xref: /openbmc/linux/net/ipv4/ip_fragment.c (revision 60772e48)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * INET		An implementation of the TCP/IP protocol suite for the LINUX
4  *		operating system.  INET is implemented using the  BSD Socket
5  *		interface as the means of communication with the user level.
6  *
7  *		The IP fragmentation functionality.
8  *
9  * Authors:	Fred N. van Kempen <waltje@uWalt.NL.Mugnet.ORG>
10  *		Alan Cox <alan@lxorguk.ukuu.org.uk>
11  *
12  * Fixes:
13  *		Alan Cox	:	Split from ip.c , see ip_input.c for history.
14  *		David S. Miller :	Begin massive cleanup...
15  *		Andi Kleen	:	Add sysctls.
16  *		xxxx		:	Overlapfrag bug.
17  *		Ultima          :       ip_expire() kernel panic.
18  *		Bill Hawes	:	Frag accounting and evictor fixes.
19  *		John McDonald	:	0 length frag bug.
20  *		Alexey Kuznetsov:	SMP races, threading, cleanup.
21  *		Patrick McHardy :	LRU queue of frag heads for evictor.
22  */
23 
24 #define pr_fmt(fmt) "IPv4: " fmt
25 
26 #include <linux/compiler.h>
27 #include <linux/module.h>
28 #include <linux/types.h>
29 #include <linux/mm.h>
30 #include <linux/jiffies.h>
31 #include <linux/skbuff.h>
32 #include <linux/list.h>
33 #include <linux/ip.h>
34 #include <linux/icmp.h>
35 #include <linux/netdevice.h>
36 #include <linux/jhash.h>
37 #include <linux/random.h>
38 #include <linux/slab.h>
39 #include <net/route.h>
40 #include <net/dst.h>
41 #include <net/sock.h>
42 #include <net/ip.h>
43 #include <net/icmp.h>
44 #include <net/checksum.h>
45 #include <net/inetpeer.h>
46 #include <net/inet_frag.h>
47 #include <linux/tcp.h>
48 #include <linux/udp.h>
49 #include <linux/inet.h>
50 #include <linux/netfilter_ipv4.h>
51 #include <net/inet_ecn.h>
52 #include <net/l3mdev.h>
53 
54 /* NOTE. Logic of IP defragmentation is parallel to corresponding IPv6
55  * code now. If you change something here, _PLEASE_ update ipv6/reassembly.c
56  * as well. Or notify me, at least. --ANK
57  */
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 	int             vif;   /* L3 master device index */
81 	unsigned int    rid;
82 	struct inet_peer *peer;
83 };
84 
85 static u8 ip4_frag_ecn(u8 tos)
86 {
87 	return 1 << (tos & INET_ECN_MASK);
88 }
89 
90 static struct inet_frags ip4_frags;
91 
92 int ip_frag_mem(struct net *net)
93 {
94 	return sum_frag_mem_limit(&net->ipv4.frags);
95 }
96 
97 static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev,
98 			 struct net_device *dev);
99 
100 struct ip4_create_arg {
101 	struct iphdr *iph;
102 	u32 user;
103 	int vif;
104 };
105 
106 static unsigned int ipqhashfn(__be16 id, __be32 saddr, __be32 daddr, u8 prot)
107 {
108 	net_get_random_once(&ip4_frags.rnd, sizeof(ip4_frags.rnd));
109 	return jhash_3words((__force u32)id << 16 | prot,
110 			    (__force u32)saddr, (__force u32)daddr,
111 			    ip4_frags.rnd);
112 }
113 
114 static unsigned int ip4_hashfn(const struct inet_frag_queue *q)
115 {
116 	const struct ipq *ipq;
117 
118 	ipq = container_of(q, struct ipq, q);
119 	return ipqhashfn(ipq->id, ipq->saddr, ipq->daddr, ipq->protocol);
120 }
121 
122 static bool ip4_frag_match(const struct inet_frag_queue *q, const void *a)
123 {
124 	const struct ipq *qp;
125 	const struct ip4_create_arg *arg = a;
126 
127 	qp = container_of(q, struct ipq, q);
128 	return	qp->id == arg->iph->id &&
129 		qp->saddr == arg->iph->saddr &&
130 		qp->daddr == arg->iph->daddr &&
131 		qp->protocol == arg->iph->protocol &&
132 		qp->user == arg->user &&
133 		qp->vif == arg->vif;
134 }
135 
136 static void ip4_frag_init(struct inet_frag_queue *q, const 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 	const 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->vif = arg->vif;
151 	qp->user = arg->user;
152 	qp->peer = q->net->max_dist ?
