xref: /linux/net/ipv6/reassembly.c (revision e190bfe56841551b1ad5abb42ebd0c4798cc8c01)
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 
46 #include <net/sock.h>
47 #include <net/snmp.h>
48 
49 #include <net/ipv6.h>
50 #include <net/ip6_route.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 #include <net/inet_frag.h>
57 
58 struct ip6frag_skb_cb
59 {
60 	struct inet6_skb_parm	h;
61 	int			offset;
62 };
63 
64 #define FRAG6_CB(skb)	((struct ip6frag_skb_cb*)((skb)->cb))
65 
66 
67 /*
68  *	Equivalent of ipv4 struct ipq
69  */
70 
71 struct frag_queue
72 {
73 	struct inet_frag_queue	q;
74 
75 	__be32			id;		/* fragment id		*/
76 	u32			user;
77 	struct in6_addr		saddr;
78 	struct in6_addr		daddr;
79 
80 	int			iif;
81 	unsigned int		csum;
82 	__u16			nhoffset;
83 };
84 
85 static struct inet_frags ip6_frags;
86 
87 int ip6_frag_nqueues(struct net *net)
88 {
89 	return net->ipv6.frags.nqueues;
90 }
91 
92 int ip6_frag_mem(struct net *net)
93 {
94 	return atomic_read(&net->ipv6.frags.mem);
95 }
96 
97 static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *prev,
98 			  struct net_device *dev);
99 
100 /*
101  * callers should be careful not to use the hash value outside the ipfrag_lock
102  * as doing so could race with ipfrag_hash_rnd being recalculated.
103  */
104 unsigned int inet6_hash_frag(__be32 id, const struct in6_addr *saddr,
105 			     const struct in6_addr *daddr, u32 rnd)
106 {
107 	u32 a, b, c;
108 
109 	a = (__force u32)saddr->s6_addr32[0];
110 	b = (__force u32)saddr->s6_addr32[1];
111 	c = (__force u32)saddr->s6_addr32[2];
112 
113 	a += JHASH_GOLDEN_RATIO;
114 	b += JHASH_GOLDEN_RATIO;
115 	c += rnd;
116 	__jhash_mix(a, b, c);
117 
118 	a += (__force u32)saddr->s6_addr32[3];
119 	b += (__force u32)daddr->s6_addr32[0];
120 	c += (__force u32)daddr->s6_addr32[1];
121 	__jhash_mix(a, b, c);
122 
123 	a += (__force u32)daddr->s6_addr32[2];
124 	b += (__force u32)daddr->s6_addr32[3];
125 	c += (__force u32)id;
126 	__jhash_mix(a, b, c);
127 
128 	return c & (INETFRAGS_HASHSZ - 1);
129 }
130 EXPORT_SYMBOL_GPL(inet6_hash_frag);
131 
132 static unsigned int ip6_hashfn(struct inet_frag_queue *q)
133 {
134 	struct frag_queue *fq;
135 
136 	fq = container_of(q, struct frag_queue, q);
137 	return inet6_hash_frag(fq->id, &fq->saddr, &fq->daddr, ip6_frags.rnd);
138 }
139 
140 int ip6_frag_match(struct inet_frag_queue *q, void *a)
141 {
142 	struct frag_queue *fq;
143 	struct ip6_create_arg *arg = a;
144 
145 	fq = container_of(q, struct frag_queue, q);
146 	return (fq->id == arg->id && fq->user == arg->user &&
147 			ipv6_addr_equal(&fq->saddr, arg->src) &&
148 			ipv6_addr_equal(&fq->daddr, arg->dst));
149 }
150 EXPORT_SYMBOL(ip6_frag_match);
151 
152 /* Memory Tracking Functions. */
153 static inline void frag_kfree_skb(struct netns_frags *nf,
154 		struct sk_buff *skb, int *work)
155 {
156 	if (work)
157 		*work -= skb->truesize;
158 	atomic_sub(skb->truesize, &nf->mem);
159 	kfree_skb(skb);
160 }
161 
162 void ip6_frag_init(struct inet_frag_queue *q, void *a)
163 {
164 	struct frag_queue *fq = container_of(q, struct frag_queue, q);
165 	struct ip6_create_arg *arg = a;
166 
167 	fq->id = arg->id;
168 	fq->user = arg->user;
169 	ipv6_addr_copy(&fq->saddr, arg->src);
170 	ipv6_addr_copy(&fq->daddr, arg->dst);
171 }
172 EXPORT_SYMBOL(ip6_frag_init);
173 
174 /* Destruction primitives. */
175 
176 static __inline__ void fq_put(struct frag_queue *fq)
177 {
178 	inet_frag_put(&fq->q, &ip6_frags);
179 }
180 
181 /* Kill fq entry. It is not destroyed immediately,
182  * because caller (and someone more) holds reference count.
