• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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 
ip6_frag_ecn(const struct ipv6hdr * ipv6h)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 
ip6_frag_init(struct inet_frag_queue * q,const void * a)82 void ip6_frag_init(struct inet_frag_queue *q, const void *a)
83 {
84 	struct frag_queue *fq = container_of(q, struct frag_queue, q);
85 	const struct frag_v6_compare_key *key = a;
86 
87 	q->key.v6 = *key;
88 	fq->ecn = 0;
89 }
90 EXPORT_SYMBOL(ip6_frag_init);
91 
ip6_expire_frag_queue(struct net * net,struct frag_queue * fq)92 void ip6_expire_frag_queue(struct net *net, struct frag_queue *fq)
93 {
94 	struct net_device *dev = NULL;
95 	struct sk_buff *head;
96 
97 	rcu_read_lock();
98 	spin_lock(&fq->q.lock);
99 
100 	if (fq->q.flags & INET_FRAG_COMPLETE)
101 		goto out;
102 
103 	inet_frag_kill(&fq->q);
104 
105 	dev = dev_get_by_index_rcu(net, fq->iif);
106 	if (!dev)
107 		goto out;
108 
109 	IP6_INC_STATS_BH(net, __in6_dev_get(dev), IPSTATS_MIB_REASMFAILS);
110 	IP6_INC_STATS_BH(net, __in6_dev_get(dev), IPSTATS_MIB_REASMTIMEOUT);
111 
112 	/* Don't send error if the first segment did not arrive. */
113 	head = fq->q.fragments;
114 	if (!(fq->q.flags & INET_FRAG_FIRST_IN) || !head)
115 		goto out;
116 
117 	/* But use as source device on which LAST ARRIVED
118 	 * segment was received. And do not use fq->dev
119 	 * pointer directly, device might already disappeared.
120 	 */
121 	head->dev = dev;
122 	skb_get(head);
123 	spin_unlock(&fq->q.lock);
124 
125 	icmpv6_send(head, ICMPV6_TIME_EXCEED, ICMPV6_EXC_FRAGTIME, 0);
126 	kfree_skb(head);
127 	goto out_rcu_unlock;
128 
129 out:
130 	spin_unlock(&fq->q.lock);
131 out_rcu_unlock:
132 	rcu_read_unlock();
133 	inet_frag_put(&fq->q);
134 }
135 EXPORT_SYMBOL(ip6_expire_frag_queue);
136 
ip6_frag_expire(unsigned long data)137 static void ip6_frag_expire(unsigned long data)
138 {
139 	struct frag_queue *fq;
140 	struct net *net;
141 
142 	fq = container_of((struct inet_frag_queue *)data, struct frag_queue, q);
143 	net = container_of(fq->q.net, struct net, ipv6.frags);
144 
145 	ip6_expire_frag_queue(net, fq);
146 }
147 
148 static struct frag_queue *
fq_find(struct net * net,__be32 id,const struct ipv6hdr * hdr,int iif)149 fq_find(struct net *net, __be32 id, const struct ipv6hdr *hdr, int iif)
150 {
151 	struct frag_v6_compare_key key = {
152 		.id = id,
153 		.saddr = hdr->saddr,
154 		.daddr = hdr->daddr,
155 		.user = IP6_DEFRAG_LOCAL_DELIVER,
156 		.iif = iif,
157 	};
158 	struct inet_frag_queue *q;
159 
160 	if (!(ipv6_addr_type(&hdr->daddr) & (IPV6_ADDR_MULTICAST |
161 					    IPV6_ADDR_LINKLOCAL)))
162 		key.iif = 0;
163 
164 	q = inet_frag_find(&net->ipv6.frags, &key);
165 	if (!q)
166 		return NULL;
167 
168 	return container_of(q, struct frag_queue, q);
169 }
170 
ip6_frag_queue(struct frag_queue * fq,struct sk_buff * skb,struct frag_hdr * fhdr,int nhoff)171 static int ip6_frag_queue(struct frag_queue *fq, struct sk_buff *skb,
