xref: /openbmc/linux/net/ipv4/ip_fragment.c (revision 060f35a317ef09101b128f399dce7ed13d019461)
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 	const struct frag_v4_compare_key *key = a;
86 	struct net *net = q->fqdir->net;
87 	struct inet_peer *p = NULL;
88 
89 	q->key.v4 = *key;
90 	qp->ecn = 0;
91 	if (q->fqdir->max_dist) {
92 		rcu_read_lock();
93 		p = inet_getpeer_v4(net->ipv4.peers, key->saddr, key->vif);
94 		if (p && !refcount_inc_not_zero(&p->refcnt))
95 			p = NULL;
96 		rcu_read_unlock();
97 	}
98 	qp->peer = p;
99 }
100 
ip4_frag_free(struct inet_frag_queue * q)101 static void ip4_frag_free(struct inet_frag_queue *q)
102 {
103 	struct ipq *qp;
104 
105 	qp = container_of(q, struct ipq, q);
106 	if (qp->peer)
107 		inet_putpeer(qp->peer);
108 }
109 
110 
111 /* Destruction primitives. */
112 
ipq_put(struct ipq * ipq)113 static void ipq_put(struct ipq *ipq)
114 {
115 	inet_frag_put(&ipq->q);
116 }
117 
118 /* Kill ipq entry. It is not destroyed immediately,
119  * because caller (and someone more) holds reference count.
120  */
ipq_kill(struct ipq * ipq)121 static void ipq_kill(struct ipq *ipq)
122 {
123 	inet_frag_kill(&ipq->q);
124 }
125 
frag_expire_skip_icmp(u32 user)126 static bool frag_expire_skip_icmp(u32 user)
127 {
128 	return user == IP_DEFRAG_AF_PACKET ||
129 	       ip_defrag_user_in_between(user, IP_DEFRAG_CONNTRACK_IN,
130 					 __IP_DEFRAG_CONNTRACK_IN_END) ||
131 	       ip_defrag_user_in_between(user, IP_DEFRAG_CONNTRACK_BRIDGE_IN,
132 					 __IP_DEFRAG_CONNTRACK_BRIDGE_IN);
133 }
134 
135 /*
136  * Oops, a fragment queue timed out.  Kill it and send an ICMP reply.
137  */
ip_expire(struct timer_list * t)138 static void ip_expire(struct timer_list *t)
139 {
140 	struct inet_frag_queue *frag = from_timer(frag, t, timer);
141 	const struct iphdr *iph;
142 	struct sk_buff *head = NULL;
143 	struct net *net;
144 	struct ipq *qp;
145 	int err;
146 
147 	qp = container_of(frag, struct ipq, q);
148 	net = qp->q.fqdir->net;
149 
150 	rcu_read_lock();
151 
152 	/* Paired with WRITE_ONCE() in fqdir_pre_exit(). */
153 	if (READ_ONCE(qp->q.fqdir->dead))
154 		goto out_rcu_unlock;
155 
156 	spin_lock(&qp->q.lock);
157 
158 	if (qp->q.flags & INET_FRAG_COMPLETE)
159 		goto out;
160 
161 	qp->q.flags |= INET_FRAG_DROP;
162 	ipq_kill(qp);
163 	__IP_INC_STATS(net, IPSTATS_MIB_REASMFAILS);
164 	__IP_INC_STATS(net, IPSTATS_MIB_REASMTIMEOUT);
165 
166 	if (!(qp->q.flags & INET_FRAG_FIRST_IN))
167 		goto out;
168 
169 	/* sk_buff::dev and sk_buff::rbnode are unionized. So we
170 	 * pull the head out of the tree in order to be able to
171 	 * deal with head->dev.
