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;
58static const char ip_frag_cache_name[] = "ip4-frags";
59
60struct ipfrag_skb_cb
61{
62	struct inet_skb_parm	h;
63	int			offset;
64};
65
66#define FRAG_CB(skb)	((struct ipfrag_skb_cb *)((skb)->cb))
67
68/* Describe an entry in the "incomplete datagrams" queue. */
69struct ipq {
70	struct inet_frag_queue q;
71
72	u32		user;
73	__be32		saddr;
74	__be32		daddr;
75	__be16		id;
76	u8		protocol;
77	u8		ecn; /* RFC3168 support */
78	int             iif;
79	unsigned int    rid;
80	struct inet_peer *peer;
81};
82
83static u8 ip4_frag_ecn(u8 tos)
84{
85	return 1 << (tos & INET_ECN_MASK);
86}
87
88static struct inet_frags ip4_frags;
89
90int ip_frag_mem(struct net *net)
91{
92	return sum_frag_mem_limit(&net->ipv4.frags);
93}
94
95static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev,
96			 struct net_device *dev);
97
98struct ip4_create_arg {
99	struct iphdr *iph;
100	u32 user;
101};
102
103static unsigned int ipqhashfn(__be16 id, __be32 saddr, __be32 daddr, u8 prot)
104{
105	net_get_random_once(&ip4_frags.rnd, sizeof(ip4_frags.rnd));
106	return jhash_3words((__force u32)id << 16 | prot,
107			    (__force u32)saddr, (__force u32)daddr,
108			    ip4_frags.rnd);
109}
110
111static unsigned int ip4_hashfn(const struct inet_frag_queue *q)
112{
113	const struct ipq *ipq;
114
115	ipq = container_of(q, struct ipq, q);
116	return ipqhashfn(ipq->id, ipq->saddr, ipq->daddr, ipq->protocol);
117}
118
119static bool ip4_frag_match(const struct inet_frag_queue *q, const void *a)
120{
121	const struct ipq *qp;
122	const struct ip4_create_arg *arg = a;
123
124	qp = container_of(q, struct ipq, q);
125	return	qp->id == arg->iph->id &&
126		qp->saddr == arg->iph->saddr &&
127		qp->daddr == arg->iph->daddr &&
128		qp->protocol == arg->iph->protocol &&
129		qp->user == arg->user;
130}
131
132static void ip4_frag_init(struct inet_frag_queue *q, const void *a)
133{
134	struct ipq *qp = container_of(q, struct ipq, q);
135	struct netns_ipv4 *ipv4 = container_of(q->net, struct netns_ipv4,
136					       frags);
137	struct net *net = container_of(ipv4, struct net, ipv4);
138
139	const struct ip4_create_arg *arg = a;
140
141	qp->protocol = arg->iph->protocol;
142	qp->id = arg->iph->id;
143	qp->ecn = ip4_frag_ecn(arg->iph->tos);
144	qp->saddr = arg->iph->saddr;
145	qp->daddr = arg->iph->daddr;
146	qp->user = arg->user;
147	qp->peer = sysctl_ipfrag_max_dist ?
148		inet_getpeer_v4(net->ipv4.peers, arg->iph->saddr, 1) : NULL;
149}
150
151static void ip4_frag_free(struct inet_frag_queue *q)
152{
153	struct ipq *qp;
154
155	qp = container_of(q, struct ipq, q);
156	if (qp->peer)
157		inet_putpeer(qp->peer);
158}
159
160
161/* Destruction primitives. */
162
163static void ipq_put(struct ipq *ipq)
164{
165	inet_frag_put(&ipq->q, &ip4_frags);
166}
167
168/* Kill ipq entry. It is not destroyed immediately,
169 * because caller (and someone more) holds reference count.
170 */
171static void ipq_kill(struct ipq *ipq)
172{
173	inet_frag_kill(&ipq->q, &ip4_frags);
174}
175
176/*
177 * Oops, a fragment queue timed out.  Kill it and send an ICMP reply.
