xref: /openbmc/linux/net/ipv6/reassembly.c (revision df687341) 
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/ipv6_frag.h>
61 #include <net/inet_ecn.h>
62 
63 static const char ip6_frag_cache_name[] = "ip6-frags";
64 
65 static u8 ip6_frag_ecn(const struct ipv6hdr *ipv6h)
66 {
67 	return 1 << (ipv6_get_dsfield(ipv6h) & INET_ECN_MASK);
68 }
69 
70 static struct inet_frags ip6_frags;
71 
72 static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *prev,
73 			  struct net_device *dev);
74 
75 static void ip6_frag_expire(struct timer_list *t)
76 {
77 	struct inet_frag_queue *frag = from_timer(frag, t, timer);
78 	struct frag_queue *fq;
79 	struct net *net;
80 
81 	fq = container_of(frag, struct frag_queue, q);
82 	net = container_of(fq->q.net, struct net, ipv6.frags);
83 
84 	ip6frag_expire_frag_queue(net, fq);
85 }
86 
87 static struct frag_queue *
88 fq_find(struct net *net, __be32 id, const struct ipv6hdr *hdr, int iif)
89 {
90 	struct frag_v6_compare_key key = {
91 		.id = id,
92 		.saddr = hdr->saddr,
93 		.daddr = hdr->daddr,
94 		.user = IP6_DEFRAG_LOCAL_DELIVER,
95 		.iif = iif,
96 	};
97 	struct inet_frag_queue *q;
98 
99 	if (!(ipv6_addr_type(&hdr->daddr) & (IPV6_ADDR_MULTICAST |
100 					    IPV6_ADDR_LINKLOCAL)))
101 		key.iif = 0;
102 
103 	q = inet_frag_find(&net->ipv6.frags, &key);
104 	if (!q)
105 		return NULL;
106 
107 	return container_of(q, struct frag_queue, q);
108 }
109 
110 static int ip6_frag_queue(struct frag_queue *fq, struct sk_buff *skb,
111 			  struct frag_hdr *fhdr, int nhoff,
112 			  u32 *prob_offset)
113 {
114 	struct sk_buff *prev, *next;
115 	struct net_device *dev;
116 	int offset, end, fragsize;
117 	struct net *net = dev_net(skb_dst(skb)->dev);
118 	u8 ecn;
119 
120 	if (fq->q.flags & INET_FRAG_COMPLETE)
121 		goto err;
122 
123 	offset = ntohs(fhdr->frag_off) & ~0x7;
124 	end = offset + (ntohs(ipv6_hdr(skb)->payload_len) -
125 			((u8 *)(fhdr + 1) - (u8 *)(ipv6_hdr(skb) + 1)));
126 
127 	if ((unsigned int)end > IPV6_MAXPLEN) {
128 		*prob_offset = (u8 *)&fhdr->frag_off - skb_network_header(skb);
129 		return -1;
130 	}
131 
132 	ecn = ip6_frag_ecn(ipv6_hdr(skb));
133 
134 	if (skb->ip_summed == CHECKSUM_COMPLETE) {
135 		const unsigned char *nh = skb_network_header(skb);
136 		skb->csum = csum_sub(skb->csum,
137 				     csum_partial(nh, (u8 *)(fhdr + 1) - nh,
138 						  0));
139 	}
140 
141 	/* Is this the final fragment? */
142 	if (!(fhdr->frag_off & htons(IP6_MF))) {
143 		/* If we already have some bits beyond end
144 		 * or have different end, the segment is corrupted.
145 		 */
146 		if (end < fq->q.len ||
147 		    ((fq->q.flags & INET_FRAG_LAST_IN) && end != fq->q.len))
148 			goto discard_fq;
149 		fq->q.flags |= INET_FRAG_LAST_IN;
150 		fq->q.len = end;
151 	} else {
152 		/* Check if the fragment is rounded to 8 bytes.
153 		 * Required by the RFC.
