xref: /openbmc/linux/net/ipv6/reassembly.c (revision b627b4ed) 
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(net, 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(net, __in6_dev_get(dev), IPSTATS_MIB_REASMTIMEOUT);
217 	IP6_INC_STATS_BH(net, __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(net, 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 	struct net *net = dev_net(skb->dst->dev);
271 
272 	if (fq->q.last_in & INET_FRAG_COMPLETE)
273 		goto err;
274 
275 	offset = ntohs(fhdr->frag_off) & ~0x7;
276 	end = offset + (ntohs(ipv6_hdr(skb)->payload_len) -
277 			((u8 *)(fhdr + 1) - (u8 *)(ipv6_hdr(skb) + 1)));
278 
279 	if ((unsigned int)end > IPV6_MAXPLEN) {
280 		IP6_INC_STATS_BH(net, ip6_dst_idev(skb->dst),
281 				 IPSTATS_MIB_INHDRERRORS);
282 		icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
283 				  ((u8 *)&fhdr->frag_off -
284 				   skb_network_header(skb)));
285 		return -1;
286 	}
287 
288 	if (skb->ip_summed == CHECKSUM_COMPLETE) {
289 		const unsigned char *nh = skb_network_header(skb);
290 		skb->csum = csum_sub(skb->csum,
291 				     csum_partial(nh, (u8 *)(fhdr + 1) - nh,
292 						  0));
293 	}
294 
295 	/* Is this the final fragment? */
296 	if (!(fhdr->frag_off & htons(IP6_MF))) {
297 		/* If we already have some bits beyond end
298 		 * or have different end, the segment is corrupted.
299 		 */
300 		if (end < fq->q.len ||
301 		    ((fq->q.last_in & INET_FRAG_LAST_IN) && end != fq->q.len))
302 			goto err;
303 		fq->q.last_in |= INET_FRAG_LAST_IN;
304 		fq->q.len = end;
305 	} else {
306 		/* Check if the fragment is rounded to 8 bytes.
307 		 * Required by the RFC.
308 		 */
309 		if (end & 0x7) {
310 			/* RFC2460 says always send parameter problem in
311 			 * this case. -DaveM
312 			 */
313 			IP6_INC_STATS_BH(net, ip6_dst_idev(skb->dst),
314 					 IPSTATS_MIB_INHDRERRORS);
315 			icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
316 					  offsetof(struct ipv6hdr, payload_len));
317 			return -1;
318 		}
319 		if (end > fq->q.len) {
320 			/* Some bits beyond end -> corruption. */
321 			if (fq->q.last_in & INET_FRAG_LAST_IN)
322 				goto err;
323 			fq->q.len = end;
324 		}
325 	}
326 
327 	if (end == offset)
328 		goto err;
329 
330 	/* Point into the IP datagram 'data' part. */
331 	if (!pskb_pull(skb, (u8 *) (fhdr + 1) - skb->data))
332 		goto err;
333 
334 	if (pskb_trim_rcsum(skb, end - offset))
335 		goto err;
336 
337 	/* Find out which fragments are in front and at the back of us
338 	 * in the chain of fragments so far.  We must know where to put
339 	 * this fragment, right?
340 	 */
341 	prev = NULL;
342 	for(next = fq->q.fragments; next != NULL; next = next->next) {
343 		if (FRAG6_CB(next)->offset >= offset)
344 			break;	/* bingo! */
345 		prev = next;
346 	}
347 
348 	/* We found where to put this one.  Check for overlap with
349 	 * preceding fragment, and, if needed, align things so that
350 	 * any overlaps are eliminated.
351 	 */
352 	if (prev) {
353 		int i = (FRAG6_CB(prev)->offset + prev->len) - offset;
354 
355 		if (i > 0) {
356 			offset += i;
357 			if (end <= offset)
358 				goto err;
359 			if (!pskb_pull(skb, i))
360 				goto err;
361 			if (skb->ip_summed != CHECKSUM_UNNECESSARY)
362 				skb->ip_summed = CHECKSUM_NONE;
363 		}
364 	}
365 
366 	/* Look for overlap with succeeding segments.
367 	 * If we can merge fragments, do it.
