nf_conntrack_reasm.c revision 86e93e470cadedda9181a2bd9aee1d9d2e5e9c0f 
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#define pr_fmt(fmt) "IPv6-nf: " fmt
18
19#include <linux/errno.h>
20#include <linux/types.h>
21#include <linux/string.h>
22#include <linux/socket.h>
23#include <linux/sockios.h>
24#include <linux/jiffies.h>
25#include <linux/net.h>
26#include <linux/list.h>
27#include <linux/netdevice.h>
28#include <linux/in6.h>
29#include <linux/ipv6.h>
30#include <linux/icmpv6.h>
31#include <linux/random.h>
32#include <linux/slab.h>
33
34#include <net/sock.h>
35#include <net/snmp.h>
36#include <net/inet_frag.h>
37
38#include <net/ipv6.h>
39#include <net/protocol.h>
40#include <net/transp_v6.h>
41#include <net/rawv6.h>
42#include <net/ndisc.h>
43#include <net/addrconf.h>
44#include <net/inet_ecn.h>
45#include <net/netfilter/ipv6/nf_conntrack_ipv6.h>
46#include <linux/sysctl.h>
47#include <linux/netfilter.h>
48#include <linux/netfilter_ipv6.h>
49#include <linux/kernel.h>
50#include <linux/module.h>
51#include <net/netfilter/ipv6/nf_defrag_ipv6.h>
52
53
54struct nf_ct_frag6_skb_cb
55{
56	struct inet6_skb_parm	h;
57	int			offset;
58	struct sk_buff		*orig;
59};
60
61#define NFCT_FRAG6_CB(skb)	((struct nf_ct_frag6_skb_cb*)((skb)->cb))
62
63static struct inet_frags nf_frags;
64
65#ifdef CONFIG_SYSCTL
66static struct ctl_table nf_ct_frag6_sysctl_table[] = {
67	{
68		.procname	= "nf_conntrack_frag6_timeout",
69		.data		= &init_net.nf_frag.frags.timeout,
70		.maxlen		= sizeof(unsigned int),
71		.mode		= 0644,
72		.proc_handler	= proc_dointvec_jiffies,
73	},
74	{
75		.procname	= "nf_conntrack_frag6_low_thresh",
76		.data		= &init_net.nf_frag.frags.low_thresh,
77		.maxlen		= sizeof(unsigned int),
78		.mode		= 0644,
79		.proc_handler	= proc_dointvec,
80	},
81	{
82		.procname	= "nf_conntrack_frag6_high_thresh",
83		.data		= &init_net.nf_frag.frags.high_thresh,
84		.maxlen		= sizeof(unsigned int),
85		.mode		= 0644,
86		.proc_handler	= proc_dointvec,
87	},
88	{ }
89};
90
91static int nf_ct_frag6_sysctl_register(struct net *net)
92{
93	struct ctl_table *table;
94	struct ctl_table_header *hdr;
95
96	table = nf_ct_frag6_sysctl_table;
97	if (!net_eq(net, &init_net)) {
98		table = kmemdup(table, sizeof(nf_ct_frag6_sysctl_table),
99				GFP_KERNEL);
100		if (table == NULL)
101			goto err_alloc;
102
103		table[0].data = &net->nf_frag.frags.timeout;
104		table[1].data = &net->nf_frag.frags.low_thresh;
105		table[2].data = &net->nf_frag.frags.high_thresh;
106	}
107
108	hdr = register_net_sysctl(net, "net/netfilter", table);
109	if (hdr == NULL)
110		goto err_reg;
111
112	net->nf_frag.sysctl.frags_hdr = hdr;
113	return 0;
114
115err_reg:
116	if (!net_eq(net, &init_net))
117		kfree(table);
118err_alloc:
119	return -ENOMEM;
120}
121
122static void __net_exit nf_ct_frags6_sysctl_unregister(struct net *net)
123{
124	struct ctl_table *table;
125
126	table = net->nf_frag.sysctl.frags_hdr->ctl_table_arg;
127	unregister_net_sysctl_table(net->nf_frag.sysctl.frags_hdr);
128	if (!net_eq(net, &init_net))
129		kfree(table);
130}
131
132#else
