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/config.h>
32 #include <linux/errno.h>
33 #include <linux/types.h>
34 #include <linux/string.h>
35 #include <linux/socket.h>
36 #include <linux/sockios.h>
37 #include <linux/jiffies.h>
38 #include <linux/net.h>
39 #include <linux/list.h>
40 #include <linux/netdevice.h>
41 #include <linux/in6.h>
42 #include <linux/ipv6.h>
43 #include <linux/icmpv6.h>
44 #include <linux/random.h>
45 #include <linux/jhash.h>
51 #include <net/protocol.h>
52 #include <net/transp_v6.h>
53 #include <net/rawv6.h>
54 #include <net/ndisc.h>
55 #include <net/addrconf.h>
57 int sysctl_ip6frag_high_thresh = 256*1024;
58 int sysctl_ip6frag_low_thresh = 192*1024;
60 int sysctl_ip6frag_time = IPV6_FRAG_TIMEOUT;
64 struct inet6_skb_parm h;
68 #define FRAG6_CB(skb) ((struct ip6frag_skb_cb*)((skb)->cb))
72 * Equivalent of ipv4 struct ipq
77 struct hlist_node list;
78 struct list_head lru_list; /* lru list member */
80 __u32 id; /* fragment id */
81 struct in6_addr saddr;
82 struct in6_addr daddr;
86 struct timer_list timer; /* expire timer */
87 struct sk_buff *fragments;
93 __u8 last_in; /* has first/last segment arrived? */
102 #define IP6Q_HASHSZ 64
104 static struct hlist_head ip6_frag_hash[IP6Q_HASHSZ];
105 static DEFINE_RWLOCK(ip6_frag_lock);
106 static u32 ip6_frag_hash_rnd;
107 static LIST_HEAD(ip6_frag_lru_list);
108 int ip6_frag_nqueues = 0;
110 static __inline__ void __fq_unlink(struct frag_queue *fq)
112 hlist_del(&fq->list);
113 list_del(&fq->lru_list);
117 static __inline__ void fq_unlink(struct frag_queue *fq)
119 write_lock(&ip6_frag_lock);
121 write_unlock(&ip6_frag_lock);
124 static unsigned int ip6qhashfn(u32 id, struct in6_addr *saddr,
125 struct in6_addr *daddr)
129 a = saddr->s6_addr32[0];
130 b = saddr->s6_addr32[1];
131 c = saddr->s6_addr32[2];
133 a += JHASH_GOLDEN_RATIO;
134 b += JHASH_GOLDEN_RATIO;
135 c += ip6_frag_hash_rnd;
136 __jhash_mix(a, b, c);
138 a += saddr->s6_addr32[3];
139 b += daddr->s6_addr32[0];
140 c += daddr->s6_addr32[1];
141 __jhash_mix(a, b, c);
143 a += daddr->s6_addr32[2];
144 b += daddr->s6_addr32[3];
146 __jhash_mix(a, b, c);
148 return c & (IP6Q_HASHSZ - 1);
151 static struct timer_list ip6_frag_secret_timer;
152 int sysctl_ip6frag_secret_interval = 10 * 60 * HZ;
154 static void ip6_frag_secret_rebuild(unsigned long dummy)
156 unsigned long now = jiffies;
159 write_lock(&ip6_frag_lock);
160 get_random_bytes(&ip6_frag_hash_rnd, sizeof(u32));
161 for (i = 0; i < IP6Q_HASHSZ; i++) {
162 struct frag_queue *q;
163 struct hlist_node *p, *n;
165 hlist_for_each_entry_safe(q, p, n, &ip6_frag_hash[i], list) {
166 unsigned int hval = ip6qhashfn(q->id,
173 /* Relink to new hash chain. */
