Linux Audio

Check our new training course

Loading...
v3.15
  1/*
  2 * INET		An implementation of the TCP/IP protocol suite for the LINUX
  3 *		operating system.  INET is implemented using the  BSD Socket
  4 *		interface as the means of communication with the user level.
  5 *
  6 *		The IP fragmentation functionality.
  7 *
  8 * Authors:	Fred N. van Kempen <waltje@uWalt.NL.Mugnet.ORG>
  9 *		Alan Cox <alan@lxorguk.ukuu.org.uk>
 10 *
 11 * Fixes:
 12 *		Alan Cox	:	Split from ip.c , see ip_input.c for history.
 13 *		David S. Miller :	Begin massive cleanup...
 14 *		Andi Kleen	:	Add sysctls.
 15 *		xxxx		:	Overlapfrag bug.
 16 *		Ultima          :       ip_expire() kernel panic.
 17 *		Bill Hawes	:	Frag accounting and evictor fixes.
 18 *		John McDonald	:	0 length frag bug.
 19 *		Alexey Kuznetsov:	SMP races, threading, cleanup.
 20 *		Patrick McHardy :	LRU queue of frag heads for evictor.
 21 */
 22
 23#define pr_fmt(fmt) "IPv4: " fmt
 24
 25#include <linux/compiler.h>
 26#include <linux/module.h>
 27#include <linux/types.h>
 28#include <linux/mm.h>
 29#include <linux/jiffies.h>
 30#include <linux/skbuff.h>
 31#include <linux/list.h>
 32#include <linux/ip.h>
 33#include <linux/icmp.h>
 34#include <linux/netdevice.h>
 35#include <linux/jhash.h>
 36#include <linux/random.h>
 37#include <linux/slab.h>
 38#include <net/route.h>
 39#include <net/dst.h>
 40#include <net/sock.h>
 41#include <net/ip.h>
 42#include <net/icmp.h>
 43#include <net/checksum.h>
 44#include <net/inetpeer.h>
 45#include <net/inet_frag.h>
 46#include <linux/tcp.h>
 47#include <linux/udp.h>
 48#include <linux/inet.h>
 49#include <linux/netfilter_ipv4.h>
 50#include <net/inet_ecn.h>
 
 51
 52/* NOTE. Logic of IP defragmentation is parallel to corresponding IPv6
 53 * code now. If you change something here, _PLEASE_ update ipv6/reassembly.c
 54 * as well. Or notify me, at least. --ANK
 55 */
 56
 57static int sysctl_ipfrag_max_dist __read_mostly = 64;
 58
 59struct ipfrag_skb_cb
 60{
 61	struct inet_skb_parm	h;
 62	int			offset;
 63};
 64
 65#define FRAG_CB(skb)	((struct ipfrag_skb_cb *)((skb)->cb))
 66
 67/* Describe an entry in the "incomplete datagrams" queue. */
 68struct ipq {
 69	struct inet_frag_queue q;
 70
 71	u32		user;
 72	__be32		saddr;
 73	__be32		daddr;
 74	__be16		id;
 75	u8		protocol;
 76	u8		ecn; /* RFC3168 support */
 
 77	int             iif;
 
 78	unsigned int    rid;
 79	struct inet_peer *peer;
 80};
 81
 82static inline u8 ip4_frag_ecn(u8 tos)
 83{
 84	return 1 << (tos & INET_ECN_MASK);
 85}
 86
 87static struct inet_frags ip4_frags;
 88
 89int ip_frag_nqueues(struct net *net)
 90{
 91	return net->ipv4.frags.nqueues;
 92}
 93
 94int ip_frag_mem(struct net *net)
 95{
 96	return sum_frag_mem_limit(&net->ipv4.frags);
 97}
 98
 99static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev,
100			 struct net_device *dev);
101
102struct ip4_create_arg {
103	struct iphdr *iph;
104	u32 user;
 
105};
106
107static unsigned int ipqhashfn(__be16 id, __be32 saddr, __be32 daddr, u8 prot)
108{
109	net_get_random_once(&ip4_frags.rnd, sizeof(ip4_frags.rnd));
110	return jhash_3words((__force u32)id << 16 | prot,
111			    (__force u32)saddr, (__force u32)daddr,
112			    ip4_frags.rnd) & (INETFRAGS_HASHSZ - 1);
113}
114
115static unsigned int ip4_hashfn(struct inet_frag_queue *q)
116{
117	struct ipq *ipq;
118
119	ipq = container_of(q, struct ipq, q);
120	return ipqhashfn(ipq->id, ipq->saddr, ipq->daddr, ipq->protocol);
121}
122
123static bool ip4_frag_match(struct inet_frag_queue *q, void *a)
124{
125	struct ipq *qp;
126	struct ip4_create_arg *arg = a;
127
128	qp = container_of(q, struct ipq, q);
129	return	qp->id == arg->iph->id &&
130		qp->saddr == arg->iph->saddr &&
131		qp->daddr == arg->iph->daddr &&
132		qp->protocol == arg->iph->protocol &&
133		qp->user == arg->user;
 
134}
135
136static void ip4_frag_init(struct inet_frag_queue *q, void *a)
137{
138	struct ipq *qp = container_of(q, struct ipq, q);
139	struct netns_ipv4 *ipv4 = container_of(q->net, struct netns_ipv4,
140					       frags);
141	struct net *net = container_of(ipv4, struct net, ipv4);
142
143	struct ip4_create_arg *arg = a;
144
145	qp->protocol = arg->iph->protocol;
146	qp->id = arg->iph->id;
147	qp->ecn = ip4_frag_ecn(arg->iph->tos);
148	qp->saddr = arg->iph->saddr;
149	qp->daddr = arg->iph->daddr;
 
150	qp->user = arg->user;
151	qp->peer = sysctl_ipfrag_max_dist ?
152		inet_getpeer_v4(net->ipv4.peers, arg->iph->saddr, 1) : NULL;
 
153}
154
155static __inline__ void ip4_frag_free(struct inet_frag_queue *q)
156{
157	struct ipq *qp;
158
159	qp = container_of(q, struct ipq, q);
160	if (qp->peer)
161		inet_putpeer(qp->peer);
162}
163
164
165/* Destruction primitives. */
166
167static __inline__ void ipq_put(struct ipq *ipq)
168{
169	inet_frag_put(&ipq->q, &ip4_frags);
170}
171
172/* Kill ipq entry. It is not destroyed immediately,
173 * because caller (and someone more) holds reference count.
174 */
175static void ipq_kill(struct ipq *ipq)
176{
177	inet_frag_kill(&ipq->q, &ip4_frags);
178}
179
180/* Memory limiting on fragments.  Evictor trashes the oldest
181 * fragment queue until we are back under the threshold.
182 */
183static void ip_evictor(struct net *net)
184{
185	int evicted;
186
187	evicted = inet_frag_evictor(&net->ipv4.frags, &ip4_frags, false);
188	if (evicted)
189		IP_ADD_STATS_BH(net, IPSTATS_MIB_REASMFAILS, evicted);
190}
191
192/*
193 * Oops, a fragment queue timed out.  Kill it and send an ICMP reply.
194 */
195static void ip_expire(unsigned long arg)
196{
197	struct ipq *qp;
198	struct net *net;
199
200	qp = container_of((struct inet_frag_queue *) arg, struct ipq, q);
201	net = container_of(qp->q.net, struct net, ipv4.frags);
202
203	spin_lock(&qp->q.lock);
204
205	if (qp->q.last_in & INET_FRAG_COMPLETE)
206		goto out;
207
208	ipq_kill(qp);
209
210	IP_INC_STATS_BH(net, IPSTATS_MIB_REASMTIMEOUT);
211	IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
212
213	if ((qp->q.last_in & INET_FRAG_FIRST_IN) && qp->q.fragments != NULL) {
214		struct sk_buff *head = qp->q.fragments;
215		const struct iphdr *iph;
216		int err;
217
 