133static int nf_ct_frag6_sysctl_register(struct net *net)
134{
135	return 0;
136}
137static void __net_exit nf_ct_frags6_sysctl_unregister(struct net *net)
138{
139}
140#endif
141
142static inline u8 ip6_frag_ecn(const struct ipv6hdr *ipv6h)
143{
144	return 1 << (ipv6_get_dsfield(ipv6h) & INET_ECN_MASK);
145}
146
147static unsigned int nf_hash_frag(__be32 id, const struct in6_addr *saddr,
148				 const struct in6_addr *daddr)
149{
150	net_get_random_once(&nf_frags.rnd, sizeof(nf_frags.rnd));
151	return jhash_3words(ipv6_addr_hash(saddr), ipv6_addr_hash(daddr),
152			    (__force u32)id, nf_frags.rnd);
153}
154
155
156static unsigned int nf_hashfn(const struct inet_frag_queue *q)
157{
158	const struct frag_queue *nq;
159
160	nq = container_of(q, struct frag_queue, q);
161	return nf_hash_frag(nq->id, &nq->saddr, &nq->daddr);
162}
163
164static void nf_skb_free(struct sk_buff *skb)
165{
166	if (NFCT_FRAG6_CB(skb)->orig)
167		kfree_skb(NFCT_FRAG6_CB(skb)->orig);
168}
169
170static void nf_ct_frag6_expire(unsigned long data)
171{
172	struct frag_queue *fq;
173	struct net *net;
174
175	fq = container_of((struct inet_frag_queue *)data, struct frag_queue, q);
176	net = container_of(fq->q.net, struct net, nf_frag.frags);
177
178	ip6_expire_frag_queue(net, fq, &nf_frags);
179}
180
181/* Creation primitives. */
182static inline struct frag_queue *fq_find(struct net *net, __be32 id,
183					 u32 user, struct in6_addr *src,
184					 struct in6_addr *dst, u8 ecn)
185{
186	struct inet_frag_queue *q;
187	struct ip6_create_arg arg;
188	unsigned int hash;
189
190	arg.id = id;
191	arg.user = user;
192	arg.src = src;
193	arg.dst = dst;
194	arg.ecn = ecn;
195
196	read_lock_bh(&nf_frags.lock);
197	hash = nf_hash_frag(id, src, dst);
198
199	q = inet_frag_find(&net->nf_frag.frags, &nf_frags, &arg, hash);
200	local_bh_enable();
201	if (IS_ERR_OR_NULL(q)) {
202		inet_frag_maybe_warn_overflow(q, pr_fmt());
203		return NULL;
204	}
205	return container_of(q, struct frag_queue, q);
206}
207
208
209static int nf_ct_frag6_queue(struct frag_queue *fq, struct sk_buff *skb,
210			     const struct frag_hdr *fhdr, int nhoff)
211{
212	struct sk_buff *prev, *next;
213	unsigned int payload_len;
214	int offset, end;
215	u8 ecn;
216
217	if (fq->q.last_in & INET_FRAG_COMPLETE) {
218		pr_debug("Already completed\n");
219		goto err;
220	}
221
222	payload_len = ntohs(ipv6_hdr(skb)->payload_len);
223
224	offset = ntohs(fhdr->frag_off) & ~0x7;
225	end = offset + (payload_len -
226			((u8 *)(fhdr + 1) - (u8 *)(ipv6_hdr(skb) + 1)));
227
228	if ((unsigned int)end > IPV6_MAXPLEN) {
229		pr_debug("offset is too large.\n");
230		return -1;
231	}
232
233	ecn = ip6_frag_ecn(ipv6_hdr(skb));
234
235	if (skb->ip_summed == CHECKSUM_COMPLETE) {
236		const unsigned char *nh = skb_network_header(skb);
237		skb->csum = csum_sub(skb->csum,
238				     csum_partial(nh, (u8 *)(fhdr + 1) - nh,
239						  0));
240	}
241
242	/* Is this the final fragment? */
243	if (!(fhdr->frag_off & htons(IP6_MF))) {
244		/* If we already have some bits beyond end
245		 * or have different end, the segment is corrupted.
