2 * IPv6 fragment reassembly
3 * Linux INET6 implementation
6 * Pedro Roque <pedro_m@yahoo.com>
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.
27 #include <linux/config.h>
28 #include <linux/errno.h>
29 #include <linux/types.h>
30 #include <linux/string.h>
31 #include <linux/socket.h>
32 #include <linux/sockios.h>
33 #include <linux/sched.h>
34 #include <linux/list.h>
35 #include <linux/net.h>
36 #include <linux/netdevice.h>
37 #include <linux/in6.h>
38 #include <linux/ipv6.h>
39 #include <linux/icmpv6.h>
40 #include <linux/random.h>
41 #include <linux/jhash.h>
47 #include <net/protocol.h>
48 #include <net/transp_v6.h>
49 #include <net/rawv6.h>
50 #include <net/ndisc.h>
51 #include <net/addrconf.h>
53 int sysctl_ip6frag_high_thresh = 256*1024;
54 int sysctl_ip6frag_low_thresh = 192*1024;
56 int sysctl_ip6frag_time = IPV6_FRAG_TIMEOUT;
60 struct inet6_skb_parm h;
64 #define FRAG6_CB(skb) ((struct ip6frag_skb_cb*)((skb)->cb))
68 * Equivalent of ipv4 struct ipq
73 struct frag_queue *next;
74 struct list_head lru_list; /* lru list member */
76 __u32 id; /* fragment id */
77 struct in6_addr saddr;
78 struct in6_addr daddr;
82 struct timer_list timer; /* expire timer */
83 struct sk_buff *fragments;
89 __u8 last_in; /* has first/last segment arrived? */
94 struct frag_queue **pprev;
99 #define IP6Q_HASHSZ 64
101 static struct frag_queue *ip6_frag_hash[IP6Q_HASHSZ];
102 static rwlock_t ip6_frag_lock = RW_LOCK_UNLOCKED;
103 static u32 ip6_frag_hash_rnd;
104 static LIST_HEAD(ip6_frag_lru_list);
105 int ip6_frag_nqueues = 0;
107 static __inline__ void __fq_unlink(struct frag_queue *fq)
110 fq->next->pprev = fq->pprev;
111 *fq->pprev = fq->next;
112 list_del(&fq->lru_list);
116 static __inline__ void fq_unlink(struct frag_queue *fq)
118 write_lock(&ip6_frag_lock);
120 write_unlock(&ip6_frag_lock);
123 static unsigned int ip6qhashfn(u32 id, struct in6_addr *saddr,
124 struct in6_addr *daddr)
128 a = saddr->s6_addr32[0];
129 b = saddr->s6_addr32[1];
130 c = saddr->s6_addr32[2];
132 a += JHASH_GOLDEN_RATIO;
133 b += JHASH_GOLDEN_RATIO;
134 c += ip6_frag_hash_rnd;
135 __jhash_mix(a, b, c);
137 a += saddr->s6_addr32[3];
138 b += daddr->s6_addr32[0];
139 c += daddr->s6_addr32[1];
140 __jhash_mix(a, b, c);
142 a += daddr->s6_addr32[2];
143 b += daddr->s6_addr32[3];
145 __jhash_mix(a, b, c);
147 return c & (IP6Q_HASHSZ - 1);
150 static struct timer_list ip6_frag_secret_timer;
151 static int ip6_frag_secret_interval = 10 * 60 * HZ;
153 static void ip6_frag_secret_rebuild(unsigned long dummy)
155 unsigned long now = jiffies;
158 write_lock(&ip6_frag_lock);
159 get_random_bytes(&ip6_frag_hash_rnd, sizeof(u32));
160 for (i = 0; i < IP6Q_HASHSZ; i++) {
161 struct frag_queue *q;
163 q = ip6_frag_hash[i];
165 struct frag_queue *next = q->next;
166 unsigned int hval = ip6qhashfn(q->id,
173 q->next->pprev = q->pprev;
176 /* Relink to new hash chain. */
177 if ((q->next = ip6_frag_hash[hval]) != NULL)
178 q->next->pprev = &q->next;
179 ip6_frag_hash[hval] = q;
180 q->pprev = &ip6_frag_hash[hval];
186 write_unlock(&ip6_frag_lock);
188 mod_timer(&ip6_frag_secret_timer, now + ip6_frag_secret_interval);
191 atomic_t ip6_frag_mem = ATOMIC_INIT(0);
193 /* Memory Tracking Functions. */
194 static inline void frag_kfree_skb(struct sk_buff *skb, int *work)
197 *work -= skb->truesize;
198 atomic_sub(skb->truesize, &ip6_frag_mem);
202 static inline void frag_free_queue(struct frag_queue *fq, int *work)
205 *work -= sizeof(struct frag_queue);
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, int *work)
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;
235 frag_kfree_skb(fp, work);
239 frag_free_queue(fq, work);
242 static __inline__ void fq_put(struct frag_queue *fq, int *work)
244 if (atomic_dec_and_test(&fq->refcnt))
245 ip6_frag_destroy(fq, work);