172 			   struct frag_hdr *fhdr, int nhoff)
173 {
174 	struct sk_buff *prev, *next;
175 	struct net_device *dev;
176 	int offset, end;
177 	struct net *net = dev_net(skb_dst(skb)->dev);
178 	u8 ecn;
179 
180 	if (fq->q.flags & INET_FRAG_COMPLETE)
181 		goto err;
182 
183 	offset = ntohs(fhdr->frag_off) & ~0x7;
184 	end = offset + (ntohs(ipv6_hdr(skb)->payload_len) -
185 			((u8 *)(fhdr + 1) - (u8 *)(ipv6_hdr(skb) + 1)));
186 
187 	if ((unsigned int)end > IPV6_MAXPLEN) {
188 		IP6_INC_STATS_BH(net, ip6_dst_idev(skb_dst(skb)),
189 				 IPSTATS_MIB_INHDRERRORS);
190 		icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
191 				  ((u8 *)&fhdr->frag_off -
192 				   skb_network_header(skb)));
193 		return -1;
194 	}
195 
196 	ecn = ip6_frag_ecn(ipv6_hdr(skb));
197 
198 	if (skb->ip_summed == CHECKSUM_COMPLETE) {
199 		const unsigned char *nh = skb_network_header(skb);
200 		skb->csum = csum_sub(skb->csum,
201 				     csum_partial(nh, (u8 *)(fhdr + 1) - nh,
202 						  0));
203 	}
204 
205 	/* Is this the final fragment? */
206 	if (!(fhdr->frag_off & htons(IP6_MF))) {
207 		/* If we already have some bits beyond end
208 		 * or have different end, the segment is corrupted.
209 		 */
210 		if (end < fq->q.len ||
211 		    ((fq->q.flags & INET_FRAG_LAST_IN) && end != fq->q.len))
212 			goto err;
213 		fq->q.flags |= INET_FRAG_LAST_IN;
214 		fq->q.len = end;
215 	} else {
216 		/* Check if the fragment is rounded to 8 bytes.
217 		 * Required by the RFC.
218 		 */
219 		if (end & 0x7) {
220 			/* RFC2460 says always send parameter problem in
221 			 * this case. -DaveM
222 			 */
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 					  offsetof(struct ipv6hdr, payload_len));
227 			return -1;
228 		}
229 		if (end > fq->q.len) {
230 			/* Some bits beyond end -> corruption. */
231 			if (fq->q.flags & INET_FRAG_LAST_IN)
232 				goto err;
233 			fq->q.len = end;
234 		}
235 	}
236 
237 	if (end == offset)
238 		goto err;
239 
240 	/* Point into the IP datagram 'data' part. */
241 	if (!pskb_pull(skb, (u8 *) (fhdr + 1) - skb->data))
242 		goto err;
243 
244 	if (pskb_trim_rcsum(skb, end - offset))
245 		goto err;
246 
247 	/* Find out which fragments are in front and at the back of us
248 	 * in the chain of fragments so far.  We must know where to put
249 	 * this fragment, right?
250 	 */
251 	prev = fq->q.fragments_tail;
252 	if (!prev || FRAG6_CB(prev)->offset < offset) {
253 		next = NULL;
254 		goto found;
255 	}
256 	prev = NULL;
257 	for (next = fq->q.fragments; next != NULL; next = next->next) {
258 		if (FRAG6_CB(next)->offset >= offset)
259 			break;	/* bingo! */
260 		prev = next;
261 	}
262 
263 found:
264 	/* RFC5722, Section 4, amended by Errata ID : 3089
265 	 *                          When reassembling an IPv6 datagram, if
266 	 *   one or more its constituent fragments is determined to be an
267 	 *   overlapping fragment, the entire datagram (and any constituent
268 	 *   fragments) MUST be silently discarded.
