1/*
2 * L2TPv3 IP encapsulation support
3 *
4 * Copyright (c) 2008,2009,2010 Katalix Systems Ltd
5 *
6 *	This program is free software; you can redistribute it and/or
7 *	modify it under the terms of the GNU General Public License
8 *	as published by the Free Software Foundation; either version
9 *	2 of the License, or (at your option) any later version.
10 */
11
12#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
13
14#include <linux/icmp.h>
15#include <linux/module.h>
16#include <linux/skbuff.h>
17#include <linux/random.h>
18#include <linux/socket.h>
19#include <linux/l2tp.h>
20#include <linux/in.h>
21#include <net/sock.h>
22#include <net/ip.h>
23#include <net/icmp.h>
24#include <net/udp.h>
25#include <net/inet_common.h>
26#include <net/inet_hashtables.h>
27#include <net/tcp_states.h>
28#include <net/protocol.h>
29#include <net/xfrm.h>
30
31#include "l2tp_core.h"
32
33struct l2tp_ip_sock {
34	/* inet_sock has to be the first member of l2tp_ip_sock */
35	struct inet_sock	inet;
36
37	u32			conn_id;
38	u32			peer_conn_id;
39};
40
41static DEFINE_RWLOCK(l2tp_ip_lock);
42static struct hlist_head l2tp_ip_table;
43static struct hlist_head l2tp_ip_bind_table;
44
45static inline struct l2tp_ip_sock *l2tp_ip_sk(const struct sock *sk)
46{
47	return (struct l2tp_ip_sock *)sk;
48}
49
50static struct sock *__l2tp_ip_bind_lookup(struct net *net, __be32 laddr, int dif, u32 tunnel_id)
51{
52	struct sock *sk;
53
54	sk_for_each_bound(sk, &l2tp_ip_bind_table) {
55		struct inet_sock *inet = inet_sk(sk);
56		struct l2tp_ip_sock *l2tp = l2tp_ip_sk(sk);
57
58		if (l2tp == NULL)
59			continue;
60
61		if ((l2tp->conn_id == tunnel_id) &&
62		    net_eq(sock_net(sk), net) &&
63		    !(inet->inet_rcv_saddr && inet->inet_rcv_saddr != laddr) &&
64		    !(sk->sk_bound_dev_if && sk->sk_bound_dev_if != dif))
65			goto found;
66	}
67
68	sk = NULL;
69found:
70	return sk;
71}
72
73static inline struct sock *l2tp_ip_bind_lookup(struct net *net, __be32 laddr, int dif, u32 tunnel_id)
74{
75	struct sock *sk = __l2tp_ip_bind_lookup(net, laddr, dif, tunnel_id);
76	if (sk)
77		sock_hold(sk);
78
79	return sk;
80}
81
82/* When processing receive frames, there are two cases to
83 * consider. Data frames consist of a non-zero session-id and an
84 * optional cookie. Control frames consist of a regular L2TP header
85 * preceded by 32-bits of zeros.
86 *
87 * L2TPv3 Session Header Over IP
88 *
89 *  0                   1                   2                   3
90 *  0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
91 * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
92 * |                           Session ID                          |
93 * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
94 * |               Cookie (optional, maximum 64 bits)...
95 * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
96 *                                                                 |
97 * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
98 *
99 * L2TPv3 Control Message Header Over IP
100 *
101 *  0                   1                   2                   3
102 *  0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
103 * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
104 * |                      (32 bits of zeros)                       |
105 * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
106 * |T|L|x|x|S|x|x|x|x|x|x|x|  Ver  |             Length            |
107 * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
108 * |                     Control Connection ID                     |
109 * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
110 * |               Ns              |               Nr              |
111 * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
112 *
113 * All control frames are passed to userspace.