153 		inet_getpeer_v4(net->ipv4.peers, arg->iph->saddr, arg->vif, 1) :
154 		NULL;
155 }
156 
157 static void ip4_frag_free(struct inet_frag_queue *q)
158 {
159 	struct ipq *qp;
160 
161 	qp = container_of(q, struct ipq, q);
162 	if (qp->peer)
163 		inet_putpeer(qp->peer);
164 }
165 
166 
167 /* Destruction primitives. */
168 
169 static void ipq_put(struct ipq *ipq)
170 {
171 	inet_frag_put(&ipq->q, &ip4_frags);
172 }
173 
174 /* Kill ipq entry. It is not destroyed immediately,
175  * because caller (and someone more) holds reference count.
176  */
177 static void ipq_kill(struct ipq *ipq)
178 {
179 	inet_frag_kill(&ipq->q, &ip4_frags);
180 }
181 
182 static bool frag_expire_skip_icmp(u32 user)
183 {
184 	return user == IP_DEFRAG_AF_PACKET ||
185 	       ip_defrag_user_in_between(user, IP_DEFRAG_CONNTRACK_IN,
186 					 __IP_DEFRAG_CONNTRACK_IN_END) ||
187 	       ip_defrag_user_in_between(user, IP_DEFRAG_CONNTRACK_BRIDGE_IN,
188 					 __IP_DEFRAG_CONNTRACK_BRIDGE_IN);
189 }
190 
191 /*
192  * Oops, a fragment queue timed out.  Kill it and send an ICMP reply.
193  */
194 static void ip_expire(struct timer_list *t)
195 {
196 	struct inet_frag_queue *frag = from_timer(frag, t, timer);
197 	struct ipq *qp;
198 	struct net *net;
199 
200 	qp = container_of(frag, struct ipq, q);
201 	net = container_of(qp->q.net, struct net, ipv4.frags);
202 
203 	rcu_read_lock();
204 	spin_lock(&qp->q.lock);
205 
206 	if (qp->q.flags & INET_FRAG_COMPLETE)
207 		goto out;
208 
209 	ipq_kill(qp);
210 	__IP_INC_STATS(net, IPSTATS_MIB_REASMFAILS);
211 
212 	if (!inet_frag_evicting(&qp->q)) {
213 		struct sk_buff *clone, *head = qp->q.fragments;
214 		const struct iphdr *iph;
215 		int err;
216 
217 		__IP_INC_STATS(net, IPSTATS_MIB_REASMTIMEOUT);
218 
219 		if (!(qp->q.flags & INET_FRAG_FIRST_IN) || !qp->q.fragments)
220 			goto out;
221 
222 		head->dev = dev_get_by_index_rcu(net, qp->iif);
223 		if (!head->dev)
224 			goto out;
225 
226 
227 		/* skb has no dst, perform route lookup again */
228 		iph = ip_hdr(head);
229 		err = ip_route_input_noref(head, iph->daddr, iph->saddr,
230 					   iph->tos, head->dev);
231 		if (err)
232 			goto out;
233 
234 		/* Only an end host needs to send an ICMP
235 		 * "Fragment Reassembly Timeout" message, per RFC792.