269 	 */
270 
271 	/* Check for overlap with preceding fragment. */
272 	if (prev &&
273 	    (FRAG6_CB(prev)->offset + prev->len) > offset)
274 		goto discard_fq;
275 
276 	/* Look for overlap with succeeding segment. */
277 	if (next && FRAG6_CB(next)->offset < end)
278 		goto discard_fq;
279 
280 	FRAG6_CB(skb)->offset = offset;
281 
282 	/* Insert this fragment in the chain of fragments. */
283 	skb->next = next;
284 	if (!next)
285 		fq->q.fragments_tail = skb;
286 	if (prev)
287 		prev->next = skb;
288 	else
289 		fq->q.fragments = skb;
290 
291 	dev = skb->dev;
292 	if (dev) {
293 		fq->iif = dev->ifindex;
294 		skb->dev = NULL;
295 	}
296 	fq->q.stamp = skb->tstamp;
297 	fq->q.meat += skb->len;
298 	fq->ecn |= ecn;
299 	add_frag_mem_limit(fq->q.net, skb->truesize);
300 
301 	/* The first fragment.
302 	 * nhoffset is obtained from the first fragment, of course.
303 	 */
304 	if (offset == 0) {
305 		fq->nhoffset = nhoff;
306 		fq->q.flags |= INET_FRAG_FIRST_IN;
307 	}
308 
309 	if (fq->q.flags == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
310 	    fq->q.meat == fq->q.len) {
311 		int res;
312 		unsigned long orefdst = skb->_skb_refdst;
313 
314 		skb->_skb_refdst = 0UL;
315 		res = ip6_frag_reasm(fq, prev, dev);
316 		skb->_skb_refdst = orefdst;
317 		return res;
318 	}
319 
320 	skb_dst_drop(skb);
321 	return -1;
322 
323 discard_fq:
324 	inet_frag_kill(&fq->q);
325 err:
326 	IP6_INC_STATS_BH(net, ip6_dst_idev(skb_dst(skb)),
327 			 IPSTATS_MIB_REASMFAILS);
328 	kfree_skb(skb);
329 	return -1;
330 }
331 
332 /*
333  *	Check if this packet is complete.
334  *	Returns NULL on failure by any reason, and pointer
335  *	to current nexthdr field in reassembled frame.
336  *
337  *	It is called with locked fq, and caller must check that
338  *	queue is eligible for reassembly i.e. it is not COMPLETE,
339  *	the last and the first frames arrived and all the bits are here.
340  */
ip6_frag_reasm(struct frag_queue * fq,struct sk_buff * prev,struct net_device * dev)341 static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *prev,
342 			  struct net_device *dev)
343 {
344 	struct net *net = container_of(fq->q.net, struct net, ipv6.frags);
345 	struct sk_buff *fp, *head = fq->q.fragments;
346 	int    payload_len;
347 	unsigned int nhoff;
348 	int sum_truesize;
349 	u8 ecn;
350 
351 	inet_frag_kill(&fq->q);
352 
353 	ecn = ip_frag_ecn_table[fq->ecn];
354 	if (unlikely(ecn == 0xff))
355 		goto out_fail;
356 
357 	/* Make the one we just received the head. */
358 	if (prev) {
359 		head = prev->next;
360 		fp = skb_clone(head, GFP_ATOMIC);
361 
362 		if (!fp)
363 			goto out_oom;
364 
365 		fp->next = head->next;
366 		if (!fp->next)
367 			fq->q.fragments_tail = fp;
368 		prev->next = fp;
369 
370 		skb_morph(head, fq->q.fragments);
371 		head->next = fq->q.fragments->next;
372 
373 		consume_skb(fq->q.fragments);
374 		fq->q.fragments = head;
375 	}
376 
377 	WARN_ON(head == NULL);
378 	WARN_ON(FRAG6_CB(head)->offset != 0);
379 
380 	/* Unfragmented part is taken from the first segment. */
381 	payload_len = ((head->data - skb_network_header(head)) -