178 */
179static void ip_expire(unsigned long arg)
180{
181	struct ipq *qp;
182	struct net *net;
183
184	qp = container_of((struct inet_frag_queue *) arg, struct ipq, q);
185	net = container_of(qp->q.net, struct net, ipv4.frags);
186
187	spin_lock(&qp->q.lock);
188
189	if (qp->q.flags & INET_FRAG_COMPLETE)
190		goto out;
191
192	ipq_kill(qp);
193	IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
194
195	if (!(qp->q.flags & INET_FRAG_EVICTED)) {
196		struct sk_buff *head = qp->q.fragments;
197		const struct iphdr *iph;
198		int err;
199
200		IP_INC_STATS_BH(net, IPSTATS_MIB_REASMTIMEOUT);
201
202		if (!(qp->q.flags & INET_FRAG_FIRST_IN) || !qp->q.fragments)
203			goto out;
204
205		rcu_read_lock();
206		head->dev = dev_get_by_index_rcu(net, qp->iif);
207		if (!head->dev)
208			goto out_rcu_unlock;
209
210		/* skb has no dst, perform route lookup again */
211		iph = ip_hdr(head);
212		err = ip_route_input_noref(head, iph->daddr, iph->saddr,
213					   iph->tos, head->dev);
214		if (err)
215			goto out_rcu_unlock;
216
217		/* Only an end host needs to send an ICMP
218		 * "Fragment Reassembly Timeout" message, per RFC792.
219		 */
220		if (qp->user == IP_DEFRAG_AF_PACKET ||
221		    ((qp->user >= IP_DEFRAG_CONNTRACK_IN) &&
222		     (qp->user <= __IP_DEFRAG_CONNTRACK_IN_END) &&
223		     (skb_rtable(head)->rt_type != RTN_LOCAL)))
224			goto out_rcu_unlock;
225
226		/* Send an ICMP "Fragment Reassembly Timeout" message. */
227		icmp_send(head, ICMP_TIME_EXCEEDED, ICMP_EXC_FRAGTIME, 0);
228out_rcu_unlock:
229		rcu_read_unlock();
230	}
231out:
232	spin_unlock(&qp->q.lock);
233	ipq_put(qp);
234}
235
236/* Find the correct entry in the "incomplete datagrams" queue for
237 * this IP datagram, and create new one, if nothing is found.
238 */
239static struct ipq *ip_find(struct net *net, struct iphdr *iph, u32 user)
240{
241	struct inet_frag_queue *q;
242	struct ip4_create_arg arg;
243	unsigned int hash;
244
245	arg.iph = iph;
246	arg.user = user;
247
248	hash = ipqhashfn(iph->id, iph->saddr, iph->daddr, iph->protocol);
249
250	q = inet_frag_find(&net->ipv4.frags, &ip4_frags, &arg, hash);
251	if (IS_ERR_OR_NULL(q)) {
252		inet_frag_maybe_warn_overflow(q, pr_fmt());
253		return NULL;
254	}
255	return container_of(q, struct ipq, q);
256}
257
258/* Is the fragment too far ahead to be part of ipq? */
259static int ip_frag_too_far(struct ipq *qp)
260{
261	struct inet_peer *peer = qp->peer;
262	unsigned int max = sysctl_ipfrag_max_dist;
263	unsigned int start, end;
264
265	int rc;
266
267	if (!peer || !max)
268		return 0;
269
270	start = qp->rid;
271	end = atomic_inc_return(&peer->rid);
272	qp->rid = end;
273
274	rc = qp->q.fragments && (end - start) > max;
275
276	if (rc) {
277		struct net *net;
278
279		net = container_of(qp->q.net, struct net, ipv4.frags);
280		IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
281	}
282
283	return rc;
284}
285
286static int ip_frag_reinit(struct ipq *qp)
287{
288	struct sk_buff *fp;
