xref: /linux/net/ipv6/reassembly.c (revision 2638eb8b50cfc16240e0bb080b9afbf541a9b39d) 
1 /*
2  *	IPv6 fragment reassembly
3  *	Linux INET6 implementation
4  *
5  *	Authors:
6  *	Pedro Roque		<roque@di.fc.ul.pt>
7  *
8  *	Based on: net/ipv4/ip_fragment.c
9  *
10  *	This program is free software; you can redistribute it and/or
11  *      modify it under the terms of the GNU General Public License
12  *      as published by the Free Software Foundation; either version
13  *      2 of the License, or (at your option) any later version.
14  */
15 
16 /*
17  *	Fixes:
18  *	Andi Kleen	Make it work with multiple hosts.
19  *			More RFC compliance.
20  *
21  *      Horst von Brand Add missing #include <linux/string.h>
22  *	Alexey Kuznetsov	SMP races, threading, cleanup.
23  *	Patrick McHardy		LRU queue of frag heads for evictor.
24  *	Mitsuru KANDA @USAGI	Register inet6_protocol{}.
25  *	David Stevens and
26  *	YOSHIFUJI,H. @USAGI	Always remove fragment header to
27  *				calculate ICV correctly.
28  */
29 
30 #define pr_fmt(fmt) "IPv6: " fmt
31 
32 #include <linux/errno.h>
33 #include <linux/types.h>
34 #include <linux/string.h>
35 #include <linux/socket.h>
36 #include <linux/sockios.h>
37 #include <linux/jiffies.h>
38 #include <linux/net.h>
39 #include <linux/list.h>
40 #include <linux/netdevice.h>
41 #include <linux/in6.h>
42 #include <linux/ipv6.h>
43 #include <linux/icmpv6.h>
44 #include <linux/random.h>
45 #include <linux/jhash.h>
46 #include <linux/skbuff.h>
47 #include <linux/slab.h>
48 #include <linux/export.h>
49 
50 #include <net/sock.h>
51 #include <net/snmp.h>
52 
53 #include <net/ipv6.h>
54 #include <net/ip6_route.h>
55 #include <net/protocol.h>
56 #include <net/transp_v6.h>
57 #include <net/rawv6.h>
58 #include <net/ndisc.h>
59 #include <net/addrconf.h>
60 #include <net/ipv6_frag.h>
61 #include <net/inet_ecn.h>
62 
63 static const char ip6_frag_cache_name[] = "ip6-frags";
64 
65 static u8 ip6_frag_ecn(const struct ipv6hdr *ipv6h)
66 {
67 	return 1 << (ipv6_get_dsfield(ipv6h) & INET_ECN_MASK);
68 }
69 
70 static struct inet_frags ip6_frags;
71 
72 static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *skb,
73 			  struct sk_buff *prev_tail, struct net_device *dev);
74 
75 static void ip6_frag_expire(struct timer_list *t)
76 {
77 	struct inet_frag_queue *frag = from_timer(frag, t, timer);
78 	struct frag_queue *fq;
79 
80 	fq = container_of(frag, struct frag_queue, q);
81 
82 	ip6frag_expire_frag_queue(fq->q.fqdir->net, fq);
83 }
84 
85 static struct frag_queue *
86 fq_find(struct net *net, __be32 id, const struct ipv6hdr *hdr, int iif)
87 {
88 	struct frag_v6_compare_key key = {
89 		.id = id,
90 		.saddr = hdr->saddr,
91 		.daddr = hdr->daddr,
92 		.user = IP6_DEFRAG_LOCAL_DELIVER,
93 		.iif = iif,
94 	};
95 	struct inet_frag_queue *q;
96 
97 	if (!(ipv6_addr_type(&hdr->daddr) & (IPV6_ADDR_MULTICAST |
98 					    IPV6_ADDR_LINKLOCAL)))
99 		key.iif = 0;
100 
101 	q = inet_frag_find(net->ipv6.fqdir, &key);
102 	if (!q)
103 		return NULL;
104 
105 	return container_of(q, struct frag_queue, q);
106 }
107 
108 static int ip6_frag_queue(struct frag_queue *fq, struct sk_buff *skb,
109 			  struct frag_hdr *fhdr, int nhoff,
110 			  u32 *prob_offset)
111 {
112 	struct net *net = dev_net(skb_dst(skb)->dev);
113 	int offset, end, fragsize;
114 	struct sk_buff *prev_tail;
115 	struct net_device *dev;
116 	int err = -ENOENT;
117 	u8 ecn;
118 
119 	if (fq->q.flags & INET_FRAG_COMPLETE)
120 		goto err;
121 
122 	err = -EINVAL;
123 	offset = ntohs(fhdr->frag_off) & ~0x7;
124 	end = offset + (ntohs(ipv6_hdr(skb)->payload_len) -
125 			((u8 *)(fhdr + 1) - (u8 *)(ipv6_hdr(skb) + 1)));
126 
127 	if ((unsigned int)end > IPV6_MAXPLEN) {
128 		*prob_offset = (u8 *)&fhdr->frag_off - skb_network_header(skb);
129 		/* note that if prob_offset is set, the skb is freed elsewhere,
130 		 * we do not free it here.