 
 
 
 
218		rcu_read_lock();
219		head->dev = dev_get_by_index_rcu(net, qp->iif);
220		if (!head->dev)
221			goto out_rcu_unlock;
222
223		/* skb has no dst, perform route lookup again */
224		iph = ip_hdr(head);
225		err = ip_route_input_noref(head, iph->daddr, iph->saddr,
226					   iph->tos, head->dev);
227		if (err)
228			goto out_rcu_unlock;
229
230		/*
231		 * Only an end host needs to send an ICMP
232		 * "Fragment Reassembly Timeout" message, per RFC792.
233		 */
234		if (qp->user == IP_DEFRAG_AF_PACKET ||
235		    ((qp->user >= IP_DEFRAG_CONNTRACK_IN) &&
236		     (qp->user <= __IP_DEFRAG_CONNTRACK_IN_END) &&
237		     (skb_rtable(head)->rt_type != RTN_LOCAL)))
238			goto out_rcu_unlock;
239
240
241		/* Send an ICMP "Fragment Reassembly Timeout" message. */
242		icmp_send(head, ICMP_TIME_EXCEEDED, ICMP_EXC_FRAGTIME, 0);
243out_rcu_unlock:
244		rcu_read_unlock();
245	}
246out:
247	spin_unlock(&qp->q.lock);
248	ipq_put(qp);
249}
250
251/* Find the correct entry in the "incomplete datagrams" queue for
252 * this IP datagram, and create new one, if nothing is found.
253 */
254static inline struct ipq *ip_find(struct net *net, struct iphdr *iph, u32 user)
 
255{
256	struct inet_frag_queue *q;
257	struct ip4_create_arg arg;
258	unsigned int hash;
259
260	arg.iph = iph;
261	arg.user = user;
 
262
263	read_lock(&ip4_frags.lock);
264	hash = ipqhashfn(iph->id, iph->saddr, iph->daddr, iph->protocol);
265
266	q = inet_frag_find(&net->ipv4.frags, &ip4_frags, &arg, hash);
267	if (IS_ERR_OR_NULL(q)) {
268		inet_frag_maybe_warn_overflow(q, pr_fmt());
269		return NULL;
270	}
271	return container_of(q, struct ipq, q);
272}
273
274/* Is the fragment too far ahead to be part of ipq? */
275static inline int ip_frag_too_far(struct ipq *qp)
276{
277	struct inet_peer *peer = qp->peer;
278	unsigned int max = sysctl_ipfrag_max_dist;
279	unsigned int start, end;
280
281	int rc;
282
283	if (!peer || !max)
284		return 0;
285
286	start = qp->rid;
287	end = atomic_inc_return(&peer->rid);
288	qp->rid = end;
289
290	rc = qp->q.fragments && (end - start) > max;
291
292	if (rc) {
293		struct net *net;
294
295		net = container_of(qp->q.net, struct net, ipv4.frags);
296		IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
297	}
298
299	return rc;
300}
301
302static int ip_frag_reinit(struct ipq *qp)
303{
304	struct sk_buff *fp;
305	unsigned int sum_truesize = 0;
306
307	if (!mod_timer(&qp->q.timer, jiffies + qp->q.net->timeout)) {
308		atomic_inc(&qp->q.refcnt);
309		return -ETIMEDOUT;
310	}
311
312	fp = qp->q.fragments;
313	do {
314		struct sk_buff *xp = fp->next;
315
316		sum_truesize += fp->truesize;
317		kfree_skb(fp);
318		fp = xp;
319	} while (fp);
320	sub_frag_mem_limit(&qp->q, sum_truesize);
321
322	qp->q.last_in = 0;
323	qp->q.len = 0;
324	qp->q.meat = 0;
325	qp->q.fragments = NULL;
326	qp->q.fragments_tail = NULL;
327	qp->iif = 0;
328	qp->ecn = 0;
329
330	return 0;
331}
332
333/* Add new segment to existing queue. */
334static int ip_frag_queue(struct ipq *qp, struct sk_buff *skb)
335{
336	struct sk_buff *prev, *next;
337	struct net_device *dev;
 
338	int flags, offset;
339	int ihl, end;
340	int err = -ENOENT;
341	u8 ecn;
342
343	if (qp->q.last_in & INET_FRAG_COMPLETE)
344		goto err;
345
346	if (!(IPCB(skb)->flags & IPSKB_FRAG_COMPLETE) &&
347	    unlikely(ip_frag_too_far(qp)) &&
348	    unlikely(err = ip_frag_reinit(qp))) {
349		ipq_kill(qp);
350		goto err;
351	}
352
353	ecn = ip4_frag_ecn(ip_hdr(skb)->tos);
354	offset = ntohs(ip_hdr(skb)->frag_off);
355	flags = offset & ~IP_OFFSET;
356	offset &= IP_OFFSET;
357	offset <<= 3;		/* offset is in 8-byte chunks */
358	ihl = ip_hdrlen(skb);
359
360	/* Determine the position of this fragment. */
361	end = offset + skb->len - ihl;
362	err = -EINVAL;
363
364	/* Is this the final fragment? */
365	if ((flags & IP_MF) == 0) {
366		/* If we already have some bits beyond end
367		 * or have different end, the segment is corrupted.
368		 */
369		if (end < qp->q.len ||
370		    ((qp->q.last_in & INET_FRAG_LAST_IN) && end != qp->q.len))
371			goto err;
372		qp->q.last_in |= INET_FRAG_LAST_IN;
373		qp->q.len = end;
374	} else {
375		if (end&7) {
376			end &= ~7;
377			if (skb->ip_summed != CHECKSUM_UNNECESSARY)
378				skb->ip_summed = CHECKSUM_NONE;
379		}
380		if (end > qp->q.len) {
381			/* Some bits beyond end -> corruption. */
382			if (qp->q.last_in & INET_FRAG_LAST_IN)
383				goto err;
384			qp->q.len = end;
385		}
386	}
387	if (end == offset)
388		goto err;
389
390	err = -ENOMEM;
391	if (pskb_pull(skb, ihl) == NULL)
392		goto err;
393
394	err = pskb_trim_rcsum(skb, end - offset);
395	if (err)
396		goto err;
397
398	/* Find out which fragments are in front and at the back of us
399	 * in the chain of fragments so far.  We must know where to put
400	 * this fragment, right?
401	 */
402	prev = qp->q.fragments_tail;
403	if (!prev || FRAG_CB(prev)->offset < offset) {
404		next = NULL;
405		goto found;
406	}
407	prev = NULL;
408	for (next = qp->q.fragments; next != NULL; next = next->next) {
409		if (FRAG_CB(next)->offset >= offset)
410			break;	/* bingo! */
411		prev = next;
412	}
413
414found:
415	/* We found where to put this one.  Check for overlap with
416	 * preceding fragment, and, if needed, align things so that
417	 * any overlaps are eliminated.
418	 */
419	if (prev) {
420		int i = (FRAG_CB(prev)->offset + prev->len) - offset;
421
422		if (i > 0) {
423			offset += i;
424			err = -EINVAL;
425			if (end <= offset)
426				goto err;
427			err = -ENOMEM;
428			if (!pskb_pull(skb, i))
429				goto err;
430			if (skb->ip_summed != CHECKSUM_UNNECESSARY)
431				skb->ip_summed = CHECKSUM_NONE;
432		}
433	}
434
435	err = -ENOMEM;
436
437	while (next && FRAG_CB(next)->offset < end) {
438		int i = end - FRAG_CB(next)->offset; /* overlap is 'i' bytes */
439
440		if (i < next->len) {
441			/* Eat head of the next overlapped fragment
442			 * and leave the loop. The next ones cannot overlap.
443			 */
444			if (!pskb_pull(next, i))
445				goto err;
446			FRAG_CB(next)->offset += i;
447			qp->q.meat -= i;
448			if (next->ip_summed != CHECKSUM_UNNECESSARY)
449				next->ip_summed = CHECKSUM_NONE;
450			break;
451		} else {
452			struct sk_buff *free_it = next;
453
454			/* Old fragment is completely overridden with
455			 * new one drop it.
456			 */
457			next = next->next;
458
459			if (prev)
460				prev->next = next;
461			else
462				qp->q.fragments = next;
463
464			qp->q.meat -= free_it->len;
465			sub_frag_mem_limit(&qp->q, free_it->truesize);
466			kfree_skb(free_it);
467		}
468	}
469
470	FRAG_CB(skb)->offset = offset;
471
472	/* Insert this fragment in the chain of fragments. */
473	skb->next = next;
474	if (!next)
475		qp->q.fragments_tail = skb;
476	if (prev)
477		prev->next = skb;
478	else
479		qp->q.fragments = skb;
480
481	dev = skb->dev;
482	if (dev) {
483		qp->iif = dev->ifindex;
484		skb->dev = NULL;
485	}
486	qp->q.stamp = skb->tstamp;
487	qp->q.meat += skb->len;
488	qp->ecn |= ecn;
489	add_frag_mem_limit(&qp->q, skb->truesize);
490	if (offset == 0)
491		qp->q.last_in |= INET_FRAG_FIRST_IN;
 