174 hlist_add_head(&q->list,
175 &ip6_frag_hash[hval]);
180 write_unlock(&ip6_frag_lock);
182 mod_timer(&ip6_frag_secret_timer, now + sysctl_ip6frag_secret_interval);
185 atomic_t ip6_frag_mem = ATOMIC_INIT(0);
187 /* Memory Tracking Functions. */
188 static inline void frag_kfree_skb(struct sk_buff *skb, int *work)
191 *work -= skb->truesize;
192 atomic_sub(skb->truesize, &ip6_frag_mem);
196 static inline void frag_free_queue(struct frag_queue *fq, int *work)
199 *work -= sizeof(struct frag_queue);
200 atomic_sub(sizeof(struct frag_queue), &ip6_frag_mem);
204 static inline struct frag_queue *frag_alloc_queue(void)
206 struct frag_queue *fq = kzalloc(sizeof(struct frag_queue), GFP_ATOMIC);
210 atomic_add(sizeof(struct frag_queue), &ip6_frag_mem);
214 /* Destruction primitives. */
216 /* Complete destruction of fq. */
217 static void ip6_frag_destroy(struct frag_queue *fq, int *work)
221 BUG_TRAP(fq->last_in&COMPLETE);
222 BUG_TRAP(del_timer(&fq->timer) == 0);
224 /* Release all fragment data. */
227 struct sk_buff *xp = fp->next;
229 frag_kfree_skb(fp, work);
233 frag_free_queue(fq, work);
236 static __inline__ void fq_put(struct frag_queue *fq, int *work)
238 if (atomic_dec_and_test(&fq->refcnt))
239 ip6_frag_destroy(fq, work);
242 /* Kill fq entry. It is not destroyed immediately,
243 * because caller (and someone more) holds reference count.
245 static __inline__ void fq_kill(struct frag_queue *fq)
247 if (del_timer(&fq->timer))
248 atomic_dec(&fq->refcnt);
250 if (!(fq->last_in & COMPLETE)) {
252 atomic_dec(&fq->refcnt);
253 fq->last_in |= COMPLETE;
257 static void ip6_evictor(void)
259 struct frag_queue *fq;
260 struct list_head *tmp;
263 work = atomic_read(&ip6_frag_mem) - sysctl_ip6frag_low_thresh;
268 read_lock(&ip6_frag_lock);
269 if (list_empty(&ip6_frag_lru_list)) {
270 read_unlock(&ip6_frag_lock);
273 tmp = ip6_frag_lru_list.next;
274 fq = list_entry(tmp, struct frag_queue, lru_list);
275 atomic_inc(&fq->refcnt);
276 read_unlock(&ip6_frag_lock);
278 spin_lock(&fq->lock);
279 if (!(fq->last_in&COMPLETE))
281 spin_unlock(&fq->lock);
284 IP6_INC_STATS_BH(IPSTATS_MIB_REASMFAILS);
288 static void ip6_frag_expire(unsigned long data)
290 struct frag_queue *fq = (struct frag_queue *) data;
291 struct net_device *dev;
293 spin_lock(&fq->lock);
295 if (fq->last_in & COMPLETE)
300 IP6_INC_STATS_BH(IPSTATS_MIB_REASMTIMEOUT);
301 IP6_INC_STATS_BH(IPSTATS_MIB_REASMFAILS);
303 /* Don't send error if the first segment did not arrive. */
304 if (!(fq->last_in&FIRST_IN) || !fq->fragments)
307 dev = dev_get_by_index(fq->iif);
312 But use as source device on which LAST ARRIVED
313 segment was received. And do not use fq->dev
314 pointer directly, device might already disappeared.
