xref: /openbmc/linux/net/ipv4/ip_fragment.c (revision b24413180f5600bcb3bb70fbed5cf186b60864bd)
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(unsigned long arg)
195 {
196 	struct ipq *qp;
197 	struct net *net;
198 
199 	qp = container_of((struct inet_frag_queue *) arg, struct ipq, q);
200 	net = container_of(qp->q.net, struct net, ipv4.frags);
201 
202 	rcu_read_lock();
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(net, IPSTATS_MIB_REASMFAILS);
210 
211 	if (!inet_frag_evicting(&qp->q)) {
212 		struct sk_buff *clone, *head = qp->q.fragments;
213 		const struct iphdr *iph;
214 		int err;
215 
216 		__IP_INC_STATS(net, IPSTATS_MIB_REASMTIMEOUT);
217 
218 		if (!(qp->q.flags & INET_FRAG_FIRST_IN) || !qp->q.fragments)
219 			goto out;
220 
221 		head->dev = dev_get_by_index_rcu(net, qp->iif);
222 		if (!head->dev)
223 			goto out;
224 
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;
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;
239 
240 		clone = skb_clone(head, GFP_ATOMIC);
241 
242 		/* Send an ICMP "Fragment Reassembly Timeout" message. */
243 		if (clone) {
244 			spin_unlock(&qp->q.lock);
245 			icmp_send(clone, ICMP_TIME_EXCEEDED,
246 				  ICMP_EXC_FRAGTIME, 0);
247 			consume_skb(clone);
248 			goto out_rcu_unlock;
249 		}
250 	}
251 out:
252 	spin_unlock(&qp->q.lock);
253 out_rcu_unlock:
254 	rcu_read_unlock();
255 	ipq_put(qp);
256 }
257 
258 /* Find the correct entry in the "incomplete datagrams" queue for
259  * this IP datagram, and create new one, if nothing is found.
260  */
261 static struct ipq *ip_find(struct net *net, struct iphdr *iph,
262 			   u32 user, int vif)
263 {
264 	struct inet_frag_queue *q;
265 	struct ip4_create_arg arg;
266 	unsigned int hash;
267 
268 	arg.iph = iph;
269 	arg.user = user;
270 	arg.vif = vif;
271 
272 	hash = ipqhashfn(iph->id, iph->saddr, iph->daddr, iph->protocol);
273 
274 	q = inet_frag_find(&net->ipv4.frags, &ip4_frags, &arg, hash);
275 	if (IS_ERR_OR_NULL(q)) {
276 		inet_frag_maybe_warn_overflow(q, pr_fmt());
277 		return NULL;
278 	}
279 	return container_of(q, struct ipq, q);
280 }
281 
282 /* Is the fragment too far ahead to be part of ipq? */
283 static int ip_frag_too_far(struct ipq *qp)
284 {
285 	struct inet_peer *peer = qp->peer;
286 	unsigned int max = qp->q.net->max_dist;
287 	unsigned int start, end;
288 
289 	int rc;
290 
291 	if (!peer || !max)
292 		return 0;
293 
294 	start = qp->rid;
295 	end = atomic_inc_return(&peer->rid);
296 	qp->rid = end;
297 
298 	rc = qp->q.fragments && (end - start) > max;
299 
300 	if (rc) {
301 		struct net *net;
302 
303 		net = container_of(qp->q.net, struct net, ipv4.frags);
304 		__IP_INC_STATS(net, IPSTATS_MIB_REASMFAILS);
305 	}
306 
307 	return rc;
308 }
309 
310 static int ip_frag_reinit(struct ipq *qp)
311 {
312 	struct sk_buff *fp;
313 	unsigned int sum_truesize = 0;
314 
315 	if (!mod_timer(&qp->q.timer, jiffies + qp->q.net->timeout)) {
316 		refcount_inc(&qp->q.refcnt);
317 		return -ETIMEDOUT;
318 	}
319 
320 	fp = qp->q.fragments;
321 	do {
322 		struct sk_buff *xp = fp->next;
323 
324 		sum_truesize += fp->truesize;
325 		kfree_skb(fp);
326 		fp = xp;
327 	} while (fp);
328 	sub_frag_mem_limit(qp->q.net, sum_truesize);
329 
330 	qp->q.flags = 0;
331 	qp->q.len = 0;
332 	qp->q.meat = 0;
