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 frag_queue *next;
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? */
98 struct frag_queue **pprev;
103 #define IP6Q_HASHSZ 64
105 static struct frag_queue *ip6_frag_hash[IP6Q_HASHSZ];
106 static rwlock_t ip6_frag_lock = RW_LOCK_UNLOCKED;
107 static u32 ip6_frag_hash_rnd;
108 static LIST_HEAD(ip6_frag_lru_list);
109 int ip6_frag_nqueues = 0;
111 static __inline__ void __fq_unlink(struct frag_queue *fq)
114 fq->next->pprev = fq->pprev;
115 *fq->pprev = fq->next;
116 list_del(&fq->lru_list);
120 static __inline__ void fq_unlink(struct frag_queue *fq)
122 write_lock(&ip6_frag_lock);
124 write_unlock(&ip6_frag_lock);
127 static unsigned int ip6qhashfn(u32 id, struct in6_addr *saddr,
128 struct in6_addr *daddr)
132 a = saddr->s6_addr32[0];
133 b = saddr->s6_addr32[1];
134 c = saddr->s6_addr32[2];
136 a += JHASH_GOLDEN_RATIO;
137 b += JHASH_GOLDEN_RATIO;
138 c += ip6_frag_hash_rnd;
139 __jhash_mix(a, b, c);
141 a += saddr->s6_addr32[3];
142 b += daddr->s6_addr32[0];
143 c += daddr->s6_addr32[1];
144 __jhash_mix(a, b, c);
146 a += daddr->s6_addr32[2];
147 b += daddr->s6_addr32[3];
149 __jhash_mix(a, b, c);
151 return c & (IP6Q_HASHSZ - 1);
154 static struct timer_list ip6_frag_secret_timer;
155 int sysctl_ip6frag_secret_interval = 10 * 60 * HZ;
157 static void ip6_frag_secret_rebuild(unsigned long dummy)
159 unsigned long now = jiffies;
162 write_lock(&ip6_frag_lock);
163 get_random_bytes(&ip6_frag_hash_rnd, sizeof(u32));
164 for (i = 0; i < IP6Q_HASHSZ; i++) {
165 struct frag_queue *q;
167 q = ip6_frag_hash[i];
169 struct frag_queue *next = q->next;
170 unsigned int hval = ip6qhashfn(q->id,
177 q->next->pprev = q->pprev;
180 /* Relink to new hash chain. */
181 if ((q->next = ip6_frag_hash[hval]) != NULL)
182 q->next->pprev = &q->next;
183 ip6_frag_hash[hval] = q;
184 q->pprev = &ip6_frag_hash[hval];
190 write_unlock(&ip6_frag_lock);
192 mod_timer(&ip6_frag_secret_timer, now + sysctl_ip6frag_secret_interval);
195 atomic_t ip6_frag_mem = ATOMIC_INIT(0);
197 /* Memory Tracking Functions. */
198 static inline void frag_kfree_skb(struct sk_buff *skb)
200 atomic_sub(skb->truesize, &ip6_frag_mem);
204 static inline void frag_free_queue(struct frag_queue *fq)
206 atomic_sub(sizeof(struct frag_queue), &ip6_frag_mem);
210 static inline struct frag_queue *frag_alloc_queue(void)
212 struct frag_queue *fq = kmalloc(sizeof(struct frag_queue), GFP_ATOMIC);
216 atomic_add(sizeof(struct frag_queue), &ip6_frag_mem);
220 /* Destruction primitives. */
222 /* Complete destruction of fq. */
223 static void ip6_frag_destroy(struct frag_queue *fq)
227 BUG_TRAP(fq->last_in&COMPLETE);
228 BUG_TRAP(del_timer(&fq->timer) == 0);
230 /* Release all fragment data. */
233 struct sk_buff *xp = fp->next;
242 static __inline__ void fq_put(struct frag_queue *fq)
244 if (atomic_dec_and_test(&fq->refcnt))
245 ip6_frag_destroy(fq);
248 /* Kill fq entry. It is not destroyed immediately,
249 * because caller (and someone more) holds reference count.
