xref: /openbmc/linux/net/ipv6/reassembly.c (revision 97599dc7)
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 
30 #define pr_fmt(fmt) "IPv6: " fmt
31 
32 #include <linux/errno.h>
33 #include <linux/types.h>
34 #include <linux/string.h>
35 #include <linux/socket.h>
36 #include <linux/sockios.h>
37 #include <linux/jiffies.h>
38 #include <linux/net.h>
39 #include <linux/list.h>
40 #include <linux/netdevice.h>
41 #include <linux/in6.h>
42 #include <linux/ipv6.h>
43 #include <linux/icmpv6.h>
44 #include <linux/random.h>
45 #include <linux/jhash.h>
46 #include <linux/skbuff.h>
47 #include <linux/slab.h>
48 #include <linux/export.h>
49 
50 #include <net/sock.h>
51 #include <net/snmp.h>
52 
53 #include <net/ipv6.h>
54 #include <net/ip6_route.h>
55 #include <net/protocol.h>
56 #include <net/transp_v6.h>
57 #include <net/rawv6.h>
58 #include <net/ndisc.h>
59 #include <net/addrconf.h>
60 #include <net/inet_frag.h>
61 
62 struct ip6frag_skb_cb
63 {
64 	struct inet6_skb_parm	h;
65 	int			offset;
66 };
67 
68 #define FRAG6_CB(skb)	((struct ip6frag_skb_cb*)((skb)->cb))
69 
70 
71 static struct inet_frags ip6_frags;
72 
73 static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *prev,
74 			  struct net_device *dev);
75 
76 /*
77  * callers should be careful not to use the hash value outside the ipfrag_lock
78  * as doing so could race with ipfrag_hash_rnd being recalculated.
79  */
80 unsigned int inet6_hash_frag(__be32 id, const struct in6_addr *saddr,
81 			     const struct in6_addr *daddr, u32 rnd)
82 {
83 	u32 c;
84 
85 	c = jhash_3words(ipv6_addr_hash(saddr), ipv6_addr_hash(daddr),
86 			 (__force u32)id, rnd);
87 
88 	return c & (INETFRAGS_HASHSZ - 1);
89 }
90 EXPORT_SYMBOL_GPL(inet6_hash_frag);
91 
92 static unsigned int ip6_hashfn(struct inet_frag_queue *q)
93 {
94 	struct frag_queue *fq;
95 
96 	fq = container_of(q, struct frag_queue, q);
97 	return inet6_hash_frag(fq->id, &fq->saddr, &fq->daddr, ip6_frags.rnd);
98 }
99 
100 bool ip6_frag_match(struct inet_frag_queue *q, void *a)
101 {
102 	struct frag_queue *fq;
103 	struct ip6_create_arg *arg = a;
104 
105 	fq = container_of(q, struct frag_queue, q);
106 	return	fq->id == arg->id &&
107 		fq->user == arg->user &&
108 		ipv6_addr_equal(&fq->saddr, arg->src) &&
109 		ipv6_addr_equal(&fq->daddr, arg->dst);
110 }
111 EXPORT_SYMBOL(ip6_frag_match);
112 
113 void ip6_frag_init(struct inet_frag_queue *q, void *a)
114 {
115 	struct frag_queue *fq = container_of(q, struct frag_queue, q);
116 	struct ip6_create_arg *arg = a;
117 
118 	fq->id = arg->id;
119 	fq->user = arg->user;
120 	fq->saddr = *arg->src;
121 	fq->daddr = *arg->dst;
122 }
123 EXPORT_SYMBOL(ip6_frag_init);
124 
125 void ip6_expire_frag_queue(struct net *net, struct frag_queue *fq,
126 			   struct inet_frags *frags)
127 {
128 	struct net_device *dev = NULL;
129 
130 	spin_lock(&fq->q.lock);
131 
132 	if (fq->q.last_in & INET_FRAG_COMPLETE)
133 		goto out;
134 
135 	inet_frag_kill(&fq->q, frags);
136 
137 	rcu_read_lock();
138 	dev = dev_get_by_index_rcu(net, fq->iif);
139 	if (!dev)
140 		goto out_rcu_unlock;
141 
142 	IP6_INC_STATS_BH(net, __in6_dev_get(dev), IPSTATS_MIB_REASMTIMEOUT);
143 	IP6_INC_STATS_BH(net, __in6_dev_get(dev), IPSTATS_MIB_REASMFAILS);
144 
145 	/* Don't send error if the first segment did not arrive. */
146 	if (!(fq->q.last_in & INET_FRAG_FIRST_IN) || !fq->q.fragments)
147 		goto out_rcu_unlock;
148 
149 	/*
150 	   But use as source device on which LAST ARRIVED
151 	   segment was received. And do not use fq->dev
152 	   pointer directly, device might already disappeared.