368 	 */
369 	while (next && FRAG6_CB(next)->offset < end) {
370 		int i = end - FRAG6_CB(next)->offset; /* overlap is 'i' bytes */
371 
372 		if (i < next->len) {
373 			/* Eat head of the next overlapped fragment
374 			 * and leave the loop. The next ones cannot overlap.
375 			 */
376 			if (!pskb_pull(next, i))
377 				goto err;
378 			FRAG6_CB(next)->offset += i;	/* next fragment */
379 			fq->q.meat -= i;
380 			if (next->ip_summed != CHECKSUM_UNNECESSARY)
381 				next->ip_summed = CHECKSUM_NONE;
382 			break;
383 		} else {
384 			struct sk_buff *free_it = next;
385 
386 			/* Old fragment is completely overridden with
387 			 * new one drop it.
388 			 */
389 			next = next->next;
390 
391 			if (prev)
392 				prev->next = next;
393 			else
394 				fq->q.fragments = next;
395 
396 			fq->q.meat -= free_it->len;
397 			frag_kfree_skb(fq->q.net, free_it, NULL);
398 		}
399 	}
400 
401 	FRAG6_CB(skb)->offset = offset;
402 
403 	/* Insert this fragment in the chain of fragments. */
404 	skb->next = next;
405 	if (prev)
406 		prev->next = skb;
407 	else
408 		fq->q.fragments = skb;
409 
410 	dev = skb->dev;
411 	if (dev) {
412 		fq->iif = dev->ifindex;
413 		skb->dev = NULL;
414 	}
415 	fq->q.stamp = skb->tstamp;
416 	fq->q.meat += skb->len;
417 	atomic_add(skb->truesize, &fq->q.net->mem);
418 
419 	/* The first fragment.
420 	 * nhoffset is obtained from the first fragment, of course.
421 	 */
422 	if (offset == 0) {
423 		fq->nhoffset = nhoff;
424 		fq->q.last_in |= INET_FRAG_FIRST_IN;
425 	}
426 
427 	if (fq->q.last_in == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
428 	    fq->q.meat == fq->q.len)
429 		return ip6_frag_reasm(fq, prev, dev);
430 
431 	write_lock(&ip6_frags.lock);
432 	list_move_tail(&fq->q.lru_list, &fq->q.net->lru_list);
433 	write_unlock(&ip6_frags.lock);
434 	return -1;
435 
436 err:
437 	IP6_INC_STATS(net, ip6_dst_idev(skb->dst),
438 		      IPSTATS_MIB_REASMFAILS);
439 	kfree_skb(skb);
440 	return -1;
441 }
442 
443 /*
444  *	Check if this packet is complete.
445  *	Returns NULL on failure by any reason, and pointer
446  *	to current nexthdr field in reassembled frame.
447  *
448  *	It is called with locked fq, and caller must check that
449  *	queue is eligible for reassembly i.e. it is not COMPLETE,
450  *	the last and the first frames arrived and all the bits are here.
451  */
452 static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *prev,
453 			  struct net_device *dev)
454 {
455 	struct net *net = container_of(fq->q.net, struct net, ipv6.frags);
456 	struct sk_buff *fp, *head = fq->q.fragments;
457 	int    payload_len;
458 	unsigned int nhoff;
459 
460 	fq_kill(fq);
461 
462 	/* Make the one we just received the head. */
463 	if (prev) {
464 		head = prev->next;
465 		fp = skb_clone(head, GFP_ATOMIC);
466 
467 		if (!fp)
468 			goto out_oom;
469 
470 		fp->next = head->next;
471 		prev->next = fp;
472 
473 		skb_morph(head, fq->q.fragments);
474 		head->next = fq->q.fragments->next;
475 
476 		kfree_skb(fq->q.fragments);
477 		fq->q.fragments = head;
478 	}
479 
480 	WARN_ON(head == NULL);
481 	WARN_ON(FRAG6_CB(head)->offset != 0);
482 
483 	/* Unfragmented part is taken from the first segment. */