251 static __inline__ void fq_kill(struct frag_queue *fq)
253 if (del_timer(&fq->timer))
254 atomic_dec(&fq->refcnt);
256 if (!(fq->last_in & COMPLETE)) {
258 atomic_dec(&fq->refcnt);
259 fq->last_in |= COMPLETE;
263 static void ip6_evictor(void)
265 struct frag_queue *fq;
266 struct list_head *tmp;
269 if (atomic_read(&ip6_frag_mem) <= sysctl_ip6frag_low_thresh)
271 read_lock(&ip6_frag_lock);
272 if (list_empty(&ip6_frag_lru_list)) {
273 read_unlock(&ip6_frag_lock);
276 tmp = ip6_frag_lru_list.next;
277 fq = list_entry(tmp, struct frag_queue, lru_list);
278 atomic_inc(&fq->refcnt);
279 read_unlock(&ip6_frag_lock);
281 spin_lock(&fq->lock);
282 if (!(fq->last_in&COMPLETE))
284 spin_unlock(&fq->lock);
287 IP6_INC_STATS_BH(ReasmFails);
291 static void ip6_frag_expire(unsigned long data)
293 struct frag_queue *fq = (struct frag_queue *) data;
295 spin_lock(&fq->lock);
297 if (fq->last_in & COMPLETE)
302 IP6_INC_STATS_BH(ReasmTimeout);
303 IP6_INC_STATS_BH(ReasmFails);
305 /* Send error only if the first segment arrived. */
306 if (fq->last_in&FIRST_IN && fq->fragments) {
307 struct net_device *dev = dev_get_by_index(fq->iif);
310 But use as source device on which LAST ARRIVED
311 segment was received. And do not use fq->dev
312 pointer directly, device might already disappeared.
315 fq->fragments->dev = dev;
316 icmpv6_send(fq->fragments, ICMPV6_TIME_EXCEED, ICMPV6_EXC_FRAGTIME, 0,
322 spin_unlock(&fq->lock);
326 /* Creation primitives. */
329 static struct frag_queue *ip6_frag_intern(unsigned int hash,
330 struct frag_queue *fq_in)
332 struct frag_queue *fq;
334 write_lock(&ip6_frag_lock);
336 for (fq = ip6_frag_hash[hash]; fq; fq = fq->next) {
337 if (fq->id == fq_in->id &&
338 !ipv6_addr_cmp(&fq_in->saddr, &fq->saddr) &&
339 !ipv6_addr_cmp(&fq_in->daddr, &fq->daddr)) {
340 atomic_inc(&fq->refcnt);
341 write_unlock(&ip6_frag_lock);
342 fq_in->last_in |= COMPLETE;
350 if (!mod_timer(&fq->timer, jiffies + sysctl_ip6frag_time))
351 atomic_inc(&fq->refcnt);
353 atomic_inc(&fq->refcnt);
354 if((fq->next = ip6_frag_hash[hash]) != NULL)
355 fq->next->pprev = &fq->next;
356 ip6_frag_hash[hash] = fq;
357 fq->pprev = &ip6_frag_hash[hash];
358 INIT_LIST_HEAD(&fq->lru_list);
359 list_add_tail(&fq->lru_list, &ip6_frag_lru_list);
361 write_unlock(&ip6_frag_lock);
366 static struct frag_queue *
367 ip6_frag_create(unsigned int hash, u32 id, struct in6_addr *src, struct in6_addr *dst)
369 struct frag_queue *fq;
371 if ((fq = frag_alloc_queue()) == NULL)
374 memset(fq, 0, sizeof(struct frag_queue));
377 ipv6_addr_copy(&fq->saddr, src);
378 ipv6_addr_copy(&fq->daddr, dst);
380 init_timer(&fq->timer);
381 fq->timer.function = ip6_frag_expire;
382 fq->timer.data = (long) fq;
383 fq->lock = SPIN_LOCK_UNLOCKED;
