xref: /openbmc/linux/net/ipv6/reassembly.c (revision 615c36f5)
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 #include <linux/slab.h>
45 #include <linux/export.h>
46 
47 #include <net/sock.h>
48 #include <net/snmp.h>
49 
50 #include <net/ipv6.h>
51 #include <net/ip6_route.h>
52 #include <net/protocol.h>
53 #include <net/transp_v6.h>
54 #include <net/rawv6.h>
55 #include <net/ndisc.h>
56 #include <net/addrconf.h>
57 #include <net/inet_frag.h>
58 
59 struct ip6frag_skb_cb
60 {
61 	struct inet6_skb_parm	h;
62 	int			offset;
63 };
64 
65 #define FRAG6_CB(skb)	((struct ip6frag_skb_cb*)((skb)->cb))
66 
67 
68 /*
69  *	Equivalent of ipv4 struct ipq
70  */
71 
72 struct frag_queue
73 {
74 	struct inet_frag_queue	q;
75 
76 	__be32			id;		/* fragment id		*/
77 	u32			user;
78 	struct in6_addr		saddr;
79 	struct in6_addr		daddr;
80 
81 	int			iif;
82 	unsigned int		csum;
83 	__u16			nhoffset;
84 };
85 
86 static struct inet_frags ip6_frags;
87 
88 int ip6_frag_nqueues(struct net *net)
89 {
90 	return net->ipv6.frags.nqueues;
91 }
92 
93 int ip6_frag_mem(struct net *net)
94 {
95 	return atomic_read(&net->ipv6.frags.mem);
96 }
97 
98 static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *prev,
99 			  struct net_device *dev);
100 
101 /*
102  * callers should be careful not to use the hash value outside the ipfrag_lock
103  * as doing so could race with ipfrag_hash_rnd being recalculated.
104  */
105 unsigned int inet6_hash_frag(__be32 id, const struct in6_addr *saddr,
106 			     const struct in6_addr *daddr, u32 rnd)
107 {
108 	u32 c;
109 
110 	c = jhash_3words((__force u32)saddr->s6_addr32[0],
111 			 (__force u32)saddr->s6_addr32[1],
112 			 (__force u32)saddr->s6_addr32[2],
113 			 rnd);
114 
115 	c = jhash_3words((__force u32)saddr->s6_addr32[3],
116 			 (__force u32)daddr->s6_addr32[0],
117 			 (__force u32)daddr->s6_addr32[1],
118 			 c);
119 
120 	c =  jhash_3words((__force u32)daddr->s6_addr32[2],
121 			  (__force u32)daddr->s6_addr32[3],
122 			  (__force u32)id,
123 			  c);
124 
125 	return c & (INETFRAGS_HASHSZ - 1);
126 }
127 EXPORT_SYMBOL_GPL(inet6_hash_frag);
128 
129 static unsigned int ip6_hashfn(struct inet_frag_queue *q)
130 {
131 	struct frag_queue *fq;
132 
133 	fq = container_of(q, struct frag_queue, q);
134 	return inet6_hash_frag(fq->id, &fq->saddr, &fq->daddr, ip6_frags.rnd);
135 }
136 
137 int ip6_frag_match(struct inet_frag_queue *q, void *a)
138 {
139 	struct frag_queue *fq;
140 	struct ip6_create_arg *arg = a;
141 
142 	fq = container_of(q, struct frag_queue, q);
143 	return (fq->id == arg->id && fq->user == arg->user &&
144 			ipv6_addr_equal(&fq->saddr, arg->src) &&
145 			ipv6_addr_equal(&fq->daddr, arg->dst));
146 }
147 EXPORT_SYMBOL(ip6_frag_match);
148 
149 void ip6_frag_init(struct inet_frag_queue *q, void *a)
150 {
151 	struct frag_queue *fq = container_of(q, struct frag_queue, q);
152 	struct ip6_create_arg *arg = a;
153 
154 	fq->id = arg->id;
155 	fq->user = arg->user;
156 	ipv6_addr_copy(&fq->saddr, arg->src);
157 	ipv6_addr_copy(&fq->daddr, arg->dst);
158 }
159 EXPORT_SYMBOL(ip6_frag_init);
160 
161 /* Destruction primitives. */
162 
163 static __inline__ void fq_put(struct frag_queue *fq)
164 {
165 	inet_frag_put(&fq->q, &ip6_frags);
166 }
167 
168 /* Kill fq entry. It is not destroyed immediately,
169  * because caller (and someone more) holds reference count.
