xref: /openbmc/linux/net/ipv4/ip_fragment.c (revision d2ba09c1)
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 /* Describe an entry in the "incomplete datagrams" queue. */
61 struct ipq {
62 	struct inet_frag_queue q;
63 
64 	u8		ecn; /* RFC3168 support */
65 	u16		max_df_size; /* largest frag with DF set seen */
66 	int             iif;
67 	unsigned int    rid;
68 	struct inet_peer *peer;
69 };
70 
71 static u8 ip4_frag_ecn(u8 tos)
72 {
73 	return 1 << (tos & INET_ECN_MASK);
74 }
75 
76 static struct inet_frags ip4_frags;
77 
78 static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev,
79 			 struct net_device *dev);
80 
81 
82 static void ip4_frag_init(struct inet_frag_queue *q, const void *a)
83 {
84 	struct ipq *qp = container_of(q, struct ipq, q);
85 	struct netns_ipv4 *ipv4 = container_of(q->net, struct netns_ipv4,
86 					       frags);
87 	struct net *net = container_of(ipv4, struct net, ipv4);
88 
89 	const struct frag_v4_compare_key *key = a;
90 
91 	q->key.v4 = *key;
92 	qp->ecn = 0;
93 	qp->peer = q->net->max_dist ?
94 		inet_getpeer_v4(net->ipv4.peers, key->saddr, key->vif, 1) :
95 		NULL;
96 }
97 
98 static void ip4_frag_free(struct inet_frag_queue *q)
99 {
100 	struct ipq *qp;
101 
102 	qp = container_of(q, struct ipq, q);
103 	if (qp->peer)
104 		inet_putpeer(qp->peer);
105 }
106 
107 
108 /* Destruction primitives. */
109 
110 static void ipq_put(struct ipq *ipq)
111 {
112 	inet_frag_put(&ipq->q);
113 }
114 
115 /* Kill ipq entry. It is not destroyed immediately,
116  * because caller (and someone more) holds reference count.
117  */
118 static void ipq_kill(struct ipq *ipq)
119 {
120 	inet_frag_kill(&ipq->q);
121 }
122 
123 static bool frag_expire_skip_icmp(u32 user)
124 {
125 	return user == IP_DEFRAG_AF_PACKET ||
126 	       ip_defrag_user_in_between(user, IP_DEFRAG_CONNTRACK_IN,
127 					 __IP_DEFRAG_CONNTRACK_IN_END) ||
128 	       ip_defrag_user_in_between(user, IP_DEFRAG_CONNTRACK_BRIDGE_IN,
129 					 __IP_DEFRAG_CONNTRACK_BRIDGE_IN);
130 }
131 
132 /*
133  * Oops, a fragment queue timed out.  Kill it and send an ICMP reply.
134  */
135 static void ip_expire(struct timer_list *t)
136 {
137 	struct inet_frag_queue *frag = from_timer(frag, t, timer);
138 	const struct iphdr *iph;
139 	struct sk_buff *head;
140 	struct net *net;
141 	struct ipq *qp;
142 	int err;
143 
144 	qp = container_of(frag, struct ipq, q);
145 	net = container_of(qp->q.net, struct net, ipv4.frags);
146 
147 	rcu_read_lock();
148 	spin_lock(&qp->q.lock);
149 
150 	if (qp->q.flags & INET_FRAG_COMPLETE)
151 		goto out;
152 
153 	ipq_kill(qp);
154 	__IP_INC_STATS(net, IPSTATS_MIB_REASMFAILS);
155 
156 	head = qp->q.fragments;
157 
158 	__IP_INC_STATS(net, IPSTATS_MIB_REASMTIMEOUT);
159 
160 	if (!(qp->q.flags & INET_FRAG_FIRST_IN) || !head)
161 		goto out;
162 
163 	head->dev = dev_get_by_index_rcu(net, qp->iif);
164 	if (!head->dev)
165 		goto out;
166 
167 
168 	/* skb has no dst, perform route lookup again */
169 	iph = ip_hdr(head);
170 	err = ip_route_input_noref(head, iph->daddr, iph->saddr,
171 					   iph->tos, head->dev);
172 	if (err)
173 		goto out;
174 
175 	/* Only an end host needs to send an ICMP
176 	 * "Fragment Reassembly Timeout" message, per RFC792.
