xref: /linux/net/ipv6/reassembly.c (revision 18f90d372cf35b387663f1567de701e5393f6eb5) 
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 	struct net *net;
80 
81 	fq = container_of(frag, struct frag_queue, q);
82 	net = container_of(fq->q.net, struct net, ipv6.frags);
83 
84 	ip6frag_expire_frag_queue(net, fq);
85 }
86 
87 static struct frag_queue *
88 fq_find(struct net *net, __be32 id, const struct ipv6hdr *hdr, int iif)
89 {
90 	struct frag_v6_compare_key key = {
91 		.id = id,
92 		.saddr = hdr->saddr,
93 		.daddr = hdr->daddr,
94 		.user = IP6_DEFRAG_LOCAL_DELIVER,
95 		.iif = iif,
96 	};
97 	struct inet_frag_queue *q;
98 
99 	if (!(ipv6_addr_type(&hdr->daddr) & (IPV6_ADDR_MULTICAST |
100 					    IPV6_ADDR_LINKLOCAL)))
101 		key.iif = 0;
102 
103 	q = inet_frag_find(&net->ipv6.frags, &key);
104 	if (!q)
105 		return NULL;
106 
107 	return container_of(q, struct frag_queue, q);
108 }
109 
110 static int ip6_frag_queue(struct frag_queue *fq, struct sk_buff *skb,
111 			  struct frag_hdr *fhdr, int nhoff,
112 			  u32 *prob_offset)
113 {
114 	struct net *net = dev_net(skb_dst(skb)->dev);
115 	int offset, end, fragsize;
116 	struct sk_buff *prev_tail;
117 	struct net_device *dev;
118 	int err = -ENOENT;
119 	u8 ecn;
120 
121 	if (fq->q.flags & INET_FRAG_COMPLETE)
122 		goto err;
123 
124 	err = -EINVAL;
125 	offset = ntohs(fhdr->frag_off) & ~0x7;
126 	end = offset + (ntohs(ipv6_hdr(skb)->payload_len) -
127 			((u8 *)(fhdr + 1) - (u8 *)(ipv6_hdr(skb) + 1)));
128 
129 	if ((unsigned int)end > IPV6_MAXPLEN) {
130 		*prob_offset = (u8 *)&fhdr->frag_off - skb_network_header(skb);
131 		/* note that if prob_offset is set, the skb is freed elsewhere,
132 		 * we do not free it here.
133 		 */
134 		return -1;
135 	}
136 
137 	ecn = ip6_frag_ecn(ipv6_hdr(skb));
138 
139 	if (skb->ip_summed == CHECKSUM_COMPLETE) {
140 		const unsigned char *nh = skb_network_header(skb);
141 		skb->csum = csum_sub(skb->csum,
142 				     csum_partial(nh, (u8 *)(fhdr + 1) - nh,
143 						  0));
144 	}
145 
146 	/* Is this the final fragment? */
147 	if (!(fhdr->frag_off & htons(IP6_MF))) {
148 		/* If we already have some bits beyond end
149 		 * or have different end, the segment is corrupted.
150 		 */
151 		if (end < fq->q.len ||
152 		    ((fq->q.flags & INET_FRAG_LAST_IN) && end != fq->q.len))
153 			goto discard_fq;
154 		fq->q.flags |= INET_FRAG_LAST_IN;
155 		fq->q.len = end;
156 	} else {
157 		/* Check if the fragment is rounded to 8 bytes.
158 		 * Required by the RFC.
