xref: /openbmc/linux/net/ipv6/reassembly.c (revision b6dcefde)
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 	u32			user;
76 	struct in6_addr		saddr;
77 	struct in6_addr		daddr;
78 
79 	int			iif;
80 	unsigned int		csum;
81 	__u16			nhoffset;
82 };
83 
84 static struct inet_frags ip6_frags;
85 
86 int ip6_frag_nqueues(struct net *net)
87 {
88 	return net->ipv6.frags.nqueues;
89 }
90 
91 int ip6_frag_mem(struct net *net)
92 {
93 	return atomic_read(&net->ipv6.frags.mem);
94 }
95 
96 static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *prev,
97 			  struct net_device *dev);
98 
99 /*
100  * callers should be careful not to use the hash value outside the ipfrag_lock
101  * as doing so could race with ipfrag_hash_rnd being recalculated.
102  */
103 unsigned int inet6_hash_frag(__be32 id, const struct in6_addr *saddr,
104 			     const struct in6_addr *daddr, u32 rnd)
105 {
106 	u32 a, b, c;
107 
108 	a = (__force u32)saddr->s6_addr32[0];
109 	b = (__force u32)saddr->s6_addr32[1];
110 	c = (__force u32)saddr->s6_addr32[2];
111 
112 	a += JHASH_GOLDEN_RATIO;
113 	b += JHASH_GOLDEN_RATIO;
114 	c += rnd;
115 	__jhash_mix(a, b, c);
116 
117 	a += (__force u32)saddr->s6_addr32[3];
118 	b += (__force u32)daddr->s6_addr32[0];
119 	c += (__force u32)daddr->s6_addr32[1];
120 	__jhash_mix(a, b, c);
121 
122 	a += (__force u32)daddr->s6_addr32[2];
123 	b += (__force u32)daddr->s6_addr32[3];
124 	c += (__force u32)id;
125 	__jhash_mix(a, b, c);
126 
127 	return c & (INETFRAGS_HASHSZ - 1);
128 }
129 EXPORT_SYMBOL_GPL(inet6_hash_frag);
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 inet6_hash_frag(fq->id, &fq->saddr, &fq->daddr, ip6_frags.rnd);
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 && fq->user == arg->user &&
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 	fq->user = arg->user;
168 	ipv6_addr_copy(&fq->saddr, arg->src);
169 	ipv6_addr_copy(&fq->daddr, arg->dst);
170 }
171 EXPORT_SYMBOL(ip6_frag_init);
172 
173 /* Destruction primitives. */
174 
175 static __inline__ void fq_put(struct frag_queue *fq)
176 {
177 	inet_frag_put(&fq->q, &ip6_frags);
178 }
179 
180 /* Kill fq entry. It is not destroyed immediately,
181  * because caller (and someone more) holds reference count.
182  */
183 static __inline__ void fq_kill(struct frag_queue *fq)
184 {
185 	inet_frag_kill(&fq->q, &ip6_frags);
186 }
187 
188 static void ip6_evictor(struct net *net, struct inet6_dev *idev)
189 {
190 	int evicted;
191 
192 	evicted = inet_frag_evictor(&net->ipv6.frags, &ip6_frags);
193 	if (evicted)
194 		IP6_ADD_STATS_BH(net, idev, IPSTATS_MIB_REASMFAILS, evicted);
195 }
196 
197 static void ip6_frag_expire(unsigned long data)
198 {
199 	struct frag_queue *fq;
200 	struct net_device *dev = NULL;
201 	struct net *net;
202 
203 	fq = container_of((struct inet_frag_queue *)data, struct frag_queue, q);
204 
205 	spin_lock(&fq->q.lock);
206 
207 	if (fq->q.last_in & INET_FRAG_COMPLETE)
208 		goto out;
209 
210 	fq_kill(fq);
211 
212 	net = container_of(fq->q.net, struct net, ipv6.frags);
213 	rcu_read_lock();
214 	dev = dev_get_by_index_rcu(net, fq->iif);
215 	if (!dev)
216 		goto out_rcu_unlock;
217 
218 	IP6_INC_STATS_BH(net, __in6_dev_get(dev), IPSTATS_MIB_REASMTIMEOUT);
219 	IP6_INC_STATS_BH(net, __in6_dev_get(dev), IPSTATS_MIB_REASMFAILS);
220 
221 	/* Don't send error if the first segment did not arrive. */
222 	if (!(fq->q.last_in & INET_FRAG_FIRST_IN) || !fq->q.fragments)
223 		goto out_rcu_unlock;
224 
225 	/*
226 	   But use as source device on which LAST ARRIVED
227 	   segment was received. And do not use fq->dev
228 	   pointer directly, device might already disappeared.
