xref: /openbmc/linux/net/ipv4/ip_fragment.c (revision bbde9fc1824aab58bc78c084163007dd6c03fe5b)
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 
57 static int sysctl_ipfrag_max_dist __read_mostly = 64;
58 static const char ip_frag_cache_name[] = "ip4-frags";
59 
60 struct ipfrag_skb_cb
61 {
62 	struct inet_skb_parm	h;
63 	int			offset;
64 };
65 
66 #define FRAG_CB(skb)	((struct ipfrag_skb_cb *)((skb)->cb))
67 
68 /* Describe an entry in the "incomplete datagrams" queue. */
69 struct ipq {
70 	struct inet_frag_queue q;
71 
72 	u32		user;
73 	__be32		saddr;
74 	__be32		daddr;
75 	__be16		id;
76 	u8		protocol;
77 	u8		ecn; /* RFC3168 support */
78 	u16		max_df_size; /* largest frag with DF set seen */
79 	int             iif;
80 	unsigned int    rid;
81 	struct inet_peer *peer;
82 };
83 
84 static u8 ip4_frag_ecn(u8 tos)
85 {
86 	return 1 << (tos & INET_ECN_MASK);
87 }
88 
89 static struct inet_frags ip4_frags;
90 
91 int ip_frag_mem(struct net *net)
92 {
93 	return sum_frag_mem_limit(&net->ipv4.frags);
94 }
95 
96 static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev,
97 			 struct net_device *dev);
98 
99 struct ip4_create_arg {
100 	struct iphdr *iph;
101 	u32 user;
102 };
103 
104 static unsigned int ipqhashfn(__be16 id, __be32 saddr, __be32 daddr, u8 prot)
105 {
106 	net_get_random_once(&ip4_frags.rnd, sizeof(ip4_frags.rnd));
107 	return jhash_3words((__force u32)id << 16 | prot,
108 			    (__force u32)saddr, (__force u32)daddr,
109 			    ip4_frags.rnd);
110 }
111 
112 static unsigned int ip4_hashfn(const struct inet_frag_queue *q)
113 {
114 	const struct ipq *ipq;
115 
116 	ipq = container_of(q, struct ipq, q);
117 	return ipqhashfn(ipq->id, ipq->saddr, ipq->daddr, ipq->protocol);
118 }
119 
120 static bool ip4_frag_match(const struct inet_frag_queue *q, const void *a)
121 {
122 	const struct ipq *qp;
123 	const struct ip4_create_arg *arg = a;
124 
125 	qp = container_of(q, struct ipq, q);
126 	return	qp->id == arg->iph->id &&
127 		qp->saddr == arg->iph->saddr &&
128 		qp->daddr == arg->iph->daddr &&
129 		qp->protocol == arg->iph->protocol &&
130 		qp->user == arg->user;
131 }
132 
133 static void ip4_frag_init(struct inet_frag_queue *q, const void *a)
134 {
135 	struct ipq *qp = container_of(q, struct ipq, q);
136 	struct netns_ipv4 *ipv4 = container_of(q->net, struct netns_ipv4,
137 					       frags);
138 	struct net *net = container_of(ipv4, struct net, ipv4);
139 
140 	const struct ip4_create_arg *arg = a;
141 
142 	qp->protocol = arg->iph->protocol;
143 	qp->id = arg->iph->id;
144 	qp->ecn = ip4_frag_ecn(arg->iph->tos);
145 	qp->saddr = arg->iph->saddr;
146 	qp->daddr = arg->iph->daddr;
147 	qp->user = arg->user;
148 	qp->peer = sysctl_ipfrag_max_dist ?
149 		inet_getpeer_v4(net->ipv4.peers, arg->iph->saddr, 1) : NULL;
150 }
151 
152 static void ip4_frag_free(struct inet_frag_queue *q)
153 {
154 	struct ipq *qp;
155 
156 	qp = container_of(q, struct ipq, q);
157 	if (qp->peer)
158 		inet_putpeer(qp->peer);
159 }
160 
161 
162 /* Destruction primitives. */
163 
164 static void ipq_put(struct ipq *ipq)
165 {
166 	inet_frag_put(&ipq->q, &ip4_frags);
167 }
168 
169 /* Kill ipq entry. It is not destroyed immediately,
170  * because caller (and someone more) holds reference count.
