Linux Audio

Check our new training course

Loading...
v6.13.7
  1// SPDX-License-Identifier: GPL-2.0
  2/*
  3 * INET		An implementation of the TCP/IP protocol suite for the LINUX
  4 *		operating system.  INET is implemented using the  BSD Socket
  5 *		interface as the means of communication with the user level.
  6 *
  7 *		The IP fragmentation functionality.
  8 *
  9 * Authors:	Fred N. van Kempen <waltje@uWalt.NL.Mugnet.ORG>
 10 *		Alan Cox <alan@lxorguk.ukuu.org.uk>
 11 *
 12 * Fixes:
 13 *		Alan Cox	:	Split from ip.c , see ip_input.c for history.
 14 *		David S. Miller :	Begin massive cleanup...
 15 *		Andi Kleen	:	Add sysctls.
 16 *		xxxx		:	Overlapfrag bug.
 17 *		Ultima          :       ip_expire() kernel panic.
 18 *		Bill Hawes	:	Frag accounting and evictor fixes.
 19 *		John McDonald	:	0 length frag bug.
 20 *		Alexey Kuznetsov:	SMP races, threading, cleanup.
 21 *		Patrick McHardy :	LRU queue of frag heads for evictor.
 22 */
 23
 24#define pr_fmt(fmt) "IPv4: " fmt
 25
 26#include <linux/compiler.h>
 27#include <linux/module.h>
 28#include <linux/types.h>
 29#include <linux/mm.h>
 30#include <linux/jiffies.h>
 31#include <linux/skbuff.h>
 32#include <linux/list.h>
 33#include <linux/ip.h>
 34#include <linux/icmp.h>
 35#include <linux/netdevice.h>
 36#include <linux/jhash.h>
 37#include <linux/random.h>
 38#include <linux/slab.h>
 39#include <net/route.h>
 40#include <net/dst.h>
 41#include <net/sock.h>
 42#include <net/ip.h>
 43#include <net/icmp.h>
 44#include <net/checksum.h>
 45#include <net/inetpeer.h>
 46#include <net/inet_frag.h>
 47#include <linux/tcp.h>
 48#include <linux/udp.h>
 49#include <linux/inet.h>
 50#include <linux/netfilter_ipv4.h>
 51#include <net/inet_ecn.h>
 52#include <net/l3mdev.h>
 53
 54/* NOTE. Logic of IP defragmentation is parallel to corresponding IPv6
 55 * code now. If you change something here, _PLEASE_ update ipv6/reassembly.c
 56 * as well. Or notify me, at least. --ANK
 57 */
 58static const char ip_frag_cache_name[] = "ip4-frags";
 59
 60/* Describe an entry in the "incomplete datagrams" queue. */
 61struct ipq {
 62	struct inet_frag_queue q;
 63
 64	u8		ecn; /* RFC3168 support */
 65	u16		max_df_size; /* largest frag with DF set seen */
 66	int             iif;
 67	unsigned int    rid;
 68	struct inet_peer *peer;
 69};
 70
 71static u8 ip4_frag_ecn(u8 tos)
 72{
 73	return 1 << (tos & INET_ECN_MASK);
 74}
 75
 76static struct inet_frags ip4_frags;
 77
 78static int ip_frag_reasm(struct ipq *qp, struct sk_buff *skb,
 79			 struct sk_buff *prev_tail, struct net_device *dev);
 80
 81
 82static void ip4_frag_init(struct inet_frag_queue *q, const void *a)
 83{
 84	struct ipq *qp = container_of(q, struct ipq, q);
 85	const struct frag_v4_compare_key *key = a;
 86	struct net *net = q->fqdir->net;
 87	struct inet_peer *p = NULL;
 
 88
 89	q->key.v4 = *key;
 90	qp->ecn = 0;
 91	if (q->fqdir->max_dist) {
 92		rcu_read_lock();
 93		p = inet_getpeer_v4(net->ipv4.peers, key->saddr, key->vif);
 94		if (p && !refcount_inc_not_zero(&p->refcnt))
 95			p = NULL;
 96		rcu_read_unlock();
 97	}
 98	qp->peer = p;
 99}
100
101static void ip4_frag_free(struct inet_frag_queue *q)
102{
103	struct ipq *qp;
104
105	qp = container_of(q, struct ipq, q);
106	if (qp->peer)
107		inet_putpeer(qp->peer);
108}
109
110
111/* Destruction primitives. */
112
113static void ipq_put(struct ipq *ipq)
114{
115	inet_frag_put(&ipq->q);
116}
117
118/* Kill ipq entry. It is not destroyed immediately,
119 * because caller (and someone more) holds reference count.
120 */
121static void ipq_kill(struct ipq *ipq)
122{
123	inet_frag_kill(&ipq->q);
124}
125
126static bool frag_expire_skip_icmp(u32 user)
127{
128	return user == IP_DEFRAG_AF_PACKET ||
129	       ip_defrag_user_in_between(user, IP_DEFRAG_CONNTRACK_IN,
130					 __IP_DEFRAG_CONNTRACK_IN_END) ||
131	       ip_defrag_user_in_between(user, IP_DEFRAG_CONNTRACK_BRIDGE_IN,
132					 __IP_DEFRAG_CONNTRACK_BRIDGE_IN);
133}
134
135/*
136 * Oops, a fragment queue timed out.  Kill it and send an ICMP reply.
137 */
138static void ip_expire(struct timer_list *t)
139{
140	enum skb_drop_reason reason = SKB_DROP_REASON_FRAG_REASM_TIMEOUT;
141	struct inet_frag_queue *frag = from_timer(frag, t, timer);
142	const struct iphdr *iph;
143	struct sk_buff *head = NULL;
144	struct net *net;
145	struct ipq *qp;
 