246		 */
247		if (end < fq->q.len ||
248		    ((fq->q.last_in & INET_FRAG_LAST_IN) && end != fq->q.len)) {
249			pr_debug("already received last fragment\n");
250			goto err;
251		}
252		fq->q.last_in |= INET_FRAG_LAST_IN;
253		fq->q.len = end;
254	} else {
255		/* Check if the fragment is rounded to 8 bytes.
256		 * Required by the RFC.
257		 */
258		if (end & 0x7) {
259			/* RFC2460 says always send parameter problem in
260			 * this case. -DaveM
261			 */
262			pr_debug("end of fragment not rounded to 8 bytes.\n");
263			return -1;
264		}
265		if (end > fq->q.len) {
266			/* Some bits beyond end -> corruption. */
267			if (fq->q.last_in & INET_FRAG_LAST_IN) {
268				pr_debug("last packet already reached.\n");
269				goto err;
270			}
271			fq->q.len = end;
272		}
273	}
274
275	if (end == offset)
276		goto err;
277
278	/* Point into the IP datagram 'data' part. */
279	if (!pskb_pull(skb, (u8 *) (fhdr + 1) - skb->data)) {
280		pr_debug("queue: message is too short.\n");
281		goto err;
282	}
283	if (pskb_trim_rcsum(skb, end - offset)) {
284		pr_debug("Can't trim\n");
285		goto err;
286	}
287
288	/* Find out which fragments are in front and at the back of us
289	 * in the chain of fragments so far.  We must know where to put
290	 * this fragment, right?
291	 */
292	prev = fq->q.fragments_tail;
293	if (!prev || NFCT_FRAG6_CB(prev)->offset < offset) {
294		next = NULL;
295		goto found;
296	}
297	prev = NULL;
298	for (next = fq->q.fragments; next != NULL; next = next->next) {
299		if (NFCT_FRAG6_CB(next)->offset >= offset)
300			break;	/* bingo! */
301		prev = next;
302	}
303
304found:
305	/* RFC5722, Section 4:
306	 *                                  When reassembling an IPv6 datagram, if
307	 *   one or more its constituent fragments is determined to be an
308	 *   overlapping fragment, the entire datagram (and any constituent
309	 *   fragments, including those not yet received) MUST be silently
310	 *   discarded.
311	 */
312
313	/* Check for overlap with preceding fragment. */
314	if (prev &&
315	    (NFCT_FRAG6_CB(prev)->offset + prev->len) > offset)
316		goto discard_fq;
317
318	/* Look for overlap with succeeding segment. */
319	if (next && NFCT_FRAG6_CB(next)->offset < end)
320		goto discard_fq;
321
322	NFCT_FRAG6_CB(skb)->offset = offset;
323
324	/* Insert this fragment in the chain of fragments. */
325	skb->next = next;
326	if (!next)
327		fq->q.fragments_tail = skb;
328	if (prev)
329		prev->next = skb;
330	else
331		fq->q.fragments = skb;
332
333	if (skb->dev) {
334		fq->iif = skb->dev->ifindex;
335		skb->dev = NULL;
336	}
337	fq->q.stamp = skb->tstamp;
338	fq->q.meat += skb->len;
339	fq->ecn |= ecn;
340	if (payload_len > fq->q.max_size)
341		fq->q.max_size = payload_len;
342	add_frag_mem_limit(&fq->q, skb->truesize);
343
344	/* The first fragment.
345	 * nhoffset is obtained from the first fragment, of course.
346	 */
347	if (offset == 0) {
348		fq->nhoffset = nhoff;
349		fq->q.last_in |= INET_FRAG_FIRST_IN;
350	}
351
352	inet_frag_lru_move(&fq->q);
353	return 0;
354
355discard_fq:
356	inet_frag_kill(&fq->q, &nf_frags);
357err:
358	return -1;
359}
360
361/*
362 *	Check if this packet is complete.
363 *	Returns NULL on failure by any reason, and pointer
364 *	to current nexthdr field in reassembled frame.
