l2tp: Add L2TPv3 IP encapsulation (no UDP) support
authorJames Chapman <jchapman@katalix.com>
Fri, 2 Apr 2010 06:19:00 +0000 (06:19 +0000)
committerDavid S. Miller <davem@davemloft.net>
Sat, 3 Apr 2010 21:56:04 +0000 (14:56 -0700)
This patch adds a new L2TPIP socket family and modifies the core to
handle the case where there is no UDP header in the L2TP
packet. L2TP/IP uses IP protocol 115. Since L2TP/UDP and L2TP/IP
packets differ in layout, the datapath packet handling code needs
changes too. Userspace uses an L2TPIP socket instead of a UDP socket
when IP encapsulation is required.

We can't use raw sockets for this because the semantics of raw sockets
don't lend themselves to the socket-per-tunnel model - we need to

Signed-off-by: David S. Miller <davem@davemloft.net>
include/linux/l2tp.h [new file with mode: 0644]
net/l2tp/Kconfig
net/l2tp/Makefile
net/l2tp/l2tp_core.c
net/l2tp/l2tp_core.h
net/l2tp/l2tp_ip.c [new file with mode: 0644]
net/l2tp/l2tp_ppp.c

diff --git a/include/linux/l2tp.h b/include/linux/l2tp.h
new file mode 100644 (file)
index 0000000..deff7bc
--- /dev/null
@@ -0,0 +1,38 @@
+/*
+ * L2TP-over-IP socket for L2TPv3.
+ *
+ * Author: James Chapman <jchapman@katalix.com>
+ */
+
+#ifndef _LINUX_L2TP_H_
+#define _LINUX_L2TP_H_
+
+#include <linux/types.h>
+#ifdef __KERNEL__
+#include <linux/socket.h>
+#include <linux/in.h>
+#endif
+
+#define IPPROTO_L2TP           115
+
+/**
+ * struct sockaddr_l2tpip - the sockaddr structure for L2TP-over-IP sockets
+ * @l2tp_family:  address family number AF_L2TPIP.
+ * @l2tp_addr:    protocol specific address information
+ * @l2tp_conn_id: connection id of tunnel
+ */
+struct sockaddr_l2tpip {
+       /* The first fields must match struct sockaddr_in */
+       sa_family_t     l2tp_family;    /* AF_INET */
+       __be16          l2tp_unused;    /* INET port number (unused) */
+       struct in_addr  l2tp_addr;      /* Internet address */
+
+       __u32           l2tp_conn_id;   /* Connection ID of tunnel */
+
+       /* Pad to size of `struct sockaddr'. */
+       unsigned char   __pad[sizeof(struct sockaddr) - sizeof(sa_family_t) -
+                             sizeof(__be16) - sizeof(struct in_addr) -
+                             sizeof(__u32)];
+};
+
+#endif
index d60758d..0a11ccf 100644 (file)
@@ -51,3 +51,20 @@ config L2TP_V3
          If you are connecting to L2TPv3 equipment, or you want to
          tunnel raw ethernet frames using L2TP, say Y here. If
          unsure, say N.
+
+config L2TP_IP
+       tristate "L2TP IP encapsulation for L2TPv3"
+       depends on L2TP_V3
+       help
+         Support for L2TP-over-IP socket family.
+
+         The L2TPv3 protocol defines two possible encapsulations for
+         L2TP frames, namely UDP and plain IP (without UDP). This
+         driver provides a new L2TPIP socket family with which
+         userspace L2TPv3 daemons may create L2TP/IP tunnel sockets
+         when UDP encapsulation is not required. When L2TP is carried
+         in IP packets, it used IP protocol number 115, so this port
+         must be enabled in firewalls.
+
+         To compile this driver as a module, choose M here. The module
+         will be called l2tp_ip.
index c91f208..ef28b16 100644 (file)
@@ -6,3 +6,4 @@ obj-$(CONFIG_L2TP) += l2tp_core.o
 
 # Build l2tp as modules if L2TP is M
 obj-$(subst y,$(CONFIG_L2TP),$(CONFIG_PPPOL2TP)) += l2tp_ppp.o
+obj-$(subst y,$(CONFIG_L2TP),$(CONFIG_L2TP_IP)) += l2tp_ip.o
index 0eee1a6..1739d04 100644 (file)
 #include <linux/inetdevice.h>
 #include <linux/skbuff.h>
 #include <linux/init.h>
+#include <linux/in.h>
 #include <linux/ip.h>
 #include <linux/udp.h>
+#include <linux/l2tp.h>
 #include <linux/hash.h>
 #include <linux/sort.h>
 #include <linux/file.h>
@@ -48,6 +50,7 @@
 #include <net/ip.h>
 #include <net/udp.h>
 #include <net/xfrm.h>
+#include <net/protocol.h>
 