171  */
172 static void ipq_kill(struct ipq *ipq)
173 {
174 	inet_frag_kill(&ipq->q, &ip4_frags);
175 }
176 
177 static bool frag_expire_skip_icmp(u32 user)
178 {
179 	return user == IP_DEFRAG_AF_PACKET ||
180 	       ip_defrag_user_in_between(user, IP_DEFRAG_CONNTRACK_IN,
181 					 __IP_DEFRAG_CONNTRACK_IN_END) ||
182 	       ip_defrag_user_in_between(user, IP_DEFRAG_CONNTRACK_BRIDGE_IN,
183 					 __IP_DEFRAG_CONNTRACK_BRIDGE_IN);
184 }
185 
186 /*
187  * Oops, a fragment queue timed out.  Kill it and send an ICMP reply.
188  */
189 static void ip_expire(unsigned long arg)
190 {
191 	struct ipq *qp;
192 	struct net *net;
193 
194 	qp = container_of((struct inet_frag_queue *) arg, struct ipq, q);
195 	net = container_of(qp->q.net, struct net, ipv4.frags);
196 
197 	spin_lock(&qp->q.lock);
198 
199 	if (qp->q.flags & INET_FRAG_COMPLETE)
200 		goto out;
201 
202 	ipq_kill(qp);
203 	IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
204 
205 	if (!inet_frag_evicting(&qp->q)) {
206 		struct sk_buff *head = qp->q.fragments;
207 		const struct iphdr *iph;
208 		int err;
209 
210 		IP_INC_STATS_BH(net, IPSTATS_MIB_REASMTIMEOUT);
211 
212 		if (!(qp->q.flags & INET_FRAG_FIRST_IN) || !qp->q.fragments)
213 			goto out;
214 
215 		rcu_read_lock();
216 		head->dev = dev_get_by_index_rcu(net, qp->iif);
217 		if (!head->dev)
218 			goto out_rcu_unlock;
219 
220 		/* skb has no dst, perform route lookup again */
221 		iph = ip_hdr(head);
222 		err = ip_route_input_noref(head, iph->daddr, iph->saddr,
223 					   iph->tos, head->dev);
224 		if (err)
225 			goto out_rcu_unlock;
226 
227 		/* Only an end host needs to send an ICMP
228 		 * "Fragment Reassembly Timeout" message, per RFC792.
229 		 */
230 		if (frag_expire_skip_icmp(qp->user) &&
231 		    (skb_rtable(head)->rt_type != RTN_LOCAL))
232 			goto out_rcu_unlock;
233 
234 		/* Send an ICMP "Fragment Reassembly Timeout" message. */
235 		icmp_send(head, ICMP_TIME_EXCEEDED, ICMP_EXC_FRAGTIME, 0);
236 out_rcu_unlock:
237 		rcu_read_unlock();
238 	}
239 out:
240 	spin_unlock(&qp->q.lock);
241 	ipq_put(qp);
242 }
243 
244 /* Find the correct entry in the "incomplete datagrams" queue for
245  * this IP datagram, and create new one, if nothing is found.