365 *
366 *	It is called with locked fq, and caller must check that
367 *	queue is eligible for reassembly i.e. it is not COMPLETE,
368 *	the last and the first frames arrived and all the bits are here.
369 */
370static struct sk_buff *
371nf_ct_frag6_reasm(struct frag_queue *fq, struct net_device *dev)
372{
373	struct sk_buff *fp, *op, *head = fq->q.fragments;
374	int    payload_len;
375	u8 ecn;
376
377	inet_frag_kill(&fq->q, &nf_frags);
378
379	WARN_ON(head == NULL);
380	WARN_ON(NFCT_FRAG6_CB(head)->offset != 0);
381
382	ecn = ip_frag_ecn_table[fq->ecn];
383	if (unlikely(ecn == 0xff))
384		goto out_fail;
385
386	/* Unfragmented part is taken from the first segment. */
387	payload_len = ((head->data - skb_network_header(head)) -
388		       sizeof(struct ipv6hdr) + fq->q.len -
389		       sizeof(struct frag_hdr));
390	if (payload_len > IPV6_MAXPLEN) {
391		pr_debug("payload len is too large.\n");
392		goto out_oversize;
393	}
394
395	/* Head of list must not be cloned. */
396	if (skb_unclone(head, GFP_ATOMIC)) {
397		pr_debug("skb is cloned but can't expand head");
398		goto out_oom;
399	}
400
401	/* If the first fragment is fragmented itself, we split
402	 * it to two chunks: the first with data and paged part
403	 * and the second, holding only fragments. */
404	if (skb_has_frag_list(head)) {
405		struct sk_buff *clone;
406		int i, plen = 0;
407
408		clone = alloc_skb(0, GFP_ATOMIC);
409		if (clone == NULL)
410			goto out_oom;
411
412		clone->next = head->next;
413		head->next = clone;
414		skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
415		skb_frag_list_init(head);
416		for (i = 0; i < skb_shinfo(head)->nr_frags; i++)
417			plen += skb_frag_size(&skb_shinfo(head)->frags[i]);
418		clone->len = clone->data_len = head->data_len - plen;
419		head->data_len -= clone->len;
420		head->len -= clone->len;
421		clone->csum = 0;
422		clone->ip_summed = head->ip_summed;
423
424		NFCT_FRAG6_CB(clone)->orig = NULL;
425		add_frag_mem_limit(&fq->q, clone->truesize);
426	}
427
428	/* We have to remove fragment header from datagram and to relocate
429	 * header in order to calculate ICV correctly. */
430	skb_network_header(head)[fq->nhoffset] = skb_transport_header(head)[0];
431	memmove(head->head + sizeof(struct frag_hdr), head->head,
432		(head->data - head->head) - sizeof(struct frag_hdr));
433	head->mac_header += sizeof(struct frag_hdr);
434	head->network_header += sizeof(struct frag_hdr);
435
436	skb_shinfo(head)->frag_list = head->next;
437	skb_reset_transport_header(head);
438	skb_push(head, head->data - skb_network_header(head));
439
440	for (fp=head->next; fp; fp = fp->next) {
441		head->data_len += fp->len;
442		head->len += fp->len;
443		if (head->ip_summed != fp->ip_summed)
444			head->ip_summed = CHECKSUM_NONE;
445		else if (head->ip_summed == CHECKSUM_COMPLETE)
446			head->csum = csum_add(head->csum, fp->csum);
447		head->truesize += fp->truesize;
448	}
449	sub_frag_mem_limit(&fq->q, head->truesize);
450
451	head->ignore_df = 1;
452	head->next = NULL;
453	head->dev = dev;
454	head->tstamp = fq->q.stamp;