114 */
115static int l2tp_ip_recv(struct sk_buff *skb)
116{
117	struct net *net = dev_net(skb->dev);
118	struct sock *sk;
119	u32 session_id;
120	u32 tunnel_id;
121	unsigned char *ptr, *optr;
122	struct l2tp_session *session;
123	struct l2tp_tunnel *tunnel = NULL;
124	int length;
125
126	if (!pskb_may_pull(skb, 4))
127		goto discard;
128
129	/* Point to L2TP header */
130	optr = ptr = skb->data;
131	session_id = ntohl(*((__be32 *) ptr));
132	ptr += 4;
133
134	/* RFC3931: L2TP/IP packets have the first 4 bytes containing
135	 * the session_id. If it is 0, the packet is a L2TP control
136	 * frame and the session_id value can be discarded.
137	 */
138	if (session_id == 0) {
139		__skb_pull(skb, 4);
140		goto pass_up;
141	}
142
143	/* Ok, this is a data packet. Lookup the session. */
144	session = l2tp_session_find(net, NULL, session_id);
145	if (session == NULL)
146		goto discard;
147
148	tunnel = session->tunnel;
149	if (tunnel == NULL)
150		goto discard;
151
152	/* Trace packet contents, if enabled */
153	if (tunnel->debug & L2TP_MSG_DATA) {
154		length = min(32u, skb->len);
155		if (!pskb_may_pull(skb, length))
156			goto discard;
157
158		/* Point to L2TP header */
159		optr = ptr = skb->data;
160		ptr += 4;
161		pr_debug("%s: ip recv\n", tunnel->name);
162		print_hex_dump_bytes("", DUMP_PREFIX_OFFSET, ptr, length);
163	}
164
165	l2tp_recv_common(session, skb, ptr, optr, 0, skb->len, tunnel->recv_payload_hook);
166
167	return 0;
168
169pass_up:
170	/* Get the tunnel_id from the L2TP header */
171	if (!pskb_may_pull(skb, 12))
172		goto discard;
173
174	if ((skb->data[0] & 0xc0) != 0xc0)
175		goto discard;
176
177	tunnel_id = ntohl(*(__be32 *) &skb->data[4]);
178	tunnel = l2tp_tunnel_find(net, tunnel_id);
179	if (tunnel != NULL)
180		sk = tunnel->sock;
181	else {
182		struct iphdr *iph = (struct iphdr *) skb_network_header(skb);
183
184		read_lock_bh(&l2tp_ip_lock);
185		sk = __l2tp_ip_bind_lookup(net, iph->daddr, 0, tunnel_id);
186		read_unlock_bh(&l2tp_ip_lock);
187	}
188
189	if (sk == NULL)
190		goto discard;
191
192	sock_hold(sk);
193
194	if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
195		goto discard_put;
196
197	nf_reset(skb);
198
199	return sk_receive_skb(sk, skb, 1);
200
201discard_put:
202	sock_put(sk);
203
204discard:
205	kfree_skb(skb);
206	return 0;
207}
208
209static int l2tp_ip_open(struct sock *sk)
210{
211	/* Prevent autobind. We don't have ports. */
212	inet_sk(sk)->inet_num = IPPROTO_L2TP;
213
214	write_lock_bh(&l2tp_ip_lock);
215	sk_add_node(sk, &l2tp_ip_table);
216	write_unlock_bh(&l2tp_ip_lock);
217
218	return 0;
219}
220
221static void l2tp_ip_close(struct sock *sk, long timeout)
222{
223	write_lock_bh(&l2tp_ip_lock);
224	hlist_del_init(&sk->sk_bind_node);
225	sk_del_node_init(sk);
226	write_unlock_bh(&l2tp_ip_lock);
227	sk_common_release(sk);
228}
229
230static void l2tp_ip_destroy_sock(struct sock *sk)
231{
232	struct sk_buff *skb;
233	struct l2tp_tunnel *tunnel = l2tp_sock_to_tunnel(sk);
234
235	while ((skb = __skb_dequeue_tail(&sk->sk_write_queue)) != NULL)