153 	 */
154 	fq->q.fragments->dev = dev;
155 	icmpv6_send(fq->q.fragments, ICMPV6_TIME_EXCEED, ICMPV6_EXC_FRAGTIME, 0);
156 out_rcu_unlock:
157 	rcu_read_unlock();
158 out:
159 	spin_unlock(&fq->q.lock);
160 	inet_frag_put(&fq->q, frags);
161 }
162 EXPORT_SYMBOL(ip6_expire_frag_queue);
163 
164 static void ip6_frag_expire(unsigned long data)
165 {
166 	struct frag_queue *fq;
167 	struct net *net;
168 
169 	fq = container_of((struct inet_frag_queue *)data, struct frag_queue, q);
170 	net = container_of(fq->q.net, struct net, ipv6.frags);
171 
172 	ip6_expire_frag_queue(net, fq, &ip6_frags);
173 }
174 
175 static __inline__ struct frag_queue *
176 fq_find(struct net *net, __be32 id, const struct in6_addr *src, const struct in6_addr *dst)
177 {
178 	struct inet_frag_queue *q;
179 	struct ip6_create_arg arg;
180 	unsigned int hash;
181 
182 	arg.id = id;
183 	arg.user = IP6_DEFRAG_LOCAL_DELIVER;
184 	arg.src = src;
185 	arg.dst = dst;
186 
187 	read_lock(&ip6_frags.lock);
188 	hash = inet6_hash_frag(id, src, dst, ip6_frags.rnd);
189 
190 	q = inet_frag_find(&net->ipv6.frags, &ip6_frags, &arg, hash);
191 	if (IS_ERR_OR_NULL(q)) {
192 		inet_frag_maybe_warn_overflow(q, pr_fmt());
193 		return NULL;
194 	}
195 	return container_of(q, struct frag_queue, q);
196 }
197 
198 static int ip6_frag_queue(struct frag_queue *fq, struct sk_buff *skb,
199 			   struct frag_hdr *fhdr, int nhoff)
200 {
201 	struct sk_buff *prev, *next;
202 	struct net_device *dev;
203 	int offset, end;
204 	struct net *net = dev_net(skb_dst(skb)->dev);
205 
206 	if (fq->q.last_in & INET_FRAG_COMPLETE)
207 		goto err;
208 
209 	offset = ntohs(fhdr->frag_off) & ~0x7;
210 	end = offset + (ntohs(ipv6_hdr(skb)->payload_len) -
211 			((u8 *)(fhdr + 1) - (u8 *)(ipv6_hdr(skb) + 1)));
212 
213 	if ((unsigned int)end > IPV6_MAXPLEN) {
214 		IP6_INC_STATS_BH(net, ip6_dst_idev(skb_dst(skb)),
215 				 IPSTATS_MIB_INHDRERRORS);
216 		icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
217 				  ((u8 *)&fhdr->frag_off -
218 				   skb_network_header(skb)));
219 		return -1;
220 	}
221 
222 	if (skb->ip_summed == CHECKSUM_COMPLETE) {
223 		const unsigned char *nh = skb_network_header(skb);
224 		skb->csum = csum_sub(skb->csum,
225 				     csum_partial(nh, (u8 *)(fhdr + 1) - nh,
226 						  0));
227 	}
228 
229 	/* Is this the final fragment? */
230 	if (!(fhdr->frag_off & htons(IP6_MF))) {
231 		/* If we already have some bits beyond end
232 		 * or have different end, the segment is corrupted.
