xref: /linux/net/ipv4/ip_fragment.c (revision cc04a46f11ea046ed53e2c832ae29e4790f7e35f)
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/vrf.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  */
57 
58 static int sysctl_ipfrag_max_dist __read_mostly = 64;
59 static const char ip_frag_cache_name[] = "ip4-frags";
60 
61 struct ipfrag_skb_cb
62 {
63 	struct inet_skb_parm	h;
64 	int			offset;
65 };
66 
67 #define FRAG_CB(skb)	((struct ipfrag_skb_cb *)((skb)->cb))
68 
69 /* Describe an entry in the "incomplete datagrams" queue. */
70 struct ipq {
71 	struct inet_frag_queue q;
72 
73 	u32		user;
74 	__be32		saddr;
75 	__be32		daddr;
76 	__be16		id;
77 	u8		protocol;
78 	u8		ecn; /* RFC3168 support */
79 	u16		max_df_size; /* largest frag with DF set seen */
80 	int             iif;
81 	int             vif;   /* VRF device index */
82 	unsigned int    rid;
83 	struct inet_peer *peer;
84 };
85 
86 static u8 ip4_frag_ecn(u8 tos)
87 {
88 	return 1 << (tos & INET_ECN_MASK);
89 }
90 
91 static struct inet_frags ip4_frags;
92 
93 int ip_frag_mem(struct net *net)
94 {
95 	return sum_frag_mem_limit(&net->ipv4.frags);
96 }
97 
98 static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev,
99 			 struct net_device *dev);
100 
101 struct ip4_create_arg {
102 	struct iphdr *iph;
103 	u32 user;
104 	int vif;
105 };
106 
107 static 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);
113 }
114 
115 static unsigned int ip4_hashfn(const struct inet_frag_queue *q)
116 {
117 	const 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 
123 static bool ip4_frag_match(const struct inet_frag_queue *q, const void *a)
124 {
125 	const struct ipq *qp;
126 	const 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 		qp->vif == arg->vif;
135 }
136 
137 static void ip4_frag_init(struct inet_frag_queue *q, const void *a)
138 {
139 	struct ipq *qp = container_of(q, struct ipq, q);
140 	struct netns_ipv4 *ipv4 = container_of(q->net, struct netns_ipv4,
141 					       frags);
142 	struct net *net = container_of(ipv4, struct net, ipv4);
143 
144 	const struct ip4_create_arg *arg = a;
145 
146 	qp->protocol = arg->iph->protocol;
147 	qp->id = arg->iph->id;
148 	qp->ecn = ip4_frag_ecn(arg->iph->tos);
149 	qp->saddr = arg->iph->saddr;
150 	qp->daddr = arg->iph->daddr;
151 	qp->vif = arg->vif;
152 	qp->user = arg->user;
153 	qp->peer = sysctl_ipfrag_max_dist ?
154 		inet_getpeer_v4(net->ipv4.peers, arg->iph->saddr, arg->vif, 1) :
155 		NULL;
156 }
157 
158 static void ip4_frag_free(struct inet_frag_queue *q)
159 {
160 	struct ipq *qp;
161 
162 	qp = container_of(q, struct ipq, q);
163 	if (qp->peer)
164 		inet_putpeer(qp->peer);
165 }
166 
167 
168 /* Destruction primitives. */
169 
170 static void ipq_put(struct ipq *ipq)
171 {
172 	inet_frag_put(&ipq->q, &ip4_frags);
173 }
174 
175 /* Kill ipq entry. It is not destroyed immediately,
176  * because caller (and someone more) holds reference count.
