ip_fragment.c revision b13d3cbfb8e8a8f53930af67d1ebf05149f32c24 
1/*
2 * INET		An implementation of the TCP/IP protocol suite for the LINUX
3 *		operating system.  INET is implemented using the  BSD Socket
4 *		interface as the means of communication with the user level.
5 *
6 *		The IP fragmentation functionality.
7 *
8 * Authors:	Fred N. van Kempen <waltje@uWalt.NL.Mugnet.ORG>
9 *		Alan Cox <alan@lxorguk.ukuu.org.uk>
10 *
11 * Fixes:
12 *		Alan Cox	:	Split from ip.c , see ip_input.c for history.
13 *		David S. Miller :	Begin massive cleanup...
14 *		Andi Kleen	:	Add sysctls.
15 *		xxxx		:	Overlapfrag bug.
16 *		Ultima          :       ip_expire() kernel panic.
17 *		Bill Hawes	:	Frag accounting and evictor fixes.
18 *		John McDonald	:	0 length frag bug.
19 *		Alexey Kuznetsov:	SMP races, threading, cleanup.
20 *		Patrick McHardy :	LRU queue of frag heads for evictor.
21 */
22
23#define pr_fmt(fmt) "IPv4: " fmt
24
25#include <linux/compiler.h>
26#include <linux/module.h>
27#include <linux/types.h>
28#include <linux/mm.h>
29#include <linux/jiffies.h>
30#include <linux/skbuff.h>
31#include <linux/list.h>
32#include <linux/ip.h>
33#include <linux/icmp.h>
34#include <linux/netdevice.h>
35#include <linux/jhash.h>
36#include <linux/random.h>
37#include <linux/slab.h>
38#include <net/route.h>
39#include <net/dst.h>
40#include <net/sock.h>
41#include <net/ip.h>
42#include <net/icmp.h>
43#include <net/checksum.h>
44#include <net/inetpeer.h>
45#include <net/inet_frag.h>
46#include <linux/tcp.h>
47#include <linux/udp.h>
48#include <linux/inet.h>
49#include <linux/netfilter_ipv4.h>
50#include <net/inet_ecn.h>
51
52/* NOTE. Logic of IP defragmentation is parallel to corresponding IPv6
53 * code now. If you change something here, _PLEASE_ update ipv6/reassembly.c
54 * as well. Or notify me, at least. --ANK
55 */
56
57static int sysctl_ipfrag_max_dist __read_mostly = 64;
58
59struct ipfrag_skb_cb
60{
61	struct inet_skb_parm	h;
62	int			offset;
63};
64
65#define FRAG_CB(skb)	((struct ipfrag_skb_cb *)((skb)->cb))
66
67/* Describe an entry in the "incomplete datagrams" queue. */
68struct ipq {
69	struct inet_frag_queue q;
70
71	u32		user;
72	__be32		saddr;
73	__be32		daddr;
74	__be16		id;
75	u8		protocol;
76	u8		ecn; /* RFC3168 support */
77	int             iif;
78	unsigned int    rid;
79	struct inet_peer *peer;
80};
81
82static inline u8 ip4_frag_ecn(u8 tos)
83{
84	return 1 << (tos & INET_ECN_MASK);
85}
86
87static struct inet_frags ip4_frags;
88
89int ip_frag_nqueues(struct net *net)
90{
91	return net->ipv4.frags.nqueues;
92}
93
94int ip_frag_mem(struct net *net)
95{
96	return sum_frag_mem_limit(&net->ipv4.frags);
97}
98
99static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev,
100			 struct net_device *dev);
101
102struct ip4_create_arg {
103	struct iphdr *iph;
104	u32 user;
105};
106
107static unsigned int ipqhashfn(__be16 id, __be32 saddr, __be32 daddr, u8 prot)
108{
109	net_get_random_once(&ip4_frags.rnd, sizeof(ip4_frags.rnd));
110	return jhash_3words((__force u32)id << 16 | prot,
111			    (__force u32)saddr, (__force u32)daddr,
112			    ip4_frags.rnd);
113}
114
115static unsigned int ip4_hashfn(const struct inet_frag_queue *q)
116{
117	const struct ipq *ipq;
118
119	ipq = container_of(q, struct ipq, q);
120	return ipqhashfn(ipq->id, ipq->saddr, ipq->daddr, ipq->protocol);
121}
122
123static bool ip4_frag_match(const struct inet_frag_queue *q, const void *a)