316 fq->fragments->dev = dev;
317 icmpv6_send(fq->fragments, ICMPV6_TIME_EXCEED, ICMPV6_EXC_FRAGTIME, 0, dev);
320 spin_unlock(&fq->lock);
324 /* Creation primitives. */
327 static struct frag_queue *ip6_frag_intern(unsigned int hash,
328 struct frag_queue *fq_in)
330 struct frag_queue *fq;
332 struct hlist_node *n;
335 write_lock(&ip6_frag_lock);
337 hlist_for_each_entry(fq, n, &ip6_frag_hash[hash], list) {
338 if (fq->id == fq_in->id &&
339 ipv6_addr_equal(&fq_in->saddr, &fq->saddr) &&
340 ipv6_addr_equal(&fq_in->daddr, &fq->daddr)) {
341 atomic_inc(&fq->refcnt);
342 write_unlock(&ip6_frag_lock);
343 fq_in->last_in |= COMPLETE;
351 if (!mod_timer(&fq->timer, jiffies + sysctl_ip6frag_time))
352 atomic_inc(&fq->refcnt);
354 atomic_inc(&fq->refcnt);
355 hlist_add_head(&fq->list, &ip6_frag_hash[hash]);
356 INIT_LIST_HEAD(&fq->lru_list);
357 list_add_tail(&fq->lru_list, &ip6_frag_lru_list);
359 write_unlock(&ip6_frag_lock);
364 static struct frag_queue *
365 ip6_frag_create(unsigned int hash, u32 id, struct in6_addr *src, struct in6_addr *dst)
367 struct frag_queue *fq;
369 if ((fq = frag_alloc_queue()) == NULL)
373 ipv6_addr_copy(&fq->saddr, src);
374 ipv6_addr_copy(&fq->daddr, dst);
376 init_timer(&fq->timer);
377 fq->timer.function = ip6_frag_expire;
378 fq->timer.data = (long) fq;
379 spin_lock_init(&fq->lock);
380 atomic_set(&fq->refcnt, 1);
382 return ip6_frag_intern(hash, fq);
385 IP6_INC_STATS_BH(IPSTATS_MIB_REASMFAILS);
389 static __inline__ struct frag_queue *
390 fq_find(u32 id, struct in6_addr *src, struct in6_addr *dst)
392 struct frag_queue *fq;
393 struct hlist_node *n;
394 unsigned int hash = ip6qhashfn(id, src, dst);
396 read_lock(&ip6_frag_lock);
397 hlist_for_each_entry(fq, n, &ip6_frag_hash[hash], list) {
399 ipv6_addr_equal(src, &fq->saddr) &&
400 ipv6_addr_equal(dst, &fq->daddr)) {
401 atomic_inc(&fq->refcnt);
402 read_unlock(&ip6_frag_lock);
406 read_unlock(&ip6_frag_lock);
408 return ip6_frag_create(hash, id, src, dst);
412 static void ip6_frag_queue(struct frag_queue *fq, struct sk_buff *skb,
413 struct frag_hdr *fhdr, int nhoff)
415 struct sk_buff *prev, *next;
418 if (fq->last_in & COMPLETE)
421 offset = ntohs(fhdr->frag_off) & ~0x7;
422 end = offset + (ntohs(skb->nh.ipv6h->payload_len) -
423 ((u8 *) (fhdr + 1) - (u8 *) (skb->nh.ipv6h + 1)));
425 if ((unsigned int)end > IPV6_MAXPLEN) {
426 IP6_INC_STATS_BH(IPSTATS_MIB_INHDRERRORS);
427 icmpv6_param_prob(skb,ICMPV6_HDR_FIELD, (u8*)&fhdr->frag_off - skb->nh.raw);
431 if (skb->ip_summed == CHECKSUM_HW)
432 skb->csum = csum_sub(skb->csum,
433 csum_partial(skb->nh.raw, (u8*)(fhdr+1)-skb->nh.raw, 0));
435 /* Is this the final fragment? */
436 if (!(fhdr->frag_off & htons(IP6_MF))) {
437 /* If we already have some bits beyond end
438 * or have different end, the segment is corrupted.
441 ((fq->last_in & LAST_IN) && end != fq->len))
443 fq->last_in |= LAST_IN;
446 /* Check if the fragment is rounded to 8 bytes.
447 * Required by the RFC.
450 /* RFC2460 says always send parameter problem in
453 IP6_INC_STATS_BH(IPSTATS_MIB_INHDRERRORS);
454 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
455 offsetof(struct ipv6hdr, payload_len));
459 /* Some bits beyond end -> corruption. */
460 if (fq->last_in & LAST_IN)
469 /* Point into the IP datagram 'data' part. */
470 if (!pskb_pull(skb, (u8 *) (fhdr + 1) - skb->data))
473 if (pskb_trim_rcsum(skb, end - offset))
476 /* Find out which fragments are in front and at the back of us
477 * in the chain of fragments so far. We must know where to put
478 * this fragment, right?
481 for(next = fq->fragments; next != NULL; next = next->next) {
482 if (FRAG6_CB(next)->offset >= offset)
487 /* We found where to put this one. Check for overlap with
488 * preceding fragment, and, if needed, align things so that
489 * any overlaps are eliminated.
