xref: /openbmc/linux/net/ipv6/reassembly.c (revision 98b3377ca77a06a7bd75a444e9f7136e9bb5112e)
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(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 	dev = dev_get_by_index(net, fq->iif);
212 	if (!dev)
213 		goto out;
214 
215 	rcu_read_lock();
216 	IP6_INC_STATS_BH(__in6_dev_get(dev), IPSTATS_MIB_REASMTIMEOUT);
217 	IP6_INC_STATS_BH(__in6_dev_get(dev), IPSTATS_MIB_REASMFAILS);
218 	rcu_read_unlock();
219 
220 	/* Don't send error if the first segment did not arrive. */
221 	if (!(fq->q.last_in & INET_FRAG_FIRST_IN) || !fq->q.fragments)
222 		goto out;
223 
224 	/*
225 	   But use as source device on which LAST ARRIVED
226 	   segment was received. And do not use fq->dev
227 	   pointer directly, device might already disappeared.
228 	 */
229 	fq->q.fragments->dev = dev;
230 	icmpv6_send(fq->q.fragments, ICMPV6_TIME_EXCEED, ICMPV6_EXC_FRAGTIME, 0, dev);
231 out:
232 	if (dev)
233 		dev_put(dev);
234 	spin_unlock(&fq->q.lock);
235 	fq_put(fq);
236 }
237 
238 static __inline__ struct frag_queue *
239 fq_find(struct net *net, __be32 id, struct in6_addr *src, struct in6_addr *dst,
240 	struct inet6_dev *idev)
241 {
242 	struct inet_frag_queue *q;
243 	struct ip6_create_arg arg;
244 	unsigned int hash;
245 
246 	arg.id = id;
247 	arg.src = src;
248 	arg.dst = dst;
249 
250 	read_lock(&ip6_frags.lock);
251 	hash = inet6_hash_frag(id, src, dst, ip6_frags.rnd);
252 
253 	q = inet_frag_find(&net->ipv6.frags, &ip6_frags, &arg, hash);
254 	if (q == NULL)
255 		goto oom;
256 
257 	return container_of(q, struct frag_queue, q);
258 
259 oom:
260 	IP6_INC_STATS_BH(idev, IPSTATS_MIB_REASMFAILS);
261 	return NULL;
262 }
263 
264 static int ip6_frag_queue(struct frag_queue *fq, struct sk_buff *skb,
265 			   struct frag_hdr *fhdr, int nhoff)
266 {
267 	struct sk_buff *prev, *next;
268 	struct net_device *dev;
269 	int offset, end;
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(ip6_dst_idev(skb->dst),
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(ip6_dst_idev(skb->dst),
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(ip6_dst_idev(skb->dst), 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 sk_buff *fp, *head = fq->q.fragments;
454 	int    payload_len;
455 	unsigned int nhoff;
456 
457 	fq_kill(fq);
458 
459 	/* Make the one we just received the head. */
460 	if (prev) {
461 		head = prev->next;
462 		fp = skb_clone(head, GFP_ATOMIC);
463 
464 		if (!fp)
465 			goto out_oom;
466 
467 		fp->next = head->next;
468 		prev->next = fp;
469 
470 		skb_morph(head, fq->q.fragments);
471 		head->next = fq->q.fragments->next;
472 
473 		kfree_skb(fq->q.fragments);
474 		fq->q.fragments = head;
475 	}
476 
477 	WARN_ON(head == NULL);
478 	WARN_ON(FRAG6_CB(head)->offset != 0);
479 
480 	/* Unfragmented part is taken from the first segment. */
481 	payload_len = ((head->data - skb_network_header(head)) -
482 		       sizeof(struct ipv6hdr) + fq->q.len -
483 		       sizeof(struct frag_hdr));
484 	if (payload_len > IPV6_MAXPLEN)
485 		goto out_oversize;
486 