124{
125	const struct ipq *qp;
126	const struct ip4_create_arg *arg = a;
127
128	qp = container_of(q, struct ipq, q);
129	return	qp->id == arg->iph->id &&
130		qp->saddr == arg->iph->saddr &&
131		qp->daddr == arg->iph->daddr &&
132		qp->protocol == arg->iph->protocol &&
133		qp->user == arg->user;
134}
135
136static void ip4_frag_init(struct inet_frag_queue *q, const void *a)
137{
138	struct ipq *qp = container_of(q, struct ipq, q);
139	struct netns_ipv4 *ipv4 = container_of(q->net, struct netns_ipv4,
140					       frags);
141	struct net *net = container_of(ipv4, struct net, ipv4);
142
143	const struct ip4_create_arg *arg = a;
144
145	qp->protocol = arg->iph->protocol;
146	qp->id = arg->iph->id;
147	qp->ecn = ip4_frag_ecn(arg->iph->tos);
148	qp->saddr = arg->iph->saddr;
149	qp->daddr = arg->iph->daddr;
150	qp->user = arg->user;
151	qp->peer = sysctl_ipfrag_max_dist ?
152		inet_getpeer_v4(net->ipv4.peers, arg->iph->saddr, 1) : NULL;
153}
154
155static __inline__ void ip4_frag_free(struct inet_frag_queue *q)
156{
157	struct ipq *qp;
158
159	qp = container_of(q, struct ipq, q);
160	if (qp->peer)
161		inet_putpeer(qp->peer);
162}
163
164
165/* Destruction primitives. */
166
167static __inline__ void ipq_put(struct ipq *ipq)
168{
169	inet_frag_put(&ipq->q, &ip4_frags);
170}
171
172/* Kill ipq entry. It is not destroyed immediately,
173 * because caller (and someone more) holds reference count.
174 */
175static void ipq_kill(struct ipq *ipq)
176{
177	inet_frag_kill(&ipq->q, &ip4_frags);
178}
179
180/*
181 * Oops, a fragment queue timed out.  Kill it and send an ICMP reply.
182 */
183static void ip_expire(unsigned long arg)
184{
185	struct ipq *qp;
186	struct net *net;
187
188	qp = container_of((struct inet_frag_queue *) arg, struct ipq, q);
189	net = container_of(qp->q.net, struct net, ipv4.frags);
190
191	spin_lock(&qp->q.lock);
192
193	if (qp->q.last_in & INET_FRAG_COMPLETE)
194		goto out;
195
196	ipq_kill(qp);
197
198	if (!(qp->q.last_in & INET_FRAG_EVICTED))
199		IP_INC_STATS_BH(net, IPSTATS_MIB_REASMTIMEOUT);
200	IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
201
202	if ((qp->q.last_in & INET_FRAG_FIRST_IN) && qp->q.fragments != NULL) {
203		struct sk_buff *head = qp->q.fragments;
204		const struct iphdr *iph;
205		int err;
206
207		rcu_read_lock();
208		head->dev = dev_get_by_index_rcu(net, qp->iif);
209		if (!head->dev)
210			goto out_rcu_unlock;
211
212		/* skb has no dst, perform route lookup again */
213		iph = ip_hdr(head);
214		err = ip_route_input_noref(head, iph->daddr, iph->saddr,
215					   iph->tos, head->dev);
216		if (err)
217			goto out_rcu_unlock;
218
219		/*
220		 * Only an end host needs to send an ICMP
221		 * "Fragment Reassembly Timeout" message, per RFC792.
222		 */
223		if (qp->user == IP_DEFRAG_AF_PACKET ||
224		    ((qp->user >= IP_DEFRAG_CONNTRACK_IN) &&
225		     (qp->user <= __IP_DEFRAG_CONNTRACK_IN_END) &&
226		     (skb_rtable(head)->rt_type != RTN_LOCAL)))
227			goto out_rcu_unlock;
228
229
230		/* Send an ICMP "Fragment Reassembly Timeout" message. */
231		icmp_send(head, ICMP_TIME_EXCEEDED, ICMP_EXC_FRAGTIME, 0);
232out_rcu_unlock:
233		rcu_read_unlock();
234	}
235out:
236	spin_unlock(&qp->q.lock);
237	ipq_put(qp);
238}
239
240/* Find the correct entry in the "incomplete datagrams" queue for
241 * this IP datagram, and create new one, if nothing is found.