170  */
171 static __inline__ void fq_kill(struct frag_queue *fq)
172 {
173 	inet_frag_kill(&fq->q, &ip6_frags);
174 }
175 
176 static void ip6_evictor(struct net *net, struct inet6_dev *idev)
177 {
178 	int evicted;
179 
180 	evicted = inet_frag_evictor(&net->ipv6.frags, &ip6_frags);
181 	if (evicted)
182 		IP6_ADD_STATS_BH(net, idev, IPSTATS_MIB_REASMFAILS, evicted);
183 }
184 
185 static void ip6_frag_expire(unsigned long data)
186 {
187 	struct frag_queue *fq;
188 	struct net_device *dev = NULL;
189 	struct net *net;
190 
191 	fq = container_of((struct inet_frag_queue *)data, struct frag_queue, q);
192 
193 	spin_lock(&fq->q.lock);
194 
195 	if (fq->q.last_in & INET_FRAG_COMPLETE)
196 		goto out;
197 
198 	fq_kill(fq);
199 
200 	net = container_of(fq->q.net, struct net, ipv6.frags);
201 	rcu_read_lock();
202 	dev = dev_get_by_index_rcu(net, fq->iif);
203 	if (!dev)
204 		goto out_rcu_unlock;
205 
206 	IP6_INC_STATS_BH(net, __in6_dev_get(dev), IPSTATS_MIB_REASMTIMEOUT);
207 	IP6_INC_STATS_BH(net, __in6_dev_get(dev), IPSTATS_MIB_REASMFAILS);
208 
209 	/* Don't send error if the first segment did not arrive. */
210 	if (!(fq->q.last_in & INET_FRAG_FIRST_IN) || !fq->q.fragments)
211 		goto out_rcu_unlock;
212 
213 	/*
214 	   But use as source device on which LAST ARRIVED
215 	   segment was received. And do not use fq->dev
216 	   pointer directly, device might already disappeared.
217 	 */
218 	fq->q.fragments->dev = dev;
219 	icmpv6_send(fq->q.fragments, ICMPV6_TIME_EXCEED, ICMPV6_EXC_FRAGTIME, 0);
220 out_rcu_unlock:
221 	rcu_read_unlock();
222 out:
223 	spin_unlock(&fq->q.lock);
224 	fq_put(fq);
225 }
226 
227 static __inline__ struct frag_queue *
228 fq_find(struct net *net, __be32 id, const struct in6_addr *src, const struct in6_addr *dst)
229 {
230 	struct inet_frag_queue *q;
231 	struct ip6_create_arg arg;
232 	unsigned int hash;
233 
234 	arg.id = id;
235 	arg.user = IP6_DEFRAG_LOCAL_DELIVER;
236 	arg.src = src;
237 	arg.dst = dst;
238 
239 	read_lock(&ip6_frags.lock);
240 	hash = inet6_hash_frag(id, src, dst, ip6_frags.rnd);
241 
242 	q = inet_frag_find(&net->ipv6.frags, &ip6_frags, &arg, hash);
243 	if (q == NULL)
244 		return NULL;
245 
246 	return container_of(q, struct frag_queue, q);
247 }
248 
249 static int ip6_frag_queue(struct frag_queue *fq, struct sk_buff *skb,
250 			   struct frag_hdr *fhdr, int nhoff)
251 {
252 	struct sk_buff *prev, *next;
253 	struct net_device *dev;
254 	int offset, end;
255 	struct net *net = dev_net(skb_dst(skb)->dev);
256 
257 	if (fq->q.last_in & INET_FRAG_COMPLETE)
258 		goto err;
259 
260 	offset = ntohs(fhdr->frag_off) & ~0x7;
261 	end = offset + (ntohs(ipv6_hdr(skb)->payload_len) -
