nf_conntrack_reasm.c revision 9e903e085262ffbf1fc44a17ac06058aca03524a 
1/*
2 * IPv6 fragment reassembly for connection tracking
3 *
4 * Copyright (C)2004 USAGI/WIDE Project
5 *
6 * Author:
7 *	Yasuyuki Kozakai @USAGI <yasuyuki.kozakai@toshiba.co.jp>
8 *
9 * Based on: net/ipv6/reassembly.c
10 *
11 * This program is free software; you can redistribute it and/or
12 * modify it under the terms of the GNU General Public License
13 * as published by the Free Software Foundation; either version
14 * 2 of the License, or (at your option) any later version.
15 */
16
17#include <linux/errno.h>
18#include <linux/types.h>
19#include <linux/string.h>
20#include <linux/socket.h>
21#include <linux/sockios.h>
22#include <linux/jiffies.h>
23#include <linux/net.h>
24#include <linux/list.h>
25#include <linux/netdevice.h>
26#include <linux/in6.h>
27#include <linux/ipv6.h>
28#include <linux/icmpv6.h>
29#include <linux/random.h>
30#include <linux/slab.h>
31
32#include <net/sock.h>
33#include <net/snmp.h>
34#include <net/inet_frag.h>
35
36#include <net/ipv6.h>
37#include <net/protocol.h>
38#include <net/transp_v6.h>
39#include <net/rawv6.h>
40#include <net/ndisc.h>
41#include <net/addrconf.h>
42#include <net/netfilter/ipv6/nf_conntrack_ipv6.h>
43#include <linux/sysctl.h>
44#include <linux/netfilter.h>
45#include <linux/netfilter_ipv6.h>
46#include <linux/kernel.h>
47#include <linux/module.h>
48#include <net/netfilter/ipv6/nf_defrag_ipv6.h>
49
50
51struct nf_ct_frag6_skb_cb
52{
53	struct inet6_skb_parm	h;
54	int			offset;
55	struct sk_buff		*orig;
56};
57
58#define NFCT_FRAG6_CB(skb)	((struct nf_ct_frag6_skb_cb*)((skb)->cb))
59
60struct nf_ct_frag6_queue
61{
62	struct inet_frag_queue	q;
63
64	__be32			id;		/* fragment id		*/
65	u32			user;
66	struct in6_addr		saddr;
67	struct in6_addr		daddr;
68
69	unsigned int		csum;
70	__u16			nhoffset;
71};
72
73static struct inet_frags nf_frags;
74static struct netns_frags nf_init_frags;
75
76#ifdef CONFIG_SYSCTL
77static struct ctl_table nf_ct_frag6_sysctl_table[] = {
78	{
79		.procname	= "nf_conntrack_frag6_timeout",
80		.data		= &nf_init_frags.timeout,
81		.maxlen		= sizeof(unsigned int),
82		.mode		= 0644,
83		.proc_handler	= proc_dointvec_jiffies,
84	},
85	{
86		.procname	= "nf_conntrack_frag6_low_thresh",
87		.data		= &nf_init_frags.low_thresh,
88		.maxlen		= sizeof(unsigned int),
89		.mode		= 0644,
90		.proc_handler	= proc_dointvec,
91	},
92	{
93		.procname	= "nf_conntrack_frag6_high_thresh",
94		.data		= &nf_init_frags.high_thresh,
95		.maxlen		= sizeof(unsigned int),
96		.mode		= 0644,
97		.proc_handler	= proc_dointvec,
98	},
99	{ }
100};
101
102static struct ctl_table_header *nf_ct_frag6_sysctl_header;
103#endif
104
105static unsigned int nf_hashfn(struct inet_frag_queue *q)
106{
107	const struct nf_ct_frag6_queue *nq;
108
109	nq = container_of(q, struct nf_ct_frag6_queue, q);
110	return inet6_hash_frag(nq->id, &nq->saddr, &nq->daddr, nf_frags.rnd);
111}
112
113static void nf_skb_free(struct sk_buff *skb)
114{
115	if (NFCT_FRAG6_CB(skb)->orig)
116		kfree_skb(NFCT_FRAG6_CB(skb)->orig);
117}
118
119/* Destruction primitives. */
120
121static __inline__ void fq_put(struct nf_ct_frag6_queue *fq)
122{
123	inet_frag_put(&fq->q, &nf_frags);
124}
125
126/* Kill fq entry. It is not destroyed immediately,
127 * because caller (and someone more) holds reference count.