 #include <asm/byteorder.h>
 #include <asm/atomic.h>
@@ -849,15 +852,21 @@ static int l2tp_build_l2tpv2_header(struct l2tp_session *session, void *buf)
 
 static int l2tp_build_l2tpv3_header(struct l2tp_session *session, void *buf)
 {
+       struct l2tp_tunnel *tunnel = session->tunnel;
        char *bufp = buf;
        char *optr = bufp;
-       u16 flags = L2TP_HDR_VER_3;
 
-       /* Setup L2TP header. */
-       *((__be16 *) bufp) = htons(flags);
-       bufp += 2;
-       *((__be16 *) bufp) = 0;
-       bufp += 2;
+       /* Setup L2TP header. The header differs slightly for UDP and
+        * IP encapsulations. For UDP, there is 4 bytes of flags.
+        */
+       if (tunnel->encap == L2TP_ENCAPTYPE_UDP) {
+               u16 flags = L2TP_HDR_VER_3;
+               *((__be16 *) bufp) = htons(flags);
+               bufp += 2;
+               *((__be16 *) bufp) = 0;
+               bufp += 2;
+       }
+
        *((__be32 *) bufp) = htonl(session->peer_session_id);
        bufp += 4;
        if (session->cookie_len) {
@@ -902,10 +911,11 @@ int l2tp_xmit_core(struct l2tp_session *session, struct sk_buff *skb, size_t dat
 
        if (session->debug & L2TP_MSG_DATA) {
                int i;
-               unsigned char *datap = skb->data + sizeof(struct udphdr);
+               int uhlen = (tunnel->encap == L2TP_ENCAPTYPE_UDP) ? sizeof(struct udphdr) : 0;
+               unsigned char *datap = skb->data + uhlen;
 
                printk(KERN_DEBUG "%s: xmit:", session->name);
-               for (i = 0; i < (len - sizeof(struct udphdr)); i++) {
+               for (i = 0; i < (len - uhlen); i++) {
                        printk(" %02X", *datap++);
                        if (i == 31) {
                                printk(" ...");
@@ -956,21 +966,23 @@ static inline void l2tp_skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
 int l2tp_xmit_skb(struct l2tp_session *session, struct sk_buff *skb, int hdr_len)
 {
        int data_len = skb->len;
-       struct sock *sk = session->tunnel->sock;
+       struct l2tp_tunnel *tunnel = session->tunnel;
+       struct sock *sk = tunnel->sock;
        struct udphdr *uh;
-       unsigned int udp_len;
        struct inet_sock *inet;
        __wsum csum;
        int old_headroom;
        int new_headroom;
        int headroom;
+       int uhlen = (tunnel->encap == L2TP_ENCAPTYPE_UDP) ? sizeof(struct udphdr) : 0;
+       int udp_len;
 
        /* Check that there's enough headroom in the skb to insert IP,
         * UDP and L2TP headers. If not enough, expand it to
         * make room. Adjust truesize.
         */
        headroom = NET_SKB_PAD + sizeof(struct iphdr) +
-               sizeof(struct udphdr) + hdr_len;
+               uhlen + hdr_len;
        old_headroom = skb_headroom(skb);
        if (skb_cow_head(skb, headroom))
                goto abort;
@@ -981,18 +993,8 @@ int l2tp_xmit_skb(struct l2tp_session *session, struct sk_buff *skb, int hdr_len
 
        /* Setup L2TP header */
        session->build_header(session, __skb_push(skb, hdr_len));
-       udp_len = sizeof(struct udphdr) + hdr_len + data_len;
-
-       /* Setup UDP header */
-       inet = inet_sk(sk);
-       __skb_push(skb, sizeof(*uh));
-       skb_reset_transport_header(skb);
-       uh = udp_hdr(skb);
-       uh->source = inet->inet_sport;
-       uh->dest = inet->inet_dport;
-       uh->len = htons(udp_len);
-       uh->check = 0;
 
