xref: /openbmc/linux/net/ipv6/reassembly.c (revision 06aa8b8a) 
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 struct ip6frag_skb_cb
64 {
65 	struct inet6_skb_parm	h;
66 	int			offset;
67 };
68 
69 #define FRAG6_CB(skb)	((struct ip6frag_skb_cb*)((skb)->cb))
70 
71 static inline u8 ip6_frag_ecn(const struct ipv6hdr *ipv6h)
72 {
73 	return 1 << (ipv6_get_dsfield(ipv6h) & INET_ECN_MASK);
74 }
75 
76 static struct inet_frags ip6_frags;
77 
78 static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *prev,
79 			  struct net_device *dev);
80 
81 /*
82  * callers should be careful not to use the hash value outside the ipfrag_lock
83  * as doing so could race with ipfrag_hash_rnd being recalculated.
84  */
85 static unsigned int inet6_hash_frag(__be32 id, const struct in6_addr *saddr,
86 				    const struct in6_addr *daddr)
87 {
88 	net_get_random_once(&ip6_frags.rnd, sizeof(ip6_frags.rnd));
89 	return jhash_3words(ipv6_addr_hash(saddr), ipv6_addr_hash(daddr),
90 			    (__force u32)id, ip6_frags.rnd);
91 }
92 
93 static unsigned int ip6_hashfn(const struct inet_frag_queue *q)
94 {
95 	const struct frag_queue *fq;
96 
97 	fq = container_of(q, struct frag_queue, q);
98 	return inet6_hash_frag(fq->id, &fq->saddr, &fq->daddr);
99 }
100 
101 bool ip6_frag_match(const struct inet_frag_queue *q, const void *a)
102 {
103 	const struct frag_queue *fq;
104 	const struct ip6_create_arg *arg = a;
105 
106 	fq = container_of(q, struct frag_queue, q);
107 	return	fq->id == arg->id &&
108 		fq->user == arg->user &&
109 		ipv6_addr_equal(&fq->saddr, arg->src) &&
110 		ipv6_addr_equal(&fq->daddr, arg->dst);
111 }
112 EXPORT_SYMBOL(ip6_frag_match);
113 
114 void ip6_frag_init(struct inet_frag_queue *q, const void *a)
115 {
116 	struct frag_queue *fq = container_of(q, struct frag_queue, q);
117 	const struct ip6_create_arg *arg = a;
118 
119 	fq->id = arg->id;
120 	fq->user = arg->user;
121 	fq->saddr = *arg->src;
122 	fq->daddr = *arg->dst;
123 	fq->ecn = arg->ecn;
124 }
125 EXPORT_SYMBOL(ip6_frag_init);
126 
127 void ip6_expire_frag_queue(struct net *net, struct frag_queue *fq,
128 			   struct inet_frags *frags)
129 {
130 	struct net_device *dev = NULL;
131 
132 	spin_lock(&fq->q.lock);
133 
134 	if (fq->q.flags & INET_FRAG_COMPLETE)
135 		goto out;
136 
137 	inet_frag_kill(&fq->q, frags);
138 
139 	rcu_read_lock();
140 	dev = dev_get_by_index_rcu(net, fq->iif);
141 	if (!dev)
142 		goto out_rcu_unlock;
143 
144 	if (!(fq->q.flags & INET_FRAG_EVICTED))
145 		IP6_INC_STATS_BH(net, __in6_dev_get(dev),
146 				 IPSTATS_MIB_REASMTIMEOUT);
147 	IP6_INC_STATS_BH(net, __in6_dev_get(dev), IPSTATS_MIB_REASMFAILS);
148 
149 	/* Don't send error if the first segment did not arrive. */
150 	if (!(fq->q.flags & INET_FRAG_FIRST_IN) || !fq->q.fragments)
151 		goto out_rcu_unlock;
152 
153 	/*
154 	   But use as source device on which LAST ARRIVED
155 	   segment was received. And do not use fq->dev
156 	   pointer directly, device might already disappeared.