128 */
129static __inline__ void fq_kill(struct nf_ct_frag6_queue *fq)
130{
131	inet_frag_kill(&fq->q, &nf_frags);
132}
133
134static void nf_ct_frag6_evictor(void)
135{
136	local_bh_disable();
137	inet_frag_evictor(&nf_init_frags, &nf_frags);
138	local_bh_enable();
139}
140
141static void nf_ct_frag6_expire(unsigned long data)
142{
143	struct nf_ct_frag6_queue *fq;
144
145	fq = container_of((struct inet_frag_queue *)data,
146			struct nf_ct_frag6_queue, q);
147
148	spin_lock(&fq->q.lock);
149
150	if (fq->q.last_in & INET_FRAG_COMPLETE)
151		goto out;
152
153	fq_kill(fq);
154
155out:
156	spin_unlock(&fq->q.lock);
157	fq_put(fq);
158}
159
160/* Creation primitives. */
161
162static __inline__ struct nf_ct_frag6_queue *
163fq_find(__be32 id, u32 user, struct in6_addr *src, struct in6_addr *dst)
164{
165	struct inet_frag_queue *q;
166	struct ip6_create_arg arg;
167	unsigned int hash;
168
169	arg.id = id;
170	arg.user = user;
171	arg.src = src;
172	arg.dst = dst;
173
174	read_lock_bh(&nf_frags.lock);
175	hash = inet6_hash_frag(id, src, dst, nf_frags.rnd);
176
177	q = inet_frag_find(&nf_init_frags, &nf_frags, &arg, hash);
178	local_bh_enable();
179	if (q == NULL)
180		goto oom;
181
182	return container_of(q, struct nf_ct_frag6_queue, q);
183
184oom:
185	pr_debug("Can't alloc new queue\n");
186	return NULL;
187}
188
189
190static int nf_ct_frag6_queue(struct nf_ct_frag6_queue *fq, struct sk_buff *skb,
191			     const struct frag_hdr *fhdr, int nhoff)
192{
193	struct sk_buff *prev, *next;
194	int offset, end;
195
196	if (fq->q.last_in & INET_FRAG_COMPLETE) {
197		pr_debug("Already completed\n");
198		goto err;
199	}
200
201	offset = ntohs(fhdr->frag_off) & ~0x7;
202	end = offset + (ntohs(ipv6_hdr(skb)->payload_len) -
203			((u8 *)(fhdr + 1) - (u8 *)(ipv6_hdr(skb) + 1)));
204
205	if ((unsigned int)end > IPV6_MAXPLEN) {
206		pr_debug("offset is too large.\n");
207		return -1;
208	}
209
210	if (skb->ip_summed == CHECKSUM_COMPLETE) {
211		const unsigned char *nh = skb_network_header(skb);
212		skb->csum = csum_sub(skb->csum,
213				     csum_partial(nh, (u8 *)(fhdr + 1) - nh,
214						  0));
215	}
216
217	/* Is this the final fragment? */
218	if (!(fhdr->frag_off & htons(IP6_MF))) {
219		/* If we already have some bits beyond end
220		 * or have different end, the segment is corrupted.
221		 */
222		if (end < fq->q.len ||
223		    ((fq->q.last_in & INET_FRAG_LAST_IN) && end != fq->q.len)) {
224			pr_debug("already received last fragment\n");
225			goto err;
226		}
227		fq->q.last_in |= INET_FRAG_LAST_IN;
228		fq->q.len = end;
229	} else {
230		/* Check if the fragment is rounded to 8 bytes.