229 	 */
230 	fq->q.fragments->dev = dev;
231 	icmpv6_send(fq->q.fragments, ICMPV6_TIME_EXCEED, ICMPV6_EXC_FRAGTIME, 0, dev);
232 out_rcu_unlock:
233 	rcu_read_unlock();
234 out:
235 	spin_unlock(&fq->q.lock);
236 	fq_put(fq);
237 }
238 
239 static __inline__ struct frag_queue *
240 fq_find(struct net *net, __be32 id, struct in6_addr *src, struct in6_addr *dst,
241 	struct inet6_dev *idev)
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 		goto oom;
258 
259 	return container_of(q, struct frag_queue, q);
260 
261 oom:
262 	IP6_INC_STATS_BH(net, idev, IPSTATS_MIB_REASMFAILS);
263 	return NULL;
264 }
265 
266 static int ip6_frag_queue(struct frag_queue *fq, struct sk_buff *skb,
267 			   struct frag_hdr *fhdr, int nhoff)
268 {
269 	struct sk_buff *prev, *next;
270 	struct net_device *dev;
271 	int offset, end;
272 	struct net *net = dev_net(skb_dst(skb)->dev);
273 
274 	if (fq->q.last_in & INET_FRAG_COMPLETE)
275 		goto err;
276 
277 	offset = ntohs(fhdr->frag_off) & ~0x7;
278 	end = offset + (ntohs(ipv6_hdr(skb)->payload_len) -
279 			((u8 *)(fhdr + 1) - (u8 *)(ipv6_hdr(skb) + 1)));
280 
281 	if ((unsigned int)end > IPV6_MAXPLEN) {
282 		IP6_INC_STATS_BH(net, ip6_dst_idev(skb_dst(skb)),
283 				 IPSTATS_MIB_INHDRERRORS);
284 		icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
285 				  ((u8 *)&fhdr->frag_off -
286 				   skb_network_header(skb)));
287 		return -1;
288 	}
289 
290 	if (skb->ip_summed == CHECKSUM_COMPLETE) {
291 		const unsigned char *nh = skb_network_header(skb);
292 		skb->csum = csum_sub(skb->csum,
293 				     csum_partial(nh, (u8 *)(fhdr + 1) - nh,
294 						  0));
295 	}
296 
297 	/* Is this the final fragment? */
298 	if (!(fhdr->frag_off & htons(IP6_MF))) {
299 		/* If we already have some bits beyond end
300 		 * or have different end, the segment is corrupted.
301 		 */
302 		if (end < fq->q.len ||
303 		    ((fq->q.last_in & INET_FRAG_LAST_IN) && end != fq->q.len))
304 			goto err;
305 		fq->q.last_in |= INET_FRAG_LAST_IN;
306 		fq->q.len = end;
307 	} else {
308 		/* Check if the fragment is rounded to 8 bytes.
309 		 * Required by the RFC.
310 		 */
311 		if (end & 0x7) {
312 			/* RFC2460 says always send parameter problem in
313 			 * this case. -DaveM
314 			 */
315 			IP6_INC_STATS_BH(net, ip6_dst_idev(skb_dst(skb)),
316 					 IPSTATS_MIB_INHDRERRORS);
317 			icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
318 					  offsetof(struct ipv6hdr, payload_len));
319 			return -1;
320 		}
321 		if (end > fq->q.len) {
322 			/* Some bits beyond end -> corruption. */
323 			if (fq->q.last_in & INET_FRAG_LAST_IN)
324 				goto err;
325 			fq->q.len = end;
326 		}
327 	}
328 
329 	if (end == offset)
330 		goto err;
331 
332 	/* Point into the IP datagram 'data' part. */
333 	if (!pskb_pull(skb, (u8 *) (fhdr + 1) - skb->data))
334 		goto err;
335 
336 	if (pskb_trim_rcsum(skb, end - offset))
337 		goto err;
338 
339 	/* Find out which fragments are in front and at the back of us
340 	 * in the chain of fragments so far.  We must know where to put
341 	 * this fragment, right?
342 	 */
343 	prev = NULL;
344 	for(next = fq->q.fragments; next != NULL; next = next->next) {
345 		if (FRAG6_CB(next)->offset >= offset)
346 			break;	/* bingo! */
347 		prev = next;
348 	}
349 
350 	/* We found where to put this one.  Check for overlap with
351 	 * preceding fragment, and, if needed, align things so that
352 	 * any overlaps are eliminated.