146
147	qp = container_of(frag, struct ipq, q);
148	net = qp->q.fqdir->net;
149
150	rcu_read_lock();
151
152	/* Paired with WRITE_ONCE() in fqdir_pre_exit(). */
153	if (READ_ONCE(qp->q.fqdir->dead))
154		goto out_rcu_unlock;
155
156	spin_lock(&qp->q.lock);
157
158	if (qp->q.flags & INET_FRAG_COMPLETE)
159		goto out;
160
161	qp->q.flags |= INET_FRAG_DROP;
162	ipq_kill(qp);
163	__IP_INC_STATS(net, IPSTATS_MIB_REASMFAILS);
164	__IP_INC_STATS(net, IPSTATS_MIB_REASMTIMEOUT);
165
166	if (!(qp->q.flags & INET_FRAG_FIRST_IN))
167		goto out;
168
169	/* sk_buff::dev and sk_buff::rbnode are unionized. So we
170	 * pull the head out of the tree in order to be able to
171	 * deal with head->dev.
172	 */
173	head = inet_frag_pull_head(&qp->q);
174	if (!head)
175		goto out;
176	head->dev = dev_get_by_index_rcu(net, qp->iif);
177	if (!head->dev)
178		goto out;
179
180
181	/* skb has no dst, perform route lookup again */
182	iph = ip_hdr(head);
183	reason = ip_route_input_noref(head, iph->daddr, iph->saddr,
184				      ip4h_dscp(iph), head->dev);
185	if (reason)
186		goto out;
187
188	/* Only an end host needs to send an ICMP
189	 * "Fragment Reassembly Timeout" message, per RFC792.
190	 */
191	reason = SKB_DROP_REASON_FRAG_REASM_TIMEOUT;
192	if (frag_expire_skip_icmp(qp->q.key.v4.user) &&
193	    (skb_rtable(head)->rt_type != RTN_LOCAL))
194		goto out;
195
196	spin_unlock(&qp->q.lock);
197	icmp_send(head, ICMP_TIME_EXCEEDED, ICMP_EXC_FRAGTIME, 0);
198	goto out_rcu_unlock;
199
200out:
201	spin_unlock(&qp->q.lock);
202out_rcu_unlock:
203	rcu_read_unlock();
204	kfree_skb_reason(head, reason);
205	ipq_put(qp);
206}
207
208/* Find the correct entry in the "incomplete datagrams" queue for
209 * this IP datagram, and create new one, if nothing is found.
210 */
211static struct ipq *ip_find(struct net *net, struct iphdr *iph,
212			   u32 user, int vif)
213{
214	struct frag_v4_compare_key key = {
215		.saddr = iph->saddr,
216		.daddr = iph->daddr,
217		.user = user,
218		.vif = vif,
219		.id = iph->id,
220		.protocol = iph->protocol,
221	};
222	struct inet_frag_queue *q;
223
224	q = inet_frag_find(net->ipv4.fqdir, &key);
225	if (!q)
226		return NULL;
227
228	return container_of(q, struct ipq, q);
229}
230
231/* Is the fragment too far ahead to be part of ipq? */
232static int ip_frag_too_far(struct ipq *qp)
233{
234	struct inet_peer *peer = qp->peer;
235	unsigned int max = qp->q.fqdir->max_dist;
236	unsigned int start, end;
237
238	int rc;
239
240	if (!peer || !max)
241		return 0;
242
243	start = qp->rid;
244	end = atomic_inc_return(&peer->rid);
245	qp->rid = end;
246
247	rc = qp->q.fragments_tail && (end - start) > max;
248
249	if (rc)
250		__IP_INC_STATS(qp->q.fqdir->net, IPSTATS_MIB_REASMFAILS);
251
252	return rc;
253}
254
255static int ip_frag_reinit(struct ipq *qp)
256{
257	unsigned int sum_truesize = 0;
258
259	if (!mod_timer(&qp->q.timer, jiffies + qp->q.fqdir->timeout)) {
260		refcount_inc(&qp->q.refcnt);
261		return -ETIMEDOUT;
262	}
263
264	sum_truesize = inet_frag_rbtree_purge(&qp->q.rb_fragments,
265					      SKB_DROP_REASON_FRAG_TOO_FAR);
266	sub_frag_mem_limit(qp->q.fqdir, sum_truesize);
267
268	qp->q.flags = 0;
269	qp->q.len = 0;
270	qp->q.meat = 0;
271	qp->q.rb_fragments = RB_ROOT;
272	qp->q.fragments_tail = NULL;
273	qp->q.last_run_head = NULL;
274	qp->iif = 0;
275	qp->ecn = 0;
276
277	return 0;
278}
279
280/* Add new segment to existing queue. */
281static int ip_frag_queue(struct ipq *qp, struct sk_buff *skb)
282{
283	struct net *net = qp->q.fqdir->net;
284	int ihl, end, flags, offset;
285	struct sk_buff *prev_tail;
286	struct net_device *dev;
287	unsigned int fragsize;
288	int err = -ENOENT;
289	SKB_DR(reason);
290	u8 ecn;
291
292	/* If reassembly is already done, @skb must be a duplicate frag. */
293	if (qp->q.flags & INET_FRAG_COMPLETE) {
294		SKB_DR_SET(reason, DUP_FRAG);
295		goto err;
296	}
297
298	if (!(IPCB(skb)->flags & IPSKB_FRAG_COMPLETE) &&
299	    unlikely(ip_frag_too_far(qp)) &&
300	    unlikely(err = ip_frag_reinit(qp))) {
301		ipq_kill(qp);
302		goto err;
303	}
304
305	ecn = ip4_frag_ecn(ip_hdr(skb)->tos);
306	offset = ntohs(ip_hdr(skb)->frag_off);
307	flags = offset & ~IP_OFFSET;
308	offset &= IP_OFFSET;
309	offset <<= 3;		/* offset is in 8-byte chunks */
310	ihl = ip_hdrlen(skb);
311
312	/* Determine the position of this fragment. */
313	end = offset + skb->len - skb_network_offset(skb) - ihl;
314	err = -EINVAL;
315
316	/* Is this the final fragment? */
317	if ((flags & IP_MF) == 0) {
318		/* If we already have some bits beyond end
319		 * or have different end, the segment is corrupted.
320		 */
321		if (end < qp->q.len ||
322		    ((qp->q.flags & INET_FRAG_LAST_IN) && end != qp->q.len))
323			goto discard_qp;
324		qp->q.flags |= INET_FRAG_LAST_IN;
325		qp->q.len = end;
326	} else {
327		if (end&7) {
328			end &= ~7;
329			if (skb->ip_summed != CHECKSUM_UNNECESSARY)
330				skb->ip_summed = CHECKSUM_NONE;
331		}
332		if (end > qp->q.len) {
333			/* Some bits beyond end -> corruption. */
334			if (qp->q.flags & INET_FRAG_LAST_IN)
335				goto discard_qp;
336			qp->q.len = end;
337		}
338	}
339	if (end == offset)
340		goto discard_qp;
341
342	err = -ENOMEM;
343	if (!pskb_pull(skb, skb_network_offset(skb) + ihl))
344		goto discard_qp;
345
346	err = pskb_trim_rcsum(skb, end - offset);
347	if (err)
348		goto discard_qp;
349
350	/* Note : skb->rbnode and skb->dev share the same location. */
351	dev = skb->dev;
352	/* Makes sure compiler wont do silly aliasing games */
353	barrier();
354
355	prev_tail = qp->q.fragments_tail;
356	err = inet_frag_queue_insert(&qp->q, skb, offset, end);
357	if (err)
358		goto insert_error;
359
360	if (dev)
361		qp->iif = dev->ifindex;
362
363	qp->q.stamp = skb->tstamp;
364	qp->q.tstamp_type = skb->tstamp_type;
365	qp->q.meat += skb->len;
366	qp->ecn |= ecn;
367	add_frag_mem_limit(qp->q.fqdir, skb->truesize);
368	if (offset == 0)
369		qp->q.flags |= INET_FRAG_FIRST_IN;
370
371	fragsize = skb->len + ihl;
372
373	if (fragsize > qp->q.max_size)
374		qp->q.max_size = fragsize;
375
376	if (ip_hdr(skb)->frag_off & htons(IP_DF) &&
377	    fragsize > qp->max_df_size)
378		qp->max_df_size = fragsize;
379
380	if (qp->q.flags == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
381	    qp->q.meat == qp->q.len) {
382		unsigned long orefdst = skb->_skb_refdst;
383
384		skb->_skb_refdst = 0UL;
385		err = ip_frag_reasm(qp, skb, prev_tail, dev);
386		skb->_skb_refdst = orefdst;
387		if (err)
388			inet_frag_kill(&qp->q);
389		return err;
390	}
391
392	skb_dst_drop(skb);
393	skb_orphan(skb);
394	return -EINPROGRESS;
395
396insert_error:
397	if (err == IPFRAG_DUP) {
398		SKB_DR_SET(reason, DUP_FRAG);
399		err = -EINVAL;
400		goto err;
401	}
402	err = -EINVAL;
403	__IP_INC_STATS(net, IPSTATS_MIB_REASM_OVERLAPS);
404discard_qp:
405	inet_frag_kill(&qp->q);
406	__IP_INC_STATS(net, IPSTATS_MIB_REASMFAILS);
407err:
408	kfree_skb_reason(skb, reason);
409	return err;
410}
411
412static bool ip_frag_coalesce_ok(const struct ipq *qp)
413{
414	return qp->q.key.v4.user == IP_DEFRAG_LOCAL_DELIVER;
415}
416
417/* Build a new IP datagram from all its fragments. */
418static int ip_frag_reasm(struct ipq *qp, struct sk_buff *skb,
419			 struct sk_buff *prev_tail, struct net_device *dev)
420{
421	struct net *net = qp->q.fqdir->net;
422	struct iphdr *iph;
423	void *reasm_data;
424	int len, err;
425	u8 ecn;
426
427	ipq_kill(qp);
428
429	ecn = ip_frag_ecn_table[qp->ecn];
430	if (unlikely(ecn == 0xff)) {
431		err = -EINVAL;
432		goto out_fail;
433	}
434
435	/* Make the one we just received the head. */
436	reasm_data = inet_frag_reasm_prepare(&qp->q, skb, prev_tail);
437	if (!reasm_data)
438		goto out_nomem;
439
440	len = ip_hdrlen(skb) + qp->q.len;
441	err = -E2BIG;
442	if (len > 65535)
443		goto out_oversize;
444
445	inet_frag_reasm_finish(&qp->q, skb, reasm_data,
446			       ip_frag_coalesce_ok(qp));
447
448	skb->dev = dev;
449	IPCB(skb)->frag_max_size = max(qp->max_df_size, qp->q.max_size);
450
451	iph = ip_hdr(skb);
452	iph->tot_len = htons(len);
453	iph->tos |= ecn;
454
455	/* When we set IP_DF on a refragmented skb we must also force a
456	 * call to ip_fragment to avoid forwarding a DF-skb of size s while
457	 * original sender only sent fragments of size f (where f < s).
458	 *
459	 * We only set DF/IPSKB_FRAG_PMTU if such DF fragment was the largest
460	 * frag seen to avoid sending tiny DF-fragments in case skb was built
461	 * from one very small df-fragment and one large non-df frag.
462	 */
463	if (qp->max_df_size == qp->q.max_size) {
464		IPCB(skb)->flags |= IPSKB_FRAG_PMTU;
465		iph->frag_off = htons(IP_DF);
466	} else {
467		iph->frag_off = 0;
468	}
469
470	ip_send_check(iph);
471
472	__IP_INC_STATS(net, IPSTATS_MIB_REASMOKS);
473	qp->q.rb_fragments = RB_ROOT;
474	qp->q.fragments_tail = NULL;
475	qp->q.last_run_head = NULL;
476	return 0;
477
478out_nomem:
479	net_dbg_ratelimited("queue_glue: no memory for gluing queue %p\n", qp);
480	err = -ENOMEM;
481	goto out_fail;
482out_oversize:
483	net_info_ratelimited("Oversized IP packet from %pI4\n", &qp->q.key.v4.saddr);
484out_fail:
485	__IP_INC_STATS(net, IPSTATS_MIB_REASMFAILS);
486	return err;
487}
488
489/* Process an incoming IP datagram fragment. */
490int ip_defrag(struct net *net, struct sk_buff *skb, u32 user)
491{
492	struct net_device *dev = skb->dev ? : skb_dst(skb)->dev;
493	int vif = l3mdev_master_ifindex_rcu(dev);
494	struct ipq *qp;
495
496	__IP_INC_STATS(net, IPSTATS_MIB_REASMREQDS);
 