262 			((u8 *)(fhdr + 1) - (u8 *)(ipv6_hdr(skb) + 1)));
263 
264 	if ((unsigned int)end > IPV6_MAXPLEN) {
265 		IP6_INC_STATS_BH(net, ip6_dst_idev(skb_dst(skb)),
266 				 IPSTATS_MIB_INHDRERRORS);
267 		icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
268 				  ((u8 *)&fhdr->frag_off -
269 				   skb_network_header(skb)));
270 		return -1;
271 	}
272 
273 	if (skb->ip_summed == CHECKSUM_COMPLETE) {
274 		const unsigned char *nh = skb_network_header(skb);
275 		skb->csum = csum_sub(skb->csum,
276 				     csum_partial(nh, (u8 *)(fhdr + 1) - nh,
277 						  0));
278 	}
279 
280 	/* Is this the final fragment? */
281 	if (!(fhdr->frag_off & htons(IP6_MF))) {
282 		/* If we already have some bits beyond end
283 		 * or have different end, the segment is corrupted.
284 		 */
285 		if (end < fq->q.len ||
286 		    ((fq->q.last_in & INET_FRAG_LAST_IN) && end != fq->q.len))
287 			goto err;
288 		fq->q.last_in |= INET_FRAG_LAST_IN;
289 		fq->q.len = end;
290 	} else {
291 		/* Check if the fragment is rounded to 8 bytes.
292 		 * Required by the RFC.
293 		 */
294 		if (end & 0x7) {
295 			/* RFC2460 says always send parameter problem in
296 			 * this case. -DaveM
297 			 */
298 			IP6_INC_STATS_BH(net, ip6_dst_idev(skb_dst(skb)),
299 					 IPSTATS_MIB_INHDRERRORS);
300 			icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
301 					  offsetof(struct ipv6hdr, payload_len));
302 			return -1;
303 		}
304 		if (end > fq->q.len) {
305 			/* Some bits beyond end -> corruption. */
306 			if (fq->q.last_in & INET_FRAG_LAST_IN)
307 				goto err;
308 			fq->q.len = end;
309 		}
310 	}
311 
312 	if (end == offset)
313 		goto err;
314 
315 	/* Point into the IP datagram 'data' part. */
316 	if (!pskb_pull(skb, (u8 *) (fhdr + 1) - skb->data))
317 		goto err;
318 
319 	if (pskb_trim_rcsum(skb, end - offset))
320 		goto err;
321 
322 	/* Find out which fragments are in front and at the back of us
323 	 * in the chain of fragments so far.  We must know where to put
324 	 * this fragment, right?
325 	 */
326 	prev = fq->q.fragments_tail;
327 	if (!prev || FRAG6_CB(prev)->offset < offset) {
328 		next = NULL;
329 		goto found;
330 	}
331 	prev = NULL;
332 	for(next = fq->q.fragments; next != NULL; next = next->next) {
333 		if (FRAG6_CB(next)->offset >= offset)
334 			break;	/* bingo! */
335 		prev = next;
336 	}
337 
338 found:
339 	/* RFC5722, Section 4:
340 	 *                                  When reassembling an IPv6 datagram, if
341 	 *   one or more its constituent fragments is determined to be an
342 	 *   overlapping fragment, the entire datagram (and any constituent
343 	 *   fragments, including those not yet received) MUST be silently
344 	 *   discarded.
