2 * L2TPv3 IP encapsulation support
4 * Copyright (c) 2008,2009,2010 Katalix Systems Ltd
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.
12 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
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>
25 #include <net/inet_common.h>
26 #include <net/inet_hashtables.h>
27 #include <net/tcp_states.h>
28 #include <net/protocol.h>
31 #include "l2tp_core.h"
34 /* inet_sock has to be the first member of l2tp_ip_sock */
35 struct inet_sock inet;
41 static DEFINE_RWLOCK(l2tp_ip_lock);
42 static struct hlist_head l2tp_ip_table;
43 static struct hlist_head l2tp_ip_bind_table;
45 static inline struct l2tp_ip_sock *l2tp_ip_sk(const struct sock *sk)
47 return (struct l2tp_ip_sock *)sk;
50 static struct sock *__l2tp_ip_bind_lookup(struct net *net, __be32 laddr, int dif, u32 tunnel_id)
52 struct hlist_node *node;
55 sk_for_each_bound(sk, node, &l2tp_ip_bind_table) {
56 struct inet_sock *inet = inet_sk(sk);
57 struct l2tp_ip_sock *l2tp = l2tp_ip_sk(sk);
62 if ((l2tp->conn_id == tunnel_id) &&
63 net_eq(sock_net(sk), net) &&
64 !(inet->inet_rcv_saddr && inet->inet_rcv_saddr != laddr) &&
65 !(sk->sk_bound_dev_if && sk->sk_bound_dev_if != dif))
74 static inline struct sock *l2tp_ip_bind_lookup(struct net *net, __be32 laddr, int dif, u32 tunnel_id)
76 struct sock *sk = __l2tp_ip_bind_lookup(net, laddr, dif, tunnel_id);
83 /* When processing receive frames, there are two cases to
84 * consider. Data frames consist of a non-zero session-id and an
85 * optional cookie. Control frames consist of a regular L2TP header
86 * preceded by 32-bits of zeros.
88 * L2TPv3 Session Header Over IP
91 * 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
92 * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
94 * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
95 * | Cookie (optional, maximum 64 bits)...
96 * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
98 * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
100 * L2TPv3 Control Message Header Over IP
103 * 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
104 * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
105 * | (32 bits of zeros) |
106 * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
107 * |T|L|x|x|S|x|x|x|x|x|x|x| Ver | Length |
108 * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
109 * | Control Connection ID |
110 * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
112 * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
114 * All control frames are passed to userspace.
116 static int l2tp_ip_recv(struct sk_buff *skb)
118 struct net *net = dev_net(skb->dev);
122 unsigned char *ptr, *optr;
123 struct l2tp_session *session;
124 struct l2tp_tunnel *tunnel = NULL;
127 /* Point to L2TP header */
128 optr = ptr = skb->data;
130 if (!pskb_may_pull(skb, 4))
133 session_id = ntohl(*((__be32 *) ptr));
136 /* RFC3931: L2TP/IP packets have the first 4 bytes containing
137 * the session_id. If it is 0, the packet is a L2TP control
138 * frame and the session_id value can be discarded.
