xref: /openbmc/linux/net/ipv6/reassembly.c (revision 36c77782)
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 	u32 c;
89 
90 	net_get_random_once(&ip6_frags.rnd, sizeof(ip6_frags.rnd));
91 	c = jhash_3words(ipv6_addr_hash(saddr), ipv6_addr_hash(daddr),
92 			 (__force u32)id, ip6_frags.rnd);
93 
94 	return c & (INETFRAGS_HASHSZ - 1);
95 }
96 
97 static unsigned int ip6_hashfn(const struct inet_frag_queue *q)
98 {
99 	const struct frag_queue *fq;
100 
101 	fq = container_of(q, struct frag_queue, q);
102 	return inet6_hash_frag(fq->id, &fq->saddr, &fq->daddr);
103 }
104 
105 bool ip6_frag_match(const struct inet_frag_queue *q, const void *a)
106 {
107 	const struct frag_queue *fq;
108 	const struct ip6_create_arg *arg = a;
109 
110 	fq = container_of(q, struct frag_queue, q);
111 	return	fq->id == arg->id &&
112 		fq->user == arg->user &&
113 		ipv6_addr_equal(&fq->saddr, arg->src) &&
114 		ipv6_addr_equal(&fq->daddr, arg->dst);
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.last_in & 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_REASMTIMEOUT);
149 	IP6_INC_STATS_BH(net, __in6_dev_get(dev), IPSTATS_MIB_REASMFAILS);
150 
151 	/* Don't send error if the first segment did not arrive. */
152 	if (!(fq->q.last_in & INET_FRAG_FIRST_IN) || !fq->q.fragments)
153 		goto out_rcu_unlock;
154 
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 __inline__ 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 	read_lock(&ip6_frags.lock);
196 	hash = inet6_hash_frag(id, src, dst);
197 
198 	q = inet_frag_find(&net->ipv6.frags, &ip6_frags, &arg, hash);
199 	if (IS_ERR_OR_NULL(q)) {
200 		inet_frag_maybe_warn_overflow(q, pr_fmt());
201 		return NULL;
202 	}
203 	return container_of(q, struct frag_queue, q);
204 }
205 
206 static int ip6_frag_queue(struct frag_queue *fq, struct sk_buff *skb,
207 			   struct frag_hdr *fhdr, int nhoff)
208 {
209 	struct sk_buff *prev, *next;
210 	struct net_device *dev;
211 	int offset, end;
212 	struct net *net = dev_net(skb_dst(skb)->dev);
213 	u8 ecn;
214 
215 	if (fq->q.last_in & INET_FRAG_COMPLETE)
216 		goto err;
217 
218 	offset = ntohs(fhdr->frag_off) & ~0x7;
219 	end = offset + (ntohs(ipv6_hdr(skb)->payload_len) -
220 			((u8 *)(fhdr + 1) - (u8 *)(ipv6_hdr(skb) + 1)));
221 
222 	if ((unsigned int)end > IPV6_MAXPLEN) {
223 		IP6_INC_STATS_BH(net, ip6_dst_idev(skb_dst(skb)),
224 				 IPSTATS_MIB_INHDRERRORS);
225 		icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
226 				  ((u8 *)&fhdr->frag_off -
227 				   skb_network_header(skb)));
228 		return -1;
229 	}
230 
231 	ecn = ip6_frag_ecn(ipv6_hdr(skb));
232 
233 	if (skb->ip_summed == CHECKSUM_COMPLETE) {
234 		const unsigned char *nh = skb_network_header(skb);
235 		skb->csum = csum_sub(skb->csum,
236 				     csum_partial(nh, (u8 *)(fhdr + 1) - nh,
237 						  0));
238 	}
239 
240 	/* Is this the final fragment? */
241 	if (!(fhdr->frag_off & htons(IP6_MF))) {
242 		/* If we already have some bits beyond end
243 		 * or have different end, the segment is corrupted.
244 		 */
245 		if (end < fq->q.len ||
246 		    ((fq->q.last_in & INET_FRAG_LAST_IN) && end != fq->q.len))
247 			goto err;
248 		fq->q.last_in |= INET_FRAG_LAST_IN;
249 		fq->q.len = end;
250 	} else {
251 		/* Check if the fragment is rounded to 8 bytes.
252 		 * Required by the RFC.