177 	 */
178 	if (frag_expire_skip_icmp(qp->q.key.v4.user) &&
179 	    (skb_rtable(head)->rt_type != RTN_LOCAL))
180 		goto out;
181 
182 	skb_get(head);
183 	spin_unlock(&qp->q.lock);
184 	icmp_send(head, ICMP_TIME_EXCEEDED, ICMP_EXC_FRAGTIME, 0);
185 	kfree_skb(head);
186 	goto out_rcu_unlock;
187 
188 out:
189 	spin_unlock(&qp->q.lock);
190 out_rcu_unlock:
191 	rcu_read_unlock();
192 	ipq_put(qp);
193 }
194 
195 /* Find the correct entry in the "incomplete datagrams" queue for
196  * this IP datagram, and create new one, if nothing is found.
197  */
198 static struct ipq *ip_find(struct net *net, struct iphdr *iph,
199 			   u32 user, int vif)
200 {
201 	struct frag_v4_compare_key key = {
202 		.saddr = iph->saddr,
203 		.daddr = iph->daddr,
204 		.user = user,
205 		.vif = vif,
206 		.id = iph->id,
207 		.protocol = iph->protocol,
208 	};
209 	struct inet_frag_queue *q;
210 
211 	q = inet_frag_find(&net->ipv4.frags, &key);
212 	if (!q)
213 		return NULL;
214 
215 	return container_of(q, struct ipq, q);
216 }
217 
218 /* Is the fragment too far ahead to be part of ipq? */
219 static int ip_frag_too_far(struct ipq *qp)
220 {
221 	struct inet_peer *peer = qp->peer;
222 	unsigned int max = qp->q.net->max_dist;
223 	unsigned int start, end;
224 
225 	int rc;
226 
227 	if (!peer || !max)
228 		return 0;
229 
230 	start = qp->rid;
231 	end = atomic_inc_return(&peer->rid);
232 	qp->rid = end;
233 
234 	rc = qp->q.fragments && (end - start) > max;
235 
236 	if (rc) {
237 		struct net *net;
238 
239 		net = container_of(qp->q.net, struct net, ipv4.frags);
240 		__IP_INC_STATS(net, IPSTATS_MIB_REASMFAILS);
241 	}
242 
243 	return rc;
244 }
245 
246 static int ip_frag_reinit(struct ipq *qp)
247 {
248 	struct sk_buff *fp;
249 	unsigned int sum_truesize = 0;
250 
251 	if (!mod_timer(&qp->q.timer, jiffies + qp->q.net->timeout)) {
252 		refcount_inc(&qp->q.refcnt);
253 		return -ETIMEDOUT;
254 	}
255 
256 	fp = qp->q.fragments;
257 	do {
258 		struct sk_buff *xp = fp->next;
259 
260 		sum_truesize += fp->truesize;
261 		kfree_skb(fp);
262 		fp = xp;
263 	} while (fp);
264 	sub_frag_mem_limit(qp->q.net, sum_truesize);
265 
266 	qp->q.flags = 0;
267 	qp->q.len = 0;
268 	qp->q.meat = 0;
269 	qp->q.fragments = NULL;
270 	qp->q.fragments_tail = NULL;
271 	qp->iif = 0;
272 	qp->ecn = 0;
273 
274 	return 0;
275 }
276 
277 /* Add new segment to existing queue. */
278 static int ip_frag_queue(struct ipq *qp, struct sk_buff *skb)
279 {
280 	struct sk_buff *prev, *next;
281 	struct net_device *dev;
282 	unsigned int fragsize;
283 	int flags, offset;
284 	int ihl, end;
285 	int err = -ENOENT;
286 	u8 ecn;
287 
288 	if (qp->q.flags & INET_FRAG_COMPLETE)
289 		goto err;
290 
291 	if (!(IPCB(skb)->flags & IPSKB_FRAG_COMPLETE) &&
292 	    unlikely(ip_frag_too_far(qp)) &&
293 	    unlikely(err = ip_frag_reinit(qp))) {
294 		ipq_kill(qp);
295 		goto err;
296 	}
297 
298 	ecn = ip4_frag_ecn(ip_hdr(skb)->tos);
299 	offset = ntohs(ip_hdr(skb)->frag_off);
300 	flags = offset & ~IP_OFFSET;
301 	offset &= IP_OFFSET;
302 	offset <<= 3;		/* offset is in 8-byte chunks */
303 	ihl = ip_hdrlen(skb);
304 
305 	/* Determine the position of this fragment. */
306 	end = offset + skb->len - skb_network_offset(skb) - ihl;
307 	err = -EINVAL;
308 
309 	/* Is this the final fragment? */
310 	if ((flags & IP_MF) == 0) {
311 		/* If we already have some bits beyond end
312 		 * or have different end, the segment is corrupted.