177  */
178 static void ipq_kill(struct ipq *ipq)
179 {
180 	inet_frag_kill(&ipq->q, &ip4_frags);
181 }
182 
183 static bool frag_expire_skip_icmp(u32 user)
184 {
185 	return user == IP_DEFRAG_AF_PACKET ||
186 	       ip_defrag_user_in_between(user, IP_DEFRAG_CONNTRACK_IN,
187 					 __IP_DEFRAG_CONNTRACK_IN_END) ||
188 	       ip_defrag_user_in_between(user, IP_DEFRAG_CONNTRACK_BRIDGE_IN,
189 					 __IP_DEFRAG_CONNTRACK_BRIDGE_IN);
190 }
191 
192 /*
193  * Oops, a fragment queue timed out.  Kill it and send an ICMP reply.
194  */
195 static 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.flags & INET_FRAG_COMPLETE)
206 		goto out;
207 
208 	ipq_kill(qp);
209 	IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
210 
211 	if (!inet_frag_evicting(&qp->q)) {
212 		struct sk_buff *head = qp->q.fragments;
213 		const struct iphdr *iph;
214 		int err;
215 
216 		IP_INC_STATS_BH(net, IPSTATS_MIB_REASMTIMEOUT);
217 
218 		if (!(qp->q.flags & INET_FRAG_FIRST_IN) || !qp->q.fragments)
219 			goto out;
220 
221 		rcu_read_lock();
222 		head->dev = dev_get_by_index_rcu(net, qp->iif);
223 		if (!head->dev)
224 			goto out_rcu_unlock;
225 
226 		/* skb has no dst, perform route lookup again */
227 		iph = ip_hdr(head);
228 		err = ip_route_input_noref(head, iph->daddr, iph->saddr,
229 					   iph->tos, head->dev);
230 		if (err)
231 			goto out_rcu_unlock;
232 
233 		/* Only an end host needs to send an ICMP
234 		 * "Fragment Reassembly Timeout" message, per RFC792.
235 		 */
236 		if (frag_expire_skip_icmp(qp->user) &&
237 		    (skb_rtable(head)->rt_type != RTN_LOCAL))
238 			goto out_rcu_unlock;
239 
240 		/* Send an ICMP "Fragment Reassembly Timeout" message. */
241 		icmp_send(head, ICMP_TIME_EXCEEDED, ICMP_EXC_FRAGTIME, 0);
242 out_rcu_unlock:
243 		rcu_read_unlock();
244 	}
245 out:
246 	spin_unlock(&qp->q.lock);
247 	ipq_put(qp);
248 }
249 
250 /* Find the correct entry in the "incomplete datagrams" queue for
251  * this IP datagram, and create new one, if nothing is found.
252  */
253 static struct ipq *ip_find(struct net *net, struct iphdr *iph,
254 			   u32 user, int vif)
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 	arg.vif = vif;
263 
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? */
275 static 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 
302 static 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.net, sum_truesize);
321 
322 	qp->q.flags = 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. */
334 static int ip_frag_queue(struct ipq *qp, struct sk_buff *skb)
335 {
336 	struct sk_buff *prev, *next;
337 	struct net_device *dev;
338 	unsigned int fragsize;
339 	int flags, offset;
340 	int ihl, end;
341 	int err = -ENOENT;
342 	u8 ecn;
343 
344 	if (qp->q.flags & INET_FRAG_COMPLETE)
345 		goto err;
346 
347 	if (!(IPCB(skb)->flags & IPSKB_FRAG_COMPLETE) &&
348 	    unlikely(ip_frag_too_far(qp)) &&
349 	    unlikely(err = ip_frag_reinit(qp))) {
350 		ipq_kill(qp);
351 		goto err;
352 	}
353 
354 	ecn = ip4_frag_ecn(ip_hdr(skb)->tos);
355 	offset = ntohs(ip_hdr(skb)->frag_off);
356 	flags = offset & ~IP_OFFSET;
357 	offset &= IP_OFFSET;
358 	offset <<= 3;		/* offset is in 8-byte chunks */
359 	ihl = ip_hdrlen(skb);
360 
361 	/* Determine the position of this fragment. */
362 	end = offset + skb->len - skb_network_offset(skb) - ihl;
363 	err = -EINVAL;
364 
365 	/* Is this the final fragment? */
366 	if ((flags & IP_MF) == 0) {
367 		/* If we already have some bits beyond end
368 		 * or have different end, the segment is corrupted.