233 		 */
234 		if (end < fq->q.len ||
235 		    ((fq->q.last_in & INET_FRAG_LAST_IN) && end != fq->q.len))
236 			goto err;
237 		fq->q.last_in |= INET_FRAG_LAST_IN;
238 		fq->q.len = end;
239 	} else {
240 		/* Check if the fragment is rounded to 8 bytes.
241 		 * Required by the RFC.
242 		 */
243 		if (end & 0x7) {
244 			/* RFC2460 says always send parameter problem in
245 			 * this case. -DaveM
246 			 */
247 			IP6_INC_STATS_BH(net, ip6_dst_idev(skb_dst(skb)),
248 					 IPSTATS_MIB_INHDRERRORS);
249 			icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
250 					  offsetof(struct ipv6hdr, payload_len));
251 			return -1;
252 		}
253 		if (end > fq->q.len) {
254 			/* Some bits beyond end -> corruption. */
255 			if (fq->q.last_in & INET_FRAG_LAST_IN)
256 				goto err;
257 			fq->q.len = end;
258 		}
259 	}
260 
261 	if (end == offset)
262 		goto err;
263 
264 	/* Point into the IP datagram 'data' part. */
265 	if (!pskb_pull(skb, (u8 *) (fhdr + 1) - skb->data))
266 		goto err;
267 
268 	if (pskb_trim_rcsum(skb, end - offset))
269 		goto err;
270 
271 	/* Find out which fragments are in front and at the back of us
272 	 * in the chain of fragments so far.  We must know where to put
273 	 * this fragment, right?
274 	 */
275 	prev = fq->q.fragments_tail;
276 	if (!prev || FRAG6_CB(prev)->offset < offset) {
277 		next = NULL;
278 		goto found;
279 	}
280 	prev = NULL;
281 	for(next = fq->q.fragments; next != NULL; next = next->next) {
282 		if (FRAG6_CB(next)->offset >= offset)
283 			break;	/* bingo! */
284 		prev = next;
285 	}
286 
287 found:
288 	/* RFC5722, Section 4, amended by Errata ID : 3089
289 	 *                          When reassembling an IPv6 datagram, if
290 	 *   one or more its constituent fragments is determined to be an
291 	 *   overlapping fragment, the entire datagram (and any constituent
292 	 *   fragments) MUST be silently discarded.
293 	 */
294 
295 	/* Check for overlap with preceding fragment. */
296 	if (prev &&
297 	    (FRAG6_CB(prev)->offset + prev->len) > offset)
298 		goto discard_fq;
299 
300 	/* Look for overlap with succeeding segment. */
301 	if (next && FRAG6_CB(next)->offset < end)
302 		goto discard_fq;
303 
304 	FRAG6_CB(skb)->offset = offset;
305 
306 	/* Insert this fragment in the chain of fragments. */
307 	skb->next = next;
308 	if (!next)
309 		fq->q.fragments_tail = skb;
310 	if (prev)
311 		prev->next = skb;
312 	else
313 		fq->q.fragments = skb;
314 
315 	dev = skb->dev;
316 	if (dev) {
317 		fq->iif = dev->ifindex;
318 		skb->dev = NULL;
319 	}
320 	fq->q.stamp = skb->tstamp;
321 	fq->q.meat += skb->len;
322 	add_frag_mem_limit(&fq->q, skb->truesize);
323 
324 	/* The first fragment.
325 	 * nhoffset is obtained from the first fragment, of course.
