Linux Audio

Check our new training course

Loading...
v5.14.15
  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	if (qp->q.fqdir->dead)
 
148		goto out_rcu_unlock;
149
150	spin_lock(&qp->q.lock);
151
152	if (qp->q.flags & INET_FRAG_COMPLETE)
153		goto out;
154
 
155	ipq_kill(qp);
156	__IP_INC_STATS(net, IPSTATS_MIB_REASMFAILS);
157	__IP_INC_STATS(net, IPSTATS_MIB_REASMTIMEOUT);
158
159	if (!(qp->q.flags & INET_FRAG_FIRST_IN))
160		goto out;
161
162	/* sk_buff::dev and sk_buff::rbnode are unionized. So we
163	 * pull the head out of the tree in order to be able to
164	 * deal with head->dev.
165	 */
166	head = inet_frag_pull_head(&qp->q);
167	if (!head)
168		goto out;
169	head->dev = dev_get_by_index_rcu(net, qp->iif);
170	if (!head->dev)
171		goto out;
172
173
174	/* skb has no dst, perform route lookup again */
175	iph = ip_hdr(head);
176	err = ip_route_input_noref(head, iph->daddr, iph->saddr,
177					   iph->tos, head->dev);
178	if (err)
179		goto out;
180
181	/* Only an end host needs to send an ICMP
182	 * "Fragment Reassembly Timeout" message, per RFC792.
183	 */
184	if (frag_expire_skip_icmp(qp->q.key.v4.user) &&
185	    (skb_rtable(head)->rt_type != RTN_LOCAL))
186		goto out;
187
188	spin_unlock(&qp->q.lock);
189	icmp_send(head, ICMP_TIME_EXCEEDED, ICMP_EXC_FRAGTIME, 0);
190	goto out_rcu_unlock;
191
192out:
193	spin_unlock(&qp->q.lock);
194out_rcu_unlock:
195	rcu_read_unlock();
196	kfree_skb(head);
197	ipq_put(qp);
198}
199
200/* Find the correct entry in the "incomplete datagrams" queue for
201 * this IP datagram, and create new one, if nothing is found.
202 */
203static struct ipq *ip_find(struct net *net, struct iphdr *iph,
204			   u32 user, int vif)
205{
206	struct frag_v4_compare_key key = {
207		.saddr = iph->saddr,
208		.daddr = iph->daddr,
209		.user = user,
210		.vif = vif,
211		.id = iph->id,
212		.protocol = iph->protocol,
213	};
214	struct inet_frag_queue *q;
215
216	q = inet_frag_find(net->ipv4.fqdir, &key);
217	if (!q)
218		return NULL;
219
220	return container_of(q, struct ipq, q);
221}
222
223/* Is the fragment too far ahead to be part of ipq? */
224static int ip_frag_too_far(struct ipq *qp)
225{
226	struct inet_peer *peer = qp->peer;
227	unsigned int max = qp->q.fqdir->max_dist;
228	unsigned int start, end;
229
230	int rc;
231
232	if (!peer || !max)
233		return 0;
234
235	start = qp->rid;
236	end = atomic_inc_return(&peer->rid);
237	qp->rid = end;
238
239	rc = qp->q.fragments_tail && (end - start) > max;
240
241	if (rc)
242		__IP_INC_STATS(qp->q.fqdir->net, IPSTATS_MIB_REASMFAILS);
243
244	return rc;
245}
246
247static int ip_frag_reinit(struct ipq *qp)
248{
249	unsigned int sum_truesize = 0;