369 		 */
370 		if (end < qp->q.len ||
371 		    ((qp->q.flags & INET_FRAG_LAST_IN) && end != qp->q.len))
372 			goto err;
373 		qp->q.flags |= INET_FRAG_LAST_IN;
374 		qp->q.len = end;
375 	} else {
376 		if (end&7) {
377 			end &= ~7;
378 			if (skb->ip_summed != CHECKSUM_UNNECESSARY)
379 				skb->ip_summed = CHECKSUM_NONE;
380 		}
381 		if (end > qp->q.len) {
382 			/* Some bits beyond end -> corruption. */
383 			if (qp->q.flags & INET_FRAG_LAST_IN)
384 				goto err;
385 			qp->q.len = end;
386 		}
387 	}
388 	if (end == offset)
389 		goto err;
390 
391 	err = -ENOMEM;
392 	if (!pskb_pull(skb, skb_network_offset(skb) + ihl))
393 		goto err;
394 
395 	err = pskb_trim_rcsum(skb, end - offset);
396 	if (err)
397 		goto err;
398 
399 	/* Find out which fragments are in front and at the back of us
400 	 * in the chain of fragments so far.  We must know where to put
401 	 * this fragment, right?
402 	 */
403 	prev = qp->q.fragments_tail;
404 	if (!prev || FRAG_CB(prev)->offset < offset) {
405 		next = NULL;
406 		goto found;
407 	}
408 	prev = NULL;
409 	for (next = qp->q.fragments; next != NULL; next = next->next) {
410 		if (FRAG_CB(next)->offset >= offset)
411 			break;	/* bingo! */
412 		prev = next;
413 	}
414 
415 found:
416 	/* We found where to put this one.  Check for overlap with
417 	 * preceding fragment, and, if needed, align things so that
418 	 * any overlaps are eliminated.
419 	 */
420 	if (prev) {
421 		int i = (FRAG_CB(prev)->offset + prev->len) - offset;
422 
423 		if (i > 0) {
424 			offset += i;
425 			err = -EINVAL;
426 			if (end <= offset)
427 				goto err;
428 			err = -ENOMEM;
429 			if (!pskb_pull(skb, i))
430 				goto err;
431 			if (skb->ip_summed != CHECKSUM_UNNECESSARY)
432 				skb->ip_summed = CHECKSUM_NONE;
433 		}
434 	}
435 
436 	err = -ENOMEM;
437 
438 	while (next && FRAG_CB(next)->offset < end) {
439 		int i = end - FRAG_CB(next)->offset; /* overlap is 'i' bytes */
440 
441 		if (i < next->len) {
442 			/* Eat head of the next overlapped fragment
443 			 * and leave the loop. The next ones cannot overlap.
444 			 */
445 			if (!pskb_pull(next, i))
446 				goto err;
447 			FRAG_CB(next)->offset += i;
448 			qp->q.meat -= i;
449 			if (next->ip_summed != CHECKSUM_UNNECESSARY)
450 				next->ip_summed = CHECKSUM_NONE;
451 			break;
452 		} else {
453 			struct sk_buff *free_it = next;
454 
455 			/* Old fragment is completely overridden with
456 			 * new one drop it.
