xref: /linux/net/ipv4/ip_fragment.c (revision a1ff5a7d78a036d6c2178ee5acd6ba4946243800)
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 
ip4_frag_ecn(u8 tos)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 *skb,
79 			 struct sk_buff *prev_tail, struct net_device *dev);
80 
81 
ip4_frag_init(struct inet_frag_queue * q,const void * a)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 net *net = q->fqdir->net;
86 
87 	const struct frag_v4_compare_key *key = a;
88 
89 	q->key.v4 = *key;
90 	qp->ecn = 0;
91 	qp->peer = q->fqdir->max_dist ?
92 		inet_getpeer_v4(net->ipv4.peers, key->saddr, key->vif, 1) :
93 		NULL;
94 }
95 
ip4_frag_free(struct inet_frag_queue * q)96 static void ip4_frag_free(struct inet_frag_queue *q)
97 {
98 	struct ipq *qp;
99 
100 	qp = container_of(q, struct ipq, q);
101 	if (qp->peer)
102 		inet_putpeer(qp->peer);
103 }
104 
105 
106 /* Destruction primitives. */
107 
ipq_put(struct ipq * ipq)108 static void ipq_put(struct ipq *ipq)
109 {
110 	inet_frag_put(&ipq->q);
111 }
112 
113 /* Kill ipq entry. It is not destroyed immediately,
114  * because caller (and someone more) holds reference count.
115  */
ipq_kill(struct ipq * ipq)116 static void ipq_kill(struct ipq *ipq)
117 {
118 	inet_frag_kill(&ipq->q);
119 }
120 
frag_expire_skip_icmp(u32 user)121 static bool frag_expire_skip_icmp(u32 user)
122 {
123 	return user == IP_DEFRAG_AF_PACKET ||
124 	       ip_defrag_user_in_between(user, IP_DEFRAG_CONNTRACK_IN,
125 					 __IP_DEFRAG_CONNTRACK_IN_END) ||
126 	       ip_defrag_user_in_between(user, IP_DEFRAG_CONNTRACK_BRIDGE_IN,
127 					 __IP_DEFRAG_CONNTRACK_BRIDGE_IN);
128 }
129 
130 /*
131  * Oops, a fragment queue timed out.  Kill it and send an ICMP reply.
132  */
ip_expire(struct timer_list * t)133 static void ip_expire(struct timer_list *t)
134 {
135 	struct inet_frag_queue *frag = from_timer(frag, t, timer);
136 	const struct iphdr *iph;
137 	struct sk_buff *head = NULL;
138 	struct net *net;
139 	struct ipq *qp;
140 	int err;
141 
142 	qp = container_of(frag, struct ipq, q);
143 	net = qp->q.fqdir->net;
144 
145 	rcu_read_lock();
146 
147 	/* Paired with WRITE_ONCE() in fqdir_pre_exit(). */
148 	if (READ_ONCE(qp->q.fqdir->dead))
149 		goto out_rcu_unlock;
150 
151 	spin_lock(&qp->q.lock);
152 
153 	if (qp->q.flags & INET_FRAG_COMPLETE)
154 		goto out;
155 
156 	qp->q.flags |= INET_FRAG_DROP;
157 	ipq_kill(qp);
158 	__IP_INC_STATS(net, IPSTATS_MIB_REASMFAILS);
159 	__IP_INC_STATS(net, IPSTATS_MIB_REASMTIMEOUT);
160 
161 	if (!(qp->q.flags & INET_FRAG_FIRST_IN))
162 		goto out;
163 
164 	/* sk_buff::dev and sk_buff::rbnode are unionized. So we
165 	 * pull the head out of the tree in order to be able to
166 	 * deal with head->dev.