236 		 */
237 		if (frag_expire_skip_icmp(qp->user) &&
238 		    (skb_rtable(head)->rt_type != RTN_LOCAL))
239 			goto out;
240 
241 		clone = skb_clone(head, GFP_ATOMIC);
242 
243 		/* Send an ICMP "Fragment Reassembly Timeout" message. */
244 		if (clone) {
245 			spin_unlock(&qp->q.lock);
246 			icmp_send(clone, ICMP_TIME_EXCEEDED,
247 				  ICMP_EXC_FRAGTIME, 0);
248 			consume_skb(clone);
249 			goto out_rcu_unlock;
250 		}
251 	}
252 out:
253 	spin_unlock(&qp->q.lock);
254 out_rcu_unlock:
255 	rcu_read_unlock();
256 	ipq_put(qp);
257 }
258 
259 /* Find the correct entry in the "incomplete datagrams" queue for
260  * this IP datagram, and create new one, if nothing is found.
261  */
262 static struct ipq *ip_find(struct net *net, struct iphdr *iph,
263 			   u32 user, int vif)
264 {
265 	struct inet_frag_queue *q;
266 	struct ip4_create_arg arg;
267 	unsigned int hash;
268 
269 	arg.iph = iph;
270 	arg.user = user;
271 	arg.vif = vif;
272 
273 	hash = ipqhashfn(iph->id, iph->saddr, iph->daddr, iph->protocol);
274 
275 	q = inet_frag_find(&net->ipv4.frags, &ip4_frags, &arg, hash);
276 	if (IS_ERR_OR_NULL(q)) {
277 		inet_frag_maybe_warn_overflow(q, pr_fmt());
278 		return NULL;
279 	}
280 	return container_of(q, struct ipq, q);
281 }
282 
283 /* Is the fragment too far ahead to be part of ipq? */
284 static int ip_frag_too_far(struct ipq *qp)
285 {
286 	struct inet_peer *peer = qp->peer;
287 	unsigned int max = qp->q.net->max_dist;
288 	unsigned int start, end;
289 
290 	int rc;
291 
292 	if (!peer || !max)
293 		return 0;
294 
295 	start = qp->rid;
296 	end = atomic_inc_return(&peer->rid);
297 	qp->rid = end;
298 
299 	rc = qp->q.fragments && (end - start) > max;
300 
301 	if (rc) {
302 		struct net *net;
303 
304 		net = container_of(qp->q.net, struct net, ipv4.frags);
305 		__IP_INC_STATS(net, IPSTATS_MIB_REASMFAILS);
306 	}
307 
308 	return rc;
309 }
310 
311 static int ip_frag_reinit(struct ipq *qp)
312 {
313 	struct sk_buff *fp;
314 	unsigned int sum_truesize = 0;
315 
316 	if (!mod_timer(&qp->q.timer, jiffies + qp->q.net->timeout)) {
317 		refcount_inc(&qp->q.refcnt);
318 		return -ETIMEDOUT;
319 	}
320 
321 	fp = qp->q.fragments;
322 	do {
323 		struct sk_buff *xp = fp->next;
324 
325 		sum_truesize += fp->truesize;
326 		kfree_skb(fp);
327 		fp = xp;
328 	} while (fp);
329 	sub_frag_mem_limit(qp->q.net, sum_truesize);
330 
331 	qp->q.flags = 0;
332 	qp->q.len = 0;
333 	qp->q.meat = 0;
334 	qp->q.fragments = NULL;
335 	qp->q.fragments_tail = NULL;
336 	qp->iif = 0;
337 	qp->ecn = 0;
338 
339 	return 0;
340 }
341 
342 /* Add new segment to existing queue. */
343 static int ip_frag_queue(struct ipq *qp, struct sk_buff *skb)
344 {
345 	struct sk_buff *prev, *next;
346 	struct net_device *dev;
347 	unsigned int fragsize;
348 	int flags, offset;
349 	int ihl, end;
350 	int err = -ENOENT;
351 	u8 ecn;
352 
353 	if (qp->q.flags & INET_FRAG_COMPLETE)
354 		goto err;
355 
356 	if (!(IPCB(skb)->flags & IPSKB_FRAG_COMPLETE) &&
357 	    unlikely(ip_frag_too_far(qp)) &&
358 	    unlikely(err = ip_frag_reinit(qp))) {
359 		ipq_kill(qp);
360 		goto err;
361 	}
362 
363 	ecn = ip4_frag_ecn(ip_hdr(skb)->tos);
364 	offset = ntohs(ip_hdr(skb)->frag_off);
365 	flags = offset & ~IP_OFFSET;
366 	offset &= IP_OFFSET;
367 	offset <<= 3;		/* offset is in 8-byte chunks */
368 	ihl = ip_hdrlen(skb);
369 
370 	/* Determine the position of this fragment. */
371 	end = offset + skb->len - skb_network_offset(skb) - ihl;
372 	err = -EINVAL;
373 
374 	/* Is this the final fragment? */
375 	if ((flags & IP_MF) == 0) {
376 		/* If we already have some bits beyond end
377 		 * or have different end, the segment is corrupted.