457 			 */
458 			next = next->next;
459 
460 			if (prev)
461 				prev->next = next;
462 			else
463 				qp->q.fragments = next;
464 
465 			qp->q.meat -= free_it->len;
466 			sub_frag_mem_limit(qp->q.net, free_it->truesize);
467 			kfree_skb(free_it);
468 		}
469 	}
470 
471 	FRAG_CB(skb)->offset = offset;
472 
473 	/* Insert this fragment in the chain of fragments. */
474 	skb->next = next;
475 	if (!next)
476 		qp->q.fragments_tail = skb;
477 	if (prev)
478 		prev->next = skb;
479 	else
480 		qp->q.fragments = skb;
481 
482 	dev = skb->dev;
483 	if (dev) {
484 		qp->iif = dev->ifindex;
485 		skb->dev = NULL;
486 	}
487 	qp->q.stamp = skb->tstamp;
488 	qp->q.meat += skb->len;
489 	qp->ecn |= ecn;
490 	add_frag_mem_limit(qp->q.net, skb->truesize);
491 	if (offset == 0)
492 		qp->q.flags |= INET_FRAG_FIRST_IN;
493 
494 	fragsize = skb->len + ihl;
495 
496 	if (fragsize > qp->q.max_size)
497 		qp->q.max_size = fragsize;
498 
499 	if (ip_hdr(skb)->frag_off & htons(IP_DF) &&
500 	    fragsize > qp->max_df_size)
501 		qp->max_df_size = fragsize;
502 
503 	if (qp->q.flags == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
504 	    qp->q.meat == qp->q.len) {
505 		unsigned long orefdst = skb->_skb_refdst;
506 
507 		skb->_skb_refdst = 0UL;
508 		err = ip_frag_reasm(qp, prev, dev);
509 		skb->_skb_refdst = orefdst;
510 		return err;
511 	}
512 
513 	skb_dst_drop(skb);
514 	return -EINPROGRESS;
515 
516 err:
517 	kfree_skb(skb);
518 	return err;
519 }
520 
521 
522 /* Build a new IP datagram from all its fragments. */
523 
524 static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev,
525 			 struct net_device *dev)
526 {
527 	struct net *net = container_of(qp->q.net, struct net, ipv4.frags);
528 	struct iphdr *iph;
529 	struct sk_buff *fp, *head = qp->q.fragments;
530 	int len;
531 	int ihlen;
532 	int err;
533 	u8 ecn;
534 
535 	ipq_kill(qp);
536 
537 	ecn = ip_frag_ecn_table[qp->ecn];
538 	if (unlikely(ecn == 0xff)) {
539 		err = -EINVAL;
540 		goto out_fail;
541 	}
542 	/* Make the one we just received the head. */
543 	if (prev) {
544 		head = prev->next;
545 		fp = skb_clone(head, GFP_ATOMIC);
546 		if (!fp)
547 			goto out_nomem;
548 
549 		fp->next = head->next;
550 		if (!fp->next)
551 			qp->q.fragments_tail = fp;
552 		prev->next = fp;
553 
554 		skb_morph(head, qp->q.fragments);
555 		head->next = qp->q.fragments->next;
556 
557 		consume_skb(qp->q.fragments);
558 		qp->q.fragments = head;
559 	}
560 
561 	WARN_ON(!head);
562 	WARN_ON(FRAG_CB(head)->offset != 0);
563 
564 	/* Allocate a new buffer for the datagram. */
565 	ihlen = ip_hdrlen(head);
566 	len = ihlen + qp->q.len;
567 
568 	err = -E2BIG;
569 	if (len > 65535)
570 		goto out_oversize;
571 
572 	/* Head of list must not be cloned. */
573 	if (skb_unclone(head, GFP_ATOMIC))
574 		goto out_nomem;
575 
576 	/* If the first fragment is fragmented itself, we split
577 	 * it to two chunks: the first with data and paged part
578 	 * and the second, holding only fragments. */
579 	if (skb_has_frag_list(head)) {
580 		struct sk_buff *clone;
581 		int i, plen = 0;
