xref: /linux/net/ipv6/reassembly.c (revision de2fe5e07d58424bc286fff3fd3c1b0bf933cd58)
1 /*
2  *	IPv6 fragment reassembly
3  *	Linux INET6 implementation
4  *
5  *	Authors:
6  *	Pedro Roque		<roque@di.fc.ul.pt>
7  *
8  *	$Id: reassembly.c,v 1.26 2001/03/07 22:00:57 davem Exp $
9  *
10  *	Based on: net/ipv4/ip_fragment.c
11  *
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.
16  */
17 
18 /*
19  *	Fixes:
20  *	Andi Kleen	Make it work with multiple hosts.
21  *			More RFC compliance.
22  *
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{}.
27  *	David Stevens and
28  *	YOSHIFUJI,H. @USAGI	Always remove fragment header to
29  *				calculate ICV correctly.
30  */
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>
46 
47 #include <net/sock.h>
48 #include <net/snmp.h>
49 
50 #include <net/ipv6.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>
56 
57 int sysctl_ip6frag_high_thresh = 256*1024;
58 int sysctl_ip6frag_low_thresh = 192*1024;
59 
60 int sysctl_ip6frag_time = IPV6_FRAG_TIMEOUT;
61 
62 struct ip6frag_skb_cb
63 {
64 	struct inet6_skb_parm	h;
65 	int			offset;
66 };
67 
68 #define FRAG6_CB(skb)	((struct ip6frag_skb_cb*)((skb)->cb))
69 
70 
71 /*
72  *	Equivalent of ipv4 struct ipq
73  */
74 
75 struct frag_queue
76 {
77 	struct hlist_node	list;
78 	struct list_head lru_list;		/* lru list member	*/
79 
80 	__u32			id;		/* fragment id		*/
81 	struct in6_addr		saddr;
82 	struct in6_addr		daddr;
83 
84 	spinlock_t		lock;
85 	atomic_t		refcnt;
86 	struct timer_list	timer;		/* expire timer		*/
87 	struct sk_buff		*fragments;
88 	int			len;
89 	int			meat;
90 	int			iif;
91 	struct timeval		stamp;
92 	unsigned int		csum;
93 	__u8			last_in;	/* has first/last segment arrived? */
94 #define COMPLETE		4
95 #define FIRST_IN		2
96 #define LAST_IN			1
97 	__u16			nhoffset;
98 };
99 
100 /* Hash table. */
101 
102 #define IP6Q_HASHSZ	64
103 
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;
109 
110 static __inline__ void __fq_unlink(struct frag_queue *fq)
111 {
112 	hlist_del(&fq->list);
113 	list_del(&fq->lru_list);
114 	ip6_frag_nqueues--;
115 }
116 
117 static __inline__ void fq_unlink(struct frag_queue *fq)
118 {
119 	write_lock(&ip6_frag_lock);
120 	__fq_unlink(fq);
121 	write_unlock(&ip6_frag_lock);
122 }
123 
124 static unsigned int ip6qhashfn(u32 id, struct in6_addr *saddr,
125 			       struct in6_addr *daddr)
126 {
127 	u32 a, b, c;
128 
129 	a = saddr->s6_addr32[0];
130 	b = saddr->s6_addr32[1];
131 	c = saddr->s6_addr32[2];
132 
133 	a += JHASH_GOLDEN_RATIO;
134 	b += JHASH_GOLDEN_RATIO;
135 	c += ip6_frag_hash_rnd;
136 	__jhash_mix(a, b, c);
137 
138 	a += saddr->s6_addr32[3];
139 	b += daddr->s6_addr32[0];
140 	c += daddr->s6_addr32[1];
141 	__jhash_mix(a, b, c);
142 
143 	a += daddr->s6_addr32[2];
144 	b += daddr->s6_addr32[3];
145 	c += id;
146 	__jhash_mix(a, b, c);
147 
148 	return c & (IP6Q_HASHSZ - 1);
149 }
150 
151 static struct timer_list ip6_frag_secret_timer;
152 int sysctl_ip6frag_secret_interval = 10 * 60 * HZ;
153 
154 static void ip6_frag_secret_rebuild(unsigned long dummy)
155 {
156 	unsigned long now = jiffies;
157 	int i;
158 
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;
164 
165 		hlist_for_each_entry_safe(q, p, n, &ip6_frag_hash[i], list) {
166 			unsigned int hval = ip6qhashfn(q->id,
167 						       &q->saddr,
168 						       &q->daddr);
169 
170 			if (hval != i) {
