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
53static const char nf_frags_cache_name[] = "nf-frags";
54
55struct nf_ct_frag6_skb_cb
56{
57	struct inet6_skb_parm	h;
58	int			offset;
59	struct sk_buff		*orig;
60};
61
62#define NFCT_FRAG6_CB(skb)	((struct nf_ct_frag6_skb_cb*)((skb)->cb))
63
64static struct inet_frags nf_frags;
65
66#ifdef CONFIG_SYSCTL
67static int zero;
68
69static struct ctl_table nf_ct_frag6_sysctl_table[] = {
70	{
71		.procname	= "nf_conntrack_frag6_timeout",
72		.data		= &init_net.nf_frag.frags.timeout,
73		.maxlen		= sizeof(unsigned int),
74		.mode		= 0644,
75		.proc_handler	= proc_dointvec_jiffies,
76	},
77	{
78		.procname	= "nf_conntrack_frag6_low_thresh",
79		.data		= &init_net.nf_frag.frags.low_thresh,
80		.maxlen		= sizeof(unsigned int),
81		.mode		= 0644,
82		.proc_handler	= proc_dointvec_minmax,
83		.extra1		= &zero,
84		.extra2		= &init_net.nf_frag.frags.high_thresh
85	},
86	{
87		.procname	= "nf_conntrack_frag6_high_thresh",
88		.data		= &init_net.nf_frag.frags.high_thresh,
89		.maxlen		= sizeof(unsigned int),
90		.mode		= 0644,
91		.proc_handler	= proc_dointvec_minmax,
92		.extra1		= &init_net.nf_frag.frags.low_thresh
93	},
94	{ }
95};
96
97static int nf_ct_frag6_sysctl_register(struct net *net)
98{
99	struct ctl_table *table;
100	struct ctl_table_header *hdr;
101
102	table = nf_ct_frag6_sysctl_table;
103	if (!net_eq(net, &init_net)) {
104		table = kmemdup(table, sizeof(nf_ct_frag6_sysctl_table),
105				GFP_KERNEL);
106		if (table == NULL)
107			goto err_alloc;
108
109		table[0].data = &net->nf_frag.frags.timeout;
110		table[1].data = &net->nf_frag.frags.low_thresh;
111		table[1].extra2 = &net->nf_frag.frags.high_thresh;
112		table[2].data = &net->nf_frag.frags.high_thresh;
113		table[2].extra1 = &net->nf_frag.frags.low_thresh;
114		table[2].extra2 = &init_net.nf_frag.frags.high_thresh;
115	}
116
117	hdr = register_net_sysctl(net, "net/netfilter", table);
118	if (hdr == NULL)
119		goto err_reg;
120
121	net->nf_frag.sysctl.frags_hdr = hdr;
122	return 0;
123
124err_reg:
125	if (!net_eq(net, &init_net))
126		kfree(table);
127err_alloc:
128	return -ENOMEM;
129}
130
131static void __net_exit nf_ct_frags6_sysctl_unregister(struct net *net)
132{
133	struct ctl_table *table;
134
135	table = net->nf_frag.sysctl.frags_hdr->ctl_table_arg;
136	unregister_net_sysctl_table(net->nf_frag.sysctl.frags_hdr);
137	if (!net_eq(net, &init_net))
138		kfree(table);
139}
140
141#else
142static int nf_ct_frag6_sysctl_register(struct net *net)
143{
144	return 0;
145}
146static void __net_exit nf_ct_frags6_sysctl_unregister(struct net *net)
147{
148}
149#endif
150
151static inline u8 ip6_frag_ecn(const struct ipv6hdr *ipv6h)
152{
153	return 1 << (ipv6_get_dsfield(ipv6h) & INET_ECN_MASK);
154}
155
156static unsigned int nf_hash_frag(__be32 id, const struct in6_addr *saddr,
157				 const struct in6_addr *daddr)
158{
159	net_get_random_once(&nf_frags.rnd, sizeof(nf_frags.rnd));
