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 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 __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, int *work)
201 *work -= skb->truesize;
202 atomic_sub(skb->truesize, &ip6_frag_mem);
206 static inline void frag_free_queue(struct frag_queue *fq, int *work)
209 *work -= sizeof(struct frag_queue);
210 atomic_sub(sizeof(struct frag_queue), &ip6_frag_mem);
214 static inline struct frag_queue *frag_alloc_queue(void)
216 struct frag_queue *fq = kmalloc(sizeof(struct frag_queue), GFP_ATOMIC);
220 atomic_add(sizeof(struct frag_queue), &ip6_frag_mem);
224 /* Destruction primitives. */
226 /* Complete destruction of fq. */
227 static void ip6_frag_destroy(struct frag_queue *fq, int *work)
231 BUG_TRAP(fq->last_in&COMPLETE);
232 BUG_TRAP(del_timer(&fq->timer) == 0);
234 /* Release all fragment data. */
237 struct sk_buff *xp = fp->next;
239 frag_kfree_skb(fp, work);
243 frag_free_queue(fq, work);
246 static __inline__ void fq_put(struct frag_queue *fq, int *work)
248 if (atomic_dec_and_test(&fq->refcnt))
249 ip6_frag_destroy(fq, work);
252 /* Kill fq entry. It is not destroyed immediately,
253 * because caller (and someone more) holds reference count.
255 static __inline__ void fq_kill(struct frag_queue *fq)
257 if (del_timer(&fq->timer))
258 atomic_dec(&fq->refcnt);
260 if (!(fq->last_in & COMPLETE)) {
262 atomic_dec(&fq->refcnt);
263 fq->last_in |= COMPLETE;
267 static void ip6_evictor(void)
269 struct frag_queue *fq;
270 struct list_head *tmp;
273 work = atomic_read(&ip6_frag_mem) - sysctl_ip6frag_low_thresh;
278 read_lock(&ip6_frag_lock);
279 if (list_empty(&ip6_frag_lru_list)) {
280 read_unlock(&ip6_frag_lock);
283 tmp = ip6_frag_lru_list.next;
284 fq = list_entry(tmp, struct frag_queue, lru_list);
285 atomic_inc(&fq->refcnt);
286 read_unlock(&ip6_frag_lock);
288 spin_lock(&fq->lock);
289 if (!(fq->last_in&COMPLETE))
291 spin_unlock(&fq->lock);
294 IP6_INC_STATS_BH(IPSTATS_MIB_REASMFAILS);
298 static void ip6_frag_expire(unsigned long data)
300 struct frag_queue *fq = (struct frag_queue *) data;
302 spin_lock(&fq->lock);
304 if (fq->last_in & COMPLETE)
309 IP6_INC_STATS_BH(IPSTATS_MIB_REASMTIMEOUT);
310 IP6_INC_STATS_BH(IPSTATS_MIB_REASMFAILS);
312 /* Send error only if the first segment arrived. */
313 if (fq->last_in&FIRST_IN && fq->fragments) {
314 struct net_device *dev = dev_get_by_index(fq->iif);
317 But use as source device on which LAST ARRIVED
318 segment was received. And do not use fq->dev
319 pointer directly, device might already disappeared.