+       /* Reset skb netfilter state */
        memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
        IPCB(skb)->flags &= ~(IPSKB_XFRM_TUNNEL_SIZE | IPSKB_XFRM_TRANSFORMED |
                              IPSKB_REROUTED);
@@ -1001,29 +1003,48 @@ int l2tp_xmit_skb(struct l2tp_session *session, struct sk_buff *skb, int hdr_len
        /* Get routing info from the tunnel socket */
        skb_dst_drop(skb);
        skb_dst_set(skb, dst_clone(__sk_dst_get(sk)));
-       l2tp_skb_set_owner_w(skb, sk);
 
-       /* Calculate UDP checksum if configured to do so */
-       if (sk->sk_no_check == UDP_CSUM_NOXMIT)
-               skb->ip_summed = CHECKSUM_NONE;
-       else if ((skb_dst(skb) && skb_dst(skb)->dev) &&
-                (!(skb_dst(skb)->dev->features & NETIF_F_V4_CSUM))) {
-               skb->ip_summed = CHECKSUM_COMPLETE;
-               csum = skb_checksum(skb, 0, udp_len, 0);
-               uh->check = csum_tcpudp_magic(inet->inet_saddr,
-                                             inet->inet_daddr,
-                                             udp_len, IPPROTO_UDP, csum);
-               if (uh->check == 0)
-                       uh->check = CSUM_MANGLED_0;
-       } else {
-               skb->ip_summed = CHECKSUM_PARTIAL;
-               skb->csum_start = skb_transport_header(skb) - skb->head;
-               skb->csum_offset = offsetof(struct udphdr, check);
-               uh->check = ~csum_tcpudp_magic(inet->inet_saddr,
-                                              inet->inet_daddr,
-                                              udp_len, IPPROTO_UDP, 0);
+       switch (tunnel->encap) {
+       case L2TP_ENCAPTYPE_UDP:
+               /* Setup UDP header */
+               inet = inet_sk(sk);
+               __skb_push(skb, sizeof(*uh));
+               skb_reset_transport_header(skb);
+               uh = udp_hdr(skb);
+               uh->source = inet->inet_sport;
+               uh->dest = inet->inet_dport;
+               udp_len = uhlen + hdr_len + data_len;
+               uh->len = htons(udp_len);
+               uh->check = 0;
+
+               /* Calculate UDP checksum if configured to do so */
+               if (sk->sk_no_check == UDP_CSUM_NOXMIT)
+                       skb->ip_summed = CHECKSUM_NONE;
+               else if ((skb_dst(skb) && skb_dst(skb)->dev) &&
+                        (!(skb_dst(skb)->dev->features & NETIF_F_V4_CSUM))) {
+                       skb->ip_summed = CHECKSUM_COMPLETE;
+                       csum = skb_checksum(skb, 0, udp_len, 0);
+                       uh->check = csum_tcpudp_magic(inet->inet_saddr,
+                                                     inet->inet_daddr,
+                                                     udp_len, IPPROTO_UDP, csum);
+                       if (uh->check == 0)
+                               uh->check = CSUM_MANGLED_0;
+               } else {
+                       skb->ip_summed = CHECKSUM_PARTIAL;
+                       skb->csum_start = skb_transport_header(skb) - skb->head;
+                       skb->csum_offset = offsetof(struct udphdr, check);
+                       uh->check = ~csum_tcpudp_magic(inet->inet_saddr,
+                                                      inet->inet_daddr,
+                                                      udp_len, IPPROTO_UDP, 0);
+               }
+               break;
+
+       case L2TP_ENCAPTYPE_IP:
+               break;
        }
 
+       l2tp_skb_set_owner_w(skb, sk);
+
        l2tp_xmit_core(session, skb, data_len);
 
 abort:
@@ -1053,9 +1074,15 @@ void l2tp_tunnel_destruct(struct sock *sk)
        /* Close all sessions */
        l2tp_tunnel_closeall(tunnel);
 
-       /* No longer an encapsulation socket. See net/ipv4/udp.c */
-       (udp_sk(sk))->encap_type = 0;
-       (udp_sk(sk))->encap_rcv = NULL;
+       switch (tunnel->encap) {
+       case L2TP_ENCAPTYPE_UDP:
+               /* No longer an encapsulation socket. See net/ipv4/udp.c */
+               (udp_sk(sk))->encap_type = 0;
+               (udp_sk(sk))->encap_rcv = NULL;
+               break;
+       case L2TP_ENCAPTYPE_IP:
+               break;
+       }
 