250
251	if (!mod_timer(&qp->q.timer, jiffies + qp->q.fqdir->timeout)) {
252		refcount_inc(&qp->q.refcnt);
253		return -ETIMEDOUT;
254	}
255
256	sum_truesize = inet_frag_rbtree_purge(&qp->q.rb_fragments);
 
257	sub_frag_mem_limit(qp->q.fqdir, sum_truesize);
258
259	qp->q.flags = 0;
260	qp->q.len = 0;
261	qp->q.meat = 0;
262	qp->q.rb_fragments = RB_ROOT;
263	qp->q.fragments_tail = NULL;
264	qp->q.last_run_head = NULL;
265	qp->iif = 0;
266	qp->ecn = 0;
267
268	return 0;
269}
270
271/* Add new segment to existing queue. */
272static int ip_frag_queue(struct ipq *qp, struct sk_buff *skb)
273{
274	struct net *net = qp->q.fqdir->net;
275	int ihl, end, flags, offset;
276	struct sk_buff *prev_tail;
277	struct net_device *dev;
278	unsigned int fragsize;
279	int err = -ENOENT;
 
280	u8 ecn;
281
282	if (qp->q.flags & INET_FRAG_COMPLETE)
 
 
283		goto err;
 
284
285	if (!(IPCB(skb)->flags & IPSKB_FRAG_COMPLETE) &&
286	    unlikely(ip_frag_too_far(qp)) &&
287	    unlikely(err = ip_frag_reinit(qp))) {
288		ipq_kill(qp);
289		goto err;
290	}
291
292	ecn = ip4_frag_ecn(ip_hdr(skb)->tos);
293	offset = ntohs(ip_hdr(skb)->frag_off);
294	flags = offset & ~IP_OFFSET;
295	offset &= IP_OFFSET;
296	offset <<= 3;		/* offset is in 8-byte chunks */
297	ihl = ip_hdrlen(skb);
298
299	/* Determine the position of this fragment. */
300	end = offset + skb->len - skb_network_offset(skb) - ihl;
301	err = -EINVAL;
302
303	/* Is this the final fragment? */
304	if ((flags & IP_MF) == 0) {
305		/* If we already have some bits beyond end
306		 * or have different end, the segment is corrupted.
307		 */
308		if (end < qp->q.len ||
309		    ((qp->q.flags & INET_FRAG_LAST_IN) && end != qp->q.len))
310			goto discard_qp;
311		qp->q.flags |= INET_FRAG_LAST_IN;
312		qp->q.len = end;
313	} else {
314		if (end&7) {
315			end &= ~7;
316			if (skb->ip_summed != CHECKSUM_UNNECESSARY)
317				skb->ip_summed = CHECKSUM_NONE;
318		}
319		if (end > qp->q.len) {
320			/* Some bits beyond end -> corruption. */
321			if (qp->q.flags & INET_FRAG_LAST_IN)
322				goto discard_qp;
323			qp->q.len = end;
324		}
325	}
326	if (end == offset)
327		goto discard_qp;
328
329	err = -ENOMEM;
330	if (!pskb_pull(skb, skb_network_offset(skb) + ihl))
331		goto discard_qp;
332
333	err = pskb_trim_rcsum(skb, end - offset);
334	if (err)
335		goto discard_qp;
336
337	/* Note : skb->rbnode and skb->dev share the same location. */
338	dev = skb->dev;
339	/* Makes sure compiler wont do silly aliasing games */
340	barrier();
341
342	prev_tail = qp->q.fragments_tail;
343	err = inet_frag_queue_insert(&qp->q, skb, offset, end);
344	if (err)
345		goto insert_error;
346
347	if (dev)
348		qp->iif = dev->ifindex;
349