497
498	/* Lookup (or create) queue header */
499	qp = ip_find(net, ip_hdr(skb), user, vif);
500	if (qp) {
501		int ret;
502
503		spin_lock(&qp->q.lock);
504
505		ret = ip_frag_queue(qp, skb);
506
507		spin_unlock(&qp->q.lock);
508		ipq_put(qp);
509		return ret;
510	}
511
512	__IP_INC_STATS(net, IPSTATS_MIB_REASMFAILS);
513	kfree_skb(skb);
514	return -ENOMEM;
515}
516EXPORT_SYMBOL(ip_defrag);
517
518struct sk_buff *ip_check_defrag(struct net *net, struct sk_buff *skb, u32 user)
519{
520	struct iphdr iph;
521	int netoff;
522	u32 len;
523
524	if (skb->protocol != htons(ETH_P_IP))
525		return skb;
526
527	netoff = skb_network_offset(skb);
528
529	if (skb_copy_bits(skb, netoff, &iph, sizeof(iph)) < 0)
530		return skb;
531
532	if (iph.ihl < 5 || iph.version != 4)
533		return skb;
534
535	len = ntohs(iph.tot_len);
536	if (skb->len < netoff + len || len < (iph.ihl * 4))
537		return skb;
538
539	if (ip_is_fragment(&iph)) {
540		skb = skb_share_check(skb, GFP_ATOMIC);
541		if (skb) {
542			if (!pskb_may_pull(skb, netoff + iph.ihl * 4)) {
543				kfree_skb(skb);
544				return NULL;
545			}
546			if (pskb_trim_rcsum(skb, netoff + len)) {
547				kfree_skb(skb);
548				return NULL;
549			}
550			memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
551			if (ip_defrag(net, skb, user))
552				return NULL;
553			skb_clear_hash(skb);
554		}
555	}
556	return skb;
557}
558EXPORT_SYMBOL(ip_check_defrag);
559
560#ifdef CONFIG_SYSCTL
561static int dist_min;
562
563static struct ctl_table ip4_frags_ns_ctl_table[] = {
564	{
565		.procname	= "ipfrag_high_thresh",
566		.maxlen		= sizeof(unsigned long),
567		.mode		= 0644,
568		.proc_handler	= proc_doulongvec_minmax,
569	},
570	{
571		.procname	= "ipfrag_low_thresh",
572		.maxlen		= sizeof(unsigned long),
573		.mode		= 0644,
574		.proc_handler	= proc_doulongvec_minmax,
575	},
576	{
577		.procname	= "ipfrag_time",
578		.maxlen		= sizeof(int),
579		.mode		= 0644,
580		.proc_handler	= proc_dointvec_jiffies,
581	},
582	{
583		.procname	= "ipfrag_max_dist",
584		.maxlen		= sizeof(int),
585		.mode		= 0644,
586		.proc_handler	= proc_dointvec_minmax,
587		.extra1		= &dist_min,
588	},
 