582 
583 		clone = alloc_skb(0, GFP_ATOMIC);
584 		if (!clone)
585 			goto out_nomem;
586 		clone->next = head->next;
587 		head->next = clone;
588 		skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
589 		skb_frag_list_init(head);
590 		for (i = 0; i < skb_shinfo(head)->nr_frags; i++)
591 			plen += skb_frag_size(&skb_shinfo(head)->frags[i]);
592 		clone->len = clone->data_len = head->data_len - plen;
593 		head->data_len -= clone->len;
594 		head->len -= clone->len;
595 		clone->csum = 0;
596 		clone->ip_summed = head->ip_summed;
597 		add_frag_mem_limit(qp->q.net, clone->truesize);
598 	}
599 
600 	skb_shinfo(head)->frag_list = head->next;
601 	skb_push(head, head->data - skb_network_header(head));
602 
603 	for (fp=head->next; fp; fp = fp->next) {
604 		head->data_len += fp->len;
605 		head->len += fp->len;
606 		if (head->ip_summed != fp->ip_summed)
607 			head->ip_summed = CHECKSUM_NONE;
608 		else if (head->ip_summed == CHECKSUM_COMPLETE)
609 			head->csum = csum_add(head->csum, fp->csum);
610 		head->truesize += fp->truesize;
611 	}
612 	sub_frag_mem_limit(qp->q.net, head->truesize);
613 
614 	head->next = NULL;
615 	head->dev = dev;
616 	head->tstamp = qp->q.stamp;
617 	IPCB(head)->frag_max_size = max(qp->max_df_size, qp->q.max_size);
618 
619 	iph = ip_hdr(head);
620 	iph->tot_len = htons(len);
621 	iph->tos |= ecn;
622 
623 	/* When we set IP_DF on a refragmented skb we must also force a
624 	 * call to ip_fragment to avoid forwarding a DF-skb of size s while
625 	 * original sender only sent fragments of size f (where f < s).
626 	 *
627 	 * We only set DF/IPSKB_FRAG_PMTU if such DF fragment was the largest
628 	 * frag seen to avoid sending tiny DF-fragments in case skb was built
629 	 * from one very small df-fragment and one large non-df frag.
630 	 */
631 	if (qp->max_df_size == qp->q.max_size) {
632 		IPCB(head)->flags |= IPSKB_FRAG_PMTU;
633 		iph->frag_off = htons(IP_DF);
634 	} else {
635 		iph->frag_off = 0;
636 	}
637 
638 	ip_send_check(iph);
639 
640 	IP_INC_STATS_BH(net, IPSTATS_MIB_REASMOKS);
641 	qp->q.fragments = NULL;
642 	qp->q.fragments_tail = NULL;
643 	return 0;
644 
645 out_nomem:
646 	net_dbg_ratelimited("queue_glue: no memory for gluing queue %p\n", qp);
647 	err = -ENOMEM;
648 	goto out_fail;
649 out_oversize:
650 	net_info_ratelimited("Oversized IP packet from %pI4\n", &qp->saddr);
651 out_fail:
652 	IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
653 	return err;
654 }
655 
656 /* Process an incoming IP datagram fragment. */
657 int ip_defrag(struct sk_buff *skb, u32 user)
658 {
659 	struct net_device *dev = skb->dev ? : skb_dst(skb)->dev;
660 	int vif = vrf_master_ifindex_rcu(dev);
661 	struct net *net = dev_net(dev);
662 	struct ipq *qp;
663 
664 	IP_INC_STATS_BH(net, IPSTATS_MIB_REASMREQDS);
665 
666 	/* Lookup (or create) queue header */
667 	qp = ip_find(net, ip_hdr(skb), user, vif);
668 	if (qp) {
669 		int ret;
670 
671 		spin_lock(&qp->q.lock);
672 
673 		ret = ip_frag_queue(qp, skb);
674 
675 		spin_unlock(&qp->q.lock);
676 		ipq_put(qp);
677 		return ret;
678 	}
679 