167 	 */
168 	head = inet_frag_pull_head(&qp->q);
169 	if (!head)
170 		goto out;
171 	head->dev = dev_get_by_index_rcu(net, qp->iif);
172 	if (!head->dev)
173 		goto out;
174 
175 
176 	/* skb has no dst, perform route lookup again */
177 	iph = ip_hdr(head);
178 	err = ip_route_input_noref(head, iph->daddr, iph->saddr,
179 					   iph->tos, head->dev);
180 	if (err)
181 		goto out;
182 
183 	/* Only an end host needs to send an ICMP
184 	 * "Fragment Reassembly Timeout" message, per RFC792.
185 	 */
186 	if (frag_expire_skip_icmp(qp->q.key.v4.user) &&
187 	    (skb_rtable(head)->rt_type != RTN_LOCAL))
188 		goto out;
189 
190 	spin_unlock(&qp->q.lock);
191 	icmp_send(head, ICMP_TIME_EXCEEDED, ICMP_EXC_FRAGTIME, 0);
192 	goto out_rcu_unlock;
193 
194 out:
195 	spin_unlock(&qp->q.lock);
196 out_rcu_unlock:
197 	rcu_read_unlock();
198 	kfree_skb_reason(head, SKB_DROP_REASON_FRAG_REASM_TIMEOUT);
199 	ipq_put(qp);
200 }
201 
202 /* Find the correct entry in the "incomplete datagrams" queue for
203  * this IP datagram, and create new one, if nothing is found.
204  */
ip_find(struct net * net,struct iphdr * iph,u32 user,int vif)205 static struct ipq *ip_find(struct net *net, struct iphdr *iph,
206 			   u32 user, int vif)
207 {
208 	struct frag_v4_compare_key key = {
209 		.saddr = iph->saddr,
210 		.daddr = iph->daddr,
211 		.user = user,
212 		.vif = vif,
213 		.id = iph->id,
214 		.protocol = iph->protocol,
215 	};
216 	struct inet_frag_queue *q;
217 
218 	q = inet_frag_find(net->ipv4.fqdir, &key);
219 	if (!q)
220 		return NULL;
221 
222 	return container_of(q, struct ipq, q);
223 }
224 
225 /* Is the fragment too far ahead to be part of ipq? */
ip_frag_too_far(struct ipq * qp)226 static int ip_frag_too_far(struct ipq *qp)
227 {
228 	struct inet_peer *peer = qp->peer;
229 	unsigned int max = qp->q.fqdir->max_dist;
230 	unsigned int start, end;
231 
232 	int rc;
233 
234 	if (!peer || !max)
235 		return 0;
236 
237 	start = qp->rid;
238 	end = atomic_inc_return(&peer->rid);
239 	qp->rid = end;
240 
241 	rc = qp->q.fragments_tail && (end - start) > max;
242 
243 	if (rc)
244 		__IP_INC_STATS(qp->q.fqdir->net, IPSTATS_MIB_REASMFAILS);
245 
246 	return rc;
247 }
248 
ip_frag_reinit(struct ipq * qp)249 static int ip_frag_reinit(struct ipq *qp)
250 {
251 	unsigned int sum_truesize = 0;
252 
253 	if (!mod_timer(&qp->q.timer, jiffies + qp->q.fqdir->timeout)) {
254 		refcount_inc(&qp->q.refcnt);
255 		return -ETIMEDOUT;
256 	}
257 
258 	sum_truesize = inet_frag_rbtree_purge(&qp->q.rb_fragments,
259 					      SKB_DROP_REASON_FRAG_TOO_FAR);
260 	sub_frag_mem_limit(qp->q.fqdir, sum_truesize);
261 
262 	qp->q.flags = 0;
263 	qp->q.len = 0;
264 	qp->q.meat = 0;
265 	qp->q.rb_fragments = RB_ROOT;
266 	qp->q.fragments_tail = NULL;