159 		 */
160 		if (end & 0x7) {
161 			/* RFC2460 says always send parameter problem in
162 			 * this case. -DaveM
163 			 */
164 			*prob_offset = offsetof(struct ipv6hdr, payload_len);
165 			return -1;
166 		}
167 		if (end > fq->q.len) {
168 			/* Some bits beyond end -> corruption. */
169 			if (fq->q.flags & INET_FRAG_LAST_IN)
170 				goto discard_fq;
171 			fq->q.len = end;
172 		}
173 	}
174 
175 	if (end == offset)
176 		goto discard_fq;
177 
178 	err = -ENOMEM;
179 	/* Point into the IP datagram 'data' part. */
180 	if (!pskb_pull(skb, (u8 *) (fhdr + 1) - skb->data))
181 		goto discard_fq;
182 
183 	err = pskb_trim_rcsum(skb, end - offset);
184 	if (err)
185 		goto discard_fq;
186 
187 	/* Note : skb->rbnode and skb->dev share the same location. */
188 	dev = skb->dev;
189 	/* Makes sure compiler wont do silly aliasing games */
190 	barrier();
191 
192 	prev_tail = fq->q.fragments_tail;
193 	err = inet_frag_queue_insert(&fq->q, skb, offset, end);
194 	if (err)
195 		goto insert_error;
196 
197 	if (dev)
198 		fq->iif = dev->ifindex;
199 
200 	fq->q.stamp = skb->tstamp;
201 	fq->q.meat += skb->len;
202 	fq->ecn |= ecn;
203 	add_frag_mem_limit(fq->q.net, skb->truesize);
204 
205 	fragsize = -skb_network_offset(skb) + skb->len;
206 	if (fragsize > fq->q.max_size)
207 		fq->q.max_size = fragsize;
208 
209 	/* The first fragment.
210 	 * nhoffset is obtained from the first fragment, of course.
211 	 */
212 	if (offset == 0) {
213 		fq->nhoffset = nhoff;
214 		fq->q.flags |= INET_FRAG_FIRST_IN;
215 	}
216 
217 	if (fq->q.flags == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
218 	    fq->q.meat == fq->q.len) {
219 		unsigned long orefdst = skb->_skb_refdst;
220 
221 		skb->_skb_refdst = 0UL;
222 		err = ip6_frag_reasm(fq, skb, prev_tail, dev);
223 		skb->_skb_refdst = orefdst;
224 		return err;
225 	}
226 
227 	skb_dst_drop(skb);
228 	return -EINPROGRESS;
229 
230 insert_error:
231 	if (err == IPFRAG_DUP) {
232 		kfree_skb(skb);
233 		return -EINVAL;
234 	}
235 	err = -EINVAL;
236 	__IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
237 			IPSTATS_MIB_REASM_OVERLAPS);
238 discard_fq:
239 	inet_frag_kill(&fq->q);
240 	__IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
241 			IPSTATS_MIB_REASMFAILS);
242 err:
243 	kfree_skb(skb);
244 	return err;
245 }
246 
247 /*
248  *	Check if this packet is complete.
249  *
250  *	It is called with locked fq, and caller must check that
251  *	queue is eligible for reassembly i.e. it is not COMPLETE,
252  *	the last and the first frames arrived and all the bits are here.
253  */
254 static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *skb,
255 			  struct sk_buff *prev_tail, struct net_device *dev)
256 {
257 	struct net *net = container_of(fq->q.net, struct net, ipv6.frags);
258 	unsigned int nhoff;
259 	void *reasm_data;
260 	int payload_len;
261 	u8 ecn;
262 
263 	inet_frag_kill(&fq->q);
264 
265 	ecn = ip_frag_ecn_table[fq->ecn];
266 	if (unlikely(ecn == 0xff))
267 		goto out_fail;
268 
269 	reasm_data = inet_frag_reasm_prepare(&fq->q, skb, prev_tail);
270 	if (!reasm_data)
271 		goto out_oom;
272 
