xref: /linux/net/ipv6/reassembly.c (revision 5e8c0fb6a95728b852d56c0a9244425d474670c0) 
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 }
113 EXPORT_SYMBOL(ip6_frag_match);
114 
115 void ip6_frag_init(struct inet_frag_queue *q, const void *a)
116 {
117 	struct frag_queue *fq = container_of(q, struct frag_queue, q);
118 	const struct ip6_create_arg *arg = a;
119 
120 	fq->id = arg->id;
121 	fq->user = arg->user;
122 	fq->saddr = *arg->src;
123 	fq->daddr = *arg->dst;
124 	fq->ecn = arg->ecn;
125 }
126 EXPORT_SYMBOL(ip6_frag_init);
127 
128 void ip6_expire_frag_queue(struct net *net, struct frag_queue *fq,
129 			   struct inet_frags *frags)
130 {
131 	struct net_device *dev = NULL;
132 
133 	spin_lock(&fq->q.lock);
134 
135 	if (fq->q.flags & INET_FRAG_COMPLETE)
136 		goto out;
137 
138 	inet_frag_kill(&fq->q, frags);
139 
140 	rcu_read_lock();
141 	dev = dev_get_by_index_rcu(net, fq->iif);
142 	if (!dev)
143 		goto out_rcu_unlock;
144 
145 	IP6_INC_STATS_BH(net, __in6_dev_get(dev), IPSTATS_MIB_REASMFAILS);
146 
147 	if (fq->q.flags & INET_FRAG_EVICTED)
148 		goto out_rcu_unlock;
149 
150 	IP6_INC_STATS_BH(net, __in6_dev_get(dev), IPSTATS_MIB_REASMTIMEOUT);
151 
152 	/* Don't send error if the first segment did not arrive. */
153 	if (!(fq->q.flags & INET_FRAG_FIRST_IN) || !fq->q.fragments)
154 		goto out_rcu_unlock;
155 
156 	/* But use as source device on which LAST ARRIVED
157 	 * segment was received. And do not use fq->dev
158 	 * pointer directly, device might already disappeared.
159 	 */
160 	fq->q.fragments->dev = dev;
161 	icmpv6_send(fq->q.fragments, ICMPV6_TIME_EXCEED, ICMPV6_EXC_FRAGTIME, 0);
162 out_rcu_unlock:
163 	rcu_read_unlock();
164 out:
165 	spin_unlock(&fq->q.lock);
166 	inet_frag_put(&fq->q, frags);
167 }
168 EXPORT_SYMBOL(ip6_expire_frag_queue);
169 
170 static void ip6_frag_expire(unsigned long data)
171 {
172 	struct frag_queue *fq;
173 	struct net *net;
174 
175 	fq = container_of((struct inet_frag_queue *)data, struct frag_queue, q);
176 	net = container_of(fq->q.net, struct net, ipv6.frags);
177 
178 	ip6_expire_frag_queue(net, fq, &ip6_frags);
179 }
180 
181 static struct frag_queue *
182 fq_find(struct net *net, __be32 id, const struct in6_addr *src,
183 	const struct in6_addr *dst, u8 ecn)
184 {
185 	struct inet_frag_queue *q;
186 	struct ip6_create_arg arg;
187 	unsigned int hash;
188 
189 	arg.id = id;
190 	arg.user = IP6_DEFRAG_LOCAL_DELIVER;
191 	arg.src = src;
192 	arg.dst = dst;
193 	arg.ecn = ecn;
194 
195 	hash = inet6_hash_frag(id, src, dst);
196 
197 	q = inet_frag_find(&net->ipv6.frags, &ip6_frags, &arg, hash);
198 	if (IS_ERR_OR_NULL(q)) {
199 		inet_frag_maybe_warn_overflow(q, pr_fmt());
200 		return NULL;
201 	}
202 	return container_of(q, struct frag_queue, q);
203 }
204 
205 static int ip6_frag_queue(struct frag_queue *fq, struct sk_buff *skb,
206 			   struct frag_hdr *fhdr, int nhoff)
207 {
208 	struct sk_buff *prev, *next;
209 	struct net_device *dev;
210 	int offset, end;
211 	struct net *net = dev_net(skb_dst(skb)->dev);
212 	u8 ecn;
213 
214 	if (fq->q.flags & INET_FRAG_COMPLETE)
215 		goto err;
216 
217 	offset = ntohs(fhdr->frag_off) & ~0x7;
218 	end = offset + (ntohs(ipv6_hdr(skb)->payload_len) -
219 			((u8 *)(fhdr + 1) - (u8 *)(ipv6_hdr(skb) + 1)));
220 
221 	if ((unsigned int)end > IPV6_MAXPLEN) {
222 		IP6_INC_STATS_BH(net, ip6_dst_idev(skb_dst(skb)),
223 				 IPSTATS_MIB_INHDRERRORS);
224 		icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
225 				  ((u8 *)&fhdr->frag_off -
226 				   skb_network_header(skb)));
227 		return -1;
228 	}
229 
230 	ecn = ip6_frag_ecn(ipv6_hdr(skb));
231 
232 	if (skb->ip_summed == CHECKSUM_COMPLETE) {
233 		const unsigned char *nh = skb_network_header(skb);
234 		skb->csum = csum_sub(skb->csum,
235 				     csum_partial(nh, (u8 *)(fhdr + 1) - nh,
236 						  0));
237 	}
238 
239 	/* Is this the final fragment? */
240 	if (!(fhdr->frag_off & htons(IP6_MF))) {
241 		/* If we already have some bits beyond end
242 		 * or have different end, the segment is corrupted.
