xref: /openbmc/linux/net/ipv6/reassembly.c (revision f42b3800) 
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/errno.h>
32 #include <linux/types.h>
33 #include <linux/string.h>
34 #include <linux/socket.h>
35 #include <linux/sockios.h>
36 #include <linux/jiffies.h>
37 #include <linux/net.h>
38 #include <linux/list.h>
39 #include <linux/netdevice.h>
40 #include <linux/in6.h>
41 #include <linux/ipv6.h>
42 #include <linux/icmpv6.h>
43 #include <linux/random.h>
44 #include <linux/jhash.h>
45 #include <linux/skbuff.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 	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 static unsigned int ip6qhashfn(__be32 id, struct in6_addr *saddr,
105 			       struct in6_addr *daddr)
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 += ip6_frags.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 
131 static unsigned int ip6_hashfn(struct inet_frag_queue *q)
132 {
133 	struct frag_queue *fq;
134 
135 	fq = container_of(q, struct frag_queue, q);
136 	return ip6qhashfn(fq->id, &fq->saddr, &fq->daddr);
137 }
138 
139 int ip6_frag_match(struct inet_frag_queue *q, void *a)
140 {
141 	struct frag_queue *fq;
142 	struct ip6_create_arg *arg = a;
143 
144 	fq = container_of(q, struct frag_queue, q);
145 	return (fq->id == arg->id &&
146 			ipv6_addr_equal(&fq->saddr, arg->src) &&
147 			ipv6_addr_equal(&fq->daddr, arg->dst));
148 }
149 EXPORT_SYMBOL(ip6_frag_match);
150 
151 /* Memory Tracking Functions. */
152 static inline void frag_kfree_skb(struct netns_frags *nf,
153 		struct sk_buff *skb, int *work)
154 {
155 	if (work)
156 		*work -= skb->truesize;
157 	atomic_sub(skb->truesize, &nf->mem);
158 	kfree_skb(skb);
159 }
160 
161 void ip6_frag_init(struct inet_frag_queue *q, void *a)
162 {
163 	struct frag_queue *fq = container_of(q, struct frag_queue, q);
164 	struct ip6_create_arg *arg = a;
165 
166 	fq->id = arg->id;
167 	ipv6_addr_copy(&fq->saddr, arg->src);
168 	ipv6_addr_copy(&fq->daddr, arg->dst);
169 }
170 EXPORT_SYMBOL(ip6_frag_init);
171 
172 /* Destruction primitives. */
173 
174 static __inline__ void fq_put(struct frag_queue *fq)
175 {
176 	inet_frag_put(&fq->q, &ip6_frags);
177 }
178 
179 /* Kill fq entry. It is not destroyed immediately,
180  * because caller (and someone more) holds reference count.