246  */
247 static struct ipq *ip_find(struct net *net, struct iphdr *iph, u32 user)
248 {
249 	struct inet_frag_queue *q;
250 	struct ip4_create_arg arg;
251 	unsigned int hash;
252 
253 	arg.iph = iph;
254 	arg.user = user;
255 
256 	hash = ipqhashfn(iph->id, iph->saddr, iph->daddr, iph->protocol);
257 
258 	q = inet_frag_find(&net->ipv4.frags, &ip4_frags, &arg, hash);
259 	if (IS_ERR_OR_NULL(q)) {
260 		inet_frag_maybe_warn_overflow(q, pr_fmt());
261 		return NULL;
262 	}
263 	return container_of(q, struct ipq, q);
264 }
265 
266 /* Is the fragment too far ahead to be part of ipq? */
267 static int ip_frag_too_far(struct ipq *qp)
268 {
269 	struct inet_peer *peer = qp->peer;
270 	unsigned int max = sysctl_ipfrag_max_dist;
271 	unsigned int start, end;
272 
273 	int rc;
274 
275 	if (!peer || !max)
276 		return 0;
277 
278 	start = qp->rid;
279 	end = atomic_inc_return(&peer->rid);
280 	qp->rid = end;
281 
282 	rc = qp->q.fragments && (end - start) > max;
283 
284 	if (rc) {
285 		struct net *net;
286 
287 		net = container_of(qp->q.net, struct net, ipv4.frags);
288 		IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
289 	}
290 
291 	return rc;
292 }
293 
294 static int ip_frag_reinit(struct ipq *qp)
295 {
296 	struct sk_buff *fp;
297 	unsigned int sum_truesize = 0;
298 
299 	if (!mod_timer(&qp->q.timer, jiffies + qp->q.net->timeout)) {
300 		atomic_inc(&qp->q.refcnt);
301 		return -ETIMEDOUT;
302 	}
303 
304 	fp = qp->q.fragments;
305 	do {
306 		struct sk_buff *xp = fp->next;
307 
308 		sum_truesize += fp->truesize;
309 		kfree_skb(fp);
310 		fp = xp;
311 	} while (fp);
312 	sub_frag_mem_limit(qp->q.net, sum_truesize);
313 
314 	qp->q.flags = 0;
315 	qp->q.len = 0;
316 	qp->q.meat = 0;
317 	qp->q.fragments = NULL;
318 	qp->q.fragments_tail = NULL;
319 	qp->iif = 0;
320 	qp->ecn = 0;
321 
322 	return 0;
323 }
324 
325 /* Add new segment to existing queue. */
326 static int ip_frag_queue(struct ipq *qp, struct sk_buff *skb)
327 {
328 	struct sk_buff *prev, *next;
329 	struct net_device *dev;
330 	unsigned int fragsize;
331 	int flags, offset;
332 	int ihl, end;
333 	int err = -ENOENT;
334 	u8 ecn;
335 
336 	if (qp->q.flags & INET_FRAG_COMPLETE)
337 		goto err;
338 
339 	if (!(IPCB(skb)->flags & IPSKB_FRAG_COMPLETE) &&
340 	    unlikely(ip_frag_too_far(qp)) &&
341 	    unlikely(err = ip_frag_reinit(qp))) {
342 		ipq_kill(qp);
343 		goto err;
344 	}
345 
346 	ecn = ip4_frag_ecn(ip_hdr(skb)->tos);
347 	offset = ntohs(ip_hdr(skb)->frag_off);
348 	flags = offset & ~IP_OFFSET;
349 	offset &= IP_OFFSET;
350 	offset <<= 3;		/* offset is in 8-byte chunks */
351 	ihl = ip_hdrlen(skb);
352 
353 	/* Determine the position of this fragment. */
354 	end = offset + skb->len - skb_network_offset(skb) - ihl;
355 	err = -EINVAL;
356 
357 	/* Is this the final fragment? */
358 	if ((flags & IP_MF) == 0) {
359 		/* If we already have some bits beyond end
360 		 * or have different end, the segment is corrupted.
361 		 */
362 		if (end < qp->q.len ||
363 		    ((qp->q.flags & INET_FRAG_LAST_IN) && end != qp->q.len))
364 			goto err;
365 		qp->q.flags |= INET_FRAG_LAST_IN;
366 		qp->q.len = end;
367 	} else {
368 		if (end&7) {
369 			end &= ~7;
370 			if (skb->ip_summed != CHECKSUM_UNNECESSARY)
371 				skb->ip_summed = CHECKSUM_NONE;
372 		}
373 		if (end > qp->q.len) {
374 			/* Some bits beyond end -> corruption. */
375 			if (qp->q.flags & INET_FRAG_LAST_IN)
376 				goto err;
377 			qp->q.len = end;
378 		}
379 	}
380 	if (end == offset)
381 		goto err;
382 
383 	err = -ENOMEM;
384 	if (!pskb_pull(skb, skb_network_offset(skb) + ihl))
385 		goto err;
386 
387 	err = pskb_trim_rcsum(skb, end - offset);
388 	if (err)
389 		goto err;
390 
391 	/* Find out which fragments are in front and at the back of us
392 	 * in the chain of fragments so far.  We must know where to put
393 	 * this fragment, right?