 
 
 
 
492
493	if (ip_hdr(skb)->frag_off & htons(IP_DF) &&
494	    skb->len + ihl > qp->q.max_size)
495		qp->q.max_size = skb->len + ihl;
496
497	if (qp->q.last_in == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
498	    qp->q.meat == qp->q.len) {
499		unsigned long orefdst = skb->_skb_refdst;
500
501		skb->_skb_refdst = 0UL;
502		err = ip_frag_reasm(qp, prev, dev);
503		skb->_skb_refdst = orefdst;
504		return err;
505	}
506
507	skb_dst_drop(skb);
508	inet_frag_lru_move(&qp->q);
509	return -EINPROGRESS;
510
511err:
512	kfree_skb(skb);
513	return err;
514}
515
516
517/* Build a new IP datagram from all its fragments. */
518
519static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev,
520			 struct net_device *dev)
521{
522	struct net *net = container_of(qp->q.net, struct net, ipv4.frags);
523	struct iphdr *iph;
524	struct sk_buff *fp, *head = qp->q.fragments;
525	int len;
526	int ihlen;
527	int err;
528	int sum_truesize;
529	u8 ecn;
530
531	ipq_kill(qp);
532
533	ecn = ip_frag_ecn_table[qp->ecn];
534	if (unlikely(ecn == 0xff)) {
535		err = -EINVAL;
536		goto out_fail;
537	}
538	/* Make the one we just received the head. */
539	if (prev) {
540		head = prev->next;
541		fp = skb_clone(head, GFP_ATOMIC);
542		if (!fp)
543			goto out_nomem;
544
545		fp->next = head->next;
546		if (!fp->next)
547			qp->q.fragments_tail = fp;
548		prev->next = fp;
549
550		skb_morph(head, qp->q.fragments);
551		head->next = qp->q.fragments->next;
552
553		consume_skb(qp->q.fragments);
554		qp->q.fragments = head;
555	}
556
557	WARN_ON(head == NULL);
558	WARN_ON(FRAG_CB(head)->offset != 0);
559
560	/* Allocate a new buffer for the datagram. */
561	ihlen = ip_hdrlen(head);
562	len = ihlen + qp->q.len;
563
564	err = -E2BIG;
565	if (len > 65535)
566		goto out_oversize;
567
568	/* Head of list must not be cloned. */
569	if (skb_unclone(head, GFP_ATOMIC))
570		goto out_nomem;
571
572	/* If the first fragment is fragmented itself, we split
573	 * it to two chunks: the first with data and paged part
574	 * and the second, holding only fragments. */
575	if (skb_has_frag_list(head)) {
576		struct sk_buff *clone;
577		int i, plen = 0;
578
579		if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL)
 
580			goto out_nomem;
581		clone->next = head->next;
582		head->next = clone;
583		skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
584		skb_frag_list_init(head);
585		for (i = 0; i < skb_shinfo(head)->nr_frags; i++)
586			plen += skb_frag_size(&skb_shinfo(head)->frags[i]);
587		clone->len = clone->data_len = head->data_len - plen;
588		head->data_len -= clone->len;
589		head->len -= clone->len;
590		clone->csum = 0;
591		clone->ip_summed = head->ip_summed;
592		add_frag_mem_limit(&qp->q, clone->truesize);
593	}
594
 
595	skb_push(head, head->data - skb_network_header(head));
596
597	sum_truesize = head->truesize;
598	for (fp = head->next; fp;) {
599		bool headstolen;
600		int delta;
601		struct sk_buff *next = fp->next;
602
603		sum_truesize += fp->truesize;
604		if (head->ip_summed != fp->ip_summed)
605			head->ip_summed = CHECKSUM_NONE;
606		else if (head->ip_summed == CHECKSUM_COMPLETE)
607			head->csum = csum_add(head->csum, fp->csum);
608
609		if (skb_try_coalesce(head, fp, &headstolen, &delta)) {
610			kfree_skb_partial(fp, headstolen);
611		} else {
612			if (!skb_shinfo(head)->frag_list)
613				skb_shinfo(head)->frag_list = fp;
614			head->data_len += fp->len;
615			head->len += fp->len;
616			head->truesize += fp->truesize;
617		}
618		fp = next;
619	}
620	sub_frag_mem_limit(&qp->q, sum_truesize);
621
622	head->next = NULL;
623	head->dev = dev;
624	head->tstamp = qp->q.stamp;
625	IPCB(head)->frag_max_size = qp->q.max_size;
626
627	iph = ip_hdr(head);
628	/* max_size != 0 implies at least one fragment had IP_DF set */
629	iph->frag_off = qp->q.max_size ? htons(IP_DF) : 0;
630	iph->tot_len = htons(len);
631	iph->tos |= ecn;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
632	IP_INC_STATS_BH(net, IPSTATS_MIB_REASMOKS);
633	qp->q.fragments = NULL;
634	qp->q.fragments_tail = NULL;
635	return 0;
636
637out_nomem:
638	LIMIT_NETDEBUG(KERN_ERR pr_fmt("queue_glue: no memory for gluing queue %p\n"),
639		       qp);
640	err = -ENOMEM;
641	goto out_fail;
642out_oversize:
643	net_info_ratelimited("Oversized IP packet from %pI4\n", &qp->saddr);
644out_fail:
645	IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
646	return err;
647}
648
649/* Process an incoming IP datagram fragment. */
650int ip_defrag(struct sk_buff *skb, u32 user)
651{
 
 
652	struct ipq *qp;
653	struct net *net;
654
655	net = skb->dev ? dev_net(skb->dev) : dev_net(skb_dst(skb)->dev);
656	IP_INC_STATS_BH(net, IPSTATS_MIB_REASMREQDS);
657
658	/* Start by cleaning up the memory. */
659	ip_evictor(net);
660
661	/* Lookup (or create) queue header */
662	if ((qp = ip_find(net, ip_hdr(skb), user)) != NULL) {
 