313 		 */
314 		if (end < qp->q.len ||
315 		    ((qp->q.flags & INET_FRAG_LAST_IN) && end != qp->q.len))
316 			goto err;
317 		qp->q.flags |= INET_FRAG_LAST_IN;
318 		qp->q.len = end;
319 	} else {
320 		if (end&7) {
321 			end &= ~7;
322 			if (skb->ip_summed != CHECKSUM_UNNECESSARY)
323 				skb->ip_summed = CHECKSUM_NONE;
324 		}
325 		if (end > qp->q.len) {
326 			/* Some bits beyond end -> corruption. */
327 			if (qp->q.flags & INET_FRAG_LAST_IN)
328 				goto err;
329 			qp->q.len = end;
330 		}
331 	}
332 	if (end == offset)
333 		goto err;
334 
335 	err = -ENOMEM;
336 	if (!pskb_pull(skb, skb_network_offset(skb) + ihl))
337 		goto err;
338 
339 	err = pskb_trim_rcsum(skb, end - offset);
340 	if (err)
341 		goto err;
342 
343 	/* Find out which fragments are in front and at the back of us
344 	 * in the chain of fragments so far.  We must know where to put
345 	 * this fragment, right?
346 	 */
347 	prev = qp->q.fragments_tail;
348 	if (!prev || prev->ip_defrag_offset < offset) {
349 		next = NULL;
350 		goto found;
351 	}
352 	prev = NULL;
353 	for (next = qp->q.fragments; next != NULL; next = next->next) {
354 		if (next->ip_defrag_offset >= offset)
355 			break;	/* bingo! */
356 		prev = next;
357 	}
358 
359 found:
360 	/* We found where to put this one.  Check for overlap with
361 	 * preceding fragment, and, if needed, align things so that
362 	 * any overlaps are eliminated.
363 	 */
364 	if (prev) {
365 		int i = (prev->ip_defrag_offset + prev->len) - offset;
366 
367 		if (i > 0) {
368 			offset += i;
369 			err = -EINVAL;
370 			if (end <= offset)
371 				goto err;
372 			err = -ENOMEM;
373 			if (!pskb_pull(skb, i))
374 				goto err;
375 			if (skb->ip_summed != CHECKSUM_UNNECESSARY)
376 				skb->ip_summed = CHECKSUM_NONE;
377 		}
378 	}
379 
380 	err = -ENOMEM;
381 
382 	while (next && next->ip_defrag_offset < end) {
383 		int i = end - next->ip_defrag_offset; /* overlap is 'i' bytes */
384 
385 		if (i < next->len) {
386 			/* Eat head of the next overlapped fragment
387 			 * and leave the loop. The next ones cannot overlap.
388 			 */
389 			if (!pskb_pull(next, i))
390 				goto err;
391 			next->ip_defrag_offset += i;
392 			qp->q.meat -= i;
393 			if (next->ip_summed != CHECKSUM_UNNECESSARY)
394 				next->ip_summed = CHECKSUM_NONE;
395 			break;
396 		} else {
397 			struct sk_buff *free_it = next;
398 
399 			/* Old fragment is completely overridden with
400 			 * new one drop it.