273 	payload_len = ((skb->data - skb_network_header(skb)) -
274 		       sizeof(struct ipv6hdr) + fq->q.len -
275 		       sizeof(struct frag_hdr));
276 	if (payload_len > IPV6_MAXPLEN)
277 		goto out_oversize;
278 
279 	/* We have to remove fragment header from datagram and to relocate
280 	 * header in order to calculate ICV correctly. */
281 	nhoff = fq->nhoffset;
282 	skb_network_header(skb)[nhoff] = skb_transport_header(skb)[0];
283 	memmove(skb->head + sizeof(struct frag_hdr), skb->head,
284 		(skb->data - skb->head) - sizeof(struct frag_hdr));
285 	if (skb_mac_header_was_set(skb))
286 		skb->mac_header += sizeof(struct frag_hdr);
287 	skb->network_header += sizeof(struct frag_hdr);
288 
289 	skb_reset_transport_header(skb);
290 
291 	inet_frag_reasm_finish(&fq->q, skb, reasm_data);
292 
293 	skb->dev = dev;
294 	ipv6_hdr(skb)->payload_len = htons(payload_len);
295 	ipv6_change_dsfield(ipv6_hdr(skb), 0xff, ecn);
296 	IP6CB(skb)->nhoff = nhoff;
297 	IP6CB(skb)->flags |= IP6SKB_FRAGMENTED;
298 	IP6CB(skb)->frag_max_size = fq->q.max_size;
299 
300 	/* Yes, and fold redundant checksum back. 8) */
301 	skb_postpush_rcsum(skb, skb_network_header(skb),
302 			   skb_network_header_len(skb));
303 
304 	rcu_read_lock();
305 	__IP6_INC_STATS(net, __in6_dev_get(dev), IPSTATS_MIB_REASMOKS);
306 	rcu_read_unlock();
307 	fq->q.rb_fragments = RB_ROOT;
308 	fq->q.fragments_tail = NULL;
309 	fq->q.last_run_head = NULL;
310 	return 1;
311 
312 out_oversize:
313 	net_dbg_ratelimited("ip6_frag_reasm: payload len = %d\n", payload_len);
314 	goto out_fail;
315 out_oom:
316 	net_dbg_ratelimited("ip6_frag_reasm: no memory for reassembly\n");
317 out_fail:
318 	rcu_read_lock();
319 	__IP6_INC_STATS(net, __in6_dev_get(dev), IPSTATS_MIB_REASMFAILS);
320 	rcu_read_unlock();
321 	inet_frag_kill(&fq->q);
322 	return -1;
323 }
324 
325 static int ipv6_frag_rcv(struct sk_buff *skb)
326 {
327 	struct frag_hdr *fhdr;
328 	struct frag_queue *fq;
329 	const struct ipv6hdr *hdr = ipv6_hdr(skb);
330 	struct net *net = dev_net(skb_dst(skb)->dev);
331 	int iif;
332 
333 	if (IP6CB(skb)->flags & IP6SKB_FRAGMENTED)
334 		goto fail_hdr;
335 
336 	__IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMREQDS);
337 
338 	/* Jumbo payload inhibits frag. header */
339 	if (hdr->payload_len == 0)
340 		goto fail_hdr;
341 
342 	if (!pskb_may_pull(skb, (skb_transport_offset(skb) +
343 				 sizeof(struct frag_hdr))))
344 		goto fail_hdr;
345 
346 	hdr = ipv6_hdr(skb);
347 	fhdr = (struct frag_hdr *)skb_transport_header(skb);
348 
349 	if (!(fhdr->frag_off & htons(0xFFF9))) {
350 		/* It is not a fragmented frame */
351 		skb->transport_header += sizeof(struct frag_hdr);
352 		__IP6_INC_STATS(net,
353 				ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMOKS);
354 
355 		IP6CB(skb)->nhoff = (u8 *)fhdr - skb_network_header(skb);
356 		IP6CB(skb)->flags |= IP6SKB_FRAGMENTED;
357 		return 1;
358 	}
359 
360 	iif = skb->dev ? skb->dev->ifindex : 0;
361 	fq = fq_find(net, fhdr->identification, hdr, iif);
362 	if (fq) {
363 		u32 prob_offset = 0;
364 		int ret;
365 
366 		spin_lock(&fq->q.lock);
367 
368 		fq->iif = iif;