172 	 */
173 	head = inet_frag_pull_head(&qp->q);
174 	if (!head)
175 		goto out;
176 	head->dev = dev_get_by_index_rcu(net, qp->iif);
177 	if (!head->dev)
178 		goto out;
179 
180 
181 	/* skb has no dst, perform route lookup again */
182 	iph = ip_hdr(head);
183 	err = ip_route_input_noref(head, iph->daddr, iph->saddr,
184 					   iph->tos, head->dev);
185 	if (err)
186 		goto out;
187 
188 	/* Only an end host needs to send an ICMP
189 	 * "Fragment Reassembly Timeout" message, per RFC792.
190 	 */
191 	if (frag_expire_skip_icmp(qp->q.key.v4.user) &&
192 	    (skb_rtable(head)->rt_type != RTN_LOCAL))
193 		goto out;
194 
195 	spin_unlock(&qp->q.lock);
196 	icmp_send(head, ICMP_TIME_EXCEEDED, ICMP_EXC_FRAGTIME, 0);
197 	goto out_rcu_unlock;
198 
199 out:
200 	spin_unlock(&qp->q.lock);
201 out_rcu_unlock:
202 	rcu_read_unlock();
203 	kfree_skb_reason(head, SKB_DROP_REASON_FRAG_REASM_TIMEOUT);
204 	ipq_put(qp);
205 }
206 
207 /* Find the correct entry in the "incomplete datagrams" queue for
208  * this IP datagram, and create new one, if nothing is found.
209  */
ip_find(struct net * net,struct iphdr * iph,u32 user,int vif)210 static struct ipq *ip_find(struct net *net, struct iphdr *iph,
211 			   u32 user, int vif)
212 {
213 	struct frag_v4_compare_key key = {
214 		.saddr = iph->saddr,
215 		.daddr = iph->daddr,
216 		.user = user,
217 		.vif = vif,
218 		.id = iph->id,
219 		.protocol = iph->protocol,
220 	};
221 	struct inet_frag_queue *q;
222 
223 	q = inet_frag_find(net->ipv4.fqdir, &key);
224 	if (!q)
225 		return NULL;
226 
227 	return container_of(q, struct ipq, q);
228 }
229 
230 /* Is the fragment too far ahead to be part of ipq? */
ip_frag_too_far(struct ipq * qp)231 static int ip_frag_too_far(struct ipq *qp)
232 {
233 	struct inet_peer *peer = qp->peer;
234 	unsigned int max = qp->q.fqdir->max_dist;
235 	unsigned int start, end;
236 
237 	int rc;
238 
239 	if (!peer || !max)
240 		return 0;
241 
242 	start = qp->rid;
243 	end = atomic_inc_return(&peer->rid);
244 	qp->rid = end;
245 
246 	rc = qp->q.fragments_tail && (end - start) > max;
247 
248 	if (rc)
249 		__IP_INC_STATS(qp->q.fqdir->net, IPSTATS_MIB_REASMFAILS);
250 
251 	return rc;
252 }
253 
ip_frag_reinit(struct ipq * qp)254 static int ip_frag_reinit(struct ipq *qp)
255 {
256 	unsigned int sum_truesize = 0;
257 
258 	if (!mod_timer(&qp->q.timer, jiffies + qp->q.fqdir->timeout)) {
259 		refcount_inc(&qp->q.refcnt);
260 		return -ETIMEDOUT;
261 	}
262 
263 	sum_truesize = inet_frag_rbtree_purge(&qp->q.rb_fragments,
264 					      SKB_DROP_REASON_FRAG_TOO_FAR);
265 	sub_frag_mem_limit(qp->q.fqdir, sum_truesize);
266 
267 	qp->q.flags = 0;
268 	qp->q.len = 0;
269 	qp->q.meat = 0;
270 	qp->q.rb_fragments = RB_ROOT;
271 	qp->q.fragments_tail = NULL;