378 		 */
379 		if (end < qp->q.len ||
380 		    ((qp->q.flags & INET_FRAG_LAST_IN) && end != qp->q.len))
381 			goto err;
382 		qp->q.flags |= INET_FRAG_LAST_IN;
383 		qp->q.len = end;
384 	} else {
385 		if (end&7) {
386 			end &= ~7;
387 			if (skb->ip_summed != CHECKSUM_UNNECESSARY)
388 				skb->ip_summed = CHECKSUM_NONE;
389 		}
390 		if (end > qp->q.len) {
391 			/* Some bits beyond end -> corruption. */
392 			if (qp->q.flags & INET_FRAG_LAST_IN)
393 				goto err;
394 			qp->q.len = end;
395 		}
396 	}
397 	if (end == offset)
398 		goto err;
399 
400 	err = -ENOMEM;
401 	if (!pskb_pull(skb, skb_network_offset(skb) + ihl))
402 		goto err;
403 
404 	err = pskb_trim_rcsum(skb, end - offset);
405 	if (err)
406 		goto err;
407 
408 	/* Find out which fragments are in front and at the back of us
409 	 * in the chain of fragments so far.  We must know where to put
410 	 * this fragment, right?
411 	 */
412 	prev = qp->q.fragments_tail;
413 	if (!prev || FRAG_CB(prev)->offset < offset) {
414 		next = NULL;
415 		goto found;
416 	}
417 	prev = NULL;
418 	for (next = qp->q.fragments; next != NULL; next = next->next) {
419 		if (FRAG_CB(next)->offset >= offset)
420 			break;	/* bingo! */
421 		prev = next;
422 	}
423 
424 found:
425 	/* We found where to put this one.  Check for overlap with
426 	 * preceding fragment, and, if needed, align things so that
427 	 * any overlaps are eliminated.
428 	 */
429 	if (prev) {
430 		int i = (FRAG_CB(prev)->offset + prev->len) - offset;
431 
432 		if (i > 0) {
433 			offset += i;
434 			err = -EINVAL;
435 			if (end <= offset)
436 				goto err;
437 			err = -ENOMEM;
438 			if (!pskb_pull(skb, i))
439 				goto err;
440 			if (skb->ip_summed != CHECKSUM_UNNECESSARY)
441 				skb->ip_summed = CHECKSUM_NONE;
442 		}
443 	}
444 
445 	err = -ENOMEM;
446 
447 	while (next && FRAG_CB(next)->offset < end) {
448 		int i = end - FRAG_CB(next)->offset; /* overlap is 'i' bytes */
449 
450 		if (i < next->len) {
451 			/* Eat head of the next overlapped fragment
452 			 * and leave the loop. The next ones cannot overlap.
453 			 */
454 			if (!pskb_pull(next, i))
455 				goto err;
456 			FRAG_CB(next)->offset += i;
457 			qp->q.meat -= i;
458 			if (next->ip_summed != CHECKSUM_UNNECESSARY)
459 				next->ip_summed = CHECKSUM_NONE;
460 			break;
461 		} else {
462 			struct sk_buff *free_it = next;
463 
464 			/* Old fragment is completely overridden with
465 			 * new one drop it.