        /* Remove hooks into tunnel socket */
        tunnel->sock = NULL;
@@ -1168,6 +1195,7 @@ int l2tp_tunnel_create(struct net *net, int fd, int version, u32 tunnel_id, u32
        struct socket *sock = NULL;
        struct sock *sk = NULL;
        struct l2tp_net *pn;
+       enum l2tp_encap_type encap = L2TP_ENCAPTYPE_UDP;
 
        /* Get the tunnel socket from the fd, which was opened by
         * the userspace L2TP daemon.
@@ -1182,18 +1210,27 @@ int l2tp_tunnel_create(struct net *net, int fd, int version, u32 tunnel_id, u32
 
        sk = sock->sk;
 
+       if (cfg != NULL)
+               encap = cfg->encap;
+
        /* Quick sanity checks */
-       err = -EPROTONOSUPPORT;
-       if (sk->sk_protocol != IPPROTO_UDP) {
-               printk(KERN_ERR "tunl %hu: fd %d wrong protocol, got %d, expected %d\n",
-                      tunnel_id, fd, sk->sk_protocol, IPPROTO_UDP);
-               goto err;
-       }
-       err = -EAFNOSUPPORT;
-       if (sock->ops->family != AF_INET) {
-               printk(KERN_ERR "tunl %hu: fd %d wrong family, got %d, expected %d\n",
-                      tunnel_id, fd, sock->ops->family, AF_INET);
-               goto err;
+       switch (encap) {
+       case L2TP_ENCAPTYPE_UDP:
+               err = -EPROTONOSUPPORT;
+               if (sk->sk_protocol != IPPROTO_UDP) {
+                       printk(KERN_ERR "tunl %hu: fd %d wrong protocol, got %d, expected %d\n",
+                              tunnel_id, fd, sk->sk_protocol, IPPROTO_UDP);
+                       goto err;
+               }
+               break;
+       case L2TP_ENCAPTYPE_IP:
+               err = -EPROTONOSUPPORT;
+               if (sk->sk_protocol != IPPROTO_L2TP) {
+                       printk(KERN_ERR "tunl %hu: fd %d wrong protocol, got %d, expected %d\n",
+                              tunnel_id, fd, sk->sk_protocol, IPPROTO_L2TP);
+                       goto err;
+               }
+               break;
        }
 
        /* Check if this socket has already been prepped */
@@ -1223,12 +1260,16 @@ int l2tp_tunnel_create(struct net *net, int fd, int version, u32 tunnel_id, u32
        tunnel->l2tp_net = net;
        pn = l2tp_pernet(net);
 
-       if (cfg)
+       if (cfg != NULL)
                tunnel->debug = cfg->debug;
 
        /* Mark socket as an encapsulation socket. See net/ipv4/udp.c */
-       udp_sk(sk)->encap_type = UDP_ENCAP_L2TPINUDP;
-       udp_sk(sk)->encap_rcv = l2tp_udp_encap_recv;
+       tunnel->encap = encap;
+       if (encap == L2TP_ENCAPTYPE_UDP) {
+               /* Mark socket as an encapsulation socket. See net/ipv4/udp.c */
+               udp_sk(sk)->encap_type = UDP_ENCAP_L2TPINUDP;
+               udp_sk(sk)->encap_rcv = l2tp_udp_encap_recv;
+       }
 
        sk->sk_user_data = tunnel;
 
@@ -1318,7 +1359,9 @@ void l2tp_session_set_header_len(struct l2tp_session *session, int version)
                if (session->send_seq)
                        session->hdr_len += 4;
        } else {
-               session->hdr_len = 8 + session->cookie_len + session->l2specific_len + session->offset;
+               session->hdr_len = 4 + session->cookie_len + session->l2specific_len + session->offset;
+               if (session->tunnel->encap == L2TP_ENCAPTYPE_UDP)
+                       session->hdr_len += 4;
        }
 
 }
index 5c53eb2..d239598 100644 (file)
@@ -48,6 +48,11 @@ enum l2tp_l2spec_type {
        L2TP_L2SPECTYPE_DEFAULT,
 };
 