487 	/* Head of list must not be cloned. */
488 	if (skb_cloned(head) && pskb_expand_head(head, 0, 0, GFP_ATOMIC))
489 		goto out_oom;
490 
491 	/* If the first fragment is fragmented itself, we split
492 	 * it to two chunks: the first with data and paged part
493 	 * and the second, holding only fragments. */
494 	if (skb_shinfo(head)->frag_list) {
495 		struct sk_buff *clone;
496 		int i, plen = 0;
497 
498 		if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL)
499 			goto out_oom;
500 		clone->next = head->next;
501 		head->next = clone;
502 		skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
503 		skb_shinfo(head)->frag_list = NULL;
504 		for (i=0; i<skb_shinfo(head)->nr_frags; i++)
505 			plen += skb_shinfo(head)->frags[i].size;
506 		clone->len = clone->data_len = head->data_len - plen;
507 		head->data_len -= clone->len;
508 		head->len -= clone->len;
509 		clone->csum = 0;
510 		clone->ip_summed = head->ip_summed;
511 		atomic_add(clone->truesize, &fq->q.net->mem);
512 	}
513 
514 	/* We have to remove fragment header from datagram and to relocate
515 	 * header in order to calculate ICV correctly. */
516 	nhoff = fq->nhoffset;
517 	skb_network_header(head)[nhoff] = skb_transport_header(head)[0];
518 	memmove(head->head + sizeof(struct frag_hdr), head->head,
519 		(head->data - head->head) - sizeof(struct frag_hdr));
520 	head->mac_header += sizeof(struct frag_hdr);
521 	head->network_header += sizeof(struct frag_hdr);
522 
523 	skb_shinfo(head)->frag_list = head->next;
524 	skb_reset_transport_header(head);
525 	skb_push(head, head->data - skb_network_header(head));
526 	atomic_sub(head->truesize, &fq->q.net->mem);
527 
528 	for (fp=head->next; fp; fp = fp->next) {
529 		head->data_len += fp->len;
530 		head->len += fp->len;
531 		if (head->ip_summed != fp->ip_summed)
532 			head->ip_summed = CHECKSUM_NONE;
533 		else if (head->ip_summed == CHECKSUM_COMPLETE)
534 			head->csum = csum_add(head->csum, fp->csum);
535 		head->truesize += fp->truesize;
536 		atomic_sub(fp->truesize, &fq->q.net->mem);
537 	}
538 
539 	head->next = NULL;
540 	head->dev = dev;
541 	head->tstamp = fq->q.stamp;
542 	ipv6_hdr(head)->payload_len = htons(payload_len);
543 	IP6CB(head)->nhoff = nhoff;
544 
545 	/* Yes, and fold redundant checksum back. 8) */
546 	if (head->ip_summed == CHECKSUM_COMPLETE)
547 		head->csum = csum_partial(skb_network_header(head),
548 					  skb_network_header_len(head),
549 					  head->csum);
550 
551 	rcu_read_lock();
552 	IP6_INC_STATS_BH(__in6_dev_get(dev), IPSTATS_MIB_REASMOKS);
553 	rcu_read_unlock();
554 	fq->q.fragments = NULL;
555 	return 1;
556 
557 out_oversize:
558 	if (net_ratelimit())
559 		printk(KERN_DEBUG "ip6_frag_reasm: payload len = %d\n", payload_len);
560 	goto out_fail;
561 out_oom:
562 	if (net_ratelimit())
563 		printk(KERN_DEBUG "ip6_frag_reasm: no memory for reassembly\n");
564 out_fail:
565 	rcu_read_lock();
566 	IP6_INC_STATS_BH(__in6_dev_get(dev), IPSTATS_MIB_REASMFAILS);
567 	rcu_read_unlock();
568 	return -1;
569 }
570 
571 static int ipv6_frag_rcv(struct sk_buff *skb)
572 {