401 			 */
402 			next = next->next;
403 
404 			if (prev)
405 				prev->next = next;
406 			else
407 				qp->q.fragments = next;
408 
409 			qp->q.meat -= free_it->len;
410 			sub_frag_mem_limit(qp->q.net, free_it->truesize);
411 			kfree_skb(free_it);
412 		}
413 	}
414 
415 	/* Note : skb->ip_defrag_offset and skb->dev share the same location */
416 	dev = skb->dev;
417 	if (dev)
418 		qp->iif = dev->ifindex;
419 	/* Makes sure compiler wont do silly aliasing games */
420 	barrier();
421 	skb->ip_defrag_offset = offset;
422 
423 	/* Insert this fragment in the chain of fragments. */
424 	skb->next = next;
425 	if (!next)
426 		qp->q.fragments_tail = skb;
427 	if (prev)
428 		prev->next = skb;
429 	else
430 		qp->q.fragments = skb;
431 
432 	qp->q.stamp = skb->tstamp;
433 	qp->q.meat += skb->len;
434 	qp->ecn |= ecn;
435 	add_frag_mem_limit(qp->q.net, skb->truesize);
436 	if (offset == 0)
437 		qp->q.flags |= INET_FRAG_FIRST_IN;
438 
439 	fragsize = skb->len + ihl;
440 
441 	if (fragsize > qp->q.max_size)
442 		qp->q.max_size = fragsize;
443 
444 	if (ip_hdr(skb)->frag_off & htons(IP_DF) &&
445 	    fragsize > qp->max_df_size)
446 		qp->max_df_size = fragsize;
447 
448 	if (qp->q.flags == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
449 	    qp->q.meat == qp->q.len) {
450 		unsigned long orefdst = skb->_skb_refdst;
451 
452 		skb->_skb_refdst = 0UL;
453 		err = ip_frag_reasm(qp, prev, dev);
454 		skb->_skb_refdst = orefdst;
455 		return err;
456 	}
457 
458 	skb_dst_drop(skb);
459 	return -EINPROGRESS;
460 
461 err:
462 	kfree_skb(skb);
463 	return err;
464 }
465 
466 
467 /* Build a new IP datagram from all its fragments. */
468 
469 static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev,
470 			 struct net_device *dev)
471 {
472 	struct net *net = container_of(qp->q.net, struct net, ipv4.frags);
473 	struct iphdr *iph;
474 	struct sk_buff *fp, *head = qp->q.fragments;
475 	int len;
476 	int ihlen;
477 	int err;
478 	u8 ecn;
479 
480 	ipq_kill(qp);
481 
482 	ecn = ip_frag_ecn_table[qp->ecn];
483 	if (unlikely(ecn == 0xff)) {
484 		err = -EINVAL;
485 		goto out_fail;
486 	}
487 	/* Make the one we just received the head. */
488 	if (prev) {
489 		head = prev->next;
490 		fp = skb_clone(head, GFP_ATOMIC);
491 		if (!fp)
492 			goto out_nomem;
493 
494 		fp->next = head->next;
495 		if (!fp->next)
496 			qp->q.fragments_tail = fp;
497 		prev->next = fp;
498 
499 		skb_morph(head, qp->q.fragments);
500 		head->next = qp->q.fragments->next;
501 
502 		consume_skb(qp->q.fragments);
503 		qp->q.fragments = head;
504 	}
505 
506 	WARN_ON(!head);
507 	WARN_ON(head->ip_defrag_offset != 0);
508 
509 	/* Allocate a new buffer for the datagram. */
510 	ihlen = ip_hdrlen(head);
511 	len = ihlen + qp->q.len;
512 
513 	err = -E2BIG;
514 	if (len > 65535)
515 		goto out_oversize;
516 
517 	/* Head of list must not be cloned. */
518 	if (skb_unclone(head, GFP_ATOMIC))
519 		goto out_nomem;
520 
521 	/* If the first fragment is fragmented itself, we split
522 	 * it to two chunks: the first with data and paged part
523 	 * and the second, holding only fragments. */
524 	if (skb_has_frag_list(head)) {
525 		struct sk_buff *clone;
526 		int i, plen = 0;
527 
528 		clone = alloc_skb(0, GFP_ATOMIC);
529 		if (!clone)
530 			goto out_nomem;
531 		clone->next = head->next;
532 		head->next = clone;
533 		skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
534 		skb_frag_list_init(head);
535 		for (i = 0; i < skb_shinfo(head)->nr_frags; i++)
536 			plen += skb_frag_size(&skb_shinfo(head)->frags[i]);
537 		clone->len = clone->data_len = head->data_len - plen;
538 		head->data_len -= clone->len;
539 		head->len -= clone->len;
540 		clone->csum = 0;
541 		clone->ip_summed = head->ip_summed;
542 		add_frag_mem_limit(qp->q.net, clone->truesize);
543 	}
544 
545 	skb_shinfo(head)->frag_list = head->next;
546 	skb_push(head, head->data - skb_network_header(head));
547 
548 	for (fp=head->next; fp; fp = fp->next) {
549 		head->data_len += fp->len;
550 		head->len += fp->len;
551 		if (head->ip_summed != fp->ip_summed)
552 			head->ip_summed = CHECKSUM_NONE;
553 		else if (head->ip_summed == CHECKSUM_COMPLETE)
554 			head->csum = csum_add(head->csum, fp->csum);
555 		head->truesize += fp->truesize;
556 	}
557 	sub_frag_mem_limit(qp->q.net, head->truesize);
558 
559 	head->next = NULL;
560 	head->dev = dev;
561 	head->tstamp = qp->q.stamp;
562 	IPCB(head)->frag_max_size = max(qp->max_df_size, qp->q.max_size);
563 
564 	iph = ip_hdr(head);
565 	iph->tot_len = htons(len);
566 	iph->tos |= ecn;
567 
568 	/* When we set IP_DF on a refragmented skb we must also force a
569 	 * call to ip_fragment to avoid forwarding a DF-skb of size s while
570 	 * original sender only sent fragments of size f (where f < s).
