xref: /linux/net/ipv6/reassembly.c (revision 273b281fa22c293963ee3e6eec418f5dda2dbc83)
1 /*
2  *	IPv6 fragment reassembly
3  *	Linux INET6 implementation
4  *
5  *	Authors:
6  *	Pedro Roque		<roque@di.fc.ul.pt>
7  *
8  *	Based on: net/ipv4/ip_fragment.c
9  *
10  *	This program is free software; you can redistribute it and/or
11  *      modify it under the terms of the GNU General Public License
12  *      as published by the Free Software Foundation; either version
13  *      2 of the License, or (at your option) any later version.
14  */
15 
16 /*
17  *	Fixes:
18  *	Andi Kleen	Make it work with multiple hosts.
19  *			More RFC compliance.
20  *
21  *      Horst von Brand Add missing #include <linux/string.h>
22  *	Alexey Kuznetsov	SMP races, threading, cleanup.
23  *	Patrick McHardy		LRU queue of frag heads for evictor.
24  *	Mitsuru KANDA @USAGI	Register inet6_protocol{}.
25  *	David Stevens and
26  *	YOSHIFUJI,H. @USAGI	Always remove fragment header to
27  *				calculate ICV correctly.
28  */
29 #include <linux/errno.h>
30 #include <linux/types.h>
31 #include <linux/string.h>
32 #include <linux/socket.h>
33 #include <linux/sockios.h>
34 #include <linux/jiffies.h>
35 #include <linux/net.h>
36 #include <linux/list.h>
37 #include <linux/netdevice.h>
38 #include <linux/in6.h>
39 #include <linux/ipv6.h>
40 #include <linux/icmpv6.h>
41 #include <linux/random.h>
42 #include <linux/jhash.h>
43 #include <linux/skbuff.h>
44 
45 #include <net/sock.h>
46 #include <net/snmp.h>
47 
48 #include <net/ipv6.h>
49 #include <net/ip6_route.h>
50 #include <net/protocol.h>
51 #include <net/transp_v6.h>
52 #include <net/rawv6.h>
53 #include <net/ndisc.h>
54 #include <net/addrconf.h>
55 #include <net/inet_frag.h>
56 
57 struct ip6frag_skb_cb
58 {
59 	struct inet6_skb_parm	h;
60 	int			offset;
61 };
62 
63 #define FRAG6_CB(skb)	((struct ip6frag_skb_cb*)((skb)->cb))
64 
65 
66 /*
67  *	Equivalent of ipv4 struct ipq
68  */
69 
70 struct frag_queue
71 {
72 	struct inet_frag_queue	q;
73 
74 	__be32			id;		/* fragment id		*/
75 	struct in6_addr		saddr;
76 	struct in6_addr		daddr;
77 
78 	int			iif;
79 	unsigned int		csum;
80 	__u16			nhoffset;
81 };
82 
83 static struct inet_frags ip6_frags;
84 
85 int ip6_frag_nqueues(struct net *net)
86 {
87 	return net->ipv6.frags.nqueues;
88 }
89 
90 int ip6_frag_mem(struct net *net)
91 {
92 	return atomic_read(&net->ipv6.frags.mem);
93 }
94 
95 static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *prev,
96 			  struct net_device *dev);
97 
98 /*
99  * callers should be careful not to use the hash value outside the ipfrag_lock
100  * as doing so could race with ipfrag_hash_rnd being recalculated.