663		int ret;
664
665		spin_lock(&qp->q.lock);
666
667		ret = ip_frag_queue(qp, skb);
668
669		spin_unlock(&qp->q.lock);
670		ipq_put(qp);
671		return ret;
672	}
673
674	IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
675	kfree_skb(skb);
676	return -ENOMEM;
677}
678EXPORT_SYMBOL(ip_defrag);
679
680struct sk_buff *ip_check_defrag(struct sk_buff *skb, u32 user)
681{
682	struct iphdr iph;
 
683	u32 len;
684
685	if (skb->protocol != htons(ETH_P_IP))
686		return skb;
687
688	if (!skb_copy_bits(skb, 0, &iph, sizeof(iph)))
 
 
689		return skb;
690
691	if (iph.ihl < 5 || iph.version != 4)
692		return skb;
693
694	len = ntohs(iph.tot_len);
695	if (skb->len < len || len < (iph.ihl * 4))
696		return skb;
697
698	if (ip_is_fragment(&iph)) {
699		skb = skb_share_check(skb, GFP_ATOMIC);
700		if (skb) {
701			if (!pskb_may_pull(skb, iph.ihl*4))
702				return skb;
703			if (pskb_trim_rcsum(skb, len))
704				return skb;
705			memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
706			if (ip_defrag(skb, user))
707				return NULL;
708			skb_clear_hash(skb);
709		}
710	}
711	return skb;
712}
713EXPORT_SYMBOL(ip_check_defrag);
714
715#ifdef CONFIG_SYSCTL
716static int zero;
717
718static struct ctl_table ip4_frags_ns_ctl_table[] = {
719	{
720		.procname	= "ipfrag_high_thresh",
721		.data		= &init_net.ipv4.frags.high_thresh,
722		.maxlen		= sizeof(int),
723		.mode		= 0644,
724		.proc_handler	= proc_dointvec
 
725	},
726	{
727		.procname	= "ipfrag_low_thresh",
728		.data		= &init_net.ipv4.frags.low_thresh,
729		.maxlen		= sizeof(int),
730		.mode		= 0644,
731		.proc_handler	= proc_dointvec
 
 
732	},
733	{
734		.procname	= "ipfrag_time",
735		.data		= &init_net.ipv4.frags.timeout,
736		.maxlen		= sizeof(int),
737		.mode		= 0644,
738		.proc_handler	= proc_dointvec_jiffies,
739	},
 
 
 
 
 
 
 
 
740	{ }
741};
742
 
 
743static struct ctl_table ip4_frags_ctl_table[] = {
744	{
745		.procname	= "ipfrag_secret_interval",
746		.data		= &ip4_frags.secret_interval,
747		.maxlen		= sizeof(int),
748		.mode		= 0644,
749		.proc_handler	= proc_dointvec_jiffies,
750	},
751	{
752		.procname	= "ipfrag_max_dist",
753		.data		= &sysctl_ipfrag_max_dist,
754		.maxlen		= sizeof(int),
755		.mode		= 0644,
756		.proc_handler	= proc_dointvec_minmax,
757		.extra1		= &zero
758	},
759	{ }
760};
761
762static int __net_init ip4_frags_ns_ctl_register(struct net *net)
763{
764	struct ctl_table *table;
765	struct ctl_table_header *hdr;
766
767	table = ip4_frags_ns_ctl_table;
768	if (!net_eq(net, &init_net)) {
769		table = kmemdup(table, sizeof(ip4_frags_ns_ctl_table), GFP_KERNEL);
770		if (table == NULL)
771			goto err_alloc;
772
773		table[0].data = &net->ipv4.frags.high_thresh;
 
 
774		table[1].data = &net->ipv4.frags.low_thresh;
 
775		table[2].data = &net->ipv4.frags.timeout;
776
777		/* Don't export sysctls to unprivileged users */
778		if (net->user_ns != &init_user_ns)
779			table[0].procname = NULL;
780	}
781
782	hdr = register_net_sysctl(net, "net/ipv4", table);
783	if (hdr == NULL)
784		goto err_reg;
785
786	net->ipv4.frags_hdr = hdr;
787	return 0;
788
789err_reg:
790	if (!net_eq(net, &init_net))
791		kfree(table);
792err_alloc:
793	return -ENOMEM;
794}
795
796static void __net_exit ip4_frags_ns_ctl_unregister(struct net *net)
797{
798	struct ctl_table *table;
799
800	table = net->ipv4.frags_hdr->ctl_table_arg;
801	unregister_net_sysctl_table(net->ipv4.frags_hdr);
802	kfree(table);
803}
804
805static void ip4_frags_ctl_register(void)
806{
807	register_net_sysctl(&init_net, "net/ipv4", ip4_frags_ctl_table);
808}
809#else
810static inline int ip4_frags_ns_ctl_register(struct net *net)
811{
812	return 0;
813}
814
815static inline void ip4_frags_ns_ctl_unregister(struct net *net)
816{
817}
818
819static inline void ip4_frags_ctl_register(void)
820{
821}
822#endif
823
824static int __net_init ipv4_frags_init_net(struct net *net)
825{
 
 
826	/* Fragment cache limits.
827	 *
828	 * The fragment memory accounting code, (tries to) account for
829	 * the real memory usage, by measuring both the size of frag
830	 * queue struct (inet_frag_queue (ipv4:ipq/ipv6:frag_queue))
831	 * and the SKB's truesize.
832	 *
833	 * A 64K fragment consumes 129736 bytes (44*2944)+200
834	 * (1500 truesize == 2944, sizeof(struct ipq) == 200)
835	 *
836	 * We will commit 4MB at one time. Should we cross that limit
837	 * we will prune down to 3MB, making room for approx 8 big 64K
838	 * fragments 8x128k.
839	 */
840	net->ipv4.frags.high_thresh = 4 * 1024 * 1024;
841	net->ipv4.frags.low_thresh  = 3 * 1024 * 1024;
842	/*
843	 * Important NOTE! Fragment queue must be destroyed before MSL expires.
844	 * RFC791 is wrong proposing to prolongate timer each fragment arrival
845	 * by TTL.
846	 */
847	net->ipv4.frags.timeout = IP_FRAG_TIME;
848
849	inet_frags_init_net(&net->ipv4.frags);
850
851	return ip4_frags_ns_ctl_register(net);
 
 
 
 
 
 
852}
853
854static void __net_exit ipv4_frags_exit_net(struct net *net)
855{
856	ip4_frags_ns_ctl_unregister(net);
857	inet_frags_exit_net(&net->ipv4.frags, &ip4_frags);
858}
859
860static struct pernet_operations ip4_frags_ops = {
861	.init = ipv4_frags_init_net,
862	.exit = ipv4_frags_exit_net,
863};
864
865void __init ipfrag_init(void)
866{
867	ip4_frags_ctl_register();
868	register_pernet_subsys(&ip4_frags_ops);
869	ip4_frags.hashfn = ip4_hashfn;
870	ip4_frags.constructor = ip4_frag_init;
871	ip4_frags.destructor = ip4_frag_free;
872	ip4_frags.skb_free = NULL;
873	ip4_frags.qsize = sizeof(struct ipq);
874	ip4_frags.match = ip4_frag_match;
875	ip4_frags.frag_expire = ip_expire;
876	ip4_frags.secret_interval = 10 * 60 * HZ;
877	inet_frags_init(&ip4_frags);
 