394 	 */
395 	prev = qp->q.fragments_tail;
396 	if (!prev || FRAG_CB(prev)->offset < offset) {
397 		next = NULL;
398 		goto found;
399 	}
400 	prev = NULL;
401 	for (next = qp->q.fragments; next != NULL; next = next->next) {
402 		if (FRAG_CB(next)->offset >= offset)
403 			break;	/* bingo! */
404 		prev = next;
405 	}
406 
407 found:
408 	/* We found where to put this one.  Check for overlap with
409 	 * preceding fragment, and, if needed, align things so that
410 	 * any overlaps are eliminated.
411 	 */
412 	if (prev) {
413 		int i = (FRAG_CB(prev)->offset + prev->len) - offset;
414 
415 		if (i > 0) {
416 			offset += i;
417 			err = -EINVAL;
418 			if (end <= offset)
419 				goto err;
420 			err = -ENOMEM;
421 			if (!pskb_pull(skb, i))
422 				goto err;
423 			if (skb->ip_summed != CHECKSUM_UNNECESSARY)
424 				skb->ip_summed = CHECKSUM_NONE;
425 		}
426 	}
427 
428 	err = -ENOMEM;
429 
430 	while (next && FRAG_CB(next)->offset < end) {
431 		int i = end - FRAG_CB(next)->offset; /* overlap is 'i' bytes */
432 
433 		if (i < next->len) {
434 			/* Eat head of the next overlapped fragment
435 			 * and leave the loop. The next ones cannot overlap.
436 			 */
437 			if (!pskb_pull(next, i))
438 				goto err;
439 			FRAG_CB(next)->offset += i;
440 			qp->q.meat -= i;
441 			if (next->ip_summed != CHECKSUM_UNNECESSARY)
442 				next->ip_summed = CHECKSUM_NONE;
443 			break;
444 		} else {
445 			struct sk_buff *free_it = next;
446 
447 			/* Old fragment is completely overridden with
448 			 * new one drop it.
449 			 */
450 			next = next->next;
451 
452 			if (prev)
453 				prev->next = next;
454 			else
455 				qp->q.fragments = next;
456 
457 			qp->q.meat -= free_it->len;
458 			sub_frag_mem_limit(qp->q.net, free_it->truesize);
459 			kfree_skb(free_it);
460 		}
461 	}
462 
463 	FRAG_CB(skb)->offset = offset;
464 
465 	/* Insert this fragment in the chain of fragments. */
466 	skb->next = next;
467 	if (!next)
468 		qp->q.fragments_tail = skb;
469 	if (prev)
470 		prev->next = skb;
471 	else
472 		qp->q.fragments = skb;
473 
474 	dev = skb->dev;
475 	if (dev) {
476 		qp->iif = dev->ifindex;
477 		skb->dev = NULL;
478 	}
479 	qp->q.stamp = skb->tstamp;
480 	qp->q.meat += skb->len;
481 	qp->ecn |= ecn;
482 	add_frag_mem_limit(qp->q.net, skb->truesize);
483 	if (offset == 0)
484 		qp->q.flags |= INET_FRAG_FIRST_IN;
485 
486 	fragsize = skb->len + ihl;
487 
488 	if (fragsize > qp->q.max_size)
489 		qp->q.max_size = fragsize;
490 
491 	if (ip_hdr(skb)->frag_off & htons(IP_DF) &&
492 	    fragsize > qp->max_df_size)
493 		qp->max_df_size = fragsize;
494 
495 	if (qp->q.flags == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
496 	    qp->q.meat == qp->q.len) {
497 		unsigned long orefdst = skb->_skb_refdst;
498 
499 		skb->_skb_refdst = 0UL;
500 		err = ip_frag_reasm(qp, prev, dev);
501 		skb->_skb_refdst = orefdst;
502 		return err;
503 	}
504 
505 	skb_dst_drop(skb);
506 	return -EINPROGRESS;
507 
508 err:
509 	kfree_skb(skb);
510 	return err;
511 }
512 
513 
514 /* Build a new IP datagram from all its fragments. */
515 
516 static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev,
517 			 struct net_device *dev)
518 {