101  */
102 unsigned int inet6_hash_frag(__be32 id, const struct in6_addr *saddr,
103 			     const struct in6_addr *daddr, u32 rnd)
104 {
105 	u32 a, b, c;
106 
107 	a = (__force u32)saddr->s6_addr32[0];
108 	b = (__force u32)saddr->s6_addr32[1];
109 	c = (__force u32)saddr->s6_addr32[2];
110 
111 	a += JHASH_GOLDEN_RATIO;
112 	b += JHASH_GOLDEN_RATIO;
113 	c += rnd;
114 	__jhash_mix(a, b, c);
115 
116 	a += (__force u32)saddr->s6_addr32[3];
117 	b += (__force u32)daddr->s6_addr32[0];
118 	c += (__force u32)daddr->s6_addr32[1];
119 	__jhash_mix(a, b, c);
120 
121 	a += (__force u32)daddr->s6_addr32[2];
122 	b += (__force u32)daddr->s6_addr32[3];
123 	c += (__force u32)id;
124 	__jhash_mix(a, b, c);
125 
126 	return c & (INETFRAGS_HASHSZ - 1);
127 }
128 EXPORT_SYMBOL_GPL(inet6_hash_frag);
129 
130 static unsigned int ip6_hashfn(struct inet_frag_queue *q)
131 {
132 	struct frag_queue *fq;
133 
134 	fq = container_of(q, struct frag_queue, q);
135 	return inet6_hash_frag(fq->id, &fq->saddr, &fq->daddr, ip6_frags.rnd);
136 }
137 
138 int ip6_frag_match(struct inet_frag_queue *q, void *a)
139 {
140 	struct frag_queue *fq;
141 	struct ip6_create_arg *arg = a;
142 
143 	fq = container_of(q, struct frag_queue, q);
144 	return (fq->id == arg->id &&
145 			ipv6_addr_equal(&fq->saddr, arg->src) &&
146 			ipv6_addr_equal(&fq->daddr, arg->dst));
147 }
148 EXPORT_SYMBOL(ip6_frag_match);
149 
150 /* Memory Tracking Functions. */
151 static inline void frag_kfree_skb(struct netns_frags *nf,
152 		struct sk_buff *skb, int *work)
153 {
154 	if (work)
155 		*work -= skb->truesize;
156 	atomic_sub(skb->truesize, &nf->mem);
157 	kfree_skb(skb);
158 }
159 
160 void ip6_frag_init(struct inet_frag_queue *q, void *a)
161 {
162 	struct frag_queue *fq = container_of(q, struct frag_queue, q);
163 	struct ip6_create_arg *arg = a;
164 
165 	fq->id = arg->id;
166 	ipv6_addr_copy(&fq->saddr, arg->src);
167 	ipv6_addr_copy(&fq->daddr, arg->dst);
168 }
169 EXPORT_SYMBOL(ip6_frag_init);
170 
171 /* Destruction primitives. */
172 
173 static __inline__ void fq_put(struct frag_queue *fq)
174 {
175 	inet_frag_put(&fq->q, &ip6_frags);
176 }
177 
178 /* Kill fq entry. It is not destroyed immediately,
179  * because caller (and someone more) holds reference count.
180  */
181 static __inline__ void fq_kill(struct frag_queue *fq)
182 {
183 	inet_frag_kill(&fq->q, &ip6_frags);
184 }
185 
186 static void ip6_evictor(struct net *net, struct inet6_dev *idev)
187 {
188 	int evicted;
189 
190 	evicted = inet_frag_evictor(&net->ipv6.frags, &ip6_frags);
191 	if (evicted)
192 		IP6_ADD_STATS_BH(net, idev, IPSTATS_MIB_REASMFAILS, evicted);
193 }
194 
195 static void ip6_frag_expire(unsigned long data)
196 {
197 	struct frag_queue *fq;
198 	struct net_device *dev = NULL;
199 	struct net *net;
200 
201 	fq = container_of((struct inet_frag_queue *)data, struct frag_queue, q);
202 
203 	spin_lock(&fq->q.lock);
204 
205 	if (fq->q.last_in & INET_FRAG_COMPLETE)
206 		goto out;
207 
208 	fq_kill(fq);
209 
210 	net = container_of(fq->q.net, struct net, ipv6.frags);
211 	rcu_read_lock();
212 	dev = dev_get_by_index_rcu(net, fq->iif);
213 	if (!dev)
214 		goto out_rcu_unlock;
215 
216 	IP6_INC_STATS_BH(net, __in6_dev_get(dev), IPSTATS_MIB_REASMTIMEOUT);
217 	IP6_INC_STATS_BH(net, __in6_dev_get(dev), IPSTATS_MIB_REASMFAILS);
218 
219 	/* Don't send error if the first segment did not arrive. */
220 	if (!(fq->q.last_in & INET_FRAG_FIRST_IN) || !fq->q.fragments)
221 		goto out_rcu_unlock;
222 
223 	/*
224 	   But use as source device on which LAST ARRIVED
225 	   segment was received. And do not use fq->dev
226 	   pointer directly, device might already disappeared.
