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