326 	 */
327 	if (offset == 0) {
328 		fq->nhoffset = nhoff;
329 		fq->q.last_in |= INET_FRAG_FIRST_IN;
330 	}
331 
332 	if (fq->q.last_in == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
333 	    fq->q.meat == fq->q.len) {
334 		int res;
335 		unsigned long orefdst = skb->_skb_refdst;
336 
337 		skb->_skb_refdst = 0UL;
338 		res = ip6_frag_reasm(fq, prev, dev);
339 		skb->_skb_refdst = orefdst;
340 		return res;
341 	}
342 
343 	skb_dst_drop(skb);
344 	inet_frag_lru_move(&fq->q);
345 	return -1;
346 
347 discard_fq:
348 	inet_frag_kill(&fq->q, &ip6_frags);
349 err:
350 	IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
351 		      IPSTATS_MIB_REASMFAILS);
352 	kfree_skb(skb);
353 	return -1;
354 }
355 
356 /*
357  *	Check if this packet is complete.
358  *	Returns NULL on failure by any reason, and pointer
359  *	to current nexthdr field in reassembled frame.
360  *
361  *	It is called with locked fq, and caller must check that
362  *	queue is eligible for reassembly i.e. it is not COMPLETE,
363  *	the last and the first frames arrived and all the bits are here.
364  */
365 static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *prev,
366 			  struct net_device *dev)
367 {
368 	struct net *net = container_of(fq->q.net, struct net, ipv6.frags);
369 	struct sk_buff *fp, *head = fq->q.fragments;
370 	int    payload_len;
371 	unsigned int nhoff;
372 	int sum_truesize;
373 
374 	inet_frag_kill(&fq->q, &ip6_frags);
375 
376 	/* Make the one we just received the head. */
377 	if (prev) {
378 		head = prev->next;
379 		fp = skb_clone(head, GFP_ATOMIC);
380 
381 		if (!fp)
382 			goto out_oom;
383 
384 		fp->next = head->next;
385 		if (!fp->next)
386 			fq->q.fragments_tail = fp;
387 		prev->next = fp;
388 
389 		skb_morph(head, fq->q.fragments);
390 		head->next = fq->q.fragments->next;
391 
392 		consume_skb(fq->q.fragments);
393 		fq->q.fragments = head;
394 	}
395 
396 	WARN_ON(head == NULL);
397 	WARN_ON(FRAG6_CB(head)->offset != 0);
398 
399 	/* Unfragmented part is taken from the first segment. */
400 	payload_len = ((head->data - skb_network_header(head)) -
401 		       sizeof(struct ipv6hdr) + fq->q.len -
402 		       sizeof(struct frag_hdr));
403 	if (payload_len > IPV6_MAXPLEN)
404 		goto out_oversize;
405 
406 	/* Head of list must not be cloned. */
407 	if (skb_unclone(head, GFP_ATOMIC))
408 		goto out_oom;
409 
410 	/* If the first fragment is fragmented itself, we split
411 	 * it to two chunks: the first with data and paged part
412 	 * and the second, holding only fragments. */
413 	if (skb_has_frag_list(head)) {
414 		struct sk_buff *clone;
415 		int i, plen = 0;
416 
417 		if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL)
418 			goto out_oom;
419 		clone->next = head->next;
420 		head->next = clone;
421 		skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
422 		skb_frag_list_init(head);
423 		for (i = 0; i < skb_shinfo(head)->nr_frags; i++)
424 			plen += skb_frag_size(&skb_shinfo(head)->frags[i]);
425 		clone->len = clone->data_len = head->data_len - plen;