227 	 */
228 	fq->q.fragments->dev = dev;
229 	icmpv6_send(fq->q.fragments, ICMPV6_TIME_EXCEED, ICMPV6_EXC_FRAGTIME, 0, dev);
230 out_rcu_unlock:
231 	rcu_read_unlock();
232 out:
233 	spin_unlock(&fq->q.lock);
234 	fq_put(fq);
235 }
236 
237 static __inline__ struct frag_queue *
238 fq_find(struct net *net, __be32 id, struct in6_addr *src, struct in6_addr *dst,
239 	struct inet6_dev *idev)
240 {
241 	struct inet_frag_queue *q;
242 	struct ip6_create_arg arg;
243 	unsigned int hash;
244 
245 	arg.id = id;
246 	arg.src = src;
247 	arg.dst = dst;
248 
249 	read_lock(&ip6_frags.lock);
250 	hash = inet6_hash_frag(id, src, dst, ip6_frags.rnd);
251 
252 	q = inet_frag_find(&net->ipv6.frags, &ip6_frags, &arg, hash);
253 	if (q == NULL)
254 		goto oom;
255 
256 	return container_of(q, struct frag_queue, q);
257 
258 oom:
259 	IP6_INC_STATS_BH(net, idev, IPSTATS_MIB_REASMFAILS);
260 	return NULL;
261 }
262 
263 static int ip6_frag_queue(struct frag_queue *fq, struct sk_buff *skb,
264 			   struct frag_hdr *fhdr, int nhoff)
265 {
266 	struct sk_buff *prev, *next;
267 	struct net_device *dev;
268 	int offset, end;
269 	struct net *net = dev_net(skb_dst(skb)->dev);
270 
271 	if (fq->q.last_in & INET_FRAG_COMPLETE)
272 		goto err;
273 
274 	offset = ntohs(fhdr->frag_off) & ~0x7;
275 	end = offset + (ntohs(ipv6_hdr(skb)->payload_len) -
276 			((u8 *)(fhdr + 1) - (u8 *)(ipv6_hdr(skb) + 1)));
277 
278 	if ((unsigned int)end > IPV6_MAXPLEN) {
279 		IP6_INC_STATS_BH(net, ip6_dst_idev(skb_dst(skb)),
280 				 IPSTATS_MIB_INHDRERRORS);
281 		icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
282 				  ((u8 *)&fhdr->frag_off -
283 				   skb_network_header(skb)));
284 		return -1;
285 	}
286 
287 	if (skb->ip_summed == CHECKSUM_COMPLETE) {
288 		const unsigned char *nh = skb_network_header(skb);
289 		skb->csum = csum_sub(skb->csum,
290 				     csum_partial(nh, (u8 *)(fhdr + 1) - nh,
291 						  0));
292 	}
293 
294 	/* Is this the final fragment? */
295 	if (!(fhdr->frag_off & htons(IP6_MF))) {
296 		/* If we already have some bits beyond end
297 		 * or have different end, the segment is corrupted.
298 		 */
299 		if (end < fq->q.len ||
300 		    ((fq->q.last_in & INET_FRAG_LAST_IN) && end != fq->q.len))
301 			goto err;
302 		fq->q.last_in |= INET_FRAG_LAST_IN;
303 		fq->q.len = end;
304 	} else {
305 		/* Check if the fragment is rounded to 8 bytes.
306 		 * Required by the RFC.
307 		 */
308 		if (end & 0x7) {
309 			/* RFC2460 says always send parameter problem in
310 			 * this case. -DaveM
311 			 */
312 			IP6_INC_STATS_BH(net, ip6_dst_idev(skb_dst(skb)),
313 					 IPSTATS_MIB_INHDRERRORS);
314 			icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
315 					  offsetof(struct ipv6hdr, payload_len));
316 			return -1;
317 		}
318 		if (end > fq->q.len) {
319 			/* Some bits beyond end -> corruption. */
320 			if (fq->q.last_in & INET_FRAG_LAST_IN)
321 				goto err;
322 			fq->q.len = end;
323 		}
324 	}
325 
326 	if (end == offset)
327 		goto err;
328 
329 	/* Point into the IP datagram 'data' part. */
330 	if (!pskb_pull(skb, (u8 *) (fhdr + 1) - skb->data))
331 		goto err;
332 
333 	if (pskb_trim_rcsum(skb, end - offset))
334 		goto err;
335 
336 	/* Find out which fragments are in front and at the back of us
337 	 * in the chain of fragments so far.  We must know where to put
338 	 * this fragment, right?