333 	qp->q.fragments = NULL;
334 	qp->q.fragments_tail = NULL;
335 	qp->iif = 0;
336 	qp->ecn = 0;
337 
338 	return 0;
339 }
340 
341 /* Add new segment to existing queue. */
342 static int ip_frag_queue(struct ipq *qp, struct sk_buff *skb)
343 {
344 	struct sk_buff *prev, *next;
345 	struct net_device *dev;
346 	unsigned int fragsize;
347 	int flags, offset;
348 	int ihl, end;
349 	int err = -ENOENT;
350 	u8 ecn;
351 
352 	if (qp->q.flags & INET_FRAG_COMPLETE)
353 		goto err;
354 
355 	if (!(IPCB(skb)->flags & IPSKB_FRAG_COMPLETE) &&
356 	    unlikely(ip_frag_too_far(qp)) &&
357 	    unlikely(err = ip_frag_reinit(qp))) {
358 		ipq_kill(qp);
359 		goto err;
360 	}
361 
362 	ecn = ip4_frag_ecn(ip_hdr(skb)->tos);
363 	offset = ntohs(ip_hdr(skb)->frag_off);
364 	flags = offset & ~IP_OFFSET;
365 	offset &= IP_OFFSET;
366 	offset <<= 3;		/* offset is in 8-byte chunks */
367 	ihl = ip_hdrlen(skb);
368 
369 	/* Determine the position of this fragment. */
370 	end = offset + skb->len - skb_network_offset(skb) - ihl;
371 	err = -EINVAL;
372 
373 	/* Is this the final fragment? */
374 	if ((flags & IP_MF) == 0) {
375 		/* If we already have some bits beyond end
376 		 * or have different end, the segment is corrupted.
377 		 */
378 		if (end < qp->q.len ||
379 		    ((qp->q.flags & INET_FRAG_LAST_IN) && end != qp->q.len))
380 			goto err;
381 		qp->q.flags |= INET_FRAG_LAST_IN;
382 		qp->q.len = end;
383 	} else {
384 		if (end&7) {
385 			end &= ~7;
386 			if (skb->ip_summed != CHECKSUM_UNNECESSARY)
387 				skb->ip_summed = CHECKSUM_NONE;
388 		}
389 		if (end > qp->q.len) {
390 			/* Some bits beyond end -> corruption. */
391 			if (qp->q.flags & INET_FRAG_LAST_IN)
392 				goto err;
393 			qp->q.len = end;
394 		}
395 	}
396 	if (end == offset)
397 		goto err;
398 
399 	err = -ENOMEM;
400 	if (!pskb_pull(skb, skb_network_offset(skb) + ihl))
401 		goto err;
402 
403 	err = pskb_trim_rcsum(skb, end - offset);
404 	if (err)
405 		goto err;
406 
407 	/* Find out which fragments are in front and at the back of us
408 	 * in the chain of fragments so far.  We must know where to put
409 	 * this fragment, right?
410 	 */
411 	prev = qp->q.fragments_tail;
412 	if (!prev || FRAG_CB(prev)->offset < offset) {
413 		next = NULL;
414 		goto found;
415 	}
416 	prev = NULL;
417 	for (next = qp->q.fragments; next != NULL; next = next->next) {
418 		if (FRAG_CB(next)->offset >= offset)
419 			break;	/* bingo! */
420 		prev = next;
421 	}
422 
423 found:
424 	/* We found where to put this one.  Check for overlap with
425 	 * preceding fragment, and, if needed, align things so that
426 	 * any overlaps are eliminated.
427 	 */
428 	if (prev) {
429 		int i = (FRAG_CB(prev)->offset + prev->len) - offset;
430 
431 		if (i > 0) {
432 			offset += i;
433 			err = -EINVAL;
434 			if (end <= offset)
435 				goto err;
436 			err = -ENOMEM;
437 			if (!pskb_pull(skb, i))
438 				goto err;
439 			if (skb->ip_summed != CHECKSUM_UNNECESSARY)
440 				skb->ip_summed = CHECKSUM_NONE;
441 		}
442 	}
443 
444 	err = -ENOMEM;
445 
446 	while (next && FRAG_CB(next)->offset < end) {
447 		int i = end - FRAG_CB(next)->offset; /* overlap is 'i' bytes */
448 
449 		if (i < next->len) {
450 			/* Eat head of the next overlapped fragment
451 			 * and leave the loop. The next ones cannot overlap.