589};
590
591/* secret interval has been deprecated */
592static int ip4_frags_secret_interval_unused;
593static struct ctl_table ip4_frags_ctl_table[] = {
594	{
595		.procname	= "ipfrag_secret_interval",
596		.data		= &ip4_frags_secret_interval_unused,
597		.maxlen		= sizeof(int),
598		.mode		= 0644,
599		.proc_handler	= proc_dointvec_jiffies,
600	},
 
601};
602
603static int __net_init ip4_frags_ns_ctl_register(struct net *net)
604{
605	struct ctl_table *table;
606	struct ctl_table_header *hdr;
607
608	table = ip4_frags_ns_ctl_table;
609	if (!net_eq(net, &init_net)) {
610		table = kmemdup(table, sizeof(ip4_frags_ns_ctl_table), GFP_KERNEL);
611		if (!table)
612			goto err_alloc;
613
614	}
615	table[0].data	= &net->ipv4.fqdir->high_thresh;
616	table[0].extra1	= &net->ipv4.fqdir->low_thresh;
617	table[1].data	= &net->ipv4.fqdir->low_thresh;
618	table[1].extra2	= &net->ipv4.fqdir->high_thresh;
619	table[2].data	= &net->ipv4.fqdir->timeout;
620	table[3].data	= &net->ipv4.fqdir->max_dist;
621
622	hdr = register_net_sysctl_sz(net, "net/ipv4", table,
623				     ARRAY_SIZE(ip4_frags_ns_ctl_table));
624	if (!hdr)
625		goto err_reg;
626
627	net->ipv4.frags_hdr = hdr;
628	return 0;
629
630err_reg:
631	if (!net_eq(net, &init_net))
632		kfree(table);
633err_alloc:
634	return -ENOMEM;
635}
636
637static void __net_exit ip4_frags_ns_ctl_unregister(struct net *net)
638{
639	const struct ctl_table *table;
640
641	table = net->ipv4.frags_hdr->ctl_table_arg;
642	unregister_net_sysctl_table(net->ipv4.frags_hdr);
643	kfree(table);
644}
645
646static void __init ip4_frags_ctl_register(void)
647{
648	register_net_sysctl(&init_net, "net/ipv4", ip4_frags_ctl_table);
649}
650#else
651static int ip4_frags_ns_ctl_register(struct net *net)
652{
653	return 0;
654}
655
656static void ip4_frags_ns_ctl_unregister(struct net *net)
657{
658}
659
660static void __init ip4_frags_ctl_register(void)
661{
662}
663#endif
664
665static int __net_init ipv4_frags_init_net(struct net *net)
666{
667	int res;
668
669	res = fqdir_init(&net->ipv4.fqdir, &ip4_frags, net);
670	if (res < 0)
671		return res;
672	/* Fragment cache limits.
673	 *
674	 * The fragment memory accounting code, (tries to) account for
675	 * the real memory usage, by measuring both the size of frag
676	 * queue struct (inet_frag_queue (ipv4:ipq/ipv6:frag_queue))
677	 * and the SKB's truesize.
678	 *
679	 * A 64K fragment consumes 129736 bytes (44*2944)+200
680	 * (1500 truesize == 2944, sizeof(struct ipq) == 200)
681	 *
682	 * We will commit 4MB at one time. Should we cross that limit
683	 * we will prune down to 3MB, making room for approx 8 big 64K
684	 * fragments 8x128k.
685	 */
686	net->ipv4.fqdir->high_thresh = 4 * 1024 * 1024;
687	net->ipv4.fqdir->low_thresh  = 3 * 1024 * 1024;
688	/*
689	 * Important NOTE! Fragment queue must be destroyed before MSL expires.
690	 * RFC791 is wrong proposing to prolongate timer each fragment arrival
691	 * by TTL.
692	 */
693	net->ipv4.fqdir->timeout = IP_FRAG_TIME;
694
695	net->ipv4.fqdir->max_dist = 64;
696
697	res = ip4_frags_ns_ctl_register(net);
698	if (res < 0)
699		fqdir_exit(net->ipv4.fqdir);
700	return res;
701}
702
703static void __net_exit ipv4_frags_pre_exit_net(struct net *net)
704{
705	fqdir_pre_exit(net->ipv4.fqdir);
706}
707
708static void __net_exit ipv4_frags_exit_net(struct net *net)
709{
710	ip4_frags_ns_ctl_unregister(net);
711	fqdir_exit(net->ipv4.fqdir);
712}
713
714static struct pernet_operations ip4_frags_ops = {
715	.init		= ipv4_frags_init_net,
716	.pre_exit	= ipv4_frags_pre_exit_net,
717	.exit		= ipv4_frags_exit_net,
718};
719
720
721static u32 ip4_key_hashfn(const void *data, u32 len, u32 seed)
722{
723	return jhash2(data,
724		      sizeof(struct frag_v4_compare_key) / sizeof(u32), seed);
725}
726
727static u32 ip4_obj_hashfn(const void *data, u32 len, u32 seed)
728{
729	const struct inet_frag_queue *fq = data;
730
731	return jhash2((const u32 *)&fq->key.v4,
732		      sizeof(struct frag_v4_compare_key) / sizeof(u32), seed);
733}
734
735static int ip4_obj_cmpfn(struct rhashtable_compare_arg *arg, const void *ptr)
736{
737	const struct frag_v4_compare_key *key = arg->key;
738	const struct inet_frag_queue *fq = ptr;
739
740	return !!memcmp(&fq->key, key, sizeof(*key));
741}
742
743static const struct rhashtable_params ip4_rhash_params = {
744	.head_offset		= offsetof(struct inet_frag_queue, node),
745	.key_offset		= offsetof(struct inet_frag_queue, key),
746	.key_len		= sizeof(struct frag_v4_compare_key),
747	.hashfn			= ip4_key_hashfn,
748	.obj_hashfn		= ip4_obj_hashfn,
749	.obj_cmpfn		= ip4_obj_cmpfn,
750	.automatic_shrinking	= true,
751};
752
753void __init ipfrag_init(void)
754{
755	ip4_frags.constructor = ip4_frag_init;
756	ip4_frags.destructor = ip4_frag_free;
757	ip4_frags.qsize = sizeof(struct ipq);
758	ip4_frags.frag_expire = ip_expire;
759	ip4_frags.frags_cache_name = ip_frag_cache_name;
760	ip4_frags.rhash_params = ip4_rhash_params;
761	if (inet_frags_init(&ip4_frags))
762		panic("IP: failed to allocate ip4_frags cache\n");
763	ip4_frags_ctl_register();
764	register_pernet_subsys(&ip4_frags_ops);
765}
v6.2
  1// SPDX-License-Identifier: GPL-2.0
  2/*
  3 * INET		An implementation of the TCP/IP protocol suite for the LINUX
  4 *		operating system.  INET is implemented using the  BSD Socket
  5 *		interface as the means of communication with the user level.
  6 *
  7 *		The IP fragmentation functionality.
  8 *
  9 * Authors:	Fred N. van Kempen <waltje@uWalt.NL.Mugnet.ORG>
 10 *		Alan Cox <alan@lxorguk.ukuu.org.uk>
 11 *
 12 * Fixes:
 13 *		Alan Cox	:	Split from ip.c , see ip_input.c for history.
 14 *		David S. Miller :	Begin massive cleanup...
 15 *		Andi Kleen	:	Add sysctls.
 16 *		xxxx		:	Overlapfrag bug.
 17 *		Ultima          :       ip_expire() kernel panic.
 18 *		Bill Hawes	:	Frag accounting and evictor fixes.
 19 *		John McDonald	:	0 length frag bug.
 20 *		Alexey Kuznetsov:	SMP races, threading, cleanup.
 21 *		Patrick McHardy :	LRU queue of frag heads for evictor.
 22 */
 23
 24#define pr_fmt(fmt) "IPv4: " fmt
 25
 26#include <linux/compiler.h>
 27#include <linux/module.h>
 28#include <linux/types.h>
 29#include <linux/mm.h>
 30#include <linux/jiffies.h>
 31#include <linux/skbuff.h>
 32#include <linux/list.h>
 33#include <linux/ip.h>
 34#include <linux/icmp.h>
 35#include <linux/netdevice.h>
 36#include <linux/jhash.h>
 37#include <linux/random.h>
 38#include <linux/slab.h>
 39#include <net/route.h>
 40#include <net/dst.h>
 41#include <net/sock.h>
 42#include <net/ip.h>
 43#include <net/icmp.h>
 44#include <net/checksum.h>
 45#include <net/inetpeer.h>
 46#include <net/inet_frag.h>
 47#include <linux/tcp.h>
 48#include <linux/udp.h>
 49#include <linux/inet.h>
 50#include <linux/netfilter_ipv4.h>
 51#include <net/inet_ecn.h>
 52#include <net/l3mdev.h>
 53
 54/* NOTE. Logic of IP defragmentation is parallel to corresponding IPv6
 55 * code now. If you change something here, _PLEASE_ update ipv6/reassembly.c
 56 * as well. Or notify me, at least. --ANK
 57 */
 58static const char ip_frag_cache_name[] = "ip4-frags";
 59
 60/* Describe an entry in the "incomplete datagrams" queue. */
 61struct ipq {
 62	struct inet_frag_queue q;
 63
 64	u8		ecn; /* RFC3168 support */
 65	u16		max_df_size; /* largest frag with DF set seen */
 66	int             iif;
 67	unsigned int    rid;
 68	struct inet_peer *peer;
 69};
 70
 71static u8 ip4_frag_ecn(u8 tos)
 72{
 73	return 1 << (tos & INET_ECN_MASK);
 74}
 75
 76static struct inet_frags ip4_frags;
 77
 78static int ip_frag_reasm(struct ipq *qp, struct sk_buff *skb,
 79			 struct sk_buff *prev_tail, struct net_device *dev);
 80
 81
 82static void ip4_frag_init(struct inet_frag_queue *q, const void *a)
 83{
 84	struct ipq *qp = container_of(q, struct ipq, q);
 