236		kfree_skb(skb);
237
238	if (tunnel) {
239		l2tp_tunnel_closeall(tunnel);
240		sock_put(sk);
241	}
242
243	sk_refcnt_debug_dec(sk);
244}
245
246static int l2tp_ip_bind(struct sock *sk, struct sockaddr *uaddr, int addr_len)
247{
248	struct inet_sock *inet = inet_sk(sk);
249	struct sockaddr_l2tpip *addr = (struct sockaddr_l2tpip *) uaddr;
250	struct net *net = sock_net(sk);
251	int ret;
252	int chk_addr_ret;
253
254	if (!sock_flag(sk, SOCK_ZAPPED))
255		return -EINVAL;
256	if (addr_len < sizeof(struct sockaddr_l2tpip))
257		return -EINVAL;
258	if (addr->l2tp_family != AF_INET)
259		return -EINVAL;
260
261	ret = -EADDRINUSE;
262	read_lock_bh(&l2tp_ip_lock);
263	if (__l2tp_ip_bind_lookup(net, addr->l2tp_addr.s_addr,
264				  sk->sk_bound_dev_if, addr->l2tp_conn_id))
265		goto out_in_use;
266
267	read_unlock_bh(&l2tp_ip_lock);
268
269	lock_sock(sk);
270	if (sk->sk_state != TCP_CLOSE || addr_len < sizeof(struct sockaddr_l2tpip))
271		goto out;
272
273	chk_addr_ret = inet_addr_type(net, addr->l2tp_addr.s_addr);
274	ret = -EADDRNOTAVAIL;
275	if (addr->l2tp_addr.s_addr && chk_addr_ret != RTN_LOCAL &&
276	    chk_addr_ret != RTN_MULTICAST && chk_addr_ret != RTN_BROADCAST)
277		goto out;
278
279	if (addr->l2tp_addr.s_addr)
280		inet->inet_rcv_saddr = inet->inet_saddr = addr->l2tp_addr.s_addr;
281	if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
282		inet->inet_saddr = 0;  /* Use device */
283	sk_dst_reset(sk);
284
285	l2tp_ip_sk(sk)->conn_id = addr->l2tp_conn_id;
286
287	write_lock_bh(&l2tp_ip_lock);
288	sk_add_bind_node(sk, &l2tp_ip_bind_table);
289	sk_del_node_init(sk);
290	write_unlock_bh(&l2tp_ip_lock);
291	ret = 0;
292	sock_reset_flag(sk, SOCK_ZAPPED);
293
294out:
295	release_sock(sk);
296
297	return ret;
298
299out_in_use:
300	read_unlock_bh(&l2tp_ip_lock);
301
302	return ret;
303}
304
305static int l2tp_ip_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
306{
307	struct sockaddr_l2tpip *lsa = (struct sockaddr_l2tpip *) uaddr;
308	int rc;
309
310	if (sock_flag(sk, SOCK_ZAPPED)) /* Must bind first - autobinding does not work */
311		return -EINVAL;
312
313	if (addr_len < sizeof(*lsa))
314		return -EINVAL;
315
316	if (ipv4_is_multicast(lsa->l2tp_addr.s_addr))
317		return -EINVAL;
318
319	rc = ip4_datagram_connect(sk, uaddr, addr_len);
320	if (rc < 0)
321		return rc;
322
323	lock_sock(sk);
324
325	l2tp_ip_sk(sk)->peer_conn_id = lsa->l2tp_conn_id;
326
327	write_lock_bh(&l2tp_ip_lock);
328	hlist_del_init(&sk->sk_bind_node);
329	sk_add_bind_node(sk, &l2tp_ip_bind_table);
330	write_unlock_bh(&l2tp_ip_lock);
331
332	release_sock(sk);
333	return rc;
334}
335
336static int l2tp_ip_disconnect(struct sock *sk, int flags)
337{
338	if (sock_flag(sk, SOCK_ZAPPED))
339		return 0;
340
341	return udp_disconnect(sk, flags);
342}
343
344static int l2tp_ip_getname(struct socket *sock, struct sockaddr *uaddr,
345			   int *uaddr_len, int peer)
346{
347	struct sock *sk		= sock->sk;
348	struct inet_sock *inet	= inet_sk(sk);
349	struct l2tp_ip_sock *lsk = l2tp_ip_sk(sk);
350	struct sockaddr_l2tpip *lsa = (struct sockaddr_l2tpip *)uaddr;