267 	qp->q.last_run_head = NULL;
268 	qp->iif = 0;
269 	qp->ecn = 0;
270 
271 	return 0;
272 }
273 
274 /* Add new segment to existing queue. */
ip_frag_queue(struct ipq * qp,struct sk_buff * skb)275 static int ip_frag_queue(struct ipq *qp, struct sk_buff *skb)
276 {
277 	struct net *net = qp->q.fqdir->net;
278 	int ihl, end, flags, offset;
279 	struct sk_buff *prev_tail;
280 	struct net_device *dev;
281 	unsigned int fragsize;
282 	int err = -ENOENT;
283 	SKB_DR(reason);
284 	u8 ecn;
285 
286 	/* If reassembly is already done, @skb must be a duplicate frag. */
287 	if (qp->q.flags & INET_FRAG_COMPLETE) {
288 		SKB_DR_SET(reason, DUP_FRAG);
289 		goto err;
290 	}
291 
292 	if (!(IPCB(skb)->flags & IPSKB_FRAG_COMPLETE) &&
293 	    unlikely(ip_frag_too_far(qp)) &&
294 	    unlikely(err = ip_frag_reinit(qp))) {
295 		ipq_kill(qp);
296 		goto err;
297 	}
298 
299 	ecn = ip4_frag_ecn(ip_hdr(skb)->tos);
300 	offset = ntohs(ip_hdr(skb)->frag_off);
301 	flags = offset & ~IP_OFFSET;
302 	offset &= IP_OFFSET;
303 	offset <<= 3;		/* offset is in 8-byte chunks */
304 	ihl = ip_hdrlen(skb);
305 
306 	/* Determine the position of this fragment. */
307 	end = offset + skb->len - skb_network_offset(skb) - ihl;
308 	err = -EINVAL;
309 
310 	/* Is this the final fragment? */
311 	if ((flags & IP_MF) == 0) {
312 		/* If we already have some bits beyond end
313 		 * or have different end, the segment is corrupted.
314 		 */
315 		if (end < qp->q.len ||
316 		    ((qp->q.flags & INET_FRAG_LAST_IN) && end != qp->q.len))
317 			goto discard_qp;
318 		qp->q.flags |= INET_FRAG_LAST_IN;
319 		qp->q.len = end;
320 	} else {
321 		if (end&7) {
322 			end &= ~7;
323 			if (skb->ip_summed != CHECKSUM_UNNECESSARY)
324 				skb->ip_summed = CHECKSUM_NONE;
325 		}
326 		if (end > qp->q.len) {
327 			/* Some bits beyond end -> corruption. */
328 			if (qp->q.flags & INET_FRAG_LAST_IN)
329 				goto discard_qp;
330 			qp->q.len = end;
331 		}
332 	}
333 	if (end == offset)
334 		goto discard_qp;
335 
336 	err = -ENOMEM;
337 	if (!pskb_pull(skb, skb_network_offset(skb) + ihl))
338 		goto discard_qp;
339 
340 	err = pskb_trim_rcsum(skb, end - offset);
341 	if (err)
342 		goto discard_qp;
343 
344 	/* Note : skb->rbnode and skb->dev share the same location. */
345 	dev = skb->dev;
346 	/* Makes sure compiler wont do silly aliasing games */
347 	barrier();
348 
349 	prev_tail = qp->q.fragments_tail;
350 	err = inet_frag_queue_insert(&qp->q, skb, offset, end);
351 	if (err)
352 		goto insert_error;
353 
354 	if (dev)
355 		qp->iif = dev->ifindex;
356 
357 	qp->q.stamp = skb->tstamp;
358 	qp->q.tstamp_type = skb->tstamp_type;
359 	qp->q.meat += skb->len;
360 	qp->ecn |= ecn;
361 	add_frag_mem_limit(qp->q.fqdir, skb->truesize);
362 	if (offset == 0)
363 		qp->q.flags |= INET_FRAG_FIRST_IN;