350	qp->q.stamp = skb->tstamp;
 
351	qp->q.meat += skb->len;
352	qp->ecn |= ecn;
353	add_frag_mem_limit(qp->q.fqdir, skb->truesize);
354	if (offset == 0)
355		qp->q.flags |= INET_FRAG_FIRST_IN;
356
357	fragsize = skb->len + ihl;
358
359	if (fragsize > qp->q.max_size)
360		qp->q.max_size = fragsize;
361
362	if (ip_hdr(skb)->frag_off & htons(IP_DF) &&
363	    fragsize > qp->max_df_size)
364		qp->max_df_size = fragsize;
365
366	if (qp->q.flags == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
367	    qp->q.meat == qp->q.len) {
368		unsigned long orefdst = skb->_skb_refdst;
369
370		skb->_skb_refdst = 0UL;
371		err = ip_frag_reasm(qp, skb, prev_tail, dev);
372		skb->_skb_refdst = orefdst;
373		if (err)
374			inet_frag_kill(&qp->q);
375		return err;
376	}
377
378	skb_dst_drop(skb);
 
379	return -EINPROGRESS;
380
381insert_error:
382	if (err == IPFRAG_DUP) {
383		kfree_skb(skb);
384		return -EINVAL;
 
385	}
386	err = -EINVAL;
387	__IP_INC_STATS(net, IPSTATS_MIB_REASM_OVERLAPS);
388discard_qp:
389	inet_frag_kill(&qp->q);
390	__IP_INC_STATS(net, IPSTATS_MIB_REASMFAILS);
391err:
392	kfree_skb(skb);
393	return err;
394}
395
396static bool ip_frag_coalesce_ok(const struct ipq *qp)
397{
398	return qp->q.key.v4.user == IP_DEFRAG_LOCAL_DELIVER;
399}
400
401/* Build a new IP datagram from all its fragments. */
402static int ip_frag_reasm(struct ipq *qp, struct sk_buff *skb,
403			 struct sk_buff *prev_tail, struct net_device *dev)
404{
405	struct net *net = qp->q.fqdir->net;
406	struct iphdr *iph;
407	void *reasm_data;
408	int len, err;
409	u8 ecn;
410
411	ipq_kill(qp);
412
413	ecn = ip_frag_ecn_table[qp->ecn];
414	if (unlikely(ecn == 0xff)) {
415		err = -EINVAL;
416		goto out_fail;
417	}
418
419	/* Make the one we just received the head. */
420	reasm_data = inet_frag_reasm_prepare(&qp->q, skb, prev_tail);
421	if (!reasm_data)
422		goto out_nomem;
423
424	len = ip_hdrlen(skb) + qp->q.len;
425	err = -E2BIG;
426	if (len > 65535)
427		goto out_oversize;
428
429	inet_frag_reasm_finish(&qp->q, skb, reasm_data,
430			       ip_frag_coalesce_ok(qp));
431
432	skb->dev = dev;
433	IPCB(skb)->frag_max_size = max(qp->max_df_size, qp->q.max_size);
434
435	iph = ip_hdr(skb);
436	iph->tot_len = htons(len);
437	iph->tos |= ecn;
438
439	/* When we set IP_DF on a refragmented skb we must also force a
440	 * call to ip_fragment to avoid forwarding a DF-skb of size s while
441	 * original sender only sent fragments of size f (where f < s).
442	 *
443	 * We only set DF/IPSKB_FRAG_PMTU if such DF fragment was the largest
444	 * frag seen to avoid sending tiny DF-fragments in case skb was built
445	 * from one very small df-fragment and one large non-df frag.
