ip_fragment.c

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