157 	 */
158 	fq->q.fragments->dev = dev;
159 	icmpv6_send(fq->q.fragments, ICMPV6_TIME_EXCEED, ICMPV6_EXC_FRAGTIME, 0);
160 out_rcu_unlock:
161 	rcu_read_unlock();
162 out:
163 	spin_unlock(&fq->q.lock);
164 	inet_frag_put(&fq->q, frags);
165 }
166 EXPORT_SYMBOL(ip6_expire_frag_queue);
167 
168 static void ip6_frag_expire(unsigned long data)
169 {
170 	struct frag_queue *fq;
171 	struct net *net;
172 
173 	fq = container_of((struct inet_frag_queue *)data, struct frag_queue, q);
174 	net = container_of(fq->q.net, struct net, ipv6.frags);
175 
176 	ip6_expire_frag_queue(net, fq, &ip6_frags);
177 }
178 
179 static __inline__ struct frag_queue *
180 fq_find(struct net *net, __be32 id, const struct in6_addr *src,
181 	const struct in6_addr *dst, u8 ecn)
182 {
183 	struct inet_frag_queue *q;
184 	struct ip6_create_arg arg;
185 	unsigned int hash;
186 
187 	arg.id = id;
188 	arg.user = IP6_DEFRAG_LOCAL_DELIVER;
189 	arg.src = src;
190 	arg.dst = dst;
191 	arg.ecn = ecn;
192 
193 	hash = inet6_hash_frag(id, src, dst);
194 
195 	q = inet_frag_find(&net->ipv6.frags, &ip6_frags, &arg, hash);
196 	if (IS_ERR_OR_NULL(q)) {
197 		inet_frag_maybe_warn_overflow(q, pr_fmt());
198 		return NULL;
199 	}
200 	return container_of(q, struct frag_queue, q);
201 }
202 
203 static int ip6_frag_queue(struct frag_queue *fq, struct sk_buff *skb,
204 			   struct frag_hdr *fhdr, int nhoff)
205 {
206 	struct sk_buff *prev, *next;
207 	struct net_device *dev;
208 	int offset, end;
209 	struct net *net = dev_net(skb_dst(skb)->dev);
210 	u8 ecn;
211 
212 	if (fq->q.flags & INET_FRAG_COMPLETE)
213 		goto err;
214 
215 	offset = ntohs(fhdr->frag_off) & ~0x7;
216 	end = offset + (ntohs(ipv6_hdr(skb)->payload_len) -
217 			((u8 *)(fhdr + 1) - (u8 *)(ipv6_hdr(skb) + 1)));
218 
219 	if ((unsigned int)end > IPV6_MAXPLEN) {
220 		IP6_INC_STATS_BH(net, ip6_dst_idev(skb_dst(skb)),
221 				 IPSTATS_MIB_INHDRERRORS);
222 		icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
223 				  ((u8 *)&fhdr->frag_off -
224 				   skb_network_header(skb)));
225 		return -1;
226 	}
227 
228 	ecn = ip6_frag_ecn(ipv6_hdr(skb));
229 
230 	if (skb->ip_summed == CHECKSUM_COMPLETE) {
231 		const unsigned char *nh = skb_network_header(skb);
232 		skb->csum = csum_sub(skb->csum,
233 				     csum_partial(nh, (u8 *)(fhdr + 1) - nh,
234 						  0));
235 	}
236 
237 	/* Is this the final fragment? */
238 	if (!(fhdr->frag_off & htons(IP6_MF))) {
239 		/* If we already have some bits beyond end
240 		 * or have different end, the segment is corrupted.
241 		 */
242 		if (end < fq->q.len ||
243 		    ((fq->q.flags & INET_FRAG_LAST_IN) && end != fq->q.len))
244 			goto err;
245 		fq->q.flags |= INET_FRAG_LAST_IN;
246 		fq->q.len = end;
247 	} else {
248 		/* Check if the fragment is rounded to 8 bytes.
249 		 * Required by the RFC.