160	return jhash_3words(ipv6_addr_hash(saddr), ipv6_addr_hash(daddr),
161			    (__force u32)id, nf_frags.rnd);
162}
163
164
165static unsigned int nf_hashfn(const struct inet_frag_queue *q)
166{
167	const struct frag_queue *nq;
168
169	nq = container_of(q, struct frag_queue, q);
170	return nf_hash_frag(nq->id, &nq->saddr, &nq->daddr);
171}
172
173static void nf_skb_free(struct sk_buff *skb)
174{
175	if (NFCT_FRAG6_CB(skb)->orig)
176		kfree_skb(NFCT_FRAG6_CB(skb)->orig);
177}
178
179static void nf_ct_frag6_expire(unsigned long data)
180{
181	struct frag_queue *fq;
182	struct net *net;
183
184	fq = container_of((struct inet_frag_queue *)data, struct frag_queue, q);
185	net = container_of(fq->q.net, struct net, nf_frag.frags);
186
187	ip6_expire_frag_queue(net, fq, &nf_frags);
188}
189
190/* Creation primitives. */
191static inline struct frag_queue *fq_find(struct net *net, __be32 id,
192					 u32 user, struct in6_addr *src,
193					 struct in6_addr *dst, u8 ecn)
194{
195	struct inet_frag_queue *q;
196	struct ip6_create_arg arg;
197	unsigned int hash;
198
199	arg.id = id;
200	arg.user = user;
201	arg.src = src;
202	arg.dst = dst;
203	arg.ecn = ecn;
204
205	local_bh_disable();
206	hash = nf_hash_frag(id, src, dst);
207
208	q = inet_frag_find(&net->nf_frag.frags, &nf_frags, &arg, hash);
209	local_bh_enable();
210	if (IS_ERR_OR_NULL(q)) {
211		inet_frag_maybe_warn_overflow(q, pr_fmt());
212		return NULL;
213	}
214	return container_of(q, struct frag_queue, q);
215}
216
217
218static int nf_ct_frag6_queue(struct frag_queue *fq, struct sk_buff *skb,
219			     const struct frag_hdr *fhdr, int nhoff)
220{
221	struct sk_buff *prev, *next;
222	unsigned int payload_len;
223	int offset, end;
224	u8 ecn;
225
226	if (fq->q.flags & INET_FRAG_COMPLETE) {
227		pr_debug("Already completed\n");
228		goto err;
229	}
230
231	payload_len = ntohs(ipv6_hdr(skb)->payload_len);
232
233	offset = ntohs(fhdr->frag_off) & ~0x7;
234	end = offset + (payload_len -
235			((u8 *)(fhdr + 1) - (u8 *)(ipv6_hdr(skb) + 1)));
236
237	if ((unsigned int)end > IPV6_MAXPLEN) {
238		pr_debug("offset is too large.\n");
239		return -1;
240	}
241
242	ecn = ip6_frag_ecn(ipv6_hdr(skb));
243
244	if (skb->ip_summed == CHECKSUM_COMPLETE) {
245		const unsigned char *nh = skb_network_header(skb);
246		skb->csum = csum_sub(skb->csum,
247				     csum_partial(nh, (u8 *)(fhdr + 1) - nh,
248						  0));
249	}
250
251	/* Is this the final fragment? */
252	if (!(fhdr->frag_off & htons(IP6_MF))) {
253		/* If we already have some bits beyond end
254		 * or have different end, the segment is corrupted.
255		 */
256		if (end < fq->q.len ||
257		    ((fq->q.flags & INET_FRAG_LAST_IN) && end != fq->q.len)) {
258			pr_debug("already received last fragment\n");
259			goto err;
260		}
261		fq->q.flags |= INET_FRAG_LAST_IN;
262		fq->q.len = end;
263	} else {
264		/* Check if the fragment is rounded to 8 bytes.
265		 * Required by the RFC.