484 	payload_len = ((head->data - skb_network_header(head)) -
485 		       sizeof(struct ipv6hdr) + fq->q.len -
486 		       sizeof(struct frag_hdr));
487 	if (payload_len > IPV6_MAXPLEN)
488 		goto out_oversize;
489 
490 	/* Head of list must not be cloned. */
491 	if (skb_cloned(head) && pskb_expand_head(head, 0, 0, GFP_ATOMIC))
492 		goto out_oom;
493 
494 	/* If the first fragment is fragmented itself, we split
495 	 * it to two chunks: the first with data and paged part
496 	 * and the second, holding only fragments. */
497 	if (skb_shinfo(head)->frag_list) {
498 		struct sk_buff *clone;
499 		int i, plen = 0;
500 
501 		if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL)
502 			goto out_oom;
503 		clone->next = head->next;
504 		head->next = clone;
505 		skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
506 		skb_shinfo(head)->frag_list = NULL;
507 		for (i=0; i<skb_shinfo(head)->nr_frags; i++)
508 			plen += skb_shinfo(head)->frags[i].size;
509 		clone->len = clone->data_len = head->data_len - plen;
510 		head->data_len -= clone->len;
511 		head->len -= clone->len;
512 		clone->csum = 0;
513 		clone->ip_summed = head->ip_summed;
514 		atomic_add(clone->truesize, &fq->q.net->mem);
515 	}
516 
517 	/* We have to remove fragment header from datagram and to relocate
518 	 * header in order to calculate ICV correctly. */
519 	nhoff = fq->nhoffset;
520 	skb_network_header(head)[nhoff] = skb_transport_header(head)[0];
521 	memmove(head->head + sizeof(struct frag_hdr), head->head,
522 		(head->data - head->head) - sizeof(struct frag_hdr));
523 	head->mac_header += sizeof(struct frag_hdr);
524 	head->network_header += sizeof(struct frag_hdr);
525 
526 	skb_shinfo(head)->frag_list = head->next;
527 	skb_reset_transport_header(head);
528 	skb_push(head, head->data - skb_network_header(head));
529 	atomic_sub(head->truesize, &fq->q.net->mem);
530 
531 	for (fp=head->next; fp; fp = fp->next) {
532 		head->data_len += fp->len;
533 		head->len += fp->len;
534 		if (head->ip_summed != fp->ip_summed)
535 			head->ip_summed = CHECKSUM_NONE;
536 		else if (head->ip_summed == CHECKSUM_COMPLETE)
537 			head->csum = csum_add(head->csum, fp->csum);
538 		head->truesize += fp->truesize;
539 		atomic_sub(fp->truesize, &fq->q.net->mem);
540 	}
541 
542 	head->next = NULL;
543 	head->dev = dev;
544 	head->tstamp = fq->q.stamp;
545 	ipv6_hdr(head)->payload_len = htons(payload_len);
546 	IP6CB(head)->nhoff = nhoff;
547 
548 	/* Yes, and fold redundant checksum back. 8) */
549 	if (head->ip_summed == CHECKSUM_COMPLETE)
550 		head->csum = csum_partial(skb_network_header(head),
551 					  skb_network_header_len(head),
552 					  head->csum);
553 
554 	rcu_read_lock();
555 	IP6_INC_STATS_BH(net, __in6_dev_get(dev), IPSTATS_MIB_REASMOKS);
556 	rcu_read_unlock();
557 	fq->q.fragments = NULL;
558 	return 1;
559 
560 out_oversize:
561 	if (net_ratelimit())
562 		printk(KERN_DEBUG "ip6_frag_reasm: payload len = %d\n", payload_len);
563 	goto out_fail;
564 out_oom:
565 	if (net_ratelimit())
566 		printk(KERN_DEBUG "ip6_frag_reasm: no memory for reassembly\n");
567 out_fail:
568 	rcu_read_lock();