243 		 */
244 		if (end < fq->q.len ||
245 		    ((fq->q.flags & INET_FRAG_LAST_IN) && end != fq->q.len))
246 			goto err;
247 		fq->q.flags |= INET_FRAG_LAST_IN;
248 		fq->q.len = end;
249 	} else {
250 		/* Check if the fragment is rounded to 8 bytes.
251 		 * Required by the RFC.
252 		 */
253 		if (end & 0x7) {
254 			/* RFC2460 says always send parameter problem in
255 			 * this case. -DaveM
256 			 */
257 			IP6_INC_STATS_BH(net, ip6_dst_idev(skb_dst(skb)),
258 					 IPSTATS_MIB_INHDRERRORS);
259 			icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
260 					  offsetof(struct ipv6hdr, payload_len));
261 			return -1;
262 		}
263 		if (end > fq->q.len) {
264 			/* Some bits beyond end -> corruption. */
265 			if (fq->q.flags & INET_FRAG_LAST_IN)
266 				goto err;
267 			fq->q.len = end;
268 		}
269 	}
270 
271 	if (end == offset)
272 		goto err;
273 
274 	/* Point into the IP datagram 'data' part. */
275 	if (!pskb_pull(skb, (u8 *) (fhdr + 1) - skb->data))
276 		goto err;
277 
278 	if (pskb_trim_rcsum(skb, end - offset))
279 		goto err;
280 
281 	/* Find out which fragments are in front and at the back of us
282 	 * in the chain of fragments so far.  We must know where to put
283 	 * this fragment, right?
284 	 */
285 	prev = fq->q.fragments_tail;
286 	if (!prev || FRAG6_CB(prev)->offset < offset) {
287 		next = NULL;
288 		goto found;
289 	}
290 	prev = NULL;
291 	for (next = fq->q.fragments; next != NULL; next = next->next) {
292 		if (FRAG6_CB(next)->offset >= offset)
293 			break;	/* bingo! */
294 		prev = next;
295 	}
296 
297 found:
298 	/* RFC5722, Section 4, amended by Errata ID : 3089
299 	 *                          When reassembling an IPv6 datagram, if
300 	 *   one or more its constituent fragments is determined to be an
301 	 *   overlapping fragment, the entire datagram (and any constituent
302 	 *   fragments) MUST be silently discarded.
303 	 */
304 
305 	/* Check for overlap with preceding fragment. */
306 	if (prev &&
307 	    (FRAG6_CB(prev)->offset + prev->len) > offset)
308 		goto discard_fq;
309 
310 	/* Look for overlap with succeeding segment. */
311 	if (next && FRAG6_CB(next)->offset < end)
312 		goto discard_fq;
313 
314 	FRAG6_CB(skb)->offset = offset;
315 
316 	/* Insert this fragment in the chain of fragments. */
317 	skb->next = next;
318 	if (!next)
319 		fq->q.fragments_tail = skb;
320 	if (prev)
321 		prev->next = skb;
322 	else
323 		fq->q.fragments = skb;
324 
325 	dev = skb->dev;
326 	if (dev) {
327 		fq->iif = dev->ifindex;
328 		skb->dev = NULL;
329 	}
330 	fq->q.stamp = skb->tstamp;
331 	fq->q.meat += skb->len;
332 	fq->ecn |= ecn;
333 	add_frag_mem_limit(&fq->q, skb->truesize);
334 
335 	/* The first fragment.