253 		 */
254 		if (end & 0x7) {
255 			/* RFC2460 says always send parameter problem in
256 			 * this case. -DaveM
257 			 */
258 			IP6_INC_STATS_BH(net, ip6_dst_idev(skb_dst(skb)),
259 					 IPSTATS_MIB_INHDRERRORS);
260 			icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
261 					  offsetof(struct ipv6hdr, payload_len));
262 			return -1;
263 		}
264 		if (end > fq->q.len) {
265 			/* Some bits beyond end -> corruption. */
266 			if (fq->q.last_in & INET_FRAG_LAST_IN)
267 				goto err;
268 			fq->q.len = end;
269 		}
270 	}
271 
272 	if (end == offset)
273 		goto err;
274 
275 	/* Point into the IP datagram 'data' part. */
276 	if (!pskb_pull(skb, (u8 *) (fhdr + 1) - skb->data))
277 		goto err;
278 
279 	if (pskb_trim_rcsum(skb, end - offset))
280 		goto err;
281 
282 	/* Find out which fragments are in front and at the back of us
283 	 * in the chain of fragments so far.  We must know where to put
284 	 * this fragment, right?
285 	 */
286 	prev = fq->q.fragments_tail;
287 	if (!prev || FRAG6_CB(prev)->offset < offset) {
288 		next = NULL;
289 		goto found;
290 	}
291 	prev = NULL;
292 	for(next = fq->q.fragments; next != NULL; next = next->next) {
293 		if (FRAG6_CB(next)->offset >= offset)
294 			break;	/* bingo! */
295 		prev = next;
296 	}
297 
298 found:
299 	/* RFC5722, Section 4, amended by Errata ID : 3089
300 	 *                          When reassembling an IPv6 datagram, if
301 	 *   one or more its constituent fragments is determined to be an
302 	 *   overlapping fragment, the entire datagram (and any constituent
303 	 *   fragments) MUST be silently discarded.
304 	 */
305 
306 	/* Check for overlap with preceding fragment. */
307 	if (prev &&
308 	    (FRAG6_CB(prev)->offset + prev->len) > offset)
309 		goto discard_fq;
310 
311 	/* Look for overlap with succeeding segment. */
312 	if (next && FRAG6_CB(next)->offset < end)
313 		goto discard_fq;
314 
315 	FRAG6_CB(skb)->offset = offset;
316 
317 	/* Insert this fragment in the chain of fragments. */
318 	skb->next = next;
319 	if (!next)
320 		fq->q.fragments_tail = skb;
321 	if (prev)
322 		prev->next = skb;
323 	else
324 		fq->q.fragments = skb;
325 
326 	dev = skb->dev;
327 	if (dev) {
328 		fq->iif = dev->ifindex;
329 		skb->dev = NULL;
330 	}
331 	fq->q.stamp = skb->tstamp;
332 	fq->q.meat += skb->len;
333 	fq->ecn |= ecn;
334 	add_frag_mem_limit(&fq->q, skb->truesize);
335 
336 	/* The first fragment.
337 	 * nhoffset is obtained from the first fragment, of course.
338 	 */
339 	if (offset == 0) {
340 		fq->nhoffset = nhoff;
341 		fq->q.last_in |= INET_FRAG_FIRST_IN;
342 	}
343 
344 	if (fq->q.last_in == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
345 	    fq->q.meat == fq->q.len) {
346 		int res;
347 		unsigned long orefdst = skb->_skb_refdst;
348 
349 		skb->_skb_refdst = 0UL;
350 		res = ip6_frag_reasm(fq, prev, dev);
351 		skb->_skb_refdst = orefdst;
352 		return res;
353 	}
354 
355 	skb_dst_drop(skb);
356 	inet_frag_lru_move(&fq->q);
357 	return -1;
358 
359 discard_fq:
360 	inet_frag_kill(&fq->q, &ip6_frags);
361 err:
362 	IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
363 		      IPSTATS_MIB_REASMFAILS);
364 	kfree_skb(skb);
365 	return -1;
366 }
367 
368 /*
369  *	Check if this packet is complete.
370  *	Returns NULL on failure by any reason, and pointer
371  *	to current nexthdr field in reassembled frame.
372  *
373  *	It is called with locked fq, and caller must check that
374  *	queue is eligible for reassembly i.e. it is not COMPLETE,
375  *	the last and the first frames arrived and all the bits are here.