183  */
184 static __inline__ void fq_kill(struct frag_queue *fq)
185 {
186 	inet_frag_kill(&fq->q, &ip6_frags);
187 }
188 
189 static void ip6_evictor(struct net *net, struct inet6_dev *idev)
190 {
191 	int evicted;
192 
193 	evicted = inet_frag_evictor(&net->ipv6.frags, &ip6_frags);
194 	if (evicted)
195 		IP6_ADD_STATS_BH(net, idev, IPSTATS_MIB_REASMFAILS, evicted);
196 }
197 
198 static void ip6_frag_expire(unsigned long data)
199 {
200 	struct frag_queue *fq;
201 	struct net_device *dev = NULL;
202 	struct net *net;
203 
204 	fq = container_of((struct inet_frag_queue *)data, struct frag_queue, q);
205 
206 	spin_lock(&fq->q.lock);
207 
208 	if (fq->q.last_in & INET_FRAG_COMPLETE)
209 		goto out;
210 
211 	fq_kill(fq);
212 
213 	net = container_of(fq->q.net, struct net, ipv6.frags);
214 	rcu_read_lock();
215 	dev = dev_get_by_index_rcu(net, fq->iif);
216 	if (!dev)
217 		goto out_rcu_unlock;
218 
219 	IP6_INC_STATS_BH(net, __in6_dev_get(dev), IPSTATS_MIB_REASMTIMEOUT);
220 	IP6_INC_STATS_BH(net, __in6_dev_get(dev), IPSTATS_MIB_REASMFAILS);
221 
222 	/* Don't send error if the first segment did not arrive. */
223 	if (!(fq->q.last_in & INET_FRAG_FIRST_IN) || !fq->q.fragments)
224 		goto out_rcu_unlock;
225 
226 	/*
227 	   But use as source device on which LAST ARRIVED
228 	   segment was received. And do not use fq->dev
229 	   pointer directly, device might already disappeared.
230 	 */
231 	fq->q.fragments->dev = dev;
232 	icmpv6_send(fq->q.fragments, ICMPV6_TIME_EXCEED, ICMPV6_EXC_FRAGTIME, 0);
233 out_rcu_unlock:
234 	rcu_read_unlock();
235 out:
236 	spin_unlock(&fq->q.lock);
237 	fq_put(fq);
238 }
239 
240 static __inline__ struct frag_queue *
241 fq_find(struct net *net, __be32 id, struct in6_addr *src, struct in6_addr *dst)
242 {
243 	struct inet_frag_queue *q;
244 	struct ip6_create_arg arg;
245 	unsigned int hash;
246 
247 	arg.id = id;
248 	arg.user = IP6_DEFRAG_LOCAL_DELIVER;
249 	arg.src = src;
250 	arg.dst = dst;
251 
252 	read_lock(&ip6_frags.lock);
253 	hash = inet6_hash_frag(id, src, dst, ip6_frags.rnd);
254 
255 	q = inet_frag_find(&net->ipv6.frags, &ip6_frags, &arg, hash);
256 	if (q == NULL)
257 		return NULL;
258 
259 	return container_of(q, struct frag_queue, q);
260 }
261 
262 static int ip6_frag_queue(struct frag_queue *fq, struct sk_buff *skb,
263 			   struct frag_hdr *fhdr, int nhoff)
264 {
265 	struct sk_buff *prev, *next;
266 	struct net_device *dev;
267 	int offset, end;
268 	struct net *net = dev_net(skb_dst(skb)->dev);
269 
270 	if (fq->q.last_in & INET_FRAG_COMPLETE)
271 		goto err;
272 
273 	offset = ntohs(fhdr->frag_off) & ~0x7;
274 	end = offset + (ntohs(ipv6_hdr(skb)->payload_len) -
275 			((u8 *)(fhdr + 1) - (u8 *)(ipv6_hdr(skb) + 1)));
276 
277 	if ((unsigned int)end > IPV6_MAXPLEN) {
278 		IP6_INC_STATS_BH(net, ip6_dst_idev(skb_dst(skb)),
279 				 IPSTATS_MIB_INHDRERRORS);
280 		icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
281 				  ((u8 *)&fhdr->frag_off -
282 				   skb_network_header(skb)));
283 		return -1;
284 	}
285 
286 	if (skb->ip_summed == CHECKSUM_COMPLETE) {
287 		const unsigned char *nh = skb_network_header(skb);
288 		skb->csum = csum_sub(skb->csum,
289 				     csum_partial(nh, (u8 *)(fhdr + 1) - nh,
290 						  0));
291 	}
292 
293 	/* Is this the final fragment? */
294 	if (!(fhdr->frag_off & htons(IP6_MF))) {
295 		/* If we already have some bits beyond end
296 		 * or have different end, the segment is corrupted.