336 	 * nhoffset is obtained from the first fragment, of course.
337 	 */
338 	if (offset == 0) {
339 		fq->nhoffset = nhoff;
340 		fq->q.flags |= INET_FRAG_FIRST_IN;
341 	}
342 
343 	if (fq->q.flags == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
344 	    fq->q.meat == fq->q.len) {
345 		int res;
346 		unsigned long orefdst = skb->_skb_refdst;
347 
348 		skb->_skb_refdst = 0UL;
349 		res = ip6_frag_reasm(fq, prev, dev);
350 		skb->_skb_refdst = orefdst;
351 		return res;
352 	}
353 
354 	skb_dst_drop(skb);
355 	return -1;
356 
357 discard_fq:
358 	inet_frag_kill(&fq->q, &ip6_frags);
359 err:
360 	IP6_INC_STATS_BH(net, ip6_dst_idev(skb_dst(skb)),
361 			 IPSTATS_MIB_REASMFAILS);
362 	kfree_skb(skb);
363 	return -1;
364 }
365 
366 /*
367  *	Check if this packet is complete.
368  *	Returns NULL on failure by any reason, and pointer
369  *	to current nexthdr field in reassembled frame.
370  *
371  *	It is called with locked fq, and caller must check that
372  *	queue is eligible for reassembly i.e. it is not COMPLETE,
373  *	the last and the first frames arrived and all the bits are here.
374  */
375 static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *prev,
376 			  struct net_device *dev)
377 {
378 	struct net *net = container_of(fq->q.net, struct net, ipv6.frags);
379 	struct sk_buff *fp, *head = fq->q.fragments;
380 	int    payload_len;
381 	unsigned int nhoff;
382 	int sum_truesize;
383 	u8 ecn;
384 
385 	inet_frag_kill(&fq->q, &ip6_frags);
386 
387 	ecn = ip_frag_ecn_table[fq->ecn];
388 	if (unlikely(ecn == 0xff))
389 		goto out_fail;
390 
391 	/* Make the one we just received the head. */
392 	if (prev) {
393 		head = prev->next;
394 		fp = skb_clone(head, GFP_ATOMIC);
395 
396 		if (!fp)
397 			goto out_oom;
398 
399 		fp->next = head->next;
400 		if (!fp->next)
401 			fq->q.fragments_tail = fp;
402 		prev->next = fp;
403 
404 		skb_morph(head, fq->q.fragments);
405 		head->next = fq->q.fragments->next;
406 
407 		consume_skb(fq->q.fragments);
408 		fq->q.fragments = head;
409 	}
410 
411 	WARN_ON(head == NULL);
412 	WARN_ON(FRAG6_CB(head)->offset != 0);
413 
414 	/* Unfragmented part is taken from the first segment. */
415 	payload_len = ((head->data - skb_network_header(head)) -
416 		       sizeof(struct ipv6hdr) + fq->q.len -
417 		       sizeof(struct frag_hdr));
418 	if (payload_len > IPV6_MAXPLEN)
419 		goto out_oversize;
420 
421 	/* Head of list must not be cloned. */
422 	if (skb_unclone(head, GFP_ATOMIC))
423 		goto out_oom;
424 
425 	/* If the first fragment is fragmented itself, we split
426 	 * it to two chunks: the first with data and paged part
427 	 * and the second, holding only fragments. */
428 	if (skb_has_frag_list(head)) {
429 		struct sk_buff *clone;
430 		int i, plen = 0;
431 
432 		clone = alloc_skb(0, GFP_ATOMIC);
433 		if (clone == NULL)
434 			goto out_oom;
435 		clone->next = head->next;
436 		head->next = clone;
437 		skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
438 		skb_frag_list_init(head);
439 		for (i = 0; i < skb_shinfo(head)->nr_frags; i++)
440 			plen += skb_frag_size(&skb_shinfo(head)->frags[i]);
441 		clone->len = clone->data_len = head->data_len - plen;