339 	 */
340 	prev = NULL;
341 	for(next = fq->q.fragments; next != NULL; next = next->next) {
342 		if (FRAG6_CB(next)->offset >= offset)
343 			break;	/* bingo! */
344 		prev = next;
345 	}
346 
347 	/* We found where to put this one.  Check for overlap with
348 	 * preceding fragment, and, if needed, align things so that
349 	 * any overlaps are eliminated.
350 	 */
351 	if (prev) {
352 		int i = (FRAG6_CB(prev)->offset + prev->len) - offset;
353 
354 		if (i > 0) {
355 			offset += i;
356 			if (end <= offset)
357 				goto err;
358 			if (!pskb_pull(skb, i))
359 				goto err;
360 			if (skb->ip_summed != CHECKSUM_UNNECESSARY)
361 				skb->ip_summed = CHECKSUM_NONE;
362 		}
363 	}
364 
365 	/* Look for overlap with succeeding segments.
366 	 * If we can merge fragments, do it.
367 	 */
368 	while (next && FRAG6_CB(next)->offset < end) {
369 		int i = end - FRAG6_CB(next)->offset; /* overlap is 'i' bytes */
370 
371 		if (i < next->len) {
372 			/* Eat head of the next overlapped fragment
373 			 * and leave the loop. The next ones cannot overlap.
374 			 */
375 			if (!pskb_pull(next, i))
376 				goto err;
377 			FRAG6_CB(next)->offset += i;	/* next fragment */
378 			fq->q.meat -= i;
379 			if (next->ip_summed != CHECKSUM_UNNECESSARY)
380 				next->ip_summed = CHECKSUM_NONE;
381 			break;
382 		} else {
383 			struct sk_buff *free_it = next;
384 
385 			/* Old fragment is completely overridden with
386 			 * new one drop it.
387 			 */
388 			next = next->next;
389 
390 			if (prev)
391 				prev->next = next;
392 			else
393 				fq->q.fragments = next;
394 
395 			fq->q.meat -= free_it->len;
396 			frag_kfree_skb(fq->q.net, free_it, NULL);
397 		}
398 	}
399 
400 	FRAG6_CB(skb)->offset = offset;
401 
402 	/* Insert this fragment in the chain of fragments. */
403 	skb->next = next;
404 	if (prev)
405 		prev->next = skb;
406 	else
407 		fq->q.fragments = skb;
408 
409 	dev = skb->dev;
410 	if (dev) {
411 		fq->iif = dev->ifindex;
412 		skb->dev = NULL;
413 	}
414 	fq->q.stamp = skb->tstamp;
415 	fq->q.meat += skb->len;
416 	atomic_add(skb->truesize, &fq->q.net->mem);
417 
418 	/* The first fragment.
419 	 * nhoffset is obtained from the first fragment, of course.
420 	 */
421 	if (offset == 0) {
422 		fq->nhoffset = nhoff;
423 		fq->q.last_in |= INET_FRAG_FIRST_IN;
424 	}
425 
426 	if (fq->q.last_in == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
427 	    fq->q.meat == fq->q.len)
428 		return ip6_frag_reasm(fq, prev, dev);
429 
430 	write_lock(&ip6_frags.lock);
431 	list_move_tail(&fq->q.lru_list, &fq->q.net->lru_list);
432 	write_unlock(&ip6_frags.lock);
433 	return -1;
434 
435 err:
436 	IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
437 		      IPSTATS_MIB_REASMFAILS);
438 	kfree_skb(skb);
439 	return -1;
440 }
441 
442 /*
443  *	Check if this packet is complete.
444  *	Returns NULL on failure by any reason, and pointer
445  *	to current nexthdr field in reassembled frame.
446  *
447  *	It is called with locked fq, and caller must check that
448  *	queue is eligible for reassembly i.e. it is not COMPLETE,
449  *	the last and the first frames arrived and all the bits are here.