426 		head->data_len -= clone->len;
427 		head->len -= clone->len;
428 		clone->csum = 0;
429 		clone->ip_summed = head->ip_summed;
430 		add_frag_mem_limit(&fq->q, clone->truesize);
431 	}
432 
433 	/* We have to remove fragment header from datagram and to relocate
434 	 * header in order to calculate ICV correctly. */
435 	nhoff = fq->nhoffset;
436 	skb_network_header(head)[nhoff] = skb_transport_header(head)[0];
437 	memmove(head->head + sizeof(struct frag_hdr), head->head,
438 		(head->data - head->head) - sizeof(struct frag_hdr));
439 	head->mac_header += sizeof(struct frag_hdr);
440 	head->network_header += sizeof(struct frag_hdr);
441 
442 	skb_reset_transport_header(head);
443 	skb_push(head, head->data - skb_network_header(head));
444 
445 	sum_truesize = head->truesize;
446 	for (fp = head->next; fp;) {
447 		bool headstolen;
448 		int delta;
449 		struct sk_buff *next = fp->next;
450 
451 		sum_truesize += fp->truesize;
452 		if (head->ip_summed != fp->ip_summed)
453 			head->ip_summed = CHECKSUM_NONE;
454 		else if (head->ip_summed == CHECKSUM_COMPLETE)
455 			head->csum = csum_add(head->csum, fp->csum);
456 
457 		if (skb_try_coalesce(head, fp, &headstolen, &delta)) {
458 			kfree_skb_partial(fp, headstolen);
459 		} else {
460 			if (!skb_shinfo(head)->frag_list)
461 				skb_shinfo(head)->frag_list = fp;
462 			head->data_len += fp->len;
463 			head->len += fp->len;
464 			head->truesize += fp->truesize;
465 		}
466 		fp = next;
467 	}
468 	sub_frag_mem_limit(&fq->q, sum_truesize);
469 
470 	head->next = NULL;
471 	head->dev = dev;
472 	head->tstamp = fq->q.stamp;
473 	ipv6_hdr(head)->payload_len = htons(payload_len);
474 	IP6CB(head)->nhoff = nhoff;
475 
476 	/* Yes, and fold redundant checksum back. 8) */
477 	if (head->ip_summed == CHECKSUM_COMPLETE)
478 		head->csum = csum_partial(skb_network_header(head),
479 					  skb_network_header_len(head),
480 					  head->csum);
481 
482 	rcu_read_lock();
483 	IP6_INC_STATS_BH(net, __in6_dev_get(dev), IPSTATS_MIB_REASMOKS);
484 	rcu_read_unlock();
485 	fq->q.fragments = NULL;
486 	fq->q.fragments_tail = NULL;
487 	return 1;
488 
489 out_oversize:
490 	net_dbg_ratelimited("ip6_frag_reasm: payload len = %d\n", payload_len);
491 	goto out_fail;
492 out_oom:
493 	net_dbg_ratelimited("ip6_frag_reasm: no memory for reassembly\n");
494 out_fail:
495 	rcu_read_lock();
496 	IP6_INC_STATS_BH(net, __in6_dev_get(dev), IPSTATS_MIB_REASMFAILS);
497 	rcu_read_unlock();
498 	return -1;
499 }
500 
501 static int ipv6_frag_rcv(struct sk_buff *skb)
502 {
503 	struct frag_hdr *fhdr;
504 	struct frag_queue *fq;
505 	const struct ipv6hdr *hdr = ipv6_hdr(skb);
506 	struct net *net = dev_net(skb_dst(skb)->dev);
507 	int evicted;
508 
509 	IP6_INC_STATS_BH(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMREQDS);
510 
511 	/* Jumbo payload inhibits frag. header */
512 	if (hdr->payload_len==0)
513 		goto fail_hdr;
514 
515 	if (!pskb_may_pull(skb, (skb_transport_offset(skb) +
516 				 sizeof(struct frag_hdr))))
517 		goto fail_hdr;