231		 * Required by the RFC.
232		 */
233		if (end & 0x7) {
234			/* RFC2460 says always send parameter problem in
235			 * this case. -DaveM
236			 */
237			pr_debug("end of fragment not rounded to 8 bytes.\n");
238			return -1;
239		}
240		if (end > fq->q.len) {
241			/* Some bits beyond end -> corruption. */
242			if (fq->q.last_in & INET_FRAG_LAST_IN) {
243				pr_debug("last packet already reached.\n");
244				goto err;
245			}
246			fq->q.len = end;
247		}
248	}
249
250	if (end == offset)
251		goto err;
252
253	/* Point into the IP datagram 'data' part. */
254	if (!pskb_pull(skb, (u8 *) (fhdr + 1) - skb->data)) {
255		pr_debug("queue: message is too short.\n");
256		goto err;
257	}
258	if (pskb_trim_rcsum(skb, end - offset)) {
259		pr_debug("Can't trim\n");
260		goto err;
261	}
262
263	/* Find out which fragments are in front and at the back of us
264	 * in the chain of fragments so far.  We must know where to put
265	 * this fragment, right?
266	 */
267	prev = fq->q.fragments_tail;
268	if (!prev || NFCT_FRAG6_CB(prev)->offset < offset) {
269		next = NULL;
270		goto found;
271	}
272	prev = NULL;
273	for (next = fq->q.fragments; next != NULL; next = next->next) {
274		if (NFCT_FRAG6_CB(next)->offset >= offset)
275			break;	/* bingo! */
276		prev = next;
277	}
278
279found:
280	/* RFC5722, Section 4:
281	 *                                  When reassembling an IPv6 datagram, if
282	 *   one or more its constituent fragments is determined to be an
283	 *   overlapping fragment, the entire datagram (and any constituent
284	 *   fragments, including those not yet received) MUST be silently
285	 *   discarded.
286	 */
287
288	/* Check for overlap with preceding fragment. */
289	if (prev &&
290	    (NFCT_FRAG6_CB(prev)->offset + prev->len) > offset)
291		goto discard_fq;
292
293	/* Look for overlap with succeeding segment. */
294	if (next && NFCT_FRAG6_CB(next)->offset < end)
295		goto discard_fq;
296
297	NFCT_FRAG6_CB(skb)->offset = offset;
298
299	/* Insert this fragment in the chain of fragments. */
300	skb->next = next;
301	if (!next)
302		fq->q.fragments_tail = skb;
303	if (prev)
304		prev->next = skb;
305	else
306		fq->q.fragments = skb;
307
308	skb->dev = NULL;
309	fq->q.stamp = skb->tstamp;
310	fq->q.meat += skb->len;
311	atomic_add(skb->truesize, &nf_init_frags.mem);
312
313	/* The first fragment.
314	 * nhoffset is obtained from the first fragment, of course.
315	 */
316	if (offset == 0) {
317		fq->nhoffset = nhoff;
318		fq->q.last_in |= INET_FRAG_FIRST_IN;
319	}
320	write_lock(&nf_frags.lock);
321	list_move_tail(&fq->q.lru_list, &nf_init_frags.lru_list);
322	write_unlock(&nf_frags.lock);
323	return 0;
324
325discard_fq:
326	fq_kill(fq);
327err:
328	return -1;
329}
330
331/*
332 *	Check if this packet is complete.
333 *	Returns NULL on failure by any reason, and pointer
334 *	to current nexthdr field in reassembled frame.
335 *
336 *	It is called with locked fq, and caller must check that
337 *	queue is eligible for reassembly i.e. it is not COMPLETE,
338 *	the last and the first frames arrived and all the bits are here.