573 	struct frag_hdr *fhdr;
574 	struct frag_queue *fq;
575 	struct ipv6hdr *hdr = ipv6_hdr(skb);
576 	struct net *net;
577 
578 	IP6_INC_STATS_BH(ip6_dst_idev(skb->dst), IPSTATS_MIB_REASMREQDS);
579 
580 	/* Jumbo payload inhibits frag. header */
581 	if (hdr->payload_len==0)
582 		goto fail_hdr;
583 
584 	if (!pskb_may_pull(skb, (skb_transport_offset(skb) +
585 				 sizeof(struct frag_hdr))))
586 		goto fail_hdr;
587 
588 	hdr = ipv6_hdr(skb);
589 	fhdr = (struct frag_hdr *)skb_transport_header(skb);
590 
591 	if (!(fhdr->frag_off & htons(0xFFF9))) {
592 		/* It is not a fragmented frame */
593 		skb->transport_header += sizeof(struct frag_hdr);
594 		IP6_INC_STATS_BH(ip6_dst_idev(skb->dst), IPSTATS_MIB_REASMOKS);
595 
596 		IP6CB(skb)->nhoff = (u8 *)fhdr - skb_network_header(skb);
597 		return 1;
598 	}
599 
600 	net = dev_net(skb->dev);
601 	if (atomic_read(&net->ipv6.frags.mem) > net->ipv6.frags.high_thresh)
602 		ip6_evictor(net, ip6_dst_idev(skb->dst));
603 
604 	if ((fq = fq_find(net, fhdr->identification, &hdr->saddr, &hdr->daddr,
605 			  ip6_dst_idev(skb->dst))) != NULL) {
606 		int ret;
607 
608 		spin_lock(&fq->q.lock);
609 
610 		ret = ip6_frag_queue(fq, skb, fhdr, IP6CB(skb)->nhoff);
611 
612 		spin_unlock(&fq->q.lock);
613 		fq_put(fq);
614 		return ret;
615 	}
616 
617 	IP6_INC_STATS_BH(ip6_dst_idev(skb->dst), IPSTATS_MIB_REASMFAILS);
618 	kfree_skb(skb);
619 	return -1;
620 
621 fail_hdr:
622 	IP6_INC_STATS(ip6_dst_idev(skb->dst), IPSTATS_MIB_INHDRERRORS);
623 	icmpv6_param_prob(skb, ICMPV6_HDR_FIELD, skb_network_header_len(skb));
624 	return -1;
625 }
626 
627 static struct inet6_protocol frag_protocol =
628 {
629 	.handler	=	ipv6_frag_rcv,
630 	.flags		=	INET6_PROTO_NOPOLICY,
631 };
632 
633 #ifdef CONFIG_SYSCTL
634 static struct ctl_table ip6_frags_ns_ctl_table[] = {
635 	{
636 		.ctl_name	= NET_IPV6_IP6FRAG_HIGH_THRESH,
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 		.ctl_name	= NET_IPV6_IP6FRAG_LOW_THRESH,
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 		.ctl_name	= NET_IPV6_IP6FRAG_TIME,
653 		.procname	= "ip6frag_time",
654 		.data		= &init_net.ipv6.frags.timeout,
655 		.maxlen		= sizeof(int),
656 		.mode		= 0644,
657 		.proc_handler	= &proc_dointvec_jiffies,
658 		.strategy	= &sysctl_jiffies,
659 	},
660 	{ }
661 };
662 
663 static struct ctl_table ip6_frags_ctl_table[] = {
664 	{
665 		.ctl_name	= NET_IPV6_IP6FRAG_SECRET_INTERVAL,
666 		.procname	= "ip6frag_secret_interval",
667 		.data		= &ip6_frags.secret_interval,
668 		.maxlen		= sizeof(int),
669 		.mode		= 0644,
670 		.proc_handler	= &proc_dointvec_jiffies,
671 		.strategy	= &sysctl_jiffies
672 	},
673 	{ }
674 };
675 
676 static int ip6_frags_ns_sysctl_register(struct net *net)
677 {
678 	struct ctl_table *table;
679 	struct ctl_table_header *hdr;
680 
681 	table = ip6_frags_ns_ctl_table;
682 	if (net != &init_net) {
683 		table = kmemdup(table, sizeof(ip6_frags_ns_ctl_table), GFP_KERNEL);
684 		if (table == NULL)
685 			goto err_alloc;
686 
687 		table[0].data = &net->ipv6.frags.high_thresh;
688 		table[1].data = &net->ipv6.frags.low_thresh;