492 int i = (FRAG6_CB(prev)->offset + prev->len) - offset;
498 if (!pskb_pull(skb, i))
500 if (skb->ip_summed != CHECKSUM_UNNECESSARY)
501 skb->ip_summed = CHECKSUM_NONE;
505 /* Look for overlap with succeeding segments.
506 * If we can merge fragments, do it.
508 while (next && FRAG6_CB(next)->offset < end) {
509 int i = end - FRAG6_CB(next)->offset; /* overlap is 'i' bytes */
512 /* Eat head of the next overlapped fragment
513 * and leave the loop. The next ones cannot overlap.
515 if (!pskb_pull(next, i))
517 FRAG6_CB(next)->offset += i; /* next fragment */
519 if (next->ip_summed != CHECKSUM_UNNECESSARY)
520 next->ip_summed = CHECKSUM_NONE;
523 struct sk_buff *free_it = next;
525 /* Old fragment is completely overridden with
533 fq->fragments = next;
535 fq->meat -= free_it->len;
536 frag_kfree_skb(free_it, NULL);
540 FRAG6_CB(skb)->offset = offset;
542 /* Insert this fragment in the chain of fragments. */
550 fq->iif = skb->dev->ifindex;
552 skb_get_timestamp(skb, &fq->stamp);
553 fq->meat += skb->len;
554 atomic_add(skb->truesize, &ip6_frag_mem);
556 /* The first fragment.
557 * nhoffset is obtained from the first fragment, of course.
560 fq->nhoffset = nhoff;
561 fq->last_in |= FIRST_IN;
563 write_lock(&ip6_frag_lock);
564 list_move_tail(&fq->lru_list, &ip6_frag_lru_list);
565 write_unlock(&ip6_frag_lock);
569 IP6_INC_STATS(IPSTATS_MIB_REASMFAILS);
574 * Check if this packet is complete.
575 * Returns NULL on failure by any reason, and pointer
576 * to current nexthdr field in reassembled frame.
578 * It is called with locked fq, and caller must check that
579 * queue is eligible for reassembly i.e. it is not COMPLETE,
580 * the last and the first frames arrived and all the bits are here.
582 static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff **skb_in,
583 struct net_device *dev)
585 struct sk_buff *fp, *head = fq->fragments;
591 BUG_TRAP(head != NULL);
592 BUG_TRAP(FRAG6_CB(head)->offset == 0);
594 /* Unfragmented part is taken from the first segment. */
595 payload_len = (head->data - head->nh.raw) - sizeof(struct ipv6hdr) + fq->len - sizeof(struct frag_hdr);
596 if (payload_len > IPV6_MAXPLEN)
599 /* Head of list must not be cloned. */
600 if (skb_cloned(head) && pskb_expand_head(head, 0, 0, GFP_ATOMIC))
603 /* If the first fragment is fragmented itself, we split
604 * it to two chunks: the first with data and paged part
605 * and the second, holding only fragments. */
606 if (skb_shinfo(head)->frag_list) {
607 struct sk_buff *clone;
610 if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL)
612 clone->next = head->next;
614 skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
615 skb_shinfo(head)->frag_list = NULL;
616 for (i=0; i<skb_shinfo(head)->nr_frags; i++)
617 plen += skb_shinfo(head)->frags[i].size;
618 clone->len = clone->data_len = head->data_len - plen;
619 head->data_len -= clone->len;
620 head->len -= clone->len;
622 clone->ip_summed = head->ip_summed;
623 atomic_add(clone->truesize, &ip6_frag_mem);
626 /* We have to remove fragment header from datagram and to relocate
627 * header in order to calculate ICV correctly. */
628 nhoff = fq->nhoffset;
629 head->nh.raw[nhoff] = head->h.raw[0];