250 		 */
251 		if (end & 0x7) {
252 			/* RFC2460 says always send parameter problem in
253 			 * this case. -DaveM
254 			 */
255 			IP6_INC_STATS_BH(net, ip6_dst_idev(skb_dst(skb)),
256 					 IPSTATS_MIB_INHDRERRORS);
257 			icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
258 					  offsetof(struct ipv6hdr, payload_len));
259 			return -1;
260 		}
261 		if (end > fq->q.len) {
262 			/* Some bits beyond end -> corruption. */
263 			if (fq->q.flags & INET_FRAG_LAST_IN)
264 				goto err;
265 			fq->q.len = end;
266 		}
267 	}
268 
269 	if (end == offset)
270 		goto err;
271 
272 	/* Point into the IP datagram 'data' part. */
273 	if (!pskb_pull(skb, (u8 *) (fhdr + 1) - skb->data))
274 		goto err;
275 
276 	if (pskb_trim_rcsum(skb, end - offset))
277 		goto err;
278 
279 	/* Find out which fragments are in front and at the back of us
280 	 * in the chain of fragments so far.  We must know where to put
281 	 * this fragment, right?
282 	 */
283 	prev = fq->q.fragments_tail;
284 	if (!prev || FRAG6_CB(prev)->offset < offset) {
285 		next = NULL;
286 		goto found;
287 	}
288 	prev = NULL;
289 	for(next = fq->q.fragments; next != NULL; next = next->next) {
290 		if (FRAG6_CB(next)->offset >= offset)
291 			break;	/* bingo! */
292 		prev = next;
293 	}
294 
295 found:
296 	/* RFC5722, Section 4, amended by Errata ID : 3089
297 	 *                          When reassembling an IPv6 datagram, if
298 	 *   one or more its constituent fragments is determined to be an
299 	 *   overlapping fragment, the entire datagram (and any constituent
300 	 *   fragments) MUST be silently discarded.
301 	 */
302 
303 	/* Check for overlap with preceding fragment. */
304 	if (prev &&
305 	    (FRAG6_CB(prev)->offset + prev->len) > offset)
306 		goto discard_fq;
307 
308 	/* Look for overlap with succeeding segment. */
309 	if (next && FRAG6_CB(next)->offset < end)
310 		goto discard_fq;
311 
312 	FRAG6_CB(skb)->offset = offset;
313 
314 	/* Insert this fragment in the chain of fragments. */
315 	skb->next = next;
316 	if (!next)
317 		fq->q.fragments_tail = skb;
318 	if (prev)
319 		prev->next = skb;
320 	else
321 		fq->q.fragments = skb;
322 
323 	dev = skb->dev;
324 	if (dev) {
325 		fq->iif = dev->ifindex;
326 		skb->dev = NULL;
327 	}
328 	fq->q.stamp = skb->tstamp;
329 	fq->q.meat += skb->len;
330 	fq->ecn |= ecn;
331 	add_frag_mem_limit(&fq->q, skb->truesize);
332 
333 	/* The first fragment.
334 	 * nhoffset is obtained from the first fragment, of course.
335 	 */
336 	if (offset == 0) {
337 		fq->nhoffset = nhoff;
338 		fq->q.flags |= INET_FRAG_FIRST_IN;
339 	}
340 
341 	if (fq->q.flags == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
342 	    fq->q.meat == fq->q.len) {
343 		int res;
344 		unsigned long orefdst = skb->_skb_refdst;
345 
346 		skb->_skb_refdst = 0UL;
347 		res = ip6_frag_reasm(fq, prev, dev);
348 		skb->_skb_refdst = orefdst;
349 		return res;
350 	}
351 
352 	skb_dst_drop(skb);
353 	return -1;
354 
355 discard_fq:
356 	inet_frag_kill(&fq->q, &ip6_frags);
357 err:
358 	IP6_INC_STATS_BH(net, ip6_dst_idev(skb_dst(skb)),
359 			 IPSTATS_MIB_REASMFAILS);
360 	kfree_skb(skb);
361 	return -1;
362 }
363 
364 /*
365  *	Check if this packet is complete.