131 		 */
132 		return -1;
133 	}
134 
135 	ecn = ip6_frag_ecn(ipv6_hdr(skb));
136 
137 	if (skb->ip_summed == CHECKSUM_COMPLETE) {
138 		const unsigned char *nh = skb_network_header(skb);
139 		skb->csum = csum_sub(skb->csum,
140 				     csum_partial(nh, (u8 *)(fhdr + 1) - nh,
141 						  0));
142 	}
143 
144 	/* Is this the final fragment? */
145 	if (!(fhdr->frag_off & htons(IP6_MF))) {
146 		/* If we already have some bits beyond end
147 		 * or have different end, the segment is corrupted.
148 		 */
149 		if (end < fq->q.len ||
150 		    ((fq->q.flags & INET_FRAG_LAST_IN) && end != fq->q.len))
151 			goto discard_fq;
152 		fq->q.flags |= INET_FRAG_LAST_IN;
153 		fq->q.len = end;
154 	} else {
155 		/* Check if the fragment is rounded to 8 bytes.
156 		 * Required by the RFC.
157 		 */
158 		if (end & 0x7) {
159 			/* RFC2460 says always send parameter problem in
160 			 * this case. -DaveM
161 			 */
162 			*prob_offset = offsetof(struct ipv6hdr, payload_len);
163 			return -1;
164 		}
165 		if (end > fq->q.len) {
166 			/* Some bits beyond end -> corruption. */
167 			if (fq->q.flags & INET_FRAG_LAST_IN)
168 				goto discard_fq;
169 			fq->q.len = end;
170 		}
171 	}
172 
173 	if (end == offset)
174 		goto discard_fq;
175 
176 	err = -ENOMEM;
177 	/* Point into the IP datagram 'data' part. */
178 	if (!pskb_pull(skb, (u8 *) (fhdr + 1) - skb->data))
179 		goto discard_fq;
180 
181 	err = pskb_trim_rcsum(skb, end - offset);
182 	if (err)
183 		goto discard_fq;
184 
185 	/* Note : skb->rbnode and skb->dev share the same location. */
186 	dev = skb->dev;
187 	/* Makes sure compiler wont do silly aliasing games */
188 	barrier();
189 
190 	prev_tail = fq->q.fragments_tail;
191 	err = inet_frag_queue_insert(&fq->q, skb, offset, end);
192 	if (err)
193 		goto insert_error;
194 
195 	if (dev)
196 		fq->iif = dev->ifindex;
197 
198 	fq->q.stamp = skb->tstamp;
199 	fq->q.meat += skb->len;
200 	fq->ecn |= ecn;
201 	add_frag_mem_limit(fq->q.fqdir, skb->truesize);
202 
203 	fragsize = -skb_network_offset(skb) + skb->len;
204 	if (fragsize > fq->q.max_size)
205 		fq->q.max_size = fragsize;
206 
207 	/* The first fragment.
208 	 * nhoffset is obtained from the first fragment, of course.
209 	 */
210 	if (offset == 0) {
211 		fq->nhoffset = nhoff;
212 		fq->q.flags |= INET_FRAG_FIRST_IN;
213 	}
214 
215 	if (fq->q.flags == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
216 	    fq->q.meat == fq->q.len) {
217 		unsigned long orefdst = skb->_skb_refdst;
218 
219 		skb->_skb_refdst = 0UL;
220 		err = ip6_frag_reasm(fq, skb, prev_tail, dev);
221 		skb->_skb_refdst = orefdst;
222 		return err;
223 	}
224 
225 	skb_dst_drop(skb);
226 	return -EINPROGRESS;
227 
228 insert_error:
229 	if (err == IPFRAG_DUP) {
230 		kfree_skb(skb);
231 		return -EINVAL;
232 	}
233 	err = -EINVAL;
234 	__IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
235 			IPSTATS_MIB_REASM_OVERLAPS);
236 discard_fq:
237 	inet_frag_kill(&fq->q);
238 	__IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
239 			IPSTATS_MIB_REASMFAILS);
240 err:
241 	kfree_skb(skb);
242 	return err;
243 }
244 
245 /*
246  *	Check if this packet is complete.
