xref: /linux/net/ipv6/reassembly.c (revision 8c749ce93ee69e789e46b3be98de9e0cbfcf8ed8)
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 #include <net/inet_ecn.h>
62 
63 static const char ip6_frag_cache_name[] = "ip6-frags";
64 
65 struct ip6frag_skb_cb {
66 	struct inet6_skb_parm	h;
67 	int			offset;
68 };
69 
70 #define FRAG6_CB(skb)	((struct ip6frag_skb_cb *)((skb)->cb))
71 
72 static u8 ip6_frag_ecn(const struct ipv6hdr *ipv6h)
73 {
74 	return 1 << (ipv6_get_dsfield(ipv6h) & INET_ECN_MASK);
75 }
76 
77 static struct inet_frags ip6_frags;
78 
79 static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *prev,
80 			  struct net_device *dev);
81 
82 /*
83  * callers should be careful not to use the hash value outside the ipfrag_lock
84  * as doing so could race with ipfrag_hash_rnd being recalculated.
85  */
86 static unsigned int inet6_hash_frag(__be32 id, const struct in6_addr *saddr,
87 				    const struct in6_addr *daddr)
88 {
89 	net_get_random_once(&ip6_frags.rnd, sizeof(ip6_frags.rnd));
90 	return jhash_3words(ipv6_addr_hash(saddr), ipv6_addr_hash(daddr),
91 			    (__force u32)id, ip6_frags.rnd);
92 }
93 
94 static unsigned int ip6_hashfn(const struct inet_frag_queue *q)
95 {
96 	const struct frag_queue *fq;
97 
98 	fq = container_of(q, struct frag_queue, q);
99 	return inet6_hash_frag(fq->id, &fq->saddr, &fq->daddr);
100 }
101 
102 bool ip6_frag_match(const struct inet_frag_queue *q, const void *a)
103 {
104 	const struct frag_queue *fq;
105 	const struct ip6_create_arg *arg = a;
106 
107 	fq = container_of(q, struct frag_queue, q);
108 	return	fq->id == arg->id &&
109 		fq->user == arg->user &&
110 		ipv6_addr_equal(&fq->saddr, arg->src) &&
111 		ipv6_addr_equal(&fq->daddr, arg->dst) &&
112 		(arg->iif == fq->iif ||
113 		 !(ipv6_addr_type(arg->dst) & (IPV6_ADDR_MULTICAST |
114 					       IPV6_ADDR_LINKLOCAL)));
115 }
116 EXPORT_SYMBOL(ip6_frag_match);
117 
118 void ip6_frag_init(struct inet_frag_queue *q, const void *a)
119 {
120 	struct frag_queue *fq = container_of(q, struct frag_queue, q);
121 	const struct ip6_create_arg *arg = a;
122 
123 	fq->id = arg->id;
124 	fq->user = arg->user;
125 	fq->saddr = *arg->src;
126 	fq->daddr = *arg->dst;
127 	fq->ecn = arg->ecn;
128 }
129 EXPORT_SYMBOL(ip6_frag_init);
130 
131 void ip6_expire_frag_queue(struct net *net, struct frag_queue *fq,
132 			   struct inet_frags *frags)
133 {
134 	struct net_device *dev = NULL;
135 
136 	spin_lock(&fq->q.lock);
137 
138 	if (fq->q.flags & INET_FRAG_COMPLETE)
139 		goto out;
140 
141 	inet_frag_kill(&fq->q, frags);
142 
143 	rcu_read_lock();
144 	dev = dev_get_by_index_rcu(net, fq->iif);
145 	if (!dev)
146 		goto out_rcu_unlock;
147 
148 	IP6_INC_STATS_BH(net, __in6_dev_get(dev), IPSTATS_MIB_REASMFAILS);
149 
150 	if (inet_frag_evicting(&fq->q))
151 		goto out_rcu_unlock;
152 
153 	IP6_INC_STATS_BH(net, __in6_dev_get(dev), IPSTATS_MIB_REASMTIMEOUT);
154 
155 	/* Don't send error if the first segment did not arrive. */
156 	if (!(fq->q.flags & INET_FRAG_FIRST_IN) || !fq->q.fragments)
157 		goto out_rcu_unlock;
158 
159 	/* But use as source device on which LAST ARRIVED
160 	 * segment was received. And do not use fq->dev
161 	 * pointer directly, device might already disappeared.