442 		head->data_len -= clone->len;
443 		head->len -= clone->len;
444 		clone->csum = 0;
445 		clone->ip_summed = head->ip_summed;
446 		add_frag_mem_limit(&fq->q, clone->truesize);
447 	}
448 
449 	/* We have to remove fragment header from datagram and to relocate
450 	 * header in order to calculate ICV correctly. */
451 	nhoff = fq->nhoffset;
452 	skb_network_header(head)[nhoff] = skb_transport_header(head)[0];
453 	memmove(head->head + sizeof(struct frag_hdr), head->head,
454 		(head->data - head->head) - sizeof(struct frag_hdr));
455 	head->mac_header += sizeof(struct frag_hdr);
456 	head->network_header += sizeof(struct frag_hdr);
457 
458 	skb_reset_transport_header(head);
459 	skb_push(head, head->data - skb_network_header(head));
460 
461 	sum_truesize = head->truesize;
462 	for (fp = head->next; fp;) {
463 		bool headstolen;
464 		int delta;
465 		struct sk_buff *next = fp->next;
466 
467 		sum_truesize += fp->truesize;
468 		if (head->ip_summed != fp->ip_summed)
469 			head->ip_summed = CHECKSUM_NONE;
470 		else if (head->ip_summed == CHECKSUM_COMPLETE)
471 			head->csum = csum_add(head->csum, fp->csum);
472 
473 		if (skb_try_coalesce(head, fp, &headstolen, &delta)) {
474 			kfree_skb_partial(fp, headstolen);
475 		} else {
476 			if (!skb_shinfo(head)->frag_list)
477 				skb_shinfo(head)->frag_list = fp;
478 			head->data_len += fp->len;
479 			head->len += fp->len;
480 			head->truesize += fp->truesize;
481 		}
482 		fp = next;
483 	}
484 	sub_frag_mem_limit(&fq->q, sum_truesize);
485 
486 	head->next = NULL;
487 	head->dev = dev;
488 	head->tstamp = fq->q.stamp;
489 	ipv6_hdr(head)->payload_len = htons(payload_len);
490 	ipv6_change_dsfield(ipv6_hdr(head), 0xff, ecn);
491 	IP6CB(head)->nhoff = nhoff;
492 	IP6CB(head)->flags |= IP6SKB_FRAGMENTED;
493 
494 	/* Yes, and fold redundant checksum back. 8) */
495 	if (head->ip_summed == CHECKSUM_COMPLETE)
496 		head->csum = csum_partial(skb_network_header(head),
497 					  skb_network_header_len(head),
498 					  head->csum);
499 
500 	rcu_read_lock();
501 	IP6_INC_STATS_BH(net, __in6_dev_get(dev), IPSTATS_MIB_REASMOKS);
502 	rcu_read_unlock();
503 	fq->q.fragments = NULL;
504 	fq->q.fragments_tail = NULL;
505 	return 1;
506 
507 out_oversize:
508 	net_dbg_ratelimited("ip6_frag_reasm: payload len = %d\n", payload_len);
509 	goto out_fail;
510 out_oom:
511 	net_dbg_ratelimited("ip6_frag_reasm: no memory for reassembly\n");
512 out_fail:
513 	rcu_read_lock();
514 	IP6_INC_STATS_BH(net, __in6_dev_get(dev), IPSTATS_MIB_REASMFAILS);
515 	rcu_read_unlock();
516 	return -1;
517 }
518 
519 static int ipv6_frag_rcv(struct sk_buff *skb)
520 {
521 	struct frag_hdr *fhdr;
522 	struct frag_queue *fq;
523 	const struct ipv6hdr *hdr = ipv6_hdr(skb);
524 	struct net *net = dev_net(skb_dst(skb)->dev);
525 
526 	if (IP6CB(skb)->flags & IP6SKB_FRAGMENTED)
527 		goto fail_hdr;
528 
529 	IP6_INC_STATS_BH(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMREQDS);
530 
531 	/* Jumbo payload inhibits frag. header */
532 	if (hdr->payload_len == 0)
533 		goto fail_hdr;
534 
535 	if (!pskb_may_pull(skb, (skb_transport_offset(skb) +
536 				 sizeof(struct frag_hdr))))
537 		goto fail_hdr;
538 
539 	hdr = ipv6_hdr(skb);