450  */
451 static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *prev,
452 			  struct net_device *dev)
453 {
454 	struct net *net = container_of(fq->q.net, struct net, ipv6.frags);
455 	struct sk_buff *fp, *head = fq->q.fragments;
456 	int    payload_len;
457 	unsigned int nhoff;
458 
459 	fq_kill(fq);
460 
461 	/* Make the one we just received the head. */
462 	if (prev) {
463 		head = prev->next;
464 		fp = skb_clone(head, GFP_ATOMIC);
465 
466 		if (!fp)
467 			goto out_oom;
468 
469 		fp->next = head->next;
470 		prev->next = fp;
471 
472 		skb_morph(head, fq->q.fragments);
473 		head->next = fq->q.fragments->next;
474 
475 		kfree_skb(fq->q.fragments);
476 		fq->q.fragments = head;
477 	}
478 
479 	WARN_ON(head == NULL);
480 	WARN_ON(FRAG6_CB(head)->offset != 0);
481 
482 	/* Unfragmented part is taken from the first segment. */
483 	payload_len = ((head->data - skb_network_header(head)) -
484 		       sizeof(struct ipv6hdr) + fq->q.len -
485 		       sizeof(struct frag_hdr));
486 	if (payload_len > IPV6_MAXPLEN)
487 		goto out_oversize;
488 
489 	/* Head of list must not be cloned. */
490 	if (skb_cloned(head) && pskb_expand_head(head, 0, 0, GFP_ATOMIC))
491 		goto out_oom;
492 
493 	/* If the first fragment is fragmented itself, we split
494 	 * it to two chunks: the first with data and paged part
495 	 * and the second, holding only fragments. */
496 	if (skb_has_frags(head)) {
497 		struct sk_buff *clone;
498 		int i, plen = 0;
499 
500 		if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL)
501 			goto out_oom;
502 		clone->next = head->next;
503 		head->next = clone;
504 		skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
505 		skb_frag_list_init(head);
506 		for (i=0; i<skb_shinfo(head)->nr_frags; i++)
507 			plen += skb_shinfo(head)->frags[i].size;
508 		clone->len = clone->data_len = head->data_len - plen;
509 		head->data_len -= clone->len;
510 		head->len -= clone->len;
511 		clone->csum = 0;
512 		clone->ip_summed = head->ip_summed;
513 		atomic_add(clone->truesize, &fq->q.net->mem);
514 	}
515 
516 	/* We have to remove fragment header from datagram and to relocate
517 	 * header in order to calculate ICV correctly. */
518 	nhoff = fq->nhoffset;
519 	skb_network_header(head)[nhoff] = skb_transport_header(head)[0];
520 	memmove(head->head + sizeof(struct frag_hdr), head->head,
521 		(head->data - head->head) - sizeof(struct frag_hdr));
522 	head->mac_header += sizeof(struct frag_hdr);
523 	head->network_header += sizeof(struct frag_hdr);
524 
525 	skb_shinfo(head)->frag_list = head->next;
526 	skb_reset_transport_header(head);
527 	skb_push(head, head->data - skb_network_header(head));
528 	atomic_sub(head->truesize, &fq->q.net->mem);
529 
530 	for (fp=head->next; fp; fp = fp->next) {
531 		head->data_len += fp->len;
532 		head->len += fp->len;
533 		if (head->ip_summed != fp->ip_summed)
534 			head->ip_summed = CHECKSUM_NONE;
535 		else if (head->ip_summed == CHECKSUM_COMPLETE)
536 			head->csum = csum_add(head->csum, fp->csum);
537 		head->truesize += fp->truesize;
538 		atomic_sub(fp->truesize, &fq->q.net->mem);
539 	}
540 
541 	head->next = NULL;
542 	head->dev = dev;
543 	head->tstamp = fq->q.stamp;
544 	ipv6_hdr(head)->payload_len = htons(payload_len);
545 	IP6CB(head)->nhoff = nhoff;
546 
547 	/* Yes, and fold redundant checksum back. 8) */
548 	if (head->ip_summed == CHECKSUM_COMPLETE)
549 		head->csum = csum_partial(skb_network_header(head),
550 					  skb_network_header_len(head),