242 */
243static inline struct ipq *ip_find(struct net *net, struct iphdr *iph, u32 user)
244{
245	struct inet_frag_queue *q;
246	struct ip4_create_arg arg;
247	unsigned int hash;
248
249	arg.iph = iph;
250	arg.user = user;
251
252	read_lock(&ip4_frags.lock);
253	hash = ipqhashfn(iph->id, iph->saddr, iph->daddr, iph->protocol);
254
255	q = inet_frag_find(&net->ipv4.frags, &ip4_frags, &arg, hash);
256	if (IS_ERR_OR_NULL(q)) {
257		inet_frag_maybe_warn_overflow(q, pr_fmt());
258		return NULL;
259	}
260	return container_of(q, struct ipq, q);
261}
262
263/* Is the fragment too far ahead to be part of ipq? */
264static inline int ip_frag_too_far(struct ipq *qp)
265{
266	struct inet_peer *peer = qp->peer;
267	unsigned int max = sysctl_ipfrag_max_dist;
268	unsigned int start, end;
269
270	int rc;
271
272	if (!peer || !max)
273		return 0;
274
275	start = qp->rid;
276	end = atomic_inc_return(&peer->rid);
277	qp->rid = end;
278
279	rc = qp->q.fragments && (end - start) > max;
280
281	if (rc) {
282		struct net *net;
283
284		net = container_of(qp->q.net, struct net, ipv4.frags);
285		IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
286	}
287
288	return rc;
289}
290
291static int ip_frag_reinit(struct ipq *qp)
292{
293	struct sk_buff *fp;
294	unsigned int sum_truesize = 0;
295
296	if (!mod_timer(&qp->q.timer, jiffies + qp->q.net->timeout)) {
297		atomic_inc(&qp->q.refcnt);
298		return -ETIMEDOUT;
299	}
300
301	fp = qp->q.fragments;
302	do {
303		struct sk_buff *xp = fp->next;
304
305		sum_truesize += fp->truesize;
306		kfree_skb(fp);
307		fp = xp;
308	} while (fp);
309	sub_frag_mem_limit(&qp->q, sum_truesize);
310
311	qp->q.last_in = 0;
312	qp->q.len = 0;
313	qp->q.meat = 0;
314	qp->q.fragments = NULL;
315	qp->q.fragments_tail = NULL;
316	qp->iif = 0;
317	qp->ecn = 0;
318
319	return 0;
320}
321
322/* Add new segment to existing queue. */
323static int ip_frag_queue(struct ipq *qp, struct sk_buff *skb)
324{
325	struct sk_buff *prev, *next;
326	struct net_device *dev;
327	int flags, offset;
328	int ihl, end;
329	int err = -ENOENT;
330	u8 ecn;
331
332	if (qp->q.last_in & INET_FRAG_COMPLETE)
333		goto err;
334
335	if (!(IPCB(skb)->flags & IPSKB_FRAG_COMPLETE) &&
336	    unlikely(ip_frag_too_far(qp)) &&
337	    unlikely(err = ip_frag_reinit(qp))) {
338		ipq_kill(qp);
339		goto err;
340	}
341
342	ecn = ip4_frag_ecn(ip_hdr(skb)->tos);
343	offset = ntohs(ip_hdr(skb)->frag_off);
344	flags = offset & ~IP_OFFSET;
345	offset &= IP_OFFSET;
346	offset <<= 3;		/* offset is in 8-byte chunks */
347	ihl = ip_hdrlen(skb);
348
349	/* Determine the position of this fragment. */
350	end = offset + skb->len - ihl;
351	err = -EINVAL;
352
353	/* Is this the final fragment? */
354	if ((flags & IP_MF) == 0) {
355		/* If we already have some bits beyond end
356		 * or have different end, the segment is corrupted.