369 		ret = ip6_frag_queue(fq, skb, fhdr, IP6CB(skb)->nhoff,
370 				     &prob_offset);
371 
372 		spin_unlock(&fq->q.lock);
373 		inet_frag_put(&fq->q);
374 		if (prob_offset) {
375 			__IP6_INC_STATS(net, __in6_dev_get_safely(skb->dev),
376 					IPSTATS_MIB_INHDRERRORS);
377 			/* icmpv6_param_prob() calls kfree_skb(skb) */
378 			icmpv6_param_prob(skb, ICMPV6_HDR_FIELD, prob_offset);
379 		}
380 		return ret;
381 	}
382 
383 	__IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMFAILS);
384 	kfree_skb(skb);
385 	return -1;
386 
387 fail_hdr:
388 	__IP6_INC_STATS(net, __in6_dev_get_safely(skb->dev),
389 			IPSTATS_MIB_INHDRERRORS);
390 	icmpv6_param_prob(skb, ICMPV6_HDR_FIELD, skb_network_header_len(skb));
391 	return -1;
392 }
393 
394 static const struct inet6_protocol frag_protocol = {
395 	.handler	=	ipv6_frag_rcv,
396 	.flags		=	INET6_PROTO_NOPOLICY,
397 };
398 
399 #ifdef CONFIG_SYSCTL
400 
401 static struct ctl_table ip6_frags_ns_ctl_table[] = {
402 	{
403 		.procname	= "ip6frag_high_thresh",
404 		.data		= &init_net.ipv6.frags.high_thresh,
405 		.maxlen		= sizeof(unsigned long),
406 		.mode		= 0644,
407 		.proc_handler	= proc_doulongvec_minmax,
408 		.extra1		= &init_net.ipv6.frags.low_thresh
409 	},
410 	{
411 		.procname	= "ip6frag_low_thresh",
412 		.data		= &init_net.ipv6.frags.low_thresh,
413 		.maxlen		= sizeof(unsigned long),
414 		.mode		= 0644,
415 		.proc_handler	= proc_doulongvec_minmax,
416 		.extra2		= &init_net.ipv6.frags.high_thresh
417 	},
418 	{
419 		.procname	= "ip6frag_time",
420 		.data		= &init_net.ipv6.frags.timeout,
421 		.maxlen		= sizeof(int),
422 		.mode		= 0644,
423 		.proc_handler	= proc_dointvec_jiffies,
424 	},
425 	{ }
426 };
427 
428 /* secret interval has been deprecated */
429 static int ip6_frags_secret_interval_unused;
430 static struct ctl_table ip6_frags_ctl_table[] = {
431 	{
432 		.procname	= "ip6frag_secret_interval",
433 		.data		= &ip6_frags_secret_interval_unused,
434 		.maxlen		= sizeof(int),
435 		.mode		= 0644,
436 		.proc_handler	= proc_dointvec_jiffies,
437 	},
438 	{ }
439 };
440 
441 static int __net_init ip6_frags_ns_sysctl_register(struct net *net)
442 {
443 	struct ctl_table *table;
444 	struct ctl_table_header *hdr;
445 
446 	table = ip6_frags_ns_ctl_table;
447 	if (!net_eq(net, &init_net)) {
448 		table = kmemdup(table, sizeof(ip6_frags_ns_ctl_table), GFP_KERNEL);
449 		if (!table)
450 			goto err_alloc;
451 
452 		table[0].data = &net->ipv6.frags.high_thresh;
453 		table[0].extra1 = &net->ipv6.frags.low_thresh;
454 		table[1].data = &net->ipv6.frags.low_thresh;
455 		table[1].extra2 = &net->ipv6.frags.high_thresh;
456 		table[2].data = &net->ipv6.frags.timeout;
457 	}
458 
459 	hdr = register_net_sysctl(net, "net/ipv6", table);
460 	if (!hdr)
461 		goto err_reg;
462 
463 	net->ipv6.sysctl.frags_hdr = hdr;
464 	return 0;
465 
466 err_reg:
467 	if (!net_eq(net, &init_net))
468 		kfree(table);
469 err_alloc:
470 	return -ENOMEM;
471 }
472 
473 static void __net_exit ip6_frags_ns_sysctl_unregister(struct net *net)
474 {
475 	struct ctl_table *table;
476 
477 	table = net->ipv6.sysctl.frags_hdr->ctl_table_arg;