551 					  head->csum);
552 
553 	rcu_read_lock();
554 	IP6_INC_STATS_BH(net, __in6_dev_get(dev), IPSTATS_MIB_REASMOKS);
555 	rcu_read_unlock();
556 	fq->q.fragments = NULL;
557 	return 1;
558 
559 out_oversize:
560 	if (net_ratelimit())
561 		printk(KERN_DEBUG "ip6_frag_reasm: payload len = %d\n", payload_len);
562 	goto out_fail;
563 out_oom:
564 	if (net_ratelimit())
565 		printk(KERN_DEBUG "ip6_frag_reasm: no memory for reassembly\n");
566 out_fail:
567 	rcu_read_lock();
568 	IP6_INC_STATS_BH(net, __in6_dev_get(dev), IPSTATS_MIB_REASMFAILS);
569 	rcu_read_unlock();
570 	return -1;
571 }
572 
573 static int ipv6_frag_rcv(struct sk_buff *skb)
574 {
575 	struct frag_hdr *fhdr;
576 	struct frag_queue *fq;
577 	struct ipv6hdr *hdr = ipv6_hdr(skb);
578 	struct net *net = dev_net(skb_dst(skb)->dev);
579 
580 	IP6_INC_STATS_BH(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMREQDS);
581 
582 	/* Jumbo payload inhibits frag. header */
583 	if (hdr->payload_len==0)
584 		goto fail_hdr;
585 
586 	if (!pskb_may_pull(skb, (skb_transport_offset(skb) +
587 				 sizeof(struct frag_hdr))))
588 		goto fail_hdr;
589 
590 	hdr = ipv6_hdr(skb);
591 	fhdr = (struct frag_hdr *)skb_transport_header(skb);
592 
593 	if (!(fhdr->frag_off & htons(0xFFF9))) {
594 		/* It is not a fragmented frame */
595 		skb->transport_header += sizeof(struct frag_hdr);
596 		IP6_INC_STATS_BH(net,
597 				 ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMOKS);
598 
599 		IP6CB(skb)->nhoff = (u8 *)fhdr - skb_network_header(skb);
600 		return 1;
601 	}
602 
603 	if (atomic_read(&net->ipv6.frags.mem) > net->ipv6.frags.high_thresh)
604 		ip6_evictor(net, ip6_dst_idev(skb_dst(skb)));
605 
606 	if ((fq = fq_find(net, fhdr->identification, &hdr->saddr, &hdr->daddr,
607 			  ip6_dst_idev(skb_dst(skb)))) != NULL) {
608 		int ret;
609 
610 		spin_lock(&fq->q.lock);
611 
612 		ret = ip6_frag_queue(fq, skb, fhdr, IP6CB(skb)->nhoff);
613 
614 		spin_unlock(&fq->q.lock);
615 		fq_put(fq);
616 		return ret;
617 	}
618 
619 	IP6_INC_STATS_BH(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMFAILS);
620 	kfree_skb(skb);
621 	return -1;
622 
623 fail_hdr:
624 	IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_INHDRERRORS);
625 	icmpv6_param_prob(skb, ICMPV6_HDR_FIELD, skb_network_header_len(skb));
626 	return -1;
627 }
628 
629 static const struct inet6_protocol frag_protocol =
630 {
631 	.handler	=	ipv6_frag_rcv,
632 	.flags		=	INET6_PROTO_NOPOLICY,
633 };
634 
635 #ifdef CONFIG_SYSCTL
636 static struct ctl_table ip6_frags_ns_ctl_table[] = {
637 	{
638 		.procname	= "ip6frag_high_thresh",
639 		.data		= &init_net.ipv6.frags.high_thresh,
640 		.maxlen		= sizeof(int),
641 		.mode		= 0644,
642 		.proc_handler	= proc_dointvec
643 	},
644 	{
645 		.procname	= "ip6frag_low_thresh",
646 		.data		= &init_net.ipv6.frags.low_thresh,
647 		.maxlen		= sizeof(int),
648 		.mode		= 0644,
649 		.proc_handler	= proc_dointvec
650 	},
651 	{
652 		.procname	= "ip6frag_time",
653 		.data		= &init_net.ipv6.frags.timeout,
654 		.maxlen		= sizeof(int),
655 		.mode		= 0644,
656 		.proc_handler	= proc_dointvec_jiffies,
657 	},
658 	{ }
659 };
660 
661 static struct ctl_table ip6_frags_ctl_table[] = {
662 	{
663 		.procname	= "ip6frag_secret_interval",
664 		.data		= &ip6_frags.secret_interval,
665 		.maxlen		= sizeof(int),
666 		.mode		= 0644,
667 		.proc_handler	= proc_dointvec_jiffies,
668 	},
669 	{ }
670 };
671 
672 static int ip6_frags_ns_sysctl_register(struct net *net)