339 */
340static struct sk_buff *
341nf_ct_frag6_reasm(struct nf_ct_frag6_queue *fq, struct net_device *dev)
342{
343	struct sk_buff *fp, *op, *head = fq->q.fragments;
344	int    payload_len;
345
346	fq_kill(fq);
347
348	WARN_ON(head == NULL);
349	WARN_ON(NFCT_FRAG6_CB(head)->offset != 0);
350
351	/* Unfragmented part is taken from the first segment. */
352	payload_len = ((head->data - skb_network_header(head)) -
353		       sizeof(struct ipv6hdr) + fq->q.len -
354		       sizeof(struct frag_hdr));
355	if (payload_len > IPV6_MAXPLEN) {
356		pr_debug("payload len is too large.\n");
357		goto out_oversize;
358	}
359
360	/* Head of list must not be cloned. */
361	if (skb_cloned(head) && pskb_expand_head(head, 0, 0, GFP_ATOMIC)) {
362		pr_debug("skb is cloned but can't expand head");
363		goto out_oom;
364	}
365
366	/* If the first fragment is fragmented itself, we split
367	 * it to two chunks: the first with data and paged part
368	 * and the second, holding only fragments. */
369	if (skb_has_frag_list(head)) {
370		struct sk_buff *clone;
371		int i, plen = 0;
372
373		if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL) {
374			pr_debug("Can't alloc skb\n");
375			goto out_oom;
376		}
377		clone->next = head->next;
378		head->next = clone;
379		skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
380		skb_frag_list_init(head);
381		for (i = 0; i < skb_shinfo(head)->nr_frags; i++)
382			plen += skb_frag_size(&skb_shinfo(head)->frags[i]);
383		clone->len = clone->data_len = head->data_len - plen;
384		head->data_len -= clone->len;
385		head->len -= clone->len;
386		clone->csum = 0;
387		clone->ip_summed = head->ip_summed;
388
389		NFCT_FRAG6_CB(clone)->orig = NULL;
390		atomic_add(clone->truesize, &nf_init_frags.mem);
391	}
392
393	/* We have to remove fragment header from datagram and to relocate
394	 * header in order to calculate ICV correctly. */
395	skb_network_header(head)[fq->nhoffset] = skb_transport_header(head)[0];
396	memmove(head->head + sizeof(struct frag_hdr), head->head,
397		(head->data - head->head) - sizeof(struct frag_hdr));
398	head->mac_header += sizeof(struct frag_hdr);
399	head->network_header += sizeof(struct frag_hdr);
400
401	skb_shinfo(head)->frag_list = head->next;
402	skb_reset_transport_header(head);
403	skb_push(head, head->data - skb_network_header(head));
404
405	for (fp=head->next; fp; fp = fp->next) {
406		head->data_len += fp->len;
407		head->len += fp->len;
408		if (head->ip_summed != fp->ip_summed)
409			head->ip_summed = CHECKSUM_NONE;
410		else if (head->ip_summed == CHECKSUM_COMPLETE)
411			head->csum = csum_add(head->csum, fp->csum);
412		head->truesize += fp->truesize;
413	}
414	atomic_sub(head->truesize, &nf_init_frags.mem);
415
416	head->next = NULL;
417	head->dev = dev;
418	head->tstamp = fq->q.stamp;
419	ipv6_hdr(head)->payload_len = htons(payload_len);
420
421	/* Yes, and fold redundant checksum back. 8) */
422	if (head->ip_summed == CHECKSUM_COMPLETE)
423		head->csum = csum_partial(skb_network_header(head),
424					  skb_network_header_len(head),
425					  head->csum);
426
427	fq->q.fragments = NULL;
428	fq->q.fragments_tail = NULL;
429
430	/* all original skbs are linked into the NFCT_FRAG6_CB(head).orig */
431	fp = skb_shinfo(head)->frag_list;
432	if (fp && NFCT_FRAG6_CB(fp)->orig == NULL)
433		/* at above code, head skb is divided into two skbs. */
434		fp = fp->next;
435
436	op = NFCT_FRAG6_CB(head)->orig;
437	for (; fp; fp = fp->next) {
438		struct sk_buff *orig = NFCT_FRAG6_CB(fp)->orig;
439
440		op->next = orig;
441		op = orig;
442		NFCT_FRAG6_CB(fp)->orig = NULL;
443	}
444
445	return head;
446
447out_oversize:
448	if (net_ratelimit())
449		printk(KERN_DEBUG "nf_ct_frag6_reasm: payload len = %d\n", payload_len);
450	goto out_fail;
451out_oom:
452	if (net_ratelimit())
453		printk(KERN_DEBUG "nf_ct_frag6_reasm: no memory for reassembly\n");
454out_fail:
455	return NULL;
456}
457
458/*
459 * find the header just before Fragment Header.