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 work = atomic_read(&ip6_frag_mem) - sysctl_ip6frag_low_thresh;
274 read_lock(&ip6_frag_lock);
275 if (list_empty(&ip6_frag_lru_list)) {
276 read_unlock(&ip6_frag_lock);
279 tmp = ip6_frag_lru_list.next;
280 fq = list_entry(tmp, struct frag_queue, lru_list);
281 atomic_inc(&fq->refcnt);
282 read_unlock(&ip6_frag_lock);
284 spin_lock(&fq->lock);
285 if (!(fq->last_in&COMPLETE))
287 spin_unlock(&fq->lock);
290 IP6_INC_STATS_BH(Ip6ReasmFails);
294 static void ip6_frag_expire(unsigned long data)
296 struct frag_queue *fq = (struct frag_queue *) data;
298 spin_lock(&fq->lock);
300 if (fq->last_in & COMPLETE)
305 IP6_INC_STATS_BH(Ip6ReasmTimeout);
306 IP6_INC_STATS_BH(Ip6ReasmFails);
308 /* Send error only if the first segment arrived. */
309 if (fq->last_in&FIRST_IN && fq->fragments) {
310 struct net_device *dev = dev_get_by_index(fq->iif);
313 But use as source device on which LAST ARRIVED
314 segment was received. And do not use fq->dev
315 pointer directly, device might already disappeared.
318 fq->fragments->dev = dev;
319 icmpv6_send(fq->fragments, ICMPV6_TIME_EXCEED, ICMPV6_EXC_FRAGTIME, 0,
325 spin_unlock(&fq->lock);
329 /* Creation primitives. */
332 static struct frag_queue *ip6_frag_intern(unsigned int hash,
333 struct frag_queue *fq_in)
335 struct frag_queue *fq;
337 write_lock(&ip6_frag_lock);
339 for (fq = ip6_frag_hash[hash]; fq; fq = fq->next) {
340 if (fq->id == fq_in->id &&
341 !ipv6_addr_cmp(&fq_in->saddr, &fq->saddr) &&
342 !ipv6_addr_cmp(&fq_in->daddr, &fq->daddr)) {
343 atomic_inc(&fq->refcnt);
344 write_unlock(&ip6_frag_lock);
345 fq_in->last_in |= COMPLETE;
353 if (!mod_timer(&fq->timer, jiffies + sysctl_ip6frag_time))
354 atomic_inc(&fq->refcnt);
356 atomic_inc(&fq->refcnt);
357 if((fq->next = ip6_frag_hash[hash]) != NULL)
358 fq->next->pprev = &fq->next;
359 ip6_frag_hash[hash] = fq;
360 fq->pprev = &ip6_frag_hash[hash];
361 INIT_LIST_HEAD(&fq->lru_list);
362 list_add_tail(&fq->lru_list, &ip6_frag_lru_list);
364 write_unlock(&ip6_frag_lock);
369 static struct frag_queue *
370 ip6_frag_create(unsigned int hash, u32 id, struct in6_addr *src, struct in6_addr *dst)
372 struct frag_queue *fq;
374 if ((fq = frag_alloc_queue()) == NULL)
377 memset(fq, 0, sizeof(struct frag_queue));
380 ipv6_addr_copy(&fq->saddr, src);
381 ipv6_addr_copy(&fq->daddr, dst);
383 /* init_timer has been done by the memset */
384 fq->timer.function = ip6_frag_expire;
385 fq->timer.data = (long) fq;
386 fq->lock = SPIN_LOCK_UNLOCKED;
387 atomic_set(&fq->refcnt, 1);
389 return ip6_frag_intern(hash, fq);
392 IP6_INC_STATS_BH(Ip6ReasmFails);
396 static __inline__ struct frag_queue *
397 fq_find(u32 id, struct in6_addr *src, struct in6_addr *dst)
399 struct frag_queue *fq;
400 unsigned int hash = ip6qhashfn(id, src, dst);
402 read_lock(&ip6_frag_lock);
403 for(fq = ip6_frag_hash[hash]; fq; fq = fq->next) {
405 !ipv6_addr_cmp(src, &fq->saddr) &&
406 !ipv6_addr_cmp(dst, &fq->daddr)) {
407 atomic_inc(&fq->refcnt);
408 read_unlock(&ip6_frag_lock);
412 read_unlock(&ip6_frag_lock);
414 return ip6_frag_create(hash, id, src, dst);
418 static void ip6_frag_queue(struct frag_queue *fq, struct sk_buff *skb,
419 struct frag_hdr *fhdr, int nhoff)
421 struct sk_buff *prev, *next;
424 if (fq->last_in & COMPLETE)
427 offset = ntohs(fhdr->frag_off) & ~0x7;
428 end = offset + (ntohs(skb->nh.ipv6h->payload_len) -
429 ((u8 *) (fhdr + 1) - (u8 *) (skb->nh.ipv6h + 1)));
431 if ((unsigned int)end >= 65536) {
432 icmpv6_param_prob(skb,ICMPV6_HDR_FIELD, (u8*)&fhdr->frag_off - skb->nh.raw);
436 if (skb->ip_summed == CHECKSUM_HW)
437 skb->csum = csum_sub(skb->csum,
438 csum_partial(skb->nh.raw, (u8*)(fhdr+1)-skb->nh.raw, 0));
440 /* Is this the final fragment? */
441 if (!(fhdr->frag_off & htons(0x0001))) {
442 /* If we already have some bits beyond end
443 * or have different end, the segment is corrupted.