353 	 */
354 	if (prev) {
355 		int i = (FRAG6_CB(prev)->offset + prev->len) - offset;
356 
357 		if (i > 0) {
358 			offset += i;
359 			if (end <= offset)
360 				goto err;
361 			if (!pskb_pull(skb, i))
362 				goto err;
363 			if (skb->ip_summed != CHECKSUM_UNNECESSARY)
364 				skb->ip_summed = CHECKSUM_NONE;
365 		}
366 	}
367 
368 	/* Look for overlap with succeeding segments.
369 	 * If we can merge fragments, do it.
370 	 */
371 	while (next && FRAG6_CB(next)->offset < end) {
372 		int i = end - FRAG6_CB(next)->offset; /* overlap is 'i' bytes */
373 
374 		if (i < next->len) {
375 			/* Eat head of the next overlapped fragment
376 			 * and leave the loop. The next ones cannot overlap.
377 			 */
378 			if (!pskb_pull(next, i))
379 				goto err;
380 			FRAG6_CB(next)->offset += i;	/* next fragment */
381 			fq->q.meat -= i;
382 			if (next->ip_summed != CHECKSUM_UNNECESSARY)
383 				next->ip_summed = CHECKSUM_NONE;
384 			break;
385 		} else {
386 			struct sk_buff *free_it = next;
387 
388 			/* Old fragment is completely overridden with
389 			 * new one drop it.
390 			 */
391 			next = next->next;
392 
393 			if (prev)
394 				prev->next = next;
395 			else
396 				fq->q.fragments = next;
397 
398 			fq->q.meat -= free_it->len;
399 			frag_kfree_skb(fq->q.net, free_it, NULL);
400 		}
401 	}
402 
403 	FRAG6_CB(skb)->offset = offset;
404 
405 	/* Insert this fragment in the chain of fragments. */
406 	skb->next = next;
407 	if (prev)
408 		prev->next = skb;
409 	else
410 		fq->q.fragments = skb;
411 
412 	dev = skb->dev;
413 	if (dev) {
414 		fq->iif = dev->ifindex;
415 		skb->dev = NULL;
416 	}
417 	fq->q.stamp = skb->tstamp;
418 	fq->q.meat += skb->len;
419 	atomic_add(skb->truesize, &fq->q.net->mem);
420 
421 	/* The first fragment.
422 	 * nhoffset is obtained from the first fragment, of course.
423 	 */
424 	if (offset == 0) {
425 		fq->nhoffset = nhoff;
426 		fq->q.last_in |= INET_FRAG_FIRST_IN;
427 	}
428 
429 	if (fq->q.last_in == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
430 	    fq->q.meat == fq->q.len)
431 		return ip6_frag_reasm(fq, prev, dev);
432 
433 	write_lock(&ip6_frags.lock);
434 	list_move_tail(&fq->q.lru_list, &fq->q.net->lru_list);
435 	write_unlock(&ip6_frags.lock);
436 	return -1;
437 
438 err:
439 	IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
440 		      IPSTATS_MIB_REASMFAILS);
441 	kfree_skb(skb);
442 	return -1;
443 }
444 
445 /*
446  *	Check if this packet is complete.
447  *	Returns NULL on failure by any reason, and pointer
448  *	to current nexthdr field in reassembled frame.
449  *
450  *	It is called with locked fq, and caller must check that
451  *	queue is eligible for reassembly i.e. it is not COMPLETE,
452  *	the last and the first frames arrived and all the bits are here.