452 			 */
453 			if (!pskb_pull(next, i))
454 				goto err;
455 			FRAG_CB(next)->offset += i;
456 			qp->q.meat -= i;
457 			if (next->ip_summed != CHECKSUM_UNNECESSARY)
458 				next->ip_summed = CHECKSUM_NONE;
459 			break;
460 		} else {
461 			struct sk_buff *free_it = next;
462 
463 			/* Old fragment is completely overridden with
464 			 * new one drop it.
465 			 */
466 			next = next->next;
467 
468 			if (prev)
469 				prev->next = next;
470 			else
471 				qp->q.fragments = next;
472 
473 			qp->q.meat -= free_it->len;
474 			sub_frag_mem_limit(qp->q.net, free_it->truesize);
475 			kfree_skb(free_it);
476 		}
477 	}
478 
479 	FRAG_CB(skb)->offset = offset;
480 
481 	/* Insert this fragment in the chain of fragments. */
482 	skb->next = next;
483 	if (!next)
484 		qp->q.fragments_tail = skb;
485 	if (prev)
486 		prev->next = skb;
487 	else
488 		qp->q.fragments = skb;
489 
490 	dev = skb->dev;
491 	if (dev) {
492 		qp->iif = dev->ifindex;
493 		skb->dev = NULL;
494 	}
495 	qp->q.stamp = skb->tstamp;
496 	qp->q.meat += skb->len;
497 	qp->ecn |= ecn;
498 	add_frag_mem_limit(qp->q.net, skb->truesize);
499 	if (offset == 0)
500 		qp->q.flags |= INET_FRAG_FIRST_IN;
501 
502 	fragsize = skb->len + ihl;
503 
504 	if (fragsize > qp->q.max_size)
505 		qp->q.max_size = fragsize;
506 
507 	if (ip_hdr(skb)->frag_off & htons(IP_DF) &&
508 	    fragsize > qp->max_df_size)
509 		qp->max_df_size = fragsize;
510 
511 	if (qp->q.flags == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
512 	    qp->q.meat == qp->q.len) {
513 		unsigned long orefdst = skb->_skb_refdst;
514 
515 		skb->_skb_refdst = 0UL;
516 		err = ip_frag_reasm(qp, prev, dev);
517 		skb->_skb_refdst = orefdst;
518 		return err;
519 	}
520 
521 	skb_dst_drop(skb);
522 	return -EINPROGRESS;
523 
524 err:
525 	kfree_skb(skb);
526 	return err;
527 }
528 
529 
530 /* Build a new IP datagram from all its fragments. */
531 
532 static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev,
533 			 struct net_device *dev)
534 {
535 	struct net *net = container_of(qp->q.net, struct net, ipv4.frags);
536 	struct iphdr *iph;
537 	struct sk_buff *fp, *head = qp->q.fragments;
538 	int len;
539 	int ihlen;
540 	int err;
541 	u8 ecn;
542 
543 	ipq_kill(qp);
544 
545 	ecn = ip_frag_ecn_table[qp->ecn];
546 	if (unlikely(ecn == 0xff)) {
547 		err = -EINVAL;
548 		goto out_fail;
549 	}
550 	/* Make the one we just received the head. */
551 	if (prev) {
552 		head = prev->next;
553 		fp = skb_clone(head, GFP_ATOMIC);
554 		if (!fp)
555 			goto out_nomem;
556 
557 		fp->next = head->next;
558 		if (!fp->next)
559 			qp->q.fragments_tail = fp;
560 		prev->next = fp;
561 
562 		skb_morph(head, qp->q.fragments);
563 		head->next = qp->q.fragments->next;
564 
565 		consume_skb(qp->q.fragments);
566 		qp->q.fragments = head;
567 	}
568 
569 	WARN_ON(!head);
570 	WARN_ON(FRAG_CB(head)->offset != 0);
571 
572 	/* Allocate a new buffer for the datagram. */
573 	ihlen = ip_hdrlen(head);
574 	len = ihlen + qp->q.len;
575 
576 	err = -E2BIG;
577 	if (len > 65535)
578 		goto out_oversize;
579 
580 	/* Head of list must not be cloned. */
581 	if (skb_unclone(head, GFP_ATOMIC))