366  *	Returns NULL on failure by any reason, and pointer
367  *	to current nexthdr field in reassembled frame.
368  *
369  *	It is called with locked fq, and caller must check that
370  *	queue is eligible for reassembly i.e. it is not COMPLETE,
371  *	the last and the first frames arrived and all the bits are here.
372  */
373 static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *prev,
374 			  struct net_device *dev)
375 {
376 	struct net *net = container_of(fq->q.net, struct net, ipv6.frags);
377 	struct sk_buff *fp, *head = fq->q.fragments;
378 	int    payload_len;
379 	unsigned int nhoff;
380 	int sum_truesize;
381 	u8 ecn;
382 
383 	inet_frag_kill(&fq->q, &ip6_frags);
384 
385 	ecn = ip_frag_ecn_table[fq->ecn];
386 	if (unlikely(ecn == 0xff))
387 		goto out_fail;
388 
389 	/* Make the one we just received the head. */
390 	if (prev) {
391 		head = prev->next;
392 		fp = skb_clone(head, GFP_ATOMIC);
393 
394 		if (!fp)
395 			goto out_oom;
396 
397 		fp->next = head->next;
398 		if (!fp->next)
399 			fq->q.fragments_tail = fp;
400 		prev->next = fp;
401 
402 		skb_morph(head, fq->q.fragments);
403 		head->next = fq->q.fragments->next;
404 
405 		consume_skb(fq->q.fragments);
406 		fq->q.fragments = head;
407 	}
408 
409 	WARN_ON(head == NULL);
410 	WARN_ON(FRAG6_CB(head)->offset != 0);
411 
412 	/* Unfragmented part is taken from the first segment. */
413 	payload_len = ((head->data - skb_network_header(head)) -
414 		       sizeof(struct ipv6hdr) + fq->q.len -
415 		       sizeof(struct frag_hdr));
416 	if (payload_len > IPV6_MAXPLEN)
417 		goto out_oversize;
418 
419 	/* Head of list must not be cloned. */
420 	if (skb_unclone(head, GFP_ATOMIC))
421 		goto out_oom;
422 
423 	/* If the first fragment is fragmented itself, we split
424 	 * it to two chunks: the first with data and paged part
425 	 * and the second, holding only fragments. */
426 	if (skb_has_frag_list(head)) {
427 		struct sk_buff *clone;
428 		int i, plen = 0;
429 
430 		if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL)
431 			goto out_oom;
432 		clone->next = head->next;
433 		head->next = clone;
434 		skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
435 		skb_frag_list_init(head);
436 		for (i = 0; i < skb_shinfo(head)->nr_frags; i++)
437 			plen += skb_frag_size(&skb_shinfo(head)->frags[i]);
438 		clone->len = clone->data_len = head->data_len - plen;
439 		head->data_len -= clone->len;
440 		head->len -= clone->len;
441 		clone->csum = 0;
442 		clone->ip_summed = head->ip_summed;
443 		add_frag_mem_limit(&fq->q, clone->truesize);
444 	}
445 
446 	/* We have to remove fragment header from datagram and to relocate
447 	 * header in order to calculate ICV correctly. */
448 	nhoff = fq->nhoffset;
449 	skb_network_header(head)[nhoff] = skb_transport_header(head)[0];
450 	memmove(head->head + sizeof(struct frag_hdr), head->head,
451 		(head->data - head->head) - sizeof(struct frag_hdr));
452 	head->mac_header += sizeof(struct frag_hdr);