171 				hlist_del(&q->list);
172 
173 				/* Relink to new hash chain. */
174 				hlist_add_head(&q->list,
175 					       &ip6_frag_hash[hval]);
176 
177 			}
178 		}
179 	}
180 	write_unlock(&ip6_frag_lock);
181 
182 	mod_timer(&ip6_frag_secret_timer, now + sysctl_ip6frag_secret_interval);
183 }
184 
185 atomic_t ip6_frag_mem = ATOMIC_INIT(0);
186 
187 /* Memory Tracking Functions. */
188 static inline void frag_kfree_skb(struct sk_buff *skb, int *work)
189 {
190 	if (work)
191 		*work -= skb->truesize;
192 	atomic_sub(skb->truesize, &ip6_frag_mem);
193 	kfree_skb(skb);
194 }
195 
196 static inline void frag_free_queue(struct frag_queue *fq, int *work)
197 {
198 	if (work)
199 		*work -= sizeof(struct frag_queue);
200 	atomic_sub(sizeof(struct frag_queue), &ip6_frag_mem);
201 	kfree(fq);
202 }
203 
204 static inline struct frag_queue *frag_alloc_queue(void)
205 {
206 	struct frag_queue *fq = kzalloc(sizeof(struct frag_queue), GFP_ATOMIC);
207 
208 	if(!fq)
209 		return NULL;
210 	atomic_add(sizeof(struct frag_queue), &ip6_frag_mem);
211 	return fq;
212 }
213 
214 /* Destruction primitives. */
215 
216 /* Complete destruction of fq. */
217 static void ip6_frag_destroy(struct frag_queue *fq, int *work)
218 {
219 	struct sk_buff *fp;
220 
221 	BUG_TRAP(fq->last_in&COMPLETE);
222 	BUG_TRAP(del_timer(&fq->timer) == 0);
223 
224 	/* Release all fragment data. */
225 	fp = fq->fragments;
226 	while (fp) {
227 		struct sk_buff *xp = fp->next;
228 
229 		frag_kfree_skb(fp, work);
230 		fp = xp;
231 	}
232 
233 	frag_free_queue(fq, work);
234 }
235 
236 static __inline__ void fq_put(struct frag_queue *fq, int *work)
237 {
238 	if (atomic_dec_and_test(&fq->refcnt))
239 		ip6_frag_destroy(fq, work);
240 }
241 
242 /* Kill fq entry. It is not destroyed immediately,
243  * because caller (and someone more) holds reference count.
244  */
245 static __inline__ void fq_kill(struct frag_queue *fq)
246 {
247 	if (del_timer(&fq->timer))
248 		atomic_dec(&fq->refcnt);
249 
250 	if (!(fq->last_in & COMPLETE)) {
251 		fq_unlink(fq);
252 		atomic_dec(&fq->refcnt);
253 		fq->last_in |= COMPLETE;
254 	}
255 }
256 
257 static void ip6_evictor(void)
258 {
259 	struct frag_queue *fq;
260 	struct list_head *tmp;
261 	int work;
262 
263 	work = atomic_read(&ip6_frag_mem) - sysctl_ip6frag_low_thresh;
264 	if (work <= 0)
265 		return;
266 
267 	while(work > 0) {
268 		read_lock(&ip6_frag_lock);
269 		if (list_empty(&ip6_frag_lru_list)) {
270 			read_unlock(&ip6_frag_lock);
271 			return;
272 		}
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);
277 
278 		spin_lock(&fq->lock);
279 		if (!(fq->last_in&COMPLETE))
280 			fq_kill(fq);
281 		spin_unlock(&fq->lock);
282 
283 		fq_put(fq, &work);
284 		IP6_INC_STATS_BH(IPSTATS_MIB_REASMFAILS);
285 	}
286 }
287 
288 static void ip6_frag_expire(unsigned long data)
289 {
290 	struct frag_queue *fq = (struct frag_queue *) data;
291 	struct net_device *dev;
292 
293 	spin_lock(&fq->lock);
294 
295 	if (fq->last_in & COMPLETE)
296 		goto out;
297 
298 	fq_kill(fq);
299 
300 	IP6_INC_STATS_BH(IPSTATS_MIB_REASMTIMEOUT);
301 	IP6_INC_STATS_BH(IPSTATS_MIB_REASMFAILS);
302 
303 	/* Don't send error if the first segment did not arrive. */
304 	if (!(fq->last_in&FIRST_IN) || !fq->fragments)
305 		goto out;
306 
307 	dev = dev_get_by_index(fq->iif);
308 	if (!dev)
309 		goto out;
310 
311 	/*
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.