297 		 */
298 		if (end < fq->q.len ||
299 		    ((fq->q.last_in & INET_FRAG_LAST_IN) && end != fq->q.len))
300 			goto err;
301 		fq->q.last_in |= INET_FRAG_LAST_IN;
302 		fq->q.len = end;
303 	} else {
304 		/* Check if the fragment is rounded to 8 bytes.
305 		 * Required by the RFC.
306 		 */
307 		if (end & 0x7) {
308 			/* RFC2460 says always send parameter problem in
309 			 * this case. -DaveM
310 			 */
311 			IP6_INC_STATS_BH(net, ip6_dst_idev(skb_dst(skb)),
312 					 IPSTATS_MIB_INHDRERRORS);
313 			icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
314 					  offsetof(struct ipv6hdr, payload_len));
315 			return -1;
316 		}
317 		if (end > fq->q.len) {
318 			/* Some bits beyond end -> corruption. */
319 			if (fq->q.last_in & INET_FRAG_LAST_IN)
320 				goto err;
321 			fq->q.len = end;
322 		}
323 	}
324 
325 	if (end == offset)
326 		goto err;
327 
328 	/* Point into the IP datagram 'data' part. */
329 	if (!pskb_pull(skb, (u8 *) (fhdr + 1) - skb->data))
330 		goto err;
331 
332 	if (pskb_trim_rcsum(skb, end - offset))
333 		goto err;
334 
335 	/* Find out which fragments are in front and at the back of us
336 	 * in the chain of fragments so far.  We must know where to put
337 	 * this fragment, right?
338 	 */
339 	prev = NULL;
340 	for(next = fq->q.fragments; next != NULL; next = next->next) {
341 		if (FRAG6_CB(next)->offset >= offset)
342 			break;	/* bingo! */
343 		prev = next;
344 	}
345 
346 	/* We found where to put this one.  Check for overlap with
347 	 * preceding fragment, and, if needed, align things so that
348 	 * any overlaps are eliminated.
349 	 */
350 	if (prev) {
351 		int i = (FRAG6_CB(prev)->offset + prev->len) - offset;
352 
353 		if (i > 0) {
354 			offset += i;
355 			if (end <= offset)
356 				goto err;
357 			if (!pskb_pull(skb, i))
358 				goto err;
359 			if (skb->ip_summed != CHECKSUM_UNNECESSARY)
360 				skb->ip_summed = CHECKSUM_NONE;
361 		}
362 	}
363 
364 	/* Look for overlap with succeeding segments.
365 	 * If we can merge fragments, do it.
366 	 */
367 	while (next && FRAG6_CB(next)->offset < end) {
368 		int i = end - FRAG6_CB(next)->offset; /* overlap is 'i' bytes */
369 
370 		if (i < next->len) {
371 			/* Eat head of the next overlapped fragment
372 			 * and leave the loop. The next ones cannot overlap.
373 			 */
374 			if (!pskb_pull(next, i))
375 				goto err;
376 			FRAG6_CB(next)->offset += i;	/* next fragment */
377 			fq->q.meat -= i;
378 			if (next->ip_summed != CHECKSUM_UNNECESSARY)
379 				next->ip_summed = CHECKSUM_NONE;
380 			break;
381 		} else {
382 			struct sk_buff *free_it = next;
383 
384 			/* Old fragment is completely overridden with
385 			 * new one drop it.