154 		 */
155 		if (end & 0x7) {
156 			/* RFC2460 says always send parameter problem in
157 			 * this case. -DaveM
158 			 */
159 			*prob_offset = offsetof(struct ipv6hdr, payload_len);
160 			return -1;
161 		}
162 		if (end > fq->q.len) {
163 			/* Some bits beyond end -> corruption. */
164 			if (fq->q.flags & INET_FRAG_LAST_IN)
165 				goto discard_fq;
166 			fq->q.len = end;
167 		}
168 	}
169 
170 	if (end == offset)
171 		goto discard_fq;
172 
173 	/* Point into the IP datagram 'data' part. */
174 	if (!pskb_pull(skb, (u8 *) (fhdr + 1) - skb->data))
175 		goto discard_fq;
176 
177 	if (pskb_trim_rcsum(skb, end - offset))
178 		goto discard_fq;
179 
180 	/* Find out which fragments are in front and at the back of us
181 	 * in the chain of fragments so far.  We must know where to put
182 	 * this fragment, right?
183 	 */
184 	prev = fq->q.fragments_tail;
185 	if (!prev || prev->ip_defrag_offset < offset) {
186 		next = NULL;
187 		goto found;
188 	}
189 	prev = NULL;
190 	for (next = fq->q.fragments; next != NULL; next = next->next) {
191 		if (next->ip_defrag_offset >= offset)
192 			break;	/* bingo! */
193 		prev = next;
194 	}
195 
196 found:
197 	/* RFC5722, Section 4, amended by Errata ID : 3089
198 	 *                          When reassembling an IPv6 datagram, if
199 	 *   one or more its constituent fragments is determined to be an
200 	 *   overlapping fragment, the entire datagram (and any constituent
201 	 *   fragments) MUST be silently discarded.
202 	 */
203 
204 	/* Check for overlap with preceding fragment. */
205 	if (prev &&
206 	    (prev->ip_defrag_offset + prev->len) > offset)
207 		goto discard_fq;
208 
209 	/* Look for overlap with succeeding segment. */
210 	if (next && next->ip_defrag_offset < end)
211 		goto discard_fq;
212 
213 	/* Note : skb->ip_defrag_offset and skb->sk share the same location */
214 	dev = skb->dev;
215 	if (dev)
216 		fq->iif = dev->ifindex;
217 	/* Makes sure compiler wont do silly aliasing games */
218 	barrier();
219 	skb->ip_defrag_offset = offset;
220 
221 	/* Insert this fragment in the chain of fragments. */
222 	skb->next = next;
223 	if (!next)
224 		fq->q.fragments_tail = skb;
225 	if (prev)
226 		prev->next = skb;
227 	else
228 		fq->q.fragments = skb;
229 
230 	fq->q.stamp = skb->tstamp;
231 	fq->q.meat += skb->len;
232 	fq->ecn |= ecn;
233 	add_frag_mem_limit(fq->q.net, skb->truesize);
234 
235 	fragsize = -skb_network_offset(skb) + skb->len;
236 	if (fragsize > fq->q.max_size)
237 		fq->q.max_size = fragsize;
238 
239 	/* The first fragment.
240 	 * nhoffset is obtained from the first fragment, of course.
241 	 */
242 	if (offset == 0) {
243 		fq->nhoffset = nhoff;
244 		fq->q.flags |= INET_FRAG_FIRST_IN;
245 	}
246 
247 	if (fq->q.flags == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
248 	    fq->q.meat == fq->q.len) {
249 		int res;
250 		unsigned long orefdst = skb->_skb_refdst;
251 
252 		skb->_skb_refdst = 0UL;
253 		res = ip6_frag_reasm(fq, prev, dev);
254 		skb->_skb_refdst = orefdst;
255 		return res;
256 	}
257 
258 	skb_dst_drop(skb);
259 	return -1;
260 
261 discard_fq:
262 	inet_frag_kill(&fq->q);
263 err:
264 	__IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
265 			IPSTATS_MIB_REASMFAILS);
266 	kfree_skb(skb);
267 	return -1;
268 }
269 
270 /*
271  *	Check if this packet is complete.
272  *	Returns NULL on failure by any reason, and pointer
273  *	to current nexthdr field in reassembled frame.
274  *
275  *	It is called with locked fq, and caller must check that
276  *	queue is eligible for reassembly i.e. it is not COMPLETE,
277  *	the last and the first frames arrived and all the bits are here.