364 
365 	fragsize = skb->len + ihl;
366 
367 	if (fragsize > qp->q.max_size)
368 		qp->q.max_size = fragsize;
369 
370 	if (ip_hdr(skb)->frag_off & htons(IP_DF) &&
371 	    fragsize > qp->max_df_size)
372 		qp->max_df_size = fragsize;
373 
374 	if (qp->q.flags == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
375 	    qp->q.meat == qp->q.len) {
376 		unsigned long orefdst = skb->_skb_refdst;
377 
378 		skb->_skb_refdst = 0UL;
379 		err = ip_frag_reasm(qp, skb, prev_tail, dev);
380 		skb->_skb_refdst = orefdst;
381 		if (err)
382 			inet_frag_kill(&qp->q);
383 		return err;
384 	}
385 
386 	skb_dst_drop(skb);
387 	skb_orphan(skb);
388 	return -EINPROGRESS;
389 
390 insert_error:
391 	if (err == IPFRAG_DUP) {
392 		SKB_DR_SET(reason, DUP_FRAG);
393 		err = -EINVAL;
394 		goto err;
395 	}
396 	err = -EINVAL;
397 	__IP_INC_STATS(net, IPSTATS_MIB_REASM_OVERLAPS);
398 discard_qp:
399 	inet_frag_kill(&qp->q);
400 	__IP_INC_STATS(net, IPSTATS_MIB_REASMFAILS);
401 err:
402 	kfree_skb_reason(skb, reason);
403 	return err;
404 }
405 
ip_frag_coalesce_ok(const struct ipq * qp)406 static bool ip_frag_coalesce_ok(const struct ipq *qp)
407 {
408 	return qp->q.key.v4.user == IP_DEFRAG_LOCAL_DELIVER;
409 }
410 
411 /* Build a new IP datagram from all its fragments. */
ip_frag_reasm(struct ipq * qp,struct sk_buff * skb,struct sk_buff * prev_tail,struct net_device * dev)412 static int ip_frag_reasm(struct ipq *qp, struct sk_buff *skb,
413 			 struct sk_buff *prev_tail, struct net_device *dev)
414 {
415 	struct net *net = qp->q.fqdir->net;
416 	struct iphdr *iph;
417 	void *reasm_data;
418 	int len, err;
419 	u8 ecn;
420 
421 	ipq_kill(qp);
422 
423 	ecn = ip_frag_ecn_table[qp->ecn];
424 	if (unlikely(ecn == 0xff)) {
425 		err = -EINVAL;
426 		goto out_fail;
427 	}
428 
429 	/* Make the one we just received the head. */
430 	reasm_data = inet_frag_reasm_prepare(&qp->q, skb, prev_tail);
431 	if (!reasm_data)
432 		goto out_nomem;
433 
434 	len = ip_hdrlen(skb) + qp->q.len;
435 	err = -E2BIG;
436 	if (len > 65535)
437 		goto out_oversize;
438 
439 	inet_frag_reasm_finish(&qp->q, skb, reasm_data,
440 			       ip_frag_coalesce_ok(qp));
441 
442 	skb->dev = dev;
443 	IPCB(skb)->frag_max_size = max(qp->max_df_size, qp->q.max_size);
444 
445 	iph = ip_hdr(skb);
446 	iph->tot_len = htons(len);
447 	iph->tos |= ecn;
448 
449 	/* When we set IP_DF on a refragmented skb we must also force a
450 	 * call to ip_fragment to avoid forwarding a DF-skb of size s while
451 	 * original sender only sent fragments of size f (where f < s).
452 	 *
453 	 * We only set DF/IPSKB_FRAG_PMTU if such DF fragment was the largest
454 	 * frag seen to avoid sending tiny DF-fragments in case skb was built
455 	 * from one very small df-fragment and one large non-df frag.
