xref: /openbmc/linux/net/ipv4/ip_fragment.c (revision 9cdb81c7)
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 
59 struct 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. */
68 struct 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 
82 /* RFC 3168 support :
83  * We want to check ECN values of all fragments, do detect invalid combinations.
84  * In ipq->ecn, we store the OR value of each ip4_frag_ecn() fragment value.
85  */
86 #define	IPFRAG_ECN_NOT_ECT	0x01 /* one frag had ECN_NOT_ECT */
87 #define	IPFRAG_ECN_ECT_1	0x02 /* one frag had ECN_ECT_1 */
88 #define	IPFRAG_ECN_ECT_0	0x04 /* one frag had ECN_ECT_0 */
89 #define	IPFRAG_ECN_CE		0x08 /* one frag had ECN_CE */
90 
91 static inline u8 ip4_frag_ecn(u8 tos)
92 {
93 	return 1 << (tos & INET_ECN_MASK);
94 }
95 
96 /* Given the OR values of all fragments, apply RFC 3168 5.3 requirements
97  * Value : 0xff if frame should be dropped.
98  *         0 or INET_ECN_CE value, to be ORed in to final iph->tos field
99  */
100 static const u8 ip4_frag_ecn_table[16] = {
101 	/* at least one fragment had CE, and others ECT_0 or ECT_1 */
102 	[IPFRAG_ECN_CE | IPFRAG_ECN_ECT_0]			= INET_ECN_CE,
103 	[IPFRAG_ECN_CE | IPFRAG_ECN_ECT_1]			= INET_ECN_CE,
104 	[IPFRAG_ECN_CE | IPFRAG_ECN_ECT_0 | IPFRAG_ECN_ECT_1]	= INET_ECN_CE,
105 
106 	/* invalid combinations : drop frame */
107 	[IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_CE] = 0xff,
108 	[IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_ECT_0] = 0xff,
109 	[IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_ECT_1] = 0xff,
110 	[IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_ECT_0 | IPFRAG_ECN_ECT_1] = 0xff,
111 	[IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_CE | IPFRAG_ECN_ECT_0] = 0xff,
112 	[IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_CE | IPFRAG_ECN_ECT_1] = 0xff,
113 	[IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_CE | IPFRAG_ECN_ECT_0 | IPFRAG_ECN_ECT_1] = 0xff,
114 };
115 
116 static struct inet_frags ip4_frags;
117 
118 int ip_frag_nqueues(struct net *net)
119 {
120 	return net->ipv4.frags.nqueues;
121 }
122 
123 int ip_frag_mem(struct net *net)
124 {
125 	return atomic_read(&net->ipv4.frags.mem);
126 }
127 
128 static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev,
129 			 struct net_device *dev);
130 
131 struct ip4_create_arg {
132 	struct iphdr *iph;
133 	u32 user;
134 };
135 
136 static unsigned int ipqhashfn(__be16 id, __be32 saddr, __be32 daddr, u8 prot)
137 {
138 	return jhash_3words((__force u32)id << 16 | prot,
139 			    (__force u32)saddr, (__force u32)daddr,
140 			    ip4_frags.rnd) & (INETFRAGS_HASHSZ - 1);
141 }
142 
143 static unsigned int ip4_hashfn(struct inet_frag_queue *q)
144 {
145 	struct ipq *ipq;
146 
147 	ipq = container_of(q, struct ipq, q);
148 	return ipqhashfn(ipq->id, ipq->saddr, ipq->daddr, ipq->protocol);
149 }
150 
151 static int ip4_frag_match(struct inet_frag_queue *q, void *a)
152 {
153 	struct ipq *qp;
154 	struct ip4_create_arg *arg = a;
155 
156 	qp = container_of(q, struct ipq, q);
157 	return	qp->id == arg->iph->id &&
158 			qp->saddr == arg->iph->saddr &&
159 			qp->daddr == arg->iph->daddr &&
160 			qp->protocol == arg->iph->protocol &&
161 			qp->user == arg->user;
162 }
163 
164 /* Memory Tracking Functions. */
165 static void frag_kfree_skb(struct netns_frags *nf, struct sk_buff *skb)
166 {
167 	atomic_sub(skb->truesize, &nf->mem);
168 	kfree_skb(skb);
169 }
170 
171 static void ip4_frag_init(struct inet_frag_queue *q, void *a)
172 {
173 	struct ipq *qp = container_of(q, struct ipq, q);
174 	struct ip4_create_arg *arg = a;
175 
176 	qp->protocol = arg->iph->protocol;
177 	qp->id = arg->iph->id;
178 	qp->ecn = ip4_frag_ecn(arg->iph->tos);
179 	qp->saddr = arg->iph->saddr;
180 	qp->daddr = arg->iph->daddr;
181 	qp->user = arg->user;
182 	qp->peer = sysctl_ipfrag_max_dist ?