384 atomic_set(&fq->refcnt, 1);
386 return ip6_frag_intern(hash, fq);
389 IP6_INC_STATS_BH(ReasmFails);
393 static __inline__ struct frag_queue *
394 fq_find(u32 id, struct in6_addr *src, struct in6_addr *dst)
396 struct frag_queue *fq;
397 unsigned int hash = ip6qhashfn(id, src, dst);
399 read_lock(&ip6_frag_lock);
400 for(fq = ip6_frag_hash[hash]; fq; fq = fq->next) {
402 !ipv6_addr_cmp(src, &fq->saddr) &&
403 !ipv6_addr_cmp(dst, &fq->daddr)) {
404 atomic_inc(&fq->refcnt);
405 read_unlock(&ip6_frag_lock);
409 read_unlock(&ip6_frag_lock);
411 return ip6_frag_create(hash, id, src, dst);
415 static void ip6_frag_queue(struct frag_queue *fq, struct sk_buff *skb,
416 struct frag_hdr *fhdr, int nhoff)
418 struct sk_buff *prev, *next;
421 if (fq->last_in & COMPLETE)
424 offset = ntohs(fhdr->frag_off) & ~0x7;
425 end = offset + (ntohs(skb->nh.ipv6h->payload_len) -
426 ((u8 *) (fhdr + 1) - (u8 *) (skb->nh.ipv6h + 1)));
428 if ((unsigned int)end > IPV6_MAXPLEN) {
429 IP6_INC_STATS_BH(InHdrErrors);
430 icmpv6_param_prob(skb,ICMPV6_HDR_FIELD, (u8*)&fhdr->frag_off - skb->nh.raw);
434 if (skb->ip_summed == CHECKSUM_HW)
435 skb->csum = csum_sub(skb->csum,
436 csum_partial(skb->nh.raw, (u8*)(fhdr+1)-skb->nh.raw, 0));
438 /* Is this the final fragment? */
439 if (!(fhdr->frag_off & htons(IP6_MF))) {
440 /* If we already have some bits beyond end
441 * or have different end, the segment is corrupted.
444 ((fq->last_in & LAST_IN) && end != fq->len))
446 fq->last_in |= LAST_IN;
449 /* Check if the fragment is rounded to 8 bytes.
450 * Required by the RFC.
453 /* RFC2460 says always send parameter problem in
456 IP6_INC_STATS_BH(InHdrErrors);
457 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
458 offsetof(struct ipv6hdr, payload_len));
462 /* Some bits beyond end -> corruption. */
463 if (fq->last_in & LAST_IN)
472 /* Point into the IP datagram 'data' part. */
473 if (!pskb_pull(skb, (u8 *) (fhdr + 1) - skb->data))
475 if (end-offset < skb->len) {
476 if (pskb_trim(skb, end - offset))
478 if (skb->ip_summed != CHECKSUM_UNNECESSARY)
479 skb->ip_summed = CHECKSUM_NONE;
482 /* Find out which fragments are in front and at the back of us
483 * in the chain of fragments so far. We must know where to put
484 * this fragment, right?
487 for(next = fq->fragments; next != NULL; next = next->next) {
488 if (FRAG6_CB(next)->offset >= offset)
493 /* We found where to put this one. Check for overlap with
494 * preceding fragment, and, if needed, align things so that
495 * any overlaps are eliminated.
498 int i = (FRAG6_CB(prev)->offset + prev->len) - offset;
504 if (!pskb_pull(skb, i))
506 if (skb->ip_summed != CHECKSUM_UNNECESSARY)
507 skb->ip_summed = CHECKSUM_NONE;
511 /* Look for overlap with succeeding segments.