351
352	memset(lsa, 0, sizeof(*lsa));
353	lsa->l2tp_family = AF_INET;
354	if (peer) {
355		if (!inet->inet_dport)
356			return -ENOTCONN;
357		lsa->l2tp_conn_id = lsk->peer_conn_id;
358		lsa->l2tp_addr.s_addr = inet->inet_daddr;
359	} else {
360		__be32 addr = inet->inet_rcv_saddr;
361		if (!addr)
362			addr = inet->inet_saddr;
363		lsa->l2tp_conn_id = lsk->conn_id;
364		lsa->l2tp_addr.s_addr = addr;
365	}
366	*uaddr_len = sizeof(*lsa);
367	return 0;
368}
369
370static int l2tp_ip_backlog_recv(struct sock *sk, struct sk_buff *skb)
371{
372	int rc;
373
374	/* Charge it to the socket, dropping if the queue is full. */
375	rc = sock_queue_rcv_skb(sk, skb);
376	if (rc < 0)
377		goto drop;
378
379	return 0;
380
381drop:
382	IP_INC_STATS(sock_net(sk), IPSTATS_MIB_INDISCARDS);
383	kfree_skb(skb);
384	return -1;
385}
386
387/* Userspace will call sendmsg() on the tunnel socket to send L2TP
388 * control frames.
389 */
390static int l2tp_ip_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
391{
392	struct sk_buff *skb;
393	int rc;
394	struct inet_sock *inet = inet_sk(sk);
395	struct rtable *rt = NULL;
396	struct flowi4 *fl4;
397	int connected = 0;
398	__be32 daddr;
399
400	lock_sock(sk);
401
402	rc = -ENOTCONN;
403	if (sock_flag(sk, SOCK_DEAD))
404		goto out;
405
406	/* Get and verify the address. */
407	if (msg->msg_name) {
408		DECLARE_SOCKADDR(struct sockaddr_l2tpip *, lip, msg->msg_name);
409		rc = -EINVAL;
410		if (msg->msg_namelen < sizeof(*lip))
411			goto out;
412
413		if (lip->l2tp_family != AF_INET) {
414			rc = -EAFNOSUPPORT;
415			if (lip->l2tp_family != AF_UNSPEC)
416				goto out;
417		}
418
419		daddr = lip->l2tp_addr.s_addr;
420	} else {
421		rc = -EDESTADDRREQ;
422		if (sk->sk_state != TCP_ESTABLISHED)
423			goto out;
424
425		daddr = inet->inet_daddr;
426		connected = 1;
427	}
428
429	/* Allocate a socket buffer */
430	rc = -ENOMEM;
431	skb = sock_wmalloc(sk, 2 + NET_SKB_PAD + sizeof(struct iphdr) +
432			   4 + len, 0, GFP_KERNEL);
433	if (!skb)
434		goto error;
435
436	/* Reserve space for headers, putting IP header on 4-byte boundary. */
437	skb_reserve(skb, 2 + NET_SKB_PAD);
438	skb_reset_network_header(skb);
439	skb_reserve(skb, sizeof(struct iphdr));
440	skb_reset_transport_header(skb);
441
442	/* Insert 0 session_id */
443	*((__be32 *) skb_put(skb, 4)) = 0;
444
445	/* Copy user data into skb */
446	rc = memcpy_from_msg(skb_put(skb, len), msg, len);
447	if (rc < 0) {
448		kfree_skb(skb);
449		goto error;
450	}
451
452	fl4 = &inet->cork.fl.u.ip4;
453	if (connected)
454		rt = (struct rtable *) __sk_dst_check(sk, 0);
455
456	rcu_read_lock();
457	if (rt == NULL) {
458		const struct ip_options_rcu *inet_opt;
459
460		inet_opt = rcu_dereference(inet->inet_opt);
461
462		/* Use correct destination address if we have options. */
463		if (inet_opt && inet_opt->opt.srr)
464			daddr = inet_opt->opt.faddr;
465
466		/* If this fails, retransmit mechanism of transport layer will
467		 * keep trying until route appears or the connection times
468		 * itself out.