162 	 */
163 	fq->q.fragments->dev = dev;
164 	icmpv6_send(fq->q.fragments, ICMPV6_TIME_EXCEED, ICMPV6_EXC_FRAGTIME, 0);
165 out_rcu_unlock:
166 	rcu_read_unlock();
167 out:
168 	spin_unlock(&fq->q.lock);
169 	inet_frag_put(&fq->q, frags);
170 }
171 EXPORT_SYMBOL(ip6_expire_frag_queue);
172 
173 static void ip6_frag_expire(unsigned long data)
174 {
175 	struct frag_queue *fq;
176 	struct net *net;
177 
178 	fq = container_of((struct inet_frag_queue *)data, struct frag_queue, q);
179 	net = container_of(fq->q.net, struct net, ipv6.frags);
180 
181 	ip6_expire_frag_queue(net, fq, &ip6_frags);
182 }
183 
184 static struct frag_queue *
185 fq_find(struct net *net, __be32 id, const struct in6_addr *src,
186 	const struct in6_addr *dst, int iif, u8 ecn)
187 {
188 	struct inet_frag_queue *q;
189 	struct ip6_create_arg arg;
190 	unsigned int hash;
191 
192 	arg.id = id;
193 	arg.user = IP6_DEFRAG_LOCAL_DELIVER;
194 	arg.src = src;
195 	arg.dst = dst;
196 	arg.iif = iif;
197 	arg.ecn = ecn;
198 
199 	hash = inet6_hash_frag(id, src, dst);
200 
201 	q = inet_frag_find(&net->ipv6.frags, &ip6_frags, &arg, hash);
202 	if (IS_ERR_OR_NULL(q)) {
203 		inet_frag_maybe_warn_overflow(q, pr_fmt());
204 		return NULL;
205 	}
206 	return container_of(q, struct frag_queue, q);
207 }
208 
209 static int ip6_frag_queue(struct frag_queue *fq, struct sk_buff *skb,
210 			   struct frag_hdr *fhdr, int nhoff)
211 {
212 	struct sk_buff *prev, *next;
213 	struct net_device *dev;
214 	int offset, end;
215 	struct net *net = dev_net(skb_dst(skb)->dev);
216 	u8 ecn;
217 
218 	if (fq->q.flags & INET_FRAG_COMPLETE)
219 		goto err;
220 
221 	offset = ntohs(fhdr->frag_off) & ~0x7;
222 	end = offset + (ntohs(ipv6_hdr(skb)->payload_len) -
223 			((u8 *)(fhdr + 1) - (u8 *)(ipv6_hdr(skb) + 1)));
224 
225 	if ((unsigned int)end > IPV6_MAXPLEN) {
226 		IP6_INC_STATS_BH(net, ip6_dst_idev(skb_dst(skb)),
227 				 IPSTATS_MIB_INHDRERRORS);
228 		icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
229 				  ((u8 *)&fhdr->frag_off -
230 				   skb_network_header(skb)));
231 		return -1;
232 	}
233 
234 	ecn = ip6_frag_ecn(ipv6_hdr(skb));
235 
236 	if (skb->ip_summed == CHECKSUM_COMPLETE) {
237 		const unsigned char *nh = skb_network_header(skb);
238 		skb->csum = csum_sub(skb->csum,
239 				     csum_partial(nh, (u8 *)(fhdr + 1) - nh,
240 						  0));
241 	}
242 
243 	/* Is this the final fragment? */
244 	if (!(fhdr->frag_off & htons(IP6_MF))) {
245 		/* If we already have some bits beyond end
246 		 * or have different end, the segment is corrupted.
247 		 */
248 		if (end < fq->q.len ||
249 		    ((fq->q.flags & INET_FRAG_LAST_IN) && end != fq->q.len))
250 			goto err;
251 		fq->q.flags |= INET_FRAG_LAST_IN;
252 		fq->q.len = end;
253 	} else {
254 		/* Check if the fragment is rounded to 8 bytes.