289	unsigned int sum_truesize = 0;
290
291	if (!mod_timer(&qp->q.timer, jiffies + qp->q.net->timeout)) {
292		atomic_inc(&qp->q.refcnt);
293		return -ETIMEDOUT;
294	}
295
296	fp = qp->q.fragments;
297	do {
298		struct sk_buff *xp = fp->next;
299
300		sum_truesize += fp->truesize;
301		kfree_skb(fp);
302		fp = xp;
303	} while (fp);
304	sub_frag_mem_limit(&qp->q, sum_truesize);
305
306	qp->q.flags = 0;
307	qp->q.len = 0;
308	qp->q.meat = 0;
309	qp->q.fragments = NULL;
310	qp->q.fragments_tail = NULL;
311	qp->iif = 0;
312	qp->ecn = 0;
313
314	return 0;
315}
316
317/* Add new segment to existing queue. */
318static int ip_frag_queue(struct ipq *qp, struct sk_buff *skb)
319{
320	struct sk_buff *prev, *next;
321	struct net_device *dev;
322	int flags, offset;
323	int ihl, end;
324	int err = -ENOENT;
325	u8 ecn;
326
327	if (qp->q.flags & INET_FRAG_COMPLETE)
328		goto err;
329
330	if (!(IPCB(skb)->flags & IPSKB_FRAG_COMPLETE) &&
331	    unlikely(ip_frag_too_far(qp)) &&
332	    unlikely(err = ip_frag_reinit(qp))) {
333		ipq_kill(qp);
334		goto err;
335	}
336
337	ecn = ip4_frag_ecn(ip_hdr(skb)->tos);
338	offset = ntohs(ip_hdr(skb)->frag_off);
339	flags = offset & ~IP_OFFSET;
340	offset &= IP_OFFSET;
341	offset <<= 3;		/* offset is in 8-byte chunks */
342	ihl = ip_hdrlen(skb);
343
344	/* Determine the position of this fragment. */
345	end = offset + skb->len - skb_network_offset(skb) - ihl;
346	err = -EINVAL;
347
348	/* Is this the final fragment? */
349	if ((flags & IP_MF) == 0) {
350		/* If we already have some bits beyond end
351		 * or have different end, the segment is corrupted.
352		 */
353		if (end < qp->q.len ||
354		    ((qp->q.flags & INET_FRAG_LAST_IN) && end != qp->q.len))
355			goto err;
356		qp->q.flags |= INET_FRAG_LAST_IN;
357		qp->q.len = end;
358	} else {
359		if (end&7) {
360			end &= ~7;
361			if (skb->ip_summed != CHECKSUM_UNNECESSARY)
362				skb->ip_summed = CHECKSUM_NONE;
363		}
364		if (end > qp->q.len) {
365			/* Some bits beyond end -> corruption. */
366			if (qp->q.flags & INET_FRAG_LAST_IN)
367				goto err;
368			qp->q.len = end;
369		}
370	}
371	if (end == offset)
372		goto err;
373
374	err = -ENOMEM;
375	if (!pskb_pull(skb, skb_network_offset(skb) + ihl))
376		goto err;
377
378	err = pskb_trim_rcsum(skb, end - offset);
379	if (err)
380		goto err;
381
382	/* Find out which fragments are in front and at the back of us
383	 * in the chain of fragments so far.  We must know where to put
384	 * this fragment, right?
385	 */
386	prev = qp->q.fragments_tail;
387	if (!prev || FRAG_CB(prev)->offset < offset) {
388		next = NULL;
389		goto found;
390	}
391	prev = NULL;
392	for (next = qp->q.fragments; next != NULL; next = next->next) {
393		if (FRAG_CB(next)->offset >= offset)
394			break;	/* bingo! */
395		prev = next;
396	}
397
398found:
399	/* We found where to put this one.  Check for overlap with
400	 * preceding fragment, and, if needed, align things so that
401	 * any overlaps are eliminated.