357		 */
358		if (end < qp->q.len ||
359		    ((qp->q.last_in & INET_FRAG_LAST_IN) && end != qp->q.len))
360			goto err;
361		qp->q.last_in |= INET_FRAG_LAST_IN;
362		qp->q.len = end;
363	} else {
364		if (end&7) {
365			end &= ~7;
366			if (skb->ip_summed != CHECKSUM_UNNECESSARY)
367				skb->ip_summed = CHECKSUM_NONE;
368		}
369		if (end > qp->q.len) {
370			/* Some bits beyond end -> corruption. */
371			if (qp->q.last_in & INET_FRAG_LAST_IN)
372				goto err;
373			qp->q.len = end;
374		}
375	}
376	if (end == offset)
377		goto err;
378
379	err = -ENOMEM;
380	if (pskb_pull(skb, ihl) == NULL)
381		goto err;
382
383	err = pskb_trim_rcsum(skb, end - offset);
384	if (err)
385		goto err;
386
387	/* Find out which fragments are in front and at the back of us
388	 * in the chain of fragments so far.  We must know where to put
389	 * this fragment, right?
390	 */
391	prev = qp->q.fragments_tail;
392	if (!prev || FRAG_CB(prev)->offset < offset) {
393		next = NULL;
394		goto found;
395	}
396	prev = NULL;
397	for (next = qp->q.fragments; next != NULL; next = next->next) {
398		if (FRAG_CB(next)->offset >= offset)
399			break;	/* bingo! */
400		prev = next;
401	}
402
403found:
404	/* We found where to put this one.  Check for overlap with
405	 * preceding fragment, and, if needed, align things so that
406	 * any overlaps are eliminated.
407	 */
408	if (prev) {
409		int i = (FRAG_CB(prev)->offset + prev->len) - offset;
410
411		if (i > 0) {
412			offset += i;
413			err = -EINVAL;
414			if (end <= offset)
415				goto err;
416			err = -ENOMEM;
417			if (!pskb_pull(skb, i))
418				goto err;
419			if (skb->ip_summed != CHECKSUM_UNNECESSARY)
420				skb->ip_summed = CHECKSUM_NONE;
421		}
422	}
423
424	err = -ENOMEM;
425
426	while (next && FRAG_CB(next)->offset < end) {
427		int i = end - FRAG_CB(next)->offset; /* overlap is 'i' bytes */
428
429		if (i < next->len) {
430			/* Eat head of the next overlapped fragment
431			 * and leave the loop. The next ones cannot overlap.
432			 */
433			if (!pskb_pull(next, i))
434				goto err;
435			FRAG_CB(next)->offset += i;
436			qp->q.meat -= i;
437			if (next->ip_summed != CHECKSUM_UNNECESSARY)
438				next->ip_summed = CHECKSUM_NONE;
439			break;
440		} else {
441			struct sk_buff *free_it = next;
442
443			/* Old fragment is completely overridden with
444			 * new one drop it.
445			 */
446			next = next->next;
447
448			if (prev)
449				prev->next = next;
450			else
451				qp->q.fragments = next;
452
453			qp->q.meat -= free_it->len;
454			sub_frag_mem_limit(&qp->q, free_it->truesize);
455			kfree_skb(free_it);
456		}
457	}
458
459	FRAG_CB(skb)->offset = offset;
460
461	/* Insert this fragment in the chain of fragments. */
462	skb->next = next;
463	if (!next)
464		qp->q.fragments_tail = skb;
465	if (prev)
466		prev->next = skb;
467	else
468		qp->q.fragments = skb;
469
470	dev = skb->dev;
471	if (dev) {
472		qp->iif = dev->ifindex;
473		skb->dev = NULL;
474	}
475	qp->q.stamp = skb->tstamp;
476	qp->q.meat += skb->len;
477	qp->ecn |= ecn;
478	add_frag_mem_limit(&qp->q, skb->truesize);
479	if (offset == 0)
480		qp->q.last_in |= INET_FRAG_FIRST_IN;
481
482	if (ip_hdr(skb)->frag_off & htons(IP_DF) &&
483	    skb->len + ihl > qp->q.max_size)
484		qp->q.max_size = skb->len + ihl;
485
486	if (qp->q.last_in == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