453  */
454 static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *prev,
455 			  struct net_device *dev)
456 {
457 	struct net *net = container_of(fq->q.net, struct net, ipv6.frags);
458 	struct sk_buff *fp, *head = fq->q.fragments;
459 	int    payload_len;
460 	unsigned int nhoff;
461 
462 	fq_kill(fq);
463 
464 	/* Make the one we just received the head. */
465 	if (prev) {
466 		head = prev->next;
467 		fp = skb_clone(head, GFP_ATOMIC);
468 
469 		if (!fp)
470 			goto out_oom;
471 
472 		fp->next = head->next;
473 		prev->next = fp;
474 
475 		skb_morph(head, fq->q.fragments);
476 		head->next = fq->q.fragments->next;
477 
478 		kfree_skb(fq->q.fragments);
479 		fq->q.fragments = head;
480 	}
481 
482 	WARN_ON(head == NULL);
483 	WARN_ON(FRAG6_CB(head)->offset != 0);
484 
485 	/* Unfragmented part is taken from the first segment. */
486 	payload_len = ((head->data - skb_network_header(head)) -
487 		       sizeof(struct ipv6hdr) + fq->q.len -
488 		       sizeof(struct frag_hdr));
489 	if (payload_len > IPV6_MAXPLEN)
490 		goto out_oversize;
491 
492 	/* Head of list must not be cloned. */
493 	if (skb_cloned(head) && pskb_expand_head(head, 0, 0, GFP_ATOMIC))
494 		goto out_oom;
495 
496 	/* If the first fragment is fragmented itself, we split
497 	 * it to two chunks: the first with data and paged part
498 	 * and the second, holding only fragments. */
499 	if (skb_has_frags(head)) {
500 		struct sk_buff *clone;
501 		int i, plen = 0;
502 
503 		if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL)
504 			goto out_oom;
505 		clone->next = head->next;
506 		head->next = clone;
507 		skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
508 		skb_frag_list_init(head);
509 		for (i=0; i<skb_shinfo(head)->nr_frags; i++)
510 			plen += skb_shinfo(head)->frags[i].size;
511 		clone->len = clone->data_len = head->data_len - plen;
512 		head->data_len -= clone->len;
513 		head->len -= clone->len;
514 		clone->csum = 0;
515 		clone->ip_summed = head->ip_summed;
516 		atomic_add(clone->truesize, &fq->q.net->mem);
517 	}
518 
519 	/* We have to remove fragment header from datagram and to relocate
520 	 * header in order to calculate ICV correctly. */
521 	nhoff = fq->nhoffset;
522 	skb_network_header(head)[nhoff] = skb_transport_header(head)[0];
523 	memmove(head->head + sizeof(struct frag_hdr), head->head,
524 		(head->data - head->head) - sizeof(struct frag_hdr));
525 	head->mac_header += sizeof(struct frag_hdr);
526 	head->network_header += sizeof(struct frag_hdr);
527 
528 	skb_shinfo(head)->frag_list = head->next;
529 	skb_reset_transport_header(head);
530 	skb_push(head, head->data - skb_network_header(head));
531 	atomic_sub(head->truesize, &fq->q.net->mem);
532 
533 	for (fp=head->next; fp; fp = fp->next) {
534 		head->data_len += fp->len;
535 		head->len += fp->len;
536 		if (head->ip_summed != fp->ip_summed)
537 			head->ip_summed = CHECKSUM_NONE;
538 		else if (head->ip_summed == CHECKSUM_COMPLETE)
539 			head->csum = csum_add(head->csum, fp->csum);
540 		head->truesize += fp->truesize;
541 		atomic_sub(fp->truesize, &fq->q.net->mem);
542 	}
543 
544 	head->next = NULL;
545 	head->dev = dev;
546 	head->tstamp = fq->q.stamp;
547 	ipv6_hdr(head)->payload_len = htons(payload_len);
548 	IP6CB(head)->nhoff = nhoff;
549 
550 	/* Yes, and fold redundant checksum back. 8) */
551 	if (head->ip_summed == CHECKSUM_COMPLETE)
552 		head->csum = csum_partial(skb_network_header(head),
553 					  skb_network_header_len(head),
554 					  head->csum);
555 
556 	rcu_read_lock();
557 	IP6_INC_STATS_BH(net, __in6_dev_get(dev), IPSTATS_MIB_REASMOKS);
558 	rcu_read_unlock();
559 	fq->q.fragments = NULL;
560 	return 1;
561 
562 out_oversize:
563 	if (net_ratelimit())
564 		printk(KERN_DEBUG "ip6_frag_reasm: payload len = %d\n", payload_len);
565 	goto out_fail;
566 out_oom:
567 	if (net_ratelimit())
568 		printk(KERN_DEBUG "ip6_frag_reasm: no memory for reassembly\n");