878}
v4.6
  1/*
  2 * INET		An implementation of the TCP/IP protocol suite for the LINUX
  3 *		operating system.  INET is implemented using the  BSD Socket
  4 *		interface as the means of communication with the user level.
  5 *
  6 *		The IP fragmentation functionality.
  7 *
  8 * Authors:	Fred N. van Kempen <waltje@uWalt.NL.Mugnet.ORG>
  9 *		Alan Cox <alan@lxorguk.ukuu.org.uk>
 10 *
 11 * Fixes:
 12 *		Alan Cox	:	Split from ip.c , see ip_input.c for history.
 13 *		David S. Miller :	Begin massive cleanup...
 14 *		Andi Kleen	:	Add sysctls.
 15 *		xxxx		:	Overlapfrag bug.
 16 *		Ultima          :       ip_expire() kernel panic.
 17 *		Bill Hawes	:	Frag accounting and evictor fixes.
 18 *		John McDonald	:	0 length frag bug.
 19 *		Alexey Kuznetsov:	SMP races, threading, cleanup.
 20 *		Patrick McHardy :	LRU queue of frag heads for evictor.
 21 */
 22
 23#define pr_fmt(fmt) "IPv4: " fmt
 24
 25#include <linux/compiler.h>
 26#include <linux/module.h>
 27#include <linux/types.h>
 28#include <linux/mm.h>
 29#include <linux/jiffies.h>
 30#include <linux/skbuff.h>
 31#include <linux/list.h>
 32#include <linux/ip.h>
 33#include <linux/icmp.h>
 34#include <linux/netdevice.h>
 35#include <linux/jhash.h>
 36#include <linux/random.h>
 37#include <linux/slab.h>
 38#include <net/route.h>
 39#include <net/dst.h>
 40#include <net/sock.h>
 41#include <net/ip.h>
 42#include <net/icmp.h>
 43#include <net/checksum.h>
 44#include <net/inetpeer.h>
 45#include <net/inet_frag.h>
 46#include <linux/tcp.h>
 47#include <linux/udp.h>
 48#include <linux/inet.h>
 49#include <linux/netfilter_ipv4.h>
 50#include <net/inet_ecn.h>
 51#include <net/l3mdev.h>
 52
 53/* NOTE. Logic of IP defragmentation is parallel to corresponding IPv6
 54 * code now. If you change something here, _PLEASE_ update ipv6/reassembly.c
 55 * as well. Or notify me, at least. --ANK
 56 */
 57static const char ip_frag_cache_name[] = "ip4-frags";
 
 58
 59struct ipfrag_skb_cb
 60{
 61	struct inet_skb_parm	h;
 62	int			offset;
 63};
 64
 65#define FRAG_CB(skb)	((struct ipfrag_skb_cb *)((skb)->cb))
 66
 67/* Describe an entry in the "incomplete datagrams" queue. */
 68struct ipq {
 69	struct inet_frag_queue q;
 70
 71	u32		user;
 72	__be32		saddr;
 73	__be32		daddr;
 74	__be16		id;
 75	u8		protocol;
 76	u8		ecn; /* RFC3168 support */
 77	u16		max_df_size; /* largest frag with DF set seen */
 78	int             iif;
 79	int             vif;   /* L3 master device index */
 80	unsigned int    rid;
 81	struct inet_peer *peer;
 82};
 83
 84static u8 ip4_frag_ecn(u8 tos)
 85{
 86	return 1 << (tos & INET_ECN_MASK);
 87}
 88
 89static struct inet_frags ip4_frags;
 90
 
 
 
 
 
 91int ip_frag_mem(struct net *net)
 92{
 93	return sum_frag_mem_limit(&net->ipv4.frags);
 94}
 95
 96static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev,
 97			 struct net_device *dev);
 98
 99struct ip4_create_arg {
100	struct iphdr *iph;
101	u32 user;
102	int vif;
103};
104
105static unsigned int ipqhashfn(__be16 id, __be32 saddr, __be32 daddr, u8 prot)
106{
107	net_get_random_once(&ip4_frags.rnd, sizeof(ip4_frags.rnd));
108	return jhash_3words((__force u32)id << 16 | prot,
109			    (__force u32)saddr, (__force u32)daddr,
110			    ip4_frags.rnd);
111}
112
113static unsigned int ip4_hashfn(const struct inet_frag_queue *q)
114{
115	const struct ipq *ipq;
116
117	ipq = container_of(q, struct ipq, q);
118	return ipqhashfn(ipq->id, ipq->saddr, ipq->daddr, ipq->protocol);
119}
120
121static bool ip4_frag_match(const struct inet_frag_queue *q, const void *a)
122{
123	const struct ipq *qp;
124	const struct ip4_create_arg *arg = a;
125
126	qp = container_of(q, struct ipq, q);
127	return	qp->id == arg->iph->id &&
128		qp->saddr == arg->iph->saddr &&
129		qp->daddr == arg->iph->daddr &&
130		qp->protocol == arg->iph->protocol &&
131		qp->user == arg->user &&
132		qp->vif == arg->vif;
133}
134
135static void ip4_frag_init(struct inet_frag_queue *q, const void *a)
136{
137	struct ipq *qp = container_of(q, struct ipq, q);
138	struct netns_ipv4 *ipv4 = container_of(q->net, struct netns_ipv4,
139					       frags);
140	struct net *net = container_of(ipv4, struct net, ipv4);
141
142	const struct ip4_create_arg *arg = a;
143
144	qp->protocol = arg->iph->protocol;
145	qp->id = arg->iph->id;
146	qp->ecn = ip4_frag_ecn(arg->iph->tos);
147	qp->saddr = arg->iph->saddr;
148	qp->daddr = arg->iph->daddr;
149	qp->vif = arg->vif;
150	qp->user = arg->user;
151	qp->peer = q->net->max_dist ?
152		inet_getpeer_v4(net->ipv4.peers, arg->iph->saddr, arg->vif, 1) :
153		NULL;
154}
155
156static void ip4_frag_free(struct inet_frag_queue *q)
157{
158	struct ipq *qp;
159
160	qp = container_of(q, struct ipq, q);
161	if (qp->peer)
162		inet_putpeer(qp->peer);
163}
164
165
166/* Destruction primitives. */
167
168static void ipq_put(struct ipq *ipq)
169{
170	inet_frag_put(&ipq->q, &ip4_frags);
171}
172
173/* Kill ipq entry. It is not destroyed immediately,
174 * because caller (and someone more) holds reference count.
175 */
176static void ipq_kill(struct ipq *ipq)
177{
178	inet_frag_kill(&ipq->q, &ip4_frags);
179}
180
181static bool frag_expire_skip_icmp(u32 user)
 
 
 
182{
183	return user == IP_DEFRAG_AF_PACKET ||
184	       ip_defrag_user_in_between(user, IP_DEFRAG_CONNTRACK_IN,
185					 __IP_DEFRAG_CONNTRACK_IN_END) ||
186	       ip_defrag_user_in_between(user, IP_DEFRAG_CONNTRACK_BRIDGE_IN,
187					 __IP_DEFRAG_CONNTRACK_BRIDGE_IN);
188}
189
190/*
191 * Oops, a fragment queue timed out.  Kill it and send an ICMP reply.
192 */
193static void ip_expire(unsigned long arg)
194{
195	struct ipq *qp;
196	struct net *net;
197
198	qp = container_of((struct inet_frag_queue *) arg, struct ipq, q);
199	net = container_of(qp->q.net, struct net, ipv4.frags);
200
201	spin_lock(&qp->q.lock);
202
203	if (qp->q.flags & INET_FRAG_COMPLETE)
204		goto out;
205
206	ipq_kill(qp);
 
 
207	IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
208
209	if (!inet_frag_evicting(&qp->q)) {
210		struct sk_buff *head = qp->q.fragments;
211		const struct iphdr *iph;
212		int err;
213
214		IP_INC_STATS_BH(net, IPSTATS_MIB_REASMTIMEOUT);
215
216		if (!(qp->q.flags & INET_FRAG_FIRST_IN) || !qp->q.fragments)
217			goto out;
218
219		rcu_read_lock();
220		head->dev = dev_get_by_index_rcu(net, qp->iif);
221		if (!head->dev)
222			goto out_rcu_unlock;
223
224		/* skb has no dst, perform route lookup again */
225		iph = ip_hdr(head);
226		err = ip_route_input_noref(head, iph->daddr, iph->saddr,
227					   iph->tos, head->dev);
228		if (err)
229			goto out_rcu_unlock;
230
231		/* Only an end host needs to send an ICMP
 