540 	fhdr = (struct frag_hdr *)skb_transport_header(skb);
541 
542 	if (!(fhdr->frag_off & htons(0xFFF9))) {
543 		/* It is not a fragmented frame */
544 		skb->transport_header += sizeof(struct frag_hdr);
545 		IP6_INC_STATS_BH(net,
546 				 ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMOKS);
547 
548 		IP6CB(skb)->nhoff = (u8 *)fhdr - skb_network_header(skb);
549 		IP6CB(skb)->flags |= IP6SKB_FRAGMENTED;
550 		return 1;
551 	}
552 
553 	fq = fq_find(net, fhdr->identification, &hdr->saddr, &hdr->daddr,
554 		     ip6_frag_ecn(hdr));
555 	if (fq != NULL) {
556 		int ret;
557 
558 		spin_lock(&fq->q.lock);
559 
560 		ret = ip6_frag_queue(fq, skb, fhdr, IP6CB(skb)->nhoff);
561 
562 		spin_unlock(&fq->q.lock);
563 		inet_frag_put(&fq->q, &ip6_frags);
564 		return ret;
565 	}
566 
567 	IP6_INC_STATS_BH(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMFAILS);
568 	kfree_skb(skb);
569 	return -1;
570 
571 fail_hdr:
572 	IP6_INC_STATS_BH(net, ip6_dst_idev(skb_dst(skb)),
573 			 IPSTATS_MIB_INHDRERRORS);
574 	icmpv6_param_prob(skb, ICMPV6_HDR_FIELD, skb_network_header_len(skb));
575 	return -1;
576 }
577 
578 static const struct inet6_protocol frag_protocol = {
579 	.handler	=	ipv6_frag_rcv,
580 	.flags		=	INET6_PROTO_NOPOLICY,
581 };
582 
583 #ifdef CONFIG_SYSCTL
584 static int zero;
585 
586 static struct ctl_table ip6_frags_ns_ctl_table[] = {
587 	{
588 		.procname	= "ip6frag_high_thresh",
589 		.data		= &init_net.ipv6.frags.high_thresh,
590 		.maxlen		= sizeof(int),
591 		.mode		= 0644,
592 		.proc_handler	= proc_dointvec_minmax,
593 		.extra1		= &init_net.ipv6.frags.low_thresh
594 	},
595 	{
596 		.procname	= "ip6frag_low_thresh",
597 		.data		= &init_net.ipv6.frags.low_thresh,
598 		.maxlen		= sizeof(int),
599 		.mode		= 0644,
600 		.proc_handler	= proc_dointvec_minmax,
601 		.extra1		= &zero,
602 		.extra2		= &init_net.ipv6.frags.high_thresh
603 	},
604 	{
605 		.procname	= "ip6frag_time",
606 		.data		= &init_net.ipv6.frags.timeout,
607 		.maxlen		= sizeof(int),
608 		.mode		= 0644,
609 		.proc_handler	= proc_dointvec_jiffies,
610 	},
611 	{ }
612 };
613 
614 /* secret interval has been deprecated */
615 static int ip6_frags_secret_interval_unused;
616 static struct ctl_table ip6_frags_ctl_table[] = {
617 	{
618 		.procname	= "ip6frag_secret_interval",
619 		.data		= &ip6_frags_secret_interval_unused,
620 		.maxlen		= sizeof(int),
621 		.mode		= 0644,
622 		.proc_handler	= proc_dointvec_jiffies,
623 	},
624 	{ }
625 };
626 
627 static int __net_init ip6_frags_ns_sysctl_register(struct net *net)
628 {
629 	struct ctl_table *table;
630 	struct ctl_table_header *hdr;
631 
632 	table = ip6_frags_ns_ctl_table;
633 	if (!net_eq(net, &init_net)) {
634 		table = kmemdup(table, sizeof(ip6_frags_ns_ctl_table), GFP_KERNEL);
635 		if (table == NULL)
636 			goto err_alloc;
637 
638 		table[0].data = &net->ipv6.frags.high_thresh;
639 		table[0].extra1 = &net->ipv6.frags.low_thresh;
640 		table[0].extra2 = &init_net.ipv6.frags.high_thresh;
641 		table[1].data = &net->ipv6.frags.low_thresh;
642 		table[1].extra2 = &net->ipv6.frags.high_thresh;
643 		table[2].data = &net->ipv6.frags.timeout;
644 
645 		/* Don't export sysctls to unprivileged users */
