xref: /linux/net/ipv6/reassembly.c (revision fd639726bf15fca8ee1a00dce8e0096d0ad9bd18)
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(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(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(struct timer_list *t)
174 {
175 	struct inet_frag_queue *frag = from_timer(frag, t, timer);
176 	struct frag_queue *fq;
177 	struct net *net;
178 
179 	fq = container_of(frag, struct frag_queue, q);
180 	net = container_of(fq->q.net, struct net, ipv6.frags);
181 
182 	ip6_expire_frag_queue(net, fq, &ip6_frags);
183 }
184 
185 static struct frag_queue *
186 fq_find(struct net *net, __be32 id, const struct in6_addr *src,
187 	const struct in6_addr *dst, int iif, u8 ecn)
188 {
189 	struct inet_frag_queue *q;
190 	struct ip6_create_arg arg;
191 	unsigned int hash;
192 
193 	arg.id = id;
194 	arg.user = IP6_DEFRAG_LOCAL_DELIVER;
195 	arg.src = src;
196 	arg.dst = dst;
197 	arg.iif = iif;
198 	arg.ecn = ecn;
199 
200 	hash = inet6_hash_frag(id, src, dst);
201 
202 	q = inet_frag_find(&net->ipv6.frags, &ip6_frags, &arg, hash);
203 	if (IS_ERR_OR_NULL(q)) {
204 		inet_frag_maybe_warn_overflow(q, pr_fmt());
205 		return NULL;
206 	}
207 	return container_of(q, struct frag_queue, q);
208 }
209 
210 static int ip6_frag_queue(struct frag_queue *fq, struct sk_buff *skb,
211 			   struct frag_hdr *fhdr, int nhoff)
212 {
213 	struct sk_buff *prev, *next;
214 	struct net_device *dev;
215 	int offset, end, fragsize;
216 	struct net *net = dev_net(skb_dst(skb)->dev);
217 	u8 ecn;
218 
219 	if (fq->q.flags & INET_FRAG_COMPLETE)
220 		goto err;
221 
222 	offset = ntohs(fhdr->frag_off) & ~0x7;
223 	end = offset + (ntohs(ipv6_hdr(skb)->payload_len) -
224 			((u8 *)(fhdr + 1) - (u8 *)(ipv6_hdr(skb) + 1)));
225 
226 	if ((unsigned int)end > IPV6_MAXPLEN) {
227 		__IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
228 				IPSTATS_MIB_INHDRERRORS);
229 		icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
230 				  ((u8 *)&fhdr->frag_off -
231 				   skb_network_header(skb)));
232 		return -1;
233 	}
234 
235 	ecn = ip6_frag_ecn(ipv6_hdr(skb));
236 
237 	if (skb->ip_summed == CHECKSUM_COMPLETE) {
238 		const unsigned char *nh = skb_network_header(skb);
239 		skb->csum = csum_sub(skb->csum,
240 				     csum_partial(nh, (u8 *)(fhdr + 1) - nh,
241 						  0));
242 	}
243 
244 	/* Is this the final fragment? */
245 	if (!(fhdr->frag_off & htons(IP6_MF))) {
246 		/* If we already have some bits beyond end
247 		 * or have different end, the segment is corrupted.
248 		 */
249 		if (end < fq->q.len ||
250 		    ((fq->q.flags & INET_FRAG_LAST_IN) && end != fq->q.len))
251 			goto err;
252 		fq->q.flags |= INET_FRAG_LAST_IN;
253 		fq->q.len = end;
254 	} else {
255 		/* Check if the fragment is rounded to 8 bytes.
256 		 * Required by the RFC.