446	 */
447	if (qp->max_df_size == qp->q.max_size) {
448		IPCB(skb)->flags |= IPSKB_FRAG_PMTU;
449		iph->frag_off = htons(IP_DF);
450	} else {
451		iph->frag_off = 0;
452	}
453
454	ip_send_check(iph);
455
456	__IP_INC_STATS(net, IPSTATS_MIB_REASMOKS);
457	qp->q.rb_fragments = RB_ROOT;
458	qp->q.fragments_tail = NULL;
459	qp->q.last_run_head = NULL;
460	return 0;
461
462out_nomem:
463	net_dbg_ratelimited("queue_glue: no memory for gluing queue %p\n", qp);
464	err = -ENOMEM;
465	goto out_fail;
466out_oversize:
467	net_info_ratelimited("Oversized IP packet from %pI4\n", &qp->q.key.v4.saddr);
468out_fail:
469	__IP_INC_STATS(net, IPSTATS_MIB_REASMFAILS);
470	return err;
471}
472
473/* Process an incoming IP datagram fragment. */
474int ip_defrag(struct net *net, struct sk_buff *skb, u32 user)
475{
476	struct net_device *dev = skb->dev ? : skb_dst(skb)->dev;
477	int vif = l3mdev_master_ifindex_rcu(dev);
478	struct ipq *qp;
479
480	__IP_INC_STATS(net, IPSTATS_MIB_REASMREQDS);
481	skb_orphan(skb);
482
483	/* Lookup (or create) queue header */
484	qp = ip_find(net, ip_hdr(skb), user, vif);
485	if (qp) {
486		int ret;
487
488		spin_lock(&qp->q.lock);
489
490		ret = ip_frag_queue(qp, skb);
491
492		spin_unlock(&qp->q.lock);
493		ipq_put(qp);
494		return ret;
495	}
496
497	__IP_INC_STATS(net, IPSTATS_MIB_REASMFAILS);
498	kfree_skb(skb);
499	return -ENOMEM;
500}
501EXPORT_SYMBOL(ip_defrag);
502
503struct sk_buff *ip_check_defrag(struct net *net, struct sk_buff *skb, u32 user)
504{
505	struct iphdr iph;
506	int netoff;
507	u32 len;
508
509	if (skb->protocol != htons(ETH_P_IP))
510		return skb;
511
512	netoff = skb_network_offset(skb);
513
514	if (skb_copy_bits(skb, netoff, &iph, sizeof(iph)) < 0)
515		return skb;
516
517	if (iph.ihl < 5 || iph.version != 4)
518		return skb;
519
520	len = ntohs(iph.tot_len);
521	if (skb->len < netoff + len || len < (iph.ihl * 4))
522		return skb;
523
524	if (ip_is_fragment(&iph)) {
525		skb = skb_share_check(skb, GFP_ATOMIC);
526		if (skb) {
527			if (!pskb_may_pull(skb, netoff + iph.ihl * 4)) {
528				kfree_skb(skb);
529				return NULL;
530			}
531			if (pskb_trim_rcsum(skb, netoff + len)) {
532				kfree_skb(skb);
533				return NULL;
534			}
535			memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
536			if (ip_defrag(net, skb, user))
537				return NULL;
538			skb_clear_hash(skb);
539		}
540	}
541	return skb;
542}
543EXPORT_SYMBOL(ip_check_defrag);
544
545#ifdef CONFIG_SYSCTL
546static int dist_min;
547
548static struct ctl_table ip4_frags_ns_ctl_table[] = {
549	{
550		.procname	= "ipfrag_high_thresh",
551		.maxlen		= sizeof(unsigned long),
552		.mode		= 0644,
553		.proc_handler	= proc_doulongvec_minmax,
554	},
555	{
556		.procname	= "ipfrag_low_thresh",
557		.maxlen		= sizeof(unsigned long),
558		.mode		= 0644,
559		.proc_handler	= proc_doulongvec_minmax,
560	},
561	{
562		.procname	= "ipfrag_time",
563		.maxlen		= sizeof(int),
564		.mode		= 0644,
565		.proc_handler	= proc_dointvec_jiffies,
566	},
567	{
568		.procname	= "ipfrag_max_dist",