455	ipv6_hdr(head)->payload_len = htons(payload_len);
456	ipv6_change_dsfield(ipv6_hdr(head), 0xff, ecn);
457	IP6CB(head)->frag_max_size = sizeof(struct ipv6hdr) + fq->q.max_size;
458
459	/* Yes, and fold redundant checksum back. 8) */
460	if (head->ip_summed == CHECKSUM_COMPLETE)
461		head->csum = csum_partial(skb_network_header(head),
462					  skb_network_header_len(head),
463					  head->csum);
464
465	fq->q.fragments = NULL;
466	fq->q.fragments_tail = NULL;
467
468	/* all original skbs are linked into the NFCT_FRAG6_CB(head).orig */
469	fp = skb_shinfo(head)->frag_list;
470	if (fp && NFCT_FRAG6_CB(fp)->orig == NULL)
471		/* at above code, head skb is divided into two skbs. */
472		fp = fp->next;
473
474	op = NFCT_FRAG6_CB(head)->orig;
475	for (; fp; fp = fp->next) {
476		struct sk_buff *orig = NFCT_FRAG6_CB(fp)->orig;
477
478		op->next = orig;
479		op = orig;
480		NFCT_FRAG6_CB(fp)->orig = NULL;
481	}
482
483	return head;
484
485out_oversize:
486	net_dbg_ratelimited("nf_ct_frag6_reasm: payload len = %d\n",
487			    payload_len);
488	goto out_fail;
489out_oom:
490	net_dbg_ratelimited("nf_ct_frag6_reasm: no memory for reassembly\n");
491out_fail:
492	return NULL;
493}
494
495/*
496 * find the header just before Fragment Header.
497 *
498 * if success return 0 and set ...
499 * (*prevhdrp): the value of "Next Header Field" in the header
500 *		just before Fragment Header.
501 * (*prevhoff): the offset of "Next Header Field" in the header
502 *		just before Fragment Header.
503 * (*fhoff)   : the offset of Fragment Header.
504 *
505 * Based on ipv6_skip_hdr() in net/ipv6/exthdr.c
506 *
507 */
508static int
509find_prev_fhdr(struct sk_buff *skb, u8 *prevhdrp, int *prevhoff, int *fhoff)
510{
511	u8 nexthdr = ipv6_hdr(skb)->nexthdr;
512	const int netoff = skb_network_offset(skb);
513	u8 prev_nhoff = netoff + offsetof(struct ipv6hdr, nexthdr);
514	int start = netoff + sizeof(struct ipv6hdr);
515	int len = skb->len - start;
516	u8 prevhdr = NEXTHDR_IPV6;
517
518	while (nexthdr != NEXTHDR_FRAGMENT) {
519		struct ipv6_opt_hdr hdr;
520		int hdrlen;
521
522		if (!ipv6_ext_hdr(nexthdr)) {
523			return -1;
524		}
525		if (nexthdr == NEXTHDR_NONE) {
526			pr_debug("next header is none\n");
527			return -1;
528		}
529		if (len < (int)sizeof(struct ipv6_opt_hdr)) {
530			pr_debug("too short\n");
531			return -1;
532		}
533		if (skb_copy_bits(skb, start, &hdr, sizeof(hdr)))
534			BUG();
535		if (nexthdr == NEXTHDR_AUTH)
536			hdrlen = (hdr.hdrlen+2)<<2;
537		else
538			hdrlen = ipv6_optlen(&hdr);
539
540		prevhdr = nexthdr;
541		prev_nhoff = start;
542
543		nexthdr = hdr.nexthdr;
544		len -= hdrlen;
545		start += hdrlen;
546	}
547
548	if (len < 0)
549		return -1;
550
551	*prevhdrp = prevhdr;
552	*prevhoff = prev_nhoff;
553	*fhoff = start;
554
555	return 0;
556}
557
558struct sk_buff *nf_ct_frag6_gather(struct sk_buff *skb, u32 user)
559{
560	struct sk_buff *clone;
561	struct net_device *dev = skb->dev;
562	struct net *net = skb_dst(skb) ? dev_net(skb_dst(skb)->dev)
563				       : dev_net(skb->dev);
564	struct frag_hdr *fhdr;
565	struct frag_queue *fq;
566	struct ipv6hdr *hdr;