257 		 */
258 		if (end & 0x7) {
259 			/* RFC2460 says always send parameter problem in
260 			 * this case. -DaveM
261 			 */
262 			__IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
263 					IPSTATS_MIB_INHDRERRORS);
264 			icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
265 					  offsetof(struct ipv6hdr, payload_len));
266 			return -1;
267 		}
268 		if (end > fq->q.len) {
269 			/* Some bits beyond end -> corruption. */
270 			if (fq->q.flags & INET_FRAG_LAST_IN)
271 				goto err;
272 			fq->q.len = end;
273 		}
274 	}
275 
276 	if (end == offset)
277 		goto err;
278 
279 	/* Point into the IP datagram 'data' part. */
280 	if (!pskb_pull(skb, (u8 *) (fhdr + 1) - skb->data))
281 		goto err;
282 
283 	if (pskb_trim_rcsum(skb, end - offset))
284 		goto err;
285 
286 	/* Find out which fragments are in front and at the back of us
287 	 * in the chain of fragments so far.  We must know where to put
288 	 * this fragment, right?
289 	 */
290 	prev = fq->q.fragments_tail;
291 	if (!prev || FRAG6_CB(prev)->offset < offset) {
292 		next = NULL;
293 		goto found;
294 	}
295 	prev = NULL;
296 	for (next = fq->q.fragments; next != NULL; next = next->next) {
297 		if (FRAG6_CB(next)->offset >= offset)
298 			break;	/* bingo! */
299 		prev = next;
300 	}
301 
302 found:
303 	/* RFC5722, Section 4, amended by Errata ID : 3089
304 	 *                          When reassembling an IPv6 datagram, if
305 	 *   one or more its constituent fragments is determined to be an
306 	 *   overlapping fragment, the entire datagram (and any constituent
307 	 *   fragments) MUST be silently discarded.
308 	 */
309 
310 	/* Check for overlap with preceding fragment. */
311 	if (prev &&
312 	    (FRAG6_CB(prev)->offset + prev->len) > offset)
313 		goto discard_fq;
314 
315 	/* Look for overlap with succeeding segment. */
316 	if (next && FRAG6_CB(next)->offset < end)
317 		goto discard_fq;
318 
319 	FRAG6_CB(skb)->offset = offset;
320 
321 	/* Insert this fragment in the chain of fragments. */
322 	skb->next = next;
323 	if (!next)
324 		fq->q.fragments_tail = skb;
325 	if (prev)
326 		prev->next = skb;
327 	else
328 		fq->q.fragments = skb;
329 
330 	dev = skb->dev;
331 	if (dev) {
332 		fq->iif = dev->ifindex;
333 		skb->dev = NULL;
334 	}
335 	fq->q.stamp = skb->tstamp;
336 	fq->q.meat += skb->len;
337 	fq->ecn |= ecn;
338 	add_frag_mem_limit(fq->q.net, skb->truesize);
339 
340 	fragsize = -skb_network_offset(skb) + skb->len;
341 	if (fragsize > fq->q.max_size)
342 		fq->q.max_size = fragsize;
343 
344 	/* The first fragment.
345 	 * nhoffset is obtained from the first fragment, of course.
346 	 */
347 	if (offset == 0) {
348 		fq->nhoffset = nhoff;
349 		fq->q.flags |= INET_FRAG_FIRST_IN;
350 	}
351 
352 	if (fq->q.flags == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
353 	    fq->q.meat == fq->q.len) {
354 		int res;
355 		unsigned long orefdst = skb->_skb_refdst;
356 
357 		skb->_skb_refdst = 0UL;
358 		res = ip6_frag_reasm(fq, prev, dev);
359 		skb->_skb_refdst = orefdst;
360 		return res;
361 	}
362 
363 	skb_dst_drop(skb);
364 	return -1;
365 
366 discard_fq:
367 	inet_frag_kill(&fq->q, &ip6_frags);
368 err:
369 	__IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
370 			IPSTATS_MIB_REASMFAILS);
371 	kfree_skb(skb);
372 	return -1;
373 }
374 
375 /*
376  *	Check if this packet is complete.
377  *	Returns NULL on failure by any reason, and pointer
378  *	to current nexthdr field in reassembled frame.