386 			 */
387 			next = next->next;
388 
389 			if (prev)
390 				prev->next = next;
391 			else
392 				fq->q.fragments = next;
393 
394 			fq->q.meat -= free_it->len;
395 			frag_kfree_skb(fq->q.net, free_it, NULL);
396 		}
397 	}
398 
399 	FRAG6_CB(skb)->offset = offset;
400 
401 	/* Insert this fragment in the chain of fragments. */
402 	skb->next = next;
403 	if (prev)
404 		prev->next = skb;
405 	else
406 		fq->q.fragments = skb;
407 
408 	dev = skb->dev;
409 	if (dev) {
410 		fq->iif = dev->ifindex;
411 		skb->dev = NULL;
412 	}
413 	fq->q.stamp = skb->tstamp;
414 	fq->q.meat += skb->len;
415 	atomic_add(skb->truesize, &fq->q.net->mem);
416 
417 	/* The first fragment.
418 	 * nhoffset is obtained from the first fragment, of course.
419 	 */
420 	if (offset == 0) {
421 		fq->nhoffset = nhoff;
422 		fq->q.last_in |= INET_FRAG_FIRST_IN;
423 	}
424 
425 	if (fq->q.last_in == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
426 	    fq->q.meat == fq->q.len)
427 		return ip6_frag_reasm(fq, prev, dev);
428 
429 	write_lock(&ip6_frags.lock);
430 	list_move_tail(&fq->q.lru_list, &fq->q.net->lru_list);
431 	write_unlock(&ip6_frags.lock);
432 	return -1;
433 
434 err:
435 	IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
436 		      IPSTATS_MIB_REASMFAILS);
437 	kfree_skb(skb);
438 	return -1;
439 }
440 
441 /*
442  *	Check if this packet is complete.
443  *	Returns NULL on failure by any reason, and pointer
444  *	to current nexthdr field in reassembled frame.
445  *
446  *	It is called with locked fq, and caller must check that
447  *	queue is eligible for reassembly i.e. it is not COMPLETE,
448  *	the last and the first frames arrived and all the bits are here.
449  */
450 static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *prev,
451 			  struct net_device *dev)
452 {
453 	struct net *net = container_of(fq->q.net, struct net, ipv6.frags);
454 	struct sk_buff *fp, *head = fq->q.fragments;
455 	int    payload_len;
456 	unsigned int nhoff;
457 
458 	fq_kill(fq);
459 
460 	/* Make the one we just received the head. */
461 	if (prev) {
462 		head = prev->next;
463 		fp = skb_clone(head, GFP_ATOMIC);
464 
465 		if (!fp)
466 			goto out_oom;
467 
468 		fp->next = head->next;
469 		prev->next = fp;
470 
471 		skb_morph(head, fq->q.fragments);
472 		head->next = fq->q.fragments->next;
473 
474 		kfree_skb(fq->q.fragments);
475 		fq->q.fragments = head;
476 	}
477 
478 	WARN_ON(head == NULL);
479 	WARN_ON(FRAG6_CB(head)->offset != 0);
480 
481 	/* Unfragmented part is taken from the first segment. */
482 	payload_len = ((head->data - skb_network_header(head)) -
483 		       sizeof(struct ipv6hdr) + fq->q.len -
484 		       sizeof(struct frag_hdr));
485 	if (payload_len > IPV6_MAXPLEN)
486 		goto out_oversize;
487 
488 	/* Head of list must not be cloned. */
489 	if (skb_cloned(head) && pskb_expand_head(head, 0, 0, GFP_ATOMIC))
490 		goto out_oom;
491 
492 	/* If the first fragment is fragmented itself, we split
493 	 * it to two chunks: the first with data and paged part
494 	 * and the second, holding only fragments. */
495 	if (skb_has_frags(head)) {
496 		struct sk_buff *clone;
497 		int i, plen = 0;
498 
499 		if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL)
500 			goto out_oom;
501 		clone->next = head->next;
502 		head->next = clone;