582 		goto out_nomem;
583 
584 	/* If the first fragment is fragmented itself, we split
585 	 * it to two chunks: the first with data and paged part
586 	 * and the second, holding only fragments. */
587 	if (skb_has_frag_list(head)) {
588 		struct sk_buff *clone;
589 		int i, plen = 0;
590 
591 		clone = alloc_skb(0, GFP_ATOMIC);
592 		if (!clone)
593 			goto out_nomem;
594 		clone->next = head->next;
595 		head->next = clone;
596 		skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
597 		skb_frag_list_init(head);
598 		for (i = 0; i < skb_shinfo(head)->nr_frags; i++)
599 			plen += skb_frag_size(&skb_shinfo(head)->frags[i]);
600 		clone->len = clone->data_len = head->data_len - plen;
601 		head->data_len -= clone->len;
602 		head->len -= clone->len;
603 		clone->csum = 0;
604 		clone->ip_summed = head->ip_summed;
605 		add_frag_mem_limit(qp->q.net, clone->truesize);
606 	}
607 
608 	skb_shinfo(head)->frag_list = head->next;
609 	skb_push(head, head->data - skb_network_header(head));
610 
611 	for (fp=head->next; fp; fp = fp->next) {
612 		head->data_len += fp->len;
613 		head->len += fp->len;
614 		if (head->ip_summed != fp->ip_summed)
615 			head->ip_summed = CHECKSUM_NONE;
616 		else if (head->ip_summed == CHECKSUM_COMPLETE)
617 			head->csum = csum_add(head->csum, fp->csum);
618 		head->truesize += fp->truesize;
619 	}
620 	sub_frag_mem_limit(qp->q.net, head->truesize);
621 
622 	head->next = NULL;
623 	head->dev = dev;
624 	head->tstamp = qp->q.stamp;
625 	IPCB(head)->frag_max_size = max(qp->max_df_size, qp->q.max_size);
626 
627 	iph = ip_hdr(head);
628 	iph->tot_len = htons(len);
629 	iph->tos |= ecn;
630 
631 	/* When we set IP_DF on a refragmented skb we must also force a
632 	 * call to ip_fragment to avoid forwarding a DF-skb of size s while
633 	 * original sender only sent fragments of size f (where f < s).
634 	 *
635 	 * We only set DF/IPSKB_FRAG_PMTU if such DF fragment was the largest
636 	 * frag seen to avoid sending tiny DF-fragments in case skb was built
637 	 * from one very small df-fragment and one large non-df frag.
638 	 */
639 	if (qp->max_df_size == qp->q.max_size) {
640 		IPCB(head)->flags |= IPSKB_FRAG_PMTU;
641 		iph->frag_off = htons(IP_DF);
642 	} else {
643 		iph->frag_off = 0;
644 	}
645 
646 	ip_send_check(iph);
647 
648 	__IP_INC_STATS(net, IPSTATS_MIB_REASMOKS);
649 	qp->q.fragments = NULL;
650 	qp->q.fragments_tail = NULL;
651 	return 0;
652 
653 out_nomem:
654 	net_dbg_ratelimited("queue_glue: no memory for gluing queue %p\n", qp);
655 	err = -ENOMEM;
656 	goto out_fail;
657 out_oversize:
658 	net_info_ratelimited("Oversized IP packet from %pI4\n", &qp->saddr);
659 out_fail:
660 	__IP_INC_STATS(net, IPSTATS_MIB_REASMFAILS);
661 	return err;
662 }
663 
664 /* Process an incoming IP datagram fragment. */
665 int ip_defrag(struct net *net, struct sk_buff *skb, u32 user)
666 {
667 	struct net_device *dev = skb->dev ? : skb_dst(skb)->dev;
668 	int vif = l3mdev_master_ifindex_rcu(dev);
669 	struct ipq *qp;
670 
671 	__IP_INC_STATS(net, IPSTATS_MIB_REASMREQDS);
672 	skb_orphan(skb);
673 
674 	/* Lookup (or create) queue header */
675 	qp = ip_find(net, ip_hdr(skb), user, vif);