512 * If we can merge fragments, do it.
514 while (next && FRAG6_CB(next)->offset < end) {
515 int i = end - FRAG6_CB(next)->offset; /* overlap is 'i' bytes */
518 /* Eat head of the next overlapped fragment
519 * and leave the loop. The next ones cannot overlap.
521 if (!pskb_pull(next, i))
523 FRAG6_CB(next)->offset += i; /* next fragment */
525 if (next->ip_summed != CHECKSUM_UNNECESSARY)
526 next->ip_summed = CHECKSUM_NONE;
529 struct sk_buff *free_it = next;
531 /* Old fragment is completely overridden with
539 fq->fragments = next;
541 fq->meat -= free_it->len;
542 frag_kfree_skb(free_it);
546 FRAG6_CB(skb)->offset = offset;
548 /* Insert this fragment in the chain of fragments. */
556 fq->iif = skb->dev->ifindex;
558 fq->stamp = skb->stamp;
559 fq->meat += skb->len;
560 atomic_add(skb->truesize, &ip6_frag_mem);
562 /* The first fragment.
563 * nhoffset is obtained from the first fragment, of course.
566 fq->nhoffset = nhoff;
567 fq->last_in |= FIRST_IN;
569 write_lock(&ip6_frag_lock);
570 list_move_tail(&fq->lru_list, &ip6_frag_lru_list);
571 write_unlock(&ip6_frag_lock);
575 IP6_INC_STATS(ReasmFails);
580 * Check if this packet is complete.
581 * Returns NULL on failure by any reason, and pointer
582 * to current nexthdr field in reassembled frame.
584 * It is called with locked fq, and caller must check that
585 * queue is eligible for reassembly i.e. it is not COMPLETE,
586 * the last and the first frames arrived and all the bits are here.
588 static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff **skb_in,
589 unsigned int *nhoffp,
590 struct net_device *dev)
592 struct sk_buff *fp, *head = fq->fragments;
598 BUG_TRAP(head != NULL);
599 BUG_TRAP(FRAG6_CB(head)->offset == 0);
601 /* Unfragmented part is taken from the first segment. */
602 payload_len = (head->data - head->nh.raw) - sizeof(struct ipv6hdr) + fq->len - sizeof(struct frag_hdr);
603 if (payload_len > IPV6_MAXPLEN)
606 /* Head of list must not be cloned. */
607 if (skb_cloned(head) && pskb_expand_head(head, 0, 0, GFP_ATOMIC))
610 /* If the first fragment is fragmented itself, we split
611 * it to two chunks: the first with data and paged part
612 * and the second, holding only fragments. */
613 if (skb_shinfo(head)->frag_list) {
614 struct sk_buff *clone;
617 if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL)
619 clone->next = head->next;
621 skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
622 skb_shinfo(head)->frag_list = NULL;
623 for (i=0; i<skb_shinfo(head)->nr_frags; i++)
624 plen += skb_shinfo(head)->frags[i].size;
625 clone->len = clone->data_len = head->data_len - plen;
626 head->data_len -= clone->len;
627 head->len -= clone->len;
629 clone->ip_summed = head->ip_summed;
630 atomic_add(clone->truesize, &ip6_frag_mem);
633 /* We have to remove fragment header from datagram and to relocate
634 * header in order to calculate ICV correctly. */
635 nhoff = fq->nhoffset;
636 head->nh.raw[nhoff] = head->h.raw[0];
637 memmove(head->head + sizeof(struct frag_hdr), head->head,