402	 */
403	if (prev) {
404		int i = (FRAG_CB(prev)->offset + prev->len) - offset;
405
406		if (i > 0) {
407			offset += i;
408			err = -EINVAL;
409			if (end <= offset)
410				goto err;
411			err = -ENOMEM;
412			if (!pskb_pull(skb, i))
413				goto err;
414			if (skb->ip_summed != CHECKSUM_UNNECESSARY)
415				skb->ip_summed = CHECKSUM_NONE;
416		}
417	}
418
419	err = -ENOMEM;
420
421	while (next && FRAG_CB(next)->offset < end) {
422		int i = end - FRAG_CB(next)->offset; /* overlap is 'i' bytes */
423
424		if (i < next->len) {
425			/* Eat head of the next overlapped fragment
426			 * and leave the loop. The next ones cannot overlap.
427			 */
428			if (!pskb_pull(next, i))
429				goto err;
430			FRAG_CB(next)->offset += i;
431			qp->q.meat -= i;
432			if (next->ip_summed != CHECKSUM_UNNECESSARY)
433				next->ip_summed = CHECKSUM_NONE;
434			break;
435		} else {
436			struct sk_buff *free_it = next;
437
438			/* Old fragment is completely overridden with
439			 * new one drop it.
440			 */
441			next = next->next;
442
443			if (prev)
444				prev->next = next;
445			else
446				qp->q.fragments = next;
447
448			qp->q.meat -= free_it->len;
449			sub_frag_mem_limit(&qp->q, free_it->truesize);
450			kfree_skb(free_it);
451		}
452	}
453
454	FRAG_CB(skb)->offset = offset;
455
456	/* Insert this fragment in the chain of fragments. */
457	skb->next = next;
458	if (!next)
459		qp->q.fragments_tail = skb;
460	if (prev)
461		prev->next = skb;
462	else
463		qp->q.fragments = skb;
464
465	dev = skb->dev;
466	if (dev) {
467		qp->iif = dev->ifindex;
468		skb->dev = NULL;
469	}
470	qp->q.stamp = skb->tstamp;
471	qp->q.meat += skb->len;
472	qp->ecn |= ecn;
473	add_frag_mem_limit(&qp->q, skb->truesize);
474	if (offset == 0)
475		qp->q.flags |= INET_FRAG_FIRST_IN;
476
477	if (ip_hdr(skb)->frag_off & htons(IP_DF) &&
478	    skb->len + ihl > qp->q.max_size)
479		qp->q.max_size = skb->len + ihl;
480
481	if (qp->q.flags == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
482	    qp->q.meat == qp->q.len) {
483		unsigned long orefdst = skb->_skb_refdst;
484
485		skb->_skb_refdst = 0UL;
486		err = ip_frag_reasm(qp, prev, dev);
487		skb->_skb_refdst = orefdst;
488		return err;
489	}
490
491	skb_dst_drop(skb);
492	return -EINPROGRESS;
493
494err:
495	kfree_skb(skb);
496	return err;
497}
498
499
500/* Build a new IP datagram from all its fragments. */
501
502static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev,
503			 struct net_device *dev)
504{
505	struct net *net = container_of(qp->q.net, struct net, ipv4.frags);
506	struct iphdr *iph;
507	struct sk_buff *fp, *head = qp->q.fragments;
508	int len;
509	int ihlen;
510	int err;
511	int sum_truesize;
512	u8 ecn;
513
514	ipq_kill(qp);
515
516	ecn = ip_frag_ecn_table[qp->ecn];
517	if (unlikely(ecn == 0xff)) {
518		err = -EINVAL;
519		goto out_fail;
520	}
521	/* Make the one we just received the head. */
522	if (prev) {
523		head = prev->next;
524		fp = skb_clone(head, GFP_ATOMIC);
525		if (!fp)
526			goto out_nomem;
527
528		fp->next = head->next;
529		if (!fp->next)
530			qp->q.fragments_tail = fp;
531		prev->next = fp;
532
533		skb_morph(head, qp->q.fragments);