272 	qp->q.last_run_head = NULL;
273 	qp->iif = 0;
274 	qp->ecn = 0;
275 
276 	return 0;
277 }
278 
279 /* Add new segment to existing queue. */
ip_frag_queue(struct ipq * qp,struct sk_buff * skb)280 static int ip_frag_queue(struct ipq *qp, struct sk_buff *skb)
281 {
282 	struct net *net = qp->q.fqdir->net;
283 	int ihl, end, flags, offset;
284 	struct sk_buff *prev_tail;
285 	struct net_device *dev;
286 	unsigned int fragsize;
287 	int err = -ENOENT;
288 	SKB_DR(reason);
289 	u8 ecn;
290 
291 	/* If reassembly is already done, @skb must be a duplicate frag. */
292 	if (qp->q.flags & INET_FRAG_COMPLETE) {
293 		SKB_DR_SET(reason, DUP_FRAG);
294 		goto err;
295 	}
296 
297 	if (!(IPCB(skb)->flags & IPSKB_FRAG_COMPLETE) &&
298 	    unlikely(ip_frag_too_far(qp)) &&
299 	    unlikely(err = ip_frag_reinit(qp))) {
300 		ipq_kill(qp);
301 		goto err;
302 	}
303 
304 	ecn = ip4_frag_ecn(ip_hdr(skb)->tos);
305 	offset = ntohs(ip_hdr(skb)->frag_off);
306 	flags = offset & ~IP_OFFSET;
307 	offset &= IP_OFFSET;
308 	offset <<= 3;		/* offset is in 8-byte chunks */
309 	ihl = ip_hdrlen(skb);
310 
311 	/* Determine the position of this fragment. */
312 	end = offset + skb->len - skb_network_offset(skb) - ihl;
313 	err = -EINVAL;
314 
315 	/* Is this the final fragment? */
316 	if ((flags & IP_MF) == 0) {
317 		/* If we already have some bits beyond end
318 		 * or have different end, the segment is corrupted.
319 		 */
320 		if (end < qp->q.len ||
321 		    ((qp->q.flags & INET_FRAG_LAST_IN) && end != qp->q.len))
322 			goto discard_qp;
323 		qp->q.flags |= INET_FRAG_LAST_IN;
324 		qp->q.len = end;
325 	} else {
326 		if (end&7) {
327 			end &= ~7;
328 			if (skb->ip_summed != CHECKSUM_UNNECESSARY)
329 				skb->ip_summed = CHECKSUM_NONE;
330 		}
331 		if (end > qp->q.len) {
332 			/* Some bits beyond end -> corruption. */
333 			if (qp->q.flags & INET_FRAG_LAST_IN)
334 				goto discard_qp;
335 			qp->q.len = end;
336 		}
337 	}
338 	if (end == offset)
339 		goto discard_qp;
340 
341 	err = -ENOMEM;
342 	if (!pskb_pull(skb, skb_network_offset(skb) + ihl))
343 		goto discard_qp;
344 
345 	err = pskb_trim_rcsum(skb, end - offset);
346 	if (err)
347 		goto discard_qp;
348 
349 	/* Note : skb->rbnode and skb->dev share the same location. */
350 	dev = skb->dev;
351 	/* Makes sure compiler wont do silly aliasing games */
352 	barrier();
353 
354 	prev_tail = qp->q.fragments_tail;
355 	err = inet_frag_queue_insert(&qp->q, skb, offset, end);
356 	if (err)
357 		goto insert_error;
358 
359 	if (dev)
360 		qp->iif = dev->ifindex;
361 
362 	qp->q.stamp = skb->tstamp;
363 	qp->q.mono_delivery_time = skb->mono_delivery_time;
364 	qp->q.meat += skb->len;
365 	qp->ecn |= ecn;
366 	add_frag_mem_limit(qp->q.fqdir, skb->truesize);