487	    qp->q.meat == qp->q.len) {
488		unsigned long orefdst = skb->_skb_refdst;
489
490		skb->_skb_refdst = 0UL;
491		err = ip_frag_reasm(qp, prev, dev);
492		skb->_skb_refdst = orefdst;
493		return err;
494	}
495
496	skb_dst_drop(skb);
497	inet_frag_lru_move(&qp->q);
498	return -EINPROGRESS;
499
500err:
501	kfree_skb(skb);
502	return err;
503}
504
505
506/* Build a new IP datagram from all its fragments. */
507
508static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev,
509			 struct net_device *dev)
510{
511	struct net *net = container_of(qp->q.net, struct net, ipv4.frags);
512	struct iphdr *iph;
513	struct sk_buff *fp, *head = qp->q.fragments;
514	int len;
515	int ihlen;
516	int err;
517	int sum_truesize;
518	u8 ecn;
519
520	ipq_kill(qp);
521
522	ecn = ip_frag_ecn_table[qp->ecn];
523	if (unlikely(ecn == 0xff)) {
524		err = -EINVAL;
525		goto out_fail;
526	}
527	/* Make the one we just received the head. */
528	if (prev) {
529		head = prev->next;
530		fp = skb_clone(head, GFP_ATOMIC);
531		if (!fp)
532			goto out_nomem;
533
534		fp->next = head->next;
535		if (!fp->next)
536			qp->q.fragments_tail = fp;
537		prev->next = fp;
538
539		skb_morph(head, qp->q.fragments);
540		head->next = qp->q.fragments->next;
541
542		consume_skb(qp->q.fragments);
543		qp->q.fragments = head;
544	}
545
546	WARN_ON(head == NULL);
547	WARN_ON(FRAG_CB(head)->offset != 0);
548
549	/* Allocate a new buffer for the datagram. */
550	ihlen = ip_hdrlen(head);
551	len = ihlen + qp->q.len;
552
553	err = -E2BIG;
554	if (len > 65535)
555		goto out_oversize;
556
557	/* Head of list must not be cloned. */
558	if (skb_unclone(head, GFP_ATOMIC))
559		goto out_nomem;
560
561	/* If the first fragment is fragmented itself, we split
562	 * it to two chunks: the first with data and paged part
563	 * and the second, holding only fragments. */
564	if (skb_has_frag_list(head)) {
565		struct sk_buff *clone;
566		int i, plen = 0;
567
568		if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL)
569			goto out_nomem;
570		clone->next = head->next;
571		head->next = clone;
572		skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
573		skb_frag_list_init(head);
574		for (i = 0; i < skb_shinfo(head)->nr_frags; i++)
575			plen += skb_frag_size(&skb_shinfo(head)->frags[i]);
576		clone->len = clone->data_len = head->data_len - plen;
577		head->data_len -= clone->len;
578		head->len -= clone->len;
579		clone->csum = 0;
580		clone->ip_summed = head->ip_summed;
581		add_frag_mem_limit(&qp->q, clone->truesize);
582	}
583
584	skb_push(head, head->data - skb_network_header(head));
585
586	sum_truesize = head->truesize;
587	for (fp = head->next; fp;) {
588		bool headstolen;
589		int delta;
590		struct sk_buff *next = fp->next;
591
592		sum_truesize += fp->truesize;
593		if (head->ip_summed != fp->ip_summed)
594			head->ip_summed = CHECKSUM_NONE;
595		else if (head->ip_summed == CHECKSUM_COMPLETE)
596			head->csum = csum_add(head->csum, fp->csum);
597
598		if (skb_try_coalesce(head, fp, &headstolen, &delta)) {
599			kfree_skb_partial(fp, headstolen);
600		} else {
601			if (!skb_shinfo(head)->frag_list)
602				skb_shinfo(head)->frag_list = fp;
603			head->data_len += fp->len;
604			head->len += fp->len;
605			head->truesize += fp->truesize;
606		}
607		fp = next;
608	}
609	sub_frag_mem_limit(&qp->q, sum_truesize);
610
611	head->next = NULL;
612	head->dev = dev;