181  */
182 static __inline__ void fq_kill(struct frag_queue *fq)
183 {
184 	inet_frag_kill(&fq->q, &ip6_frags);
185 }
186 
187 static void ip6_evictor(struct net *net, struct inet6_dev *idev)
188 {
189 	int evicted;
190 
191 	evicted = inet_frag_evictor(&net->ipv6.frags, &ip6_frags);
192 	if (evicted)
193 		IP6_ADD_STATS_BH(idev, IPSTATS_MIB_REASMFAILS, evicted);
194 }
195 
196 static void ip6_frag_expire(unsigned long data)
197 {
198 	struct frag_queue *fq;
199 	struct net_device *dev = NULL;
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 	dev = dev_get_by_index(&init_net, fq->iif);
211 	if (!dev)
212 		goto out;
213 
214 	rcu_read_lock();
215 	IP6_INC_STATS_BH(__in6_dev_get(dev), IPSTATS_MIB_REASMTIMEOUT);
216 	IP6_INC_STATS_BH(__in6_dev_get(dev), IPSTATS_MIB_REASMFAILS);
217 	rcu_read_unlock();
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;
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:
231 	if (dev)
232 		dev_put(dev);
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 	hash = ip6qhashfn(id, src, dst);
249 
250 	q = inet_frag_find(&net->ipv6.frags, &ip6_frags, &arg, hash);
251 	if (q == NULL)
252 		goto oom;
253 
254 	return container_of(q, struct frag_queue, q);
255 
256 oom:
257 	IP6_INC_STATS_BH(idev, IPSTATS_MIB_REASMFAILS);
258 	return NULL;
259 }
260 
261 static int ip6_frag_queue(struct frag_queue *fq, struct sk_buff *skb,
262 			   struct frag_hdr *fhdr, int nhoff)
263 {
264 	struct sk_buff *prev, *next;
265 	struct net_device *dev;
266 	int offset, end;
267 
268 	if (fq->q.last_in & INET_FRAG_COMPLETE)
269 		goto err;
270 
271 	offset = ntohs(fhdr->frag_off) & ~0x7;
272 	end = offset + (ntohs(ipv6_hdr(skb)->payload_len) -
273 			((u8 *)(fhdr + 1) - (u8 *)(ipv6_hdr(skb) + 1)));
274 
275 	if ((unsigned int)end > IPV6_MAXPLEN) {
276 		IP6_INC_STATS_BH(ip6_dst_idev(skb->dst),
277 				 IPSTATS_MIB_INHDRERRORS);
278 		icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
279 				  ((u8 *)&fhdr->frag_off -
280 				   skb_network_header(skb)));
281 		return -1;
282 	}
283 
284 	if (skb->ip_summed == CHECKSUM_COMPLETE) {
285 		const unsigned char *nh = skb_network_header(skb);
286 		skb->csum = csum_sub(skb->csum,
287 				     csum_partial(nh, (u8 *)(fhdr + 1) - nh,
288 						  0));
289 	}
290 
291 	/* Is this the final fragment? */
292 	if (!(fhdr->frag_off & htons(IP6_MF))) {
293 		/* If we already have some bits beyond end
294 		 * or have different end, the segment is corrupted.
295 		 */
296 		if (end < fq->q.len ||
297 		    ((fq->q.last_in & INET_FRAG_LAST_IN) && end != fq->q.len))
298 			goto err;
299 		fq->q.last_in |= INET_FRAG_LAST_IN;
300 		fq->q.len = end;
301 	} else {
302 		/* Check if the fragment is rounded to 8 bytes.
303 		 * Required by the RFC.
304 		 */
305 		if (end & 0x7) {
306 			/* RFC2460 says always send parameter problem in
307 			 * this case. -DaveM
308 			 */
309 			IP6_INC_STATS_BH(ip6_dst_idev(skb->dst),
310 					 IPSTATS_MIB_INHDRERRORS);
311 			icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
312 					  offsetof(struct ipv6hdr, payload_len));
313 			return -1;
314 		}
315 		if (end > fq->q.len) {
316 			/* Some bits beyond end -> corruption. */
317 			if (fq->q.last_in & INET_FRAG_LAST_IN)
318 				goto err;
319 			fq->q.len = end;
320 		}
321 	}
322 
323 	if (end == offset)
324 		goto err;
325 
326 	/* Point into the IP datagram 'data' part. */
327 	if (!pskb_pull(skb, (u8 *) (fhdr + 1) - skb->data))
328 		goto err;
329 
330 	if (pskb_trim_rcsum(skb, end - offset))
331 		goto err;
332 
333 	/* Find out which fragments are in front and at the back of us
334 	 * in the chain of fragments so far.  We must know where to put
335 	 * this fragment, right?
336 	 */
337 	prev = NULL;
338 	for(next = fq->q.fragments; next != NULL; next = next->next) {
339 		if (FRAG6_CB(next)->offset >= offset)
340 			break;	/* bingo! */
341 		prev = next;
342 	}
343 
344 	/* We found where to put this one.  Check for overlap with
345 	 * preceding fragment, and, if needed, align things so that
346 	 * any overlaps are eliminated.