638 (head->data - head->head) - sizeof(struct frag_hdr));
639 head->mac.raw += sizeof(struct frag_hdr);
640 head->nh.raw += sizeof(struct frag_hdr);
642 skb_shinfo(head)->frag_list = head->next;
643 head->h.raw = head->data;
644 skb_push(head, head->data - head->nh.raw);
645 atomic_sub(head->truesize, &ip6_frag_mem);
647 for (fp=head->next; fp; fp = fp->next) {
648 head->data_len += fp->len;
649 head->len += fp->len;
650 if (head->ip_summed != fp->ip_summed)
651 head->ip_summed = CHECKSUM_NONE;
652 else if (head->ip_summed == CHECKSUM_HW)
653 head->csum = csum_add(head->csum, fp->csum);
654 head->truesize += fp->truesize;
655 atomic_sub(fp->truesize, &ip6_frag_mem);
660 head->stamp = fq->stamp;
661 head->nh.ipv6h->payload_len = ntohs(payload_len);
665 /* Yes, and fold redundant checksum back. 8) */
666 if (head->ip_summed == CHECKSUM_HW)
667 head->csum = csum_partial(head->nh.raw, head->h.raw-head->nh.raw, head->csum);
669 IP6_INC_STATS_BH(ReasmOKs);
670 fq->fragments = NULL;
676 printk(KERN_DEBUG "ip6_frag_reasm: payload len = %d\n", payload_len);
680 printk(KERN_DEBUG "ip6_frag_reasm: no memory for reassembly\n");
682 IP6_INC_STATS_BH(ReasmFails);
686 static int ipv6_frag_rcv(struct sk_buff **skbp, unsigned int *nhoffp)
688 struct sk_buff *skb = *skbp;
689 struct net_device *dev = skb->dev;
690 struct frag_hdr *fhdr;
691 struct frag_queue *fq;
696 IP6_INC_STATS_BH(ReasmReqds);
698 /* Jumbo payload inhibits frag. header */
699 if (hdr->payload_len==0) {
700 IP6_INC_STATS(InHdrErrors);
701 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD, skb->h.raw-skb->nh.raw);
704 if (!pskb_may_pull(skb, (skb->h.raw-skb->data)+sizeof(struct frag_hdr))) {
705 IP6_INC_STATS(InHdrErrors);
706 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD, skb->h.raw-skb->nh.raw);
711 fhdr = (struct frag_hdr *)skb->h.raw;
713 if (!(fhdr->frag_off & htons(0xFFF9))) {
714 /* It is not a fragmented frame */
715 skb->h.raw += sizeof(struct frag_hdr);
716 IP6_INC_STATS_BH(ReasmOKs);
718 *nhoffp = (u8*)fhdr - skb->nh.raw;
722 if (atomic_read(&ip6_frag_mem) > sysctl_ip6frag_high_thresh)
725 if ((fq = fq_find(fhdr->identification, &hdr->saddr, &hdr->daddr)) != NULL) {
728 spin_lock(&fq->lock);
730 ip6_frag_queue(fq, skb, fhdr, *nhoffp);
732 if (fq->last_in == (FIRST_IN|LAST_IN) &&
734 ret = ip6_frag_reasm(fq, skbp, nhoffp, dev);
736 spin_unlock(&fq->lock);
741 IP6_INC_STATS_BH(ReasmFails);
746 static struct inet6_protocol frag_protocol =
748 .handler = ipv6_frag_rcv,
749 .flags = INET6_PROTO_NOPOLICY,
752 void __init ipv6_frag_init(void)
754 if (inet6_add_protocol(&frag_protocol, IPPROTO_FRAGMENT) < 0)
755 printk(KERN_ERR "ipv6_frag_init: Could not register protocol\n");
757 ip6_frag_hash_rnd = (u32) ((num_physpages ^ (num_physpages>>7)) ^
758 (jiffies ^ (jiffies >> 6)));
760 init_timer(&ip6_frag_secret_timer);
761 ip6_frag_secret_timer.function = ip6_frag_secret_rebuild;
762 ip6_frag_secret_timer.expires = jiffies + sysctl_ip6frag_secret_interval;
763 add_timer(&ip6_frag_secret_timer);