382 		       sizeof(struct ipv6hdr) + fq->q.len -
383 		       sizeof(struct frag_hdr));
384 	if (payload_len > IPV6_MAXPLEN)
385 		goto out_oversize;
386 
387 	/* Head of list must not be cloned. */
388 	if (skb_unclone(head, GFP_ATOMIC))
389 		goto out_oom;
390 
391 	/* If the first fragment is fragmented itself, we split
392 	 * it to two chunks: the first with data and paged part
393 	 * and the second, holding only fragments. */
394 	if (skb_has_frag_list(head)) {
395 		struct sk_buff *clone;
396 		int i, plen = 0;
397 
398 		clone = alloc_skb(0, GFP_ATOMIC);
399 		if (!clone)
400 			goto out_oom;
401 		clone->next = head->next;
402 		head->next = clone;
403 		skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
404 		skb_frag_list_init(head);
405 		for (i = 0; i < skb_shinfo(head)->nr_frags; i++)
406 			plen += skb_frag_size(&skb_shinfo(head)->frags[i]);
407 		clone->len = clone->data_len = head->data_len - plen;
408 		head->data_len -= clone->len;
409 		head->len -= clone->len;
410 		clone->csum = 0;
411 		clone->ip_summed = head->ip_summed;
412 		add_frag_mem_limit(fq->q.net, clone->truesize);
413 	}
414 
415 	/* We have to remove fragment header from datagram and to relocate
416 	 * header in order to calculate ICV correctly. */
417 	nhoff = fq->nhoffset;
418 	skb_network_header(head)[nhoff] = skb_transport_header(head)[0];
419 	memmove(head->head + sizeof(struct frag_hdr), head->head,
420 		(head->data - head->head) - sizeof(struct frag_hdr));
421 	head->mac_header += sizeof(struct frag_hdr);
422 	head->network_header += sizeof(struct frag_hdr);
423 
424 	skb_reset_transport_header(head);
425 	skb_push(head, head->data - skb_network_header(head));
426 
427 	sum_truesize = head->truesize;
428 	for (fp = head->next; fp;) {
429 		bool headstolen;
430 		int delta;
431 		struct sk_buff *next = fp->next;
432 
433 		sum_truesize += fp->truesize;
434 		if (head->ip_summed != fp->ip_summed)
435 			head->ip_summed = CHECKSUM_NONE;
436 		else if (head->ip_summed == CHECKSUM_COMPLETE)
437 			head->csum = csum_add(head->csum, fp->csum);
438 
439 		if (skb_try_coalesce(head, fp, &headstolen, &delta)) {
440 			kfree_skb_partial(fp, headstolen);
441 		} else {
442 			if (!skb_shinfo(head)->frag_list)
443 				skb_shinfo(head)->frag_list = fp;
444 			head->data_len += fp->len;
445 			head->len += fp->len;
446 			head->truesize += fp->truesize;
447 		}
448 		fp = next;
449 	}
450 	sub_frag_mem_limit(fq->q.net, sum_truesize);
451 
452 	head->next = NULL;
453 	head->dev = dev;
454 	head->tstamp = fq->q.stamp;
455 	ipv6_hdr(head)->payload_len = htons(payload_len);
456 	ipv6_change_dsfield(ipv6_hdr(head), 0xff, ecn);
457 	IP6CB(head)->nhoff = nhoff;
458 	IP6CB(head)->flags |= IP6SKB_FRAGMENTED;
459 
460 	/* Yes, and fold redundant checksum back. 8) */
461 	skb_postpush_rcsum(head, skb_network_header(head),
462 			   skb_network_header_len(head));
463 
464 	rcu_read_lock();
465 	IP6_INC_STATS_BH(net, __in6_dev_get(dev), IPSTATS_MIB_REASMOKS);
466 	rcu_read_unlock();
467 	fq->q.fragments = NULL;
468 	fq->q.rb_fragments = RB_ROOT;