518 
519 	hdr = ipv6_hdr(skb);
520 	fhdr = (struct frag_hdr *)skb_transport_header(skb);
521 
522 	if (!(fhdr->frag_off & htons(0xFFF9))) {
523 		/* It is not a fragmented frame */
524 		skb->transport_header += sizeof(struct frag_hdr);
525 		IP6_INC_STATS_BH(net,
526 				 ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMOKS);
527 
528 		IP6CB(skb)->nhoff = (u8 *)fhdr - skb_network_header(skb);
529 		return 1;
530 	}
531 
532 	evicted = inet_frag_evictor(&net->ipv6.frags, &ip6_frags, false);
533 	if (evicted)
534 		IP6_ADD_STATS_BH(net, ip6_dst_idev(skb_dst(skb)),
535 				 IPSTATS_MIB_REASMFAILS, evicted);
536 
537 	fq = fq_find(net, fhdr->identification, &hdr->saddr, &hdr->daddr);
538 	if (fq != NULL) {
539 		int ret;
540 
541 		spin_lock(&fq->q.lock);
542 
543 		ret = ip6_frag_queue(fq, skb, fhdr, IP6CB(skb)->nhoff);
544 
545 		spin_unlock(&fq->q.lock);
546 		inet_frag_put(&fq->q, &ip6_frags);
547 		return ret;
548 	}
549 
550 	IP6_INC_STATS_BH(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMFAILS);
551 	kfree_skb(skb);
552 	return -1;
553 
554 fail_hdr:
555 	IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_INHDRERRORS);
556 	icmpv6_param_prob(skb, ICMPV6_HDR_FIELD, skb_network_header_len(skb));
557 	return -1;
558 }
559 
560 static const struct inet6_protocol frag_protocol =
561 {
562 	.handler	=	ipv6_frag_rcv,
563 	.flags		=	INET6_PROTO_NOPOLICY,
564 };
565 
566 #ifdef CONFIG_SYSCTL
567 static struct ctl_table ip6_frags_ns_ctl_table[] = {
568 	{
569 		.procname	= "ip6frag_high_thresh",
570 		.data		= &init_net.ipv6.frags.high_thresh,
571 		.maxlen		= sizeof(int),
572 		.mode		= 0644,
573 		.proc_handler	= proc_dointvec
574 	},
575 	{
576 		.procname	= "ip6frag_low_thresh",
577 		.data		= &init_net.ipv6.frags.low_thresh,
578 		.maxlen		= sizeof(int),
579 		.mode		= 0644,
580 		.proc_handler	= proc_dointvec
581 	},
582 	{
583 		.procname	= "ip6frag_time",
584 		.data		= &init_net.ipv6.frags.timeout,
585 		.maxlen		= sizeof(int),
586 		.mode		= 0644,
587 		.proc_handler	= proc_dointvec_jiffies,
588 	},
589 	{ }
590 };
591 
592 static struct ctl_table ip6_frags_ctl_table[] = {
593 	{
594 		.procname	= "ip6frag_secret_interval",
595 		.data		= &ip6_frags.secret_interval,
596 		.maxlen		= sizeof(int),
597 		.mode		= 0644,
598 		.proc_handler	= proc_dointvec_jiffies,
599 	},
600 	{ }
601 };
602 
603 static int __net_init ip6_frags_ns_sysctl_register(struct net *net)
604 {
605 	struct ctl_table *table;
606 	struct ctl_table_header *hdr;
607 
608 	table = ip6_frags_ns_ctl_table;
609 	if (!net_eq(net, &init_net)) {
610 		table = kmemdup(table, sizeof(ip6_frags_ns_ctl_table), GFP_KERNEL);
611 		if (table == NULL)
612 			goto err_alloc;
613 
614 		table[0].data = &net->ipv6.frags.high_thresh;
615 		table[1].data = &net->ipv6.frags.low_thresh;
616 		table[2].data = &net->ipv6.frags.timeout;
617 
618 		/* Don't export sysctls to unprivileged users */
619 		if (net->user_ns != &init_user_ns)
620 			table[0].procname = NULL;
621 	}
622 
623 	hdr = register_net_sysctl(net, "net/ipv6", table);