453 	head->network_header += sizeof(struct frag_hdr);
454 
455 	skb_reset_transport_header(head);
456 	skb_push(head, head->data - skb_network_header(head));
457 
458 	sum_truesize = head->truesize;
459 	for (fp = head->next; fp;) {
460 		bool headstolen;
461 		int delta;
462 		struct sk_buff *next = fp->next;
463 
464 		sum_truesize += fp->truesize;
465 		if (head->ip_summed != fp->ip_summed)
466 			head->ip_summed = CHECKSUM_NONE;
467 		else if (head->ip_summed == CHECKSUM_COMPLETE)
468 			head->csum = csum_add(head->csum, fp->csum);
469 
470 		if (skb_try_coalesce(head, fp, &headstolen, &delta)) {
471 			kfree_skb_partial(fp, headstolen);
472 		} else {
473 			if (!skb_shinfo(head)->frag_list)
474 				skb_shinfo(head)->frag_list = fp;
475 			head->data_len += fp->len;
476 			head->len += fp->len;
477 			head->truesize += fp->truesize;
478 		}
479 		fp = next;
480 	}
481 	sub_frag_mem_limit(&fq->q, sum_truesize);
482 
483 	head->next = NULL;
484 	head->dev = dev;
485 	head->tstamp = fq->q.stamp;
486 	ipv6_hdr(head)->payload_len = htons(payload_len);
487 	ipv6_change_dsfield(ipv6_hdr(head), 0xff, ecn);
488 	IP6CB(head)->nhoff = nhoff;
489 	IP6CB(head)->flags |= IP6SKB_FRAGMENTED;
490 
491 	/* Yes, and fold redundant checksum back. 8) */
492 	if (head->ip_summed == CHECKSUM_COMPLETE)
493 		head->csum = csum_partial(skb_network_header(head),
494 					  skb_network_header_len(head),
495 					  head->csum);
496 
497 	rcu_read_lock();
498 	IP6_INC_STATS_BH(net, __in6_dev_get(dev), IPSTATS_MIB_REASMOKS);
499 	rcu_read_unlock();
500 	fq->q.fragments = NULL;
501 	fq->q.fragments_tail = NULL;
502 	return 1;
503 
504 out_oversize:
505 	net_dbg_ratelimited("ip6_frag_reasm: payload len = %d\n", payload_len);
506 	goto out_fail;
507 out_oom:
508 	net_dbg_ratelimited("ip6_frag_reasm: no memory for reassembly\n");
509 out_fail:
510 	rcu_read_lock();
511 	IP6_INC_STATS_BH(net, __in6_dev_get(dev), IPSTATS_MIB_REASMFAILS);
512 	rcu_read_unlock();
513 	return -1;
514 }
515 
516 static int ipv6_frag_rcv(struct sk_buff *skb)
517 {
518 	struct frag_hdr *fhdr;
519 	struct frag_queue *fq;
520 	const struct ipv6hdr *hdr = ipv6_hdr(skb);
521 	struct net *net = dev_net(skb_dst(skb)->dev);
522 
523 	if (IP6CB(skb)->flags & IP6SKB_FRAGMENTED)
524 		goto fail_hdr;
525 
526 	IP6_INC_STATS_BH(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMREQDS);
527 
528 	/* Jumbo payload inhibits frag. header */
529 	if (hdr->payload_len==0)
530 		goto fail_hdr;
531 
532 	if (!pskb_may_pull(skb, (skb_transport_offset(skb) +
533 				 sizeof(struct frag_hdr))))
534 		goto fail_hdr;
535 
536 	hdr = ipv6_hdr(skb);
537 	fhdr = (struct frag_hdr *)skb_transport_header(skb);
538 
539 	if (!(fhdr->frag_off & htons(0xFFF9))) {
540 		/* It is not a fragmented frame */
541 		skb->transport_header += sizeof(struct frag_hdr);
542 		IP6_INC_STATS_BH(net,
543 				 ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMOKS);
544 
545 		IP6CB(skb)->nhoff = (u8 *)fhdr - skb_network_header(skb);