 85	struct net *net = q->fqdir->net;
 86
 87	const struct frag_v4_compare_key *key = a;
 88
 89	q->key.v4 = *key;
 90	qp->ecn = 0;
 91	qp->peer = q->fqdir->max_dist ?
 92		inet_getpeer_v4(net->ipv4.peers, key->saddr, key->vif, 1) :
 93		NULL;
 
 
 
 
 
 94}
 95
 96static void ip4_frag_free(struct inet_frag_queue *q)
 97{
 98	struct ipq *qp;
 99
100	qp = container_of(q, struct ipq, q);
101	if (qp->peer)
102		inet_putpeer(qp->peer);
103}
104
105
106/* Destruction primitives. */
107
108static void ipq_put(struct ipq *ipq)
109{
110	inet_frag_put(&ipq->q);
111}
112
113/* Kill ipq entry. It is not destroyed immediately,
114 * because caller (and someone more) holds reference count.
115 */
116static void ipq_kill(struct ipq *ipq)
117{
118	inet_frag_kill(&ipq->q);
119}
120
121static bool frag_expire_skip_icmp(u32 user)
122{
123	return user == IP_DEFRAG_AF_PACKET ||
124	       ip_defrag_user_in_between(user, IP_DEFRAG_CONNTRACK_IN,
125					 __IP_DEFRAG_CONNTRACK_IN_END) ||
126	       ip_defrag_user_in_between(user, IP_DEFRAG_CONNTRACK_BRIDGE_IN,
127					 __IP_DEFRAG_CONNTRACK_BRIDGE_IN);
128}
129
130/*
131 * Oops, a fragment queue timed out.  Kill it and send an ICMP reply.
132 */
133static void ip_expire(struct timer_list *t)
134{
 
135	struct inet_frag_queue *frag = from_timer(frag, t, timer);
136	const struct iphdr *iph;
137	struct sk_buff *head = NULL;
138	struct net *net;
139	struct ipq *qp;
140	int err;
141
142	qp = container_of(frag, struct ipq, q);
143	net = qp->q.fqdir->net;
144
145	rcu_read_lock();
146
147	/* Paired with WRITE_ONCE() in fqdir_pre_exit(). */
148	if (READ_ONCE(qp->q.fqdir->dead))
149		goto out_rcu_unlock;
150
151	spin_lock(&qp->q.lock);
152
153	if (qp->q.flags & INET_FRAG_COMPLETE)
154		goto out;
155
156	qp->q.flags |= INET_FRAG_DROP;
157	ipq_kill(qp);
158	__IP_INC_STATS(net, IPSTATS_MIB_REASMFAILS);
159	__IP_INC_STATS(net, IPSTATS_MIB_REASMTIMEOUT);
160
161	if (!(qp->q.flags & INET_FRAG_FIRST_IN))
162		goto out;
163
164	/* sk_buff::dev and sk_buff::rbnode are unionized. So we
165	 * pull the head out of the tree in order to be able to
166	 * deal with head->dev.
167	 */
168	head = inet_frag_pull_head(&qp->q);
169	if (!head)
170		goto out;
171	head->dev = dev_get_by_index_rcu(net, qp->iif);
172	if (!head->dev)
173		goto out;
174
175
176	/* skb has no dst, perform route lookup again */
177	iph = ip_hdr(head);
178	err = ip_route_input_noref(head, iph->daddr, iph->saddr,
179					   iph->tos, head->dev);
180	if (err)
181		goto out;
182
183	/* Only an end host needs to send an ICMP
184	 * "Fragment Reassembly Timeout" message, per RFC792.
185	 */
 