278  */
279 static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *prev,
280 			  struct net_device *dev)
281 {
282 	struct net *net = container_of(fq->q.net, struct net, ipv6.frags);
283 	struct sk_buff *fp, *head = fq->q.fragments;
284 	int    payload_len, delta;
285 	unsigned int nhoff;
286 	int sum_truesize;
287 	u8 ecn;
288 
289 	inet_frag_kill(&fq->q);
290 
291 	ecn = ip_frag_ecn_table[fq->ecn];
292 	if (unlikely(ecn == 0xff))
293 		goto out_fail;
294 
295 	/* Make the one we just received the head. */
296 	if (prev) {
297 		head = prev->next;
298 		fp = skb_clone(head, GFP_ATOMIC);
299 
300 		if (!fp)
301 			goto out_oom;
302 
303 		fp->next = head->next;
304 		if (!fp->next)
305 			fq->q.fragments_tail = fp;
306 		prev->next = fp;
307 
308 		skb_morph(head, fq->q.fragments);
309 		head->next = fq->q.fragments->next;
310 
311 		consume_skb(fq->q.fragments);
312 		fq->q.fragments = head;
313 	}
314 
315 	WARN_ON(head == NULL);
316 	WARN_ON(head->ip_defrag_offset != 0);
317 
318 	/* Unfragmented part is taken from the first segment. */
319 	payload_len = ((head->data - skb_network_header(head)) -
320 		       sizeof(struct ipv6hdr) + fq->q.len -
321 		       sizeof(struct frag_hdr));
322 	if (payload_len > IPV6_MAXPLEN)
323 		goto out_oversize;
324 
325 	delta = - head->truesize;
326 
327 	/* Head of list must not be cloned. */
328 	if (skb_unclone(head, GFP_ATOMIC))
329 		goto out_oom;
330 
331 	delta += head->truesize;
332 	if (delta)
333 		add_frag_mem_limit(fq->q.net, delta);
334 
335 	/* If the first fragment is fragmented itself, we split
336 	 * it to two chunks: the first with data and paged part
337 	 * and the second, holding only fragments. */
338 	if (skb_has_frag_list(head)) {
339 		struct sk_buff *clone;
340 		int i, plen = 0;
341 
342 		clone = alloc_skb(0, GFP_ATOMIC);
343 		if (!clone)
344 			goto out_oom;
345 		clone->next = head->next;
346 		head->next = clone;
347 		skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
348 		skb_frag_list_init(head);
349 		for (i = 0; i < skb_shinfo(head)->nr_frags; i++)
350 			plen += skb_frag_size(&skb_shinfo(head)->frags[i]);
351 		clone->len = clone->data_len = head->data_len - plen;
352 		head->data_len -= clone->len;
353 		head->len -= clone->len;
354 		clone->csum = 0;
355 		clone->ip_summed = head->ip_summed;
356 		add_frag_mem_limit(fq->q.net, clone->truesize);
357 	}
358 
359 	/* We have to remove fragment header from datagram and to relocate
360 	 * header in order to calculate ICV correctly. */
361 	nhoff = fq->nhoffset;
362 	skb_network_header(head)[nhoff] = skb_transport_header(head)[0];
363 	memmove(head->head + sizeof(struct frag_hdr), head->head,
364 		(head->data - head->head) - sizeof(struct frag_hdr));
365 	if (skb_mac_header_was_set(head))
366 		head->mac_header += sizeof(struct frag_hdr);
367 	head->network_header += sizeof(struct frag_hdr);
368 
369 	skb_reset_transport_header(head);
370 	skb_push(head, head->data - skb_network_header(head));
371 
372 	sum_truesize = head->truesize;
373 	for (fp = head->next; fp;) {
374 		bool headstolen;
375 		int delta;
376 		struct sk_buff *next = fp->next;