546 		IP6CB(skb)->flags |= IP6SKB_FRAGMENTED;
547 		return 1;
548 	}
549 
550 	fq = fq_find(net, fhdr->identification, &hdr->saddr, &hdr->daddr,
551 		     ip6_frag_ecn(hdr));
552 	if (fq != NULL) {
553 		int ret;
554 
555 		spin_lock(&fq->q.lock);
556 
557 		ret = ip6_frag_queue(fq, skb, fhdr, IP6CB(skb)->nhoff);
558 
559 		spin_unlock(&fq->q.lock);
560 		inet_frag_put(&fq->q, &ip6_frags);
561 		return ret;
562 	}
563 
564 	IP6_INC_STATS_BH(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMFAILS);
565 	kfree_skb(skb);
566 	return -1;
567 
568 fail_hdr:
569 	IP6_INC_STATS_BH(net, ip6_dst_idev(skb_dst(skb)),
570 			 IPSTATS_MIB_INHDRERRORS);
571 	icmpv6_param_prob(skb, ICMPV6_HDR_FIELD, skb_network_header_len(skb));
572 	return -1;
573 }
574 
575 static const struct inet6_protocol frag_protocol =
576 {
577 	.handler	=	ipv6_frag_rcv,
578 	.flags		=	INET6_PROTO_NOPOLICY,
579 };
580 
581 #ifdef CONFIG_SYSCTL
582 static int zero;
583 
584 static struct ctl_table ip6_frags_ns_ctl_table[] = {
585 	{
586 		.procname	= "ip6frag_high_thresh",
587 		.data		= &init_net.ipv6.frags.high_thresh,
588 		.maxlen		= sizeof(int),
589 		.mode		= 0644,
590 		.proc_handler	= proc_dointvec_minmax,
591 		.extra1		= &init_net.ipv6.frags.low_thresh
592 	},
593 	{
594 		.procname	= "ip6frag_low_thresh",
595 		.data		= &init_net.ipv6.frags.low_thresh,
596 		.maxlen		= sizeof(int),
597 		.mode		= 0644,
598 		.proc_handler	= proc_dointvec_minmax,
599 		.extra1		= &zero,
600 		.extra2		= &init_net.ipv6.frags.high_thresh
601 	},
602 	{
603 		.procname	= "ip6frag_time",
604 		.data		= &init_net.ipv6.frags.timeout,
605 		.maxlen		= sizeof(int),
606 		.mode		= 0644,
607 		.proc_handler	= proc_dointvec_jiffies,
608 	},
609 	{ }
610 };
611 
612 /* secret interval has been deprecated */
613 static int ip6_frags_secret_interval_unused;
614 static struct ctl_table ip6_frags_ctl_table[] = {
615 	{
616 		.procname	= "ip6frag_secret_interval",
617 		.data		= &ip6_frags_secret_interval_unused,
618 		.maxlen		= sizeof(int),
619 		.mode		= 0644,
620 		.proc_handler	= proc_dointvec_jiffies,
621 	},
622 	{ }
623 };
624 
625 static int __net_init ip6_frags_ns_sysctl_register(struct net *net)
626 {
627 	struct ctl_table *table;
628 	struct ctl_table_header *hdr;
629 
630 	table = ip6_frags_ns_ctl_table;
631 	if (!net_eq(net, &init_net)) {
632 		table = kmemdup(table, sizeof(ip6_frags_ns_ctl_table), GFP_KERNEL);
633 		if (table == NULL)
634 			goto err_alloc;
635 
636 		table[0].data = &net->ipv6.frags.high_thresh;
637 		table[0].extra1 = &net->ipv6.frags.low_thresh;
638 		table[0].extra2 = &init_net.ipv6.frags.high_thresh;
639 		table[1].data = &net->ipv6.frags.low_thresh;
640 		table[1].extra2 = &net->ipv6.frags.high_thresh;
641 		table[2].data = &net->ipv6.frags.timeout;
642 
643 		/* Don't export sysctls to unprivileged users */
644 		if (net->user_ns != &init_user_ns)
645 			table[0].procname = NULL;
646 	}
647 
648 	hdr = register_net_sysctl(net, "net/ipv6", table);