347 	 */
348 	if (prev) {
349 		int i = (FRAG6_CB(prev)->offset + prev->len) - offset;
350 
351 		if (i > 0) {
352 			offset += i;
353 			if (end <= offset)
354 				goto err;
355 			if (!pskb_pull(skb, i))
356 				goto err;
357 			if (skb->ip_summed != CHECKSUM_UNNECESSARY)
358 				skb->ip_summed = CHECKSUM_NONE;
359 		}
360 	}
361 
362 	/* Look for overlap with succeeding segments.
363 	 * If we can merge fragments, do it.
364 	 */
365 	while (next && FRAG6_CB(next)->offset < end) {
366 		int i = end - FRAG6_CB(next)->offset; /* overlap is 'i' bytes */
367 
368 		if (i < next->len) {
369 			/* Eat head of the next overlapped fragment
370 			 * and leave the loop. The next ones cannot overlap.
371 			 */
372 			if (!pskb_pull(next, i))
373 				goto err;
374 			FRAG6_CB(next)->offset += i;	/* next fragment */
375 			fq->q.meat -= i;
376 			if (next->ip_summed != CHECKSUM_UNNECESSARY)
377 				next->ip_summed = CHECKSUM_NONE;
378 			break;
379 		} else {
380 			struct sk_buff *free_it = next;
381 
382 			/* Old fragment is completely overridden with
383 			 * new one drop it.
384 			 */
385 			next = next->next;
386 
387 			if (prev)
388 				prev->next = next;
389 			else
390 				fq->q.fragments = next;
391 
392 			fq->q.meat -= free_it->len;
393 			frag_kfree_skb(fq->q.net, free_it, NULL);
394 		}
395 	}
396 
397 	FRAG6_CB(skb)->offset = offset;
398 
399 	/* Insert this fragment in the chain of fragments. */
400 	skb->next = next;
401 	if (prev)
402 		prev->next = skb;
403 	else
404 		fq->q.fragments = skb;
405 
406 	dev = skb->dev;
407 	if (dev) {
408 		fq->iif = dev->ifindex;
409 		skb->dev = NULL;
410 	}
411 	fq->q.stamp = skb->tstamp;
412 	fq->q.meat += skb->len;
413 	atomic_add(skb->truesize, &fq->q.net->mem);
414 
415 	/* The first fragment.
416 	 * nhoffset is obtained from the first fragment, of course.
417 	 */
418 	if (offset == 0) {
419 		fq->nhoffset = nhoff;
420 		fq->q.last_in |= INET_FRAG_FIRST_IN;
421 	}
422 
423 	if (fq->q.last_in == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
424 	    fq->q.meat == fq->q.len)
425 		return ip6_frag_reasm(fq, prev, dev);
426 
427 	write_lock(&ip6_frags.lock);
428 	list_move_tail(&fq->q.lru_list, &fq->q.net->lru_list);
429 	write_unlock(&ip6_frags.lock);
430 	return -1;
431 
432 err:
433 	IP6_INC_STATS(ip6_dst_idev(skb->dst), IPSTATS_MIB_REASMFAILS);
434 	kfree_skb(skb);
435 	return -1;
436 }
437 
438 /*
439  *	Check if this packet is complete.
440  *	Returns NULL on failure by any reason, and pointer
441  *	to current nexthdr field in reassembled frame.
442  *
443  *	It is called with locked fq, and caller must check that
444  *	queue is eligible for reassembly i.e. it is not COMPLETE,
445  *	the last and the first frames arrived and all the bits are here.