446 ((fq->last_in & LAST_IN) && end != fq->len))
448 fq->last_in |= LAST_IN;
451 /* Check if the fragment is rounded to 8 bytes.
452 * Required by the RFC.
455 /* RFC2460 says always send parameter problem in
458 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
459 offsetof(struct ipv6hdr, payload_len));
463 /* Some bits beyond end -> corruption. */
464 if (fq->last_in & LAST_IN)
473 /* Point into the IP datagram 'data' part. */
474 if (!pskb_pull(skb, (u8 *) (fhdr + 1) - skb->data))
476 if (end-offset < skb->len) {
477 if (pskb_trim(skb, end - offset))
479 if (skb->ip_summed != CHECKSUM_UNNECESSARY)
480 skb->ip_summed = CHECKSUM_NONE;
483 /* Find out which fragments are in front and at the back of us
484 * in the chain of fragments so far. We must know where to put
485 * this fragment, right?
488 for(next = fq->fragments; next != NULL; next = next->next) {
489 if (FRAG6_CB(next)->offset >= offset)
494 /* We found where to put this one. Check for overlap with
495 * preceding fragment, and, if needed, align things so that
496 * any overlaps are eliminated.
499 int i = (FRAG6_CB(prev)->offset + prev->len) - offset;
505 if (!pskb_pull(skb, i))
507 if (skb->ip_summed != CHECKSUM_UNNECESSARY)
508 skb->ip_summed = CHECKSUM_NONE;
512 /* Look for overlap with succeeding segments.
513 * If we can merge fragments, do it.
515 while (next && FRAG6_CB(next)->offset < end) {
516 int i = end - FRAG6_CB(next)->offset; /* overlap is 'i' bytes */
519 /* Eat head of the next overlapped fragment
520 * and leave the loop. The next ones cannot overlap.
522 if (!pskb_pull(next, i))
524 FRAG6_CB(next)->offset += i; /* next fragment */
526 if (next->ip_summed != CHECKSUM_UNNECESSARY)
527 next->ip_summed = CHECKSUM_NONE;
530 struct sk_buff *free_it = next;
532 /* Old fragmnet is completely overridden with
540 fq->fragments = next;
542 fq->meat -= free_it->len;
543 frag_kfree_skb(free_it, NULL);
547 FRAG6_CB(skb)->offset = offset;
549 /* Insert this fragment in the chain of fragments. */
557 fq->iif = skb->dev->ifindex;
559 fq->stamp = skb->stamp;
560 fq->meat += skb->len;
561 atomic_add(skb->truesize, &ip6_frag_mem);
563 /* The first fragment.
564 * nhoffset is obtained from the first fragment, of course.