186	if (frag_expire_skip_icmp(qp->q.key.v4.user) &&
187	    (skb_rtable(head)->rt_type != RTN_LOCAL))
188		goto out;
189
190	spin_unlock(&qp->q.lock);
191	icmp_send(head, ICMP_TIME_EXCEEDED, ICMP_EXC_FRAGTIME, 0);
192	goto out_rcu_unlock;
193
194out:
195	spin_unlock(&qp->q.lock);
196out_rcu_unlock:
197	rcu_read_unlock();
198	kfree_skb_reason(head, SKB_DROP_REASON_FRAG_REASM_TIMEOUT);
199	ipq_put(qp);
200}
201
202/* Find the correct entry in the "incomplete datagrams" queue for
203 * this IP datagram, and create new one, if nothing is found.
204 */
205static struct ipq *ip_find(struct net *net, struct iphdr *iph,
206			   u32 user, int vif)
207{
208	struct frag_v4_compare_key key = {
209		.saddr = iph->saddr,
210		.daddr = iph->daddr,
211		.user = user,
212		.vif = vif,
213		.id = iph->id,
214		.protocol = iph->protocol,
215	};
216	struct inet_frag_queue *q;
217
218	q = inet_frag_find(net->ipv4.fqdir, &key);
219	if (!q)
220		return NULL;
221
222	return container_of(q, struct ipq, q);
223}
224
225/* Is the fragment too far ahead to be part of ipq? */
226static int ip_frag_too_far(struct ipq *qp)
227{
228	struct inet_peer *peer = qp->peer;
229	unsigned int max = qp->q.fqdir->max_dist;
230	unsigned int start, end;
231
232	int rc;
233
234	if (!peer || !max)
235		return 0;
236
237	start = qp->rid;
238	end = atomic_inc_return(&peer->rid);
239	qp->rid = end;
240
241	rc = qp->q.fragments_tail && (end - start) > max;
242
243	if (rc)
244		__IP_INC_STATS(qp->q.fqdir->net, IPSTATS_MIB_REASMFAILS);
245
246	return rc;
247}
248
249static int ip_frag_reinit(struct ipq *qp)
250{
251	unsigned int sum_truesize = 0;
252
253	if (!mod_timer(&qp->q.timer, jiffies + qp->q.fqdir->timeout)) {
254		refcount_inc(&qp->q.refcnt);
255		return -ETIMEDOUT;
256	}
257
258	sum_truesize = inet_frag_rbtree_purge(&qp->q.rb_fragments,
259					      SKB_DROP_REASON_FRAG_TOO_FAR);
260	sub_frag_mem_limit(qp->q.fqdir, sum_truesize);
261
262	qp->q.flags = 0;
263	qp->q.len = 0;
264	qp->q.meat = 0;
265	qp->q.rb_fragments = RB_ROOT;
266	qp->q.fragments_tail = NULL;
267	qp->q.last_run_head = NULL;
268	qp->iif = 0;
269	qp->ecn = 0;
270
271	return 0;
272}
273
274/* Add new segment to existing queue. */
275static int ip_frag_queue(struct ipq *qp, struct sk_buff *skb)
276{
277	struct net *net = qp->q.fqdir->net;
278	int ihl, end, flags, offset;
279	struct sk_buff *prev_tail;
280	struct net_device *dev;
281	unsigned int fragsize;
282	int err = -ENOENT;
283	SKB_DR(reason);
284	u8 ecn;
285
286	/* If reassembly is already done, @skb must be a duplicate frag. */
287	if (qp->q.flags & INET_FRAG_COMPLETE) {
288		SKB_DR_SET(reason, DUP_FRAG);
289		goto err;
290	}
291
292	if (!(IPCB(skb)->flags & IPSKB_FRAG_COMPLETE) &&
293	    unlikely(ip_frag_too_far(qp)) &&
294	    unlikely(err = ip_frag_reinit(qp))) {
295		ipq_kill(qp);
296		goto err;
297	}
298
299	ecn = ip4_frag_ecn(ip_hdr(skb)->tos);
300	offset = ntohs(ip_hdr(skb)->frag_off);
301	flags = offset & ~IP_OFFSET;
302	offset &= IP_OFFSET;
303	offset <<= 3;		/* offset is in 8-byte chunks */
304	ihl = ip_hdrlen(skb);
305
306	/* Determine the position of this fragment. */
307	end = offset + skb->len - skb_network_offset(skb) - ihl;
308	err = -EINVAL;
309
310	/* Is this the final fragment? */
311	if ((flags & IP_MF) == 0) {
312		/* If we already have some bits beyond end
313		 * or have different end, the segment is corrupted.
314		 */
315		if (end < qp->q.len ||
316		    ((qp->q.flags & INET_FRAG_LAST_IN) && end != qp->q.len))
317			goto discard_qp;
318		qp->q.flags |= INET_FRAG_LAST_IN;
319		qp->q.len = end;
320	} else {
321		if (end&7) {
322			end &= ~7;
323			if (skb->ip_summed != CHECKSUM_UNNECESSARY)
324				skb->ip_summed = CHECKSUM_NONE;
325		}
326		if (end > qp->q.len) {
327			/* Some bits beyond end -> corruption. */
328			if (qp->q.flags & INET_FRAG_LAST_IN)
329				goto discard_qp;
330			qp->q.len = end;
331		}
332	}
333	if (end == offset)
334		goto discard_qp;
335
336	err = -ENOMEM;
337	if (!pskb_pull(skb, skb_network_offset(skb) + ihl))
338		goto discard_qp;
339
340	err = pskb_trim_rcsum(skb, end - offset);
341	if (err)
342		goto discard_qp;
343
344	/* Note : skb->rbnode and skb->dev share the same location. */
345	dev = skb->dev;
346	/* Makes sure compiler wont do silly aliasing games */
347	barrier();
348
349	prev_tail = qp->q.fragments_tail;
350	err = inet_frag_queue_insert(&qp->q, skb, offset, end);
351	if (err)
352		goto insert_error;
353
354	if (dev)
355		qp->iif = dev->ifindex;
356
357	qp->q.stamp = skb->tstamp;
358	qp->q.mono_delivery_time = skb->mono_delivery_time;
359	qp->q.meat += skb->len;
360	qp->ecn |= ecn;
361	add_frag_mem_limit(qp->q.fqdir, skb->truesize);
362	if (offset == 0)
363		qp->q.flags |= INET_FRAG_FIRST_IN;
364
365	fragsize = skb->len + ihl;
366
367	if (fragsize > qp->q.max_size)
368		qp->q.max_size = fragsize;
369
370	if (ip_hdr(skb)->frag_off & htons(IP_DF) &&
371	    fragsize > qp->max_df_size)
372		qp->max_df_size = fragsize;
373
374	if (qp->q.flags == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
375	    qp->q.meat == qp->q.len) {
376		unsigned long orefdst = skb->_skb_refdst;
377
378		skb->_skb_refdst = 0UL;
379		err = ip_frag_reasm(qp, skb, prev_tail, dev);
380		skb->_skb_refdst = orefdst;
381		if (err)
382			inet_frag_kill(&qp->q);
383		return err;
384	}
385
386	skb_dst_drop(skb);
 