266		 */
267		if (end & 0x7) {
268			/* RFC2460 says always send parameter problem in
269			 * this case. -DaveM
270			 */
271			pr_debug("end of fragment not rounded to 8 bytes.\n");
272			return -1;
273		}
274		if (end > fq->q.len) {
275			/* Some bits beyond end -> corruption. */
276			if (fq->q.flags & INET_FRAG_LAST_IN) {
277				pr_debug("last packet already reached.\n");
278				goto err;
279			}
280			fq->q.len = end;
281		}
282	}
283
284	if (end == offset)
285		goto err;
286
287	/* Point into the IP datagram 'data' part. */
288	if (!pskb_pull(skb, (u8 *) (fhdr + 1) - skb->data)) {
289		pr_debug("queue: message is too short.\n");
290		goto err;
291	}
292	if (pskb_trim_rcsum(skb, end - offset)) {
293		pr_debug("Can't trim\n");
294		goto err;
295	}
296
297	/* Find out which fragments are in front and at the back of us
298	 * in the chain of fragments so far.  We must know where to put
299	 * this fragment, right?
300	 */
301	prev = fq->q.fragments_tail;
302	if (!prev || NFCT_FRAG6_CB(prev)->offset < offset) {
303		next = NULL;
304		goto found;
305	}
306	prev = NULL;
307	for (next = fq->q.fragments; next != NULL; next = next->next) {
308		if (NFCT_FRAG6_CB(next)->offset >= offset)
309			break;	/* bingo! */
310		prev = next;
311	}
312
313found:
314	/* RFC5722, Section 4:
315	 *                                  When reassembling an IPv6 datagram, if
316	 *   one or more its constituent fragments is determined to be an
317	 *   overlapping fragment, the entire datagram (and any constituent
318	 *   fragments, including those not yet received) MUST be silently
319	 *   discarded.
320	 */
321
322	/* Check for overlap with preceding fragment. */
323	if (prev &&
324	    (NFCT_FRAG6_CB(prev)->offset + prev->len) > offset)
325		goto discard_fq;
326
327	/* Look for overlap with succeeding segment. */
328	if (next && NFCT_FRAG6_CB(next)->offset < end)
329		goto discard_fq;
330
331	NFCT_FRAG6_CB(skb)->offset = offset;
332
333	/* Insert this fragment in the chain of fragments. */
334	skb->next = next;
335	if (!next)
336		fq->q.fragments_tail = skb;
337	if (prev)
338		prev->next = skb;
339	else
340		fq->q.fragments = skb;
341
342	if (skb->dev) {
343		fq->iif = skb->dev->ifindex;
344		skb->dev = NULL;
345	}
346	fq->q.stamp = skb->tstamp;
347	fq->q.meat += skb->len;
348	fq->ecn |= ecn;
349	if (payload_len > fq->q.max_size)
350		fq->q.max_size = payload_len;
351	add_frag_mem_limit(&fq->q, skb->truesize);
352
353	/* The first fragment.
354	 * nhoffset is obtained from the first fragment, of course.
355	 */
356	if (offset == 0) {
357		fq->nhoffset = nhoff;
358		fq->q.flags |= INET_FRAG_FIRST_IN;
359	}
360
361	return 0;
362
363discard_fq:
364	inet_frag_kill(&fq->q, &nf_frags);
365err:
366	return -1;
367}
368
369/*
370 *	Check if this packet is complete.
371 *	Returns NULL on failure by any reason, and pointer
372 *	to current nexthdr field in reassembled frame.
373 *
374 *	It is called with locked fq, and caller must check that
375 *	queue is eligible for reassembly i.e. it is not COMPLETE,
376 *	the last and the first frames arrived and all the bits are here.
