frags.lru_list);
464 write_unlock(&ip6_frags.lock);
468 IP6_INC_STATS(ip6_dst_idev(skb->dst), IPSTATS_MIB_REASMFAILS);
474 * Check if this packet is complete.
475 * Returns NULL on failure by any reason, and pointer
476 * to current nexthdr field in reassembled frame.
478 * It is called with locked fq, and caller must check that
479 * queue is eligible for reassembly i.e. it is not COMPLETE,
480 * the last and the first frames arrived and all the bits are here.
482 static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *prev,
483 struct net_device *dev)
485 struct sk_buff *fp, *head = fq->q.fragments;
491 /* Make the one we just received the head. */
494 fp = skb_clone(head, GFP_ATOMIC);
499 fp->next = head->next;
502 skb_morph(head, fq->q.fragments);
503 head->next = fq->q.fragments->next;
505 kfree_skb(fq->q.fragments);
506 fq->q.fragments = head;
509 BUG_TRAP(head != NULL);
510 BUG_TRAP(FRAG6_CB(head)->offset == 0);
512 /* Unfragmented part is taken from the first segment. */
513 payload_len = ((head->data - skb_network_header(head)) -
514 sizeof(struct ipv6hdr) + fq->q.len -
515 sizeof(struct frag_hdr));
516 if (payload_len > IPV6_MAXPLEN)
519 /* Head of list must not be cloned. */
520 if (skb_cloned(head) && pskb_expand_head(head, 0, 0, GFP_ATOMIC))
523 /* If the first fragment is fragmented itself, we split
524 * it to two chunks: the first with data and paged part
525 * and the second, holding only fragments. */
526 if (skb_shinfo(head)->frag_list) {
527 struct sk_buff *clone;
530 if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL)
532 clone->next = head->next;
534 skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
535 skb_shinfo(head)->frag_list = NULL;
536 for (i=0; i<skb_shinfo(head)->nr_frags; i++)
537 plen += skb_shinfo(head)->frags[i].size;
538 clone->len = clone->data_len = head->data_len - plen;
539 head->data_len -= clone->len;
540 head->len -= clone->len;
542 clone->ip_summed = head->ip_summed;
543 atomic_add(clone->truesize, &ip6_frags.mem);
546 /* We have to remove fragment header from datagram and to relocate
547 * header in order to calculate ICV correctly. */
548 nhoff = fq->nhoffset;
549 skb_network_header(head)[nhoff] = skb_transport_header(head)[0];
550 memmove(head->head + sizeof(struct frag_hdr), head->head,
551 (head->data - head->head) - sizeof(struct frag_hdr));
552 head->mac_header += sizeof(struct frag_hdr);
553 head->network_header += sizeof(struct frag_hdr);
555 skb_shinfo(head)->frag_list = head->next;
556 skb_reset_transport_header(head);
557 skb_push(head, head->data - skb_network_header(head));
558 atomic_sub(head->truesize, &ip6_frags.mem);
560 for (fp=head->next; fp; fp = fp->next) {
561 head->data_len += fp->len;
562 head->len += fp->len;
563 if (head->ip_summed != fp->ip_summed)
564 head->ip_summed = CHECKSUM_NONE;
565 else if (head->ip_summed == CHECKSUM_COMPLETE)
566 head->csum = csum_add(head->csum, fp->csum);
567 head->truesize += fp->truesize;
568 atomic_sub(fp->truesize, &ip6_frags.mem);
573 head->tstamp = fq->q.stamp;
574 ipv6_hdr(head)->payload_len = htons(payload_len);
575 IP6CB(head)->nhoff = nhoff;
577 /* Yes, and fold redundant checksum back. 8) */
578 if (head->ip_summed == CHECKSUM_COMPLETE)
579 head->csum = csum_partial(skb_network_header(head),
580 skb_network_header_len(head),
584 IP6_INC_STATS_BH(__in6_dev_get(dev), IPSTATS_MIB_REASMOKS);
586 fq->q.fragments = NULL;
591 printk(KERN_DEBUG "ip6_frag_reasm: payload len = %d\n", payload_len);
595 printk(KERN_DEBUG "ip6_frag_reasm: no memory for reassembly\n");
598 IP6_INC_STATS_BH(__in6_dev_get(dev), IPSTATS_MIB_REASMFAILS);
603 static int ipv6_frag_rcv(struct sk_buff *skb)
605 struct frag_hdr *fhdr;
606 struct frag_queue *fq;
607 struct ipv6hdr *hdr = ipv6_hdr(skb);
609 IP6_INC_STATS_BH(ip6_dst_idev(skb->dst), IPSTATS_MIB_REASMREQDS);
611 /* Jumbo payload inhibits frag. header */
612 if (hdr->payload_len==0) {
613 IP6_INC_STATS(ip6_dst_idev(skb->dst), IPSTATS_MIB_INHDRERRORS);
614 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
615 skb_network_header_len(skb));
618 if (!pskb_may_pull(skb, (skb_transport_offset(skb) +
619 sizeof(struct frag_hdr)))) {
620 IP6_INC_STATS(ip6_dst_idev(skb->dst), IPSTATS_MIB_INHDRERRORS);
621 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
622 skb_network_header_len(skb));
627 fhdr = (struct frag_hdr *)skb_transport_header(skb);
629 if (!(fhdr->frag_off & htons(0xFFF9))) {
630 /* It is not a fragmented frame */
631 skb->transport_header += sizeof(struct frag_hdr);
632 IP6_INC_STATS_BH(ip6_dst_idev(skb->dst), IPSTATS_MIB_REASMOKS);
634 IP6CB(skb)->nhoff = (u8 *)fhdr - skb_network_header(skb);
638 if (atomic_read(&ip6_frags.mem) > ip6_frags_ctl.high_thresh)
639 ip6_evictor(ip6_dst_idev(skb->dst));
641 if ((fq = fq_find(fhdr->identification, &hdr->saddr, &hdr->daddr,
642 ip6_dst_idev(skb->dst))) != NULL) {
645 spin_lock(&fq->q.lock);
647 ret = ip6_frag_queue(fq, skb, fhdr, IP6CB(skb)->nhoff);
649 spin_unlock(&fq->q.lock);
654 IP6_INC_STATS_BH(ip6_dst_idev(skb->dst), IPSTATS_MIB_REASMFAILS);
659 static struct inet6_protocol frag_protocol =
661 .handler = ipv6_frag_rcv,
662 .flags = INET6_PROTO_NOPOLICY,
665 void __init ipv6_frag_init(void)
667 if (inet6_add_protocol(&frag_protocol, IPPROTO_FRAGMENT) < 0)
668 printk(KERN_ERR "ipv6_frag_init: Could not register protocol\n");
670 ip6_frags.ctl = &ip6_frags_ctl;
671 ip6_frags.hashfn = ip6_hashfn;
672 ip6_frags.constructor = ip6_frag_init;
673 ip6_frags.destructor = ip6_frag_free;
674 ip6_frags.skb_free = NULL;
675 ip6_frags.qsize = sizeof(struct frag_queue);
676 ip6_frags.equal = ip6_frag_equal;
677 ip6_frags.frag_expire = ip6_frag_expire;
678 inet_frags_init(&ip6_frags);