345 	 */
346 
347 	/* Check for overlap with preceding fragment. */
348 	if (prev &&
349 	    (FRAG6_CB(prev)->offset + prev->len) > offset)
350 		goto discard_fq;
351 
352 	/* Look for overlap with succeeding segment. */
353 	if (next && FRAG6_CB(next)->offset < end)
354 		goto discard_fq;
355 
356 	FRAG6_CB(skb)->offset = offset;
357 
358 	/* Insert this fragment in the chain of fragments. */
359 	skb->next = next;
360 	if (!next)
361 		fq->q.fragments_tail = skb;
362 	if (prev)
363 		prev->next = skb;
364 	else
365 		fq->q.fragments = skb;
366 
367 	dev = skb->dev;
368 	if (dev) {
369 		fq->iif = dev->ifindex;
370 		skb->dev = NULL;
371 	}
372 	fq->q.stamp = skb->tstamp;
373 	fq->q.meat += skb->len;
374 	atomic_add(skb->truesize, &fq->q.net->mem);
375 
376 	/* The first fragment.
377 	 * nhoffset is obtained from the first fragment, of course.
378 	 */
379 	if (offset == 0) {
380 		fq->nhoffset = nhoff;
381 		fq->q.last_in |= INET_FRAG_FIRST_IN;
382 	}
383 
384 	if (fq->q.last_in == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
385 	    fq->q.meat == fq->q.len)
386 		return ip6_frag_reasm(fq, prev, dev);
387 
388 	write_lock(&ip6_frags.lock);
389 	list_move_tail(&fq->q.lru_list, &fq->q.net->lru_list);
390 	write_unlock(&ip6_frags.lock);
391 	return -1;
392 
393 discard_fq:
394 	fq_kill(fq);
395 err:
396 	IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
397 		      IPSTATS_MIB_REASMFAILS);
398 	kfree_skb(skb);
399 	return -1;
400 }
401 
402 /*
403  *	Check if this packet is complete.
404  *	Returns NULL on failure by any reason, and pointer
405  *	to current nexthdr field in reassembled frame.
406  *
407  *	It is called with locked fq, and caller must check that
408  *	queue is eligible for reassembly i.e. it is not COMPLETE,
409  *	the last and the first frames arrived and all the bits are here.
410  */
411 static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *prev,
412 			  struct net_device *dev)
413 {
414 	struct net *net = container_of(fq->q.net, struct net, ipv6.frags);
415 	struct sk_buff *fp, *head = fq->q.fragments;
416 	int    payload_len;
417 	unsigned int nhoff;
418 
419 	fq_kill(fq);
420 
421 	/* Make the one we just received the head. */
422 	if (prev) {
423 		head = prev->next;
424 		fp = skb_clone(head, GFP_ATOMIC);
425 
426 		if (!fp)
427 			goto out_oom;
428 
429 		fp->next = head->next;
430 		if (!fp->next)
431 			fq->q.fragments_tail = fp;
432 		prev->next = fp;
433 
434 		skb_morph(head, fq->q.fragments);
435 		head->next = fq->q.fragments->next;
436 
437 		kfree_skb(fq->q.fragments);
438 		fq->q.fragments = head;
439 	}
440 
441 	WARN_ON(head == NULL);
442 	WARN_ON(FRAG6_CB(head)->offset != 0);
443 
444 	/* Unfragmented part is taken from the first segment. */
445 	payload_len = ((head->data - skb_network_header(head)) -
446 		       sizeof(struct ipv6hdr) + fq->q.len -
447 		       sizeof(struct frag_hdr));
448 	if (payload_len > IPV6_MAXPLEN)
449 		goto out_oversize;
450 
451 	/* Head of list must not be cloned. */
452 	if (skb_cloned(head) && pskb_expand_head(head, 0, 0, GFP_ATOMIC))
453 		goto out_oom;