377 
378 		sum_truesize += fp->truesize;
379 		if (head->ip_summed != fp->ip_summed)
380 			head->ip_summed = CHECKSUM_NONE;
381 		else if (head->ip_summed == CHECKSUM_COMPLETE)
382 			head->csum = csum_add(head->csum, fp->csum);
383 
384 		if (skb_try_coalesce(head, fp, &headstolen, &delta)) {
385 			kfree_skb_partial(fp, headstolen);
386 		} else {
387 			fp->sk = NULL;
388 			if (!skb_shinfo(head)->frag_list)
389 				skb_shinfo(head)->frag_list = fp;
390 			head->data_len += fp->len;
391 			head->len += fp->len;
392 			head->truesize += fp->truesize;
393 		}
394 		fp = next;
395 	}
396 	sub_frag_mem_limit(fq->q.net, sum_truesize);
397 
398 	skb_mark_not_on_list(head);
399 	head->dev = dev;
400 	head->tstamp = fq->q.stamp;
401 	ipv6_hdr(head)->payload_len = htons(payload_len);
402 	ipv6_change_dsfield(ipv6_hdr(head), 0xff, ecn);
403 	IP6CB(head)->nhoff = nhoff;
404 	IP6CB(head)->flags |= IP6SKB_FRAGMENTED;
405 	IP6CB(head)->frag_max_size = fq->q.max_size;
406 
407 	/* Yes, and fold redundant checksum back. 8) */
408 	skb_postpush_rcsum(head, skb_network_header(head),
409 			   skb_network_header_len(head));
410 
411 	rcu_read_lock();
412 	__IP6_INC_STATS(net, __in6_dev_get(dev), IPSTATS_MIB_REASMOKS);
413 	rcu_read_unlock();
414 	fq->q.fragments = NULL;
415 	fq->q.rb_fragments = RB_ROOT;
416 	fq->q.fragments_tail = NULL;
417 	return 1;
418 
419 out_oversize:
420 	net_dbg_ratelimited("ip6_frag_reasm: payload len = %d\n", payload_len);
421 	goto out_fail;
422 out_oom:
423 	net_dbg_ratelimited("ip6_frag_reasm: no memory for reassembly\n");
424 out_fail:
425 	rcu_read_lock();
426 	__IP6_INC_STATS(net, __in6_dev_get(dev), IPSTATS_MIB_REASMFAILS);
427 	rcu_read_unlock();
428 	inet_frag_kill(&fq->q);
429 	return -1;
430 }
431 
432 static int ipv6_frag_rcv(struct sk_buff *skb)
433 {
434 	struct frag_hdr *fhdr;
435 	struct frag_queue *fq;
436 	const struct ipv6hdr *hdr = ipv6_hdr(skb);
437 	struct net *net = dev_net(skb_dst(skb)->dev);
438 	int iif;
439 
440 	if (IP6CB(skb)->flags & IP6SKB_FRAGMENTED)
441 		goto fail_hdr;
442 
443 	__IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMREQDS);
444 
445 	/* Jumbo payload inhibits frag. header */
446 	if (hdr->payload_len == 0)
447 		goto fail_hdr;
448 
449 	if (!pskb_may_pull(skb, (skb_transport_offset(skb) +
450 				 sizeof(struct frag_hdr))))
451 		goto fail_hdr;
452 
453 	hdr = ipv6_hdr(skb);
454 	fhdr = (struct frag_hdr *)skb_transport_header(skb);
455 
456 	if (!(fhdr->frag_off & htons(0xFFF9))) {
457 		/* It is not a fragmented frame */
458 		skb->transport_header += sizeof(struct frag_hdr);
459 		__IP6_INC_STATS(net,
460 				ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMOKS);
461 
462 		IP6CB(skb)->nhoff = (u8 *)fhdr - skb_network_header(skb);
463 		IP6CB(skb)->flags |= IP6SKB_FRAGMENTED;
464 		return 1;
465 	}
466 
467 	if (skb->len - skb_network_offset(skb) < IPV6_MIN_MTU &&
468 	    fhdr->frag_off & htons(IP6_MF))
469 		goto fail_hdr;
470 
471 	iif = skb->dev ? skb->dev->ifindex : 0;
472 	fq = fq_find(net, fhdr->identification, hdr, iif);
473 	if (fq) {
474 		u32 prob_offset = 0;
475 		int ret;