469 	fq->q.fragments_tail = NULL;
470 	return 1;
471 
472 out_oversize:
473 	net_dbg_ratelimited("ip6_frag_reasm: payload len = %d\n", payload_len);
474 	goto out_fail;
475 out_oom:
476 	net_dbg_ratelimited("ip6_frag_reasm: no memory for reassembly\n");
477 out_fail:
478 	rcu_read_lock();
479 	IP6_INC_STATS_BH(net, __in6_dev_get(dev), IPSTATS_MIB_REASMFAILS);
480 	rcu_read_unlock();
481 	return -1;
482 }
483 
ipv6_frag_rcv(struct sk_buff * skb)484 static int ipv6_frag_rcv(struct sk_buff *skb)
485 {
486 	struct frag_hdr *fhdr;
487 	struct frag_queue *fq;
488 	const struct ipv6hdr *hdr = ipv6_hdr(skb);
489 	struct net *net = dev_net(skb_dst(skb)->dev);
490 	int iif;
491 
492 	if (IP6CB(skb)->flags & IP6SKB_FRAGMENTED)
493 		goto fail_hdr;
494 
495 	IP6_INC_STATS_BH(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMREQDS);
496 
497 	/* Jumbo payload inhibits frag. header */
498 	if (hdr->payload_len == 0)
499 		goto fail_hdr;
500 
501 	if (!pskb_may_pull(skb, (skb_transport_offset(skb) +
502 				 sizeof(struct frag_hdr))))
503 		goto fail_hdr;
504 
505 	hdr = ipv6_hdr(skb);
506 	fhdr = (struct frag_hdr *)skb_transport_header(skb);
507 
508 	if (!(fhdr->frag_off & htons(0xFFF9))) {
509 		/* It is not a fragmented frame */
510 		skb->transport_header += sizeof(struct frag_hdr);
511 		IP6_INC_STATS_BH(net,
512 				 ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMOKS);
513 
514 		IP6CB(skb)->nhoff = (u8 *)fhdr - skb_network_header(skb);
515 		IP6CB(skb)->flags |= IP6SKB_FRAGMENTED;
516 		return 1;
517 	}
518 
519 	if (skb->len - skb_network_offset(skb) < IPV6_MIN_MTU &&
520 	    fhdr->frag_off & htons(IP6_MF))
521 		goto fail_hdr;
522 
523 	iif = skb->dev ? skb->dev->ifindex : 0;
524 	fq = fq_find(net, fhdr->identification, hdr, iif);
525 	if (fq) {
526 		int ret;
527 
528 		spin_lock(&fq->q.lock);
529 
530 		fq->iif = iif;
531 		ret = ip6_frag_queue(fq, skb, fhdr, IP6CB(skb)->nhoff);
532 
533 		spin_unlock(&fq->q.lock);
534 		inet_frag_put(&fq->q);
535 		return ret;
536 	}
537 
538 	IP6_INC_STATS_BH(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMFAILS);
539 	kfree_skb(skb);
540 	return -1;
541 
542 fail_hdr:
543 	IP6_INC_STATS_BH(net, ip6_dst_idev(skb_dst(skb)),
544 			 IPSTATS_MIB_INHDRERRORS);
545 	icmpv6_param_prob(skb, ICMPV6_HDR_FIELD, skb_network_header_len(skb));
546 	return -1;
547 }
548 
549 static const struct inet6_protocol frag_protocol = {
550 	.handler	=	ipv6_frag_rcv,
551 	.flags		=	INET6_PROTO_NOPOLICY,
552 };
553 
554 #ifdef CONFIG_SYSCTL
555 
556 static struct ctl_table ip6_frags_ns_ctl_table[] = {
557 	{
558 		.procname	= "ip6frag_high_thresh",
559 		.data		= &init_net.ipv6.frags.high_thresh,
560 		.maxlen		= sizeof(unsigned long),
561 		.mode		= 0644,
562 		.proc_handler	= proc_doulongvec_minmax,
563 		.extra1		= &init_net.ipv6.frags.low_thresh
564 	},
565 	{
566 		.procname	= "ip6frag_low_thresh",
567 		.data		= &init_net.ipv6.frags.low_thresh,
568 		.maxlen		= sizeof(unsigned long),
569 		.mode		= 0644,
570 		.proc_handler	= proc_doulongvec_minmax,
571 		.extra2		= &init_net.ipv6.frags.high_thresh
572 	},
573 	{
574 		.procname	= "ip6frag_time",