569 	IP6_INC_STATS_BH(net, __in6_dev_get(dev), IPSTATS_MIB_REASMFAILS);
570 	rcu_read_unlock();
571 	return -1;
572 }
573 
574 static int ipv6_frag_rcv(struct sk_buff *skb)
575 {
576 	struct frag_hdr *fhdr;
577 	struct frag_queue *fq;
578 	struct ipv6hdr *hdr = ipv6_hdr(skb);
579 	struct net *net = dev_net(skb->dst->dev);
580 
581 	IP6_INC_STATS_BH(net, ip6_dst_idev(skb->dst), IPSTATS_MIB_REASMREQDS);
582 
583 	/* Jumbo payload inhibits frag. header */
584 	if (hdr->payload_len==0)
585 		goto fail_hdr;
586 
587 	if (!pskb_may_pull(skb, (skb_transport_offset(skb) +
588 				 sizeof(struct frag_hdr))))
589 		goto fail_hdr;
590 
591 	hdr = ipv6_hdr(skb);
592 	fhdr = (struct frag_hdr *)skb_transport_header(skb);
593 
594 	if (!(fhdr->frag_off & htons(0xFFF9))) {
595 		/* It is not a fragmented frame */
596 		skb->transport_header += sizeof(struct frag_hdr);
597 		IP6_INC_STATS_BH(net,
598 				 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 	if (atomic_read(&net->ipv6.frags.mem) > net->ipv6.frags.high_thresh)
605 		ip6_evictor(net, ip6_dst_idev(skb->dst));
606 
607 	if ((fq = fq_find(net, fhdr->identification, &hdr->saddr, &hdr->daddr,
608 			  ip6_dst_idev(skb->dst))) != NULL) {
609 		int ret;
610 
611 		spin_lock(&fq->q.lock);
612 
613 		ret = ip6_frag_queue(fq, skb, fhdr, IP6CB(skb)->nhoff);
614 
615 		spin_unlock(&fq->q.lock);
616 		fq_put(fq);
617 		return ret;
618 	}
619 
620 	IP6_INC_STATS_BH(net, ip6_dst_idev(skb->dst), IPSTATS_MIB_REASMFAILS);
621 	kfree_skb(skb);
622 	return -1;
623 
624 fail_hdr:
625 	IP6_INC_STATS(net, ip6_dst_idev(skb->dst), IPSTATS_MIB_INHDRERRORS);
626 	icmpv6_param_prob(skb, ICMPV6_HDR_FIELD, skb_network_header_len(skb));
627 	return -1;
628 }
629 
630 static struct inet6_protocol frag_protocol =
631 {
632 	.handler	=	ipv6_frag_rcv,
633 	.flags		=	INET6_PROTO_NOPOLICY,
634 };
635 
636 #ifdef CONFIG_SYSCTL
637 static struct ctl_table ip6_frags_ns_ctl_table[] = {
638 	{
639 		.ctl_name	= NET_IPV6_IP6FRAG_HIGH_THRESH,
640 		.procname	= "ip6frag_high_thresh",
641 		.data		= &init_net.ipv6.frags.high_thresh,
642 		.maxlen		= sizeof(int),
643 		.mode		= 0644,
644 		.proc_handler	= proc_dointvec
645 	},
646 	{
647 		.ctl_name	= NET_IPV6_IP6FRAG_LOW_THRESH,
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 		.ctl_name	= NET_IPV6_IP6FRAG_TIME,
656 		.procname	= "ip6frag_time",
657 		.data		= &init_net.ipv6.frags.timeout,
658 		.maxlen		= sizeof(int),
659 		.mode		= 0644,
660 		.proc_handler	= proc_dointvec_jiffies,
661 		.strategy	= sysctl_jiffies,
662 	},
663 	{ }
664 };
665 
666 static struct ctl_table ip6_frags_ctl_table[] = {
667 	{
668 		.ctl_name	= NET_IPV6_IP6FRAG_SECRET_INTERVAL,
669 		.procname	= "ip6frag_secret_interval",
670 		.data		= &ip6_frags.secret_interval,
671 		.maxlen		= sizeof(int),
672 		.mode		= 0644,
673 		.proc_handler	= proc_dointvec_jiffies,
674 		.strategy	= sysctl_jiffies
675 	},
676 	{ }
677 };
678 
679 static int ip6_frags_ns_sysctl_register(struct net *net)
680 {
681 	struct ctl_table *table;
682 	struct ctl_table_header *hdr;
683 
684 	table = ip6_frags_ns_ctl_table;
685 	if (net != &init_net) {
686 		table = kmemdup(table, sizeof(ip6_frags_ns_ctl_table), GFP_KERNEL);
687 		if (table == NULL)