255 		 * Required by the RFC.
256 		 */
257 		if (end & 0x7) {
258 			/* RFC2460 says always send parameter problem in
259 			 * this case. -DaveM
260 			 */
261 			IP6_INC_STATS_BH(net, ip6_dst_idev(skb_dst(skb)),
262 					 IPSTATS_MIB_INHDRERRORS);
263 			icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
264 					  offsetof(struct ipv6hdr, payload_len));
265 			return -1;
266 		}
267 		if (end > fq->q.len) {
268 			/* Some bits beyond end -> corruption. */
269 			if (fq->q.flags & INET_FRAG_LAST_IN)
270 				goto err;
271 			fq->q.len = end;
272 		}
273 	}
274 
275 	if (end == offset)
276 		goto err;
277 
278 	/* Point into the IP datagram 'data' part. */
279 	if (!pskb_pull(skb, (u8 *) (fhdr + 1) - skb->data))
280 		goto err;
281 
282 	if (pskb_trim_rcsum(skb, end - offset))
283 		goto err;
284 
285 	/* Find out which fragments are in front and at the back of us
286 	 * in the chain of fragments so far.  We must know where to put
287 	 * this fragment, right?
288 	 */
289 	prev = fq->q.fragments_tail;
290 	if (!prev || FRAG6_CB(prev)->offset < offset) {
291 		next = NULL;
292 		goto found;
293 	}
294 	prev = NULL;
295 	for (next = fq->q.fragments; next != NULL; next = next->next) {
296 		if (FRAG6_CB(next)->offset >= offset)
297 			break;	/* bingo! */
298 		prev = next;
299 	}
300 
301 found:
302 	/* RFC5722, Section 4, amended by Errata ID : 3089
303 	 *                          When reassembling an IPv6 datagram, if
304 	 *   one or more its constituent fragments is determined to be an
305 	 *   overlapping fragment, the entire datagram (and any constituent
306 	 *   fragments) MUST be silently discarded.
307 	 */
308 
309 	/* Check for overlap with preceding fragment. */
310 	if (prev &&
311 	    (FRAG6_CB(prev)->offset + prev->len) > offset)
312 		goto discard_fq;
313 
314 	/* Look for overlap with succeeding segment. */
315 	if (next && FRAG6_CB(next)->offset < end)
316 		goto discard_fq;
317 
318 	FRAG6_CB(skb)->offset = offset;
319 
320 	/* Insert this fragment in the chain of fragments. */
321 	skb->next = next;
322 	if (!next)
323 		fq->q.fragments_tail = skb;
324 	if (prev)
325 		prev->next = skb;
326 	else
327 		fq->q.fragments = skb;
328 
329 	dev = skb->dev;
330 	if (dev) {
331 		fq->iif = dev->ifindex;
332 		skb->dev = NULL;
333 	}
334 	fq->q.stamp = skb->tstamp;
335 	fq->q.meat += skb->len;
336 	fq->ecn |= ecn;
337 	add_frag_mem_limit(fq->q.net, skb->truesize);
338 
339 	/* The first fragment.
340 	 * nhoffset is obtained from the first fragment, of course.
341 	 */
342 	if (offset == 0) {
343 		fq->nhoffset = nhoff;
344 		fq->q.flags |= INET_FRAG_FIRST_IN;
345 	}
346 
347 	if (fq->q.flags == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
348 	    fq->q.meat == fq->q.len) {
349 		int res;
350 		unsigned long orefdst = skb->_skb_refdst;
351 
352 		skb->_skb_refdst = 0UL;
353 		res = ip6_frag_reasm(fq, prev, dev);
354 		skb->_skb_refdst = orefdst;
355 		return res;
356 	}
357 
358 	skb_dst_drop(skb);
359 	return -1;
360 
361 discard_fq:
362 	inet_frag_kill(&fq->q, &ip6_frags);
363 err:
364 	IP6_INC_STATS_BH(net, ip6_dst_idev(skb_dst(skb)),
365 			 IPSTATS_MIB_REASMFAILS);
366 	kfree_skb(skb);
367 	return -1;
368 }
369 
370 /*
371  *	Check if this packet is complete.