624 	if (hdr == NULL)
625 		goto err_reg;
626 
627 	net->ipv6.sysctl.frags_hdr = hdr;
628 	return 0;
629 
630 err_reg:
631 	if (!net_eq(net, &init_net))
632 		kfree(table);
633 err_alloc:
634 	return -ENOMEM;
635 }
636 
637 static void __net_exit ip6_frags_ns_sysctl_unregister(struct net *net)
638 {
639 	struct ctl_table *table;
640 
641 	table = net->ipv6.sysctl.frags_hdr->ctl_table_arg;
642 	unregister_net_sysctl_table(net->ipv6.sysctl.frags_hdr);
643 	if (!net_eq(net, &init_net))
644 		kfree(table);
645 }
646 
647 static struct ctl_table_header *ip6_ctl_header;
648 
649 static int ip6_frags_sysctl_register(void)
650 {
651 	ip6_ctl_header = register_net_sysctl(&init_net, "net/ipv6",
652 			ip6_frags_ctl_table);
653 	return ip6_ctl_header == NULL ? -ENOMEM : 0;
654 }
655 
656 static void ip6_frags_sysctl_unregister(void)
657 {
658 	unregister_net_sysctl_table(ip6_ctl_header);
659 }
660 #else
661 static inline int ip6_frags_ns_sysctl_register(struct net *net)
662 {
663 	return 0;
664 }
665 
666 static inline void ip6_frags_ns_sysctl_unregister(struct net *net)
667 {
668 }
669 
670 static inline int ip6_frags_sysctl_register(void)
671 {
672 	return 0;
673 }
674 
675 static inline void ip6_frags_sysctl_unregister(void)
676 {
677 }
678 #endif
679 
680 static int __net_init ipv6_frags_init_net(struct net *net)
681 {
682 	net->ipv6.frags.high_thresh = IPV6_FRAG_HIGH_THRESH;
683 	net->ipv6.frags.low_thresh = IPV6_FRAG_LOW_THRESH;
684 	net->ipv6.frags.timeout = IPV6_FRAG_TIMEOUT;
685 
686 	inet_frags_init_net(&net->ipv6.frags);
687 
688 	return ip6_frags_ns_sysctl_register(net);
689 }
690 
691 static void __net_exit ipv6_frags_exit_net(struct net *net)
692 {
693 	ip6_frags_ns_sysctl_unregister(net);
694 	inet_frags_exit_net(&net->ipv6.frags, &ip6_frags);
695 }
696 
697 static struct pernet_operations ip6_frags_ops = {
698 	.init = ipv6_frags_init_net,
699 	.exit = ipv6_frags_exit_net,
700 };
701 
702 int __init ipv6_frag_init(void)
703 {
704 	int ret;
705 
706 	ret = inet6_add_protocol(&frag_protocol, IPPROTO_FRAGMENT);
707 	if (ret)
708 		goto out;
709 
710 	ret = ip6_frags_sysctl_register();
711 	if (ret)
712 		goto err_sysctl;
713 
714 	ret = register_pernet_subsys(&ip6_frags_ops);
715 	if (ret)
716 		goto err_pernet;
717 
718 	ip6_frags.hashfn = ip6_hashfn;
719 	ip6_frags.constructor = ip6_frag_init;
720 	ip6_frags.destructor = NULL;
721 	ip6_frags.skb_free = NULL;
722 	ip6_frags.qsize = sizeof(struct frag_queue);
723 	ip6_frags.match = ip6_frag_match;
724 	ip6_frags.frag_expire = ip6_frag_expire;
725 	ip6_frags.secret_interval = 10 * 60 * HZ;
726 	inet_frags_init(&ip6_frags);
727 out:
728 	return ret;
729 
730 err_pernet:
731 	ip6_frags_sysctl_unregister();
732 err_sysctl:
733 	inet6_del_protocol(&frag_protocol, IPPROTO_FRAGMENT);
734 	goto out;
735 }
736 
737 void ipv6_frag_exit(void)
738 {
739 	inet_frags_fini(&ip6_frags);
740 	ip6_frags_sysctl_unregister();
741 	unregister_pernet_subsys(&ip6_frags_ops);
742 	inet6_del_protocol(&frag_protocol, IPPROTO_FRAGMENT);
743 }
744