379  *
380  *	It is called with locked fq, and caller must check that
381  *	queue is eligible for reassembly i.e. it is not COMPLETE,
382  *	the last and the first frames arrived and all the bits are here.
383  */
384 static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *prev,
385 			  struct net_device *dev)
386 {
387 	struct net *net = container_of(fq->q.net, struct net, ipv6.frags);
388 	struct sk_buff *fp, *head = fq->q.fragments;
389 	int    payload_len;
390 	unsigned int nhoff;
391 	int sum_truesize;
392 	u8 ecn;
393 
394 	inet_frag_kill(&fq->q, &ip6_frags);
395 
396 	ecn = ip_frag_ecn_table[fq->ecn];
397 	if (unlikely(ecn == 0xff))
398 		goto out_fail;
399 
400 	/* Make the one we just received the head. */
401 	if (prev) {
402 		head = prev->next;
403 		fp = skb_clone(head, GFP_ATOMIC);
404 
405 		if (!fp)
406 			goto out_oom;
407 
408 		fp->next = head->next;
409 		if (!fp->next)
410 			fq->q.fragments_tail = fp;
411 		prev->next = fp;
412 
413 		skb_morph(head, fq->q.fragments);
414 		head->next = fq->q.fragments->next;
415 
416 		consume_skb(fq->q.fragments);
417 		fq->q.fragments = head;
418 	}
419 
420 	WARN_ON(head == NULL);
421 	WARN_ON(FRAG6_CB(head)->offset != 0);
422 
423 	/* Unfragmented part is taken from the first segment. */
424 	payload_len = ((head->data - skb_network_header(head)) -
425 		       sizeof(struct ipv6hdr) + fq->q.len -
426 		       sizeof(struct frag_hdr));
427 	if (payload_len > IPV6_MAXPLEN)
428 		goto out_oversize;
429 
430 	/* Head of list must not be cloned. */
431 	if (skb_unclone(head, GFP_ATOMIC))
432 		goto out_oom;
433 
434 	/* If the first fragment is fragmented itself, we split
435 	 * it to two chunks: the first with data and paged part
436 	 * and the second, holding only fragments. */
437 	if (skb_has_frag_list(head)) {
438 		struct sk_buff *clone;
439 		int i, plen = 0;
440 
441 		clone = alloc_skb(0, GFP_ATOMIC);
442 		if (!clone)
443 			goto out_oom;
444 		clone->next = head->next;
445 		head->next = clone;
446 		skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
447 		skb_frag_list_init(head);
448 		for (i = 0; i < skb_shinfo(head)->nr_frags; i++)
449 			plen += skb_frag_size(&skb_shinfo(head)->frags[i]);
450 		clone->len = clone->data_len = head->data_len - plen;
451 		head->data_len -= clone->len;
452 		head->len -= clone->len;
453 		clone->csum = 0;
454 		clone->ip_summed = head->ip_summed;
455 		add_frag_mem_limit(fq->q.net, clone->truesize);
456 	}
457 
458 	/* We have to remove fragment header from datagram and to relocate
459 	 * header in order to calculate ICV correctly. */
460 	nhoff = fq->nhoffset;
461 	skb_network_header(head)[nhoff] = skb_transport_header(head)[0];
462 	memmove(head->head + sizeof(struct frag_hdr), head->head,
463 		(head->data - head->head) - sizeof(struct frag_hdr));
464 	if (skb_mac_header_was_set(head))
465 		head->mac_header += sizeof(struct frag_hdr);
466 	head->network_header += sizeof(struct frag_hdr);
467 