689 		table[2].data = &net->ipv6.frags.timeout;
690 	}
691 
692 	hdr = register_net_sysctl_table(net, net_ipv6_ctl_path, table);
693 	if (hdr == NULL)
694 		goto err_reg;
695 
696 	net->ipv6.sysctl.frags_hdr = hdr;
697 	return 0;
698 
699 err_reg:
700 	if (net != &init_net)
701 		kfree(table);
702 err_alloc:
703 	return -ENOMEM;
704 }
705 
706 static void ip6_frags_ns_sysctl_unregister(struct net *net)
707 {
708 	struct ctl_table *table;
709 
710 	table = net->ipv6.sysctl.frags_hdr->ctl_table_arg;
711 	unregister_net_sysctl_table(net->ipv6.sysctl.frags_hdr);
712 	kfree(table);
713 }
714 
715 static struct ctl_table_header *ip6_ctl_header;
716 
717 static int ip6_frags_sysctl_register(void)
718 {
719 	ip6_ctl_header = register_net_sysctl_rotable(net_ipv6_ctl_path,
720 			ip6_frags_ctl_table);
721 	return ip6_ctl_header == NULL ? -ENOMEM : 0;
722 }
723 
724 static void ip6_frags_sysctl_unregister(void)
725 {
726 	unregister_net_sysctl_table(ip6_ctl_header);
727 }
728 #else
729 static inline int ip6_frags_ns_sysctl_register(struct net *net)
730 {
731 	return 0;
732 }
733 
734 static inline void ip6_frags_ns_sysctl_unregister(struct net *net)
735 {
736 }
737 
738 static inline int ip6_frags_sysctl_register(void)
739 {
740 	return 0;
741 }
742 
743 static inline void ip6_frags_sysctl_unregister(void)
744 {
745 }
746 #endif
747 
748 static int ipv6_frags_init_net(struct net *net)
749 {
750 	net->ipv6.frags.high_thresh = 256 * 1024;
751 	net->ipv6.frags.low_thresh = 192 * 1024;
752 	net->ipv6.frags.timeout = IPV6_FRAG_TIMEOUT;
753 
754 	inet_frags_init_net(&net->ipv6.frags);
755 
756 	return ip6_frags_ns_sysctl_register(net);
757 }
758 
759 static void ipv6_frags_exit_net(struct net *net)
760 {
761 	ip6_frags_ns_sysctl_unregister(net);
762 	inet_frags_exit_net(&net->ipv6.frags, &ip6_frags);
763 }
764 
765 static struct pernet_operations ip6_frags_ops = {
766 	.init = ipv6_frags_init_net,
767 	.exit = ipv6_frags_exit_net,
768 };
769 
770 int __init ipv6_frag_init(void)
771 {
772 	int ret;
773 
774 	ret = inet6_add_protocol(&frag_protocol, IPPROTO_FRAGMENT);
775 	if (ret)
776 		goto out;
777 
778 	ret = ip6_frags_sysctl_register();
779 	if (ret)
780 		goto err_sysctl;
781 
782 	ret = register_pernet_subsys(&ip6_frags_ops);
783 	if (ret)
784 		goto err_pernet;
785 
786 	ip6_frags.hashfn = ip6_hashfn;
787 	ip6_frags.constructor = ip6_frag_init;
788 	ip6_frags.destructor = NULL;
789 	ip6_frags.skb_free = NULL;
790 	ip6_frags.qsize = sizeof(struct frag_queue);
791 	ip6_frags.match = ip6_frag_match;
792 	ip6_frags.frag_expire = ip6_frag_expire;
793 	ip6_frags.secret_interval = 10 * 60 * HZ;
794 	inet_frags_init(&ip6_frags);
795 out:
796 	return ret;
797 
798 err_pernet:
799 	ip6_frags_sysctl_unregister();
800 err_sysctl:
801 	inet6_del_protocol(&frag_protocol, IPPROTO_FRAGMENT);
802 	goto out;
803 }
804 
805 void ipv6_frag_exit(void)
806 {
807 	inet_frags_fini(&ip6_frags);
808 	ip6_frags_sysctl_unregister();
809 	unregister_pernet_subsys(&ip6_frags_ops);
810 	inet6_del_protocol(&frag_protocol, IPPROTO_FRAGMENT);
811 }
812