680 	IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
681 	kfree_skb(skb);
682 	return -ENOMEM;
683 }
684 EXPORT_SYMBOL(ip_defrag);
685 
686 struct sk_buff *ip_check_defrag(struct sk_buff *skb, u32 user)
687 {
688 	struct iphdr iph;
689 	int netoff;
690 	u32 len;
691 
692 	if (skb->protocol != htons(ETH_P_IP))
693 		return skb;
694 
695 	netoff = skb_network_offset(skb);
696 
697 	if (skb_copy_bits(skb, netoff, &iph, sizeof(iph)) < 0)
698 		return skb;
699 
700 	if (iph.ihl < 5 || iph.version != 4)
701 		return skb;
702 
703 	len = ntohs(iph.tot_len);
704 	if (skb->len < netoff + len || len < (iph.ihl * 4))
705 		return skb;
706 
707 	if (ip_is_fragment(&iph)) {
708 		skb = skb_share_check(skb, GFP_ATOMIC);
709 		if (skb) {
710 			if (!pskb_may_pull(skb, netoff + iph.ihl * 4))
711 				return skb;
712 			if (pskb_trim_rcsum(skb, netoff + len))
713 				return skb;
714 			memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
715 			if (ip_defrag(skb, user))
716 				return NULL;
717 			skb_clear_hash(skb);
718 		}
719 	}
720 	return skb;
721 }
722 EXPORT_SYMBOL(ip_check_defrag);
723 
724 #ifdef CONFIG_SYSCTL
725 static int zero;
726 
727 static struct ctl_table ip4_frags_ns_ctl_table[] = {
728 	{
729 		.procname	= "ipfrag_high_thresh",
730 		.data		= &init_net.ipv4.frags.high_thresh,
731 		.maxlen		= sizeof(int),
732 		.mode		= 0644,
733 		.proc_handler	= proc_dointvec_minmax,
734 		.extra1		= &init_net.ipv4.frags.low_thresh
735 	},
736 	{
737 		.procname	= "ipfrag_low_thresh",
738 		.data		= &init_net.ipv4.frags.low_thresh,
739 		.maxlen		= sizeof(int),
740 		.mode		= 0644,
741 		.proc_handler	= proc_dointvec_minmax,
742 		.extra1		= &zero,
743 		.extra2		= &init_net.ipv4.frags.high_thresh
744 	},
745 	{
746 		.procname	= "ipfrag_time",
747 		.data		= &init_net.ipv4.frags.timeout,
748 		.maxlen		= sizeof(int),
749 		.mode		= 0644,
750 		.proc_handler	= proc_dointvec_jiffies,
751 	},
752 	{ }
753 };
754 
755 /* secret interval has been deprecated */
756 static int ip4_frags_secret_interval_unused;
757 static struct ctl_table ip4_frags_ctl_table[] = {
758 	{
759 		.procname	= "ipfrag_secret_interval",
760 		.data		= &ip4_frags_secret_interval_unused,
761 		.maxlen		= sizeof(int),
762 		.mode		= 0644,
763 		.proc_handler	= proc_dointvec_jiffies,
764 	},
765 	{
766 		.procname	= "ipfrag_max_dist",
767 		.data		= &sysctl_ipfrag_max_dist,
768 		.maxlen		= sizeof(int),
769 		.mode		= 0644,
770 		.proc_handler	= proc_dointvec_minmax,
771 		.extra1		= &zero
772 	},
773 	{ }
774 };
775 
776 static int __net_init ip4_frags_ns_ctl_register(struct net *net)
777 {
778 	struct ctl_table *table;
779 	struct ctl_table_header *hdr;
780 
781 	table = ip4_frags_ns_ctl_table;
782 	if (!net_eq(net, &init_net)) {
783 		table = kmemdup(table, sizeof(ip4_frags_ns_ctl_table), GFP_KERNEL);
784 		if (!table)
785 			goto err_alloc;
786 
787 		table[0].data = &net->ipv4.frags.high_thresh;
788 		table[0].extra1 = &net->ipv4.frags.low_thresh;
789 		table[0].extra2 = &init_net.ipv4.frags.high_thresh;
790 		table[1].data = &net->ipv4.frags.low_thresh;