247  *
248  *	It is called with locked fq, and caller must check that
249  *	queue is eligible for reassembly i.e. it is not COMPLETE,
250  *	the last and the first frames arrived and all the bits are here.
251  */
252 static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *skb,
253 			  struct sk_buff *prev_tail, struct net_device *dev)
254 {
255 	struct net *net = fq->q.fqdir->net;
256 	unsigned int nhoff;
257 	void *reasm_data;
258 	int payload_len;
259 	u8 ecn;
260 
261 	inet_frag_kill(&fq->q);
262 
263 	ecn = ip_frag_ecn_table[fq->ecn];
264 	if (unlikely(ecn == 0xff))
265 		goto out_fail;
266 
267 	reasm_data = inet_frag_reasm_prepare(&fq->q, skb, prev_tail);
268 	if (!reasm_data)
269 		goto out_oom;
270 
271 	payload_len = ((skb->data - skb_network_header(skb)) -
272 		       sizeof(struct ipv6hdr) + fq->q.len -
273 		       sizeof(struct frag_hdr));
274 	if (payload_len > IPV6_MAXPLEN)
275 		goto out_oversize;
276 
277 	/* We have to remove fragment header from datagram and to relocate
278 	 * header in order to calculate ICV correctly. */
279 	nhoff = fq->nhoffset;
280 	skb_network_header(skb)[nhoff] = skb_transport_header(skb)[0];
281 	memmove(skb->head + sizeof(struct frag_hdr), skb->head,
282 		(skb->data - skb->head) - sizeof(struct frag_hdr));
283 	if (skb_mac_header_was_set(skb))
284 		skb->mac_header += sizeof(struct frag_hdr);
285 	skb->network_header += sizeof(struct frag_hdr);
286 
287 	skb_reset_transport_header(skb);
288 
289 	inet_frag_reasm_finish(&fq->q, skb, reasm_data);
290 
291 	skb->dev = dev;
292 	ipv6_hdr(skb)->payload_len = htons(payload_len);
293 	ipv6_change_dsfield(ipv6_hdr(skb), 0xff, ecn);
294 	IP6CB(skb)->nhoff = nhoff;
295 	IP6CB(skb)->flags |= IP6SKB_FRAGMENTED;
296 	IP6CB(skb)->frag_max_size = fq->q.max_size;
297 
298 	/* Yes, and fold redundant checksum back. 8) */
299 	skb_postpush_rcsum(skb, skb_network_header(skb),
300 			   skb_network_header_len(skb));
301 
302 	rcu_read_lock();
303 	__IP6_INC_STATS(net, __in6_dev_get(dev), IPSTATS_MIB_REASMOKS);
304 	rcu_read_unlock();
305 	fq->q.rb_fragments = RB_ROOT;
306 	fq->q.fragments_tail = NULL;
307 	fq->q.last_run_head = NULL;
308 	return 1;
309 
310 out_oversize:
311 	net_dbg_ratelimited("ip6_frag_reasm: payload len = %d\n", payload_len);
312 	goto out_fail;
313 out_oom:
314 	net_dbg_ratelimited("ip6_frag_reasm: no memory for reassembly\n");
315 out_fail:
316 	rcu_read_lock();
317 	__IP6_INC_STATS(net, __in6_dev_get(dev), IPSTATS_MIB_REASMFAILS);
318 	rcu_read_unlock();
319 	inet_frag_kill(&fq->q);
320 	return -1;
321 }
322 
323 static int ipv6_frag_rcv(struct sk_buff *skb)
324 {
325 	struct frag_hdr *fhdr;
326 	struct frag_queue *fq;
327 	const struct ipv6hdr *hdr = ipv6_hdr(skb);
328 	struct net *net = dev_net(skb_dst(skb)->dev);
329 	int iif;
330 
331 	if (IP6CB(skb)->flags & IP6SKB_FRAGMENTED)
332 		goto fail_hdr;
333 
334 	__IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMREQDS);
335 
336 	/* Jumbo payload inhibits frag. header */
337 	if (hdr->payload_len == 0)
338 		goto fail_hdr;
339 
340 	if (!pskb_may_pull(skb, (skb_transport_offset(skb) +
341 				 sizeof(struct frag_hdr))))
342 		goto fail_hdr;
343 
344 	hdr = ipv6_hdr(skb);
345 	fhdr = (struct frag_hdr *)skb_transport_header(skb);
346 