315 	 */
316 	fq->fragments->dev = dev;
317 	icmpv6_send(fq->fragments, ICMPV6_TIME_EXCEED, ICMPV6_EXC_FRAGTIME, 0, dev);
318 	dev_put(dev);
319 out:
320 	spin_unlock(&fq->lock);
321 	fq_put(fq, NULL);
322 }
323 
324 /* Creation primitives. */
325 
326 
327 static struct frag_queue *ip6_frag_intern(unsigned int hash,
328 					  struct frag_queue *fq_in)
329 {
330 	struct frag_queue *fq;
331 #ifdef CONFIG_SMP
332 	struct hlist_node *n;
333 #endif
334 
335 	write_lock(&ip6_frag_lock);
336 #ifdef CONFIG_SMP
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;
344 			fq_put(fq_in, NULL);
345 			return fq;
346 		}
347 	}
348 #endif
349 	fq = fq_in;
350 
351 	if (!mod_timer(&fq->timer, jiffies + sysctl_ip6frag_time))
352 		atomic_inc(&fq->refcnt);
353 
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);
358 	ip6_frag_nqueues++;
359 	write_unlock(&ip6_frag_lock);
360 	return fq;
361 }
362 
363 
364 static struct frag_queue *
365 ip6_frag_create(unsigned int hash, u32 id, struct in6_addr *src, struct in6_addr *dst)
366 {
367 	struct frag_queue *fq;
368 
369 	if ((fq = frag_alloc_queue()) == NULL)
370 		goto oom;
371 
372 	fq->id = id;
373 	ipv6_addr_copy(&fq->saddr, src);
374 	ipv6_addr_copy(&fq->daddr, dst);
375 
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);
381 
382 	return ip6_frag_intern(hash, fq);
383 
384 oom:
385 	IP6_INC_STATS_BH(IPSTATS_MIB_REASMFAILS);
386 	return NULL;
387 }
388 
389 static __inline__ struct frag_queue *
390 fq_find(u32 id, struct in6_addr *src, struct in6_addr *dst)
391 {
392 	struct frag_queue *fq;
393 	struct hlist_node *n;
394 	unsigned int hash = ip6qhashfn(id, src, dst);
395 
396 	read_lock(&ip6_frag_lock);
397 	hlist_for_each_entry(fq, n, &ip6_frag_hash[hash], list) {
398 		if (fq->id == id &&
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);
403 			return fq;
404 		}
405 	}
406 	read_unlock(&ip6_frag_lock);
407 
408 	return ip6_frag_create(hash, id, src, dst);
409 }
410 
411 
412 static void ip6_frag_queue(struct frag_queue *fq, struct sk_buff *skb,
413 			   struct frag_hdr *fhdr, int nhoff)
414 {
415 	struct sk_buff *prev, *next;
416 	int offset, end;
417 
418 	if (fq->last_in & COMPLETE)
419 		goto err;
420 
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)));
424 
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);
428  		return;
429 	}
430 
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));
434 
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.
439 		 */
440 		if (end < fq->len ||
441 		    ((fq->last_in & LAST_IN) && end != fq->len))
442 			goto err;
443 		fq->last_in |= LAST_IN;
444 		fq->len = end;
445 	} else {
446 		/* Check if the fragment is rounded to 8 bytes.
447 		 * Required by the RFC.