446  */
447 static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *prev,
448 			  struct net_device *dev)
449 {
450 	struct sk_buff *fp, *head = fq->q.fragments;
451 	int    payload_len;
452 	unsigned int nhoff;
453 
454 	fq_kill(fq);
455 
456 	/* Make the one we just received the head. */
457 	if (prev) {
458 		head = prev->next;
459 		fp = skb_clone(head, GFP_ATOMIC);
460 
461 		if (!fp)
462 			goto out_oom;
463 
464 		fp->next = head->next;
465 		prev->next = fp;
466 
467 		skb_morph(head, fq->q.fragments);
468 		head->next = fq->q.fragments->next;
469 
470 		kfree_skb(fq->q.fragments);
471 		fq->q.fragments = head;
472 	}
473 
474 	BUG_TRAP(head != NULL);
475 	BUG_TRAP(FRAG6_CB(head)->offset == 0);
476 
477 	/* Unfragmented part is taken from the first segment. */
478 	payload_len = ((head->data - skb_network_header(head)) -
479 		       sizeof(struct ipv6hdr) + fq->q.len -
480 		       sizeof(struct frag_hdr));
481 	if (payload_len > IPV6_MAXPLEN)
482 		goto out_oversize;
483 
484 	/* Head of list must not be cloned. */
485 	if (skb_cloned(head) && pskb_expand_head(head, 0, 0, GFP_ATOMIC))
486 		goto out_oom;
487 
488 	/* If the first fragment is fragmented itself, we split
489 	 * it to two chunks: the first with data and paged part
490 	 * and the second, holding only fragments. */
491 	if (skb_shinfo(head)->frag_list) {
492 		struct sk_buff *clone;
493 		int i, plen = 0;
494 
495 		if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL)
496 			goto out_oom;
497 		clone->next = head->next;
498 		head->next = clone;
499 		skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
500 		skb_shinfo(head)->frag_list = NULL;
501 		for (i=0; i<skb_shinfo(head)->nr_frags; i++)
502 			plen += skb_shinfo(head)->frags[i].size;
503 		clone->len = clone->data_len = head->data_len - plen;
504 		head->data_len -= clone->len;
505 		head->len -= clone->len;
506 		clone->csum = 0;
507 		clone->ip_summed = head->ip_summed;
508 		atomic_add(clone->truesize, &fq->q.net->mem);
509 	}
510 
511 	/* We have to remove fragment header from datagram and to relocate
512 	 * header in order to calculate ICV correctly. */
513 	nhoff = fq->nhoffset;
514 	skb_network_header(head)[nhoff] = skb_transport_header(head)[0];
515 	memmove(head->head + sizeof(struct frag_hdr), head->head,
516 		(head->data - head->head) - sizeof(struct frag_hdr));
517 	head->mac_header += sizeof(struct frag_hdr);
518 	head->network_header += sizeof(struct frag_hdr);
519 
520 	skb_shinfo(head)->frag_list = head->next;
521 	skb_reset_transport_header(head);
522 	skb_push(head, head->data - skb_network_header(head));
523 	atomic_sub(head->truesize, &fq->q.net->mem);
524 
525 	for (fp=head->next; fp; fp = fp->next) {
526 		head->data_len += fp->len;
527 		head->len += fp->len;
528 		if (head->ip_summed != fp->ip_summed)
529 			head->ip_summed = CHECKSUM_NONE;
530 		else if (head->ip_summed == CHECKSUM_COMPLETE)
531 			head->csum = csum_add(head->csum, fp->csum);
532 		head->truesize += fp->truesize;