468 	skb_reset_transport_header(head);
469 	skb_push(head, head->data - skb_network_header(head));
470 
471 	sum_truesize = head->truesize;
472 	for (fp = head->next; fp;) {
473 		bool headstolen;
474 		int delta;
475 		struct sk_buff *next = fp->next;
476 
477 		sum_truesize += fp->truesize;
478 		if (head->ip_summed != fp->ip_summed)
479 			head->ip_summed = CHECKSUM_NONE;
480 		else if (head->ip_summed == CHECKSUM_COMPLETE)
481 			head->csum = csum_add(head->csum, fp->csum);
482 
483 		if (skb_try_coalesce(head, fp, &headstolen, &delta)) {
484 			kfree_skb_partial(fp, headstolen);
485 		} else {
486 			if (!skb_shinfo(head)->frag_list)
487 				skb_shinfo(head)->frag_list = fp;
488 			head->data_len += fp->len;
489 			head->len += fp->len;
490 			head->truesize += fp->truesize;
491 		}
492 		fp = next;
493 	}
494 	sub_frag_mem_limit(fq->q.net, sum_truesize);
495 
496 	head->next = NULL;
497 	head->dev = dev;
498 	head->tstamp = fq->q.stamp;
499 	ipv6_hdr(head)->payload_len = htons(payload_len);
500 	ipv6_change_dsfield(ipv6_hdr(head), 0xff, ecn);
501 	IP6CB(head)->nhoff = nhoff;
502 	IP6CB(head)->flags |= IP6SKB_FRAGMENTED;
503 	IP6CB(head)->frag_max_size = fq->q.max_size;
504 
505 	/* Yes, and fold redundant checksum back. 8) */
506 	skb_postpush_rcsum(head, skb_network_header(head),
507 			   skb_network_header_len(head));
508 
509 	rcu_read_lock();
510 	__IP6_INC_STATS(net, __in6_dev_get(dev), IPSTATS_MIB_REASMOKS);
511 	rcu_read_unlock();
512 	fq->q.fragments = NULL;
513 	fq->q.fragments_tail = NULL;
514 	return 1;
515 
516 out_oversize:
517 	net_dbg_ratelimited("ip6_frag_reasm: payload len = %d\n", payload_len);
518 	goto out_fail;
519 out_oom:
520 	net_dbg_ratelimited("ip6_frag_reasm: no memory for reassembly\n");
521 out_fail:
522 	rcu_read_lock();
523 	__IP6_INC_STATS(net, __in6_dev_get(dev), IPSTATS_MIB_REASMFAILS);
524 	rcu_read_unlock();
525 	return -1;
526 }
527 
528 static int ipv6_frag_rcv(struct sk_buff *skb)
529 {
530 	struct frag_hdr *fhdr;
531 	struct frag_queue *fq;
532 	const struct ipv6hdr *hdr = ipv6_hdr(skb);
533 	struct net *net = dev_net(skb_dst(skb)->dev);
534 
535 	if (IP6CB(skb)->flags & IP6SKB_FRAGMENTED)
536 		goto fail_hdr;
537 
538 	__IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMREQDS);
539 
540 	/* Jumbo payload inhibits frag. header */
541 	if (hdr->payload_len == 0)
542 		goto fail_hdr;
543 
544 	if (!pskb_may_pull(skb, (skb_transport_offset(skb) +
545 				 sizeof(struct frag_hdr))))
546 		goto fail_hdr;
547 
548 	hdr = ipv6_hdr(skb);
549 	fhdr = (struct frag_hdr *)skb_transport_header(skb);
550 
551 	if (!(fhdr->frag_off & htons(0xFFF9))) {
552 		/* It is not a fragmented frame */
553 		skb->transport_header += sizeof(struct frag_hdr);
554 		__IP6_INC_STATS(net,
555 				ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMOKS);
556 
557 		IP6CB(skb)->nhoff = (u8 *)fhdr - skb_network_header(skb);
558 		IP6CB(skb)->flags |= IP6SKB_FRAGMENTED;
559 		return 1;
560 	}