448 		 */
449 		if (end & 0x7) {
450 			/* RFC2460 says always send parameter problem in
451 			 * this case. -DaveM
452 			 */
453 			IP6_INC_STATS_BH(IPSTATS_MIB_INHDRERRORS);
454 			icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
455 					  offsetof(struct ipv6hdr, payload_len));
456 			return;
457 		}
458 		if (end > fq->len) {
459 			/* Some bits beyond end -> corruption. */
460 			if (fq->last_in & LAST_IN)
461 				goto err;
462 			fq->len = end;
463 		}
464 	}
465 
466 	if (end == offset)
467 		goto err;
468 
469 	/* Point into the IP datagram 'data' part. */
470 	if (!pskb_pull(skb, (u8 *) (fhdr + 1) - skb->data))
471 		goto err;
472 
473 	if (pskb_trim_rcsum(skb, end - offset))
474 		goto err;
475 
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?
479 	 */
480 	prev = NULL;
481 	for(next = fq->fragments; next != NULL; next = next->next) {
482 		if (FRAG6_CB(next)->offset >= offset)
483 			break;	/* bingo! */
484 		prev = next;
485 	}
486 
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.
490 	 */
491 	if (prev) {
492 		int i = (FRAG6_CB(prev)->offset + prev->len) - offset;
493 
494 		if (i > 0) {
495 			offset += i;
496 			if (end <= offset)
497 				goto err;
498 			if (!pskb_pull(skb, i))
499 				goto err;
500 			if (skb->ip_summed != CHECKSUM_UNNECESSARY)
501 				skb->ip_summed = CHECKSUM_NONE;
502 		}
503 	}
504 
505 	/* Look for overlap with succeeding segments.
506 	 * If we can merge fragments, do it.
507 	 */
508 	while (next && FRAG6_CB(next)->offset < end) {
509 		int i = end - FRAG6_CB(next)->offset; /* overlap is 'i' bytes */
510 
511 		if (i < next->len) {
512 			/* Eat head of the next overlapped fragment
513 			 * and leave the loop. The next ones cannot overlap.
514 			 */
515 			if (!pskb_pull(next, i))
516 				goto err;
517 			FRAG6_CB(next)->offset += i;	/* next fragment */
518 			fq->meat -= i;
519 			if (next->ip_summed != CHECKSUM_UNNECESSARY)
520 				next->ip_summed = CHECKSUM_NONE;
521 			break;
522 		} else {
523 			struct sk_buff *free_it = next;
524 
525 			/* Old fragment is completely overridden with
526 			 * new one drop it.
527 			 */
528 			next = next->next;
529 
530 			if (prev)
531 				prev->next = next;
532 			else
533 				fq->fragments = next;
534 
535 			fq->meat -= free_it->len;
536 			frag_kfree_skb(free_it, NULL);
537 		}
538 	}
539 
540 	FRAG6_CB(skb)->offset = offset;
541 
542 	/* Insert this fragment in the chain of fragments. */
543 	skb->next = next;
544 	if (prev)
545 		prev->next = skb;
546 	else
547 		fq->fragments = skb;
548 
549 	if (skb->dev)
550 		fq->iif = skb->dev->ifindex;
551 	skb->dev = NULL;
552 	skb_get_timestamp(skb, &fq->stamp);
553 	fq->meat += skb->len;
554 	atomic_add(skb->truesize, &ip6_frag_mem);
555 
556 	/* The first fragment.
557 	 * nhoffset is obtained from the first fragment, of course.
558 	 */
559 	if (offset == 0) {
560 		fq->nhoffset = nhoff;
561 		fq->last_in |= FIRST_IN;
562 	}
563 	write_lock(&ip6_frag_lock);
564 	list_move_tail(&fq->lru_list, &ip6_frag_lru_list);
565 	write_unlock(&ip6_frag_lock);
566 	return;
567 
568 err:
569 	IP6_INC_STATS(IPSTATS_MIB_REASMFAILS);
570 	kfree_skb(skb);
571 }
572 
573 /*
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.
577  *
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.