569		.maxlen		= sizeof(int),
570		.mode		= 0644,
571		.proc_handler	= proc_dointvec_minmax,
572		.extra1		= &dist_min,
573	},
574	{ }
575};
576
577/* secret interval has been deprecated */
578static int ip4_frags_secret_interval_unused;
579static struct ctl_table ip4_frags_ctl_table[] = {
580	{
581		.procname	= "ipfrag_secret_interval",
582		.data		= &ip4_frags_secret_interval_unused,
583		.maxlen		= sizeof(int),
584		.mode		= 0644,
585		.proc_handler	= proc_dointvec_jiffies,
586	},
587	{ }
588};
589
590static int __net_init ip4_frags_ns_ctl_register(struct net *net)
591{
592	struct ctl_table *table;
593	struct ctl_table_header *hdr;
594
595	table = ip4_frags_ns_ctl_table;
596	if (!net_eq(net, &init_net)) {
597		table = kmemdup(table, sizeof(ip4_frags_ns_ctl_table), GFP_KERNEL);
598		if (!table)
599			goto err_alloc;
600
601	}
602	table[0].data	= &net->ipv4.fqdir->high_thresh;
603	table[0].extra1	= &net->ipv4.fqdir->low_thresh;
604	table[1].data	= &net->ipv4.fqdir->low_thresh;
605	table[1].extra2	= &net->ipv4.fqdir->high_thresh;
606	table[2].data	= &net->ipv4.fqdir->timeout;
607	table[3].data	= &net->ipv4.fqdir->max_dist;
608
609	hdr = register_net_sysctl(net, "net/ipv4", table);
 
610	if (!hdr)
611		goto err_reg;
612
613	net->ipv4.frags_hdr = hdr;
614	return 0;
615
616err_reg:
617	if (!net_eq(net, &init_net))
618		kfree(table);
619err_alloc:
620	return -ENOMEM;
621}
622
623static void __net_exit ip4_frags_ns_ctl_unregister(struct net *net)
624{
625	struct ctl_table *table;
626
627	table = net->ipv4.frags_hdr->ctl_table_arg;
628	unregister_net_sysctl_table(net->ipv4.frags_hdr);
629	kfree(table);
630}
631
632static void __init ip4_frags_ctl_register(void)
633{
634	register_net_sysctl(&init_net, "net/ipv4", ip4_frags_ctl_table);
635}
636#else
637static int ip4_frags_ns_ctl_register(struct net *net)
638{
639	return 0;
640}
641
642static void ip4_frags_ns_ctl_unregister(struct net *net)
643{
644}
645
646static void __init ip4_frags_ctl_register(void)
647{
648}
649#endif
650
651static int __net_init ipv4_frags_init_net(struct net *net)
652{
653	int res;
654
655	res = fqdir_init(&net->ipv4.fqdir, &ip4_frags, net);
656	if (res < 0)
657		return res;
658	/* Fragment cache limits.
659	 *
660	 * The fragment memory accounting code, (tries to) account for
661	 * the real memory usage, by measuring both the size of frag
662	 * queue struct (inet_frag_queue (ipv4:ipq/ipv6:frag_queue))
663	 * and the SKB's truesize.
664	 *
665	 * A 64K fragment consumes 129736 bytes (44*2944)+200
666	 * (1500 truesize == 2944, sizeof(struct ipq) == 200)
667	 *
668	 * We will commit 4MB at one time. Should we cross that limit
669	 * we will prune down to 3MB, making room for approx 8 big 64K
670	 * fragments 8x128k.
671	 */
672	net->ipv4.fqdir->high_thresh = 4 * 1024 * 1024;
673	net->ipv4.fqdir->low_thresh  = 3 * 1024 * 1024;
674	/*
675	 * Important NOTE! Fragment queue must be destroyed before MSL expires.
676	 * RFC791 is wrong proposing to prolongate timer each fragment arrival
677	 * by TTL.