140 if (session_id == 0) {
145 /* Ok, this is a data packet. Lookup the session. */
146 session = l2tp_session_find(net, NULL, session_id);
150 tunnel = session->tunnel;
154 /* Trace packet contents, if enabled */
155 if (tunnel->debug & L2TP_MSG_DATA) {
156 length = min(32u, skb->len);
157 if (!pskb_may_pull(skb, length))
160 pr_debug("%s: ip recv\n", tunnel->name);
161 print_hex_dump_bytes("", DUMP_PREFIX_OFFSET, ptr, length);
164 l2tp_recv_common(session, skb, ptr, optr, 0, skb->len, tunnel->recv_payload_hook);
169 /* Get the tunnel_id from the L2TP header */
170 if (!pskb_may_pull(skb, 12))
173 if ((skb->data[0] & 0xc0) != 0xc0)
176 tunnel_id = ntohl(*(__be32 *) &skb->data[4]);
177 tunnel = l2tp_tunnel_find(net, tunnel_id);
181 struct iphdr *iph = (struct iphdr *) skb_network_header(skb);
183 read_lock_bh(&l2tp_ip_lock);
184 sk = __l2tp_ip_bind_lookup(net, iph->daddr, 0, tunnel_id);
185 read_unlock_bh(&l2tp_ip_lock);
193 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
198 return sk_receive_skb(sk, skb, 1);
208 static int l2tp_ip_open(struct sock *sk)
210 /* Prevent autobind. We don't have ports. */
211 inet_sk(sk)->inet_num = IPPROTO_L2TP;
213 write_lock_bh(&l2tp_ip_lock);
214 sk_add_node(sk, &l2tp_ip_table);
215 write_unlock_bh(&l2tp_ip_lock);
220 static void l2tp_ip_close(struct sock *sk, long timeout)
222 write_lock_bh(&l2tp_ip_lock);
223 hlist_del_init(&sk->sk_bind_node);
224 sk_del_node_init(sk);
225 write_unlock_bh(&l2tp_ip_lock);
226 sk_common_release(sk);
229 static void l2tp_ip_destroy_sock(struct sock *sk)
233 while ((skb = __skb_dequeue_tail(&sk->sk_write_queue)) != NULL)
236 sk_refcnt_debug_dec(sk);
239 static int l2tp_ip_bind(struct sock *sk, struct sockaddr *uaddr, int addr_len)
241 struct inet_sock *inet = inet_sk(sk);
242 struct sockaddr_l2tpip *addr = (struct sockaddr_l2tpip *) uaddr;
243 struct net *net = sock_net(sk);
247 if (!sock_flag(sk, SOCK_ZAPPED))
249 if (addr_len < sizeof(struct sockaddr_l2tpip))
251 if (addr->l2tp_family != AF_INET)
255 read_lock_bh(&l2tp_ip_lock);
256 if (__l2tp_ip_bind_lookup(net, addr->l2tp_addr.s_addr,
257 sk->sk_bound_dev_if, addr->l2tp_conn_id))
260 read_unlock_bh(&l2tp_ip_lock);
263 if (sk->sk_state != TCP_CLOSE || addr_len < sizeof(struct sockaddr_l2tpip))
266 chk_addr_ret = inet_addr_type(net, addr->l2tp_addr.s_addr);
267 ret = -EADDRNOTAVAIL;
268 if (addr->l2tp_addr.s_addr && chk_addr_ret != RTN_LOCAL &&
269 chk_addr_ret != RTN_MULTICAST && chk_addr_ret != RTN_BROADCAST)
272 if (addr->l2tp_addr.s_addr)
273 inet->inet_rcv_saddr = inet->inet_saddr = addr->l2tp_addr.s_addr;
274 if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
275 inet->inet_saddr = 0; /* Use device */
278 l2tp_ip_sk(sk)->conn_id = addr->l2tp_conn_id;
280 write_lock_bh(&l2tp_ip_lock);
281 sk_add_bind_node(sk, &l2tp_ip_bind_table);
282 sk_del_node_init(sk);
283 write_unlock_bh(&l2tp_ip_lock);
285 sock_reset_flag(sk, SOCK_ZAPPED);
293 read_unlock_bh(&l2tp_ip_lock);
298 static int l2tp_ip_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
300 struct sockaddr_l2tpip *lsa = (struct sockaddr_l2tpip *) uaddr;
303 if (sock_flag(sk, SOCK_ZAPPED)) /* Must bind first - autobinding does not work */
306 if (addr_len < sizeof(*lsa))
309 if (ipv4_is_multicast(lsa->l2tp_addr.s_addr))