367 	if (offset == 0)
368 		qp->q.flags |= INET_FRAG_FIRST_IN;
369 
370 	fragsize = skb->len + ihl;
371 
372 	if (fragsize > qp->q.max_size)
373 		qp->q.max_size = fragsize;
374 
375 	if (ip_hdr(skb)->frag_off & htons(IP_DF) &&
376 	    fragsize > qp->max_df_size)
377 		qp->max_df_size = fragsize;
378 
379 	if (qp->q.flags == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
380 	    qp->q.meat == qp->q.len) {
381 		unsigned long orefdst = skb->_skb_refdst;
382 
383 		skb->_skb_refdst = 0UL;
384 		err = ip_frag_reasm(qp, skb, prev_tail, dev);
385 		skb->_skb_refdst = orefdst;
386 		if (err)
387 			inet_frag_kill(&qp->q);
388 		return err;
389 	}
390 
391 	skb_dst_drop(skb);
392 	skb_orphan(skb);
393 	return -EINPROGRESS;
394 
395 insert_error:
396 	if (err == IPFRAG_DUP) {
397 		SKB_DR_SET(reason, DUP_FRAG);
398 		err = -EINVAL;
399 		goto err;
400 	}
401 	err = -EINVAL;
402 	__IP_INC_STATS(net, IPSTATS_MIB_REASM_OVERLAPS);
403 discard_qp:
404 	inet_frag_kill(&qp->q);
405 	__IP_INC_STATS(net, IPSTATS_MIB_REASMFAILS);
406 err:
407 	kfree_skb_reason(skb, reason);
408 	return err;
409 }
410 
ip_frag_coalesce_ok(const struct ipq * qp)411 static bool ip_frag_coalesce_ok(const struct ipq *qp)
412 {
413 	return qp->q.key.v4.user == IP_DEFRAG_LOCAL_DELIVER;
414 }
415 
416 /* 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)417 static int ip_frag_reasm(struct ipq *qp, struct sk_buff *skb,
418 			 struct sk_buff *prev_tail, struct net_device *dev)
419 {
420 	struct net *net = qp->q.fqdir->net;
421 	struct iphdr *iph;
422 	void *reasm_data;
423 	int len, err;
424 	u8 ecn;
425 
426 	ipq_kill(qp);
427 
428 	ecn = ip_frag_ecn_table[qp->ecn];
429 	if (unlikely(ecn == 0xff)) {
430 		err = -EINVAL;
431 		goto out_fail;
432 	}
433 
434 	/* Make the one we just received the head. */
435 	reasm_data = inet_frag_reasm_prepare(&qp->q, skb, prev_tail);
436 	if (!reasm_data)
437 		goto out_nomem;
438 
439 	len = ip_hdrlen(skb) + qp->q.len;
440 	err = -E2BIG;
441 	if (len > 65535)
442 		goto out_oversize;
443 
444 	inet_frag_reasm_finish(&qp->q, skb, reasm_data,
445 			       ip_frag_coalesce_ok(qp));
446 
447 	skb->dev = dev;
448 	IPCB(skb)->frag_max_size = max(qp->max_df_size, qp->q.max_size);
449 
450 	iph = ip_hdr(skb);
451 	iph->tot_len = htons(len);
452 	iph->tos |= ecn;
453 
454 	/* When we set IP_DF on a refragmented skb we must also force a
455 	 * call to ip_fragment to avoid forwarding a DF-skb of size s while
456 	 * original sender only sent fragments of size f (where f < s).
457 	 *
458 	 * We only set DF/IPSKB_FRAG_PMTU if such DF fragment was the largest
459 	 * frag seen to avoid sending tiny DF-fragments in case skb was built
460 	 * from one very small df-fragment and one large non-df frag.