183 		inet_getpeer_v4(arg->iph->saddr, 1) : NULL;
184 }
185 
186 static __inline__ void ip4_frag_free(struct inet_frag_queue *q)
187 {
188 	struct ipq *qp;
189 
190 	qp = container_of(q, struct ipq, q);
191 	if (qp->peer)
192 		inet_putpeer(qp->peer);
193 }
194 
195 
196 /* Destruction primitives. */
197 
198 static __inline__ void ipq_put(struct ipq *ipq)
199 {
200 	inet_frag_put(&ipq->q, &ip4_frags);
201 }
202 
203 /* Kill ipq entry. It is not destroyed immediately,
204  * because caller (and someone more) holds reference count.
205  */
206 static void ipq_kill(struct ipq *ipq)
207 {
208 	inet_frag_kill(&ipq->q, &ip4_frags);
209 }
210 
211 /* Memory limiting on fragments.  Evictor trashes the oldest
212  * fragment queue until we are back under the threshold.
213  */
214 static void ip_evictor(struct net *net)
215 {
216 	int evicted;
217 
218 	evicted = inet_frag_evictor(&net->ipv4.frags, &ip4_frags);
219 	if (evicted)
220 		IP_ADD_STATS_BH(net, IPSTATS_MIB_REASMFAILS, evicted);
221 }
222 
223 /*
224  * Oops, a fragment queue timed out.  Kill it and send an ICMP reply.
225  */
226 static void ip_expire(unsigned long arg)
227 {
228 	struct ipq *qp;
229 	struct net *net;
230 
231 	qp = container_of((struct inet_frag_queue *) arg, struct ipq, q);
232 	net = container_of(qp->q.net, struct net, ipv4.frags);
233 
234 	spin_lock(&qp->q.lock);
235 
236 	if (qp->q.last_in & INET_FRAG_COMPLETE)
237 		goto out;
238 
239 	ipq_kill(qp);
240 
241 	IP_INC_STATS_BH(net, IPSTATS_MIB_REASMTIMEOUT);
242 	IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
243 
244 	if ((qp->q.last_in & INET_FRAG_FIRST_IN) && qp->q.fragments != NULL) {
245 		struct sk_buff *head = qp->q.fragments;
246 		const struct iphdr *iph;
247 		int err;
248 
249 		rcu_read_lock();
250 		head->dev = dev_get_by_index_rcu(net, qp->iif);
251 		if (!head->dev)
252 			goto out_rcu_unlock;
253 
254 		/* skb dst is stale, drop it, and perform route lookup again */
255 		skb_dst_drop(head);
256 		iph = ip_hdr(head);
257 		err = ip_route_input_noref(head, iph->daddr, iph->saddr,
258 					   iph->tos, head->dev);
259 		if (err)
260 			goto out_rcu_unlock;
261 
262 		/*
263 		 * Only an end host needs to send an ICMP
264 		 * "Fragment Reassembly Timeout" message, per RFC792.
265 		 */
266 		if (qp->user == IP_DEFRAG_AF_PACKET ||
267 		    (qp->user == IP_DEFRAG_CONNTRACK_IN &&
268 		     skb_rtable(head)->rt_type != RTN_LOCAL))
269 			goto out_rcu_unlock;
270 
271 
272 		/* Send an ICMP "Fragment Reassembly Timeout" message. */
273 		icmp_send(head, ICMP_TIME_EXCEEDED, ICMP_EXC_FRAGTIME, 0);
274 out_rcu_unlock:
275 		rcu_read_unlock();
276 	}
277 out:
278 	spin_unlock(&qp->q.lock);
279 	ipq_put(qp);
280 }
281 
282 /* Find the correct entry in the "incomplete datagrams" queue for
283  * this IP datagram, and create new one, if nothing is found.