649 	if (hdr == NULL)
650 		goto err_reg;
651 
652 	net->ipv6.sysctl.frags_hdr = hdr;
653 	return 0;
654 
655 err_reg:
656 	if (!net_eq(net, &init_net))
657 		kfree(table);
658 err_alloc:
659 	return -ENOMEM;
660 }
661 
662 static void __net_exit ip6_frags_ns_sysctl_unregister(struct net *net)
663 {
664 	struct ctl_table *table;
665 
666 	table = net->ipv6.sysctl.frags_hdr->ctl_table_arg;
667 	unregister_net_sysctl_table(net->ipv6.sysctl.frags_hdr);
668 	if (!net_eq(net, &init_net))
669 		kfree(table);
670 }
671 
672 static struct ctl_table_header *ip6_ctl_header;
673 
674 static int ip6_frags_sysctl_register(void)
675 {
676 	ip6_ctl_header = register_net_sysctl(&init_net, "net/ipv6",
677 			ip6_frags_ctl_table);
678 	return ip6_ctl_header == NULL ? -ENOMEM : 0;
679 }
680 
681 static void ip6_frags_sysctl_unregister(void)
682 {
683 	unregister_net_sysctl_table(ip6_ctl_header);
684 }
685 #else
686 static inline int ip6_frags_ns_sysctl_register(struct net *net)
687 {
688 	return 0;
689 }
690 
691 static inline void ip6_frags_ns_sysctl_unregister(struct net *net)
692 {
693 }
694 
695 static inline int ip6_frags_sysctl_register(void)
696 {
697 	return 0;
698 }
699 
700 static inline void ip6_frags_sysctl_unregister(void)
701 {
702 }
703 #endif
704 
705 static int __net_init ipv6_frags_init_net(struct net *net)
706 {
707 	net->ipv6.frags.high_thresh = IPV6_FRAG_HIGH_THRESH;
708 	net->ipv6.frags.low_thresh = IPV6_FRAG_LOW_THRESH;
709 	net->ipv6.frags.timeout = IPV6_FRAG_TIMEOUT;
710 
711 	inet_frags_init_net(&net->ipv6.frags);
712 
713 	return ip6_frags_ns_sysctl_register(net);
714 }
715 
716 static void __net_exit ipv6_frags_exit_net(struct net *net)
717 {
718 	ip6_frags_ns_sysctl_unregister(net);
719 	inet_frags_exit_net(&net->ipv6.frags, &ip6_frags);
720 }
721 
722 static struct pernet_operations ip6_frags_ops = {
723 	.init = ipv6_frags_init_net,
724 	.exit = ipv6_frags_exit_net,
725 };
726 
727 int __init ipv6_frag_init(void)
728 {
729 	int ret;
730 
731 	ret = inet6_add_protocol(&frag_protocol, IPPROTO_FRAGMENT);
732 	if (ret)
733 		goto out;
734 
735 	ret = ip6_frags_sysctl_register();
736 	if (ret)
737 		goto err_sysctl;
738 
739 	ret = register_pernet_subsys(&ip6_frags_ops);
740 	if (ret)
741 		goto err_pernet;
742 
743 	ip6_frags.hashfn = ip6_hashfn;
744 	ip6_frags.constructor = ip6_frag_init;
745 	ip6_frags.destructor = NULL;
746 	ip6_frags.skb_free = NULL;
747 	ip6_frags.qsize = sizeof(struct frag_queue);
748 	ip6_frags.match = ip6_frag_match;
749 	ip6_frags.frag_expire = ip6_frag_expire;
750 	inet_frags_init(&ip6_frags);
751 out:
752 	return ret;
753 
754 err_pernet:
755 	ip6_frags_sysctl_unregister();
756 err_sysctl:
757 	inet6_del_protocol(&frag_protocol, IPPROTO_FRAGMENT);
758 	goto out;
759 }
760 
761 void ipv6_frag_exit(void)
762 {
763 	inet_frags_fini(&ip6_frags);
764 	ip6_frags_sysctl_unregister();
765 	unregister_pernet_subsys(&ip6_frags_ops);
766 	inet6_del_protocol(&frag_protocol, IPPROTO_FRAGMENT);
767 }
768