561 
562 	fq = fq_find(net, fhdr->identification, &hdr->saddr, &hdr->daddr,
563 		     skb->dev ? skb->dev->ifindex : 0, ip6_frag_ecn(hdr));
564 	if (fq) {
565 		int ret;
566 
567 		spin_lock(&fq->q.lock);
568 
569 		ret = ip6_frag_queue(fq, skb, fhdr, IP6CB(skb)->nhoff);
570 
571 		spin_unlock(&fq->q.lock);
572 		inet_frag_put(&fq->q, &ip6_frags);
573 		return ret;
574 	}
575 
576 	__IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMFAILS);
577 	kfree_skb(skb);
578 	return -1;
579 
580 fail_hdr:
581 	__IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
582 			IPSTATS_MIB_INHDRERRORS);
583 	icmpv6_param_prob(skb, ICMPV6_HDR_FIELD, skb_network_header_len(skb));
584 	return -1;
585 }
586 
587 static const struct inet6_protocol frag_protocol = {
588 	.handler	=	ipv6_frag_rcv,
589 	.flags		=	INET6_PROTO_NOPOLICY,
590 };
591 
592 #ifdef CONFIG_SYSCTL
593 static int zero;
594 
595 static struct ctl_table ip6_frags_ns_ctl_table[] = {
596 	{
597 		.procname	= "ip6frag_high_thresh",
598 		.data		= &init_net.ipv6.frags.high_thresh,
599 		.maxlen		= sizeof(int),
600 		.mode		= 0644,
601 		.proc_handler	= proc_dointvec_minmax,
602 		.extra1		= &init_net.ipv6.frags.low_thresh
603 	},
604 	{
605 		.procname	= "ip6frag_low_thresh",
606 		.data		= &init_net.ipv6.frags.low_thresh,
607 		.maxlen		= sizeof(int),
608 		.mode		= 0644,
609 		.proc_handler	= proc_dointvec_minmax,
610 		.extra1		= &zero,
611 		.extra2		= &init_net.ipv6.frags.high_thresh
612 	},
613 	{
614 		.procname	= "ip6frag_time",
615 		.data		= &init_net.ipv6.frags.timeout,
616 		.maxlen		= sizeof(int),
617 		.mode		= 0644,
618 		.proc_handler	= proc_dointvec_jiffies,
619 	},
620 	{ }
621 };
622 
623 /* secret interval has been deprecated */
624 static int ip6_frags_secret_interval_unused;
625 static struct ctl_table ip6_frags_ctl_table[] = {
626 	{
627 		.procname	= "ip6frag_secret_interval",
628 		.data		= &ip6_frags_secret_interval_unused,
629 		.maxlen		= sizeof(int),
630 		.mode		= 0644,
631 		.proc_handler	= proc_dointvec_jiffies,
632 	},
633 	{ }
634 };
635 
636 static int __net_init ip6_frags_ns_sysctl_register(struct net *net)
637 {
638 	struct ctl_table *table;
639 	struct ctl_table_header *hdr;
640 
641 	table = ip6_frags_ns_ctl_table;
642 	if (!net_eq(net, &init_net)) {
643 		table = kmemdup(table, sizeof(ip6_frags_ns_ctl_table), GFP_KERNEL);
644 		if (!table)
645 			goto err_alloc;
646 
647 		table[0].data = &net->ipv6.frags.high_thresh;
648 		table[0].extra1 = &net->ipv6.frags.low_thresh;
649 		table[0].extra2 = &init_net.ipv6.frags.high_thresh;
650 		table[1].data = &net->ipv6.frags.low_thresh;
651 		table[1].extra2 = &net->ipv6.frags.high_thresh;
652 		table[2].data = &net->ipv6.frags.timeout;
653 
654 		/* Don't export sysctls to unprivileged users */
655 		if (net->user_ns != &init_user_ns)
656 			table[0].procname = NULL;
657 	}
658 
659 	hdr = register_net_sysctl(net, "net/ipv6", table);
660 	if (!hdr)
661 		goto err_reg;
662 