456 	 */
457 	if (qp->max_df_size == qp->q.max_size) {
458 		IPCB(skb)->flags |= IPSKB_FRAG_PMTU;
459 		iph->frag_off = htons(IP_DF);
460 	} else {
461 		iph->frag_off = 0;
462 	}
463 
464 	ip_send_check(iph);
465 
466 	__IP_INC_STATS(net, IPSTATS_MIB_REASMOKS);
467 	qp->q.rb_fragments = RB_ROOT;
468 	qp->q.fragments_tail = NULL;
469 	qp->q.last_run_head = NULL;
470 	return 0;
471 
472 out_nomem:
473 	net_dbg_ratelimited("queue_glue: no memory for gluing queue %p\n", qp);
474 	err = -ENOMEM;
475 	goto out_fail;
476 out_oversize:
477 	net_info_ratelimited("Oversized IP packet from %pI4\n", &qp->q.key.v4.saddr);
478 out_fail:
479 	__IP_INC_STATS(net, IPSTATS_MIB_REASMFAILS);
480 	return err;
481 }
482 
483 /* Process an incoming IP datagram fragment. */
ip_defrag(struct net * net,struct sk_buff * skb,u32 user)484 int ip_defrag(struct net *net, struct sk_buff *skb, u32 user)
485 {
486 	struct net_device *dev = skb->dev ? : skb_dst(skb)->dev;
487 	int vif = l3mdev_master_ifindex_rcu(dev);
488 	struct ipq *qp;
489 
490 	__IP_INC_STATS(net, IPSTATS_MIB_REASMREQDS);
491 
492 	/* Lookup (or create) queue header */
493 	qp = ip_find(net, ip_hdr(skb), user, vif);
494 	if (qp) {
495 		int ret;
496 
497 		spin_lock(&qp->q.lock);
498 
499 		ret = ip_frag_queue(qp, skb);
500 
501 		spin_unlock(&qp->q.lock);
502 		ipq_put(qp);
503 		return ret;
504 	}
505 
506 	__IP_INC_STATS(net, IPSTATS_MIB_REASMFAILS);
507 	kfree_skb(skb);
508 	return -ENOMEM;
509 }
510 EXPORT_SYMBOL(ip_defrag);
511 
ip_check_defrag(struct net * net,struct sk_buff * skb,u32 user)512 struct sk_buff *ip_check_defrag(struct net *net, struct sk_buff *skb, u32 user)
513 {
514 	struct iphdr iph;
515 	int netoff;
516 	u32 len;
517 
518 	if (skb->protocol != htons(ETH_P_IP))
519 		return skb;
520 
521 	netoff = skb_network_offset(skb);
522 
523 	if (skb_copy_bits(skb, netoff, &iph, sizeof(iph)) < 0)
524 		return skb;
525 
526 	if (iph.ihl < 5 || iph.version != 4)
527 		return skb;
528 
529 	len = ntohs(iph.tot_len);
530 	if (skb->len < netoff + len || len < (iph.ihl * 4))
531 		return skb;
532 
533 	if (ip_is_fragment(&iph)) {
534 		skb = skb_share_check(skb, GFP_ATOMIC);
535 		if (skb) {
536 			if (!pskb_may_pull(skb, netoff + iph.ihl * 4)) {
537 				kfree_skb(skb);
538 				return NULL;
539 			}
540 			if (pskb_trim_rcsum(skb, netoff + len)) {
541 				kfree_skb(skb);
542 				return NULL;
543 			}
544 			memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
545 			if (ip_defrag(net, skb, user))
546 				return NULL;
547 			skb_clear_hash(skb);
548 		}
549 	}
550 	return skb;
551 }
552 EXPORT_SYMBOL(ip_check_defrag);
553 
554 #ifdef CONFIG_SYSCTL
555 static int dist_min;
556 
557 static struct ctl_table ip4_frags_ns_ctl_table[] = {
558 	{
559 		.procname	= "ipfrag_high_thresh",
560 		.maxlen		= sizeof(unsigned long),
561 		.mode		= 0644,
562 		.proc_handler	= proc_doulongvec_minmax,
563 	},
564 	{
565 		.procname	= "ipfrag_low_thresh",
566 		.maxlen		= sizeof(unsigned long),
567 		.mode		= 0644,
568 		.proc_handler	= proc_doulongvec_minmax,
569 	},
570 	{
571 		.procname	= "ipfrag_time",
572 		.maxlen		= sizeof(int),
573 		.mode		= 0644,
574 		.proc_handler	= proc_dointvec_jiffies,