567	int fhoff, nhoff;
568	u8 prevhdr;
569	struct sk_buff *ret_skb = NULL;
570
571	/* Jumbo payload inhibits frag. header */
572	if (ipv6_hdr(skb)->payload_len == 0) {
573		pr_debug("payload len = 0\n");
574		return skb;
575	}
576
577	if (find_prev_fhdr(skb, &prevhdr, &nhoff, &fhoff) < 0)
578		return skb;
579
580	clone = skb_clone(skb, GFP_ATOMIC);
581	if (clone == NULL) {
582		pr_debug("Can't clone skb\n");
583		return skb;
584	}
585
586	NFCT_FRAG6_CB(clone)->orig = skb;
587
588	if (!pskb_may_pull(clone, fhoff + sizeof(*fhdr))) {
589		pr_debug("message is too short.\n");
590		goto ret_orig;
591	}
592
593	skb_set_transport_header(clone, fhoff);
594	hdr = ipv6_hdr(clone);
595	fhdr = (struct frag_hdr *)skb_transport_header(clone);
596
597	fq = fq_find(net, fhdr->identification, user, &hdr->saddr, &hdr->daddr,
598		     ip6_frag_ecn(hdr));
599	if (fq == NULL) {
600		pr_debug("Can't find and can't create new queue\n");
601		goto ret_orig;
602	}
603
604	spin_lock_bh(&fq->q.lock);
605
606	if (nf_ct_frag6_queue(fq, clone, fhdr, nhoff) < 0) {
607		spin_unlock_bh(&fq->q.lock);
608		pr_debug("Can't insert skb to queue\n");
609		inet_frag_put(&fq->q, &nf_frags);
610		goto ret_orig;
611	}
612
613	if (fq->q.last_in == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
614	    fq->q.meat == fq->q.len) {
615		ret_skb = nf_ct_frag6_reasm(fq, dev);
616		if (ret_skb == NULL)
617			pr_debug("Can't reassemble fragmented packets\n");
618	}
619	spin_unlock_bh(&fq->q.lock);
620
621	inet_frag_put(&fq->q, &nf_frags);
622	return ret_skb;
623
624ret_orig:
625	kfree_skb(clone);
626	return skb;
627}
628
629void nf_ct_frag6_consume_orig(struct sk_buff *skb)
630{
631	struct sk_buff *s, *s2;
632
633	for (s = NFCT_FRAG6_CB(skb)->orig; s;) {
634		s2 = s->next;
635		s->next = NULL;
636		consume_skb(s);
637		s = s2;
638	}
639}
640
641static int nf_ct_net_init(struct net *net)
642{
643	net->nf_frag.frags.high_thresh = IPV6_FRAG_HIGH_THRESH;
644	net->nf_frag.frags.low_thresh = IPV6_FRAG_LOW_THRESH;
645	net->nf_frag.frags.timeout = IPV6_FRAG_TIMEOUT;
646	inet_frags_init_net(&net->nf_frag.frags);
647
648	return nf_ct_frag6_sysctl_register(net);
649}
650
651static void nf_ct_net_exit(struct net *net)
652{
653	nf_ct_frags6_sysctl_unregister(net);
654	inet_frags_exit_net(&net->nf_frag.frags, &nf_frags);
655}
656
657static struct pernet_operations nf_ct_net_ops = {
658	.init = nf_ct_net_init,
659	.exit = nf_ct_net_exit,
660};
661
662int nf_ct_frag6_init(void)
663{
664	int ret = 0;
665
666	nf_frags.hashfn = nf_hashfn;
667	nf_frags.constructor = ip6_frag_init;
668	nf_frags.destructor = NULL;
669	nf_frags.skb_free = nf_skb_free;
670	nf_frags.qsize = sizeof(struct frag_queue);
671	nf_frags.match = ip6_frag_match;
672	nf_frags.frag_expire = nf_ct_frag6_expire;
673	nf_frags.secret_interval = 10 * 60 * HZ;
674	inet_frags_init(&nf_frags);
675
676	ret = register_pernet_subsys(&nf_ct_net_ops);
677	if (ret)
678		inet_frags_fini(&nf_frags);
679
680	return ret;
681}
682
683void nf_ct_frag6_cleanup(void)
684{
685	unregister_pernet_subsys(&nf_ct_net_ops);
686	inet_frags_fini(&nf_frags);
687}
688