+enum l2tp_encap_type {
+       L2TP_ENCAPTYPE_UDP,
+       L2TP_ENCAPTYPE_IP,
+};
+
 struct sk_buff;
 
 struct l2tp_stats {
@@ -155,6 +160,7 @@ struct l2tp_session {
 struct l2tp_tunnel_cfg {
        int                     debug;          /* bitmask of debug message
                                                 * categories */
+       enum l2tp_encap_type    encap;
 };
 
 struct l2tp_tunnel {
@@ -170,6 +176,7 @@ struct l2tp_tunnel {
        char                    name[20];       /* for logging */
        int                     debug;          /* bitmask of debug message
                                                 * categories */
+       enum l2tp_encap_type    encap;
        struct l2tp_stats       stats;
 
        struct list_head        list;           /* Keep a list of all tunnels */
diff --git a/net/l2tp/l2tp_ip.c b/net/l2tp/l2tp_ip.c
new file mode 100644 (file)
index 0000000..75bf784
--- /dev/null
@@ -0,0 +1,679 @@
+/*
+ * L2TPv3 IP encapsulation support
+ *
+ * Copyright (c) 2008,2009,2010 Katalix Systems Ltd
+ *
+ *     This program is free software; you can redistribute it and/or
+ *     modify it under the terms of the GNU General Public License
+ *     as published by the Free Software Foundation; either version
+ *     2 of the License, or (at your option) any later version.
+ */
+
+#include <linux/icmp.h>
+#include <linux/module.h>
+#include <linux/skbuff.h>
+#include <linux/random.h>
+#include <linux/socket.h>
+#include <linux/l2tp.h>
+#include <linux/in.h>
+#include <net/sock.h>
+#include <net/ip.h>
+#include <net/icmp.h>
+#include <net/udp.h>
+#include <net/inet_common.h>
+#include <net/inet_hashtables.h>
+#include <net/tcp_states.h>
+#include <net/protocol.h>
+#include <net/xfrm.h>
+
+#include "l2tp_core.h"
+
+struct l2tp_ip_sock {
+       /* inet_sock has to be the first member of l2tp_ip_sock */
+       struct inet_sock        inet;
+
+       __u32                   conn_id;
+       __u32                   peer_conn_id;
+
+       __u64                   tx_packets;
+       __u64                   tx_bytes;
+       __u64                   tx_errors;
+       __u64                   rx_packets;
+       __u64                   rx_bytes;
+       __u64                   rx_errors;
+};
+
+static DEFINE_RWLOCK(l2tp_ip_lock);
+static struct hlist_head l2tp_ip_table;
+static struct hlist_head l2tp_ip_bind_table;
+
+static inline struct l2tp_ip_sock *l2tp_ip_sk(const struct sock *sk)
+{
+       return (struct l2tp_ip_sock *)sk;
+}
+
+static struct sock *__l2tp_ip_bind_lookup(struct net *net, __be32 laddr, int dif, u32 tunnel_id)
+{
+       struct hlist_node *node;
+       struct sock *sk;
+
+       sk_for_each_bound(sk, node, &l2tp_ip_bind_table) {
+               struct inet_sock *inet = inet_sk(sk);
+               struct l2tp_ip_sock *l2tp = l2tp_ip_sk(sk);
+
+               if (l2tp == NULL)
+                       continue;
+
+               if ((l2tp->conn_id == tunnel_id) &&
+#ifdef CONFIG_NET_NS
+                   (sk->sk_net == net) &&
+#endif
+                   !(inet->inet_rcv_saddr && inet->inet_rcv_saddr != laddr) &&
+                   !(sk->sk_bound_dev_if && sk->sk_bound_dev_if != dif))
+                       goto found;
+       }
+
+       sk = NULL;
+found:
+       return sk;
+}
+
+static inline struct sock *l2tp_ip_bind_lookup(struct net *net, __be32 laddr, int dif, u32 tunnel_id)
+{
+       struct sock *sk = __l2tp_ip_bind_lookup(net, laddr, dif, tunnel_id);
+       if (sk)
+               sock_hold(sk);
+
+       return sk;
+}
+
+/* When processing receive frames, there are two cases to
+ * consider. Data frames consist of a non-zero session-id and an
+ * optional cookie. Control frames consist of a regular L2TP header
+ * preceded by 32-bits of zeros.
+ *
+ * L2TPv3 Session Header Over IP
+ *
+ *  0                   1                   2                   3
+ *  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
+ * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+ * |                           Session ID                          |
+ * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+ * |               Cookie (optional, maximum 64 bits)...
+ * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+ *                                                                 |
+ * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+ *
+ * L2TPv3 Control Message Header Over IP
+ *
+ *  0                   1                   2                   3
+ *  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
+ * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+ * |                      (32 bits of zeros)                       |
+ * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+ * |T|L|x|x|S|x|x|x|x|x|x|x|  Ver  |             Length            |
+ * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+ * |                     Control Connection ID                     |
+ * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+ * |               Ns              |               Nr              |
+ * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+ *
+ * All control frames are passed to userspace.
+ */
+static int l2tp_ip_recv(struct sk_buff *skb)
+{
+       struct sock *sk;
+       u32 session_id;
+       u32 tunnel_id;
+       unsigned char *ptr, *optr;
+       struct l2tp_session *session;
+       struct l2tp_tunnel *tunnel = NULL;
+       int length;
+       int offset;
+
+       /* Point to L2TP header */
+       optr = ptr = skb->data;
+
+       if (!pskb_may_pull(skb, 4))
+               goto discard;
+
+       session_id = ntohl(*((__be32 *) ptr));
+       ptr += 4;
+
+       /* RFC3931: L2TP/IP packets have the first 4 bytes containing
+        * the session_id. If it is 0, the packet is a L2TP control
+        * frame and the session_id value can be discarded.
+        */
+       if (session_id == 0) {
+               __skb_pull(skb, 4);
+               goto pass_up;
+       }
+
+       /* Ok, this is a data packet. Lookup the session. */
+       session = l2tp_session_find(&init_net, NULL, session_id);
+       if (session == NULL)
+               goto discard;
+
+       tunnel = session->tunnel;
+       if (tunnel == NULL)
+               goto discard;
+
+       /* Trace packet contents, if enabled */
+       if (tunnel->debug & L2TP_MSG_DATA) {
+               length = min(32u, skb->len);
+               if (!pskb_may_pull(skb, length))
+                       goto discard;
+
+               printk(KERN_DEBUG "%s: ip recv: ", tunnel->name);
+
+               offset = 0;
+               do {
+                       printk(" %02X", ptr[offset]);
+               } while (++offset < length);
+
+               printk("\n");
+       }
+
+       l2tp_recv_common(session, skb, ptr, optr, 0, skb->len, tunnel->recv_payload_hook);
+
+       return 0;
+
+pass_up:
+       /* Get the tunnel_id from the L2TP header */
+       if (!pskb_may_pull(skb, 12))
+               goto discard;
+
+       if ((skb->data[0] & 0xc0) != 0xc0)
+               goto discard;
+
+       tunnel_id = ntohl(*(__be32 *) &skb->data[4]);
+       tunnel = l2tp_tunnel_find(&init_net, tunnel_id);
+       if (tunnel != NULL)
+               sk = tunnel->sock;
+       else {
+               struct iphdr *iph = (struct iphdr *) skb_network_header(skb);
+
+               read_lock_bh(&l2tp_ip_lock);
+               sk = __l2tp_ip_bind_lookup(&init_net, iph->daddr, 0, tunnel_id);
+               read_unlock_bh(&l2tp_ip_lock);
+       }
+
+       if (sk == NULL)
+               goto discard;
+
+       sock_hold(sk);
+
+       if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
+               goto discard_put;
+
+       nf_reset(skb);
+
+       return sk_receive_skb(sk, skb, 1);
+
+discard_put:
+       sock_put(sk);
+
+discard:
+       kfree_skb(skb);
+       return 0;
+}
+
+static int l2tp_ip_open(struct sock *sk)
+{
+       /* Prevent autobind. We don't have ports. */
+       inet_sk(sk)->inet_num = IPPROTO_L2TP;
+
+       write_lock_bh(&l2tp_ip_lock);
+       sk_add_node(sk, &l2tp_ip_table);
+       write_unlock_bh(&l2tp_ip_lock);
+
+       return 0;
+}
+
+static void l2tp_ip_close(struct sock *sk, long timeout)
+{