372  *	Returns NULL on failure by any reason, and pointer
373  *	to current nexthdr field in reassembled frame.
374  *
375  *	It is called with locked fq, and caller must check that
376  *	queue is eligible for reassembly i.e. it is not COMPLETE,
377  *	the last and the first frames arrived and all the bits are here.
378  */
379 static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *prev,
380 			  struct net_device *dev)
381 {
382 	struct net *net = container_of(fq->q.net, struct net, ipv6.frags);
383 	struct sk_buff *fp, *head = fq->q.fragments;
384 	int    payload_len;
385 	unsigned int nhoff;
386 	int sum_truesize;
387 	u8 ecn;
388 
389 	inet_frag_kill(&fq->q, &ip6_frags);
390 
391 	ecn = ip_frag_ecn_table[fq->ecn];
392 	if (unlikely(ecn == 0xff))
393 		goto out_fail;
394 
395 	/* Make the one we just received the head. */
396 	if (prev) {
397 		head = prev->next;
398 		fp = skb_clone(head, GFP_ATOMIC);
399 
400 		if (!fp)
401 			goto out_oom;
402 
403 		fp->next = head->next;
404 		if (!fp->next)
405 			fq->q.fragments_tail = fp;
406 		prev->next = fp;
407 
408 		skb_morph(head, fq->q.fragments);
409 		head->next = fq->q.fragments->next;
410 
411 		consume_skb(fq->q.fragments);
412 		fq->q.fragments = head;
413 	}
414 
415 	WARN_ON(head == NULL);
416 	WARN_ON(FRAG6_CB(head)->offset != 0);
417 
418 	/* Unfragmented part is taken from the first segment. */
419 	payload_len = ((head->data - skb_network_header(head)) -
420 		       sizeof(struct ipv6hdr) + fq->q.len -
421 		       sizeof(struct frag_hdr));
422 	if (payload_len > IPV6_MAXPLEN)
423 		goto out_oversize;
424 
425 	/* Head of list must not be cloned. */
426 	if (skb_unclone(head, GFP_ATOMIC))
427 		goto out_oom;
428 
429 	/* If the first fragment is fragmented itself, we split
430 	 * it to two chunks: the first with data and paged part
431 	 * and the second, holding only fragments. */
432 	if (skb_has_frag_list(head)) {
433 		struct sk_buff *clone;
434 		int i, plen = 0;
435 
436 		clone = alloc_skb(0, GFP_ATOMIC);
437 		if (!clone)
438 			goto out_oom;
439 		clone->next = head->next;
440 		head->next = clone;
441 		skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
442 		skb_frag_list_init(head);
443 		for (i = 0; i < skb_shinfo(head)->nr_frags; i++)
444 			plen += skb_frag_size(&skb_shinfo(head)->frags[i]);
445 		clone->len = clone->data_len = head->data_len - plen;
446 		head->data_len -= clone->len;
447 		head->len -= clone->len;
448 		clone->csum = 0;
449 		clone->ip_summed = head->ip_summed;
450 		add_frag_mem_limit(fq->q.net, clone->truesize);
451 	}
452 
453 	/* We have to remove fragment header from datagram and to relocate
454 	 * header in order to calculate ICV correctly. */
455 	nhoff = fq->nhoffset;
456 	skb_network_header(head)[nhoff] = skb_transport_header(head)[0];
457 	memmove(head->head + sizeof(struct frag_hdr), head->head,
458 		(head->data - head->head) - sizeof(struct frag_hdr));
459 	head->mac_header += sizeof(struct frag_hdr);
460 	head->network_header += sizeof(struct frag_hdr);