347 	if (!(fhdr->frag_off & htons(0xFFF9))) {
348 		/* It is not a fragmented frame */
349 		skb->transport_header += sizeof(struct frag_hdr);
350 		__IP6_INC_STATS(net,
351 				ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMOKS);
352 
353 		IP6CB(skb)->nhoff = (u8 *)fhdr - skb_network_header(skb);
354 		IP6CB(skb)->flags |= IP6SKB_FRAGMENTED;
355 		return 1;
356 	}
357 
358 	iif = skb->dev ? skb->dev->ifindex : 0;
359 	fq = fq_find(net, fhdr->identification, hdr, iif);
360 	if (fq) {
361 		u32 prob_offset = 0;
362 		int ret;
363 
364 		spin_lock(&fq->q.lock);
365 
366 		fq->iif = iif;
367 		ret = ip6_frag_queue(fq, skb, fhdr, IP6CB(skb)->nhoff,
368 				     &prob_offset);
369 
370 		spin_unlock(&fq->q.lock);
371 		inet_frag_put(&fq->q);
372 		if (prob_offset) {
373 			__IP6_INC_STATS(net, __in6_dev_get_safely(skb->dev),
374 					IPSTATS_MIB_INHDRERRORS);
375 			/* icmpv6_param_prob() calls kfree_skb(skb) */
376 			icmpv6_param_prob(skb, ICMPV6_HDR_FIELD, prob_offset);
377 		}
378 		return ret;
379 	}
380 
381 	__IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMFAILS);
382 	kfree_skb(skb);
383 	return -1;
384 
385 fail_hdr:
386 	__IP6_INC_STATS(net, __in6_dev_get_safely(skb->dev),
387 			IPSTATS_MIB_INHDRERRORS);
388 	icmpv6_param_prob(skb, ICMPV6_HDR_FIELD, skb_network_header_len(skb));
389 	return -1;
390 }
391 
392 static const struct inet6_protocol frag_protocol = {
393 	.handler	=	ipv6_frag_rcv,
394 	.flags		=	INET6_PROTO_NOPOLICY,
395 };
396 
397 #ifdef CONFIG_SYSCTL
398 
399 static struct ctl_table ip6_frags_ns_ctl_table[] = {
400 	{
401 		.procname	= "ip6frag_high_thresh",
402 		.maxlen		= sizeof(unsigned long),
403 		.mode		= 0644,
404 		.proc_handler	= proc_doulongvec_minmax,
405 	},
406 	{
407 		.procname	= "ip6frag_low_thresh",
408 		.maxlen		= sizeof(unsigned long),
409 		.mode		= 0644,
410 		.proc_handler	= proc_doulongvec_minmax,
411 	},
412 	{
413 		.procname	= "ip6frag_time",
414 		.maxlen		= sizeof(int),
415 		.mode		= 0644,
416 		.proc_handler	= proc_dointvec_jiffies,
417 	},
418 	{ }
419 };
420 
421 /* secret interval has been deprecated */
422 static int ip6_frags_secret_interval_unused;
423 static struct ctl_table ip6_frags_ctl_table[] = {
424 	{
425 		.procname	= "ip6frag_secret_interval",
426 		.data		= &ip6_frags_secret_interval_unused,
427 		.maxlen		= sizeof(int),
428 		.mode		= 0644,
429 		.proc_handler	= proc_dointvec_jiffies,
430 	},
431 	{ }
432 };
433 
434 static int __net_init ip6_frags_ns_sysctl_register(struct net *net)
435 {
436 	struct ctl_table *table;
437 	struct ctl_table_header *hdr;
438 
439 	table = ip6_frags_ns_ctl_table;
440 	if (!net_eq(net, &init_net)) {
441 		table = kmemdup(table, sizeof(ip6_frags_ns_ctl_table), GFP_KERNEL);
442 		if (!table)
443 			goto err_alloc;
444 
445 	}
446 	table[0].data	= &net->ipv6.fqdir->high_thresh;
447 	table[0].extra1	= &net->ipv6.fqdir->low_thresh;
448 	table[1].data	= &net->ipv6.fqdir->low_thresh;
449 	table[1].extra2	= &net->ipv6.fqdir->high_thresh;
450 	table[2].data	= &net->ipv6.fqdir->timeout;
451 
452 	hdr = register_net_sysctl(net, "net/ipv6", table);
453 	if (!hdr)
454 		goto err_reg;
455 
456 	net->ipv6.sysctl.frags_hdr = hdr;
457 	return 0;
458 
459 err_reg:
460 	if (!net_eq(net, &init_net))
461 		kfree(table);
462 err_alloc:
463 	return -ENOMEM;
464 }
465 
466 static void __net_exit ip6_frags_ns_sysctl_unregister(struct net *net)
467 {
468 	struct ctl_table *table;
469 
470 	table = net->ipv6.sysctl.frags_hdr->ctl_table_arg;
471 	unregister_net_sysctl_table(net->ipv6.sysctl.frags_hdr);
472 	if (!net_eq(net, &init_net))
473 		kfree(table);
474 }
475 
476 static struct ctl_table_header *ip6_ctl_header;
477 
478 static int ip6_frags_sysctl_register(void)
479 {
480 	ip6_ctl_header = register_net_sysctl(&init_net, "net/ipv6",
481 			ip6_frags_ctl_table);
482 	return ip6_ctl_header == NULL ? -ENOMEM : 0;
483 }
484 
485 static void ip6_frags_sysctl_unregister(void)
486 {
487 	unregister_net_sysctl_table(ip6_ctl_header);
488 }
489 #else
490 static int ip6_frags_ns_sysctl_register(struct net *net)
491 {
492 	return 0;
493 }
494 
495 static void ip6_frags_ns_sysctl_unregister(struct net *net)
496 {
497 }
498 
499 static int ip6_frags_sysctl_register(void)
500 {
501 	return 0;
502 }
503 
504 static void ip6_frags_sysctl_unregister(void)
505 {
506 }
507 #endif
508 
509 static int __net_init ipv6_frags_init_net(struct net *net)
510 {
511 	int res;
512 
513 	res = fqdir_init(&net->ipv6.fqdir, &ip6_frags, net);
514 	if (res < 0)
515 		return res;
516 
517 	net->ipv6.fqdir->high_thresh = IPV6_FRAG_HIGH_THRESH;
518 	net->ipv6.fqdir->low_thresh = IPV6_FRAG_LOW_THRESH;
519 	net->ipv6.fqdir->timeout = IPV6_FRAG_TIMEOUT;
520 
521 	res = ip6_frags_ns_sysctl_register(net);
522 	if (res < 0)
523 		fqdir_exit(net->ipv6.fqdir);
524 	return res;
525 }
526 
527 static void __net_exit ipv6_frags_exit_net(struct net *net)
528 {
529 	ip6_frags_ns_sysctl_unregister(net);
530 	fqdir_exit(net->ipv6.fqdir);
531 }
532 
533 static struct pernet_operations ip6_frags_ops = {
534 	.init = ipv6_frags_init_net,
535 	.exit = ipv6_frags_exit_net,
536 };
537 
538 static const struct rhashtable_params ip6_rhash_params = {
539 	.head_offset		= offsetof(struct inet_frag_queue, node),
540 	.hashfn			= ip6frag_key_hashfn,
541 	.obj_hashfn		= ip6frag_obj_hashfn,
542 	.obj_cmpfn		= ip6frag_obj_cmpfn,
543 	.automatic_shrinking	= true,
544 };
545 
546 int __init ipv6_frag_init(void)
547 {
548 	int ret;
549 
550 	ip6_frags.constructor = ip6frag_init;
551 	ip6_frags.destructor = NULL;
552 	ip6_frags.qsize = sizeof(struct frag_queue);
553 	ip6_frags.frag_expire = ip6_frag_expire;
554 	ip6_frags.frags_cache_name = ip6_frag_cache_name;
555 	ip6_frags.rhash_params = ip6_rhash_params;
556 	ret = inet_frags_init(&ip6_frags);
557 	if (ret)
558 		goto out;
559 
560 	ret = inet6_add_protocol(&frag_protocol, IPPROTO_FRAGMENT);
561 	if (ret)
562 		goto err_protocol;
563 
564 	ret = ip6_frags_sysctl_register();
565 	if (ret)
566 		goto err_sysctl;
567 
568 	ret = register_pernet_subsys(&ip6_frags_ops);
569 	if (ret)
570 		goto err_pernet;
571 
572 out:
573 	return ret;
574 
575 err_pernet:
576 	ip6_frags_sysctl_unregister();
577 err_sysctl:
578 	inet6_del_protocol(&frag_protocol, IPPROTO_FRAGMENT);
579 err_protocol:
580 	inet_frags_fini(&ip6_frags);
581 	goto out;
582 }
583 
584 void ipv6_frag_exit(void)
585 {
586 	ip6_frags_sysctl_unregister();
587 	unregister_pernet_subsys(&ip6_frags_ops);
588 	inet6_del_protocol(&frag_protocol, IPPROTO_FRAGMENT);
589 	inet_frags_fini(&ip6_frags);
590 }
591