+       write_lock_bh(&l2tp_ip_lock);
+       hlist_del_init(&sk->sk_bind_node);
+       hlist_del_init(&sk->sk_node);
+       write_unlock_bh(&l2tp_ip_lock);
+       sk_common_release(sk);
+}
+
+static void l2tp_ip_destroy_sock(struct sock *sk)
+{
+       struct sk_buff *skb;
+
+       while ((skb = __skb_dequeue_tail(&sk->sk_write_queue)) != NULL)
+               kfree_skb(skb);
+
+       sk_refcnt_debug_dec(sk);
+}
+
+static int l2tp_ip_bind(struct sock *sk, struct sockaddr *uaddr, int addr_len)
+{
+       struct inet_sock *inet = inet_sk(sk);
+       struct sockaddr_l2tpip *addr = (struct sockaddr_l2tpip *) uaddr;
+       int ret = -EINVAL;
+       int chk_addr_ret;
+
+       ret = -EADDRINUSE;
+       read_lock_bh(&l2tp_ip_lock);
+       if (__l2tp_ip_bind_lookup(&init_net, addr->l2tp_addr.s_addr, sk->sk_bound_dev_if, addr->l2tp_conn_id))
+               goto out_in_use;
+
+       read_unlock_bh(&l2tp_ip_lock);
+
+       lock_sock(sk);
+       if (sk->sk_state != TCP_CLOSE || addr_len < sizeof(struct sockaddr_l2tpip))
+               goto out;
+
+       chk_addr_ret = inet_addr_type(&init_net, addr->l2tp_addr.s_addr);
+       ret = -EADDRNOTAVAIL;
+       if (addr->l2tp_addr.s_addr && chk_addr_ret != RTN_LOCAL &&
+           chk_addr_ret != RTN_MULTICAST && chk_addr_ret != RTN_BROADCAST)
+               goto out;
+
+       inet->inet_rcv_saddr = inet->inet_saddr = addr->l2tp_addr.s_addr;
+       if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
+               inet->inet_saddr = 0;  /* Use device */
+       sk_dst_reset(sk);
+
+       l2tp_ip_sk(sk)->conn_id = addr->l2tp_conn_id;
+
+       write_lock_bh(&l2tp_ip_lock);
+       sk_add_bind_node(sk, &l2tp_ip_bind_table);
+       sk_del_node_init(sk);
+       write_unlock_bh(&l2tp_ip_lock);
+       ret = 0;
+out:
+       release_sock(sk);
+
+       return ret;
+
+out_in_use:
+       read_unlock_bh(&l2tp_ip_lock);
+
+       return ret;
+}
+
+static int l2tp_ip_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
+{
+       int rc;
+       struct inet_sock *inet = inet_sk(sk);
+       struct sockaddr_l2tpip *lsa = (struct sockaddr_l2tpip *) uaddr;
+       struct rtable *rt;
+       __be32 saddr;
+       int oif;
+
+       rc = -EINVAL;
+       if (addr_len < sizeof(*lsa))
+               goto out;
+
+       rc = -EAFNOSUPPORT;
+       if (lsa->l2tp_family != AF_INET)
+               goto out;
+
+       sk_dst_reset(sk);
+
+       oif = sk->sk_bound_dev_if;
+       saddr = inet->inet_saddr;
+
+       rc = -EINVAL;
+       if (ipv4_is_multicast(lsa->l2tp_addr.s_addr))
+               goto out;
+
+       rc = ip_route_connect(&rt, lsa->l2tp_addr.s_addr, saddr,
+                             RT_CONN_FLAGS(sk), oif,
+                             IPPROTO_L2TP,
+                             0, 0, sk, 1);
+       if (rc) {
+               if (rc == -ENETUNREACH)
+                       IP_INC_STATS_BH(&init_net, IPSTATS_MIB_OUTNOROUTES);
+               goto out;
+       }
+
+       rc = -ENETUNREACH;
+       if (rt->rt_flags & (RTCF_MULTICAST | RTCF_BROADCAST)) {
+               ip_rt_put(rt);
+               goto out;
+       }
+
+       l2tp_ip_sk(sk)->peer_conn_id = lsa->l2tp_conn_id;
+
+       if (!inet->inet_saddr)
+               inet->inet_saddr = rt->rt_src;
+       if (!inet->inet_rcv_saddr)
+               inet->inet_rcv_saddr = rt->rt_src;
+       inet->inet_daddr = rt->rt_dst;
+       sk->sk_state = TCP_ESTABLISHED;
+       inet->inet_id = jiffies;
+
+       sk_dst_set(sk, &rt->u.dst);
+
+       write_lock_bh(&l2tp_ip_lock);
+       hlist_del_init(&sk->sk_bind_node);
+       sk_add_bind_node(sk, &l2tp_ip_bind_table);
+       write_unlock_bh(&l2tp_ip_lock);
+
+       rc = 0;
+out:
+       return rc;
+}
+
+static int l2tp_ip_getname(struct socket *sock, struct sockaddr *uaddr,
+                          int *uaddr_len, int peer)
+{
+       struct sock *sk         = sock->sk;
+       struct inet_sock *inet  = inet_sk(sk);
+       struct l2tp_ip_sock *lsk = l2tp_ip_sk(sk);
+       struct sockaddr_l2tpip *lsa = (struct sockaddr_l2tpip *)uaddr;
+
+       memset(lsa, 0, sizeof(*lsa));