461 	 */
462 	if (qp->max_df_size == qp->q.max_size) {
463 		IPCB(skb)->flags |= IPSKB_FRAG_PMTU;
464 		iph->frag_off = htons(IP_DF);
465 	} else {
466 		iph->frag_off = 0;
467 	}
468 
469 	ip_send_check(iph);
470 
471 	__IP_INC_STATS(net, IPSTATS_MIB_REASMOKS);
472 	qp->q.rb_fragments = RB_ROOT;
473 	qp->q.fragments_tail = NULL;
474 	qp->q.last_run_head = NULL;
475 	return 0;
476 
477 out_nomem:
478 	net_dbg_ratelimited("queue_glue: no memory for gluing queue %p\n", qp);
479 	err = -ENOMEM;
480 	goto out_fail;
481 out_oversize:
482 	net_info_ratelimited("Oversized IP packet from %pI4\n", &qp->q.key.v4.saddr);
483 out_fail:
484 	__IP_INC_STATS(net, IPSTATS_MIB_REASMFAILS);
485 	return err;
486 }
487 
488 /* Process an incoming IP datagram fragment. */
ip_defrag(struct net * net,struct sk_buff * skb,u32 user)489 int ip_defrag(struct net *net, struct sk_buff *skb, u32 user)
490 {
491 	struct net_device *dev = skb->dev ? : skb_dst(skb)->dev;
492 	int vif = l3mdev_master_ifindex_rcu(dev);
493 	struct ipq *qp;
494 
495 	__IP_INC_STATS(net, IPSTATS_MIB_REASMREQDS);
496 
497 	/* Lookup (or create) queue header */
498 	qp = ip_find(net, ip_hdr(skb), user, vif);
499 	if (qp) {
500 		int ret;
501 
502 		spin_lock(&qp->q.lock);
503 
504 		ret = ip_frag_queue(qp, skb);
505 
506 		spin_unlock(&qp->q.lock);
507 		ipq_put(qp);
508 		return ret;
509 	}
510 
511 	__IP_INC_STATS(net, IPSTATS_MIB_REASMFAILS);
512 	kfree_skb(skb);
513 	return -ENOMEM;
514 }
515 EXPORT_SYMBOL(ip_defrag);
516 
ip_check_defrag(struct net * net,struct sk_buff * skb,u32 user)517 struct sk_buff *ip_check_defrag(struct net *net, struct sk_buff *skb, u32 user)
518 {
519 	struct iphdr iph;
520 	int netoff;
521 	u32 len;
522 
523 	if (skb->protocol != htons(ETH_P_IP))
524 		return skb;
525 
526 	netoff = skb_network_offset(skb);
527 
528 	if (skb_copy_bits(skb, netoff, &iph, sizeof(iph)) < 0)
529 		return skb;
530 
531 	if (iph.ihl < 5 || iph.version != 4)
532 		return skb;
533 
534 	len = ntohs(iph.tot_len);
535 	if (skb->len < netoff + len || len < (iph.ihl * 4))
536 		return skb;
537 
538 	if (ip_is_fragment(&iph)) {
539 		skb = skb_share_check(skb, GFP_ATOMIC);
540 		if (skb) {
541 			if (!pskb_may_pull(skb, netoff + iph.ihl * 4)) {
542 				kfree_skb(skb);
543 				return NULL;
544 			}
545 			if (pskb_trim_rcsum(skb, netoff + len)) {
546 				kfree_skb(skb);
547 				return NULL;
548 			}
549 			memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
550 			if (ip_defrag(net, skb, user))
551 				return NULL;
552 			skb_clear_hash(skb);
553 		}
554 	}
555 	return skb;
556 }
557 EXPORT_SYMBOL(ip_check_defrag);
558 
559 #ifdef CONFIG_SYSCTL
560 static int dist_min;
561 
562 static struct ctl_table ip4_frags_ns_ctl_table[] = {
563 	{
564 		.procname	= "ipfrag_high_thresh",
565 		.maxlen		= sizeof(unsigned long),
566 		.mode		= 0644,
567 		.proc_handler	= proc_doulongvec_minmax,
568 	},
569 	{
570 		.procname	= "ipfrag_low_thresh",
571 		.maxlen		= sizeof(unsigned long),
572 		.mode		= 0644,
573 		.proc_handler	= proc_doulongvec_minmax,
574 	},
575 	{
576 		.procname	= "ipfrag_time",
577 		.maxlen		= sizeof(int),
578 		.mode		= 0644,
579 		.proc_handler	= proc_dointvec_jiffies,
580 	},