466 			 */
467 			next = next->next;
468 
469 			if (prev)
470 				prev->next = next;
471 			else
472 				qp->q.fragments = next;
473 
474 			qp->q.meat -= free_it->len;
475 			sub_frag_mem_limit(qp->q.net, free_it->truesize);
476 			kfree_skb(free_it);
477 		}
478 	}
479 
480 	FRAG_CB(skb)->offset = offset;
481 
482 	/* Insert this fragment in the chain of fragments. */
483 	skb->next = next;
484 	if (!next)
485 		qp->q.fragments_tail = skb;
486 	if (prev)
487 		prev->next = skb;
488 	else
489 		qp->q.fragments = skb;
490 
491 	dev = skb->dev;
492 	if (dev) {
493 		qp->iif = dev->ifindex;
494 		skb->dev = NULL;
495 	}
496 	qp->q.stamp = skb->tstamp;
497 	qp->q.meat += skb->len;
498 	qp->ecn |= ecn;
499 	add_frag_mem_limit(qp->q.net, skb->truesize);
500 	if (offset == 0)
501 		qp->q.flags |= INET_FRAG_FIRST_IN;
502 
503 	fragsize = skb->len + ihl;
504 
505 	if (fragsize > qp->q.max_size)
506 		qp->q.max_size = fragsize;
507 
508 	if (ip_hdr(skb)->frag_off & htons(IP_DF) &&
509 	    fragsize > qp->max_df_size)
510 		qp->max_df_size = fragsize;
511 
512 	if (qp->q.flags == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
513 	    qp->q.meat == qp->q.len) {
514 		unsigned long orefdst = skb->_skb_refdst;
515 
516 		skb->_skb_refdst = 0UL;
517 		err = ip_frag_reasm(qp, prev, dev);
518 		skb->_skb_refdst = orefdst;
519 		return err;
520 	}
521 
522 	skb_dst_drop(skb);
523 	return -EINPROGRESS;
524 
525 err:
526 	kfree_skb(skb);
527 	return err;
528 }
529 
530 
531 /* Build a new IP datagram from all its fragments. */
532 
533 static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev,
534 			 struct net_device *dev)
535 {
536 	struct net *net = container_of(qp->q.net, struct net, ipv4.frags);
537 	struct iphdr *iph;
538 	struct sk_buff *fp, *head = qp->q.fragments;
539 	int len;
540 	int ihlen;
541 	int err;
542 	u8 ecn;
543 
544 	ipq_kill(qp);
545 
546 	ecn = ip_frag_ecn_table[qp->ecn];
547 	if (unlikely(ecn == 0xff)) {
548 		err = -EINVAL;
549 		goto out_fail;
550 	}
551 	/* Make the one we just received the head. */
552 	if (prev) {
553 		head = prev->next;
554 		fp = skb_clone(head, GFP_ATOMIC);
555 		if (!fp)
556 			goto out_nomem;
557 
558 		fp->next = head->next;
559 		if (!fp->next)
560 			qp->q.fragments_tail = fp;
561 		prev->next = fp;
562 
563 		skb_morph(head, qp->q.fragments);
564 		head->next = qp->q.fragments->next;
565 
566 		consume_skb(qp->q.fragments);
567 		qp->q.fragments = head;
568 	}
569 
570 	WARN_ON(!head);
571 	WARN_ON(FRAG_CB(head)->offset != 0);
572 
573 	/* Allocate a new buffer for the datagram. */
574 	ihlen = ip_hdrlen(head);
575 	len = ihlen + qp->q.len;
576 
577 	err = -E2BIG;
578 	if (len > 65535)
579 		goto out_oversize;
580 
581 	/* Head of list must not be cloned. */
582 	if (skb_unclone(head, GFP_ATOMIC))
583 		goto out_nomem;
584 
585 	/* If the first fragment is fragmented itself, we split
586 	 * it to two chunks: the first with data and paged part
587 	 * and the second, holding only fragments. */
588 	if (skb_has_frag_list(head)) {
589 		struct sk_buff *clone;
590 		int i, plen = 0;