469		 */
470		rt = ip_route_output_ports(sock_net(sk), fl4, sk,
471					   daddr, inet->inet_saddr,
472					   inet->inet_dport, inet->inet_sport,
473					   sk->sk_protocol, RT_CONN_FLAGS(sk),
474					   sk->sk_bound_dev_if);
475		if (IS_ERR(rt))
476			goto no_route;
477		if (connected) {
478			sk_setup_caps(sk, &rt->dst);
479		} else {
480			skb_dst_set(skb, &rt->dst);
481			goto xmit;
482		}
483	}
484
485	/* We dont need to clone dst here, it is guaranteed to not disappear.
486	 *  __dev_xmit_skb() might force a refcount if needed.
487	 */
488	skb_dst_set_noref(skb, &rt->dst);
489
490xmit:
491	/* Queue the packet to IP for output */
492	rc = ip_queue_xmit(sk, skb, &inet->cork.fl);
493	rcu_read_unlock();
494
495error:
496	if (rc >= 0)
497		rc = len;
498
499out:
500	release_sock(sk);
501	return rc;
502
503no_route:
504	rcu_read_unlock();
505	IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTNOROUTES);
506	kfree_skb(skb);
507	rc = -EHOSTUNREACH;
508	goto out;
509}
510
511static int l2tp_ip_recvmsg(struct sock *sk, struct msghdr *msg,
512			   size_t len, int noblock, int flags, int *addr_len)
513{
514	struct inet_sock *inet = inet_sk(sk);
515	size_t copied = 0;
516	int err = -EOPNOTSUPP;
517	DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name);
518	struct sk_buff *skb;
519
520	if (flags & MSG_OOB)
521		goto out;
522
523	skb = skb_recv_datagram(sk, flags, noblock, &err);
524	if (!skb)
525		goto out;
526
527	copied = skb->len;
528	if (len < copied) {
529		msg->msg_flags |= MSG_TRUNC;
530		copied = len;
531	}
532
533	err = skb_copy_datagram_msg(skb, 0, msg, copied);
534	if (err)
535		goto done;
536
537	sock_recv_timestamp(msg, sk, skb);
538
539	/* Copy the address. */
540	if (sin) {
541		sin->sin_family = AF_INET;
542		sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
543		sin->sin_port = 0;
544		memset(&sin->sin_zero, 0, sizeof(sin->sin_zero));
545		*addr_len = sizeof(*sin);
546	}
547	if (inet->cmsg_flags)
548		ip_cmsg_recv(msg, skb);
549	if (flags & MSG_TRUNC)
550		copied = skb->len;
551done:
552	skb_free_datagram(sk, skb);
553out:
554	return err ? err : copied;
555}
556
557static struct proto l2tp_ip_prot = {
558	.name		   = "L2TP/IP",
559	.owner		   = THIS_MODULE,
560	.init		   = l2tp_ip_open,
561	.close		   = l2tp_ip_close,
562	.bind		   = l2tp_ip_bind,
563	.connect	   = l2tp_ip_connect,
564	.disconnect	   = l2tp_ip_disconnect,
565	.ioctl		   = udp_ioctl,
566	.destroy	   = l2tp_ip_destroy_sock,
567	.setsockopt	   = ip_setsockopt,
568	.getsockopt	   = ip_getsockopt,