476 
477 		spin_lock(&fq->q.lock);
478 
479 		fq->iif = iif;
480 		ret = ip6_frag_queue(fq, skb, fhdr, IP6CB(skb)->nhoff,
481 				     &prob_offset);
482 
483 		spin_unlock(&fq->q.lock);
484 		inet_frag_put(&fq->q);
485 		if (prob_offset) {
486 			__IP6_INC_STATS(net, __in6_dev_get_safely(skb->dev),
487 					IPSTATS_MIB_INHDRERRORS);
488 			icmpv6_param_prob(skb, ICMPV6_HDR_FIELD, prob_offset);
489 		}
490 		return ret;
491 	}
492 
493 	__IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMFAILS);
494 	kfree_skb(skb);
495 	return -1;
496 
497 fail_hdr:
498 	__IP6_INC_STATS(net, __in6_dev_get_safely(skb->dev),
499 			IPSTATS_MIB_INHDRERRORS);
500 	icmpv6_param_prob(skb, ICMPV6_HDR_FIELD, skb_network_header_len(skb));
501 	return -1;
502 }
503 
504 static const struct inet6_protocol frag_protocol = {
505 	.handler	=	ipv6_frag_rcv,
506 	.flags		=	INET6_PROTO_NOPOLICY,
507 };
508 
509 #ifdef CONFIG_SYSCTL
510 
511 static struct ctl_table ip6_frags_ns_ctl_table[] = {
512 	{
513 		.procname	= "ip6frag_high_thresh",
514 		.data		= &init_net.ipv6.frags.high_thresh,
515 		.maxlen		= sizeof(unsigned long),
516 		.mode		= 0644,
517 		.proc_handler	= proc_doulongvec_minmax,
518 		.extra1		= &init_net.ipv6.frags.low_thresh
519 	},
520 	{
521 		.procname	= "ip6frag_low_thresh",
522 		.data		= &init_net.ipv6.frags.low_thresh,
523 		.maxlen		= sizeof(unsigned long),
524 		.mode		= 0644,
525 		.proc_handler	= proc_doulongvec_minmax,
526 		.extra2		= &init_net.ipv6.frags.high_thresh
527 	},
528 	{
529 		.procname	= "ip6frag_time",
530 		.data		= &init_net.ipv6.frags.timeout,
531 		.maxlen		= sizeof(int),
532 		.mode		= 0644,
533 		.proc_handler	= proc_dointvec_jiffies,
534 	},
535 	{ }
536 };
537 
538 /* secret interval has been deprecated */
539 static int ip6_frags_secret_interval_unused;
540 static struct ctl_table ip6_frags_ctl_table[] = {
541 	{
542 		.procname	= "ip6frag_secret_interval",
543 		.data		= &ip6_frags_secret_interval_unused,
544 		.maxlen		= sizeof(int),
545 		.mode		= 0644,
546 		.proc_handler	= proc_dointvec_jiffies,
547 	},
548 	{ }
549 };
550 
551 static int __net_init ip6_frags_ns_sysctl_register(struct net *net)
552 {
553 	struct ctl_table *table;
554 	struct ctl_table_header *hdr;
555 
556 	table = ip6_frags_ns_ctl_table;
557 	if (!net_eq(net, &init_net)) {
558 		table = kmemdup(table, sizeof(ip6_frags_ns_ctl_table), GFP_KERNEL);
559 		if (!table)
560 			goto err_alloc;
561 
562 		table[0].data = &net->ipv6.frags.high_thresh;
563 		table[0].extra1 = &net->ipv6.frags.low_thresh;
564 		table[1].data = &net->ipv6.frags.low_thresh;
565 		table[1].extra2 = &net->ipv6.frags.high_thresh;
566 		table[2].data = &net->ipv6.frags.timeout;
567 	}
568 
569 	hdr = register_net_sysctl(net, "net/ipv6", table);
570 	if (!hdr)
571 		goto err_reg;
572 
573 	net->ipv6.sysctl.frags_hdr = hdr;
574 	return 0;
575 
576 err_reg:
577 	if (!net_eq(net, &init_net))
578 		kfree(table);
579 err_alloc:
580 	return -ENOMEM;
581 }
582 
583 static void __net_exit ip6_frags_ns_sysctl_unregister(struct net *net)
584 {
585 	struct ctl_table *table;
586 
587 	table = net->ipv6.sysctl.frags_hdr->ctl_table_arg;