478 	unregister_net_sysctl_table(net->ipv6.sysctl.frags_hdr);
479 	if (!net_eq(net, &init_net))
480 		kfree(table);
481 }
482 
483 static struct ctl_table_header *ip6_ctl_header;
484 
485 static int ip6_frags_sysctl_register(void)
486 {
487 	ip6_ctl_header = register_net_sysctl(&init_net, "net/ipv6",
488 			ip6_frags_ctl_table);
489 	return ip6_ctl_header == NULL ? -ENOMEM : 0;
490 }
491 
492 static void ip6_frags_sysctl_unregister(void)
493 {
494 	unregister_net_sysctl_table(ip6_ctl_header);
495 }
496 #else
497 static int ip6_frags_ns_sysctl_register(struct net *net)
498 {
499 	return 0;
500 }
501 
502 static void ip6_frags_ns_sysctl_unregister(struct net *net)
503 {
504 }
505 
506 static int ip6_frags_sysctl_register(void)
507 {
508 	return 0;
509 }
510 
511 static void ip6_frags_sysctl_unregister(void)
512 {
513 }
514 #endif
515 
516 static int __net_init ipv6_frags_init_net(struct net *net)
517 {
518 	int res;
519 
520 	net->ipv6.frags.high_thresh = IPV6_FRAG_HIGH_THRESH;
521 	net->ipv6.frags.low_thresh = IPV6_FRAG_LOW_THRESH;
522 	net->ipv6.frags.timeout = IPV6_FRAG_TIMEOUT;
523 	net->ipv6.frags.f = &ip6_frags;
524 
525 	res = inet_frags_init_net(&net->ipv6.frags);
526 	if (res < 0)
527 		return res;
528 
529 	res = ip6_frags_ns_sysctl_register(net);
530 	if (res < 0)
531 		inet_frags_exit_net(&net->ipv6.frags);
532 	return res;
533 }
534 
535 static void __net_exit ipv6_frags_exit_net(struct net *net)
536 {
537 	ip6_frags_ns_sysctl_unregister(net);
538 	inet_frags_exit_net(&net->ipv6.frags);
539 }
540 
541 static struct pernet_operations ip6_frags_ops = {
542 	.init = ipv6_frags_init_net,
543 	.exit = ipv6_frags_exit_net,
544 };
545 
546 static const struct rhashtable_params ip6_rhash_params = {
547 	.head_offset		= offsetof(struct inet_frag_queue, node),
548 	.hashfn			= ip6frag_key_hashfn,
549 	.obj_hashfn		= ip6frag_obj_hashfn,
550 	.obj_cmpfn		= ip6frag_obj_cmpfn,
551 	.automatic_shrinking	= true,
552 };
553 
554 int __init ipv6_frag_init(void)
555 {
556 	int ret;
557 
558 	ip6_frags.constructor = ip6frag_init;
559 	ip6_frags.destructor = NULL;
560 	ip6_frags.qsize = sizeof(struct frag_queue);
561 	ip6_frags.frag_expire = ip6_frag_expire;
562 	ip6_frags.frags_cache_name = ip6_frag_cache_name;
563 	ip6_frags.rhash_params = ip6_rhash_params;
564 	ret = inet_frags_init(&ip6_frags);
565 	if (ret)
566 		goto out;
567 
568 	ret = inet6_add_protocol(&frag_protocol, IPPROTO_FRAGMENT);
569 	if (ret)
570 		goto err_protocol;
571 
572 	ret = ip6_frags_sysctl_register();
573 	if (ret)
574 		goto err_sysctl;
575 
576 	ret = register_pernet_subsys(&ip6_frags_ops);
577 	if (ret)
578 		goto err_pernet;
579 
580 out:
581 	return ret;
582 
583 err_pernet:
584 	ip6_frags_sysctl_unregister();
585 err_sysctl:
586 	inet6_del_protocol(&frag_protocol, IPPROTO_FRAGMENT);
587 err_protocol:
588 	inet_frags_fini(&ip6_frags);
589 	goto out;
590 }
591 
592 void ipv6_frag_exit(void)
593 {
594 	inet_frags_fini(&ip6_frags);
595 	ip6_frags_sysctl_unregister();
596 	unregister_pernet_subsys(&ip6_frags_ops);
597 	inet6_del_protocol(&frag_protocol, IPPROTO_FRAGMENT);
598 }
599