534		head->next = qp->q.fragments->next;
535
536		consume_skb(qp->q.fragments);
537		qp->q.fragments = head;
538	}
539
540	WARN_ON(!head);
541	WARN_ON(FRAG_CB(head)->offset != 0);
542
543	/* Allocate a new buffer for the datagram. */
544	ihlen = ip_hdrlen(head);
545	len = ihlen + qp->q.len;
546
547	err = -E2BIG;
548	if (len > 65535)
549		goto out_oversize;
550
551	/* Head of list must not be cloned. */
552	if (skb_unclone(head, GFP_ATOMIC))
553		goto out_nomem;
554
555	/* If the first fragment is fragmented itself, we split
556	 * it to two chunks: the first with data and paged part
557	 * and the second, holding only fragments. */
558	if (skb_has_frag_list(head)) {
559		struct sk_buff *clone;
560		int i, plen = 0;
561
562		clone = alloc_skb(0, GFP_ATOMIC);
563		if (!clone)
564			goto out_nomem;
565		clone->next = head->next;
566		head->next = clone;
567		skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
568		skb_frag_list_init(head);
569		for (i = 0; i < skb_shinfo(head)->nr_frags; i++)
570			plen += skb_frag_size(&skb_shinfo(head)->frags[i]);
571		clone->len = clone->data_len = head->data_len - plen;
572		head->data_len -= clone->len;
573		head->len -= clone->len;
574		clone->csum = 0;
575		clone->ip_summed = head->ip_summed;
576		add_frag_mem_limit(&qp->q, clone->truesize);
577	}
578
579	skb_push(head, head->data - skb_network_header(head));
580
581	sum_truesize = head->truesize;
582	for (fp = head->next; fp;) {
583		bool headstolen;
584		int delta;
585		struct sk_buff *next = fp->next;
586
587		sum_truesize += fp->truesize;
588		if (head->ip_summed != fp->ip_summed)
589			head->ip_summed = CHECKSUM_NONE;
590		else if (head->ip_summed == CHECKSUM_COMPLETE)
591			head->csum = csum_add(head->csum, fp->csum);
592
593		if (skb_try_coalesce(head, fp, &headstolen, &delta)) {
594			kfree_skb_partial(fp, headstolen);
595		} else {
596			if (!skb_shinfo(head)->frag_list)
597				skb_shinfo(head)->frag_list = fp;
598			head->data_len += fp->len;
599			head->len += fp->len;
600			head->truesize += fp->truesize;
601		}
602		fp = next;
603	}
604	sub_frag_mem_limit(&qp->q, sum_truesize);
605
606	head->next = NULL;
607	head->dev = dev;
608	head->tstamp = qp->q.stamp;
609	IPCB(head)->frag_max_size = qp->q.max_size;
610
611	iph = ip_hdr(head);
612	/* max_size != 0 implies at least one fragment had IP_DF set */
613	iph->frag_off = qp->q.max_size ? htons(IP_DF) : 0;
614	iph->tot_len = htons(len);
615	iph->tos |= ecn;
616
617	ip_send_check(iph);
618
619	IP_INC_STATS_BH(net, IPSTATS_MIB_REASMOKS);
620	qp->q.fragments = NULL;
621	qp->q.fragments_tail = NULL;
622	return 0;
623
624out_nomem:
625	net_dbg_ratelimited("queue_glue: no memory for gluing queue %p\n", qp);
626	err = -ENOMEM;
627	goto out_fail;
628out_oversize:
629	net_info_ratelimited("Oversized IP packet from %pI4\n", &qp->saddr);
630out_fail:
631	IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
632	return err;
633}
634
635/* Process an incoming IP datagram fragment. */
636int ip_defrag(struct sk_buff *skb, u32 user)
637{
638	struct ipq *qp;
639	struct net *net;
640
641	net = skb->dev ? dev_net(skb->dev) : dev_net(skb_dst(skb)->dev);