461 
462 	skb_reset_transport_header(head);
463 	skb_push(head, head->data - skb_network_header(head));
464 
465 	sum_truesize = head->truesize;
466 	for (fp = head->next; fp;) {
467 		bool headstolen;
468 		int delta;
469 		struct sk_buff *next = fp->next;
470 
471 		sum_truesize += fp->truesize;
472 		if (head->ip_summed != fp->ip_summed)
473 			head->ip_summed = CHECKSUM_NONE;
474 		else if (head->ip_summed == CHECKSUM_COMPLETE)
475 			head->csum = csum_add(head->csum, fp->csum);
476 
477 		if (skb_try_coalesce(head, fp, &headstolen, &delta)) {
478 			kfree_skb_partial(fp, headstolen);
479 		} else {
480 			if (!skb_shinfo(head)->frag_list)
481 				skb_shinfo(head)->frag_list = fp;
482 			head->data_len += fp->len;
483 			head->len += fp->len;
484 			head->truesize += fp->truesize;
485 		}
486 		fp = next;
487 	}
488 	sub_frag_mem_limit(fq->q.net, sum_truesize);
489 
490 	head->next = NULL;
491 	head->dev = dev;
492 	head->tstamp = fq->q.stamp;
493 	ipv6_hdr(head)->payload_len = htons(payload_len);
494 	ipv6_change_dsfield(ipv6_hdr(head), 0xff, ecn);
495 	IP6CB(head)->nhoff = nhoff;
496 	IP6CB(head)->flags |= IP6SKB_FRAGMENTED;
497 
498 	/* Yes, and fold redundant checksum back. 8) */
499 	if (head->ip_summed == CHECKSUM_COMPLETE)
500 		head->csum = csum_partial(skb_network_header(head),
501 					  skb_network_header_len(head),
502 					  head->csum);
503 
504 	rcu_read_lock();
505 	IP6_INC_STATS_BH(net, __in6_dev_get(dev), IPSTATS_MIB_REASMOKS);
506 	rcu_read_unlock();
507 	fq->q.fragments = NULL;
508 	fq->q.fragments_tail = NULL;
509 	return 1;
510 
511 out_oversize:
512 	net_dbg_ratelimited("ip6_frag_reasm: payload len = %d\n", payload_len);
513 	goto out_fail;
514 out_oom:
515 	net_dbg_ratelimited("ip6_frag_reasm: no memory for reassembly\n");
516 out_fail:
517 	rcu_read_lock();
518 	IP6_INC_STATS_BH(net, __in6_dev_get(dev), IPSTATS_MIB_REASMFAILS);
519 	rcu_read_unlock();
520 	return -1;
521 }
522 
523 static int ipv6_frag_rcv(struct sk_buff *skb)
524 {
525 	struct frag_hdr *fhdr;
526 	struct frag_queue *fq;
527 	const struct ipv6hdr *hdr = ipv6_hdr(skb);
528 	struct net *net = dev_net(skb_dst(skb)->dev);
529 
530 	if (IP6CB(skb)->flags & IP6SKB_FRAGMENTED)
531 		goto fail_hdr;
532 
533 	IP6_INC_STATS_BH(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMREQDS);
534 
535 	/* Jumbo payload inhibits frag. header */
536 	if (hdr->payload_len == 0)
537 		goto fail_hdr;
538 
539 	if (!pskb_may_pull(skb, (skb_transport_offset(skb) +
540 				 sizeof(struct frag_hdr))))
541 		goto fail_hdr;
542 
543 	hdr = ipv6_hdr(skb);
544 	fhdr = (struct frag_hdr *)skb_transport_header(skb);
545 
546 	if (!(fhdr->frag_off & htons(0xFFF9))) {
547 		/* It is not a fragmented frame */
548 		skb->transport_header += sizeof(struct frag_hdr);
549 		IP6_INC_STATS_BH(net,
550 				 ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMOKS);
551 
552 		IP6CB(skb)->nhoff = (u8 *)fhdr - skb_network_header(skb);
553 		IP6CB(skb)->flags |= IP6SKB_FRAGMENTED;
554 		return 1;
555 	}
556 