571 	 *
572 	 * We only set DF/IPSKB_FRAG_PMTU if such DF fragment was the largest
573 	 * frag seen to avoid sending tiny DF-fragments in case skb was built
574 	 * from one very small df-fragment and one large non-df frag.
575 	 */
576 	if (qp->max_df_size == qp->q.max_size) {
577 		IPCB(head)->flags |= IPSKB_FRAG_PMTU;
578 		iph->frag_off = htons(IP_DF);
579 	} else {
580 		iph->frag_off = 0;
581 	}
582 
583 	ip_send_check(iph);
584 
585 	__IP_INC_STATS(net, IPSTATS_MIB_REASMOKS);
586 	qp->q.fragments = NULL;
587 	qp->q.fragments_tail = NULL;
588 	return 0;
589 
590 out_nomem:
591 	net_dbg_ratelimited("queue_glue: no memory for gluing queue %p\n", qp);
592 	err = -ENOMEM;
593 	goto out_fail;
594 out_oversize:
595 	net_info_ratelimited("Oversized IP packet from %pI4\n", &qp->q.key.v4.saddr);
596 out_fail:
597 	__IP_INC_STATS(net, IPSTATS_MIB_REASMFAILS);
598 	return err;
599 }
600 
601 /* Process an incoming IP datagram fragment. */
602 int ip_defrag(struct net *net, struct sk_buff *skb, u32 user)
603 {
604 	struct net_device *dev = skb->dev ? : skb_dst(skb)->dev;
605 	int vif = l3mdev_master_ifindex_rcu(dev);
606 	struct ipq *qp;
607 
608 	__IP_INC_STATS(net, IPSTATS_MIB_REASMREQDS);
609 	skb_orphan(skb);
610 
611 	/* Lookup (or create) queue header */
612 	qp = ip_find(net, ip_hdr(skb), user, vif);
613 	if (qp) {
614 		int ret;
615 
616 		spin_lock(&qp->q.lock);
617 
618 		ret = ip_frag_queue(qp, skb);
619 
620 		spin_unlock(&qp->q.lock);
621 		ipq_put(qp);
622 		return ret;
623 	}
624 
625 	__IP_INC_STATS(net, IPSTATS_MIB_REASMFAILS);
626 	kfree_skb(skb);
627 	return -ENOMEM;
628 }
629 EXPORT_SYMBOL(ip_defrag);
630 
631 struct sk_buff *ip_check_defrag(struct net *net, struct sk_buff *skb, u32 user)
632 {
633 	struct iphdr iph;
634 	int netoff;
635 	u32 len;
636 
637 	if (skb->protocol != htons(ETH_P_IP))
638 		return skb;
639 
640 	netoff = skb_network_offset(skb);
641 
642 	if (skb_copy_bits(skb, netoff, &iph, sizeof(iph)) < 0)
643 		return skb;
644 
645 	if (iph.ihl < 5 || iph.version != 4)
646 		return skb;
647 
648 	len = ntohs(iph.tot_len);
649 	if (skb->len < netoff + len || len < (iph.ihl * 4))
650 		return skb;
651 
652 	if (ip_is_fragment(&iph)) {
653 		skb = skb_share_check(skb, GFP_ATOMIC);
654 		if (skb) {
655 			if (!pskb_may_pull(skb, netoff + iph.ihl * 4))
656 				return skb;
657 			if (pskb_trim_rcsum(skb, netoff + len))
658 				return skb;
659 			memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
660 			if (ip_defrag(net, skb, user))
661 				return NULL;
662 			skb_clear_hash(skb);
663 		}
664 	}
665 	return skb;
666 }
667 EXPORT_SYMBOL(ip_check_defrag);
668 
669 #ifdef CONFIG_SYSCTL
670 static int dist_min;
671 
672 static struct ctl_table ip4_frags_ns_ctl_table[] = {
673 	{
674 		.procname	= "ipfrag_high_thresh",
675 		.data		= &init_net.ipv4.frags.high_thresh,
676 		.maxlen		= sizeof(unsigned long),
677 		.mode		= 0644,