232		 * "Fragment Reassembly Timeout" message, per RFC792.
233		 */
234		if (frag_expire_skip_icmp(qp->user) &&
235		    (skb_rtable(head)->rt_type != RTN_LOCAL))
 
 
236			goto out_rcu_unlock;
237
 
238		/* Send an ICMP "Fragment Reassembly Timeout" message. */
239		icmp_send(head, ICMP_TIME_EXCEEDED, ICMP_EXC_FRAGTIME, 0);
240out_rcu_unlock:
241		rcu_read_unlock();
242	}
243out:
244	spin_unlock(&qp->q.lock);
245	ipq_put(qp);
246}
247
248/* Find the correct entry in the "incomplete datagrams" queue for
249 * this IP datagram, and create new one, if nothing is found.
250 */
251static struct ipq *ip_find(struct net *net, struct iphdr *iph,
252			   u32 user, int vif)
253{
254	struct inet_frag_queue *q;
255	struct ip4_create_arg arg;
256	unsigned int hash;
257
258	arg.iph = iph;
259	arg.user = user;
260	arg.vif = vif;
261
 
262	hash = ipqhashfn(iph->id, iph->saddr, iph->daddr, iph->protocol);
263
264	q = inet_frag_find(&net->ipv4.frags, &ip4_frags, &arg, hash);
265	if (IS_ERR_OR_NULL(q)) {
266		inet_frag_maybe_warn_overflow(q, pr_fmt());
267		return NULL;
268	}
269	return container_of(q, struct ipq, q);
270}
271
272/* Is the fragment too far ahead to be part of ipq? */
273static int ip_frag_too_far(struct ipq *qp)
274{
275	struct inet_peer *peer = qp->peer;
276	unsigned int max = qp->q.net->max_dist;
277	unsigned int start, end;
278
279	int rc;
280
281	if (!peer || !max)
282		return 0;
283
284	start = qp->rid;
285	end = atomic_inc_return(&peer->rid);
286	qp->rid = end;
287
288	rc = qp->q.fragments && (end - start) > max;
289
290	if (rc) {
291		struct net *net;
292
293		net = container_of(qp->q.net, struct net, ipv4.frags);
294		IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
295	}
296
297	return rc;
298}
299
300static int ip_frag_reinit(struct ipq *qp)
301{
302	struct sk_buff *fp;
303	unsigned int sum_truesize = 0;
304
305	if (!mod_timer(&qp->q.timer, jiffies + qp->q.net->timeout)) {
306		atomic_inc(&qp->q.refcnt);
307		return -ETIMEDOUT;
308	}
309
310	fp = qp->q.fragments;
311	do {
312		struct sk_buff *xp = fp->next;
313
314		sum_truesize += fp->truesize;
315		kfree_skb(fp);
316		fp = xp;
317	} while (fp);
318	sub_frag_mem_limit(qp->q.net, sum_truesize);
319
320	qp->q.flags = 0;
321	qp->q.len = 0;
322	qp->q.meat = 0;
323	qp->q.fragments = NULL;
324	qp->q.fragments_tail = NULL;
325	qp->iif = 0;
326	qp->ecn = 0;
327
328	return 0;
329}
330
331/* Add new segment to existing queue. */
332static int ip_frag_queue(struct ipq *qp, struct sk_buff *skb)
333{
334	struct sk_buff *prev, *next;
335	struct net_device *dev;
336	unsigned int fragsize;
337	int flags, offset;
338	int ihl, end;
339	int err = -ENOENT;
340	u8 ecn;
341
342	if (qp->q.flags & INET_FRAG_COMPLETE)
343		goto err;
344
345	if (!(IPCB(skb)->flags & IPSKB_FRAG_COMPLETE) &&
346	    unlikely(ip_frag_too_far(qp)) &&
347	    unlikely(err = ip_frag_reinit(qp))) {
348		ipq_kill(qp);
349		goto err;
350	}
351
352	ecn = ip4_frag_ecn(ip_hdr(skb)->tos);
353	offset = ntohs(ip_hdr(skb)->frag_off);
354	flags = offset & ~IP_OFFSET;
355	offset &= IP_OFFSET;
356	offset <<= 3;		/* offset is in 8-byte chunks */
357	ihl = ip_hdrlen(skb);
358
359	/* Determine the position of this fragment. */
360	end = offset + skb->len - skb_network_offset(skb) - ihl;
361	err = -EINVAL;
362
363	/* Is this the final fragment? */
364	if ((flags & IP_MF) == 0) {
365		/* If we already have some bits beyond end
366		 * or have different end, the segment is corrupted.
367		 */
368		if (end < qp->q.len ||
369		    ((qp->q.flags & INET_FRAG_LAST_IN) && end != qp->q.len))
370			goto err;
371		qp->q.flags |= INET_FRAG_LAST_IN;
372		qp->q.len = end;
373	} else {
374		if (end&7) {
375			end &= ~7;
376			if (skb->ip_summed != CHECKSUM_UNNECESSARY)
377				skb->ip_summed = CHECKSUM_NONE;
378		}
379		if (end > qp->q.len) {
380			/* Some bits beyond end -> corruption. */
381			if (qp->q.flags & INET_FRAG_LAST_IN)
382				goto err;
383			qp->q.len = end;
384		}
385	}
386	if (end == offset)
387		goto err;
388
389	err = -ENOMEM;
390	if (!pskb_pull(skb, skb_network_offset(skb) + ihl))
391		goto err;
392
393	err = pskb_trim_rcsum(skb, end - offset);
394	if (err)
395		goto err;
396
397	/* Find out which fragments are in front and at the back of us
398	 * in the chain of fragments so far.  We must know where to put
399	 * this fragment, right?
400	 */
401	prev = qp->q.fragments_tail;
402	if (!prev || FRAG_CB(prev)->offset < offset) {
403		next = NULL;
404		goto found;
405	}
406	prev = NULL;
407	for (next = qp->q.fragments; next != NULL; next = next->next) {
408		if (FRAG_CB(next)->offset >= offset)
409			break;	/* bingo! */
410		prev = next;
411	}
412
413found:
414	/* We found where to put this one.  Check for overlap with
415	 * preceding fragment, and, if needed, align things so that
416	 * any overlaps are eliminated.
417	 */
418	if (prev) {
419		int i = (FRAG_CB(prev)->offset + prev->len) - offset;
420
421		if (i > 0) {
422			offset += i;
423			err = -EINVAL;
424			if (end <= offset)
425				goto err;
426			err = -ENOMEM;
427			if (!pskb_pull(skb, i))
428				goto err;
429			if (skb->ip_summed != CHECKSUM_UNNECESSARY)
430				skb->ip_summed = CHECKSUM_NONE;
431		}
432	}
433
434	err = -ENOMEM;
435
436	while (next && FRAG_CB(next)->offset < end) {
437		int i = end - FRAG_CB(next)->offset; /* overlap is 'i' bytes */
438
439		if (i < next->len) {
440			/* Eat head of the next overlapped fragment
441			 * and leave the loop. The next ones cannot overlap.
442			 */
443			if (!pskb_pull(next, i))
444				goto err;
445			FRAG_CB(next)->offset += i;
446			qp->q.meat -= i;
447			if (next->ip_summed != CHECKSUM_UNNECESSARY)
448				next->ip_summed = CHECKSUM_NONE;
449			break;
450		} else {
451			struct sk_buff *free_it = next;
452
453			/* Old fragment is completely overridden with
454			 * new one drop it.
455			 */
456			next = next->next;
457
458			if (prev)
459				prev->next = next;
460			else
461				qp->q.fragments = next;
462
463			qp->q.meat -= free_it->len;
464			sub_frag_mem_limit(qp->q.net, free_it->truesize);
465			kfree_skb(free_it);
466		}
467	}
468
469	FRAG_CB(skb)->offset = offset;
470
471	/* Insert this fragment in the chain of fragments. */
472	skb->next = next;
473	if (!next)
474		qp->q.fragments_tail = skb;
475	if (prev)
476		prev->next = skb;
477	else
478		qp->q.fragments = skb;
479
480	dev = skb->dev;
481	if (dev) {
482		qp->iif = dev->ifindex;
483		skb->dev = NULL;
484	}
485	qp->q.stamp = skb->tstamp;
486	qp->q.meat += skb->len;
487	qp->ecn |= ecn;
488	add_frag_mem_limit(qp->q.net, skb->truesize);
489	if (offset == 0)
490		qp->q.flags |= INET_FRAG_FIRST_IN;
491
492	fragsize = skb->len + ihl;
493
494	if (fragsize > qp->q.max_size)
495		qp->q.max_size = fragsize;
496
497	if (ip_hdr(skb)->frag_off & htons(IP_DF) &&
498	    fragsize > qp->max_df_size)
499		qp->max_df_size = fragsize;
500
501	if (qp->q.flags == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
502	    qp->q.meat == qp->q.len) {
503		unsigned long orefdst = skb->_skb_refdst;
504
505		skb->_skb_refdst = 0UL;
506		err = ip_frag_reasm(qp, prev, dev);
507		skb->_skb_refdst = orefdst;
508		return err;
509	}
510
511	skb_dst_drop(skb);
 