646 		if (net->user_ns != &init_user_ns)
647 			table[0].procname = NULL;
648 	}
649 
650 	hdr = register_net_sysctl(net, "net/ipv6", table);
651 	if (hdr == NULL)
652 		goto err_reg;
653 
654 	net->ipv6.sysctl.frags_hdr = hdr;
655 	return 0;
656 
657 err_reg:
658 	if (!net_eq(net, &init_net))
659 		kfree(table);
660 err_alloc:
661 	return -ENOMEM;
662 }
663 
664 static void __net_exit ip6_frags_ns_sysctl_unregister(struct net *net)
665 {
666 	struct ctl_table *table;
667 
668 	table = net->ipv6.sysctl.frags_hdr->ctl_table_arg;
669 	unregister_net_sysctl_table(net->ipv6.sysctl.frags_hdr);
670 	if (!net_eq(net, &init_net))
671 		kfree(table);
672 }
673 
674 static struct ctl_table_header *ip6_ctl_header;
675 
676 static int ip6_frags_sysctl_register(void)
677 {
678 	ip6_ctl_header = register_net_sysctl(&init_net, "net/ipv6",
679 			ip6_frags_ctl_table);
680 	return ip6_ctl_header == NULL ? -ENOMEM : 0;
681 }
682 
683 static void ip6_frags_sysctl_unregister(void)
684 {
685 	unregister_net_sysctl_table(ip6_ctl_header);
686 }
687 #else
688 static int ip6_frags_ns_sysctl_register(struct net *net)
689 {
690 	return 0;
691 }
692 
693 static void ip6_frags_ns_sysctl_unregister(struct net *net)
694 {
695 }
696 
697 static int ip6_frags_sysctl_register(void)
698 {
699 	return 0;
700 }
701 
702 static void ip6_frags_sysctl_unregister(void)
703 {
704 }
705 #endif
706 
707 static int __net_init ipv6_frags_init_net(struct net *net)
708 {
709 	net->ipv6.frags.high_thresh = IPV6_FRAG_HIGH_THRESH;
710 	net->ipv6.frags.low_thresh = IPV6_FRAG_LOW_THRESH;
711 	net->ipv6.frags.timeout = IPV6_FRAG_TIMEOUT;
712 
713 	inet_frags_init_net(&net->ipv6.frags);
714 
715 	return ip6_frags_ns_sysctl_register(net);
716 }
717 
718 static void __net_exit ipv6_frags_exit_net(struct net *net)
719 {
720 	ip6_frags_ns_sysctl_unregister(net);
721 	inet_frags_exit_net(&net->ipv6.frags, &ip6_frags);
722 }
723 
724 static struct pernet_operations ip6_frags_ops = {
725 	.init = ipv6_frags_init_net,
726 	.exit = ipv6_frags_exit_net,
727 };
728 
729 int __init ipv6_frag_init(void)
730 {
731 	int ret;
732 
733 	ret = inet6_add_protocol(&frag_protocol, IPPROTO_FRAGMENT);
734 	if (ret)
735 		goto out;
736 
737 	ret = ip6_frags_sysctl_register();
738 	if (ret)
739 		goto err_sysctl;
740 
741 	ret = register_pernet_subsys(&ip6_frags_ops);
742 	if (ret)
743 		goto err_pernet;
744 
745 	ip6_frags.hashfn = ip6_hashfn;
746 	ip6_frags.constructor = ip6_frag_init;
747 	ip6_frags.destructor = NULL;
748 	ip6_frags.skb_free = NULL;
749 	ip6_frags.qsize = sizeof(struct frag_queue);
750 	ip6_frags.match = ip6_frag_match;
751 	ip6_frags.frag_expire = ip6_frag_expire;
752 	ip6_frags.frags_cache_name = ip6_frag_cache_name;
753 	ret = inet_frags_init(&ip6_frags);
754 	if (ret)
755 		goto err_pernet;
756 out:
757 	return ret;
758 
759 err_pernet:
760 	ip6_frags_sysctl_unregister();
761 err_sysctl:
762 	inet6_del_protocol(&frag_protocol, IPPROTO_FRAGMENT);
763 	goto out;
764 }
765 
766 void ipv6_frag_exit(void)
767 {
768 	inet_frags_fini(&ip6_frags);
769 	ip6_frags_sysctl_unregister();
770 	unregister_pernet_subsys(&ip6_frags_ops);
771 	inet6_del_protocol(&frag_protocol, IPPROTO_FRAGMENT);
772 }
773