678 		.proc_handler	= proc_doulongvec_minmax,
679 		.extra1		= &init_net.ipv4.frags.low_thresh
680 	},
681 	{
682 		.procname	= "ipfrag_low_thresh",
683 		.data		= &init_net.ipv4.frags.low_thresh,
684 		.maxlen		= sizeof(unsigned long),
685 		.mode		= 0644,
686 		.proc_handler	= proc_doulongvec_minmax,
687 		.extra2		= &init_net.ipv4.frags.high_thresh
688 	},
689 	{
690 		.procname	= "ipfrag_time",
691 		.data		= &init_net.ipv4.frags.timeout,
692 		.maxlen		= sizeof(int),
693 		.mode		= 0644,
694 		.proc_handler	= proc_dointvec_jiffies,
695 	},
696 	{
697 		.procname	= "ipfrag_max_dist",
698 		.data		= &init_net.ipv4.frags.max_dist,
699 		.maxlen		= sizeof(int),
700 		.mode		= 0644,
701 		.proc_handler	= proc_dointvec_minmax,
702 		.extra1		= &dist_min,
703 	},
704 	{ }
705 };
706 
707 /* secret interval has been deprecated */
708 static int ip4_frags_secret_interval_unused;
709 static struct ctl_table ip4_frags_ctl_table[] = {
710 	{
711 		.procname	= "ipfrag_secret_interval",
712 		.data		= &ip4_frags_secret_interval_unused,
713 		.maxlen		= sizeof(int),
714 		.mode		= 0644,
715 		.proc_handler	= proc_dointvec_jiffies,
716 	},
717 	{ }
718 };
719 
720 static int __net_init ip4_frags_ns_ctl_register(struct net *net)
721 {
722 	struct ctl_table *table;
723 	struct ctl_table_header *hdr;
724 
725 	table = ip4_frags_ns_ctl_table;
726 	if (!net_eq(net, &init_net)) {
727 		table = kmemdup(table, sizeof(ip4_frags_ns_ctl_table), GFP_KERNEL);
728 		if (!table)
729 			goto err_alloc;
730 
731 		table[0].data = &net->ipv4.frags.high_thresh;
732 		table[0].extra1 = &net->ipv4.frags.low_thresh;
733 		table[0].extra2 = &init_net.ipv4.frags.high_thresh;
734 		table[1].data = &net->ipv4.frags.low_thresh;
735 		table[1].extra2 = &net->ipv4.frags.high_thresh;
736 		table[2].data = &net->ipv4.frags.timeout;
737 		table[3].data = &net->ipv4.frags.max_dist;
738 	}
739 
740 	hdr = register_net_sysctl(net, "net/ipv4", table);
741 	if (!hdr)
742 		goto err_reg;
743 
744 	net->ipv4.frags_hdr = hdr;
745 	return 0;
746 
747 err_reg:
748 	if (!net_eq(net, &init_net))
749 		kfree(table);
750 err_alloc:
751 	return -ENOMEM;
752 }
753 
754 static void __net_exit ip4_frags_ns_ctl_unregister(struct net *net)
755 {
756 	struct ctl_table *table;
757 
758 	table = net->ipv4.frags_hdr->ctl_table_arg;
759 	unregister_net_sysctl_table(net->ipv4.frags_hdr);
760 	kfree(table);
761 }
762 
763 static void __init ip4_frags_ctl_register(void)
764 {
765 	register_net_sysctl(&init_net, "net/ipv4", ip4_frags_ctl_table);
766 }
767 #else
768 static int ip4_frags_ns_ctl_register(struct net *net)
769 {
770 	return 0;
771 }
772 
773 static void ip4_frags_ns_ctl_unregister(struct net *net)
774 {
775 }
776 
777 static void __init ip4_frags_ctl_register(void)
778 {
779 }
780 #endif
781 
782 static int __net_init ipv4_frags_init_net(struct net *net)
783 {
784 	int res;
785 
786 	/* Fragment cache limits.