512	return -EINPROGRESS;
513
514err:
515	kfree_skb(skb);
516	return err;
517}
518
519
520/* Build a new IP datagram from all its fragments. */
521
522static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev,
523			 struct net_device *dev)
524{
525	struct net *net = container_of(qp->q.net, struct net, ipv4.frags);
526	struct iphdr *iph;
527	struct sk_buff *fp, *head = qp->q.fragments;
528	int len;
529	int ihlen;
530	int err;
 
531	u8 ecn;
532
533	ipq_kill(qp);
534
535	ecn = ip_frag_ecn_table[qp->ecn];
536	if (unlikely(ecn == 0xff)) {
537		err = -EINVAL;
538		goto out_fail;
539	}
540	/* Make the one we just received the head. */
541	if (prev) {
542		head = prev->next;
543		fp = skb_clone(head, GFP_ATOMIC);
544		if (!fp)
545			goto out_nomem;
546
547		fp->next = head->next;
548		if (!fp->next)
549			qp->q.fragments_tail = fp;
550		prev->next = fp;
551
552		skb_morph(head, qp->q.fragments);
553		head->next = qp->q.fragments->next;
554
555		consume_skb(qp->q.fragments);
556		qp->q.fragments = head;
557	}
558
559	WARN_ON(!head);
560	WARN_ON(FRAG_CB(head)->offset != 0);
561
562	/* Allocate a new buffer for the datagram. */
563	ihlen = ip_hdrlen(head);
564	len = ihlen + qp->q.len;
565
566	err = -E2BIG;
567	if (len > 65535)
568		goto out_oversize;
569
570	/* Head of list must not be cloned. */
571	if (skb_unclone(head, GFP_ATOMIC))
572		goto out_nomem;
573
574	/* If the first fragment is fragmented itself, we split
575	 * it to two chunks: the first with data and paged part
576	 * and the second, holding only fragments. */
577	if (skb_has_frag_list(head)) {
578		struct sk_buff *clone;
579		int i, plen = 0;
580
581		clone = alloc_skb(0, GFP_ATOMIC);
582		if (!clone)
583			goto out_nomem;
584		clone->next = head->next;
585		head->next = clone;
586		skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
587		skb_frag_list_init(head);
588		for (i = 0; i < skb_shinfo(head)->nr_frags; i++)
589			plen += skb_frag_size(&skb_shinfo(head)->frags[i]);
590		clone->len = clone->data_len = head->data_len - plen;
591		head->data_len -= clone->len;
592		head->len -= clone->len;
593		clone->csum = 0;
594		clone->ip_summed = head->ip_summed;
595		add_frag_mem_limit(qp->q.net, clone->truesize);
596	}
597
598	skb_shinfo(head)->frag_list = head->next;
599	skb_push(head, head->data - skb_network_header(head));
600
601	for (fp=head->next; fp; fp = fp->next) {
602		head->data_len += fp->len;
603		head->len += fp->len;
 
 
 
 
604		if (head->ip_summed != fp->ip_summed)
605			head->ip_summed = CHECKSUM_NONE;
606		else if (head->ip_summed == CHECKSUM_COMPLETE)
607			head->csum = csum_add(head->csum, fp->csum);
608		head->truesize += fp->truesize;
 
 
 
 
 
 
 
 
 
 
609	}
610	sub_frag_mem_limit(qp->q.net, head->truesize);
611
612	head->next = NULL;
613	head->dev = dev;
614	head->tstamp = qp->q.stamp;
615	IPCB(head)->frag_max_size = max(qp->max_df_size, qp->q.max_size);
616
617	iph = ip_hdr(head);
 
 
618	iph->tot_len = htons(len);
619	iph->tos |= ecn;
620
621	/* When we set IP_DF on a refragmented skb we must also force a
622	 * call to ip_fragment to avoid forwarding a DF-skb of size s while
623	 * original sender only sent fragments of size f (where f < s).
624	 *
625	 * We only set DF/IPSKB_FRAG_PMTU if such DF fragment was the largest
626	 * frag seen to avoid sending tiny DF-fragments in case skb was built
627	 * from one very small df-fragment and one large non-df frag.
628	 */
629	if (qp->max_df_size == qp->q.max_size) {
630		IPCB(head)->flags |= IPSKB_FRAG_PMTU;
631		iph->frag_off = htons(IP_DF);
632	} else {
633		iph->frag_off = 0;
634	}
635
636	ip_send_check(iph);
637
638	IP_INC_STATS_BH(net, IPSTATS_MIB_REASMOKS);
639	qp->q.fragments = NULL;
640	qp->q.fragments_tail = NULL;
641	return 0;
642
643out_nomem:
644	net_dbg_ratelimited("queue_glue: no memory for gluing queue %p\n", qp);
 
645	err = -ENOMEM;
646	goto out_fail;
647out_oversize:
648	net_info_ratelimited("Oversized IP packet from %pI4\n", &qp->saddr);
649out_fail:
650	IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
651	return err;
652}
653
654/* Process an incoming IP datagram fragment. */
655int ip_defrag(struct net *net, struct sk_buff *skb, u32 user)
656{
657	struct net_device *dev = skb->dev ? : skb_dst(skb)->dev;
658	int vif = l3mdev_master_ifindex_rcu(dev);
659	struct ipq *qp;
 