312 rc = ip4_datagram_connect(sk, uaddr, addr_len);
318 l2tp_ip_sk(sk)->peer_conn_id = lsa->l2tp_conn_id;
320 write_lock_bh(&l2tp_ip_lock);
321 hlist_del_init(&sk->sk_bind_node);
322 sk_add_bind_node(sk, &l2tp_ip_bind_table);
323 write_unlock_bh(&l2tp_ip_lock);
329 static int l2tp_ip_disconnect(struct sock *sk, int flags)
331 if (sock_flag(sk, SOCK_ZAPPED))
334 return udp_disconnect(sk, flags);
337 static int l2tp_ip_getname(struct socket *sock, struct sockaddr *uaddr,
338 int *uaddr_len, int peer)
340 struct sock *sk = sock->sk;
341 struct inet_sock *inet = inet_sk(sk);
342 struct l2tp_ip_sock *lsk = l2tp_ip_sk(sk);
343 struct sockaddr_l2tpip *lsa = (struct sockaddr_l2tpip *)uaddr;
345 memset(lsa, 0, sizeof(*lsa));
346 lsa->l2tp_family = AF_INET;
348 if (!inet->inet_dport)
350 lsa->l2tp_conn_id = lsk->peer_conn_id;
351 lsa->l2tp_addr.s_addr = inet->inet_daddr;
353 __be32 addr = inet->inet_rcv_saddr;
355 addr = inet->inet_saddr;
356 lsa->l2tp_conn_id = lsk->conn_id;
357 lsa->l2tp_addr.s_addr = addr;
359 *uaddr_len = sizeof(*lsa);
363 static int l2tp_ip_backlog_recv(struct sock *sk, struct sk_buff *skb)
367 /* Charge it to the socket, dropping if the queue is full. */
368 rc = sock_queue_rcv_skb(sk, skb);
375 IP_INC_STATS(sock_net(sk), IPSTATS_MIB_INDISCARDS);
380 /* Userspace will call sendmsg() on the tunnel socket to send L2TP
383 static int l2tp_ip_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg, size_t len)
387 struct inet_sock *inet = inet_sk(sk);
388 struct rtable *rt = NULL;
396 if (sock_flag(sk, SOCK_DEAD))
399 /* Get and verify the address. */
401 struct sockaddr_l2tpip *lip = (struct sockaddr_l2tpip *) msg->msg_name;
403 if (msg->msg_namelen < sizeof(*lip))
406 if (lip->l2tp_family != AF_INET) {
408 if (lip->l2tp_family != AF_UNSPEC)
412 daddr = lip->l2tp_addr.s_addr;
415 if (sk->sk_state != TCP_ESTABLISHED)
418 daddr = inet->inet_daddr;
422 /* Allocate a socket buffer */
424 skb = sock_wmalloc(sk, 2 + NET_SKB_PAD + sizeof(struct iphdr) +
425 4 + len, 0, GFP_KERNEL);
429 /* Reserve space for headers, putting IP header on 4-byte boundary. */
430 skb_reserve(skb, 2 + NET_SKB_PAD);
431 skb_reset_network_header(skb);
432 skb_reserve(skb, sizeof(struct iphdr));
433 skb_reset_transport_header(skb);
435 /* Insert 0 session_id */
436 *((__be32 *) skb_put(skb, 4)) = 0;
438 /* Copy user data into skb */
439 rc = memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len);
445 fl4 = &inet->cork.fl.u.ip4;
447 rt = (struct rtable *) __sk_dst_check(sk, 0);
451 const struct ip_options_rcu *inet_opt;
453 inet_opt = rcu_dereference(inet->inet_opt);
455 /* Use correct destination address if we have options. */
456 if (inet_opt && inet_opt->opt.srr)
457 daddr = inet_opt->opt.faddr;
459 /* If this fails, retransmit mechanism of transport layer will
460 * keep trying until route appears or the connection times
463 rt = ip_route_output_ports(sock_net(sk), fl4, sk,
464 daddr, inet->inet_saddr,
465 inet->inet_dport, inet->inet_sport,
466 sk->sk_protocol, RT_CONN_FLAGS(sk),
467 sk->sk_bound_dev_if);
471 sk_setup_caps(sk, &rt->dst);
473 skb_dst_set(skb, &rt->dst);
478 /* We dont need to clone dst here, it is guaranteed to not disappear.
479 * __dev_xmit_skb() might force a refcount if needed.