676 	if (qp) {
677 		int ret;
678 
679 		spin_lock(&qp->q.lock);
680 
681 		ret = ip_frag_queue(qp, skb);
682 
683 		spin_unlock(&qp->q.lock);
684 		ipq_put(qp);
685 		return ret;
686 	}
687 
688 	__IP_INC_STATS(net, IPSTATS_MIB_REASMFAILS);
689 	kfree_skb(skb);
690 	return -ENOMEM;
691 }
692 EXPORT_SYMBOL(ip_defrag);
693 
694 struct sk_buff *ip_check_defrag(struct net *net, struct sk_buff *skb, u32 user)
695 {
696 	struct iphdr iph;
697 	int netoff;
698 	u32 len;
699 
700 	if (skb->protocol != htons(ETH_P_IP))
701 		return skb;
702 
703 	netoff = skb_network_offset(skb);
704 
705 	if (skb_copy_bits(skb, netoff, &iph, sizeof(iph)) < 0)
706 		return skb;
707 
708 	if (iph.ihl < 5 || iph.version != 4)
709 		return skb;
710 
711 	len = ntohs(iph.tot_len);
712 	if (skb->len < netoff + len || len < (iph.ihl * 4))
713 		return skb;
714 
715 	if (ip_is_fragment(&iph)) {
716 		skb = skb_share_check(skb, GFP_ATOMIC);
717 		if (skb) {
718 			if (!pskb_may_pull(skb, netoff + iph.ihl * 4))
719 				return skb;
720 			if (pskb_trim_rcsum(skb, netoff + len))
721 				return skb;
722 			memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
723 			if (ip_defrag(net, skb, user))
724 				return NULL;
725 			skb_clear_hash(skb);
726 		}
727 	}
728 	return skb;
729 }
730 EXPORT_SYMBOL(ip_check_defrag);
731 
732 #ifdef CONFIG_SYSCTL
733 static int zero;
734 
735 static struct ctl_table ip4_frags_ns_ctl_table[] = {
736 	{
737 		.procname	= "ipfrag_high_thresh",
738 		.data		= &init_net.ipv4.frags.high_thresh,
739 		.maxlen		= sizeof(int),
740 		.mode		= 0644,
741 		.proc_handler	= proc_dointvec_minmax,
742 		.extra1		= &init_net.ipv4.frags.low_thresh
743 	},
744 	{
745 		.procname	= "ipfrag_low_thresh",
746 		.data		= &init_net.ipv4.frags.low_thresh,
747 		.maxlen		= sizeof(int),
748 		.mode		= 0644,
749 		.proc_handler	= proc_dointvec_minmax,
750 		.extra1		= &zero,
751 		.extra2		= &init_net.ipv4.frags.high_thresh
752 	},
753 	{
754 		.procname	= "ipfrag_time",
755 		.data		= &init_net.ipv4.frags.timeout,
756 		.maxlen		= sizeof(int),
757 		.mode		= 0644,
758 		.proc_handler	= proc_dointvec_jiffies,
759 	},
760 	{
761 		.procname	= "ipfrag_max_dist",
762 		.data		= &init_net.ipv4.frags.max_dist,
763 		.maxlen		= sizeof(int),
764 		.mode		= 0644,
765 		.proc_handler	= proc_dointvec_minmax,
766 		.extra1		= &zero
767 	},
768 	{ }
769 };
770 
771 /* secret interval has been deprecated */
772 static int ip4_frags_secret_interval_unused;
773 static struct ctl_table ip4_frags_ctl_table[] = {
774 	{
775 		.procname	= "ipfrag_secret_interval",
776 		.data		= &ip4_frags_secret_interval_unused,
777 		.maxlen		= sizeof(int),
778 		.mode		= 0644,
779 		.proc_handler	= proc_dointvec_jiffies,
780 	},
781 	{ }
782 };
783 
784 static int __net_init ip4_frags_ns_ctl_register(struct net *net)
785 {
786 	struct ctl_table *table;
787 	struct ctl_table_header *hdr;
788 
789 	table = ip4_frags_ns_ctl_table;
790 	if (!net_eq(net, &init_net)) {
791 		table = kmemdup(table, sizeof(ip4_frags_ns_ctl_table), GFP_KERNEL);
792 		if (!table)
793 			goto err_alloc;
794 
795 		table[0].data = &net->ipv4.frags.high_thresh;