569	.sendmsg	   = l2tp_ip_sendmsg,
570	.recvmsg	   = l2tp_ip_recvmsg,
571	.backlog_rcv	   = l2tp_ip_backlog_recv,
572	.hash		   = inet_hash,
573	.unhash		   = inet_unhash,
574	.obj_size	   = sizeof(struct l2tp_ip_sock),
575#ifdef CONFIG_COMPAT
576	.compat_setsockopt = compat_ip_setsockopt,
577	.compat_getsockopt = compat_ip_getsockopt,
578#endif
579};
580
581static const struct proto_ops l2tp_ip_ops = {
582	.family		   = PF_INET,
583	.owner		   = THIS_MODULE,
584	.release	   = inet_release,
585	.bind		   = inet_bind,
586	.connect	   = inet_dgram_connect,
587	.socketpair	   = sock_no_socketpair,
588	.accept		   = sock_no_accept,
589	.getname	   = l2tp_ip_getname,
590	.poll		   = datagram_poll,
591	.ioctl		   = inet_ioctl,
592	.listen		   = sock_no_listen,
593	.shutdown	   = inet_shutdown,
594	.setsockopt	   = sock_common_setsockopt,
595	.getsockopt	   = sock_common_getsockopt,
596	.sendmsg	   = inet_sendmsg,
597	.recvmsg	   = sock_common_recvmsg,
598	.mmap		   = sock_no_mmap,
599	.sendpage	   = sock_no_sendpage,
600#ifdef CONFIG_COMPAT
601	.compat_setsockopt = compat_sock_common_setsockopt,
602	.compat_getsockopt = compat_sock_common_getsockopt,
603#endif
604};
605
606static struct inet_protosw l2tp_ip_protosw = {
607	.type		= SOCK_DGRAM,
608	.protocol	= IPPROTO_L2TP,
609	.prot		= &l2tp_ip_prot,
610	.ops		= &l2tp_ip_ops,
611};
612
613static struct net_protocol l2tp_ip_protocol __read_mostly = {
614	.handler	= l2tp_ip_recv,
615	.netns_ok	= 1,
616};
617
618static int __init l2tp_ip_init(void)
619{
620	int err;
621
622	pr_info("L2TP IP encapsulation support (L2TPv3)\n");
623
624	err = proto_register(&l2tp_ip_prot, 1);
625	if (err != 0)
626		goto out;
627
628	err = inet_add_protocol(&l2tp_ip_protocol, IPPROTO_L2TP);
629	if (err)
630		goto out1;
631
632	inet_register_protosw(&l2tp_ip_protosw);
633	return 0;
634
635out1:
636	proto_unregister(&l2tp_ip_prot);
637out:
638	return err;
639}
640
641static void __exit l2tp_ip_exit(void)
642{
643	inet_unregister_protosw(&l2tp_ip_protosw);
644	inet_del_protocol(&l2tp_ip_protocol, IPPROTO_L2TP);
645	proto_unregister(&l2tp_ip_prot);
646}
647
648module_init(l2tp_ip_init);
649module_exit(l2tp_ip_exit);
650
651MODULE_LICENSE("GPL");
652MODULE_AUTHOR("James Chapman <jchapman@katalix.com>");
653MODULE_DESCRIPTION("L2TP over IP");
654MODULE_VERSION("1.0");
655
656/* Use the value of SOCK_DGRAM (2) directory, because __stringify doesn't like
657 * enums
658 */
659MODULE_ALIAS_NET_PF_PROTO_TYPE(PF_INET, 2, IPPROTO_L2TP);
660MODULE_ALIAS_NET_PF_PROTO(PF_INET, IPPROTO_L2TP);
661