591 
592 		clone = alloc_skb(0, GFP_ATOMIC);
593 		if (!clone)
594 			goto out_nomem;
595 		clone->next = head->next;
596 		head->next = clone;
597 		skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
598 		skb_frag_list_init(head);
599 		for (i = 0; i < skb_shinfo(head)->nr_frags; i++)
600 			plen += skb_frag_size(&skb_shinfo(head)->frags[i]);
601 		clone->len = clone->data_len = head->data_len - plen;
602 		head->data_len -= clone->len;
603 		head->len -= clone->len;
604 		clone->csum = 0;
605 		clone->ip_summed = head->ip_summed;
606 		add_frag_mem_limit(qp->q.net, clone->truesize);
607 	}
608 
609 	skb_shinfo(head)->frag_list = head->next;
610 	skb_push(head, head->data - skb_network_header(head));
611 
612 	for (fp=head->next; fp; fp = fp->next) {
613 		head->data_len += fp->len;
614 		head->len += fp->len;
615 		if (head->ip_summed != fp->ip_summed)
616 			head->ip_summed = CHECKSUM_NONE;
617 		else if (head->ip_summed == CHECKSUM_COMPLETE)
618 			head->csum = csum_add(head->csum, fp->csum);
619 		head->truesize += fp->truesize;
620 	}
621 	sub_frag_mem_limit(qp->q.net, head->truesize);
622 
623 	head->next = NULL;
624 	head->dev = dev;
625 	head->tstamp = qp->q.stamp;
626 	IPCB(head)->frag_max_size = max(qp->max_df_size, qp->q.max_size);
627 
628 	iph = ip_hdr(head);
629 	iph->tot_len = htons(len);
630 	iph->tos |= ecn;
631 
632 	/* When we set IP_DF on a refragmented skb we must also force a
633 	 * call to ip_fragment to avoid forwarding a DF-skb of size s while
634 	 * original sender only sent fragments of size f (where f < s).
635 	 *
636 	 * We only set DF/IPSKB_FRAG_PMTU if such DF fragment was the largest
637 	 * frag seen to avoid sending tiny DF-fragments in case skb was built
638 	 * from one very small df-fragment and one large non-df frag.
639 	 */
640 	if (qp->max_df_size == qp->q.max_size) {
641 		IPCB(head)->flags |= IPSKB_FRAG_PMTU;
642 		iph->frag_off = htons(IP_DF);
643 	} else {
644 		iph->frag_off = 0;
645 	}
646 
647 	ip_send_check(iph);
648 
649 	__IP_INC_STATS(net, IPSTATS_MIB_REASMOKS);
650 	qp->q.fragments = NULL;
651 	qp->q.fragments_tail = NULL;
652 	return 0;
653 
654 out_nomem:
655 	net_dbg_ratelimited("queue_glue: no memory for gluing queue %p\n", qp);
656 	err = -ENOMEM;
657 	goto out_fail;
658 out_oversize:
659 	net_info_ratelimited("Oversized IP packet from %pI4\n", &qp->saddr);
660 out_fail:
661 	__IP_INC_STATS(net, IPSTATS_MIB_REASMFAILS);
662 	return err;
663 }
664 
665 /* Process an incoming IP datagram fragment. */
666 int ip_defrag(struct net *net, struct sk_buff *skb, u32 user)
667 {
668 	struct net_device *dev = skb->dev ? : skb_dst(skb)->dev;
669 	int vif = l3mdev_master_ifindex_rcu(dev);
670 	struct ipq *qp;
671 
672 	__IP_INC_STATS(net, IPSTATS_MIB_REASMREQDS);
673 	skb_orphan(skb);
674 
675 	/* Lookup (or create) queue header */
676 	qp = ip_find(net, ip_hdr(skb), user, vif);
677 	if (qp) {
678 		int ret;
679 
680 		spin_lock(&qp->q.lock);
681 
682 		ret = ip_frag_queue(qp, skb);
683 
684 		spin_unlock(&qp->q.lock);
685 		ipq_put(qp);
686 		return ret;