688 			goto err_alloc;
689 
690 		table[0].data = &net->ipv6.frags.high_thresh;
691 		table[1].data = &net->ipv6.frags.low_thresh;
692 		table[2].data = &net->ipv6.frags.timeout;
693 	}
694 
695 	hdr = register_net_sysctl_table(net, net_ipv6_ctl_path, table);
696 	if (hdr == NULL)
697 		goto err_reg;
698 
699 	net->ipv6.sysctl.frags_hdr = hdr;
700 	return 0;
701 
702 err_reg:
703 	if (net != &init_net)
704 		kfree(table);
705 err_alloc:
706 	return -ENOMEM;
707 }
708 
709 static void ip6_frags_ns_sysctl_unregister(struct net *net)
710 {
711 	struct ctl_table *table;
712 
713 	table = net->ipv6.sysctl.frags_hdr->ctl_table_arg;
714 	unregister_net_sysctl_table(net->ipv6.sysctl.frags_hdr);
715 	kfree(table);
716 }
717 
718 static struct ctl_table_header *ip6_ctl_header;
719 
720 static int ip6_frags_sysctl_register(void)
721 {
722 	ip6_ctl_header = register_net_sysctl_rotable(net_ipv6_ctl_path,
723 			ip6_frags_ctl_table);
724 	return ip6_ctl_header == NULL ? -ENOMEM : 0;
725 }
726 
727 static void ip6_frags_sysctl_unregister(void)
728 {
729 	unregister_net_sysctl_table(ip6_ctl_header);
730 }
731 #else
732 static inline int ip6_frags_ns_sysctl_register(struct net *net)
733 {
734 	return 0;
735 }
736 
737 static inline void ip6_frags_ns_sysctl_unregister(struct net *net)
738 {
739 }
740 
741 static inline int ip6_frags_sysctl_register(void)
742 {
743 	return 0;
744 }
745 
746 static inline void ip6_frags_sysctl_unregister(void)
747 {
748 }
749 #endif
750 
751 static int ipv6_frags_init_net(struct net *net)
752 {
753 	net->ipv6.frags.high_thresh = 256 * 1024;
754 	net->ipv6.frags.low_thresh = 192 * 1024;
755 	net->ipv6.frags.timeout = IPV6_FRAG_TIMEOUT;
756 
757 	inet_frags_init_net(&net->ipv6.frags);
758 
759 	return ip6_frags_ns_sysctl_register(net);
760 }
761 
762 static void ipv6_frags_exit_net(struct net *net)
763 {
764 	ip6_frags_ns_sysctl_unregister(net);
765 	inet_frags_exit_net(&net->ipv6.frags, &ip6_frags);
766 }
767 
768 static struct pernet_operations ip6_frags_ops = {
769 	.init = ipv6_frags_init_net,
770 	.exit = ipv6_frags_exit_net,
771 };
772 
773 int __init ipv6_frag_init(void)
774 {
775 	int ret;
776 
777 	ret = inet6_add_protocol(&frag_protocol, IPPROTO_FRAGMENT);
778 	if (ret)
779 		goto out;
780 
781 	ret = ip6_frags_sysctl_register();
782 	if (ret)
783 		goto err_sysctl;
784 
785 	ret = register_pernet_subsys(&ip6_frags_ops);
786 	if (ret)
787 		goto err_pernet;
788 
789 	ip6_frags.hashfn = ip6_hashfn;
790 	ip6_frags.constructor = ip6_frag_init;
791 	ip6_frags.destructor = NULL;
792 	ip6_frags.skb_free = NULL;
793 	ip6_frags.qsize = sizeof(struct frag_queue);
794 	ip6_frags.match = ip6_frag_match;
795 	ip6_frags.frag_expire = ip6_frag_expire;
796 	ip6_frags.secret_interval = 10 * 60 * HZ;
797 	inet_frags_init(&ip6_frags);
798 out:
799 	return ret;
800 
801 err_pernet:
802 	ip6_frags_sysctl_unregister();
803 err_sysctl:
804 	inet6_del_protocol(&frag_protocol, IPPROTO_FRAGMENT);
805 	goto out;
806 }
807 
808 void ipv6_frag_exit(void)
809 {
810 	inet_frags_fini(&ip6_frags);
811 	ip6_frags_sysctl_unregister();
812 	unregister_pernet_subsys(&ip6_frags_ops);
813 	inet6_del_protocol(&frag_protocol, IPPROTO_FRAGMENT);
814 }
815