575 		.data		= &init_net.ipv6.frags.timeout,
576 		.maxlen		= sizeof(int),
577 		.mode		= 0644,
578 		.proc_handler	= proc_dointvec_jiffies,
579 	},
580 	{ }
581 };
582 
583 /* secret interval has been deprecated */
584 static int ip6_frags_secret_interval_unused;
585 static struct ctl_table ip6_frags_ctl_table[] = {
586 	{
587 		.procname	= "ip6frag_secret_interval",
588 		.data		= &ip6_frags_secret_interval_unused,
589 		.maxlen		= sizeof(int),
590 		.mode		= 0644,
591 		.proc_handler	= proc_dointvec_jiffies,
592 	},
593 	{ }
594 };
595 
ip6_frags_ns_sysctl_register(struct net * net)596 static int __net_init ip6_frags_ns_sysctl_register(struct net *net)
597 {
598 	struct ctl_table *table;
599 	struct ctl_table_header *hdr;
600 
601 	table = ip6_frags_ns_ctl_table;
602 	if (!net_eq(net, &init_net)) {
603 		table = kmemdup(table, sizeof(ip6_frags_ns_ctl_table), GFP_KERNEL);
604 		if (!table)
605 			goto err_alloc;
606 
607 		table[0].data = &net->ipv6.frags.high_thresh;
608 		table[0].extra1 = &net->ipv6.frags.low_thresh;
609 		table[0].extra2 = &init_net.ipv6.frags.high_thresh;
610 		table[1].data = &net->ipv6.frags.low_thresh;
611 		table[1].extra2 = &net->ipv6.frags.high_thresh;
612 		table[2].data = &net->ipv6.frags.timeout;
613 
614 		/* Don't export sysctls to unprivileged users */
615 		if (net->user_ns != &init_user_ns)
616 			table[0].procname = NULL;
617 	}
618 
619 	hdr = register_net_sysctl(net, "net/ipv6", table);
620 	if (!hdr)
621 		goto err_reg;
622 
623 	net->ipv6.sysctl.frags_hdr = hdr;
624 	return 0;
625 
626 err_reg:
627 	if (!net_eq(net, &init_net))
628 		kfree(table);
629 err_alloc:
630 	return -ENOMEM;
631 }
632 
ip6_frags_ns_sysctl_unregister(struct net * net)633 static void __net_exit ip6_frags_ns_sysctl_unregister(struct net *net)
634 {
635 	struct ctl_table *table;
636 
637 	table = net->ipv6.sysctl.frags_hdr->ctl_table_arg;
638 	unregister_net_sysctl_table(net->ipv6.sysctl.frags_hdr);
639 	if (!net_eq(net, &init_net))
640 		kfree(table);
641 }
642 
643 static struct ctl_table_header *ip6_ctl_header;
644 
ip6_frags_sysctl_register(void)645 static int ip6_frags_sysctl_register(void)
646 {
647 	ip6_ctl_header = register_net_sysctl(&init_net, "net/ipv6",
648 			ip6_frags_ctl_table);
649 	return ip6_ctl_header == NULL ? -ENOMEM : 0;
650 }
651 
ip6_frags_sysctl_unregister(void)652 static void ip6_frags_sysctl_unregister(void)
653 {
654 	unregister_net_sysctl_table(ip6_ctl_header);
655 }
656 #else
ip6_frags_ns_sysctl_register(struct net * net)657 static int ip6_frags_ns_sysctl_register(struct net *net)
658 {
659 	return 0;
660 }
661 
ip6_frags_ns_sysctl_unregister(struct net * net)662 static void ip6_frags_ns_sysctl_unregister(struct net *net)
663 {
664 }
665 
ip6_frags_sysctl_register(void)666 static int ip6_frags_sysctl_register(void)
667 {
668 	return 0;
669 }
670 
ip6_frags_sysctl_unregister(void)671 static void ip6_frags_sysctl_unregister(void)
672 {
673 }
674 #endif
675 
ipv6_frags_init_net(struct net * net)676 static int __net_init ipv6_frags_init_net(struct net *net)
677 {
678 	int res;
679 