387	return -EINPROGRESS;
388
389insert_error:
390	if (err == IPFRAG_DUP) {
391		SKB_DR_SET(reason, DUP_FRAG);
392		err = -EINVAL;
393		goto err;
394	}
395	err = -EINVAL;
396	__IP_INC_STATS(net, IPSTATS_MIB_REASM_OVERLAPS);
397discard_qp:
398	inet_frag_kill(&qp->q);
399	__IP_INC_STATS(net, IPSTATS_MIB_REASMFAILS);
400err:
401	kfree_skb_reason(skb, reason);
402	return err;
403}
404
405static bool ip_frag_coalesce_ok(const struct ipq *qp)
406{
407	return qp->q.key.v4.user == IP_DEFRAG_LOCAL_DELIVER;
408}
409
410/* Build a new IP datagram from all its fragments. */
411static int ip_frag_reasm(struct ipq *qp, struct sk_buff *skb,
412			 struct sk_buff *prev_tail, struct net_device *dev)
413{
414	struct net *net = qp->q.fqdir->net;
415	struct iphdr *iph;
416	void *reasm_data;
417	int len, err;
418	u8 ecn;
419
420	ipq_kill(qp);
421
422	ecn = ip_frag_ecn_table[qp->ecn];
423	if (unlikely(ecn == 0xff)) {
424		err = -EINVAL;
425		goto out_fail;
426	}
427
428	/* Make the one we just received the head. */
429	reasm_data = inet_frag_reasm_prepare(&qp->q, skb, prev_tail);
430	if (!reasm_data)
431		goto out_nomem;
432
433	len = ip_hdrlen(skb) + qp->q.len;
434	err = -E2BIG;
435	if (len > 65535)
436		goto out_oversize;
437
438	inet_frag_reasm_finish(&qp->q, skb, reasm_data,
439			       ip_frag_coalesce_ok(qp));
440
441	skb->dev = dev;
442	IPCB(skb)->frag_max_size = max(qp->max_df_size, qp->q.max_size);
443
444	iph = ip_hdr(skb);
445	iph->tot_len = htons(len);
446	iph->tos |= ecn;
447
448	/* When we set IP_DF on a refragmented skb we must also force a
449	 * call to ip_fragment to avoid forwarding a DF-skb of size s while
450	 * original sender only sent fragments of size f (where f < s).
451	 *
452	 * We only set DF/IPSKB_FRAG_PMTU if such DF fragment was the largest
453	 * frag seen to avoid sending tiny DF-fragments in case skb was built
454	 * from one very small df-fragment and one large non-df frag.
455	 */
456	if (qp->max_df_size == qp->q.max_size) {
457		IPCB(skb)->flags |= IPSKB_FRAG_PMTU;
458		iph->frag_off = htons(IP_DF);
459	} else {
460		iph->frag_off = 0;
461	}
462
463	ip_send_check(iph);
464
465	__IP_INC_STATS(net, IPSTATS_MIB_REASMOKS);
466	qp->q.rb_fragments = RB_ROOT;
467	qp->q.fragments_tail = NULL;
468	qp->q.last_run_head = NULL;
469	return 0;
470
471out_nomem:
472	net_dbg_ratelimited("queue_glue: no memory for gluing queue %p\n", qp);
473	err = -ENOMEM;
474	goto out_fail;
475out_oversize:
476	net_info_ratelimited("Oversized IP packet from %pI4\n", &qp->q.key.v4.saddr);
477out_fail:
478	__IP_INC_STATS(net, IPSTATS_MIB_REASMFAILS);
479	return err;
480}
481
482/* Process an incoming IP datagram fragment. */
483int ip_defrag(struct net *net, struct sk_buff *skb, u32 user)
484{
485	struct net_device *dev = skb->dev ? : skb_dst(skb)->dev;
486	int vif = l3mdev_master_ifindex_rcu(dev);
487	struct ipq *qp;
488
489	__IP_INC_STATS(net, IPSTATS_MIB_REASMREQDS);
490	skb_orphan(skb);
491
492	/* Lookup (or create) queue header */
493	qp = ip_find(net, ip_hdr(skb), user, vif);
494	if (qp) {
495		int ret;
496
497		spin_lock(&qp->q.lock);
498
499		ret = ip_frag_queue(qp, skb);
500
501		spin_unlock(&qp->q.lock);
502		ipq_put(qp);
503		return ret;
504	}
505
506	__IP_INC_STATS(net, IPSTATS_MIB_REASMFAILS);
507	kfree_skb(skb);
508	return -ENOMEM;
509}
510EXPORT_SYMBOL(ip_defrag);
511
512struct sk_buff *ip_check_defrag(struct net *net, struct sk_buff *skb, u32 user)
513{
514	struct iphdr iph;
515	int netoff;
516	u32 len;
517
518	if (skb->protocol != htons(ETH_P_IP))
519		return skb;
520
521	netoff = skb_network_offset(skb);
522
523	if (skb_copy_bits(skb, netoff, &iph, sizeof(iph)) < 0)
524		return skb;
525
526	if (iph.ihl < 5 || iph.version != 4)
527		return skb;
528
529	len = ntohs(iph.tot_len);
530	if (skb->len < netoff + len || len < (iph.ihl * 4))
531		return skb;
532
533	if (ip_is_fragment(&iph)) {
534		skb = skb_share_check(skb, GFP_ATOMIC);
535		if (skb) {
536			if (!pskb_may_pull(skb, netoff + iph.ihl * 4)) {
537				kfree_skb(skb);
538				return NULL;
539			}
540			if (pskb_trim_rcsum(skb, netoff + len)) {
541				kfree_skb(skb);
542				return NULL;
543			}
544			memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
545			if (ip_defrag(net, skb, user))
546				return NULL;
547			skb_clear_hash(skb);
548		}
549	}
550	return skb;
551}
552EXPORT_SYMBOL(ip_check_defrag);
553
554#ifdef CONFIG_SYSCTL
555static int dist_min;
556
557static struct ctl_table ip4_frags_ns_ctl_table[] = {
558	{
559		.procname	= "ipfrag_high_thresh",
560		.maxlen		= sizeof(unsigned long),
561		.mode		= 0644,
562		.proc_handler	= proc_doulongvec_minmax,
563	},
564	{
565		.procname	= "ipfrag_low_thresh",
566		.maxlen		= sizeof(unsigned long),
567		.mode		= 0644,
568		.proc_handler	= proc_doulongvec_minmax,
569	},
570	{
571		.procname	= "ipfrag_time",
572		.maxlen		= sizeof(int),
573		.mode		= 0644,
574		.proc_handler	= proc_dointvec_jiffies,
575	},
576	{
577		.procname	= "ipfrag_max_dist",
578		.maxlen		= sizeof(int),
579		.mode		= 0644,
580		.proc_handler	= proc_dointvec_minmax,
581		.extra1		= &dist_min,
582	},
583	{ }
584};
585
586/* secret interval has been deprecated */
587static int ip4_frags_secret_interval_unused;
588static struct ctl_table ip4_frags_ctl_table[] = {
589	{
590		.procname	= "ipfrag_secret_interval",
591		.data		= &ip4_frags_secret_interval_unused,
592		.maxlen		= sizeof(int),
593		.mode		= 0644,
594		.proc_handler	= proc_dointvec_jiffies,
595	},
596	{ }
597};
598
599static int __net_init ip4_frags_ns_ctl_register(struct net *net)
600{
601	struct ctl_table *table;
602	struct ctl_table_header *hdr;
603
604	table = ip4_frags_ns_ctl_table;
605	if (!net_eq(net, &init_net)) {
606		table = kmemdup(table, sizeof(ip4_frags_ns_ctl_table), GFP_KERNEL);
607		if (!table)
608			goto err_alloc;
609
610	}
611	table[0].data	= &net->ipv4.fqdir->high_thresh;
612	table[0].extra1	= &net->ipv4.fqdir->low_thresh;
613	table[1].data	= &net->ipv4.fqdir->low_thresh;
614	table[1].extra2	= &net->ipv4.fqdir->high_thresh;
615	table[2].data	= &net->ipv4.fqdir->timeout;
616	table[3].data	= &net->ipv4.fqdir->max_dist;
617
618	hdr = register_net_sysctl(net, "net/ipv4", table);
 