460 *
461 * if success return 0 and set ...
462 * (*prevhdrp): the value of "Next Header Field" in the header
463 *		just before Fragment Header.
464 * (*prevhoff): the offset of "Next Header Field" in the header
465 *		just before Fragment Header.
466 * (*fhoff)   : the offset of Fragment Header.
467 *
468 * Based on ipv6_skip_hdr() in net/ipv6/exthdr.c
469 *
470 */
471static int
472find_prev_fhdr(struct sk_buff *skb, u8 *prevhdrp, int *prevhoff, int *fhoff)
473{
474	u8 nexthdr = ipv6_hdr(skb)->nexthdr;
475	const int netoff = skb_network_offset(skb);
476	u8 prev_nhoff = netoff + offsetof(struct ipv6hdr, nexthdr);
477	int start = netoff + sizeof(struct ipv6hdr);
478	int len = skb->len - start;
479	u8 prevhdr = NEXTHDR_IPV6;
480
481	while (nexthdr != NEXTHDR_FRAGMENT) {
482		struct ipv6_opt_hdr hdr;
483		int hdrlen;
484
485		if (!ipv6_ext_hdr(nexthdr)) {
486			return -1;
487		}
488		if (nexthdr == NEXTHDR_NONE) {
489			pr_debug("next header is none\n");
490			return -1;
491		}
492		if (len < (int)sizeof(struct ipv6_opt_hdr)) {
493			pr_debug("too short\n");
494			return -1;
495		}
496		if (skb_copy_bits(skb, start, &hdr, sizeof(hdr)))
497			BUG();
498		if (nexthdr == NEXTHDR_AUTH)
499			hdrlen = (hdr.hdrlen+2)<<2;
500		else
501			hdrlen = ipv6_optlen(&hdr);
502
503		prevhdr = nexthdr;
504		prev_nhoff = start;
505
506		nexthdr = hdr.nexthdr;
507		len -= hdrlen;
508		start += hdrlen;
509	}
510
511	if (len < 0)
512		return -1;
513
514	*prevhdrp = prevhdr;
515	*prevhoff = prev_nhoff;
516	*fhoff = start;
517
518	return 0;
519}
520
521struct sk_buff *nf_ct_frag6_gather(struct sk_buff *skb, u32 user)
522{
523	struct sk_buff *clone;
524	struct net_device *dev = skb->dev;
525	struct frag_hdr *fhdr;
526	struct nf_ct_frag6_queue *fq;
527	struct ipv6hdr *hdr;
528	int fhoff, nhoff;
529	u8 prevhdr;
530	struct sk_buff *ret_skb = NULL;
531
532	/* Jumbo payload inhibits frag. header */
533	if (ipv6_hdr(skb)->payload_len == 0) {
534		pr_debug("payload len = 0\n");
535		return skb;
536	}
537
538	if (find_prev_fhdr(skb, &prevhdr, &nhoff, &fhoff) < 0)
539		return skb;
540
541	clone = skb_clone(skb, GFP_ATOMIC);
542	if (clone == NULL) {
543		pr_debug("Can't clone skb\n");
544		return skb;
545	}
546
547	NFCT_FRAG6_CB(clone)->orig = skb;
548
549	if (!pskb_may_pull(clone, fhoff + sizeof(*fhdr))) {
550		pr_debug("message is too short.\n");
551		goto ret_orig;
552	}
553
554	skb_set_transport_header(clone, fhoff);
555	hdr = ipv6_hdr(clone);
556	fhdr = (struct frag_hdr *)skb_transport_header(clone);
557
558	if (atomic_read(&nf_init_frags.mem) > nf_init_frags.high_thresh)