660
 
661	IP_INC_STATS_BH(net, IPSTATS_MIB_REASMREQDS);
662	skb_orphan(skb);
 
 
663
664	/* Lookup (or create) queue header */
665	qp = ip_find(net, ip_hdr(skb), user, vif);
666	if (qp) {
667		int ret;
668
669		spin_lock(&qp->q.lock);
670
671		ret = ip_frag_queue(qp, skb);
672
673		spin_unlock(&qp->q.lock);
674		ipq_put(qp);
675		return ret;
676	}
677
678	IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
679	kfree_skb(skb);
680	return -ENOMEM;
681}
682EXPORT_SYMBOL(ip_defrag);
683
684struct sk_buff *ip_check_defrag(struct net *net, struct sk_buff *skb, u32 user)
685{
686	struct iphdr iph;
687	int netoff;
688	u32 len;
689
690	if (skb->protocol != htons(ETH_P_IP))
691		return skb;
692
693	netoff = skb_network_offset(skb);
694
695	if (skb_copy_bits(skb, netoff, &iph, sizeof(iph)) < 0)
696		return skb;
697
698	if (iph.ihl < 5 || iph.version != 4)
699		return skb;
700
701	len = ntohs(iph.tot_len);
702	if (skb->len < netoff + len || len < (iph.ihl * 4))
703		return skb;
704
705	if (ip_is_fragment(&iph)) {
706		skb = skb_share_check(skb, GFP_ATOMIC);
707		if (skb) {
708			if (!pskb_may_pull(skb, netoff + iph.ihl * 4))
709				return skb;
710			if (pskb_trim_rcsum(skb, netoff + len))
711				return skb;
712			memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
713			if (ip_defrag(net, skb, user))
714				return NULL;
715			skb_clear_hash(skb);
716		}
717	}
718	return skb;
719}
720EXPORT_SYMBOL(ip_check_defrag);
721
722#ifdef CONFIG_SYSCTL
723static int zero;
724
725static struct ctl_table ip4_frags_ns_ctl_table[] = {
726	{
727		.procname	= "ipfrag_high_thresh",
728		.data		= &init_net.ipv4.frags.high_thresh,
729		.maxlen		= sizeof(int),
730		.mode		= 0644,
731		.proc_handler	= proc_dointvec_minmax,
732		.extra1		= &init_net.ipv4.frags.low_thresh
733	},
734	{
735		.procname	= "ipfrag_low_thresh",
736		.data		= &init_net.ipv4.frags.low_thresh,
737		.maxlen		= sizeof(int),
738		.mode		= 0644,
739		.proc_handler	= proc_dointvec_minmax,
740		.extra1		= &zero,
741		.extra2		= &init_net.ipv4.frags.high_thresh
742	},
743	{
744		.procname	= "ipfrag_time",
745		.data		= &init_net.ipv4.frags.timeout,
746		.maxlen		= sizeof(int),
747		.mode		= 0644,
748		.proc_handler	= proc_dointvec_jiffies,
749	},
750	{
751		.procname	= "ipfrag_max_dist",
752		.data		= &init_net.ipv4.frags.max_dist,
753		.maxlen		= sizeof(int),
754		.mode		= 0644,
755		.proc_handler	= proc_dointvec_minmax,
756		.extra1		= &zero
757	},
758	{ }
759};
760
761/* secret interval has been deprecated */
762static int ip4_frags_secret_interval_unused;
763static struct ctl_table ip4_frags_ctl_table[] = {
764	{
765		.procname	= "ipfrag_secret_interval",
766		.data		= &ip4_frags_secret_interval_unused,
767		.maxlen		= sizeof(int),
768		.mode		= 0644,
769		.proc_handler	= proc_dointvec_jiffies,
770	},
 
 
 
 
 
 
 
 
771	{ }
772};
773
774static int __net_init ip4_frags_ns_ctl_register(struct net *net)
775{
776	struct ctl_table *table;
777	struct ctl_table_header *hdr;
778
779	table = ip4_frags_ns_ctl_table;
780	if (!net_eq(net, &init_net)) {
781		table = kmemdup(table, sizeof(ip4_frags_ns_ctl_table), GFP_KERNEL);
782		if (!table)
783			goto err_alloc;
784
785		table[0].data = &net->ipv4.frags.high_thresh;
786		table[0].extra1 = &net->ipv4.frags.low_thresh;
787		table[0].extra2 = &init_net.ipv4.frags.high_thresh;
788		table[1].data = &net->ipv4.frags.low_thresh;
789		table[1].extra2 = &net->ipv4.frags.high_thresh;
790		table[2].data = &net->ipv4.frags.timeout;
791		table[3].data = &net->ipv4.frags.max_dist;
 
 
 
792	}
793
794	hdr = register_net_sysctl(net, "net/ipv4", table);
795	if (!hdr)
796		goto err_reg;
797
798	net->ipv4.frags_hdr = hdr;
799	return 0;
800
801err_reg:
802	if (!net_eq(net, &init_net))
803		kfree(table);
804err_alloc:
805	return -ENOMEM;
806}
807
808static void __net_exit ip4_frags_ns_ctl_unregister(struct net *net)
809{
810	struct ctl_table *table;
811
812	table = net->ipv4.frags_hdr->ctl_table_arg;
813	unregister_net_sysctl_table(net->ipv4.frags_hdr);
814	kfree(table);
815}
816
817static void __init ip4_frags_ctl_register(void)
818{
819	register_net_sysctl(&init_net, "net/ipv4", ip4_frags_ctl_table);
820}
821#else
822static int ip4_frags_ns_ctl_register(struct net *net)
823{
824	return 0;
825}
826
827static void ip4_frags_ns_ctl_unregister(struct net *net)
828{
829}
830
831static void __init ip4_frags_ctl_register(void)
832{
833}
834#endif
835
836static int __net_init ipv4_frags_init_net(struct net *net)
837{
838	int res;
839
840	/* Fragment cache limits.
841	 *
842	 * The fragment memory accounting code, (tries to) account for
843	 * the real memory usage, by measuring both the size of frag
844	 * queue struct (inet_frag_queue (ipv4:ipq/ipv6:frag_queue))
845	 * and the SKB's truesize.
846	 *
847	 * A 64K fragment consumes 129736 bytes (44*2944)+200
848	 * (1500 truesize == 2944, sizeof(struct ipq) == 200)
849	 *
850	 * We will commit 4MB at one time. Should we cross that limit
851	 * we will prune down to 3MB, making room for approx 8 big 64K
852	 * fragments 8x128k.
853	 */
854	net->ipv4.frags.high_thresh = 4 * 1024 * 1024;
855	net->ipv4.frags.low_thresh  = 3 * 1024 * 1024;
856	/*
857	 * Important NOTE! Fragment queue must be destroyed before MSL expires.
858	 * RFC791 is wrong proposing to prolongate timer each fragment arrival
859	 * by TTL.
860	 */
861	net->ipv4.frags.timeout = IP_FRAG_TIME;
862
863	net->ipv4.frags.max_dist = 64;
864
865	res = inet_frags_init_net(&net->ipv4.frags);
866	if (res)
867		return res;
868	res = ip4_frags_ns_ctl_register(net);
869	if (res)
870		inet_frags_uninit_net(&net->ipv4.frags);
871	return res;
872}
873
874static void __net_exit ipv4_frags_exit_net(struct net *net)
875{
876	ip4_frags_ns_ctl_unregister(net);
877	inet_frags_exit_net(&net->ipv4.frags, &ip4_frags);
878}
879
880static struct pernet_operations ip4_frags_ops = {
881	.init = ipv4_frags_init_net,
882	.exit = ipv4_frags_exit_net,
883};
884
885void __init ipfrag_init(void)
886{
887	ip4_frags_ctl_register();
888	register_pernet_subsys(&ip4_frags_ops);
889	ip4_frags.hashfn = ip4_hashfn;
890	ip4_frags.constructor = ip4_frag_init;
891	ip4_frags.destructor = ip4_frag_free;
 
892	ip4_frags.qsize = sizeof(struct ipq);
893	ip4_frags.match = ip4_frag_match;
894	ip4_frags.frag_expire = ip_expire;
895	ip4_frags.frags_cache_name = ip_frag_cache_name;
896	if (inet_frags_init(&ip4_frags))
897		panic("IP: failed to allocate ip4_frags cache\n");
898}