673 {
674 	struct ctl_table *table;
675 	struct ctl_table_header *hdr;
676 
677 	table = ip6_frags_ns_ctl_table;
678 	if (!net_eq(net, &init_net)) {
679 		table = kmemdup(table, sizeof(ip6_frags_ns_ctl_table), GFP_KERNEL);
680 		if (table == NULL)
681 			goto err_alloc;
682 
683 		table[0].data = &net->ipv6.frags.high_thresh;
684 		table[1].data = &net->ipv6.frags.low_thresh;
685 		table[2].data = &net->ipv6.frags.timeout;
686 	}
687 
688 	hdr = register_net_sysctl_table(net, net_ipv6_ctl_path, table);
689 	if (hdr == NULL)
690 		goto err_reg;
691 
692 	net->ipv6.sysctl.frags_hdr = hdr;
693 	return 0;
694 
695 err_reg:
696 	if (!net_eq(net, &init_net))
697 		kfree(table);
698 err_alloc:
699 	return -ENOMEM;
700 }
701 
702 static void ip6_frags_ns_sysctl_unregister(struct net *net)
703 {
704 	struct ctl_table *table;
705 
706 	table = net->ipv6.sysctl.frags_hdr->ctl_table_arg;
707 	unregister_net_sysctl_table(net->ipv6.sysctl.frags_hdr);
708 	kfree(table);
709 }
710 
711 static struct ctl_table_header *ip6_ctl_header;
712 
713 static int ip6_frags_sysctl_register(void)
714 {
715 	ip6_ctl_header = register_net_sysctl_rotable(net_ipv6_ctl_path,
716 			ip6_frags_ctl_table);
717 	return ip6_ctl_header == NULL ? -ENOMEM : 0;
718 }
719 
720 static void ip6_frags_sysctl_unregister(void)
721 {
722 	unregister_net_sysctl_table(ip6_ctl_header);
723 }
724 #else
725 static inline int ip6_frags_ns_sysctl_register(struct net *net)
726 {
727 	return 0;
728 }
729 
730 static inline void ip6_frags_ns_sysctl_unregister(struct net *net)
731 {
732 }
733 
734 static inline int ip6_frags_sysctl_register(void)
735 {
736 	return 0;
737 }
738 
739 static inline void ip6_frags_sysctl_unregister(void)
740 {
741 }
742 #endif
743 
744 static int ipv6_frags_init_net(struct net *net)
745 {
746 	net->ipv6.frags.high_thresh = 256 * 1024;
747 	net->ipv6.frags.low_thresh = 192 * 1024;
748 	net->ipv6.frags.timeout = IPV6_FRAG_TIMEOUT;
749 
750 	inet_frags_init_net(&net->ipv6.frags);
751 
752 	return ip6_frags_ns_sysctl_register(net);
753 }
754 
755 static void ipv6_frags_exit_net(struct net *net)
756 {
757 	ip6_frags_ns_sysctl_unregister(net);
758 	inet_frags_exit_net(&net->ipv6.frags, &ip6_frags);
759 }
760 
761 static struct pernet_operations ip6_frags_ops = {
762 	.init = ipv6_frags_init_net,
763 	.exit = ipv6_frags_exit_net,
764 };
765 
766 int __init ipv6_frag_init(void)
767 {
768 	int ret;
769 
770 	ret = inet6_add_protocol(&frag_protocol, IPPROTO_FRAGMENT);
771 	if (ret)
772 		goto out;
773 
774 	ret = ip6_frags_sysctl_register();
775 	if (ret)
776 		goto err_sysctl;
777 
778 	ret = register_pernet_subsys(&ip6_frags_ops);
779 	if (ret)
780 		goto err_pernet;
781 
782 	ip6_frags.hashfn = ip6_hashfn;
783 	ip6_frags.constructor = ip6_frag_init;
784 	ip6_frags.destructor = NULL;
785 	ip6_frags.skb_free = NULL;
786 	ip6_frags.qsize = sizeof(struct frag_queue);
787 	ip6_frags.match = ip6_frag_match;
788 	ip6_frags.frag_expire = ip6_frag_expire;
789 	ip6_frags.secret_interval = 10 * 60 * HZ;
790 	inet_frags_init(&ip6_frags);
791 out:
792 	return ret;
793 
794 err_pernet:
795 	ip6_frags_sysctl_unregister();
796 err_sysctl:
797 	inet6_del_protocol(&frag_protocol, IPPROTO_FRAGMENT);
798 	goto out;
799 }
800 
801 void ipv6_frag_exit(void)
802 {
803 	inet_frags_fini(&ip6_frags);
804 	ip6_frags_sysctl_unregister();
805 	unregister_pernet_subsys(&ip6_frags_ops);
806 	inet6_del_protocol(&frag_protocol, IPPROTO_FRAGMENT);
807 }
808