642	IP_INC_STATS_BH(net, IPSTATS_MIB_REASMREQDS);
643
644	/* Lookup (or create) queue header */
645	qp = ip_find(net, ip_hdr(skb), user);
646	if (qp) {
647		int ret;
648
649		spin_lock(&qp->q.lock);
650
651		ret = ip_frag_queue(qp, skb);
652
653		spin_unlock(&qp->q.lock);
654		ipq_put(qp);
655		return ret;
656	}
657
658	IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
659	kfree_skb(skb);
660	return -ENOMEM;
661}
662EXPORT_SYMBOL(ip_defrag);
663
664struct sk_buff *ip_check_defrag(struct sk_buff *skb, u32 user)
665{
666	struct iphdr iph;
667	int netoff;
668	u32 len;
669
670	if (skb->protocol != htons(ETH_P_IP))
671		return skb;
672
673	netoff = skb_network_offset(skb);
674
675	if (skb_copy_bits(skb, netoff, &iph, sizeof(iph)) < 0)
676		return skb;
677
678	if (iph.ihl < 5 || iph.version != 4)
679		return skb;
680
681	len = ntohs(iph.tot_len);
682	if (skb->len < netoff + len || len < (iph.ihl * 4))
683		return skb;
684
685	if (ip_is_fragment(&iph)) {
686		skb = skb_share_check(skb, GFP_ATOMIC);
687		if (skb) {
688			if (!pskb_may_pull(skb, netoff + iph.ihl * 4))
689				return skb;
690			if (pskb_trim_rcsum(skb, netoff + len))
691				return skb;
692			memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
693			if (ip_defrag(skb, user))
694				return NULL;
695			skb_clear_hash(skb);
696		}
697	}
698	return skb;
699}
700EXPORT_SYMBOL(ip_check_defrag);
701
702#ifdef CONFIG_SYSCTL
703static int zero;
704
705static struct ctl_table ip4_frags_ns_ctl_table[] = {
706	{
707		.procname	= "ipfrag_high_thresh",
708		.data		= &init_net.ipv4.frags.high_thresh,
709		.maxlen		= sizeof(int),
710		.mode		= 0644,
711		.proc_handler	= proc_dointvec_minmax,
712		.extra1		= &init_net.ipv4.frags.low_thresh
713	},
714	{
715		.procname	= "ipfrag_low_thresh",
716		.data		= &init_net.ipv4.frags.low_thresh,
717		.maxlen		= sizeof(int),
718		.mode		= 0644,
719		.proc_handler	= proc_dointvec_minmax,
720		.extra1		= &zero,
721		.extra2		= &init_net.ipv4.frags.high_thresh
722	},
723	{
724		.procname	= "ipfrag_time",
725		.data		= &init_net.ipv4.frags.timeout,
726		.maxlen		= sizeof(int),
727		.mode		= 0644,
728		.proc_handler	= proc_dointvec_jiffies,
729	},
730	{ }
731};
732
733/* secret interval has been deprecated */
734static int ip4_frags_secret_interval_unused;
735static struct ctl_table ip4_frags_ctl_table[] = {
736	{
737		.procname	= "ipfrag_secret_interval",
738		.data		= &ip4_frags_secret_interval_unused,
739		.maxlen		= sizeof(int),
740		.mode		= 0644,
741		.proc_handler	= proc_dointvec_jiffies,
742	},
743	{
744		.procname	= "ipfrag_max_dist",
745		.data		= &sysctl_ipfrag_max_dist,
746		.maxlen		= sizeof(int),
747		.mode		= 0644,
748		.proc_handler	= proc_dointvec_minmax,
749		.extra1		= &zero
750	},
751	{ }
752};
753
754static int __net_init ip4_frags_ns_ctl_register(struct net *net)
755{
756	struct ctl_table *table;
757	struct ctl_table_header *hdr;
758
759	table = ip4_frags_ns_ctl_table;
760	if (!net_eq(net, &init_net)) {
761		table = kmemdup(table, sizeof(ip4_frags_ns_ctl_table), GFP_KERNEL);
762		if (!table)
763			goto err_alloc;
764
765		table[0].data = &net->ipv4.frags.high_thresh;