377 */
378static struct sk_buff *
379nf_ct_frag6_reasm(struct frag_queue *fq, struct net_device *dev)
380{
381	struct sk_buff *fp, *op, *head = fq->q.fragments;
382	int    payload_len;
383	u8 ecn;
384
385	inet_frag_kill(&fq->q, &nf_frags);
386
387	WARN_ON(head == NULL);
388	WARN_ON(NFCT_FRAG6_CB(head)->offset != 0);
389
390	ecn = ip_frag_ecn_table[fq->ecn];
391	if (unlikely(ecn == 0xff))
392		goto out_fail;
393
394	/* Unfragmented part is taken from the first segment. */
395	payload_len = ((head->data - skb_network_header(head)) -
396		       sizeof(struct ipv6hdr) + fq->q.len -
397		       sizeof(struct frag_hdr));
398	if (payload_len > IPV6_MAXPLEN) {
399		pr_debug("payload len is too large.\n");
400		goto out_oversize;
401	}
402
403	/* Head of list must not be cloned. */
404	if (skb_unclone(head, GFP_ATOMIC)) {
405		pr_debug("skb is cloned but can't expand head");
406		goto out_oom;
407	}
408
409	/* If the first fragment is fragmented itself, we split
410	 * it to two chunks: the first with data and paged part
411	 * and the second, holding only fragments. */
412	if (skb_has_frag_list(head)) {
413		struct sk_buff *clone;
414		int i, plen = 0;
415
416		clone = alloc_skb(0, GFP_ATOMIC);
417		if (clone == NULL)
418			goto out_oom;
419
420		clone->next = head->next;
421		head->next = clone;
422		skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
423		skb_frag_list_init(head);
424		for (i = 0; i < skb_shinfo(head)->nr_frags; i++)
425			plen += skb_frag_size(&skb_shinfo(head)->frags[i]);
426		clone->len = clone->data_len = head->data_len - plen;
427		head->data_len -= clone->len;
428		head->len -= clone->len;
429		clone->csum = 0;
430		clone->ip_summed = head->ip_summed;
431
432		NFCT_FRAG6_CB(clone)->orig = NULL;
433		add_frag_mem_limit(&fq->q, clone->truesize);
434	}
435
436	/* We have to remove fragment header from datagram and to relocate
437	 * header in order to calculate ICV correctly. */
438	skb_network_header(head)[fq->nhoffset] = skb_transport_header(head)[0];
439	memmove(head->head + sizeof(struct frag_hdr), head->head,
440		(head->data - head->head) - sizeof(struct frag_hdr));
441	head->mac_header += sizeof(struct frag_hdr);
442	head->network_header += sizeof(struct frag_hdr);
443
444	skb_shinfo(head)->frag_list = head->next;
445	skb_reset_transport_header(head);
446	skb_push(head, head->data - skb_network_header(head));
447
448	for (fp=head->next; fp; fp = fp->next) {
449		head->data_len += fp->len;
450		head->len += fp->len;
451		if (head->ip_summed != fp->ip_summed)
452			head->ip_summed = CHECKSUM_NONE;
453		else if (head->ip_summed == CHECKSUM_COMPLETE)
454			head->csum = csum_add(head->csum, fp->csum);
455		head->truesize += fp->truesize;
456	}
457	sub_frag_mem_limit(&fq->q, head->truesize);
458
459	head->ignore_df = 1;
460	head->next = NULL;
461	head->dev = dev;
462	head->tstamp = fq->q.stamp;
463	ipv6_hdr(head)->payload_len = htons(payload_len);
464	ipv6_change_dsfield(ipv6_hdr(head), 0xff, ecn);
465	IP6CB(head)->frag_max_size = sizeof(struct ipv6hdr) + fq->q.max_size;
466
467	/* Yes, and fold redundant checksum back. 8) */
468	if (head->ip_summed == CHECKSUM_COMPLETE)
469		head->csum = csum_partial(skb_network_header(head),
470					  skb_network_header_len(head),
471					  head->csum);
472
473	fq->q.fragments = NULL;
474	fq->q.fragments_tail = NULL;
475
476	/* all original skbs are linked into the NFCT_FRAG6_CB(head).orig */
477	fp = skb_shinfo(head)->frag_list;
478	if (fp && NFCT_FRAG6_CB(fp)->orig == NULL)
479		/* at above code, head skb is divided into two skbs. */
480		fp = fp->next;
481
482	op = NFCT_FRAG6_CB(head)->orig;
483	for (; fp; fp = fp->next) {
484		struct sk_buff *orig = NFCT_FRAG6_CB(fp)->orig;
485
486		op->next = orig;
487		op = orig;
488		NFCT_FRAG6_CB(fp)->orig = NULL;
489	}
490
491	return head;
492
493out_oversize:
494	net_dbg_ratelimited("nf_ct_frag6_reasm: payload len = %d\n",
495			    payload_len);
496	goto out_fail;
497out_oom:
498	net_dbg_ratelimited("nf_ct_frag6_reasm: no memory for reassembly\n");
499out_fail:
500	return NULL;
501}
502
503/*
504 * find the header just before Fragment Header.