680 	net->ipv6.frags.high_thresh = IPV6_FRAG_HIGH_THRESH;
681 	net->ipv6.frags.low_thresh = IPV6_FRAG_LOW_THRESH;
682 	net->ipv6.frags.timeout = IPV6_FRAG_TIMEOUT;
683 	net->ipv6.frags.f = &ip6_frags;
684 
685 	res = inet_frags_init_net(&net->ipv6.frags);
686 	if (res < 0)
687 		return res;
688 
689 	res = ip6_frags_ns_sysctl_register(net);
690 	if (res < 0)
691 		inet_frags_exit_net(&net->ipv6.frags);
692 	return res;
693 }
694 
ipv6_frags_exit_net(struct net * net)695 static void __net_exit ipv6_frags_exit_net(struct net *net)
696 {
697 	ip6_frags_ns_sysctl_unregister(net);
698 	inet_frags_exit_net(&net->ipv6.frags);
699 }
700 
701 static struct pernet_operations ip6_frags_ops = {
702 	.init = ipv6_frags_init_net,
703 	.exit = ipv6_frags_exit_net,
704 };
705 
ip6_key_hashfn(const void * data,u32 len,u32 seed)706 static u32 ip6_key_hashfn(const void *data, u32 len, u32 seed)
707 {
708 	return jhash2(data,
709 		      sizeof(struct frag_v6_compare_key) / sizeof(u32), seed);
710 }
711 
ip6_obj_hashfn(const void * data,u32 len,u32 seed)712 static u32 ip6_obj_hashfn(const void *data, u32 len, u32 seed)
713 {
714 	const struct inet_frag_queue *fq = data;
715 
716 	return jhash2((const u32 *)&fq->key.v6,
717 		      sizeof(struct frag_v6_compare_key) / sizeof(u32), seed);
718 }
719 
ip6_obj_cmpfn(struct rhashtable_compare_arg * arg,const void * ptr)720 static int ip6_obj_cmpfn(struct rhashtable_compare_arg *arg, const void *ptr)
721 {
722 	const struct frag_v6_compare_key *key = arg->key;
723 	const struct inet_frag_queue *fq = ptr;
724 
725 	return !!memcmp(&fq->key, key, sizeof(*key));
726 }
727 
728 const struct rhashtable_params ip6_rhash_params = {
729 	.head_offset		= offsetof(struct inet_frag_queue, node),
730 	.hashfn			= ip6_key_hashfn,
731 	.obj_hashfn		= ip6_obj_hashfn,
732 	.obj_cmpfn		= ip6_obj_cmpfn,
733 	.automatic_shrinking	= true,
734 };
735 EXPORT_SYMBOL(ip6_rhash_params);
736 
ipv6_frag_init(void)737 int __init ipv6_frag_init(void)
738 {
739 	int ret;
740 
741 	ip6_frags.constructor = ip6_frag_init;
742 	ip6_frags.destructor = NULL;
743 	ip6_frags.qsize = sizeof(struct frag_queue);
744 	ip6_frags.frag_expire = ip6_frag_expire;
745 	ip6_frags.frags_cache_name = ip6_frag_cache_name;
746 	ip6_frags.rhash_params = ip6_rhash_params;
747 	ret = inet_frags_init(&ip6_frags);
748 	if (ret)
749 		goto out;
750 
751 	ret = inet6_add_protocol(&frag_protocol, IPPROTO_FRAGMENT);
752 	if (ret)
753 		goto err_protocol;
754 
755 	ret = ip6_frags_sysctl_register();
756 	if (ret)
757 		goto err_sysctl;
758 
759 	ret = register_pernet_subsys(&ip6_frags_ops);
760 	if (ret)
761 		goto err_pernet;
762 
763 out:
764 	return ret;
765 
766 err_pernet:
767 	ip6_frags_sysctl_unregister();
768 err_sysctl:
769 	inet6_del_protocol(&frag_protocol, IPPROTO_FRAGMENT);
770 err_protocol:
771 	inet_frags_fini(&ip6_frags);
772 	goto out;
773 }
774 
ipv6_frag_exit(void)775 void ipv6_frag_exit(void)
776 {
777 	ip6_frags_sysctl_unregister();
778 	unregister_pernet_subsys(&ip6_frags_ops);
779 	inet6_del_protocol(&frag_protocol, IPPROTO_FRAGMENT);
780 	inet_frags_fini(&ip6_frags);
781 }
782