284  */
285 static inline struct ipq *ip_find(struct net *net, struct iphdr *iph, u32 user)
286 {
287 	struct inet_frag_queue *q;
288 	struct ip4_create_arg arg;
289 	unsigned int hash;
290 
291 	arg.iph = iph;
292 	arg.user = user;
293 
294 	read_lock(&ip4_frags.lock);
295 	hash = ipqhashfn(iph->id, iph->saddr, iph->daddr, iph->protocol);
296 
297 	q = inet_frag_find(&net->ipv4.frags, &ip4_frags, &arg, hash);
298 	if (q == NULL)
299 		goto out_nomem;
300 
301 	return container_of(q, struct ipq, q);
302 
303 out_nomem:
304 	LIMIT_NETDEBUG(KERN_ERR pr_fmt("ip_frag_create: no memory left !\n"));
305 	return NULL;
306 }
307 
308 /* Is the fragment too far ahead to be part of ipq? */
309 static inline int ip_frag_too_far(struct ipq *qp)
310 {
311 	struct inet_peer *peer = qp->peer;
312 	unsigned int max = sysctl_ipfrag_max_dist;
313 	unsigned int start, end;
314 
315 	int rc;
316 
317 	if (!peer || !max)
318 		return 0;
319 
320 	start = qp->rid;
321 	end = atomic_inc_return(&peer->rid);
322 	qp->rid = end;
323 
324 	rc = qp->q.fragments && (end - start) > max;
325 
326 	if (rc) {
327 		struct net *net;
328 
329 		net = container_of(qp->q.net, struct net, ipv4.frags);
330 		IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
331 	}
332 
333 	return rc;
334 }
335 
336 static int ip_frag_reinit(struct ipq *qp)
337 {
338 	struct sk_buff *fp;
339 
340 	if (!mod_timer(&qp->q.timer, jiffies + qp->q.net->timeout)) {
341 		atomic_inc(&qp->q.refcnt);
342 		return -ETIMEDOUT;
343 	}
344 
345 	fp = qp->q.fragments;
346 	do {
347 		struct sk_buff *xp = fp->next;
348 		frag_kfree_skb(qp->q.net, fp);
349 		fp = xp;
350 	} while (fp);
351 
352 	qp->q.last_in = 0;
353 	qp->q.len = 0;
354 	qp->q.meat = 0;
355 	qp->q.fragments = NULL;
356 	qp->q.fragments_tail = NULL;
357 	qp->iif = 0;
358 	qp->ecn = 0;
359 
360 	return 0;
361 }
362 
363 /* Add new segment to existing queue. */
364 static int ip_frag_queue(struct ipq *qp, struct sk_buff *skb)
365 {
366 	struct sk_buff *prev, *next;
367 	struct net_device *dev;
368 	int flags, offset;
369 	int ihl, end;
370 	int err = -ENOENT;
371 	u8 ecn;
372 
373 	if (qp->q.last_in & INET_FRAG_COMPLETE)
374 		goto err;
375 
376 	if (!(IPCB(skb)->flags & IPSKB_FRAG_COMPLETE) &&
377 	    unlikely(ip_frag_too_far(qp)) &&
378 	    unlikely(err = ip_frag_reinit(qp))) {
379 		ipq_kill(qp);
380 		goto err;
381 	}
382 
383 	ecn = ip4_frag_ecn(ip_hdr(skb)->tos);
384 	offset = ntohs(ip_hdr(skb)->frag_off);
385 	flags = offset & ~IP_OFFSET;
386 	offset &= IP_OFFSET;
387 	offset <<= 3;		/* offset is in 8-byte chunks */
388 	ihl = ip_hdrlen(skb);
389 
390 	/* Determine the position of this fragment. */
391 	end = offset + skb->len - ihl;
392 	err = -EINVAL;
393 
394 	/* Is this the final fragment? */
395 	if ((flags & IP_MF) == 0) {
396 		/* If we already have some bits beyond end
397 		 * or have different end, the segment is corrupted.