678	 */
679	net->ipv4.fqdir->timeout = IP_FRAG_TIME;
680
681	net->ipv4.fqdir->max_dist = 64;
682
683	res = ip4_frags_ns_ctl_register(net);
684	if (res < 0)
685		fqdir_exit(net->ipv4.fqdir);
686	return res;
687}
688
689static void __net_exit ipv4_frags_pre_exit_net(struct net *net)
690{
691	fqdir_pre_exit(net->ipv4.fqdir);
692}
693
694static void __net_exit ipv4_frags_exit_net(struct net *net)
695{
696	ip4_frags_ns_ctl_unregister(net);
697	fqdir_exit(net->ipv4.fqdir);
698}
699
700static struct pernet_operations ip4_frags_ops = {
701	.init		= ipv4_frags_init_net,
702	.pre_exit	= ipv4_frags_pre_exit_net,
703	.exit		= ipv4_frags_exit_net,
704};
705
706
707static u32 ip4_key_hashfn(const void *data, u32 len, u32 seed)
708{
709	return jhash2(data,
710		      sizeof(struct frag_v4_compare_key) / sizeof(u32), seed);
711}
712
713static u32 ip4_obj_hashfn(const void *data, u32 len, u32 seed)
714{
715	const struct inet_frag_queue *fq = data;
716
717	return jhash2((const u32 *)&fq->key.v4,
718		      sizeof(struct frag_v4_compare_key) / sizeof(u32), seed);
719}
720
721static int ip4_obj_cmpfn(struct rhashtable_compare_arg *arg, const void *ptr)
722{
723	const struct frag_v4_compare_key *key = arg->key;
724	const struct inet_frag_queue *fq = ptr;
725
726	return !!memcmp(&fq->key, key, sizeof(*key));
727}
728
729static const struct rhashtable_params ip4_rhash_params = {
730	.head_offset		= offsetof(struct inet_frag_queue, node),
731	.key_offset		= offsetof(struct inet_frag_queue, key),
732	.key_len		= sizeof(struct frag_v4_compare_key),
733	.hashfn			= ip4_key_hashfn,
734	.obj_hashfn		= ip4_obj_hashfn,
735	.obj_cmpfn		= ip4_obj_cmpfn,
736	.automatic_shrinking	= true,
737};
738
739void __init ipfrag_init(void)
740{
741	ip4_frags.constructor = ip4_frag_init;
742	ip4_frags.destructor = ip4_frag_free;
743	ip4_frags.qsize = sizeof(struct ipq);
744	ip4_frags.frag_expire = ip_expire;
745	ip4_frags.frags_cache_name = ip_frag_cache_name;
746	ip4_frags.rhash_params = ip4_rhash_params;
747	if (inet_frags_init(&ip4_frags))
748		panic("IP: failed to allocate ip4_frags cache\n");
749	ip4_frags_ctl_register();
750	register_pernet_subsys(&ip4_frags_ops);
751}
v6.9.4
  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	skb_orphan(skb);
388	return -EINPROGRESS;
389
390insert_error:
391	if (err == IPFRAG_DUP) {
392		SKB_DR_SET(reason, DUP_FRAG);
393		err = -EINVAL;
394		goto err;
395	}
396	err = -EINVAL;
397	__IP_INC_STATS(net, IPSTATS_MIB_REASM_OVERLAPS);
398discard_qp:
399	inet_frag_kill(&qp->q);
400	__IP_INC_STATS(net, IPSTATS_MIB_REASMFAILS);
401err:
402	kfree_skb_reason(skb, reason);
403	return err;
404}
405
406static bool ip_frag_coalesce_ok(const struct ipq *qp)
407{
408	return qp->q.key.v4.user == IP_DEFRAG_LOCAL_DELIVER;
409}
410
411/* Build a new IP datagram from all its fragments. */
412static int ip_frag_reasm(struct ipq *qp, struct sk_buff *skb,
413			 struct sk_buff *prev_tail, struct net_device *dev)
414{
415	struct net *net = qp->q.fqdir->net;
416	struct iphdr *iph;
417	void *reasm_data;
418	int len, err;
419	u8 ecn;
420
421	ipq_kill(qp);
422
423	ecn = ip_frag_ecn_table[qp->ecn];
424	if (unlikely(ecn == 0xff)) {
425		err = -EINVAL;
426		goto out_fail;
427	}
428
429	/* Make the one we just received the head. */
430	reasm_data = inet_frag_reasm_prepare(&qp->q, skb, prev_tail);
431	if (!reasm_data)
432		goto out_nomem;
433
434	len = ip_hdrlen(skb) + qp->q.len;
435	err = -E2BIG;
436	if (len > 65535)
437		goto out_oversize;
438
439	inet_frag_reasm_finish(&qp->q, skb, reasm_data,
440			       ip_frag_coalesce_ok(qp));
441
442	skb->dev = dev;
443	IPCB(skb)->frag_max_size = max(qp->max_df_size, qp->q.max_size);
444
445	iph = ip_hdr(skb);
446	iph->tot_len = htons(len);
447	iph->tos |= ecn;
448
449	/* When we set IP_DF on a refragmented skb we must also force a
450	 * call to ip_fragment to avoid forwarding a DF-skb of size s while
451	 * original sender only sent fragments of size f (where f < s).