519 	struct net *net = container_of(qp->q.net, struct net, ipv4.frags);
520 	struct iphdr *iph;
521 	struct sk_buff *fp, *head = qp->q.fragments;
522 	int len;
523 	int ihlen;
524 	int err;
525 	u8 ecn;
526 
527 	ipq_kill(qp);
528 
529 	ecn = ip_frag_ecn_table[qp->ecn];
530 	if (unlikely(ecn == 0xff)) {
531 		err = -EINVAL;
532 		goto out_fail;
533 	}
534 	/* Make the one we just received the head. */
535 	if (prev) {
536 		head = prev->next;
537 		fp = skb_clone(head, GFP_ATOMIC);
538 		if (!fp)
539 			goto out_nomem;
540 
541 		fp->next = head->next;
542 		if (!fp->next)
543 			qp->q.fragments_tail = fp;
544 		prev->next = fp;
545 
546 		skb_morph(head, qp->q.fragments);
547 		head->next = qp->q.fragments->next;
548 
549 		consume_skb(qp->q.fragments);
550 		qp->q.fragments = head;
551 	}
552 
553 	WARN_ON(!head);
554 	WARN_ON(FRAG_CB(head)->offset != 0);
555 
556 	/* Allocate a new buffer for the datagram. */
557 	ihlen = ip_hdrlen(head);
558 	len = ihlen + qp->q.len;
559 
560 	err = -E2BIG;
561 	if (len > 65535)
562 		goto out_oversize;
563 
564 	/* Head of list must not be cloned. */
565 	if (skb_unclone(head, GFP_ATOMIC))
566 		goto out_nomem;
567 
568 	/* If the first fragment is fragmented itself, we split
569 	 * it to two chunks: the first with data and paged part
570 	 * and the second, holding only fragments. */
571 	if (skb_has_frag_list(head)) {
572 		struct sk_buff *clone;
573 		int i, plen = 0;
574 
575 		clone = alloc_skb(0, GFP_ATOMIC);
576 		if (!clone)
577 			goto out_nomem;
578 		clone->next = head->next;
579 		head->next = clone;
580 		skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
581 		skb_frag_list_init(head);
582 		for (i = 0; i < skb_shinfo(head)->nr_frags; i++)
583 			plen += skb_frag_size(&skb_shinfo(head)->frags[i]);
584 		clone->len = clone->data_len = head->data_len - plen;
585 		head->data_len -= clone->len;
586 		head->len -= clone->len;
587 		clone->csum = 0;
588 		clone->ip_summed = head->ip_summed;
589 		add_frag_mem_limit(qp->q.net, clone->truesize);
590 	}
591 
592 	skb_shinfo(head)->frag_list = head->next;
593 	skb_push(head, head->data - skb_network_header(head));
594 
595 	for (fp=head->next; fp; fp = fp->next) {
596 		head->data_len += fp->len;
597 		head->len += fp->len;
598 		if (head->ip_summed != fp->ip_summed)
599 			head->ip_summed = CHECKSUM_NONE;
600 		else if (head->ip_summed == CHECKSUM_COMPLETE)
601 			head->csum = csum_add(head->csum, fp->csum);
602 		head->truesize += fp->truesize;
603 	}
604 	sub_frag_mem_limit(qp->q.net, head->truesize);
605 
606 	head->next = NULL;
607 	head->dev = dev;
608 	head->tstamp = qp->q.stamp;
609 	IPCB(head)->frag_max_size = max(qp->max_df_size, qp->q.max_size);
610 
611 	iph = ip_hdr(head);
612 	iph->tot_len = htons(len);
613 	iph->tos |= ecn;
614 
615 	/* When we set IP_DF on a refragmented skb we must also force a
616 	 * call to ip_fragment to avoid forwarding a DF-skb of size s while
617 	 * original sender only sent fragments of size f (where f < s).
618 	 *
619 	 * We only set DF/IPSKB_FRAG_PMTU if such DF fragment was the largest
620 	 * frag seen to avoid sending tiny DF-fragments in case skb was built
621 	 * from one very small df-fragment and one large non-df frag.