567 fq->nhoffset = nhoff;
568 fq->last_in |= FIRST_IN;
570 write_lock(&ip6_frag_lock);
571 list_move_tail(&fq->lru_list, &ip6_frag_lru_list);
572 write_unlock(&ip6_frag_lock);
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 struct net_device *dev)
591 struct sk_buff *fp, *head = fq->fragments;
592 int remove_fraghdr = 0;
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;
603 nhoff = head->h.raw - head->nh.raw;
605 if (payload_len > 65535) {
607 if (payload_len > 65535)
612 /* Head of list must not be cloned. */
613 if (skb_cloned(head) && pskb_expand_head(head, 0, 0, GFP_ATOMIC))
616 /* If the first fragment is fragmented itself, we split
617 * it to two chunks: the first with data and paged part
618 * and the second, holding only fragments. */
619 if (skb_shinfo(head)->frag_list) {
620 struct sk_buff *clone;
623 if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL)
625 clone->next = head->next;
627 skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
628 skb_shinfo(head)->frag_list = NULL;
629 for (i=0; i<skb_shinfo(head)->nr_frags; i++)
630 plen += skb_shinfo(head)->frags[i].size;
631 clone->len = clone->data_len = head->data_len - plen;
632 head->data_len -= clone->len;
633 head->len -= clone->len;
635 clone->ip_summed = head->ip_summed;
636 atomic_add(clone->truesize, &ip6_frag_mem);
639 /* Normally we do not remove frag header from datagram, but
640 * we have to do this and to relocate header, when payload
642 if (remove_fraghdr) {
643 nhoff = fq->nhoffset;
644 head->nh.raw[nhoff] = head->h.raw[0];
645 memmove(head->head+8, head->head, (head->data-head->head)-8);
649 ((struct frag_hdr*)head->h.raw)->frag_off = 0;
652 skb_shinfo(head)->frag_list = head->next;
653 head->h.raw = head->data;
654 skb_push(head, head->data - head->nh.raw);
655 atomic_sub(head->truesize, &ip6_frag_mem);
657 for (fp=head->next; fp; fp = fp->next) {
658 head->data_len += fp->len;
659 head->len += fp->len;
660 if (head->ip_summed != fp->ip_summed)
661 head->ip_summed = CHECKSUM_NONE;
662 else if (head->ip_summed == CHECKSUM_HW)
663 head->csum = csum_add(head->csum, fp->csum);
664 head->truesize += fp->truesize;
665 atomic_sub(fp->truesize, &ip6_frag_mem);
670 head->stamp = fq->stamp;
671 head->nh.ipv6h->payload_len = ntohs(payload_len);
675 /* Yes, and fold redundant checksum back. 8) */
676 if (head->ip_summed == CHECKSUM_HW)
677 head->csum = csum_partial(head->nh.raw, head->h.raw-head->nh.raw, head->csum);
679 IP6_INC_STATS_BH(Ip6ReasmOKs);
680 fq->fragments = NULL;
685 printk(KERN_DEBUG "ip6_frag_reasm: payload len = %d\n", payload_len);
689 printk(KERN_DEBUG "ip6_frag_reasm: no memory for reassembly\n");
691 IP6_INC_STATS_BH(Ip6ReasmFails);
695 int ipv6_reassembly(struct sk_buff **skbp, int nhoff)
697 struct sk_buff *skb = *skbp;
698 struct net_device *dev = skb->dev;
699 struct frag_hdr *fhdr;
700 struct frag_queue *fq;
705 IP6_INC_STATS_BH(Ip6ReasmReqds);
707 /* Jumbo payload inhibits frag. header */
708 if (hdr->payload_len==0) {
709 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD, skb->h.raw-skb->nh.raw);
712 if (!pskb_may_pull(skb, (skb->h.raw-skb->data)+sizeof(struct frag_hdr))) {
713 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD, skb->h.raw-skb->nh.raw);
718 fhdr = (struct frag_hdr *)skb->h.raw;
720 if (!(fhdr->frag_off & htons(0xFFF9))) {
721 /* It is not a fragmented frame */
722 skb->h.raw += sizeof(struct frag_hdr);
723 IP6_INC_STATS_BH(Ip6ReasmOKs);
725 return (u8*)fhdr - skb->nh.raw;
728 if (atomic_read(&ip6_frag_mem) > sysctl_ip6frag_high_thresh)
731 if ((fq = fq_find(fhdr->identification, &hdr->saddr, &hdr->daddr)) != NULL) {
734 spin_lock(&fq->lock);
736 ip6_frag_queue(fq, skb, fhdr, nhoff);
738 if (fq->last_in == (FIRST_IN|LAST_IN) &&
740 ret = ip6_frag_reasm(fq, skbp, dev);
742 spin_unlock(&fq->lock);
747 IP6_INC_STATS_BH(Ip6ReasmFails);
752 void __init ipv6_frag_init(void)
754 ip6_frag_hash_rnd = (u32) ((num_physpages ^ (num_physpages>>7)) ^
755 (jiffies ^ (jiffies >> 6)));
757 init_timer(&ip6_frag_secret_timer);
758 ip6_frag_secret_timer.function = ip6_frag_secret_rebuild;
759 ip6_frag_secret_timer.expires = jiffies + ip6_frag_secret_interval;
760 add_timer(&ip6_frag_secret_timer);