322 fq->fragments->dev = dev;
323 icmpv6_send(fq->fragments, ICMPV6_TIME_EXCEED, ICMPV6_EXC_FRAGTIME, 0,
329 spin_unlock(&fq->lock);
333 /* Creation primitives. */
336 static struct frag_queue *ip6_frag_intern(unsigned int hash,
337 struct frag_queue *fq_in)
339 struct frag_queue *fq;
341 write_lock(&ip6_frag_lock);
343 for (fq = ip6_frag_hash[hash]; fq; fq = fq->next) {
344 if (fq->id == fq_in->id &&
345 ipv6_addr_equal(&fq_in->saddr, &fq->saddr) &&
346 ipv6_addr_equal(&fq_in->daddr, &fq->daddr)) {
347 atomic_inc(&fq->refcnt);
348 write_unlock(&ip6_frag_lock);
349 fq_in->last_in |= COMPLETE;
357 if (!mod_timer(&fq->timer, jiffies + sysctl_ip6frag_time))
358 atomic_inc(&fq->refcnt);
360 atomic_inc(&fq->refcnt);
361 if((fq->next = ip6_frag_hash[hash]) != NULL)
362 fq->next->pprev = &fq->next;
363 ip6_frag_hash[hash] = fq;
364 fq->pprev = &ip6_frag_hash[hash];
365 INIT_LIST_HEAD(&fq->lru_list);
366 list_add_tail(&fq->lru_list, &ip6_frag_lru_list);
368 write_unlock(&ip6_frag_lock);
373 static struct frag_queue *
374 ip6_frag_create(unsigned int hash, u32 id, struct in6_addr *src, struct in6_addr *dst)
376 struct frag_queue *fq;
378 if ((fq = frag_alloc_queue()) == NULL)
381 memset(fq, 0, sizeof(struct frag_queue));
384 ipv6_addr_copy(&fq->saddr, src);
385 ipv6_addr_copy(&fq->daddr, dst);
387 init_timer(&fq->timer);
388 fq->timer.function = ip6_frag_expire;
389 fq->timer.data = (long) fq;
390 spin_lock_init(&fq->lock);
391 atomic_set(&fq->refcnt, 1);
393 return ip6_frag_intern(hash, fq);
396 IP6_INC_STATS_BH(IPSTATS_MIB_REASMFAILS);
400 static __inline__ struct frag_queue *
401 fq_find(u32 id, struct in6_addr *src, struct in6_addr *dst)
403 struct frag_queue *fq;
404 unsigned int hash = ip6qhashfn(id, src, dst);
406 read_lock(&ip6_frag_lock);
407 for(fq = ip6_frag_hash[hash]; fq; fq = fq->next) {
409 ipv6_addr_equal(src, &fq->saddr) &&
410 ipv6_addr_equal(dst, &fq->daddr)) {
411 atomic_inc(&fq->refcnt);
412 read_unlock(&ip6_frag_lock);
416 read_unlock(&ip6_frag_lock);
418 return ip6_frag_create(hash, id, src, dst);
422 static void ip6_frag_queue(struct frag_queue *fq, struct sk_buff *skb,
423 struct frag_hdr *fhdr, int nhoff)
425 struct sk_buff *prev, *next;
428 if (fq->last_in & COMPLETE)
431 offset = ntohs(fhdr->frag_off) & ~0x7;
432 end = offset + (ntohs(skb->nh.ipv6h->payload_len) -
433 ((u8 *) (fhdr + 1) - (u8 *) (skb->nh.ipv6h + 1)));
435 if ((unsigned int)end > IPV6_MAXPLEN) {
436 IP6_INC_STATS_BH(IPSTATS_MIB_INHDRERRORS);
437 icmpv6_param_prob(skb,ICMPV6_HDR_FIELD, (u8*)&fhdr->frag_off - skb->nh.raw);
441 if (skb->ip_summed == CHECKSUM_HW)
442 skb->csum = csum_sub(skb->csum,
443 csum_partial(skb->nh.raw, (u8*)(fhdr+1)-skb->nh.raw, 0));
445 /* Is this the final fragment? */
446 if (!(fhdr->frag_off & htons(IP6_MF))) {
447 /* If we already have some bits beyond end
448 * or have different end, the segment is corrupted.
451 ((fq->last_in & LAST_IN) && end != fq->len))
453 fq->last_in |= LAST_IN;
456 /* Check if the fragment is rounded to 8 bytes.
457 * Required by the RFC.
460 /* RFC2460 says always send parameter problem in
463 IP6_INC_STATS_BH(IPSTATS_MIB_INHDRERRORS);
464 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
465 offsetof(struct ipv6hdr, payload_len));
469 /* Some bits beyond end -> corruption. */
470 if (fq->last_in & LAST_IN)
479 /* Point into the IP datagram 'data' part. */
480 if (!pskb_pull(skb, (u8 *) (fhdr + 1) - skb->data))
483 if (pskb_trim_rcsum(skb, end - offset))
486 /* Find out which fragments are in front and at the back of us
487 * in the chain of fragments so far. We must know where to put
488 * this fragment, right?