398 		 */
399 		if (end < qp->q.len ||
400 		    ((qp->q.last_in & INET_FRAG_LAST_IN) && end != qp->q.len))
401 			goto err;
402 		qp->q.last_in |= INET_FRAG_LAST_IN;
403 		qp->q.len = end;
404 	} else {
405 		if (end&7) {
406 			end &= ~7;
407 			if (skb->ip_summed != CHECKSUM_UNNECESSARY)
408 				skb->ip_summed = CHECKSUM_NONE;
409 		}
410 		if (end > qp->q.len) {
411 			/* Some bits beyond end -> corruption. */
412 			if (qp->q.last_in & INET_FRAG_LAST_IN)
413 				goto err;
414 			qp->q.len = end;
415 		}
416 	}
417 	if (end == offset)
418 		goto err;
419 
420 	err = -ENOMEM;
421 	if (pskb_pull(skb, ihl) == NULL)
422 		goto err;
423 
424 	err = pskb_trim_rcsum(skb, end - offset);
425 	if (err)
426 		goto err;
427 
428 	/* Find out which fragments are in front and at the back of us
429 	 * in the chain of fragments so far.  We must know where to put
430 	 * this fragment, right?
431 	 */
432 	prev = qp->q.fragments_tail;
433 	if (!prev || FRAG_CB(prev)->offset < offset) {
434 		next = NULL;
435 		goto found;
436 	}
437 	prev = NULL;
438 	for (next = qp->q.fragments; next != NULL; next = next->next) {
439 		if (FRAG_CB(next)->offset >= offset)
440 			break;	/* bingo! */
441 		prev = next;
442 	}
443 
444 found:
445 	/* We found where to put this one.  Check for overlap with
446 	 * preceding fragment, and, if needed, align things so that
447 	 * any overlaps are eliminated.
448 	 */
449 	if (prev) {
450 		int i = (FRAG_CB(prev)->offset + prev->len) - offset;
451 
452 		if (i > 0) {
453 			offset += i;
454 			err = -EINVAL;
455 			if (end <= offset)
456 				goto err;
457 			err = -ENOMEM;
458 			if (!pskb_pull(skb, i))
459 				goto err;
460 			if (skb->ip_summed != CHECKSUM_UNNECESSARY)
461 				skb->ip_summed = CHECKSUM_NONE;
462 		}
463 	}
464 
465 	err = -ENOMEM;
466 
467 	while (next && FRAG_CB(next)->offset < end) {
468 		int i = end - FRAG_CB(next)->offset; /* overlap is 'i' bytes */
469 
470 		if (i < next->len) {
471 			/* Eat head of the next overlapped fragment
472 			 * and leave the loop. The next ones cannot overlap.
473 			 */
474 			if (!pskb_pull(next, i))
475 				goto err;
476 			FRAG_CB(next)->offset += i;
477 			qp->q.meat -= i;
478 			if (next->ip_summed != CHECKSUM_UNNECESSARY)
479 				next->ip_summed = CHECKSUM_NONE;
480 			break;
481 		} else {
482 			struct sk_buff *free_it = next;
483 
484 			/* Old fragment is completely overridden with
485 			 * new one drop it.