557 	fq = fq_find(net, fhdr->identification, &hdr->saddr, &hdr->daddr,
558 		     skb->dev ? skb->dev->ifindex : 0, ip6_frag_ecn(hdr));
559 	if (fq) {
560 		int ret;
561 
562 		spin_lock(&fq->q.lock);
563 
564 		ret = ip6_frag_queue(fq, skb, fhdr, IP6CB(skb)->nhoff);
565 
566 		spin_unlock(&fq->q.lock);
567 		inet_frag_put(&fq->q, &ip6_frags);
568 		return ret;
569 	}
570 
571 	IP6_INC_STATS_BH(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMFAILS);
572 	kfree_skb(skb);
573 	return -1;
574 
575 fail_hdr:
576 	IP6_INC_STATS_BH(net, ip6_dst_idev(skb_dst(skb)),
577 			 IPSTATS_MIB_INHDRERRORS);
578 	icmpv6_param_prob(skb, ICMPV6_HDR_FIELD, skb_network_header_len(skb));
579 	return -1;
580 }
581 
582 static const struct inet6_protocol frag_protocol = {
583 	.handler	=	ipv6_frag_rcv,
584 	.flags		=	INET6_PROTO_NOPOLICY,
585 };
586 
587 #ifdef CONFIG_SYSCTL
588 static int zero;
589 
590 static struct ctl_table ip6_frags_ns_ctl_table[] = {
591 	{
592 		.procname	= "ip6frag_high_thresh",
593 		.data		= &init_net.ipv6.frags.high_thresh,
594 		.maxlen		= sizeof(int),
595 		.mode		= 0644,
596 		.proc_handler	= proc_dointvec_minmax,
597 		.extra1		= &init_net.ipv6.frags.low_thresh
598 	},
599 	{
600 		.procname	= "ip6frag_low_thresh",
601 		.data		= &init_net.ipv6.frags.low_thresh,
602 		.maxlen		= sizeof(int),
603 		.mode		= 0644,
604 		.proc_handler	= proc_dointvec_minmax,
605 		.extra1		= &zero,
606 		.extra2		= &init_net.ipv6.frags.high_thresh
607 	},
608 	{
609 		.procname	= "ip6frag_time",
610 		.data		= &init_net.ipv6.frags.timeout,
611 		.maxlen		= sizeof(int),
612 		.mode		= 0644,
613 		.proc_handler	= proc_dointvec_jiffies,
614 	},
615 	{ }
616 };
617 
618 /* secret interval has been deprecated */
619 static int ip6_frags_secret_interval_unused;
620 static struct ctl_table ip6_frags_ctl_table[] = {
621 	{
622 		.procname	= "ip6frag_secret_interval",
623 		.data		= &ip6_frags_secret_interval_unused,
624 		.maxlen		= sizeof(int),
625 		.mode		= 0644,
626 		.proc_handler	= proc_dointvec_jiffies,
627 	},
628 	{ }
629 };
630 
631 static int __net_init ip6_frags_ns_sysctl_register(struct net *net)
632 {
633 	struct ctl_table *table;
634 	struct ctl_table_header *hdr;
635 
636 	table = ip6_frags_ns_ctl_table;
637 	if (!net_eq(net, &init_net)) {
638 		table = kmemdup(table, sizeof(ip6_frags_ns_ctl_table), GFP_KERNEL);
639 		if (!table)
640 			goto err_alloc;
641 
642 		table[0].data = &net->ipv6.frags.high_thresh;
643 		table[0].extra1 = &net->ipv6.frags.low_thresh;
644 		table[0].extra2 = &init_net.ipv6.frags.high_thresh;
645 		table[1].data = &net->ipv6.frags.low_thresh;
646 		table[1].extra2 = &net->ipv6.frags.high_thresh;
647 		table[2].data = &net->ipv6.frags.timeout;
648 
649 		/* Don't export sysctls to unprivileged users */
650 		if (net->user_ns != &init_user_ns)
651 			table[0].procname = NULL;
652 	}
653 
654 	hdr = register_net_sysctl(net, "net/ipv6", table);
655 	if (!hdr)
656 		goto err_reg;
657 
658 	net->ipv6.sysctl.frags_hdr = hdr;
659 	return 0;
660 
661 err_reg:
662 	if (!net_eq(net, &init_net))
663 		kfree(table);
664 err_alloc:
665 	return -ENOMEM;
666 }
667 
668 static void __net_exit ip6_frags_ns_sysctl_unregister(struct net *net)
669 {
670 	struct ctl_table *table;
671 
672 	table = net->ipv6.sysctl.frags_hdr->ctl_table_arg;
673 	unregister_net_sysctl_table(net->ipv6.sysctl.frags_hdr);
674 	if (!net_eq(net, &init_net))
675 		kfree(table);
676 }
677 
678 static struct ctl_table_header *ip6_ctl_header;
679 
680 static int ip6_frags_sysctl_register(void)
681 {
682 	ip6_ctl_header = register_net_sysctl(&init_net, "net/ipv6",
683 			ip6_frags_ctl_table);
684 	return ip6_ctl_header == NULL ? -ENOMEM : 0;
685 }
686 
687 static void ip6_frags_sysctl_unregister(void)
688 {
689 	unregister_net_sysctl_table(ip6_ctl_header);
690 }
691 #else
692 static int ip6_frags_ns_sysctl_register(struct net *net)
693 {
694 	return 0;
695 }
696 
697 static void ip6_frags_ns_sysctl_unregister(struct net *net)
698 {
699 }
700 
701 static int ip6_frags_sysctl_register(void)
702 {
703 	return 0;
704 }
705 
706 static void ip6_frags_sysctl_unregister(void)
707 {
708 }
709 #endif
710 
711 static int __net_init ipv6_frags_init_net(struct net *net)
712 {
713 	int res;
714 
715 	net->ipv6.frags.high_thresh = IPV6_FRAG_HIGH_THRESH;
716 	net->ipv6.frags.low_thresh = IPV6_FRAG_LOW_THRESH;
717 	net->ipv6.frags.timeout = IPV6_FRAG_TIMEOUT;
718 
719 	res = inet_frags_init_net(&net->ipv6.frags);
720 	if (res)
721 		return res;
722 	res = ip6_frags_ns_sysctl_register(net);
723 	if (res)
724 		inet_frags_uninit_net(&net->ipv6.frags);
725 	return res;
726 }
727 
728 static void __net_exit ipv6_frags_exit_net(struct net *net)
729 {
730 	ip6_frags_ns_sysctl_unregister(net);
731 	inet_frags_exit_net(&net->ipv6.frags, &ip6_frags);
732 }
733 
734 static struct pernet_operations ip6_frags_ops = {
735 	.init = ipv6_frags_init_net,
736 	.exit = ipv6_frags_exit_net,
737 };
738 
739 int __init ipv6_frag_init(void)
740 {
741 	int ret;
742 
743 	ret = inet6_add_protocol(&frag_protocol, IPPROTO_FRAGMENT);
744 	if (ret)
745 		goto out;
746 
747 	ret = ip6_frags_sysctl_register();
748 	if (ret)
749 		goto err_sysctl;
750 
751 	ret = register_pernet_subsys(&ip6_frags_ops);
752 	if (ret)
753 		goto err_pernet;
754 
755 	ip6_frags.hashfn = ip6_hashfn;
756 	ip6_frags.constructor = ip6_frag_init;
757 	ip6_frags.destructor = NULL;
758 	ip6_frags.qsize = sizeof(struct frag_queue);
759 	ip6_frags.match = ip6_frag_match;
760 	ip6_frags.frag_expire = ip6_frag_expire;
761 	ip6_frags.frags_cache_name = ip6_frag_cache_name;
762 	ret = inet_frags_init(&ip6_frags);
763 	if (ret)
764 		goto err_pernet;
765 out:
766 	return ret;
767 
768 err_pernet:
769 	ip6_frags_sysctl_unregister();
770 err_sysctl:
771 	inet6_del_protocol(&frag_protocol, IPPROTO_FRAGMENT);
772 	goto out;
773 }
774 
775 void ipv6_frag_exit(void)
776 {
777 	inet_frags_fini(&ip6_frags);
778 	ip6_frags_sysctl_unregister();
779 	unregister_pernet_subsys(&ip6_frags_ops);
780 	inet6_del_protocol(&frag_protocol, IPPROTO_FRAGMENT);
781 }
782