505 *
506 * if success return 0 and set ...
507 * (*prevhdrp): the value of "Next Header Field" in the header
508 *		just before Fragment Header.
509 * (*prevhoff): the offset of "Next Header Field" in the header
510 *		just before Fragment Header.
511 * (*fhoff)   : the offset of Fragment Header.
512 *
513 * Based on ipv6_skip_hdr() in net/ipv6/exthdr.c
514 *
515 */
516static int
517find_prev_fhdr(struct sk_buff *skb, u8 *prevhdrp, int *prevhoff, int *fhoff)
518{
519	u8 nexthdr = ipv6_hdr(skb)->nexthdr;
520	const int netoff = skb_network_offset(skb);
521	u8 prev_nhoff = netoff + offsetof(struct ipv6hdr, nexthdr);
522	int start = netoff + sizeof(struct ipv6hdr);
523	int len = skb->len - start;
524	u8 prevhdr = NEXTHDR_IPV6;
525
526	while (nexthdr != NEXTHDR_FRAGMENT) {
527		struct ipv6_opt_hdr hdr;
528		int hdrlen;
529
530		if (!ipv6_ext_hdr(nexthdr)) {
531			return -1;
532		}
533		if (nexthdr == NEXTHDR_NONE) {
534			pr_debug("next header is none\n");
535			return -1;
536		}
537		if (len < (int)sizeof(struct ipv6_opt_hdr)) {
538			pr_debug("too short\n");
539			return -1;
540		}
541		if (skb_copy_bits(skb, start, &hdr, sizeof(hdr)))
542			BUG();
543		if (nexthdr == NEXTHDR_AUTH)
544			hdrlen = (hdr.hdrlen+2)<<2;
545		else
546			hdrlen = ipv6_optlen(&hdr);
547
548		prevhdr = nexthdr;
549		prev_nhoff = start;
550
551		nexthdr = hdr.nexthdr;
552		len -= hdrlen;
553		start += hdrlen;
554	}
555
556	if (len < 0)
557		return -1;
558
559	*prevhdrp = prevhdr;
560	*prevhoff = prev_nhoff;
561	*fhoff = start;
562
563	return 0;
564}
565
566struct sk_buff *nf_ct_frag6_gather(struct sk_buff *skb, u32 user)
567{
568	struct sk_buff *clone;
569	struct net_device *dev = skb->dev;
570	struct net *net = skb_dst(skb) ? dev_net(skb_dst(skb)->dev)
571				       : dev_net(skb->dev);
572	struct frag_hdr *fhdr;
573	struct frag_queue *fq;
574	struct ipv6hdr *hdr;
575	int fhoff, nhoff;
576	u8 prevhdr;
577	struct sk_buff *ret_skb = NULL;
578
579	/* Jumbo payload inhibits frag. header */
580	if (ipv6_hdr(skb)->payload_len == 0) {
581		pr_debug("payload len = 0\n");
582		return skb;
583	}
584
585	if (find_prev_fhdr(skb, &prevhdr, &nhoff, &fhoff) < 0)
586		return skb;
587
588	clone = skb_clone(skb, GFP_ATOMIC);
589	if (clone == NULL) {
590		pr_debug("Can't clone skb\n");
591		return skb;
592	}
593
594	NFCT_FRAG6_CB(clone)->orig = skb;
595
596	if (!pskb_may_pull(clone, fhoff + sizeof(*fhdr))) {
597		pr_debug("message is too short.\n");
598		goto ret_orig;
599	}
600
601	skb_set_transport_header(clone, fhoff);