486 			 */
487 			next = next->next;
488 
489 			if (prev)
490 				prev->next = next;
491 			else
492 				qp->q.fragments = next;
493 
494 			qp->q.meat -= free_it->len;
495 			frag_kfree_skb(qp->q.net, free_it);
496 		}
497 	}
498 
499 	FRAG_CB(skb)->offset = offset;
500 
501 	/* Insert this fragment in the chain of fragments. */
502 	skb->next = next;
503 	if (!next)
504 		qp->q.fragments_tail = skb;
505 	if (prev)
506 		prev->next = skb;
507 	else
508 		qp->q.fragments = skb;
509 
510 	dev = skb->dev;
511 	if (dev) {
512 		qp->iif = dev->ifindex;
513 		skb->dev = NULL;
514 	}
515 	qp->q.stamp = skb->tstamp;
516 	qp->q.meat += skb->len;
517 	qp->ecn |= ecn;
518 	atomic_add(skb->truesize, &qp->q.net->mem);
519 	if (offset == 0)
520 		qp->q.last_in |= INET_FRAG_FIRST_IN;
521 
522 	if (qp->q.last_in == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
523 	    qp->q.meat == qp->q.len)
524 		return ip_frag_reasm(qp, prev, dev);
525 
526 	write_lock(&ip4_frags.lock);
527 	list_move_tail(&qp->q.lru_list, &qp->q.net->lru_list);
528 	write_unlock(&ip4_frags.lock);
529 	return -EINPROGRESS;
530 
531 err:
532 	kfree_skb(skb);
533 	return err;
534 }
535 
536 
537 /* Build a new IP datagram from all its fragments. */
538 
539 static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev,
540 			 struct net_device *dev)
541 {
542 	struct net *net = container_of(qp->q.net, struct net, ipv4.frags);
543 	struct iphdr *iph;
544 	struct sk_buff *fp, *head = qp->q.fragments;
545 	int len;
546 	int ihlen;
547 	int err;
548 	u8 ecn;
549 
550 	ipq_kill(qp);
551 
552 	ecn = ip4_frag_ecn_table[qp->ecn];
553 	if (unlikely(ecn == 0xff)) {
554 		err = -EINVAL;
555 		goto out_fail;
556 	}
557 	/* Make the one we just received the head. */
558 	if (prev) {
559 		head = prev->next;
560 		fp = skb_clone(head, GFP_ATOMIC);
561 		if (!fp)
562 			goto out_nomem;
563 
564 		fp->next = head->next;
565 		if (!fp->next)
566 			qp->q.fragments_tail = fp;
567 		prev->next = fp;
568 
569 		skb_morph(head, qp->q.fragments);
570 		head->next = qp->q.fragments->next;
571 
572 		kfree_skb(qp->q.fragments);
573 		qp->q.fragments = head;
574 	}
575 
576 	WARN_ON(head == NULL);
577 	WARN_ON(FRAG_CB(head)->offset != 0);
578 
579 	/* Allocate a new buffer for the datagram. */
580 	ihlen = ip_hdrlen(head);
581 	len = ihlen + qp->q.len;
582 
583 	err = -E2BIG;
584 	if (len > 65535)
585 		goto out_oversize;
586 
587 	/* Head of list must not be cloned. */
588 	if (skb_cloned(head) && pskb_expand_head(head, 0, 0, GFP_ATOMIC))
589 		goto out_nomem;
590 
591 	/* If the first fragment is fragmented itself, we split
592 	 * it to two chunks: the first with data and paged part
593 	 * and the second, holding only fragments. */
594 	if (skb_has_frag_list(head)) {
595 		struct sk_buff *clone;
596 		int i, plen = 0;
597 
598 		if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL)
599 			goto out_nomem;
600 		clone->next = head->next;
601 		head->next = clone;
602 		skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
603 		skb_frag_list_init(head);
604 		for (i = 0; i < skb_shinfo(head)->nr_frags; i++)
605 			plen += skb_frag_size(&skb_shinfo(head)->frags[i]);
606 		clone->len = clone->data_len = head->data_len - plen;
607 		head->data_len -= clone->len;
608 		head->len -= clone->len;
609 		clone->csum = 0;
610 		clone->ip_summed = head->ip_summed;