575 	},
576 	{
577 		.procname	= "ipfrag_max_dist",
578 		.maxlen		= sizeof(int),
579 		.mode		= 0644,
580 		.proc_handler	= proc_dointvec_minmax,
581 		.extra1		= &dist_min,
582 	},
583 };
584 
585 /* secret interval has been deprecated */
586 static int ip4_frags_secret_interval_unused;
587 static struct ctl_table ip4_frags_ctl_table[] = {
588 	{
589 		.procname	= "ipfrag_secret_interval",
590 		.data		= &ip4_frags_secret_interval_unused,
591 		.maxlen		= sizeof(int),
592 		.mode		= 0644,
593 		.proc_handler	= proc_dointvec_jiffies,
594 	},
595 };
596 
ip4_frags_ns_ctl_register(struct net * net)597 static int __net_init ip4_frags_ns_ctl_register(struct net *net)
598 {
599 	struct ctl_table *table;
600 	struct ctl_table_header *hdr;
601 
602 	table = ip4_frags_ns_ctl_table;
603 	if (!net_eq(net, &init_net)) {
604 		table = kmemdup(table, sizeof(ip4_frags_ns_ctl_table), GFP_KERNEL);
605 		if (!table)
606 			goto err_alloc;
607 
608 	}
609 	table[0].data	= &net->ipv4.fqdir->high_thresh;
610 	table[0].extra1	= &net->ipv4.fqdir->low_thresh;
611 	table[1].data	= &net->ipv4.fqdir->low_thresh;
612 	table[1].extra2	= &net->ipv4.fqdir->high_thresh;
613 	table[2].data	= &net->ipv4.fqdir->timeout;
614 	table[3].data	= &net->ipv4.fqdir->max_dist;
615 
616 	hdr = register_net_sysctl_sz(net, "net/ipv4", table,
617 				     ARRAY_SIZE(ip4_frags_ns_ctl_table));
618 	if (!hdr)
619 		goto err_reg;
620 
621 	net->ipv4.frags_hdr = hdr;
622 	return 0;
623 
624 err_reg:
625 	if (!net_eq(net, &init_net))
626 		kfree(table);
627 err_alloc:
628 	return -ENOMEM;
629 }
630 
ip4_frags_ns_ctl_unregister(struct net * net)631 static void __net_exit ip4_frags_ns_ctl_unregister(struct net *net)
632 {
633 	const struct ctl_table *table;
634 
635 	table = net->ipv4.frags_hdr->ctl_table_arg;
636 	unregister_net_sysctl_table(net->ipv4.frags_hdr);
637 	kfree(table);
638 }
639 
ip4_frags_ctl_register(void)640 static void __init ip4_frags_ctl_register(void)
641 {
642 	register_net_sysctl(&init_net, "net/ipv4", ip4_frags_ctl_table);
643 }
644 #else
ip4_frags_ns_ctl_register(struct net * net)645 static int ip4_frags_ns_ctl_register(struct net *net)
646 {
647 	return 0;
648 }
649 
ip4_frags_ns_ctl_unregister(struct net * net)650 static void ip4_frags_ns_ctl_unregister(struct net *net)
651 {
652 }
653 
ip4_frags_ctl_register(void)654 static void __init ip4_frags_ctl_register(void)
655 {
656 }
657 #endif
658 
ipv4_frags_init_net(struct net * net)659 static int __net_init ipv4_frags_init_net(struct net *net)
660 {
661 	int res;
662 
663 	res = fqdir_init(&net->ipv4.fqdir, &ip4_frags, net);
664 	if (res < 0)
665 		return res;
666 	/* Fragment cache limits.
667 	 *
668 	 * The fragment memory accounting code, (tries to) account for
669 	 * the real memory usage, by measuring both the size of frag
670 	 * queue struct (inet_frag_queue (ipv4:ipq/ipv6:frag_queue))
671 	 * and the SKB's truesize.
672 	 *
673 	 * A 64K fragment consumes 129736 bytes (44*2944)+200
674 	 * (1500 truesize == 2944, sizeof(struct ipq) == 200)
675 	 *
676 	 * We will commit 4MB at one time. Should we cross that limit
677 	 * we will prune down to 3MB, making room for approx 8 big 64K
678 	 * fragments 8x128k.