454 
455 	/* If the first fragment is fragmented itself, we split
456 	 * it to two chunks: the first with data and paged part
457 	 * and the second, holding only fragments. */
458 	if (skb_has_frag_list(head)) {
459 		struct sk_buff *clone;
460 		int i, plen = 0;
461 
462 		if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL)
463 			goto out_oom;
464 		clone->next = head->next;
465 		head->next = clone;
466 		skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
467 		skb_frag_list_init(head);
468 		for (i = 0; i < skb_shinfo(head)->nr_frags; i++)
469 			plen += skb_frag_size(&skb_shinfo(head)->frags[i]);
470 		clone->len = clone->data_len = head->data_len - plen;
471 		head->data_len -= clone->len;
472 		head->len -= clone->len;
473 		clone->csum = 0;
474 		clone->ip_summed = head->ip_summed;
475 		atomic_add(clone->truesize, &fq->q.net->mem);
476 	}
477 
478 	/* We have to remove fragment header from datagram and to relocate
479 	 * header in order to calculate ICV correctly. */
480 	nhoff = fq->nhoffset;
481 	skb_network_header(head)[nhoff] = skb_transport_header(head)[0];
482 	memmove(head->head + sizeof(struct frag_hdr), head->head,
483 		(head->data - head->head) - sizeof(struct frag_hdr));
484 	head->mac_header += sizeof(struct frag_hdr);
485 	head->network_header += sizeof(struct frag_hdr);
486 
487 	skb_shinfo(head)->frag_list = head->next;
488 	skb_reset_transport_header(head);
489 	skb_push(head, head->data - skb_network_header(head));
490 
491 	for (fp=head->next; fp; fp = fp->next) {
492 		head->data_len += fp->len;
493 		head->len += fp->len;
494 		if (head->ip_summed != fp->ip_summed)
495 			head->ip_summed = CHECKSUM_NONE;
496 		else if (head->ip_summed == CHECKSUM_COMPLETE)
497 			head->csum = csum_add(head->csum, fp->csum);
498 		head->truesize += fp->truesize;
499 	}
500 	atomic_sub(head->truesize, &fq->q.net->mem);
501 
502 	head->next = NULL;
503 	head->dev = dev;
504 	head->tstamp = fq->q.stamp;
505 	ipv6_hdr(head)->payload_len = htons(payload_len);
506 	IP6CB(head)->nhoff = nhoff;
507 
508 	/* Yes, and fold redundant checksum back. 8) */
509 	if (head->ip_summed == CHECKSUM_COMPLETE)
510 		head->csum = csum_partial(skb_network_header(head),
511 					  skb_network_header_len(head),
512 					  head->csum);
513 
514 	rcu_read_lock();
515 	IP6_INC_STATS_BH(net, __in6_dev_get(dev), IPSTATS_MIB_REASMOKS);
516 	rcu_read_unlock();
517 	fq->q.fragments = NULL;
518 	fq->q.fragments_tail = NULL;
519 	return 1;
520 
521 out_oversize:
522 	if (net_ratelimit())
523 		printk(KERN_DEBUG "ip6_frag_reasm: payload len = %d\n", payload_len);
524 	goto out_fail;
525 out_oom:
526 	if (net_ratelimit())
527 		printk(KERN_DEBUG "ip6_frag_reasm: no memory for reassembly\n");
528 out_fail:
529 	rcu_read_lock();
530 	IP6_INC_STATS_BH(net, __in6_dev_get(dev), IPSTATS_MIB_REASMFAILS);
531 	rcu_read_unlock();
532 	return -1;
533 }
534 
535 static int ipv6_frag_rcv(struct sk_buff *skb)
536 {
537 	struct frag_hdr *fhdr;
538 	struct frag_queue *fq;
539 	const struct ipv6hdr *hdr = ipv6_hdr(skb);
540 	struct net *net = dev_net(skb_dst(skb)->dev);