481 skb_dst_set_noref(skb, &rt->dst);
484 /* Queue the packet to IP for output */
485 rc = ip_queue_xmit(skb, &inet->cork.fl);
498 IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTNOROUTES);
504 static int l2tp_ip_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
505 size_t len, int noblock, int flags, int *addr_len)
507 struct inet_sock *inet = inet_sk(sk);
509 int err = -EOPNOTSUPP;
510 struct sockaddr_in *sin = (struct sockaddr_in *)msg->msg_name;
517 *addr_len = sizeof(*sin);
519 skb = skb_recv_datagram(sk, flags, noblock, &err);
525 msg->msg_flags |= MSG_TRUNC;
529 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
533 sock_recv_timestamp(msg, sk, skb);
535 /* Copy the address. */
537 sin->sin_family = AF_INET;
538 sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
540 memset(&sin->sin_zero, 0, sizeof(sin->sin_zero));
542 if (inet->cmsg_flags)
543 ip_cmsg_recv(msg, skb);
544 if (flags & MSG_TRUNC)
547 skb_free_datagram(sk, skb);
549 return err ? err : copied;
552 static struct proto l2tp_ip_prot = {
554 .owner = THIS_MODULE,
555 .init = l2tp_ip_open,
556 .close = l2tp_ip_close,
557 .bind = l2tp_ip_bind,
558 .connect = l2tp_ip_connect,
559 .disconnect = l2tp_ip_disconnect,
561 .destroy = l2tp_ip_destroy_sock,
562 .setsockopt = ip_setsockopt,
563 .getsockopt = ip_getsockopt,
564 .sendmsg = l2tp_ip_sendmsg,
565 .recvmsg = l2tp_ip_recvmsg,
566 .backlog_rcv = l2tp_ip_backlog_recv,
568 .unhash = inet_unhash,
569 .obj_size = sizeof(struct l2tp_ip_sock),
571 .compat_setsockopt = compat_ip_setsockopt,
572 .compat_getsockopt = compat_ip_getsockopt,
576 static const struct proto_ops l2tp_ip_ops = {
578 .owner = THIS_MODULE,
579 .release = inet_release,
581 .connect = inet_dgram_connect,
582 .socketpair = sock_no_socketpair,
583 .accept = sock_no_accept,
584 .getname = l2tp_ip_getname,
585 .poll = datagram_poll,
587 .listen = sock_no_listen,
588 .shutdown = inet_shutdown,
589 .setsockopt = sock_common_setsockopt,
590 .getsockopt = sock_common_getsockopt,
591 .sendmsg = inet_sendmsg,
592 .recvmsg = sock_common_recvmsg,
593 .mmap = sock_no_mmap,
594 .sendpage = sock_no_sendpage,
596 .compat_setsockopt = compat_sock_common_setsockopt,
597 .compat_getsockopt = compat_sock_common_getsockopt,
601 static struct inet_protosw l2tp_ip_protosw = {
603 .protocol = IPPROTO_L2TP,
604 .prot = &l2tp_ip_prot,
609 static struct net_protocol l2tp_ip_protocol __read_mostly = {
610 .handler = l2tp_ip_recv,
614 static int __init l2tp_ip_init(void)
618 pr_info("L2TP IP encapsulation support (L2TPv3)\n");
620 err = proto_register(&l2tp_ip_prot, 1);
624 err = inet_add_protocol(&l2tp_ip_protocol, IPPROTO_L2TP);
628 inet_register_protosw(&l2tp_ip_protosw);
632 proto_unregister(&l2tp_ip_prot);
637 static void __exit l2tp_ip_exit(void)
639 inet_unregister_protosw(&l2tp_ip_protosw);
640 inet_del_protocol(&l2tp_ip_protocol, IPPROTO_L2TP);
641 proto_unregister(&l2tp_ip_prot);
644 module_init(l2tp_ip_init);
645 module_exit(l2tp_ip_exit);
647 MODULE_LICENSE("GPL");
648 MODULE_AUTHOR("James Chapman <jchapman@katalix.com>");
649 MODULE_DESCRIPTION("L2TP over IP");
650 MODULE_VERSION("1.0");
652 /* Use the value of SOCK_DGRAM (2) directory, because __stringify doesn't like
655 MODULE_ALIAS_NET_PF_PROTO_TYPE(PF_INET, 2, IPPROTO_L2TP);