376  */
377 static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *prev,
378 			  struct net_device *dev)
379 {
380 	struct net *net = container_of(fq->q.net, struct net, ipv6.frags);
381 	struct sk_buff *fp, *head = fq->q.fragments;
382 	int    payload_len;
383 	unsigned int nhoff;
384 	int sum_truesize;
385 	u8 ecn;
386 
387 	inet_frag_kill(&fq->q, &ip6_frags);
388 
389 	ecn = ip_frag_ecn_table[fq->ecn];
390 	if (unlikely(ecn == 0xff))
391 		goto out_fail;
392 
393 	/* Make the one we just received the head. */
394 	if (prev) {
395 		head = prev->next;
396 		fp = skb_clone(head, GFP_ATOMIC);
397 
398 		if (!fp)
399 			goto out_oom;
400 
401 		fp->next = head->next;
402 		if (!fp->next)
403 			fq->q.fragments_tail = fp;
404 		prev->next = fp;
405 
406 		skb_morph(head, fq->q.fragments);
407 		head->next = fq->q.fragments->next;
408 
409 		consume_skb(fq->q.fragments);
410 		fq->q.fragments = head;
411 	}
412 
413 	WARN_ON(head == NULL);
414 	WARN_ON(FRAG6_CB(head)->offset != 0);
415 
416 	/* Unfragmented part is taken from the first segment. */
417 	payload_len = ((head->data - skb_network_header(head)) -
418 		       sizeof(struct ipv6hdr) + fq->q.len -
419 		       sizeof(struct frag_hdr));
420 	if (payload_len > IPV6_MAXPLEN)
421 		goto out_oversize;
422 
423 	/* Head of list must not be cloned. */
424 	if (skb_unclone(head, GFP_ATOMIC))
425 		goto out_oom;
426 
427 	/* If the first fragment is fragmented itself, we split
428 	 * it to two chunks: the first with data and paged part
429 	 * and the second, holding only fragments. */
430 	if (skb_has_frag_list(head)) {
431 		struct sk_buff *clone;
432 		int i, plen = 0;
433 
434 		if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL)
435 			goto out_oom;
436 		clone->next = head->next;
437 		head->next = clone;
438 		skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
439 		skb_frag_list_init(head);
440 		for (i = 0; i < skb_shinfo(head)->nr_frags; i++)
441 			plen += skb_frag_size(&skb_shinfo(head)->frags[i]);
442 		clone->len = clone->data_len = head->data_len - plen;
443 		head->data_len -= clone->len;
444 		head->len -= clone->len;
445 		clone->csum = 0;
446 		clone->ip_summed = head->ip_summed;
447 		add_frag_mem_limit(&fq->q, clone->truesize);
448 	}
449 
450 	/* We have to remove fragment header from datagram and to relocate
451 	 * header in order to calculate ICV correctly. */
452 	nhoff = fq->nhoffset;
453 	skb_network_header(head)[nhoff] = skb_transport_header(head)[0];
454 	memmove(head->head + sizeof(struct frag_hdr), head->head,
455 		(head->data - head->head) - sizeof(struct frag_hdr));
456 	head->mac_header += sizeof(struct frag_hdr);
457 	head->network_header += sizeof(struct frag_hdr);
458 
459 	skb_reset_transport_header(head);
460 	skb_push(head, head->data - skb_network_header(head));
461 
462 	sum_truesize = head->truesize;
463 	for (fp = head->next; fp;) {
464 		bool headstolen;
465 		int delta;
466 		struct sk_buff *next = fp->next;
467 
468 		sum_truesize += fp->truesize;
469 		if (head->ip_summed != fp->ip_summed)
470 			head->ip_summed = CHECKSUM_NONE;
471 		else if (head->ip_summed == CHECKSUM_COMPLETE)
472 			head->csum = csum_add(head->csum, fp->csum);
473 
474 		if (skb_try_coalesce(head, fp, &headstolen, &delta)) {
475 			kfree_skb_partial(fp, headstolen);
476 		} else {
477 			if (!skb_shinfo(head)->frag_list)
478 				skb_shinfo(head)->frag_list = fp;
479 			head->data_len += fp->len;
480 			head->len += fp->len;