541 
542 	IP6_INC_STATS_BH(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMREQDS);
543 
544 	/* Jumbo payload inhibits frag. header */
545 	if (hdr->payload_len==0)
546 		goto fail_hdr;
547 
548 	if (!pskb_may_pull(skb, (skb_transport_offset(skb) +
549 				 sizeof(struct frag_hdr))))
550 		goto fail_hdr;
551 
552 	hdr = ipv6_hdr(skb);
553 	fhdr = (struct frag_hdr *)skb_transport_header(skb);
554 
555 	if (!(fhdr->frag_off & htons(0xFFF9))) {
556 		/* It is not a fragmented frame */
557 		skb->transport_header += sizeof(struct frag_hdr);
558 		IP6_INC_STATS_BH(net,
559 				 ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMOKS);
560 
561 		IP6CB(skb)->nhoff = (u8 *)fhdr - skb_network_header(skb);
562 		return 1;
563 	}
564 
565 	if (atomic_read(&net->ipv6.frags.mem) > net->ipv6.frags.high_thresh)
566 		ip6_evictor(net, ip6_dst_idev(skb_dst(skb)));
567 
568 	fq = fq_find(net, fhdr->identification, &hdr->saddr, &hdr->daddr);
569 	if (fq != NULL) {
570 		int ret;
571 
572 		spin_lock(&fq->q.lock);
573 
574 		ret = ip6_frag_queue(fq, skb, fhdr, IP6CB(skb)->nhoff);
575 
576 		spin_unlock(&fq->q.lock);
577 		fq_put(fq);
578 		return ret;
579 	}
580 
581 	IP6_INC_STATS_BH(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMFAILS);
582 	kfree_skb(skb);
583 	return -1;
584 
585 fail_hdr:
586 	IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_INHDRERRORS);
587 	icmpv6_param_prob(skb, ICMPV6_HDR_FIELD, skb_network_header_len(skb));
588 	return -1;
589 }
590 
591 static const struct inet6_protocol frag_protocol =
592 {
593 	.handler	=	ipv6_frag_rcv,
594 	.flags		=	INET6_PROTO_NOPOLICY,
595 };
596 
597 #ifdef CONFIG_SYSCTL
598 static struct ctl_table ip6_frags_ns_ctl_table[] = {
599 	{
600 		.procname	= "ip6frag_high_thresh",
601 		.data		= &init_net.ipv6.frags.high_thresh,
602 		.maxlen		= sizeof(int),
603 		.mode		= 0644,
604 		.proc_handler	= proc_dointvec
605 	},
606 	{
607 		.procname	= "ip6frag_low_thresh",
608 		.data		= &init_net.ipv6.frags.low_thresh,
609 		.maxlen		= sizeof(int),
610 		.mode		= 0644,
611 		.proc_handler	= proc_dointvec
612 	},
613 	{
614 		.procname	= "ip6frag_time",
615 		.data		= &init_net.ipv6.frags.timeout,
616 		.maxlen		= sizeof(int),
617 		.mode		= 0644,
618 		.proc_handler	= proc_dointvec_jiffies,
619 	},
620 	{ }
621 };
622 
623 static struct ctl_table ip6_frags_ctl_table[] = {
624 	{
625 		.procname	= "ip6frag_secret_interval",
626 		.data		= &ip6_frags.secret_interval,
627 		.maxlen		= sizeof(int),
628 		.mode		= 0644,
629 		.proc_handler	= proc_dointvec_jiffies,
630 	},
631 	{ }
632 };
633 
634 static int __net_init ip6_frags_ns_sysctl_register(struct net *net)
635 {
636 	struct ctl_table *table;
637 	struct ctl_table_header *hdr;
638 
639 	table = ip6_frags_ns_ctl_table;
640 	if (!net_eq(net, &init_net)) {
641 		table = kmemdup(table, sizeof(ip6_frags_ns_ctl_table), GFP_KERNEL);
642 		if (table == NULL)
643 			goto err_alloc;
644 
645 		table[0].data = &net->ipv6.frags.high_thresh;
646 		table[1].data = &net->ipv6.frags.low_thresh;
647 		table[2].data = &net->ipv6.frags.timeout;
648 	}
649 