611 		atomic_add(clone->truesize, &qp->q.net->mem);
612 	}
613 
614 	skb_shinfo(head)->frag_list = head->next;
615 	skb_push(head, head->data - skb_network_header(head));
616 
617 	for (fp=head->next; fp; fp = fp->next) {
618 		head->data_len += fp->len;
619 		head->len += fp->len;
620 		if (head->ip_summed != fp->ip_summed)
621 			head->ip_summed = CHECKSUM_NONE;
622 		else if (head->ip_summed == CHECKSUM_COMPLETE)
623 			head->csum = csum_add(head->csum, fp->csum);
624 		head->truesize += fp->truesize;
625 	}
626 	atomic_sub(head->truesize, &qp->q.net->mem);
627 
628 	head->next = NULL;
629 	head->dev = dev;
630 	head->tstamp = qp->q.stamp;
631 
632 	iph = ip_hdr(head);
633 	iph->frag_off = 0;
634 	iph->tot_len = htons(len);
635 	iph->tos |= ecn;
636 	IP_INC_STATS_BH(net, IPSTATS_MIB_REASMOKS);
637 	qp->q.fragments = NULL;
638 	qp->q.fragments_tail = NULL;
639 	return 0;
640 
641 out_nomem:
642 	LIMIT_NETDEBUG(KERN_ERR pr_fmt("queue_glue: no memory for gluing queue %p\n"),
643 		       qp);
644 	err = -ENOMEM;
645 	goto out_fail;
646 out_oversize:
647 	if (net_ratelimit())
648 		pr_info("Oversized IP packet from %pI4\n", &qp->saddr);
649 out_fail:
650 	IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
651 	return err;
652 }
653 
654 /* Process an incoming IP datagram fragment. */
655 int ip_defrag(struct sk_buff *skb, u32 user)
656 {
657 	struct ipq *qp;
658 	struct net *net;
659 
660 	net = skb->dev ? dev_net(skb->dev) : dev_net(skb_dst(skb)->dev);
661 	IP_INC_STATS_BH(net, IPSTATS_MIB_REASMREQDS);
662 
663 	/* Start by cleaning up the memory. */
664 	if (atomic_read(&net->ipv4.frags.mem) > net->ipv4.frags.high_thresh)
665 		ip_evictor(net);
666 
667 	/* Lookup (or create) queue header */
668 	if ((qp = ip_find(net, ip_hdr(skb), user)) != NULL) {
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 	const struct iphdr *iph;
689 	u32 len;
690 
691 	if (skb->protocol != htons(ETH_P_IP))
692 		return skb;
693 
694 	if (!pskb_may_pull(skb, sizeof(struct iphdr)))
695 		return skb;
696 
697 	iph = ip_hdr(skb);
698 	if (iph->ihl < 5 || iph->version != 4)
699 		return skb;
700 	if (!pskb_may_pull(skb, iph->ihl*4))
701 		return skb;
702 	iph = ip_hdr(skb);
703 	len = ntohs(iph->tot_len);
704 	if (skb->len < len || len < (iph->ihl * 4))
705 		return skb;
706 
707 	if (ip_is_fragment(ip_hdr(skb))) {
708 		skb = skb_share_check(skb, GFP_ATOMIC);
709 		if (skb) {
710 			if (pskb_trim_rcsum(skb, len))
711 				return skb;
712 			memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
713 			if (ip_defrag(skb, user))
714 				return NULL;
715 			skb->rxhash = 0;
716 		}
717 	}
718 	return skb;
719 }
720 EXPORT_SYMBOL(ip_check_defrag);
721 
722 #ifdef CONFIG_SYSCTL
723 static int zero;
724 
725 static 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
732 	},
733 	{
734 		.procname	= "ipfrag_low_thresh",
735 		.data		= &init_net.ipv4.frags.low_thresh,
736 		.maxlen		= sizeof(int),
737 		.mode		= 0644,
738 		.proc_handler	= proc_dointvec
739 	},
740 	{
741 		.procname	= "ipfrag_time",
742 		.data		= &init_net.ipv4.frags.timeout,
743 		.maxlen		= sizeof(int),
744 		.mode		= 0644,
745 		.proc_handler	= proc_dointvec_jiffies,
746 	},
747 	{ }
748 };
749 
750 static struct ctl_table ip4_frags_ctl_table[] = {
751 	{
752 		.procname	= "ipfrag_secret_interval",
753 		.data		= &ip4_frags.secret_interval,