533 		atomic_sub(fp->truesize, &fq->q.net->mem);
534 	}
535 
536 	head->next = NULL;
537 	head->dev = dev;
538 	head->tstamp = fq->q.stamp;
539 	ipv6_hdr(head)->payload_len = htons(payload_len);
540 	IP6CB(head)->nhoff = nhoff;
541 
542 	/* Yes, and fold redundant checksum back. 8) */
543 	if (head->ip_summed == CHECKSUM_COMPLETE)
544 		head->csum = csum_partial(skb_network_header(head),
545 					  skb_network_header_len(head),
546 					  head->csum);
547 
548 	rcu_read_lock();
549 	IP6_INC_STATS_BH(__in6_dev_get(dev), IPSTATS_MIB_REASMOKS);
550 	rcu_read_unlock();
551 	fq->q.fragments = NULL;
552 	return 1;
553 
554 out_oversize:
555 	if (net_ratelimit())
556 		printk(KERN_DEBUG "ip6_frag_reasm: payload len = %d\n", payload_len);
557 	goto out_fail;
558 out_oom:
559 	if (net_ratelimit())
560 		printk(KERN_DEBUG "ip6_frag_reasm: no memory for reassembly\n");
561 out_fail:
562 	rcu_read_lock();
563 	IP6_INC_STATS_BH(__in6_dev_get(dev), IPSTATS_MIB_REASMFAILS);
564 	rcu_read_unlock();
565 	return -1;
566 }
567 
568 static int ipv6_frag_rcv(struct sk_buff *skb)
569 {
570 	struct frag_hdr *fhdr;
571 	struct frag_queue *fq;
572 	struct ipv6hdr *hdr = ipv6_hdr(skb);
573 	struct net *net;
574 
575 	IP6_INC_STATS_BH(ip6_dst_idev(skb->dst), IPSTATS_MIB_REASMREQDS);
576 
577 	/* Jumbo payload inhibits frag. header */
578 	if (hdr->payload_len==0) {
579 		IP6_INC_STATS(ip6_dst_idev(skb->dst), IPSTATS_MIB_INHDRERRORS);
580 		icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
581 				  skb_network_header_len(skb));
582 		return -1;
583 	}
584 	if (!pskb_may_pull(skb, (skb_transport_offset(skb) +
585 				 sizeof(struct frag_hdr)))) {
586 		IP6_INC_STATS(ip6_dst_idev(skb->dst), IPSTATS_MIB_INHDRERRORS);
587 		icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
588 				  skb_network_header_len(skb));
589 		return -1;
590 	}
591 
592 	hdr = ipv6_hdr(skb);
593 	fhdr = (struct frag_hdr *)skb_transport_header(skb);
594 
595 	if (!(fhdr->frag_off & htons(0xFFF9))) {
596 		/* It is not a fragmented frame */
597 		skb->transport_header += sizeof(struct frag_hdr);
598 		IP6_INC_STATS_BH(ip6_dst_idev(skb->dst), IPSTATS_MIB_REASMOKS);
599 
600 		IP6CB(skb)->nhoff = (u8 *)fhdr - skb_network_header(skb);
601 		return 1;
602 	}
603 
604 	net = dev_net(skb->dev);
605 	if (atomic_read(&net->ipv6.frags.mem) > net->ipv6.frags.high_thresh)
606 		ip6_evictor(net, ip6_dst_idev(skb->dst));
607 
608 	if ((fq = fq_find(net, fhdr->identification, &hdr->saddr, &hdr->daddr,
609 			  ip6_dst_idev(skb->dst))) != NULL) {
610 		int ret;
611 
612 		spin_lock(&fq->q.lock);
613 
614 		ret = ip6_frag_queue(fq, skb, fhdr, IP6CB(skb)->nhoff);
615 
616 		spin_unlock(&fq->q.lock);
617 		fq_put(fq);
618 		return ret;
619 	}
620 
621 	IP6_INC_STATS_BH(ip6_dst_idev(skb->dst), IPSTATS_MIB_REASMFAILS);
622 	kfree_skb(skb);
623 	return -1;
624 }
625 
626 static struct inet6_protocol frag_protocol =
627 {
628 	.handler	=	ipv6_frag_rcv,
629 	.flags		=	INET6_PROTO_NOPOLICY,
630 };
631 