687 	}
688 
689 	__IP_INC_STATS(net, IPSTATS_MIB_REASMFAILS);
690 	kfree_skb(skb);
691 	return -ENOMEM;
692 }
693 EXPORT_SYMBOL(ip_defrag);
694 
695 struct sk_buff *ip_check_defrag(struct net *net, struct sk_buff *skb, u32 user)
696 {
697 	struct iphdr iph;
698 	int netoff;
699 	u32 len;
700 
701 	if (skb->protocol != htons(ETH_P_IP))
702 		return skb;
703 
704 	netoff = skb_network_offset(skb);
705 
706 	if (skb_copy_bits(skb, netoff, &iph, sizeof(iph)) < 0)
707 		return skb;
708 
709 	if (iph.ihl < 5 || iph.version != 4)
710 		return skb;
711 
712 	len = ntohs(iph.tot_len);
713 	if (skb->len < netoff + len || len < (iph.ihl * 4))
714 		return skb;
715 
716 	if (ip_is_fragment(&iph)) {
717 		skb = skb_share_check(skb, GFP_ATOMIC);
718 		if (skb) {
719 			if (!pskb_may_pull(skb, netoff + iph.ihl * 4))
720 				return skb;
721 			if (pskb_trim_rcsum(skb, netoff + len))
722 				return skb;
723 			memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
724 			if (ip_defrag(net, skb, user))
725 				return NULL;
726 			skb_clear_hash(skb);
727 		}
728 	}
729 	return skb;
730 }
731 EXPORT_SYMBOL(ip_check_defrag);
732 
733 #ifdef CONFIG_SYSCTL
734 static int zero;
735 
736 static struct ctl_table ip4_frags_ns_ctl_table[] = {
737 	{
738 		.procname	= "ipfrag_high_thresh",
739 		.data		= &init_net.ipv4.frags.high_thresh,
740 		.maxlen		= sizeof(int),
741 		.mode		= 0644,
742 		.proc_handler	= proc_dointvec_minmax,
743 		.extra1		= &init_net.ipv4.frags.low_thresh
744 	},
745 	{
746 		.procname	= "ipfrag_low_thresh",
747 		.data		= &init_net.ipv4.frags.low_thresh,
748 		.maxlen		= sizeof(int),
749 		.mode		= 0644,
750 		.proc_handler	= proc_dointvec_minmax,
751 		.extra1		= &zero,
752 		.extra2		= &init_net.ipv4.frags.high_thresh
753 	},
754 	{
755 		.procname	= "ipfrag_time",
756 		.data		= &init_net.ipv4.frags.timeout,
757 		.maxlen		= sizeof(int),
758 		.mode		= 0644,
759 		.proc_handler	= proc_dointvec_jiffies,
760 	},
761 	{
762 		.procname	= "ipfrag_max_dist",
763 		.data		= &init_net.ipv4.frags.max_dist,
764 		.maxlen		= sizeof(int),
765 		.mode		= 0644,
766 		.proc_handler	= proc_dointvec_minmax,
767 		.extra1		= &zero
768 	},
769 	{ }
770 };
771 
772 /* secret interval has been deprecated */
773 static int ip4_frags_secret_interval_unused;
774 static struct ctl_table ip4_frags_ctl_table[] = {
775 	{
776 		.procname	= "ipfrag_secret_interval",
777 		.data		= &ip4_frags_secret_interval_unused,
778 		.maxlen		= sizeof(int),
779 		.mode		= 0644,
780 		.proc_handler	= proc_dointvec_jiffies,
781 	},
782 	{ }
783 };
784 
785 static int __net_init ip4_frags_ns_ctl_register(struct net *net)
786 {
787 	struct ctl_table *table;
788 	struct ctl_table_header *hdr;
789 
790 	table = ip4_frags_ns_ctl_table;
791 	if (!net_eq(net, &init_net)) {
792 		table = kmemdup(table, sizeof(ip4_frags_ns_ctl_table), GFP_KERNEL);
793 		if (!table)
794 			goto err_alloc;
795 
796 		table[0].data = &net->ipv4.frags.high_thresh;
797 		table[0].extra1 = &net->ipv4.frags.low_thresh;
798 		table[0].extra2 = &init_net.ipv4.frags.high_thresh;