452	 *
453	 * We only set DF/IPSKB_FRAG_PMTU if such DF fragment was the largest
454	 * frag seen to avoid sending tiny DF-fragments in case skb was built
455	 * from one very small df-fragment and one large non-df frag.
456	 */
457	if (qp->max_df_size == qp->q.max_size) {
458		IPCB(skb)->flags |= IPSKB_FRAG_PMTU;
459		iph->frag_off = htons(IP_DF);
460	} else {
461		iph->frag_off = 0;
462	}
463
464	ip_send_check(iph);
465
466	__IP_INC_STATS(net, IPSTATS_MIB_REASMOKS);
467	qp->q.rb_fragments = RB_ROOT;
468	qp->q.fragments_tail = NULL;
469	qp->q.last_run_head = NULL;
470	return 0;
471
472out_nomem:
473	net_dbg_ratelimited("queue_glue: no memory for gluing queue %p\n", qp);
474	err = -ENOMEM;
475	goto out_fail;
476out_oversize:
477	net_info_ratelimited("Oversized IP packet from %pI4\n", &qp->q.key.v4.saddr);
478out_fail:
479	__IP_INC_STATS(net, IPSTATS_MIB_REASMFAILS);
480	return err;
481}
482
483/* Process an incoming IP datagram fragment. */
484int ip_defrag(struct net *net, struct sk_buff *skb, u32 user)
485{
486	struct net_device *dev = skb->dev ? : skb_dst(skb)->dev;
487	int vif = l3mdev_master_ifindex_rcu(dev);
488	struct ipq *qp;
489
490	__IP_INC_STATS(net, IPSTATS_MIB_REASMREQDS);
 
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_sz(net, "net/ipv4", table,
619				     ARRAY_SIZE(ip4_frags_ns_ctl_table));
620	if (!hdr)
621		goto err_reg;
622
623	net->ipv4.frags_hdr = hdr;
624	return 0;
625
626err_reg:
627	if (!net_eq(net, &init_net))
628		kfree(table);
629err_alloc:
630	return -ENOMEM;
631}
632
633static void __net_exit ip4_frags_ns_ctl_unregister(struct net *net)
634{
635	struct ctl_table *table;
636
637	table = net->ipv4.frags_hdr->ctl_table_arg;
638	unregister_net_sysctl_table(net->ipv4.frags_hdr);
639	kfree(table);
640}
641
642static void __init ip4_frags_ctl_register(void)
643{
644	register_net_sysctl(&init_net, "net/ipv4", ip4_frags_ctl_table);
645}
646#else
647static int ip4_frags_ns_ctl_register(struct net *net)
648{
649	return 0;
650}
651
652static void ip4_frags_ns_ctl_unregister(struct net *net)
653{
654}
655
656static void __init ip4_frags_ctl_register(void)
657{
658}
659#endif
660
661static int __net_init ipv4_frags_init_net(struct net *net)
662{
663	int res;
664
665	res = fqdir_init(&net->ipv4.fqdir, &ip4_frags, net);
666	if (res < 0)
667		return res;
668	/* Fragment cache limits.