796 		table[0].extra1 = &net->ipv4.frags.low_thresh;
797 		table[0].extra2 = &init_net.ipv4.frags.high_thresh;
798 		table[1].data = &net->ipv4.frags.low_thresh;
799 		table[1].extra2 = &net->ipv4.frags.high_thresh;
800 		table[2].data = &net->ipv4.frags.timeout;
801 		table[3].data = &net->ipv4.frags.max_dist;
802 	}
803 
804 	hdr = register_net_sysctl(net, "net/ipv4", table);
805 	if (!hdr)
806 		goto err_reg;
807 
808 	net->ipv4.frags_hdr = hdr;
809 	return 0;
810 
811 err_reg:
812 	if (!net_eq(net, &init_net))
813 		kfree(table);
814 err_alloc:
815 	return -ENOMEM;
816 }
817 
818 static void __net_exit ip4_frags_ns_ctl_unregister(struct net *net)
819 {
820 	struct ctl_table *table;
821 
822 	table = net->ipv4.frags_hdr->ctl_table_arg;
823 	unregister_net_sysctl_table(net->ipv4.frags_hdr);
824 	kfree(table);
825 }
826 
827 static void __init ip4_frags_ctl_register(void)
828 {
829 	register_net_sysctl(&init_net, "net/ipv4", ip4_frags_ctl_table);
830 }
831 #else
832 static int ip4_frags_ns_ctl_register(struct net *net)
833 {
834 	return 0;
835 }
836 
837 static void ip4_frags_ns_ctl_unregister(struct net *net)
838 {
839 }
840 
841 static void __init ip4_frags_ctl_register(void)
842 {
843 }
844 #endif
845 
846 static int __net_init ipv4_frags_init_net(struct net *net)
847 {
848 	/* Fragment cache limits.
849 	 *
850 	 * The fragment memory accounting code, (tries to) account for
851 	 * the real memory usage, by measuring both the size of frag
852 	 * queue struct (inet_frag_queue (ipv4:ipq/ipv6:frag_queue))
853 	 * and the SKB's truesize.
854 	 *
855 	 * A 64K fragment consumes 129736 bytes (44*2944)+200
856 	 * (1500 truesize == 2944, sizeof(struct ipq) == 200)
857 	 *
858 	 * We will commit 4MB at one time. Should we cross that limit
859 	 * we will prune down to 3MB, making room for approx 8 big 64K
860 	 * fragments 8x128k.
861 	 */
862 	net->ipv4.frags.high_thresh = 4 * 1024 * 1024;
863 	net->ipv4.frags.low_thresh  = 3 * 1024 * 1024;
864 	/*
865 	 * Important NOTE! Fragment queue must be destroyed before MSL expires.
866 	 * RFC791 is wrong proposing to prolongate timer each fragment arrival
867 	 * by TTL.
868 	 */
869 	net->ipv4.frags.timeout = IP_FRAG_TIME;
870 
871 	net->ipv4.frags.max_dist = 64;
872 
873 	inet_frags_init_net(&net->ipv4.frags);
874 
875 	return ip4_frags_ns_ctl_register(net);
876 }
877 
878 static void __net_exit ipv4_frags_exit_net(struct net *net)
879 {
880 	ip4_frags_ns_ctl_unregister(net);
881 	inet_frags_exit_net(&net->ipv4.frags, &ip4_frags);
882 }
883 
884 static struct pernet_operations ip4_frags_ops = {
885 	.init = ipv4_frags_init_net,
886 	.exit = ipv4_frags_exit_net,
887 };
888 
889 void __init ipfrag_init(void)
890 {
891 	ip4_frags_ctl_register();
892 	register_pernet_subsys(&ip4_frags_ops);
893 	ip4_frags.hashfn = ip4_hashfn;
894 	ip4_frags.constructor = ip4_frag_init;
895 	ip4_frags.destructor = ip4_frag_free;
896 	ip4_frags.qsize = sizeof(struct ipq);
897 	ip4_frags.match = ip4_frag_match;
898 	ip4_frags.frag_expire = ip_expire;
899 	ip4_frags.frags_cache_name = ip_frag_cache_name;
900 	if (inet_frags_init(&ip4_frags))
901 		panic("IP: failed to allocate ip4_frags cache\n");
902 }
903