503 		skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
504 		skb_frag_list_init(head);
505 		for (i=0; i<skb_shinfo(head)->nr_frags; i++)
506 			plen += skb_shinfo(head)->frags[i].size;
507 		clone->len = clone->data_len = head->data_len - plen;
508 		head->data_len -= clone->len;
509 		head->len -= clone->len;
510 		clone->csum = 0;
511 		clone->ip_summed = head->ip_summed;
512 		atomic_add(clone->truesize, &fq->q.net->mem);
513 	}
514 
515 	/* We have to remove fragment header from datagram and to relocate
516 	 * header in order to calculate ICV correctly. */
517 	nhoff = fq->nhoffset;
518 	skb_network_header(head)[nhoff] = skb_transport_header(head)[0];
519 	memmove(head->head + sizeof(struct frag_hdr), head->head,
520 		(head->data - head->head) - sizeof(struct frag_hdr));
521 	head->mac_header += sizeof(struct frag_hdr);
522 	head->network_header += sizeof(struct frag_hdr);
523 
524 	skb_shinfo(head)->frag_list = head->next;
525 	skb_reset_transport_header(head);
526 	skb_push(head, head->data - skb_network_header(head));
527 	atomic_sub(head->truesize, &fq->q.net->mem);
528 
529 	for (fp=head->next; fp; fp = fp->next) {
530 		head->data_len += fp->len;
531 		head->len += fp->len;
532 		if (head->ip_summed != fp->ip_summed)
533 			head->ip_summed = CHECKSUM_NONE;
534 		else if (head->ip_summed == CHECKSUM_COMPLETE)
535 			head->csum = csum_add(head->csum, fp->csum);
536 		head->truesize += fp->truesize;
537 		atomic_sub(fp->truesize, &fq->q.net->mem);
538 	}
539 
540 	head->next = NULL;
541 	head->dev = dev;
542 	head->tstamp = fq->q.stamp;
543 	ipv6_hdr(head)->payload_len = htons(payload_len);
544 	IP6CB(head)->nhoff = nhoff;
545 
546 	/* Yes, and fold redundant checksum back. 8) */
547 	if (head->ip_summed == CHECKSUM_COMPLETE)
548 		head->csum = csum_partial(skb_network_header(head),
549 					  skb_network_header_len(head),
550 					  head->csum);
551 
552 	rcu_read_lock();
553 	IP6_INC_STATS_BH(net, __in6_dev_get(dev), IPSTATS_MIB_REASMOKS);
554 	rcu_read_unlock();
555 	fq->q.fragments = NULL;
556 	return 1;
557 
558 out_oversize:
559 	if (net_ratelimit())
560 		printk(KERN_DEBUG "ip6_frag_reasm: payload len = %d\n", payload_len);
561 	goto out_fail;
562 out_oom:
563 	if (net_ratelimit())
564 		printk(KERN_DEBUG "ip6_frag_reasm: no memory for reassembly\n");
565 out_fail:
566 	rcu_read_lock();
567 	IP6_INC_STATS_BH(net, __in6_dev_get(dev), IPSTATS_MIB_REASMFAILS);
568 	rcu_read_unlock();
569 	return -1;
570 }
571 
572 static int ipv6_frag_rcv(struct sk_buff *skb)
573 {
574 	struct frag_hdr *fhdr;
575 	struct frag_queue *fq;
576 	struct ipv6hdr *hdr = ipv6_hdr(skb);
577 	struct net *net = dev_net(skb_dst(skb)->dev);
578 
579 	IP6_INC_STATS_BH(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMREQDS);
580 
581 	/* Jumbo payload inhibits frag. header */
582 	if (hdr->payload_len==0)
583 		goto fail_hdr;
584 
585 	if (!pskb_may_pull(skb, (skb_transport_offset(skb) +
586 				 sizeof(struct frag_hdr))))
587 		goto fail_hdr;
588 
589 	hdr = ipv6_hdr(skb);
590 	fhdr = (struct frag_hdr *)skb_transport_header(skb);
591 
592 	if (!(fhdr->frag_off & htons(0xFFF9))) {
593 		/* It is not a fragmented frame */
594 		skb->transport_header += sizeof(struct frag_hdr);
595 		IP6_INC_STATS_BH(net,
596 				 ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMOKS);
597 