588 	unregister_net_sysctl_table(net->ipv6.sysctl.frags_hdr);
589 	if (!net_eq(net, &init_net))
590 		kfree(table);
591 }
592 
593 static struct ctl_table_header *ip6_ctl_header;
594 
595 static int ip6_frags_sysctl_register(void)
596 {
597 	ip6_ctl_header = register_net_sysctl(&init_net, "net/ipv6",
598 			ip6_frags_ctl_table);
599 	return ip6_ctl_header == NULL ? -ENOMEM : 0;
600 }
601 
602 static void ip6_frags_sysctl_unregister(void)
603 {
604 	unregister_net_sysctl_table(ip6_ctl_header);
605 }
606 #else
607 static int ip6_frags_ns_sysctl_register(struct net *net)
608 {
609 	return 0;
610 }
611 
612 static void ip6_frags_ns_sysctl_unregister(struct net *net)
613 {
614 }
615 
616 static int ip6_frags_sysctl_register(void)
617 {
618 	return 0;
619 }
620 
621 static void ip6_frags_sysctl_unregister(void)
622 {
623 }
624 #endif
625 
626 static int __net_init ipv6_frags_init_net(struct net *net)
627 {
628 	int res;
629 
630 	net->ipv6.frags.high_thresh = IPV6_FRAG_HIGH_THRESH;
631 	net->ipv6.frags.low_thresh = IPV6_FRAG_LOW_THRESH;
632 	net->ipv6.frags.timeout = IPV6_FRAG_TIMEOUT;
633 	net->ipv6.frags.f = &ip6_frags;
634 
635 	res = inet_frags_init_net(&net->ipv6.frags);
636 	if (res < 0)
637 		return res;
638 
639 	res = ip6_frags_ns_sysctl_register(net);
640 	if (res < 0)
641 		inet_frags_exit_net(&net->ipv6.frags);
642 	return res;
643 }
644 
645 static void __net_exit ipv6_frags_exit_net(struct net *net)
646 {
647 	ip6_frags_ns_sysctl_unregister(net);
648 	inet_frags_exit_net(&net->ipv6.frags);
649 }
650 
651 static struct pernet_operations ip6_frags_ops = {
652 	.init = ipv6_frags_init_net,
653 	.exit = ipv6_frags_exit_net,
654 };
655 
656 static const struct rhashtable_params ip6_rhash_params = {
657 	.head_offset		= offsetof(struct inet_frag_queue, node),
658 	.hashfn			= ip6frag_key_hashfn,
659 	.obj_hashfn		= ip6frag_obj_hashfn,
660 	.obj_cmpfn		= ip6frag_obj_cmpfn,
661 	.automatic_shrinking	= true,
662 };
663 
664 int __init ipv6_frag_init(void)
665 {
666 	int ret;
667 
668 	ip6_frags.constructor = ip6frag_init;
669 	ip6_frags.destructor = NULL;
670 	ip6_frags.qsize = sizeof(struct frag_queue);
671 	ip6_frags.frag_expire = ip6_frag_expire;
672 	ip6_frags.frags_cache_name = ip6_frag_cache_name;
673 	ip6_frags.rhash_params = ip6_rhash_params;
674 	ret = inet_frags_init(&ip6_frags);
675 	if (ret)
676 		goto out;
677 
678 	ret = inet6_add_protocol(&frag_protocol, IPPROTO_FRAGMENT);
679 	if (ret)
680 		goto err_protocol;
681 
682 	ret = ip6_frags_sysctl_register();
683 	if (ret)
684 		goto err_sysctl;
685 
686 	ret = register_pernet_subsys(&ip6_frags_ops);
687 	if (ret)
688 		goto err_pernet;
689 
690 out:
691 	return ret;
692 
693 err_pernet:
694 	ip6_frags_sysctl_unregister();
695 err_sysctl:
696 	inet6_del_protocol(&frag_protocol, IPPROTO_FRAGMENT);
697 err_protocol:
698 	inet_frags_fini(&ip6_frags);
699 	goto out;
700 }
701 
702 void ipv6_frag_exit(void)
703 {
704 	inet_frags_fini(&ip6_frags);
705 	ip6_frags_sysctl_unregister();
706 	unregister_pernet_subsys(&ip6_frags_ops);
707 	inet6_del_protocol(&frag_protocol, IPPROTO_FRAGMENT);
708 }
709