481 			head->truesize += fp->truesize;
482 		}
483 		fp = next;
484 	}
485 	sub_frag_mem_limit(&fq->q, sum_truesize);
486 
487 	head->next = NULL;
488 	head->dev = dev;
489 	head->tstamp = fq->q.stamp;
490 	ipv6_hdr(head)->payload_len = htons(payload_len);
491 	ipv6_change_dsfield(ipv6_hdr(head), 0xff, ecn);
492 	IP6CB(head)->nhoff = nhoff;
493 	IP6CB(head)->flags |= IP6SKB_FRAGMENTED;
494 
495 	/* Yes, and fold redundant checksum back. 8) */
496 	if (head->ip_summed == CHECKSUM_COMPLETE)
497 		head->csum = csum_partial(skb_network_header(head),
498 					  skb_network_header_len(head),
499 					  head->csum);
500 
501 	rcu_read_lock();
502 	IP6_INC_STATS_BH(net, __in6_dev_get(dev), IPSTATS_MIB_REASMOKS);
503 	rcu_read_unlock();
504 	fq->q.fragments = NULL;
505 	fq->q.fragments_tail = NULL;
506 	return 1;
507 
508 out_oversize:
509 	net_dbg_ratelimited("ip6_frag_reasm: payload len = %d\n", payload_len);
510 	goto out_fail;
511 out_oom:
512 	net_dbg_ratelimited("ip6_frag_reasm: no memory for reassembly\n");
513 out_fail:
514 	rcu_read_lock();
515 	IP6_INC_STATS_BH(net, __in6_dev_get(dev), IPSTATS_MIB_REASMFAILS);
516 	rcu_read_unlock();
517 	return -1;
518 }
519 
520 static int ipv6_frag_rcv(struct sk_buff *skb)
521 {
522 	struct frag_hdr *fhdr;
523 	struct frag_queue *fq;
524 	const struct ipv6hdr *hdr = ipv6_hdr(skb);
525 	struct net *net = dev_net(skb_dst(skb)->dev);
526 	int evicted;
527 
528 	if (IP6CB(skb)->flags & IP6SKB_FRAGMENTED)
529 		goto fail_hdr;
530 
531 	IP6_INC_STATS_BH(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMREQDS);
532 
533 	/* Jumbo payload inhibits frag. header */
534 	if (hdr->payload_len==0)
535 		goto fail_hdr;
536 
537 	if (!pskb_may_pull(skb, (skb_transport_offset(skb) +
538 				 sizeof(struct frag_hdr))))
539 		goto fail_hdr;
540 
541 	hdr = ipv6_hdr(skb);
542 	fhdr = (struct frag_hdr *)skb_transport_header(skb);
543 
544 	if (!(fhdr->frag_off & htons(0xFFF9))) {
545 		/* It is not a fragmented frame */
546 		skb->transport_header += sizeof(struct frag_hdr);
547 		IP6_INC_STATS_BH(net,
548 				 ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMOKS);
549 
550 		IP6CB(skb)->nhoff = (u8 *)fhdr - skb_network_header(skb);
551 		IP6CB(skb)->flags |= IP6SKB_FRAGMENTED;
552 		return 1;
553 	}
554 
555 	evicted = inet_frag_evictor(&net->ipv6.frags, &ip6_frags, false);
556 	if (evicted)
557 		IP6_ADD_STATS_BH(net, ip6_dst_idev(skb_dst(skb)),
558 				 IPSTATS_MIB_REASMFAILS, evicted);
559 
560 	fq = fq_find(net, fhdr->identification, &hdr->saddr, &hdr->daddr,
561 		     ip6_frag_ecn(hdr));
562 	if (fq != NULL) {
563 		int ret;
564 
565 		spin_lock(&fq->q.lock);
566 
567 		ret = ip6_frag_queue(fq, skb, fhdr, IP6CB(skb)->nhoff);
568 
569 		spin_unlock(&fq->q.lock);
570 		inet_frag_put(&fq->q, &ip6_frags);
571 		return ret;
572 	}
573 
574 	IP6_INC_STATS_BH(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMFAILS);
575 	kfree_skb(skb);
576 	return -1;
577 
578 fail_hdr:
579 	IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_INHDRERRORS);
580 	icmpv6_param_prob(skb, ICMPV6_HDR_FIELD, skb_network_header_len(skb));
581 	return -1;
582 }
583 
584 static const struct inet6_protocol frag_protocol =
585 {
586 	.handler	=	ipv6_frag_rcv,
587 	.flags		=	INET6_PROTO_NOPOLICY,
588 };
589 
590 #ifdef CONFIG_SYSCTL
591 static struct ctl_table ip6_frags_ns_ctl_table[] = {
592 	{
593 		.procname	= "ip6frag_high_thresh",
594 		.data		= &init_net.ipv6.frags.high_thresh,
595 		.maxlen		= sizeof(int),