491 for(next = fq->fragments; next != NULL; next = next->next) {
492 if (FRAG6_CB(next)->offset >= offset)
497 /* We found where to put this one. Check for overlap with
498 * preceding fragment, and, if needed, align things so that
499 * any overlaps are eliminated.
502 int i = (FRAG6_CB(prev)->offset + prev->len) - offset;
508 if (!pskb_pull(skb, i))
510 if (skb->ip_summed != CHECKSUM_UNNECESSARY)
511 skb->ip_summed = CHECKSUM_NONE;
515 /* Look for overlap with succeeding segments.
516 * If we can merge fragments, do it.
518 while (next && FRAG6_CB(next)->offset < end) {
519 int i = end - FRAG6_CB(next)->offset; /* overlap is 'i' bytes */
522 /* Eat head of the next overlapped fragment
523 * and leave the loop. The next ones cannot overlap.
525 if (!pskb_pull(next, i))
527 FRAG6_CB(next)->offset += i; /* next fragment */
529 if (next->ip_summed != CHECKSUM_UNNECESSARY)
530 next->ip_summed = CHECKSUM_NONE;
533 struct sk_buff *free_it = next;
535 /* Old fragment is completely overridden with
543 fq->fragments = next;
545 fq->meat -= free_it->len;
546 frag_kfree_skb(free_it, NULL);
550 FRAG6_CB(skb)->offset = offset;
552 /* Insert this fragment in the chain of fragments. */
560 fq->iif = skb->dev->ifindex;
562 skb_get_timestamp(skb, &fq->stamp);
563 fq->meat += skb->len;
564 atomic_add(skb->truesize, &ip6_frag_mem);
566 /* The first fragment.
567 * nhoffset is obtained from the first fragment, of course.
570 fq->nhoffset = nhoff;
571 fq->last_in |= FIRST_IN;
573 write_lock(&ip6_frag_lock);
574 list_move_tail(&fq->lru_list, &ip6_frag_lru_list);
575 write_unlock(&ip6_frag_lock);
579 IP6_INC_STATS(IPSTATS_MIB_REASMFAILS);
584 * Check if this packet is complete.
585 * Returns NULL on failure by any reason, and pointer
586 * to current nexthdr field in reassembled frame.
588 * It is called with locked fq, and caller must check that
589 * queue is eligible for reassembly i.e. it is not COMPLETE,
590 * the last and the first frames arrived and all the bits are here.
592 static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff **skb_in,
593 unsigned int *nhoffp,
594 struct net_device *dev)
596 struct sk_buff *fp, *head = fq->fragments;
602 BUG_TRAP(head != NULL);
603 BUG_TRAP(FRAG6_CB(head)->offset == 0);
605 /* Unfragmented part is taken from the first segment. */
606 payload_len = (head->data - head->nh.raw) - sizeof(struct ipv6hdr) + fq->len - sizeof(struct frag_hdr);
607 if (payload_len > IPV6_MAXPLEN)
610 /* Head of list must not be cloned. */
611 if (skb_cloned(head) && pskb_expand_head(head, 0, 0, GFP_ATOMIC))
614 /* If the first fragment is fragmented itself, we split
615 * it to two chunks: the first with data and paged part
616 * and the second, holding only fragments. */
617 if (skb_shinfo(head)->frag_list) {
618 struct sk_buff *clone;
621 if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL)
623 clone->next = head->next;
625 skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
626 skb_shinfo(head)->frag_list = NULL;
627 for (i=0; i<skb_shinfo(head)->nr_frags; i++)
628 plen += skb_shinfo(head)->frags[i].size;
629 clone->len = clone->data_len = head->data_len - plen;
630 head->data_len -= clone->len;
631 head->len -= clone->len;
633 clone->ip_summed = head->ip_summed;
634 atomic_add(clone->truesize, &ip6_frag_mem);
637 /* We have to remove fragment header from datagram and to relocate