679 	 */
680 	net->ipv4.fqdir->high_thresh = 4 * 1024 * 1024;
681 	net->ipv4.fqdir->low_thresh  = 3 * 1024 * 1024;
682 	/*
683 	 * Important NOTE! Fragment queue must be destroyed before MSL expires.
684 	 * RFC791 is wrong proposing to prolongate timer each fragment arrival
685 	 * by TTL.
686 	 */
687 	net->ipv4.fqdir->timeout = IP_FRAG_TIME;
688 
689 	net->ipv4.fqdir->max_dist = 64;
690 
691 	res = ip4_frags_ns_ctl_register(net);
692 	if (res < 0)
693 		fqdir_exit(net->ipv4.fqdir);
694 	return res;
695 }
696 
ipv4_frags_pre_exit_net(struct net * net)697 static void __net_exit ipv4_frags_pre_exit_net(struct net *net)
698 {
699 	fqdir_pre_exit(net->ipv4.fqdir);
700 }
701 
ipv4_frags_exit_net(struct net * net)702 static void __net_exit ipv4_frags_exit_net(struct net *net)
703 {
704 	ip4_frags_ns_ctl_unregister(net);
705 	fqdir_exit(net->ipv4.fqdir);
706 }
707 
708 static struct pernet_operations ip4_frags_ops = {
709 	.init		= ipv4_frags_init_net,
710 	.pre_exit	= ipv4_frags_pre_exit_net,
711 	.exit		= ipv4_frags_exit_net,
712 };
713 
714 
ip4_key_hashfn(const void * data,u32 len,u32 seed)715 static u32 ip4_key_hashfn(const void *data, u32 len, u32 seed)
716 {
717 	return jhash2(data,
718 		      sizeof(struct frag_v4_compare_key) / sizeof(u32), seed);
719 }
720 
ip4_obj_hashfn(const void * data,u32 len,u32 seed)721 static u32 ip4_obj_hashfn(const void *data, u32 len, u32 seed)
722 {
723 	const struct inet_frag_queue *fq = data;
724 
725 	return jhash2((const u32 *)&fq->key.v4,
726 		      sizeof(struct frag_v4_compare_key) / sizeof(u32), seed);
727 }
728 
ip4_obj_cmpfn(struct rhashtable_compare_arg * arg,const void * ptr)729 static int ip4_obj_cmpfn(struct rhashtable_compare_arg *arg, const void *ptr)
730 {
731 	const struct frag_v4_compare_key *key = arg->key;
732 	const struct inet_frag_queue *fq = ptr;
733 
734 	return !!memcmp(&fq->key, key, sizeof(*key));
735 }
736 
737 static const struct rhashtable_params ip4_rhash_params = {
738 	.head_offset		= offsetof(struct inet_frag_queue, node),
739 	.key_offset		= offsetof(struct inet_frag_queue, key),
740 	.key_len		= sizeof(struct frag_v4_compare_key),
741 	.hashfn			= ip4_key_hashfn,
742 	.obj_hashfn		= ip4_obj_hashfn,
743 	.obj_cmpfn		= ip4_obj_cmpfn,
744 	.automatic_shrinking	= true,
745 };
746 
ipfrag_init(void)747 void __init ipfrag_init(void)
748 {
749 	ip4_frags.constructor = ip4_frag_init;
750 	ip4_frags.destructor = ip4_frag_free;
751 	ip4_frags.qsize = sizeof(struct ipq);
752 	ip4_frags.frag_expire = ip_expire;
753 	ip4_frags.frags_cache_name = ip_frag_cache_name;
754 	ip4_frags.rhash_params = ip4_rhash_params;
755 	if (inet_frags_init(&ip4_frags))
756 		panic("IP: failed to allocate ip4_frags cache\n");
757 	ip4_frags_ctl_register();
758 	register_pernet_subsys(&ip4_frags_ops);
759 }
760