598 		IP6CB(skb)->nhoff = (u8 *)fhdr - skb_network_header(skb);
599 		return 1;
600 	}
601 
602 	if (atomic_read(&net->ipv6.frags.mem) > net->ipv6.frags.high_thresh)
603 		ip6_evictor(net, ip6_dst_idev(skb_dst(skb)));
604 
605 	fq = fq_find(net, fhdr->identification, &hdr->saddr, &hdr->daddr);
606 	if (fq != NULL) {
607 		int ret;
608 
609 		spin_lock(&fq->q.lock);
610 
611 		ret = ip6_frag_queue(fq, skb, fhdr, IP6CB(skb)->nhoff);
612 
613 		spin_unlock(&fq->q.lock);
614 		fq_put(fq);
615 		return ret;
616 	}
617 
618 	IP6_INC_STATS_BH(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMFAILS);
619 	kfree_skb(skb);
620 	return -1;
621 
622 fail_hdr:
623 	IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_INHDRERRORS);
624 	icmpv6_param_prob(skb, ICMPV6_HDR_FIELD, skb_network_header_len(skb));
625 	return -1;
626 }
627 
628 static const struct inet6_protocol frag_protocol =
629 {
630 	.handler	=	ipv6_frag_rcv,
631 	.flags		=	INET6_PROTO_NOPOLICY,
632 };
633 
634 #ifdef CONFIG_SYSCTL
635 static struct ctl_table ip6_frags_ns_ctl_table[] = {
636 	{
637 		.procname	= "ip6frag_high_thresh",
638 		.data		= &init_net.ipv6.frags.high_thresh,
639 		.maxlen		= sizeof(int),
640 		.mode		= 0644,
641 		.proc_handler	= proc_dointvec
642 	},
643 	{
644 		.procname	= "ip6frag_low_thresh",
645 		.data		= &init_net.ipv6.frags.low_thresh,
646 		.maxlen		= sizeof(int),
647 		.mode		= 0644,
648 		.proc_handler	= proc_dointvec
649 	},
650 	{
651 		.procname	= "ip6frag_time",
652 		.data		= &init_net.ipv6.frags.timeout,
653 		.maxlen		= sizeof(int),
654 		.mode		= 0644,
655 		.proc_handler	= proc_dointvec_jiffies,
656 	},
657 	{ }
658 };
659 
660 static struct ctl_table ip6_frags_ctl_table[] = {
661 	{
662 		.procname	= "ip6frag_secret_interval",
663 		.data		= &ip6_frags.secret_interval,
664 		.maxlen		= sizeof(int),
665 		.mode		= 0644,
666 		.proc_handler	= proc_dointvec_jiffies,
667 	},
668 	{ }
669 };
670 
671 static int __net_init ip6_frags_ns_sysctl_register(struct net *net)
672 {
673 	struct ctl_table *table;
674 	struct ctl_table_header *hdr;
675 
676 	table = ip6_frags_ns_ctl_table;
677 	if (!net_eq(net, &init_net)) {
678 		table = kmemdup(table, sizeof(ip6_frags_ns_ctl_table), GFP_KERNEL);
679 		if (table == NULL)
680 			goto err_alloc;
681 
682 		table[0].data = &net->ipv6.frags.high_thresh;
683 		table[1].data = &net->ipv6.frags.low_thresh;
684 		table[2].data = &net->ipv6.frags.timeout;
685 	}
686 
687 	hdr = register_net_sysctl_table(net, net_ipv6_ctl_path, table);
688 	if (hdr == NULL)
689 		goto err_reg;
690 
691 	net->ipv6.sysctl.frags_hdr = hdr;
692 	return 0;
693 
694 err_reg:
695 	if (!net_eq(net, &init_net))
696 		kfree(table);
697 err_alloc:
698 	return -ENOMEM;
699 }
700 
701 static void __net_exit ip6_frags_ns_sysctl_unregister(struct net *net)
702 {
703 	struct ctl_table *table;
704 
705 	table = net->ipv6.sysctl.frags_hdr->ctl_table_arg;
706 	unregister_net_sysctl_table(net->ipv6.sysctl.frags_hdr);
707 	if (!net_eq(net, &init_net))
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 __net_init ipv6_frags_init_net(struct net *net)
745 {
746 	net->ipv6.frags.high_thresh = IPV6_FRAG_HIGH_THRESH;
747 	net->ipv6.frags.low_thresh = IPV6_FRAG_LOW_THRESH;
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 __net_exit 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