559		nf_ct_frag6_evictor();
560
561	fq = fq_find(fhdr->identification, user, &hdr->saddr, &hdr->daddr);
562	if (fq == NULL) {
563		pr_debug("Can't find and can't create new queue\n");
564		goto ret_orig;
565	}
566
567	spin_lock_bh(&fq->q.lock);
568
569	if (nf_ct_frag6_queue(fq, clone, fhdr, nhoff) < 0) {
570		spin_unlock_bh(&fq->q.lock);
571		pr_debug("Can't insert skb to queue\n");
572		fq_put(fq);
573		goto ret_orig;
574	}
575
576	if (fq->q.last_in == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
577	    fq->q.meat == fq->q.len) {
578		ret_skb = nf_ct_frag6_reasm(fq, dev);
579		if (ret_skb == NULL)
580			pr_debug("Can't reassemble fragmented packets\n");
581	}
582	spin_unlock_bh(&fq->q.lock);
583
584	fq_put(fq);
585	return ret_skb;
586
587ret_orig:
588	kfree_skb(clone);
589	return skb;
590}
591
592void nf_ct_frag6_output(unsigned int hooknum, struct sk_buff *skb,
593			struct net_device *in, struct net_device *out,
594			int (*okfn)(struct sk_buff *))
595{
596	struct sk_buff *s, *s2;
597
598	for (s = NFCT_FRAG6_CB(skb)->orig; s;) {
599		nf_conntrack_put_reasm(s->nfct_reasm);
600		nf_conntrack_get_reasm(skb);
601		s->nfct_reasm = skb;
602
603		s2 = s->next;
604		s->next = NULL;
605
606		NF_HOOK_THRESH(NFPROTO_IPV6, hooknum, s, in, out, okfn,
607			       NF_IP6_PRI_CONNTRACK_DEFRAG + 1);
608		s = s2;
609	}
610	nf_conntrack_put_reasm(skb);
611}
612
613int nf_ct_frag6_init(void)
614{
615	nf_frags.hashfn = nf_hashfn;
616	nf_frags.constructor = ip6_frag_init;
617	nf_frags.destructor = NULL;
618	nf_frags.skb_free = nf_skb_free;
619	nf_frags.qsize = sizeof(struct nf_ct_frag6_queue);
620	nf_frags.match = ip6_frag_match;
621	nf_frags.frag_expire = nf_ct_frag6_expire;
622	nf_frags.secret_interval = 10 * 60 * HZ;
623	nf_init_frags.timeout = IPV6_FRAG_TIMEOUT;
624	nf_init_frags.high_thresh = IPV6_FRAG_HIGH_THRESH;
625	nf_init_frags.low_thresh = IPV6_FRAG_LOW_THRESH;
626	inet_frags_init_net(&nf_init_frags);
627	inet_frags_init(&nf_frags);
628
629#ifdef CONFIG_SYSCTL
630	nf_ct_frag6_sysctl_header = register_sysctl_paths(nf_net_netfilter_sysctl_path,
631							  nf_ct_frag6_sysctl_table);
632	if (!nf_ct_frag6_sysctl_header) {
633		inet_frags_fini(&nf_frags);
634		return -ENOMEM;
635	}
636#endif
637
638	return 0;
639}
640
641void nf_ct_frag6_cleanup(void)
642{
643#ifdef CONFIG_SYSCTL
644	unregister_sysctl_table(nf_ct_frag6_sysctl_header);
645	nf_ct_frag6_sysctl_header = NULL;
646#endif
647	inet_frags_fini(&nf_frags);
648
649	nf_init_frags.low_thresh = 0;
650	nf_ct_frag6_evictor();
651}
652