799 		table[1].data = &net->ipv4.frags.low_thresh;
800 		table[1].extra2 = &net->ipv4.frags.high_thresh;
801 		table[2].data = &net->ipv4.frags.timeout;
802 		table[3].data = &net->ipv4.frags.max_dist;
803 	}
804 
805 	hdr = register_net_sysctl(net, "net/ipv4", table);
806 	if (!hdr)
807 		goto err_reg;
808 
809 	net->ipv4.frags_hdr = hdr;
810 	return 0;
811 
812 err_reg:
813 	if (!net_eq(net, &init_net))
814 		kfree(table);
815 err_alloc:
816 	return -ENOMEM;
817 }
818 
819 static void __net_exit ip4_frags_ns_ctl_unregister(struct net *net)
820 {
821 	struct ctl_table *table;
822 
823 	table = net->ipv4.frags_hdr->ctl_table_arg;
824 	unregister_net_sysctl_table(net->ipv4.frags_hdr);
825 	kfree(table);
826 }
827 
828 static void __init ip4_frags_ctl_register(void)
829 {
830 	register_net_sysctl(&init_net, "net/ipv4", ip4_frags_ctl_table);
831 }
832 #else
833 static int ip4_frags_ns_ctl_register(struct net *net)
834 {
835 	return 0;
836 }
837 
838 static void ip4_frags_ns_ctl_unregister(struct net *net)
839 {
840 }
841 
842 static void __init ip4_frags_ctl_register(void)
843 {
844 }
845 #endif
846 
847 static int __net_init ipv4_frags_init_net(struct net *net)
848 {
849 	/* Fragment cache limits.
850 	 *
851 	 * The fragment memory accounting code, (tries to) account for
852 	 * the real memory usage, by measuring both the size of frag
853 	 * queue struct (inet_frag_queue (ipv4:ipq/ipv6:frag_queue))
854 	 * and the SKB's truesize.
855 	 *
856 	 * A 64K fragment consumes 129736 bytes (44*2944)+200
857 	 * (1500 truesize == 2944, sizeof(struct ipq) == 200)
858 	 *
859 	 * We will commit 4MB at one time. Should we cross that limit
860 	 * we will prune down to 3MB, making room for approx 8 big 64K
861 	 * fragments 8x128k.
862 	 */
863 	net->ipv4.frags.high_thresh = 4 * 1024 * 1024;
864 	net->ipv4.frags.low_thresh  = 3 * 1024 * 1024;
865 	/*
866 	 * Important NOTE! Fragment queue must be destroyed before MSL expires.
867 	 * RFC791 is wrong proposing to prolongate timer each fragment arrival
868 	 * by TTL.
869 	 */
870 	net->ipv4.frags.timeout = IP_FRAG_TIME;
871 
872 	net->ipv4.frags.max_dist = 64;
873 
874 	inet_frags_init_net(&net->ipv4.frags);
875 
876 	return ip4_frags_ns_ctl_register(net);
877 }
878 
879 static void __net_exit ipv4_frags_exit_net(struct net *net)
880 {
881 	ip4_frags_ns_ctl_unregister(net);
882 	inet_frags_exit_net(&net->ipv4.frags, &ip4_frags);
883 }
884 
885 static struct pernet_operations ip4_frags_ops = {
886 	.init = ipv4_frags_init_net,
887 	.exit = ipv4_frags_exit_net,
888 	.async = true,
889 };
890 
891 void __init ipfrag_init(void)
892 {
893 	ip4_frags_ctl_register();
894 	register_pernet_subsys(&ip4_frags_ops);
895 	ip4_frags.hashfn = ip4_hashfn;
896 	ip4_frags.constructor = ip4_frag_init;
897 	ip4_frags.destructor = ip4_frag_free;
898 	ip4_frags.qsize = sizeof(struct ipq);
899 	ip4_frags.match = ip4_frag_match;
900 	ip4_frags.frag_expire = ip_expire;
901 	ip4_frags.frags_cache_name = ip_frag_cache_name;
902 	if (inet_frags_init(&ip4_frags))
903 		panic("IP: failed to allocate ip4_frags cache\n");
904 }
905