581 	{
582 		.procname	= "ipfrag_max_dist",
583 		.maxlen		= sizeof(int),
584 		.mode		= 0644,
585 		.proc_handler	= proc_dointvec_minmax,
586 		.extra1		= &dist_min,
587 	},
588 	{ }
589 };
590 
591 /* secret interval has been deprecated */
592 static int ip4_frags_secret_interval_unused;
593 static struct ctl_table ip4_frags_ctl_table[] = {
594 	{
595 		.procname	= "ipfrag_secret_interval",
596 		.data		= &ip4_frags_secret_interval_unused,
597 		.maxlen		= sizeof(int),
598 		.mode		= 0644,
599 		.proc_handler	= proc_dointvec_jiffies,
600 	},
601 	{ }
602 };
603 
ip4_frags_ns_ctl_register(struct net * net)604 static int __net_init ip4_frags_ns_ctl_register(struct net *net)
605 {
606 	struct ctl_table *table;
607 	struct ctl_table_header *hdr;
608 
609 	table = ip4_frags_ns_ctl_table;
610 	if (!net_eq(net, &init_net)) {
611 		table = kmemdup(table, sizeof(ip4_frags_ns_ctl_table), GFP_KERNEL);
612 		if (!table)
613 			goto err_alloc;
614 
615 	}
616 	table[0].data	= &net->ipv4.fqdir->high_thresh;
617 	table[0].extra1	= &net->ipv4.fqdir->low_thresh;
618 	table[1].data	= &net->ipv4.fqdir->low_thresh;
619 	table[1].extra2	= &net->ipv4.fqdir->high_thresh;
620 	table[2].data	= &net->ipv4.fqdir->timeout;
621 	table[3].data	= &net->ipv4.fqdir->max_dist;
622 
623 	hdr = register_net_sysctl_sz(net, "net/ipv4", table,
624 				     ARRAY_SIZE(ip4_frags_ns_ctl_table));
625 	if (!hdr)
626 		goto err_reg;
627 
628 	net->ipv4.frags_hdr = hdr;
629 	return 0;
630 
631 err_reg:
632 	if (!net_eq(net, &init_net))
633 		kfree(table);
634 err_alloc:
635 	return -ENOMEM;
636 }
637 
ip4_frags_ns_ctl_unregister(struct net * net)638 static void __net_exit ip4_frags_ns_ctl_unregister(struct net *net)
639 {
640 	struct ctl_table *table;
641 
642 	table = net->ipv4.frags_hdr->ctl_table_arg;
643 	unregister_net_sysctl_table(net->ipv4.frags_hdr);
644 	kfree(table);
645 }
646 
ip4_frags_ctl_register(void)647 static void __init ip4_frags_ctl_register(void)
648 {
649 	register_net_sysctl(&init_net, "net/ipv4", ip4_frags_ctl_table);
650 }
651 #else
ip4_frags_ns_ctl_register(struct net * net)652 static int ip4_frags_ns_ctl_register(struct net *net)
653 {
654 	return 0;
655 }
656 
ip4_frags_ns_ctl_unregister(struct net * net)657 static void ip4_frags_ns_ctl_unregister(struct net *net)
658 {
659 }
660 
ip4_frags_ctl_register(void)661 static void __init ip4_frags_ctl_register(void)
662 {
663 }
664 #endif
665 
ipv4_frags_init_net(struct net * net)666 static int __net_init ipv4_frags_init_net(struct net *net)
667 {
668 	int res;
669 
670 	res = fqdir_init(&net->ipv4.fqdir, &ip4_frags, net);
671 	if (res < 0)
672 		return res;
673 	/* Fragment cache limits.
674 	 *
675 	 * The fragment memory accounting code, (tries to) account for
676 	 * the real memory usage, by measuring both the size of frag
677 	 * queue struct (inet_frag_queue (ipv4:ipq/ipv6:frag_queue))
678 	 * and the SKB's truesize.
679 	 *
680 	 * A 64K fragment consumes 129736 bytes (44*2944)+200
681 	 * (1500 truesize == 2944, sizeof(struct ipq) == 200)
682 	 *
683 	 * We will commit 4MB at one time. Should we cross that limit
684 	 * we will prune down to 3MB, making room for approx 8 big 64K
685 	 * fragments 8x128k.