569 out_fail:
570 	rcu_read_lock();
571 	IP6_INC_STATS_BH(net, __in6_dev_get(dev), IPSTATS_MIB_REASMFAILS);
572 	rcu_read_unlock();
573 	return -1;
574 }
575 
576 static int ipv6_frag_rcv(struct sk_buff *skb)
577 {
578 	struct frag_hdr *fhdr;
579 	struct frag_queue *fq;
580 	struct ipv6hdr *hdr = ipv6_hdr(skb);
581 	struct net *net = dev_net(skb_dst(skb)->dev);
582 
583 	IP6_INC_STATS_BH(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMREQDS);
584 
585 	/* Jumbo payload inhibits frag. header */
586 	if (hdr->payload_len==0)
587 		goto fail_hdr;
588 
589 	if (!pskb_may_pull(skb, (skb_transport_offset(skb) +
590 				 sizeof(struct frag_hdr))))
591 		goto fail_hdr;
592 
593 	hdr = ipv6_hdr(skb);
594 	fhdr = (struct frag_hdr *)skb_transport_header(skb);
595 
596 	if (!(fhdr->frag_off & htons(0xFFF9))) {
597 		/* It is not a fragmented frame */
598 		skb->transport_header += sizeof(struct frag_hdr);
599 		IP6_INC_STATS_BH(net,
600 				 ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMOKS);
601 
602 		IP6CB(skb)->nhoff = (u8 *)fhdr - skb_network_header(skb);
603 		return 1;
604 	}
605 
606 	if (atomic_read(&net->ipv6.frags.mem) > net->ipv6.frags.high_thresh)
607 		ip6_evictor(net, ip6_dst_idev(skb_dst(skb)));
608 
609 	if ((fq = fq_find(net, fhdr->identification, &hdr->saddr, &hdr->daddr,
610 			  ip6_dst_idev(skb_dst(skb)))) != NULL) {
611 		int ret;
612 
613 		spin_lock(&fq->q.lock);
614 
615 		ret = ip6_frag_queue(fq, skb, fhdr, IP6CB(skb)->nhoff);
616 
617 		spin_unlock(&fq->q.lock);
618 		fq_put(fq);
619 		return ret;
620 	}
621 
622 	IP6_INC_STATS_BH(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMFAILS);
623 	kfree_skb(skb);
624 	return -1;
625 
626 fail_hdr:
627 	IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_INHDRERRORS);
628 	icmpv6_param_prob(skb, ICMPV6_HDR_FIELD, skb_network_header_len(skb));
629 	return -1;
630 }
631 
632 static const struct inet6_protocol frag_protocol =
633 {
634 	.handler	=	ipv6_frag_rcv,
635 	.flags		=	INET6_PROTO_NOPOLICY,
636 };
637 
638 #ifdef CONFIG_SYSCTL
639 static struct ctl_table ip6_frags_ns_ctl_table[] = {
640 	{
641 		.procname	= "ip6frag_high_thresh",
642 		.data		= &init_net.ipv6.frags.high_thresh,
643 		.maxlen		= sizeof(int),
644 		.mode		= 0644,
645 		.proc_handler	= proc_dointvec
646 	},
647 	{
648 		.procname	= "ip6frag_low_thresh",
649 		.data		= &init_net.ipv6.frags.low_thresh,
650 		.maxlen		= sizeof(int),
651 		.mode		= 0644,
652 		.proc_handler	= proc_dointvec
653 	},
654 	{
655 		.procname	= "ip6frag_time",
656 		.data		= &init_net.ipv6.frags.timeout,
657 		.maxlen		= sizeof(int),
658 		.mode		= 0644,
659 		.proc_handler	= proc_dointvec_jiffies,
660 	},
661 	{ }
662 };
663 
664 static struct ctl_table ip6_frags_ctl_table[] = {
665 	{
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 	},
672 	{ }
673 };
674 
675 static int ip6_frags_ns_sysctl_register(struct net *net)
676 {
677 	struct ctl_table *table;
678 	struct ctl_table_header *hdr;
679 
680 	table = ip6_frags_ns_ctl_table;
681 	if (!net_eq(net, &init_net)) {
682 		table = kmemdup(table, sizeof(ip6_frags_ns_ctl_table), GFP_KERNEL);
683 		if (table == NULL)
684 			goto err_alloc;
685 
686 		table[0].data = &net->ipv6.frags.high_thresh;
687 		table[1].data = &net->ipv6.frags.low_thresh;
688 		table[2].data = &net->ipv6.frags.timeout;
689 	}
690 
691 	hdr = register_net_sysctl_table(net, net_ipv6_ctl_path, table);
692 	if (hdr == NULL)
693 		goto err_reg;
694 
695 	net->ipv6.sysctl.frags_hdr = hdr;
696 	return 0;
697 
698 err_reg:
699 	if (!net_eq(net, &init_net))
700 		kfree(table);
701 err_alloc:
702 	return -ENOMEM;
703 }
704 
705 static void ip6_frags_ns_sysctl_unregister(struct net *net)
706 {
707 	struct ctl_table *table;
708 
709 	table = net->ipv6.sysctl.frags_hdr->ctl_table_arg;
710 	unregister_net_sysctl_table(net->ipv6.sysctl.frags_hdr);
711 	if (!net_eq(net, &init_net))
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