622 	 */
623 	if (qp->max_df_size == qp->q.max_size) {
624 		IPCB(head)->flags |= IPSKB_FRAG_PMTU;
625 		iph->frag_off = htons(IP_DF);
626 	} else {
627 		iph->frag_off = 0;
628 	}
629 
630 	ip_send_check(iph);
631 
632 	IP_INC_STATS_BH(net, IPSTATS_MIB_REASMOKS);
633 	qp->q.fragments = NULL;
634 	qp->q.fragments_tail = NULL;
635 	return 0;
636 
637 out_nomem:
638 	net_dbg_ratelimited("queue_glue: no memory for gluing queue %p\n", qp);
639 	err = -ENOMEM;
640 	goto out_fail;
641 out_oversize:
642 	net_info_ratelimited("Oversized IP packet from %pI4\n", &qp->saddr);
643 out_fail:
644 	IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
645 	return err;
646 }
647 
648 /* Process an incoming IP datagram fragment. */
649 int ip_defrag(struct sk_buff *skb, u32 user)
650 {
651 	struct ipq *qp;
652 	struct net *net;
653 
654 	net = skb->dev ? dev_net(skb->dev) : dev_net(skb_dst(skb)->dev);
655 	IP_INC_STATS_BH(net, IPSTATS_MIB_REASMREQDS);
656 
657 	/* Lookup (or create) queue header */
658 	qp = ip_find(net, ip_hdr(skb), user);
659 	if (qp) {
660 		int ret;
661 
662 		spin_lock(&qp->q.lock);
663 
664 		ret = ip_frag_queue(qp, skb);
665 
666 		spin_unlock(&qp->q.lock);
667 		ipq_put(qp);
668 		return ret;
669 	}
670 
671 	IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
672 	kfree_skb(skb);
673 	return -ENOMEM;
674 }
675 EXPORT_SYMBOL(ip_defrag);
676 
677 struct sk_buff *ip_check_defrag(struct sk_buff *skb, u32 user)
678 {
679 	struct iphdr iph;
680 	int netoff;
681 	u32 len;
682 
683 	if (skb->protocol != htons(ETH_P_IP))
684 		return skb;
685 
686 	netoff = skb_network_offset(skb);
687 
688 	if (skb_copy_bits(skb, netoff, &iph, sizeof(iph)) < 0)
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 < netoff + 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, netoff + iph.ihl * 4))
702 				return skb;
703 			if (pskb_trim_rcsum(skb, netoff + 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 }
713 EXPORT_SYMBOL(ip_check_defrag);
714 
715 #ifdef CONFIG_SYSCTL
716 static int zero;
717 
718 static 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_minmax,
725 		.extra1		= &init_net.ipv4.frags.low_thresh
726 	},
727 	{
728 		.procname	= "ipfrag_low_thresh",
729 		.data		= &init_net.ipv4.frags.low_thresh,
730 		.maxlen		= sizeof(int),
731 		.mode		= 0644,
732 		.proc_handler	= proc_dointvec_minmax,
733 		.extra1		= &zero,
734 		.extra2		= &init_net.ipv4.frags.high_thresh
735 	},
736 	{
737 		.procname	= "ipfrag_time",
738 		.data		= &init_net.ipv4.frags.timeout,
739 		.maxlen		= sizeof(int),
740 		.mode		= 0644,
741 		.proc_handler	= proc_dointvec_jiffies,
742 	},
743 	{ }
744 };
745 
746 /* secret interval has been deprecated */
747 static int ip4_frags_secret_interval_unused;
748 static struct ctl_table ip4_frags_ctl_table[] = {
749 	{
750 		.procname	= "ipfrag_secret_interval",
751 		.data		= &ip4_frags_secret_interval_unused,
752 		.maxlen		= sizeof(int),
753 		.mode		= 0644,
754 		.proc_handler	= proc_dointvec_jiffies,
755 	},
756 	{
757 		.procname	= "ipfrag_max_dist",
758 		.data		= &sysctl_ipfrag_max_dist,
759 		.maxlen		= sizeof(int),
760 		.mode		= 0644,
761 		.proc_handler	= proc_dointvec_minmax,
762 		.extra1		= &zero
763 	},
764 	{ }
765 };
766 
767 static int __net_init ip4_frags_ns_ctl_register(struct net *net)
768 {
769 	struct ctl_table *table;
770 	struct ctl_table_header *hdr;
771 
772 	table = ip4_frags_ns_ctl_table;
773 	if (!net_eq(net, &init_net)) {