787 	 *
788 	 * The fragment memory accounting code, (tries to) account for
789 	 * the real memory usage, by measuring both the size of frag
790 	 * queue struct (inet_frag_queue (ipv4:ipq/ipv6:frag_queue))
791 	 * and the SKB's truesize.
792 	 *
793 	 * A 64K fragment consumes 129736 bytes (44*2944)+200
794 	 * (1500 truesize == 2944, sizeof(struct ipq) == 200)
795 	 *
796 	 * We will commit 4MB at one time. Should we cross that limit
797 	 * we will prune down to 3MB, making room for approx 8 big 64K
798 	 * fragments 8x128k.
799 	 */
800 	net->ipv4.frags.high_thresh = 4 * 1024 * 1024;
801 	net->ipv4.frags.low_thresh  = 3 * 1024 * 1024;
802 	/*
803 	 * Important NOTE! Fragment queue must be destroyed before MSL expires.
804 	 * RFC791 is wrong proposing to prolongate timer each fragment arrival
805 	 * by TTL.
806 	 */
807 	net->ipv4.frags.timeout = IP_FRAG_TIME;
808 
809 	net->ipv4.frags.max_dist = 64;
810 	net->ipv4.frags.f = &ip4_frags;
811 
812 	res = inet_frags_init_net(&net->ipv4.frags);
813 	if (res < 0)
814 		return res;
815 	res = ip4_frags_ns_ctl_register(net);
816 	if (res < 0)
817 		inet_frags_exit_net(&net->ipv4.frags);
818 	return res;
819 }
820 
821 static void __net_exit ipv4_frags_exit_net(struct net *net)
822 {
823 	ip4_frags_ns_ctl_unregister(net);
824 	inet_frags_exit_net(&net->ipv4.frags);
825 }
826 
827 static struct pernet_operations ip4_frags_ops = {
828 	.init = ipv4_frags_init_net,
829 	.exit = ipv4_frags_exit_net,
830 };
831 
832 
833 static u32 ip4_key_hashfn(const void *data, u32 len, u32 seed)
834 {
835 	return jhash2(data,
836 		      sizeof(struct frag_v4_compare_key) / sizeof(u32), seed);
837 }
838 
839 static u32 ip4_obj_hashfn(const void *data, u32 len, u32 seed)
840 {
841 	const struct inet_frag_queue *fq = data;
842 
843 	return jhash2((const u32 *)&fq->key.v4,
844 		      sizeof(struct frag_v4_compare_key) / sizeof(u32), seed);
845 }
846 
847 static int ip4_obj_cmpfn(struct rhashtable_compare_arg *arg, const void *ptr)
848 {
849 	const struct frag_v4_compare_key *key = arg->key;
850 	const struct inet_frag_queue *fq = ptr;
851 
852 	return !!memcmp(&fq->key, key, sizeof(*key));
853 }
854 
855 static const struct rhashtable_params ip4_rhash_params = {
856 	.head_offset		= offsetof(struct inet_frag_queue, node),
857 	.key_offset		= offsetof(struct inet_frag_queue, key),
858 	.key_len		= sizeof(struct frag_v4_compare_key),
859 	.hashfn			= ip4_key_hashfn,
860 	.obj_hashfn		= ip4_obj_hashfn,
861 	.obj_cmpfn		= ip4_obj_cmpfn,
862 	.automatic_shrinking	= true,
863 };
864 
865 void __init ipfrag_init(void)
866 {
867 	ip4_frags.constructor = ip4_frag_init;
868 	ip4_frags.destructor = ip4_frag_free;
869 	ip4_frags.qsize = sizeof(struct ipq);
870 	ip4_frags.frag_expire = ip_expire;
871 	ip4_frags.frags_cache_name = ip_frag_cache_name;
872 	ip4_frags.rhash_params = ip4_rhash_params;
873 	if (inet_frags_init(&ip4_frags))
874 		panic("IP: failed to allocate ip4_frags cache\n");
875 	ip4_frags_ctl_register();
876 	register_pernet_subsys(&ip4_frags_ops);
877 }
878