686 	 */
687 	net->ipv4.fqdir->high_thresh = 4 * 1024 * 1024;
688 	net->ipv4.fqdir->low_thresh  = 3 * 1024 * 1024;
689 	/*
690 	 * Important NOTE! Fragment queue must be destroyed before MSL expires.
691 	 * RFC791 is wrong proposing to prolongate timer each fragment arrival
692 	 * by TTL.
693 	 */
694 	net->ipv4.fqdir->timeout = IP_FRAG_TIME;
695 
696 	net->ipv4.fqdir->max_dist = 64;
697 
698 	res = ip4_frags_ns_ctl_register(net);
699 	if (res < 0)
700 		fqdir_exit(net->ipv4.fqdir);
701 	return res;
702 }
703 
ipv4_frags_pre_exit_net(struct net * net)704 static void __net_exit ipv4_frags_pre_exit_net(struct net *net)
705 {
706 	fqdir_pre_exit(net->ipv4.fqdir);
707 }
708 
ipv4_frags_exit_net(struct net * net)709 static void __net_exit ipv4_frags_exit_net(struct net *net)
710 {
711 	ip4_frags_ns_ctl_unregister(net);
712 	fqdir_exit(net->ipv4.fqdir);
713 }
714 
715 static struct pernet_operations ip4_frags_ops = {
716 	.init		= ipv4_frags_init_net,
717 	.pre_exit	= ipv4_frags_pre_exit_net,
718 	.exit		= ipv4_frags_exit_net,
719 };
720 
721 
ip4_key_hashfn(const void * data,u32 len,u32 seed)722 static u32 ip4_key_hashfn(const void *data, u32 len, u32 seed)
723 {
724 	return jhash2(data,
725 		      sizeof(struct frag_v4_compare_key) / sizeof(u32), seed);
726 }
727 
ip4_obj_hashfn(const void * data,u32 len,u32 seed)728 static u32 ip4_obj_hashfn(const void *data, u32 len, u32 seed)
729 {
730 	const struct inet_frag_queue *fq = data;
731 
732 	return jhash2((const u32 *)&fq->key.v4,
733 		      sizeof(struct frag_v4_compare_key) / sizeof(u32), seed);
734 }
735 
ip4_obj_cmpfn(struct rhashtable_compare_arg * arg,const void * ptr)736 static int ip4_obj_cmpfn(struct rhashtable_compare_arg *arg, const void *ptr)
737 {
738 	const struct frag_v4_compare_key *key = arg->key;
739 	const struct inet_frag_queue *fq = ptr;
740 
741 	return !!memcmp(&fq->key, key, sizeof(*key));
742 }
743 
744 static const struct rhashtable_params ip4_rhash_params = {
745 	.head_offset		= offsetof(struct inet_frag_queue, node),
746 	.key_offset		= offsetof(struct inet_frag_queue, key),
747 	.key_len		= sizeof(struct frag_v4_compare_key),
748 	.hashfn			= ip4_key_hashfn,
749 	.obj_hashfn		= ip4_obj_hashfn,
750 	.obj_cmpfn		= ip4_obj_cmpfn,
751 	.automatic_shrinking	= true,
752 };
753 
ipfrag_init(void)754 void __init ipfrag_init(void)
755 {
756 	ip4_frags.constructor = ip4_frag_init;
757 	ip4_frags.destructor = ip4_frag_free;
758 	ip4_frags.qsize = sizeof(struct ipq);
759 	ip4_frags.frag_expire = ip_expire;
760 	ip4_frags.frags_cache_name = ip_frag_cache_name;
761 	ip4_frags.rhash_params = ip4_rhash_params;
762 	if (inet_frags_init(&ip4_frags))
763 		panic("IP: failed to allocate ip4_frags cache\n");
764 	ip4_frags_ctl_register();
765 	register_pernet_subsys(&ip4_frags_ops);
766 }
767