669	 *
670	 * The fragment memory accounting code, (tries to) account for
671	 * the real memory usage, by measuring both the size of frag
672	 * queue struct (inet_frag_queue (ipv4:ipq/ipv6:frag_queue))
673	 * and the SKB's truesize.
674	 *
675	 * A 64K fragment consumes 129736 bytes (44*2944)+200
676	 * (1500 truesize == 2944, sizeof(struct ipq) == 200)
677	 *
678	 * We will commit 4MB at one time. Should we cross that limit
679	 * we will prune down to 3MB, making room for approx 8 big 64K
680	 * fragments 8x128k.
681	 */
682	net->ipv4.fqdir->high_thresh = 4 * 1024 * 1024;
683	net->ipv4.fqdir->low_thresh  = 3 * 1024 * 1024;
684	/*
685	 * Important NOTE! Fragment queue must be destroyed before MSL expires.
686	 * RFC791 is wrong proposing to prolongate timer each fragment arrival
687	 * by TTL.
688	 */
689	net->ipv4.fqdir->timeout = IP_FRAG_TIME;
690
691	net->ipv4.fqdir->max_dist = 64;
692
693	res = ip4_frags_ns_ctl_register(net);
694	if (res < 0)
695		fqdir_exit(net->ipv4.fqdir);
696	return res;
697}
698
699static void __net_exit ipv4_frags_pre_exit_net(struct net *net)
700{
701	fqdir_pre_exit(net->ipv4.fqdir);
702}
703
704static void __net_exit ipv4_frags_exit_net(struct net *net)
705{
706	ip4_frags_ns_ctl_unregister(net);
707	fqdir_exit(net->ipv4.fqdir);
708}
709
710static struct pernet_operations ip4_frags_ops = {
711	.init		= ipv4_frags_init_net,
712	.pre_exit	= ipv4_frags_pre_exit_net,
713	.exit		= ipv4_frags_exit_net,
714};
715
716
717static u32 ip4_key_hashfn(const void *data, u32 len, u32 seed)
718{
719	return jhash2(data,
720		      sizeof(struct frag_v4_compare_key) / sizeof(u32), seed);
721}
722
723static u32 ip4_obj_hashfn(const void *data, u32 len, u32 seed)
724{
725	const struct inet_frag_queue *fq = data;
726
727	return jhash2((const u32 *)&fq->key.v4,
728		      sizeof(struct frag_v4_compare_key) / sizeof(u32), seed);
729}
730
731static int ip4_obj_cmpfn(struct rhashtable_compare_arg *arg, const void *ptr)
732{
733	const struct frag_v4_compare_key *key = arg->key;
734	const struct inet_frag_queue *fq = ptr;
735
736	return !!memcmp(&fq->key, key, sizeof(*key));
737}
738
739static const struct rhashtable_params ip4_rhash_params = {
740	.head_offset		= offsetof(struct inet_frag_queue, node),
741	.key_offset		= offsetof(struct inet_frag_queue, key),
742	.key_len		= sizeof(struct frag_v4_compare_key),
743	.hashfn			= ip4_key_hashfn,
744	.obj_hashfn		= ip4_obj_hashfn,
745	.obj_cmpfn		= ip4_obj_cmpfn,
746	.automatic_shrinking	= true,
747};
748
749void __init ipfrag_init(void)
750{
751	ip4_frags.constructor = ip4_frag_init;
752	ip4_frags.destructor = ip4_frag_free;
753	ip4_frags.qsize = sizeof(struct ipq);
754	ip4_frags.frag_expire = ip_expire;
755	ip4_frags.frags_cache_name = ip_frag_cache_name;
756	ip4_frags.rhash_params = ip4_rhash_params;
757	if (inet_frags_init(&ip4_frags))
758		panic("IP: failed to allocate ip4_frags cache\n");
759	ip4_frags_ctl_register();
760	register_pernet_subsys(&ip4_frags_ops);
761}