766		table[0].extra1 = &net->ipv4.frags.low_thresh;
767		table[0].extra2 = &init_net.ipv4.frags.high_thresh;
768		table[1].data = &net->ipv4.frags.low_thresh;
769		table[1].extra2 = &net->ipv4.frags.high_thresh;
770		table[2].data = &net->ipv4.frags.timeout;
771
772		/* Don't export sysctls to unprivileged users */
773		if (net->user_ns != &init_user_ns)
774			table[0].procname = NULL;
775	}
776
777	hdr = register_net_sysctl(net, "net/ipv4", table);
778	if (!hdr)
779		goto err_reg;
780
781	net->ipv4.frags_hdr = hdr;
782	return 0;
783
784err_reg:
785	if (!net_eq(net, &init_net))
786		kfree(table);
787err_alloc:
788	return -ENOMEM;
789}
790
791static void __net_exit ip4_frags_ns_ctl_unregister(struct net *net)
792{
793	struct ctl_table *table;
794
795	table = net->ipv4.frags_hdr->ctl_table_arg;
796	unregister_net_sysctl_table(net->ipv4.frags_hdr);
797	kfree(table);
798}
799
800static void __init ip4_frags_ctl_register(void)
801{
802	register_net_sysctl(&init_net, "net/ipv4", ip4_frags_ctl_table);
803}
804#else
805static int ip4_frags_ns_ctl_register(struct net *net)
806{
807	return 0;
808}
809
810static void ip4_frags_ns_ctl_unregister(struct net *net)
811{
812}
813
814static void __init ip4_frags_ctl_register(void)
815{
816}
817#endif
818
819static int __net_init ipv4_frags_init_net(struct net *net)
820{
821	/* Fragment cache limits.
822	 *
823	 * The fragment memory accounting code, (tries to) account for
824	 * the real memory usage, by measuring both the size of frag
825	 * queue struct (inet_frag_queue (ipv4:ipq/ipv6:frag_queue))
826	 * and the SKB's truesize.
827	 *
828	 * A 64K fragment consumes 129736 bytes (44*2944)+200
829	 * (1500 truesize == 2944, sizeof(struct ipq) == 200)
830	 *
831	 * We will commit 4MB at one time. Should we cross that limit
832	 * we will prune down to 3MB, making room for approx 8 big 64K
833	 * fragments 8x128k.
834	 */
835	net->ipv4.frags.high_thresh = 4 * 1024 * 1024;
836	net->ipv4.frags.low_thresh  = 3 * 1024 * 1024;
837	/*
838	 * Important NOTE! Fragment queue must be destroyed before MSL expires.
839	 * RFC791 is wrong proposing to prolongate timer each fragment arrival
840	 * by TTL.
841	 */
842	net->ipv4.frags.timeout = IP_FRAG_TIME;
843
844	inet_frags_init_net(&net->ipv4.frags);
845
846	return ip4_frags_ns_ctl_register(net);
847}
848
849static void __net_exit ipv4_frags_exit_net(struct net *net)
850{
851	ip4_frags_ns_ctl_unregister(net);
852	inet_frags_exit_net(&net->ipv4.frags, &ip4_frags);
853}
854
855static struct pernet_operations ip4_frags_ops = {
856	.init = ipv4_frags_init_net,
857	.exit = ipv4_frags_exit_net,
858};
859
860void __init ipfrag_init(void)
861{
862	ip4_frags_ctl_register();
863	register_pernet_subsys(&ip4_frags_ops);
864	ip4_frags.hashfn = ip4_hashfn;
865	ip4_frags.constructor = ip4_frag_init;
866	ip4_frags.destructor = ip4_frag_free;
867	ip4_frags.skb_free = NULL;
868	ip4_frags.qsize = sizeof(struct ipq);
869	ip4_frags.match = ip4_frag_match;
870	ip4_frags.frag_expire = ip_expire;
871	ip4_frags.frags_cache_name = ip_frag_cache_name;
872	if (inet_frags_init(&ip4_frags))
873		panic("IP: failed to allocate ip4_frags cache\n");
874}
875