630 memmove(head->head + sizeof(struct frag_hdr), head->head,
631 (head->data - head->head) - sizeof(struct frag_hdr));
632 head->mac.raw += sizeof(struct frag_hdr);
633 head->nh.raw += sizeof(struct frag_hdr);
635 skb_shinfo(head)->frag_list = head->next;
636 head->h.raw = head->data;
637 skb_push(head, head->data - head->nh.raw);
638 atomic_sub(head->truesize, &ip6_frag_mem);
640 for (fp=head->next; fp; fp = fp->next) {
641 head->data_len += fp->len;
642 head->len += fp->len;
643 if (head->ip_summed != fp->ip_summed)
644 head->ip_summed = CHECKSUM_NONE;
645 else if (head->ip_summed == CHECKSUM_HW)
646 head->csum = csum_add(head->csum, fp->csum);
647 head->truesize += fp->truesize;
648 atomic_sub(fp->truesize, &ip6_frag_mem);
653 skb_set_timestamp(head, &fq->stamp);
654 head->nh.ipv6h->payload_len = htons(payload_len);
655 IP6CB(head)->nhoff = nhoff;
659 /* Yes, and fold redundant checksum back. 8) */
660 if (head->ip_summed == CHECKSUM_HW)
661 head->csum = csum_partial(head->nh.raw, head->h.raw-head->nh.raw, head->csum);
663 IP6_INC_STATS_BH(IPSTATS_MIB_REASMOKS);
664 fq->fragments = NULL;
669 printk(KERN_DEBUG "ip6_frag_reasm: payload len = %d\n", payload_len);
673 printk(KERN_DEBUG "ip6_frag_reasm: no memory for reassembly\n");
675 IP6_INC_STATS_BH(IPSTATS_MIB_REASMFAILS);
679 static int ipv6_frag_rcv(struct sk_buff **skbp)
681 struct sk_buff *skb = *skbp;
682 struct net_device *dev = skb->dev;
683 struct frag_hdr *fhdr;
684 struct frag_queue *fq;
689 IP6_INC_STATS_BH(IPSTATS_MIB_REASMREQDS);
691 /* Jumbo payload inhibits frag. header */
692 if (hdr->payload_len==0) {
693 IP6_INC_STATS(IPSTATS_MIB_INHDRERRORS);
694 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD, skb->h.raw-skb->nh.raw);
697 if (!pskb_may_pull(skb, (skb->h.raw-skb->data)+sizeof(struct frag_hdr))) {
698 IP6_INC_STATS(IPSTATS_MIB_INHDRERRORS);
699 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD, skb->h.raw-skb->nh.raw);
704 fhdr = (struct frag_hdr *)skb->h.raw;
706 if (!(fhdr->frag_off & htons(0xFFF9))) {
707 /* It is not a fragmented frame */
708 skb->h.raw += sizeof(struct frag_hdr);
709 IP6_INC_STATS_BH(IPSTATS_MIB_REASMOKS);
711 IP6CB(skb)->nhoff = (u8*)fhdr - skb->nh.raw;
715 if (atomic_read(&ip6_frag_mem) > sysctl_ip6frag_high_thresh)
718 if ((fq = fq_find(fhdr->identification, &hdr->saddr, &hdr->daddr)) != NULL) {
721 spin_lock(&fq->lock);
723 ip6_frag_queue(fq, skb, fhdr, IP6CB(skb)->nhoff);
725 if (fq->last_in == (FIRST_IN|LAST_IN) &&
727 ret = ip6_frag_reasm(fq, skbp, dev);
729 spin_unlock(&fq->lock);
734 IP6_INC_STATS_BH(IPSTATS_MIB_REASMFAILS);
739 static struct inet6_protocol frag_protocol =
741 .handler = ipv6_frag_rcv,
742 .flags = INET6_PROTO_NOPOLICY,
745 void __init ipv6_frag_init(void)
747 if (inet6_add_protocol(&frag_protocol, IPPROTO_FRAGMENT) < 0)
748 printk(KERN_ERR "ipv6_frag_init: Could not register protocol\n");
750 ip6_frag_hash_rnd = (u32) ((num_physpages ^ (num_physpages>>7)) ^
751 (jiffies ^ (jiffies >> 6)));
753 init_timer(&ip6_frag_secret_timer);
754 ip6_frag_secret_timer.function = ip6_frag_secret_rebuild;
755 ip6_frag_secret_timer.expires = jiffies + sysctl_ip6frag_secret_interval;
756 add_timer(&ip6_frag_secret_timer);