613	head->tstamp = qp->q.stamp;
614	IPCB(head)->frag_max_size = qp->q.max_size;
615
616	iph = ip_hdr(head);
617	/* max_size != 0 implies at least one fragment had IP_DF set */
618	iph->frag_off = qp->q.max_size ? htons(IP_DF) : 0;
619	iph->tot_len = htons(len);
620	iph->tos |= ecn;
621	IP_INC_STATS_BH(net, IPSTATS_MIB_REASMOKS);
622	qp->q.fragments = NULL;
623	qp->q.fragments_tail = NULL;
624	return 0;
625
626out_nomem:
627	LIMIT_NETDEBUG(KERN_ERR pr_fmt("queue_glue: no memory for gluing queue %p\n"),
628		       qp);
629	err = -ENOMEM;
630	goto out_fail;
631out_oversize:
632	net_info_ratelimited("Oversized IP packet from %pI4\n", &qp->saddr);
633out_fail:
634	IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
635	return err;
636}
637
638/* Process an incoming IP datagram fragment. */
639int ip_defrag(struct sk_buff *skb, u32 user)
640{
641	struct ipq *qp;
642	struct net *net;
643
644	net = skb->dev ? dev_net(skb->dev) : dev_net(skb_dst(skb)->dev);
645	IP_INC_STATS_BH(net, IPSTATS_MIB_REASMREQDS);
646
647	/* Lookup (or create) queue header */
648	if ((qp = ip_find(net, ip_hdr(skb), user)) != NULL) {
649		int ret;
650
651		spin_lock(&qp->q.lock);
652
653		ret = ip_frag_queue(qp, skb);
654
655		spin_unlock(&qp->q.lock);
656		ipq_put(qp);
657		return ret;
658	}
659
660	IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
661	kfree_skb(skb);
662	return -ENOMEM;
663}
664EXPORT_SYMBOL(ip_defrag);
665
666struct sk_buff *ip_check_defrag(struct sk_buff *skb, u32 user)
667{
668	struct iphdr iph;
669	u32 len;
670
671	if (skb->protocol != htons(ETH_P_IP))
672		return skb;
673
674	if (!skb_copy_bits(skb, 0, &iph, sizeof(iph)))
675		return skb;
676
677	if (iph.ihl < 5 || iph.version != 4)
678		return skb;
679
680	len = ntohs(iph.tot_len);
681	if (skb->len < len || len < (iph.ihl * 4))
682		return skb;
683
684	if (ip_is_fragment(&iph)) {
685		skb = skb_share_check(skb, GFP_ATOMIC);
686		if (skb) {
687			if (!pskb_may_pull(skb, iph.ihl*4))
688				return skb;
689			if (pskb_trim_rcsum(skb, len))
690				return skb;
691			memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
692			if (ip_defrag(skb, user))
693				return NULL;
694			skb_clear_hash(skb);
695		}
696	}
697	return skb;
698}
699EXPORT_SYMBOL(ip_check_defrag);
700
701#ifdef CONFIG_SYSCTL
702static int zero;
703
704static struct ctl_table ip4_frags_ns_ctl_table[] = {
705	{
706		.procname	= "ipfrag_high_thresh",
707		.data		= &init_net.ipv4.frags.high_thresh,
708		.maxlen		= sizeof(int),
709		.mode		= 0644,
710		.proc_handler	= proc_dointvec
711	},
712	{
713		.procname	= "ipfrag_low_thresh",
714		.data		= &init_net.ipv4.frags.low_thresh,
715		.maxlen		= sizeof(int),
716		.mode		= 0644,
717		.proc_handler	= proc_dointvec
718	},
719	{
720		.procname	= "ipfrag_time",
721		.data		= &init_net.ipv4.frags.timeout,
722		.maxlen		= sizeof(int),
723		.mode		= 0644,
724		.proc_handler	= proc_dointvec_jiffies,
725	},
726	{ }
727};
728
729static struct ctl_table ip4_frags_ctl_table[] = {
730	{
731		.procname	= "ipfrag_secret_interval",
732		.data		= &ip4_frags.secret_interval,
733		.maxlen		= sizeof(int),
734		.mode		= 0644,
735		.proc_handler	= proc_dointvec_jiffies,
736	},
737	{
738		.procname	= "ipfrag_max_dist",
739		.data		= &sysctl_ipfrag_max_dist,
740		.maxlen		= sizeof(int),
741		.mode		= 0644,
742		.proc_handler	= proc_dointvec_minmax,
743		.extra1		= &zero