650 	hdr = register_net_sysctl_table(net, net_ipv6_ctl_path, table);
651 	if (hdr == NULL)
652 		goto err_reg;
653 
654 	net->ipv6.sysctl.frags_hdr = hdr;
655 	return 0;
656 
657 err_reg:
658 	if (!net_eq(net, &init_net))
659 		kfree(table);
660 err_alloc:
661 	return -ENOMEM;
662 }
663 
664 static void __net_exit ip6_frags_ns_sysctl_unregister(struct net *net)
665 {
666 	struct ctl_table *table;
667 
668 	table = net->ipv6.sysctl.frags_hdr->ctl_table_arg;
669 	unregister_net_sysctl_table(net->ipv6.sysctl.frags_hdr);
670 	if (!net_eq(net, &init_net))
671 		kfree(table);
672 }
673 
674 static struct ctl_table_header *ip6_ctl_header;
675 
676 static int ip6_frags_sysctl_register(void)
677 {
678 	ip6_ctl_header = register_net_sysctl_rotable(net_ipv6_ctl_path,
679 			ip6_frags_ctl_table);
680 	return ip6_ctl_header == NULL ? -ENOMEM : 0;
681 }
682 
683 static void ip6_frags_sysctl_unregister(void)
684 {
685 	unregister_net_sysctl_table(ip6_ctl_header);
686 }
687 #else
688 static inline int ip6_frags_ns_sysctl_register(struct net *net)
689 {
690 	return 0;
691 }
692 
693 static inline void ip6_frags_ns_sysctl_unregister(struct net *net)
694 {
695 }
696 
697 static inline int ip6_frags_sysctl_register(void)
698 {
699 	return 0;
700 }
701 
702 static inline void ip6_frags_sysctl_unregister(void)
703 {
704 }
705 #endif
706 
707 static int __net_init ipv6_frags_init_net(struct net *net)
708 {
709 	net->ipv6.frags.high_thresh = IPV6_FRAG_HIGH_THRESH;
710 	net->ipv6.frags.low_thresh = IPV6_FRAG_LOW_THRESH;
711 	net->ipv6.frags.timeout = IPV6_FRAG_TIMEOUT;
712 
713 	inet_frags_init_net(&net->ipv6.frags);
714 
715 	return ip6_frags_ns_sysctl_register(net);
716 }
717 
718 static void __net_exit ipv6_frags_exit_net(struct net *net)
719 {
720 	ip6_frags_ns_sysctl_unregister(net);
721 	inet_frags_exit_net(&net->ipv6.frags, &ip6_frags);
722 }
723 
724 static struct pernet_operations ip6_frags_ops = {
725 	.init = ipv6_frags_init_net,
726 	.exit = ipv6_frags_exit_net,
727 };
728 
729 int __init ipv6_frag_init(void)
730 {
731 	int ret;
732 
733 	ret = inet6_add_protocol(&frag_protocol, IPPROTO_FRAGMENT);
734 	if (ret)
735 		goto out;
736 
737 	ret = ip6_frags_sysctl_register();
738 	if (ret)
739 		goto err_sysctl;
740 
741 	ret = register_pernet_subsys(&ip6_frags_ops);
742 	if (ret)
743 		goto err_pernet;
744 
745 	ip6_frags.hashfn = ip6_hashfn;
746 	ip6_frags.constructor = ip6_frag_init;
747 	ip6_frags.destructor = NULL;
748 	ip6_frags.skb_free = NULL;
749 	ip6_frags.qsize = sizeof(struct frag_queue);
750 	ip6_frags.match = ip6_frag_match;
751 	ip6_frags.frag_expire = ip6_frag_expire;
752 	ip6_frags.secret_interval = 10 * 60 * HZ;
753 	inet_frags_init(&ip6_frags);
754 out:
755 	return ret;
756 
757 err_pernet:
758 	ip6_frags_sysctl_unregister();
759 err_sysctl:
760 	inet6_del_protocol(&frag_protocol, IPPROTO_FRAGMENT);
761 	goto out;
762 }
763 
764 void ipv6_frag_exit(void)
765 {
766 	inet_frags_fini(&ip6_frags);
767 	ip6_frags_sysctl_unregister();
768 	unregister_pernet_subsys(&ip6_frags_ops);
769 	inet6_del_protocol(&frag_protocol, IPPROTO_FRAGMENT);
770 }
771