754 		.maxlen		= sizeof(int),
755 		.mode		= 0644,
756 		.proc_handler	= proc_dointvec_jiffies,
757 	},
758 	{
759 		.procname	= "ipfrag_max_dist",
760 		.data		= &sysctl_ipfrag_max_dist,
761 		.maxlen		= sizeof(int),
762 		.mode		= 0644,
763 		.proc_handler	= proc_dointvec_minmax,
764 		.extra1		= &zero
765 	},
766 	{ }
767 };
768 
769 static int __net_init ip4_frags_ns_ctl_register(struct net *net)
770 {
771 	struct ctl_table *table;
772 	struct ctl_table_header *hdr;
773 
774 	table = ip4_frags_ns_ctl_table;
775 	if (!net_eq(net, &init_net)) {
776 		table = kmemdup(table, sizeof(ip4_frags_ns_ctl_table), GFP_KERNEL);
777 		if (table == NULL)
778 			goto err_alloc;
779 
780 		table[0].data = &net->ipv4.frags.high_thresh;
781 		table[1].data = &net->ipv4.frags.low_thresh;
782 		table[2].data = &net->ipv4.frags.timeout;
783 	}
784 
785 	hdr = register_net_sysctl_table(net, net_ipv4_ctl_path, table);
786 	if (hdr == NULL)
787 		goto err_reg;
788 
789 	net->ipv4.frags_hdr = hdr;
790 	return 0;
791 
792 err_reg:
793 	if (!net_eq(net, &init_net))
794 		kfree(table);
795 err_alloc:
796 	return -ENOMEM;
797 }
798 
799 static void __net_exit ip4_frags_ns_ctl_unregister(struct net *net)
800 {
801 	struct ctl_table *table;
802 
803 	table = net->ipv4.frags_hdr->ctl_table_arg;
804 	unregister_net_sysctl_table(net->ipv4.frags_hdr);
805 	kfree(table);
806 }
807 
808 static void ip4_frags_ctl_register(void)
809 {
810 	register_net_sysctl_rotable(net_ipv4_ctl_path, ip4_frags_ctl_table);
811 }
812 #else
813 static inline int ip4_frags_ns_ctl_register(struct net *net)
814 {
815 	return 0;
816 }
817 
818 static inline void ip4_frags_ns_ctl_unregister(struct net *net)
819 {
820 }
821 
822 static inline void ip4_frags_ctl_register(void)
823 {
824 }
825 #endif
826 
827 static int __net_init ipv4_frags_init_net(struct net *net)
828 {
829 	/*
830 	 * Fragment cache limits. We will commit 256K at one time. Should we
831 	 * cross that limit we will prune down to 192K. This should cope with
832 	 * even the most extreme cases without allowing an attacker to
833 	 * measurably harm machine performance.
834 	 */
835 	net->ipv4.frags.high_thresh = 256 * 1024;
836 	net->ipv4.frags.low_thresh = 192 * 1024;
837 	/*
838 	 * Important NOTE! Fragment queue must be destroyed before MSL expires.
839 	 * RFC791 is wrong proposing to prolongate timer each fragment arrival
840 	 * by TTL.
841 	 */
842 	net->ipv4.frags.timeout = IP_FRAG_TIME;
843 
844 	inet_frags_init_net(&net->ipv4.frags);
845 
846 	return ip4_frags_ns_ctl_register(net);
847 }
848 
849 static void __net_exit ipv4_frags_exit_net(struct net *net)
850 {
851 	ip4_frags_ns_ctl_unregister(net);
852 	inet_frags_exit_net(&net->ipv4.frags, &ip4_frags);
853 }
854 
855 static struct pernet_operations ip4_frags_ops = {
856 	.init = ipv4_frags_init_net,
857 	.exit = ipv4_frags_exit_net,
858 };
859 
860 void __init ipfrag_init(void)
861 {
862 	ip4_frags_ctl_register();
863 	register_pernet_subsys(&ip4_frags_ops);
864 	ip4_frags.hashfn = ip4_hashfn;
865 	ip4_frags.constructor = ip4_frag_init;
866 	ip4_frags.destructor = ip4_frag_free;
867 	ip4_frags.skb_free = NULL;
868 	ip4_frags.qsize = sizeof(struct ipq);
869 	ip4_frags.match = ip4_frag_match;
870 	ip4_frags.frag_expire = ip_expire;
871 	ip4_frags.secret_interval = 10 * 60 * HZ;
872 	inet_frags_init(&ip4_frags);
873 }
874