638 * header in order to calculate ICV correctly. */
639 nhoff = fq->nhoffset;
640 head->nh.raw[nhoff] = head->h.raw[0];
641 memmove(head->head + sizeof(struct frag_hdr), head->head,
642 (head->data - head->head) - sizeof(struct frag_hdr));
643 head->mac.raw += sizeof(struct frag_hdr);
644 head->nh.raw += sizeof(struct frag_hdr);
646 skb_shinfo(head)->frag_list = head->next;
647 head->h.raw = head->data;
648 skb_push(head, head->data - head->nh.raw);
649 atomic_sub(head->truesize, &ip6_frag_mem);
651 for (fp=head->next; fp; fp = fp->next) {
652 head->data_len += fp->len;
653 head->len += fp->len;
654 if (head->ip_summed != fp->ip_summed)
655 head->ip_summed = CHECKSUM_NONE;
656 else if (head->ip_summed == CHECKSUM_HW)
657 head->csum = csum_add(head->csum, fp->csum);
658 head->truesize += fp->truesize;
659 atomic_sub(fp->truesize, &ip6_frag_mem);
664 skb_set_timestamp(head, &fq->stamp);
665 head->nh.ipv6h->payload_len = htons(payload_len);
669 /* Yes, and fold redundant checksum back. 8) */
670 if (head->ip_summed == CHECKSUM_HW)
671 head->csum = csum_partial(head->nh.raw, head->h.raw-head->nh.raw, head->csum);
673 IP6_INC_STATS_BH(IPSTATS_MIB_REASMOKS);
674 fq->fragments = NULL;
680 printk(KERN_DEBUG "ip6_frag_reasm: payload len = %d\n", payload_len);
684 printk(KERN_DEBUG "ip6_frag_reasm: no memory for reassembly\n");
686 IP6_INC_STATS_BH(IPSTATS_MIB_REASMFAILS);
690 static int ipv6_frag_rcv(struct sk_buff **skbp, unsigned int *nhoffp)
692 struct sk_buff *skb = *skbp;
693 struct net_device *dev = skb->dev;
694 struct frag_hdr *fhdr;
695 struct frag_queue *fq;
700 IP6_INC_STATS_BH(IPSTATS_MIB_REASMREQDS);
702 /* Jumbo payload inhibits frag. header */
703 if (hdr->payload_len==0) {
704 IP6_INC_STATS(IPSTATS_MIB_INHDRERRORS);
705 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD, skb->h.raw-skb->nh.raw);
708 if (!pskb_may_pull(skb, (skb->h.raw-skb->data)+sizeof(struct frag_hdr))) {
709 IP6_INC_STATS(IPSTATS_MIB_INHDRERRORS);
710 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD, skb->h.raw-skb->nh.raw);
715 fhdr = (struct frag_hdr *)skb->h.raw;
717 if (!(fhdr->frag_off & htons(0xFFF9))) {
718 /* It is not a fragmented frame */
719 skb->h.raw += sizeof(struct frag_hdr);
720 IP6_INC_STATS_BH(IPSTATS_MIB_REASMOKS);
722 *nhoffp = (u8*)fhdr - skb->nh.raw;
726 if (atomic_read(&ip6_frag_mem) > sysctl_ip6frag_high_thresh)
729 if ((fq = fq_find(fhdr->identification, &hdr->saddr, &hdr->daddr)) != NULL) {
732 spin_lock(&fq->lock);
734 ip6_frag_queue(fq, skb, fhdr, *nhoffp);
736 if (fq->last_in == (FIRST_IN|LAST_IN) &&
738 ret = ip6_frag_reasm(fq, skbp, nhoffp, dev);
740 spin_unlock(&fq->lock);
745 IP6_INC_STATS_BH(IPSTATS_MIB_REASMFAILS);
750 static struct inet6_protocol frag_protocol =
752 .handler = ipv6_frag_rcv,
753 .flags = INET6_PROTO_NOPOLICY,
756 void __init ipv6_frag_init(void)
758 if (inet6_add_protocol(&frag_protocol, IPPROTO_FRAGMENT) < 0)
759 printk(KERN_ERR "ipv6_frag_init: Could not register protocol\n");
761 ip6_frag_hash_rnd = (u32) ((num_physpages ^ (num_physpages>>7)) ^
762 (jiffies ^ (jiffies >> 6)));
764 init_timer(&ip6_frag_secret_timer);
765 ip6_frag_secret_timer.function = ip6_frag_secret_rebuild;
766 ip6_frag_secret_timer.expires = jiffies + sysctl_ip6frag_secret_interval;
767 add_timer(&ip6_frag_secret_timer);