774 		table = kmemdup(table, sizeof(ip4_frags_ns_ctl_table), GFP_KERNEL);
775 		if (!table)
776 			goto err_alloc;
777 
778 		table[0].data = &net->ipv4.frags.high_thresh;
779 		table[0].extra1 = &net->ipv4.frags.low_thresh;
780 		table[0].extra2 = &init_net.ipv4.frags.high_thresh;
781 		table[1].data = &net->ipv4.frags.low_thresh;
782 		table[1].extra2 = &net->ipv4.frags.high_thresh;
783 		table[2].data = &net->ipv4.frags.timeout;
784 
785 		/* Don't export sysctls to unprivileged users */
786 		if (net->user_ns != &init_user_ns)
787 			table[0].procname = NULL;
788 	}
789 
790 	hdr = register_net_sysctl(net, "net/ipv4", table);
791 	if (!hdr)
792 		goto err_reg;
793 
794 	net->ipv4.frags_hdr = hdr;
795 	return 0;
796 
797 err_reg:
798 	if (!net_eq(net, &init_net))
799 		kfree(table);
800 err_alloc:
801 	return -ENOMEM;
802 }
803 
804 static void __net_exit ip4_frags_ns_ctl_unregister(struct net *net)
805 {
806 	struct ctl_table *table;
807 
808 	table = net->ipv4.frags_hdr->ctl_table_arg;
809 	unregister_net_sysctl_table(net->ipv4.frags_hdr);
810 	kfree(table);
811 }
812 
813 static void __init ip4_frags_ctl_register(void)
814 {
815 	register_net_sysctl(&init_net, "net/ipv4", ip4_frags_ctl_table);
816 }
817 #else
818 static int ip4_frags_ns_ctl_register(struct net *net)
819 {
820 	return 0;
821 }
822 
823 static void ip4_frags_ns_ctl_unregister(struct net *net)
824 {
825 }
826 
827 static void __init ip4_frags_ctl_register(void)
828 {
829 }
830 #endif
831 
832 static int __net_init ipv4_frags_init_net(struct net *net)
833 {
834 	/* Fragment cache limits.
835 	 *
836 	 * The fragment memory accounting code, (tries to) account for
837 	 * the real memory usage, by measuring both the size of frag
838 	 * queue struct (inet_frag_queue (ipv4:ipq/ipv6:frag_queue))
839 	 * and the SKB's truesize.
840 	 *
841 	 * A 64K fragment consumes 129736 bytes (44*2944)+200
842 	 * (1500 truesize == 2944, sizeof(struct ipq) == 200)
843 	 *
844 	 * We will commit 4MB at one time. Should we cross that limit
845 	 * we will prune down to 3MB, making room for approx 8 big 64K
846 	 * fragments 8x128k.
847 	 */
848 	net->ipv4.frags.high_thresh = 4 * 1024 * 1024;
849 	net->ipv4.frags.low_thresh  = 3 * 1024 * 1024;
850 	/*
851 	 * Important NOTE! Fragment queue must be destroyed before MSL expires.
852 	 * RFC791 is wrong proposing to prolongate timer each fragment arrival
853 	 * by TTL.
854 	 */
855 	net->ipv4.frags.timeout = IP_FRAG_TIME;
856 
857 	inet_frags_init_net(&net->ipv4.frags);
858 
859 	return ip4_frags_ns_ctl_register(net);
860 }
861 
862 static void __net_exit ipv4_frags_exit_net(struct net *net)
863 {
864 	ip4_frags_ns_ctl_unregister(net);
865 	inet_frags_exit_net(&net->ipv4.frags, &ip4_frags);
866 }
867 
868 static struct pernet_operations ip4_frags_ops = {
869 	.init = ipv4_frags_init_net,
870 	.exit = ipv4_frags_exit_net,
871 };
872 
873 void __init ipfrag_init(void)
874 {
875 	ip4_frags_ctl_register();
876 	register_pernet_subsys(&ip4_frags_ops);
877 	ip4_frags.hashfn = ip4_hashfn;
878 	ip4_frags.constructor = ip4_frag_init;
879 	ip4_frags.destructor = ip4_frag_free;
880 	ip4_frags.skb_free = NULL;
881 	ip4_frags.qsize = sizeof(struct ipq);
882 	ip4_frags.match = ip4_frag_match;
883 	ip4_frags.frag_expire = ip_expire;
884 	ip4_frags.frags_cache_name = ip_frag_cache_name;
885 	if (inet_frags_init(&ip4_frags))
886 		panic("IP: failed to allocate ip4_frags cache\n");
887 }
888