602	hdr = ipv6_hdr(clone);
603	fhdr = (struct frag_hdr *)skb_transport_header(clone);
604
605	fq = fq_find(net, fhdr->identification, user, &hdr->saddr, &hdr->daddr,
606		     ip6_frag_ecn(hdr));
607	if (fq == NULL) {
608		pr_debug("Can't find and can't create new queue\n");
609		goto ret_orig;
610	}
611
612	spin_lock_bh(&fq->q.lock);
613
614	if (nf_ct_frag6_queue(fq, clone, fhdr, nhoff) < 0) {
615		spin_unlock_bh(&fq->q.lock);
616		pr_debug("Can't insert skb to queue\n");
617		inet_frag_put(&fq->q, &nf_frags);
618		goto ret_orig;
619	}
620
621	if (fq->q.flags == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
622	    fq->q.meat == fq->q.len) {
623		ret_skb = nf_ct_frag6_reasm(fq, dev);
624		if (ret_skb == NULL)
625			pr_debug("Can't reassemble fragmented packets\n");
626	}
627	spin_unlock_bh(&fq->q.lock);
628
629	inet_frag_put(&fq->q, &nf_frags);
630	return ret_skb;
631
632ret_orig:
633	kfree_skb(clone);
634	return skb;
635}
636
637void nf_ct_frag6_consume_orig(struct sk_buff *skb)
638{
639	struct sk_buff *s, *s2;
640
641	for (s = NFCT_FRAG6_CB(skb)->orig; s;) {
642		s2 = s->next;
643		s->next = NULL;
644		consume_skb(s);
645		s = s2;
646	}
647}
648
649static int nf_ct_net_init(struct net *net)
650{
651	net->nf_frag.frags.high_thresh = IPV6_FRAG_HIGH_THRESH;
652	net->nf_frag.frags.low_thresh = IPV6_FRAG_LOW_THRESH;
653	net->nf_frag.frags.timeout = IPV6_FRAG_TIMEOUT;
654	inet_frags_init_net(&net->nf_frag.frags);
655
656	return nf_ct_frag6_sysctl_register(net);
657}
658
659static void nf_ct_net_exit(struct net *net)
660{
661	nf_ct_frags6_sysctl_unregister(net);
662	inet_frags_exit_net(&net->nf_frag.frags, &nf_frags);
663}
664
665static struct pernet_operations nf_ct_net_ops = {
666	.init = nf_ct_net_init,
667	.exit = nf_ct_net_exit,
668};
669
670int nf_ct_frag6_init(void)
671{
672	int ret = 0;
673
674	nf_frags.hashfn = nf_hashfn;
675	nf_frags.constructor = ip6_frag_init;
676	nf_frags.destructor = NULL;
677	nf_frags.skb_free = nf_skb_free;
678	nf_frags.qsize = sizeof(struct frag_queue);
679	nf_frags.match = ip6_frag_match;
680	nf_frags.frag_expire = nf_ct_frag6_expire;
681	nf_frags.frags_cache_name = nf_frags_cache_name;
682	ret = inet_frags_init(&nf_frags);
683	if (ret)
684		goto out;
685	ret = register_pernet_subsys(&nf_ct_net_ops);
686	if (ret)
687		inet_frags_fini(&nf_frags);
688
689out:
690	return ret;
691}
692
693void nf_ct_frag6_cleanup(void)
694{
695	unregister_pernet_subsys(&nf_ct_net_ops);
696	inet_frags_fini(&nf_frags);
697}
698