791 		table[1].extra2 = &net->ipv4.frags.high_thresh;
792 		table[2].data = &net->ipv4.frags.timeout;
793 
794 		/* Don't export sysctls to unprivileged users */
795 		if (net->user_ns != &init_user_ns)
796 			table[0].procname = NULL;
797 	}
798 
799 	hdr = register_net_sysctl(net, "net/ipv4", table);
800 	if (!hdr)
801 		goto err_reg;
802 
803 	net->ipv4.frags_hdr = hdr;
804 	return 0;
805 
806 err_reg:
807 	if (!net_eq(net, &init_net))
808 		kfree(table);
809 err_alloc:
810 	return -ENOMEM;
811 }
812 
813 static void __net_exit ip4_frags_ns_ctl_unregister(struct net *net)
814 {
815 	struct ctl_table *table;
816 
817 	table = net->ipv4.frags_hdr->ctl_table_arg;
818 	unregister_net_sysctl_table(net->ipv4.frags_hdr);
819 	kfree(table);
820 }
821 
822 static void __init ip4_frags_ctl_register(void)
823 {
824 	register_net_sysctl(&init_net, "net/ipv4", ip4_frags_ctl_table);
825 }
826 #else
827 static int ip4_frags_ns_ctl_register(struct net *net)
828 {
829 	return 0;
830 }
831 
832 static void ip4_frags_ns_ctl_unregister(struct net *net)
833 {
834 }
835 
836 static void __init ip4_frags_ctl_register(void)
837 {
838 }
839 #endif
840 
841 static int __net_init ipv4_frags_init_net(struct net *net)
842 {
843 	/* Fragment cache limits.
844 	 *
845 	 * The fragment memory accounting code, (tries to) account for
846 	 * the real memory usage, by measuring both the size of frag
847 	 * queue struct (inet_frag_queue (ipv4:ipq/ipv6:frag_queue))
848 	 * and the SKB's truesize.
849 	 *
850 	 * A 64K fragment consumes 129736 bytes (44*2944)+200
851 	 * (1500 truesize == 2944, sizeof(struct ipq) == 200)
852 	 *
853 	 * We will commit 4MB at one time. Should we cross that limit
854 	 * we will prune down to 3MB, making room for approx 8 big 64K
855 	 * fragments 8x128k.
856 	 */
857 	net->ipv4.frags.high_thresh = 4 * 1024 * 1024;
858 	net->ipv4.frags.low_thresh  = 3 * 1024 * 1024;
859 	/*
860 	 * Important NOTE! Fragment queue must be destroyed before MSL expires.
861 	 * RFC791 is wrong proposing to prolongate timer each fragment arrival
862 	 * by TTL.
863 	 */
864 	net->ipv4.frags.timeout = IP_FRAG_TIME;
865 
866 	inet_frags_init_net(&net->ipv4.frags);
867 
868 	return ip4_frags_ns_ctl_register(net);
869 }
870 
871 static void __net_exit ipv4_frags_exit_net(struct net *net)
872 {
873 	ip4_frags_ns_ctl_unregister(net);
874 	inet_frags_exit_net(&net->ipv4.frags, &ip4_frags);
875 }
876 
877 static struct pernet_operations ip4_frags_ops = {
878 	.init = ipv4_frags_init_net,
879 	.exit = ipv4_frags_exit_net,
880 };
881 
882 void __init ipfrag_init(void)
883 {
884 	ip4_frags_ctl_register();
885 	register_pernet_subsys(&ip4_frags_ops);
886 	ip4_frags.hashfn = ip4_hashfn;
887 	ip4_frags.constructor = ip4_frag_init;
888 	ip4_frags.destructor = ip4_frag_free;
889 	ip4_frags.skb_free = NULL;
890 	ip4_frags.qsize = sizeof(struct ipq);
891 	ip4_frags.match = ip4_frag_match;
892 	ip4_frags.frag_expire = ip_expire;
893 	ip4_frags.frags_cache_name = ip_frag_cache_name;
894 	if (inet_frags_init(&ip4_frags))
895 		panic("IP: failed to allocate ip4_frags cache\n");
896 }
897