744	},
745	{ }
746};
747
748static int __net_init ip4_frags_ns_ctl_register(struct net *net)
749{
750	struct ctl_table *table;
751	struct ctl_table_header *hdr;
752
753	table = ip4_frags_ns_ctl_table;
754	if (!net_eq(net, &init_net)) {
755		table = kmemdup(table, sizeof(ip4_frags_ns_ctl_table), GFP_KERNEL);
756		if (table == NULL)
757			goto err_alloc;
758
759		table[0].data = &net->ipv4.frags.high_thresh;
760		table[1].data = &net->ipv4.frags.low_thresh;
761		table[2].data = &net->ipv4.frags.timeout;
762
763		/* Don't export sysctls to unprivileged users */
764		if (net->user_ns != &init_user_ns)
765			table[0].procname = NULL;
766	}
767
768	hdr = register_net_sysctl(net, "net/ipv4", table);
769	if (hdr == NULL)
770		goto err_reg;
771
772	net->ipv4.frags_hdr = hdr;
773	return 0;
774
775err_reg:
776	if (!net_eq(net, &init_net))
777		kfree(table);
778err_alloc:
779	return -ENOMEM;
780}
781
782static void __net_exit ip4_frags_ns_ctl_unregister(struct net *net)
783{
784	struct ctl_table *table;
785
786	table = net->ipv4.frags_hdr->ctl_table_arg;
787	unregister_net_sysctl_table(net->ipv4.frags_hdr);
788	kfree(table);
789}
790
791static void ip4_frags_ctl_register(void)
792{
793	register_net_sysctl(&init_net, "net/ipv4", ip4_frags_ctl_table);
794}
795#else
796static inline int ip4_frags_ns_ctl_register(struct net *net)
797{
798	return 0;
799}
800
801static inline void ip4_frags_ns_ctl_unregister(struct net *net)
802{
803}
804
805static inline void ip4_frags_ctl_register(void)
806{
807}
808#endif
809
810static int __net_init ipv4_frags_init_net(struct net *net)
811{
812	/* Fragment cache limits.
813	 *
814	 * The fragment memory accounting code, (tries to) account for
815	 * the real memory usage, by measuring both the size of frag
816	 * queue struct (inet_frag_queue (ipv4:ipq/ipv6:frag_queue))
817	 * and the SKB's truesize.
818	 *
819	 * A 64K fragment consumes 129736 bytes (44*2944)+200
820	 * (1500 truesize == 2944, sizeof(struct ipq) == 200)
821	 *
822	 * We will commit 4MB at one time. Should we cross that limit
823	 * we will prune down to 3MB, making room for approx 8 big 64K
824	 * fragments 8x128k.
825	 */
826	net->ipv4.frags.high_thresh = 4 * 1024 * 1024;
827	net->ipv4.frags.low_thresh  = 3 * 1024 * 1024;
828	/*
829	 * Important NOTE! Fragment queue must be destroyed before MSL expires.
830	 * RFC791 is wrong proposing to prolongate timer each fragment arrival
831	 * by TTL.
832	 */
833	net->ipv4.frags.timeout = IP_FRAG_TIME;
834
835	inet_frags_init_net(&net->ipv4.frags);
836
837	return ip4_frags_ns_ctl_register(net);
838}
839
840static void __net_exit ipv4_frags_exit_net(struct net *net)
841{
842	ip4_frags_ns_ctl_unregister(net);
843	inet_frags_exit_net(&net->ipv4.frags, &ip4_frags);
844}
845
846static struct pernet_operations ip4_frags_ops = {
847	.init = ipv4_frags_init_net,
848	.exit = ipv4_frags_exit_net,
849};
850
851void __init ipfrag_init(void)
852{
853	ip4_frags_ctl_register();
854	register_pernet_subsys(&ip4_frags_ops);
855	ip4_frags.hashfn = ip4_hashfn;
856	ip4_frags.constructor = ip4_frag_init;
857	ip4_frags.destructor = ip4_frag_free;
858	ip4_frags.skb_free = NULL;
859	ip4_frags.qsize = sizeof(struct ipq);
860	ip4_frags.match = ip4_frag_match;
861	ip4_frags.frag_expire = ip_expire;
862	ip4_frags.secret_interval = 10 * 60 * HZ;
863	inet_frags_init(&ip4_frags);
864}
865