663 	net->ipv6.sysctl.frags_hdr = hdr;
664 	return 0;
665 
666 err_reg:
667 	if (!net_eq(net, &init_net))
668 		kfree(table);
669 err_alloc:
670 	return -ENOMEM;
671 }
672 
673 static void __net_exit ip6_frags_ns_sysctl_unregister(struct net *net)
674 {
675 	struct ctl_table *table;
676 
677 	table = net->ipv6.sysctl.frags_hdr->ctl_table_arg;
678 	unregister_net_sysctl_table(net->ipv6.sysctl.frags_hdr);
679 	if (!net_eq(net, &init_net))
680 		kfree(table);
681 }
682 
683 static struct ctl_table_header *ip6_ctl_header;
684 
685 static int ip6_frags_sysctl_register(void)
686 {
687 	ip6_ctl_header = register_net_sysctl(&init_net, "net/ipv6",
688 			ip6_frags_ctl_table);
689 	return ip6_ctl_header == NULL ? -ENOMEM : 0;
690 }
691 
692 static void ip6_frags_sysctl_unregister(void)
693 {
694 	unregister_net_sysctl_table(ip6_ctl_header);
695 }
696 #else
697 static int ip6_frags_ns_sysctl_register(struct net *net)
698 {
699 	return 0;
700 }
701 
702 static void ip6_frags_ns_sysctl_unregister(struct net *net)
703 {
704 }
705 
706 static int ip6_frags_sysctl_register(void)
707 {
708 	return 0;
709 }
710 
711 static void ip6_frags_sysctl_unregister(void)
712 {
713 }
714 #endif
715 
716 static int __net_init ipv6_frags_init_net(struct net *net)
717 {
718 	net->ipv6.frags.high_thresh = IPV6_FRAG_HIGH_THRESH;
719 	net->ipv6.frags.low_thresh = IPV6_FRAG_LOW_THRESH;
720 	net->ipv6.frags.timeout = IPV6_FRAG_TIMEOUT;
721 
722 	inet_frags_init_net(&net->ipv6.frags);
723 
724 	return ip6_frags_ns_sysctl_register(net);
725 }
726 
727 static void __net_exit ipv6_frags_exit_net(struct net *net)
728 {
729 	ip6_frags_ns_sysctl_unregister(net);
730 	inet_frags_exit_net(&net->ipv6.frags, &ip6_frags);
731 }
732 
733 static struct pernet_operations ip6_frags_ops = {
734 	.init = ipv6_frags_init_net,
735 	.exit = ipv6_frags_exit_net,
736 };
737 
738 int __init ipv6_frag_init(void)
739 {
740 	int ret;
741 
742 	ret = inet6_add_protocol(&frag_protocol, IPPROTO_FRAGMENT);
743 	if (ret)
744 		goto out;
745 
746 	ret = ip6_frags_sysctl_register();
747 	if (ret)
748 		goto err_sysctl;
749 
750 	ret = register_pernet_subsys(&ip6_frags_ops);
751 	if (ret)
752 		goto err_pernet;
753 
754 	ip6_frags.hashfn = ip6_hashfn;
755 	ip6_frags.constructor = ip6_frag_init;
756 	ip6_frags.destructor = NULL;
757 	ip6_frags.qsize = sizeof(struct frag_queue);
758 	ip6_frags.match = ip6_frag_match;
759 	ip6_frags.frag_expire = ip6_frag_expire;
760 	ip6_frags.frags_cache_name = ip6_frag_cache_name;
761 	ret = inet_frags_init(&ip6_frags);
762 	if (ret)
763 		goto err_pernet;
764 out:
765 	return ret;
766 
767 err_pernet:
768 	ip6_frags_sysctl_unregister();
769 err_sysctl:
770 	inet6_del_protocol(&frag_protocol, IPPROTO_FRAGMENT);
771 	goto out;
772 }
773 
774 void ipv6_frag_exit(void)
775 {
776 	inet_frags_fini(&ip6_frags);
777 	ip6_frags_sysctl_unregister();
778 	unregister_pernet_subsys(&ip6_frags_ops);
779 	inet6_del_protocol(&frag_protocol, IPPROTO_FRAGMENT);
780 }
781