596 		.mode		= 0644,
597 		.proc_handler	= proc_dointvec
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
605 	},
606 	{
607 		.procname	= "ip6frag_time",
608 		.data		= &init_net.ipv6.frags.timeout,
609 		.maxlen		= sizeof(int),
610 		.mode		= 0644,
611 		.proc_handler	= proc_dointvec_jiffies,
612 	},
613 	{ }
614 };
615 
616 static struct ctl_table ip6_frags_ctl_table[] = {
617 	{
618 		.procname	= "ip6frag_secret_interval",
619 		.data		= &ip6_frags.secret_interval,
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[1].data = &net->ipv6.frags.low_thresh;
640 		table[2].data = &net->ipv6.frags.timeout;
641 
642 		/* Don't export sysctls to unprivileged users */
643 		if (net->user_ns != &init_user_ns)
644 			table[0].procname = NULL;
645 	}
646 
647 	hdr = register_net_sysctl(net, "net/ipv6", table);
648 	if (hdr == NULL)
649 		goto err_reg;
650 
651 	net->ipv6.sysctl.frags_hdr = hdr;
652 	return 0;
653 
654 err_reg:
655 	if (!net_eq(net, &init_net))
656 		kfree(table);
657 err_alloc:
658 	return -ENOMEM;
659 }
660 
661 static void __net_exit ip6_frags_ns_sysctl_unregister(struct net *net)
662 {
663 	struct ctl_table *table;
664 
665 	table = net->ipv6.sysctl.frags_hdr->ctl_table_arg;
666 	unregister_net_sysctl_table(net->ipv6.sysctl.frags_hdr);
667 	if (!net_eq(net, &init_net))
668 		kfree(table);
669 }
670 
671 static struct ctl_table_header *ip6_ctl_header;
672 
673 static int ip6_frags_sysctl_register(void)
674 {
675 	ip6_ctl_header = register_net_sysctl(&init_net, "net/ipv6",
676 			ip6_frags_ctl_table);
677 	return ip6_ctl_header == NULL ? -ENOMEM : 0;
678 }
679 
680 static void ip6_frags_sysctl_unregister(void)
681 {
682 	unregister_net_sysctl_table(ip6_ctl_header);
683 }
684 #else
685 static inline int ip6_frags_ns_sysctl_register(struct net *net)
686 {
687 	return 0;
688 }
689 
690 static inline void ip6_frags_ns_sysctl_unregister(struct net *net)
691 {
692 }
693 
694 static inline int ip6_frags_sysctl_register(void)
695 {
696 	return 0;
697 }
698 
699 static inline void ip6_frags_sysctl_unregister(void)
700 {
701 }
702 #endif
703 
704 static int __net_init ipv6_frags_init_net(struct net *net)
705 {
706 	net->ipv6.frags.high_thresh = IPV6_FRAG_HIGH_THRESH;
707 	net->ipv6.frags.low_thresh = IPV6_FRAG_LOW_THRESH;
708 	net->ipv6.frags.timeout = IPV6_FRAG_TIMEOUT;
709 
710 	inet_frags_init_net(&net->ipv6.frags);
711 
712 	return ip6_frags_ns_sysctl_register(net);
713 }
714 
715 static void __net_exit ipv6_frags_exit_net(struct net *net)
716 {
717 	ip6_frags_ns_sysctl_unregister(net);
718 	inet_frags_exit_net(&net->ipv6.frags, &ip6_frags);
719 }
720 
721 static struct pernet_operations ip6_frags_ops = {
722 	.init = ipv6_frags_init_net,
723 	.exit = ipv6_frags_exit_net,
724 };
725 
726 int __init ipv6_frag_init(void)
727 {
728 	int ret;
729 
730 	ret = inet6_add_protocol(&frag_protocol, IPPROTO_FRAGMENT);
731 	if (ret)
732 		goto out;
733 
734 	ret = ip6_frags_sysctl_register();
735 	if (ret)
736 		goto err_sysctl;
737 
738 	ret = register_pernet_subsys(&ip6_frags_ops);
739 	if (ret)
740 		goto err_pernet;
741 
742 	ip6_frags.hashfn = ip6_hashfn;
743 	ip6_frags.constructor = ip6_frag_init;
744 	ip6_frags.destructor = NULL;
745 	ip6_frags.skb_free = NULL;
746 	ip6_frags.qsize = sizeof(struct frag_queue);
747 	ip6_frags.match = ip6_frag_match;
748 	ip6_frags.frag_expire = ip6_frag_expire;
749 	ip6_frags.secret_interval = 10 * 60 * HZ;
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