581  */
582 static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff **skb_in,
583 			  struct net_device *dev)
584 {
585 	struct sk_buff *fp, *head = fq->fragments;
586 	int    payload_len;
587 	unsigned int nhoff;
588 
589 	fq_kill(fq);
590 
591 	BUG_TRAP(head != NULL);
592 	BUG_TRAP(FRAG6_CB(head)->offset == 0);
593 
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)
597 		goto out_oversize;
598 
599 	/* Head of list must not be cloned. */
600 	if (skb_cloned(head) && pskb_expand_head(head, 0, 0, GFP_ATOMIC))
601 		goto out_oom;
602 
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;
608 		int i, plen = 0;
609 
610 		if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL)
611 			goto out_oom;
612 		clone->next = head->next;
613 		head->next = clone;
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;
621 		clone->csum = 0;
622 		clone->ip_summed = head->ip_summed;
623 		atomic_add(clone->truesize, &ip6_frag_mem);
624 	}
625 
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);
634 
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);
639 
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);
649 	}
650 
651 	head->next = NULL;
652 	head->dev = dev;
653 	skb_set_timestamp(head, &fq->stamp);
654 	head->nh.ipv6h->payload_len = htons(payload_len);
655 	IP6CB(head)->nhoff = nhoff;
656 
657 	*skb_in = head;
658 
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);
662 
663 	IP6_INC_STATS_BH(IPSTATS_MIB_REASMOKS);
664 	fq->fragments = NULL;
665 	return 1;
666 
667 out_oversize:
668 	if (net_ratelimit())
669 		printk(KERN_DEBUG "ip6_frag_reasm: payload len = %d\n", payload_len);
670 	goto out_fail;
671 out_oom:
672 	if (net_ratelimit())
673 		printk(KERN_DEBUG "ip6_frag_reasm: no memory for reassembly\n");
674 out_fail:
675 	IP6_INC_STATS_BH(IPSTATS_MIB_REASMFAILS);
676 	return -1;
677 }
678 
679 static int ipv6_frag_rcv(struct sk_buff **skbp)
680 {
681 	struct sk_buff *skb = *skbp;
682 	struct net_device *dev = skb->dev;
683 	struct frag_hdr *fhdr;
684 	struct frag_queue *fq;
685 	struct ipv6hdr *hdr;
686 
687 	hdr = skb->nh.ipv6h;
688 
689 	IP6_INC_STATS_BH(IPSTATS_MIB_REASMREQDS);
690 
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);
695 		return -1;
696 	}
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);
700 		return -1;
701 	}
702 
703 	hdr = skb->nh.ipv6h;
704 	fhdr = (struct frag_hdr *)skb->h.raw;
705 
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);
710 
711 		IP6CB(skb)->nhoff = (u8*)fhdr - skb->nh.raw;
712 		return 1;
713 	}
714 
715 	if (atomic_read(&ip6_frag_mem) > sysctl_ip6frag_high_thresh)
716 		ip6_evictor();
717 
718 	if ((fq = fq_find(fhdr->identification, &hdr->saddr, &hdr->daddr)) != NULL) {
719 		int ret = -1;
720 
721 		spin_lock(&fq->lock);
722 
723 		ip6_frag_queue(fq, skb, fhdr, IP6CB(skb)->nhoff);
724 
725 		if (fq->last_in == (FIRST_IN|LAST_IN) &&
726 		    fq->meat == fq->len)
727 			ret = ip6_frag_reasm(fq, skbp, dev);
728 
729 		spin_unlock(&fq->lock);
730 		fq_put(fq, NULL);
731 		return ret;
732 	}
733 
734 	IP6_INC_STATS_BH(IPSTATS_MIB_REASMFAILS);
735 	kfree_skb(skb);
736 	return -1;
737 }
738 
739 static struct inet6_protocol frag_protocol =
740 {
741 	.handler	=	ipv6_frag_rcv,
742 	.flags		=	INET6_PROTO_NOPOLICY,
743 };
744 
745 void __init ipv6_frag_init(void)
746 {
747 	if (inet6_add_protocol(&frag_protocol, IPPROTO_FRAGMENT) < 0)
748 		printk(KERN_ERR "ipv6_frag_init: Could not register protocol\n");
749 
750 	ip6_frag_hash_rnd = (u32) ((num_physpages ^ (num_physpages>>7)) ^
751 				   (jiffies ^ (jiffies >> 6)));
752 
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);
757 }
758