619	if (!hdr)
620		goto err_reg;
621
622	net->ipv4.frags_hdr = hdr;
623	return 0;
624
625err_reg:
626	if (!net_eq(net, &init_net))
627		kfree(table);
628err_alloc:
629	return -ENOMEM;
630}
631
632static void __net_exit ip4_frags_ns_ctl_unregister(struct net *net)
633{
634	struct ctl_table *table;
635
636	table = net->ipv4.frags_hdr->ctl_table_arg;
637	unregister_net_sysctl_table(net->ipv4.frags_hdr);
638	kfree(table);
639}
640
641static void __init ip4_frags_ctl_register(void)
642{
643	register_net_sysctl(&init_net, "net/ipv4", ip4_frags_ctl_table);
644}
645#else
646static int ip4_frags_ns_ctl_register(struct net *net)
647{
648	return 0;
649}
650
651static void ip4_frags_ns_ctl_unregister(struct net *net)
652{
653}
654
655static void __init ip4_frags_ctl_register(void)
656{
657}
658#endif
659
660static int __net_init ipv4_frags_init_net(struct net *net)
661{
662	int res;
663
664	res = fqdir_init(&net->ipv4.fqdir, &ip4_frags, net);
665	if (res < 0)
666		return res;
667	/* Fragment cache limits.
668	 *
669	 * The fragment memory accounting code, (tries to) account for
670	 * the real memory usage, by measuring both the size of frag
671	 * queue struct (inet_frag_queue (ipv4:ipq/ipv6:frag_queue))
672	 * and the SKB's truesize.
673	 *
674	 * A 64K fragment consumes 129736 bytes (44*2944)+200
675	 * (1500 truesize == 2944, sizeof(struct ipq) == 200)
676	 *
677	 * We will commit 4MB at one time. Should we cross that limit
678	 * we will prune down to 3MB, making room for approx 8 big 64K
679	 * fragments 8x128k.
680	 */
681	net->ipv4.fqdir->high_thresh = 4 * 1024 * 1024;
682	net->ipv4.fqdir->low_thresh  = 3 * 1024 * 1024;
683	/*
684	 * Important NOTE! Fragment queue must be destroyed before MSL expires.
685	 * RFC791 is wrong proposing to prolongate timer each fragment arrival
686	 * by TTL.
687	 */
688	net->ipv4.fqdir->timeout = IP_FRAG_TIME;
689
690	net->ipv4.fqdir->max_dist = 64;
691
692	res = ip4_frags_ns_ctl_register(net);
693	if (res < 0)
694		fqdir_exit(net->ipv4.fqdir);
695	return res;
696}
697
698static void __net_exit ipv4_frags_pre_exit_net(struct net *net)
699{
700	fqdir_pre_exit(net->ipv4.fqdir);
701}
702
703static void __net_exit ipv4_frags_exit_net(struct net *net)
704{
705	ip4_frags_ns_ctl_unregister(net);
706	fqdir_exit(net->ipv4.fqdir);
707}
708
709static struct pernet_operations ip4_frags_ops = {
710	.init		= ipv4_frags_init_net,
711	.pre_exit	= ipv4_frags_pre_exit_net,
712	.exit		= ipv4_frags_exit_net,
713};
714
715
716static u32 ip4_key_hashfn(const void *data, u32 len, u32 seed)
717{
718	return jhash2(data,
719		      sizeof(struct frag_v4_compare_key) / sizeof(u32), seed);
720}
721
722static u32 ip4_obj_hashfn(const void *data, u32 len, u32 seed)
723{
724	const struct inet_frag_queue *fq = data;
725
726	return jhash2((const u32 *)&fq->key.v4,
727		      sizeof(struct frag_v4_compare_key) / sizeof(u32), seed);
728}
729
730static int ip4_obj_cmpfn(struct rhashtable_compare_arg *arg, const void *ptr)
731{
732	const struct frag_v4_compare_key *key = arg->key;
733	const struct inet_frag_queue *fq = ptr;
734
735	return !!memcmp(&fq->key, key, sizeof(*key));
736}
737
738static const struct rhashtable_params ip4_rhash_params = {
739	.head_offset		= offsetof(struct inet_frag_queue, node),
740	.key_offset		= offsetof(struct inet_frag_queue, key),
741	.key_len		= sizeof(struct frag_v4_compare_key),
742	.hashfn			= ip4_key_hashfn,
743	.obj_hashfn		= ip4_obj_hashfn,
744	.obj_cmpfn		= ip4_obj_cmpfn,
745	.automatic_shrinking	= true,
746};
747
748void __init ipfrag_init(void)
749{
750	ip4_frags.constructor = ip4_frag_init;
751	ip4_frags.destructor = ip4_frag_free;
752	ip4_frags.qsize = sizeof(struct ipq);
753	ip4_frags.frag_expire = ip_expire;
754	ip4_frags.frags_cache_name = ip_frag_cache_name;
755	ip4_frags.rhash_params = ip4_rhash_params;
756	if (inet_frags_init(&ip4_frags))
757		panic("IP: failed to allocate ip4_frags cache\n");
758	ip4_frags_ctl_register();
759	register_pernet_subsys(&ip4_frags_ops);
760}