632 #ifdef CONFIG_SYSCTL
633 static struct ctl_table ip6_frags_ctl_table[] = {
634 	{
635 		.ctl_name	= NET_IPV6_IP6FRAG_HIGH_THRESH,
636 		.procname	= "ip6frag_high_thresh",
637 		.data		= &init_net.ipv6.frags.high_thresh,
638 		.maxlen		= sizeof(int),
639 		.mode		= 0644,
640 		.proc_handler	= &proc_dointvec
641 	},
642 	{
643 		.ctl_name	= NET_IPV6_IP6FRAG_LOW_THRESH,
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 		.ctl_name	= NET_IPV6_IP6FRAG_TIME,
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 		.strategy	= &sysctl_jiffies,
658 	},
659 	{
660 		.ctl_name	= NET_IPV6_IP6FRAG_SECRET_INTERVAL,
661 		.procname	= "ip6frag_secret_interval",
662 		.data		= &ip6_frags.secret_interval,
663 		.maxlen		= sizeof(int),
664 		.mode		= 0644,
665 		.proc_handler	= &proc_dointvec_jiffies,
666 		.strategy	= &sysctl_jiffies
667 	},
668 	{ }
669 };
670 
671 static int ip6_frags_sysctl_register(struct net *net)
672 {
673 	struct ctl_table *table;
674 	struct ctl_table_header *hdr;
675 
676 	table = ip6_frags_ctl_table;
677 	if (net != &init_net) {
678 		table = kmemdup(table, sizeof(ip6_frags_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 		table[3].mode &= ~0222;
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 != &init_net)
697 		kfree(table);
698 err_alloc:
699 	return -ENOMEM;
700 }
701 
702 static void ip6_frags_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 #else
711 static inline int ip6_frags_sysctl_register(struct net *net)
712 {
713 	return 0;
714 }
715 
716 static inline void ip6_frags_sysctl_unregister(struct net *net)
717 {
718 }
719 #endif
720 
721 static int ipv6_frags_init_net(struct net *net)
722 {
723 	net->ipv6.frags.high_thresh = 256 * 1024;
724 	net->ipv6.frags.low_thresh = 192 * 1024;
725 	net->ipv6.frags.timeout = IPV6_FRAG_TIMEOUT;
726 
727 	inet_frags_init_net(&net->ipv6.frags);
728 
729 	return ip6_frags_sysctl_register(net);
730 }
731 
732 static void ipv6_frags_exit_net(struct net *net)
733 {
734 	ip6_frags_sysctl_unregister(net);
735 	inet_frags_exit_net(&net->ipv6.frags, &ip6_frags);
736 }
737 
738 static struct pernet_operations ip6_frags_ops = {
739 	.init = ipv6_frags_init_net,
740 	.exit = ipv6_frags_exit_net,
741 };
742 
743 int __init ipv6_frag_init(void)
744 {
745 	int ret;
746 
747 	ret = inet6_add_protocol(&frag_protocol, IPPROTO_FRAGMENT);
748 	if (ret)
749 		goto out;
750 
751 	register_pernet_subsys(&ip6_frags_ops);
752 
753 	ip6_frags.hashfn = ip6_hashfn;
754 	ip6_frags.constructor = ip6_frag_init;
755 	ip6_frags.destructor = NULL;
756 	ip6_frags.skb_free = NULL;
757 	ip6_frags.qsize = sizeof(struct frag_queue);
758 	ip6_frags.match = ip6_frag_match;
759 	ip6_frags.frag_expire = ip6_frag_expire;
760 	ip6_frags.secret_interval = 10 * 60 * HZ;
761 	inet_frags_init(&ip6_frags);
762 out:
763 	return ret;
764 }
765 
766 void ipv6_frag_exit(void)
767 {
768 	inet_frags_fini(&ip6_frags);
769 	unregister_pernet_subsys(&ip6_frags_ops);
770 	inet6_del_protocol(&frag_protocol, IPPROTO_FRAGMENT);
771 }
772