+       lsa->l2tp_family = AF_INET;
+       if (peer) {
+               if (!inet->inet_dport)
+                       return -ENOTCONN;
+               lsa->l2tp_conn_id = lsk->peer_conn_id;
+               lsa->l2tp_addr.s_addr = inet->inet_daddr;
+       } else {
+               __be32 addr = inet->inet_rcv_saddr;
+               if (!addr)
+                       addr = inet->inet_saddr;
+               lsa->l2tp_conn_id = lsk->conn_id;
+               lsa->l2tp_addr.s_addr = addr;
+       }
+       *uaddr_len = sizeof(*lsa);
+       return 0;
+}
+
+static int l2tp_ip_backlog_recv(struct sock *sk, struct sk_buff *skb)
+{
+       int rc;
+
+       if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
+               goto drop;
+
+       nf_reset(skb);
+
+       /* Charge it to the socket, dropping if the queue is full. */
+       rc = sock_queue_rcv_skb(sk, skb);
+       if (rc < 0)
+               goto drop;
+
+       return 0;
+
+drop:
+       IP_INC_STATS(&init_net, IPSTATS_MIB_INDISCARDS);
+       kfree_skb(skb);
+       return -1;
+}
+
+/* Userspace will call sendmsg() on the tunnel socket to send L2TP
+ * control frames.
+ */
+static int l2tp_ip_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg, size_t len)
+{
+       struct sk_buff *skb;
+       int rc;
+       struct l2tp_ip_sock *lsa = l2tp_ip_sk(sk);
+       struct inet_sock *inet = inet_sk(sk);
+       struct ip_options *opt = inet->opt;
+       struct rtable *rt = NULL;
+       int connected = 0;
+       __be32 daddr;
+
+       if (sock_flag(sk, SOCK_DEAD))
+               return -ENOTCONN;
+
+       /* Get and verify the address. */
+       if (msg->msg_name) {
+               struct sockaddr_l2tpip *lip = (struct sockaddr_l2tpip *) msg->msg_name;
+               if (msg->msg_namelen < sizeof(*lip))
+                       return -EINVAL;
+
+               if (lip->l2tp_family != AF_INET) {
+                       if (lip->l2tp_family != AF_UNSPEC)
+                               return -EAFNOSUPPORT;
+               }
+
+               daddr = lip->l2tp_addr.s_addr;
+       } else {
+               if (sk->sk_state != TCP_ESTABLISHED)
+                       return -EDESTADDRREQ;
+
+               daddr = inet->inet_daddr;
+               connected = 1;
+       }
+
+       /* Allocate a socket buffer */
+       rc = -ENOMEM;
+       skb = sock_wmalloc(sk, 2 + NET_SKB_PAD + sizeof(struct iphdr) +
+                          4 + len, 0, GFP_KERNEL);
+       if (!skb)
+               goto error;
+
+       /* Reserve space for headers, putting IP header on 4-byte boundary. */
+       skb_reserve(skb, 2 + NET_SKB_PAD);
+       skb_reset_network_header(skb);
+       skb_reserve(skb, sizeof(struct iphdr));
+       skb_reset_transport_header(skb);
+
+       /* Insert 0 session_id */
+       *((__be32 *) skb_put(skb, 4)) = 0;
+
+       /* Copy user data into skb */
+       rc = memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len);
+       if (rc < 0) {
+               kfree_skb(skb);
+               goto error;
+       }
+
+       if (connected)
+               rt = (struct rtable *) __sk_dst_check(sk, 0);
+
+       if (rt == NULL) {
+               /* Use correct destination address if we have options. */
+               if (opt && opt->srr)
+                       daddr = opt->faddr;
+
+               {
+                       struct flowi fl = { .oif = sk->sk_bound_dev_if,
+                                           .nl_u = { .ip4_u = {
+                                                       .daddr = daddr,
+                                                       .saddr = inet->inet_saddr,
+                                                       .tos = RT_CONN_FLAGS(sk) } },
+                                           .proto = sk->sk_protocol,
+                                           .flags = inet_sk_flowi_flags(sk),
+                                           .uli_u = { .ports = {
+                                                        .sport = inet->inet_sport,
+                                                        .dport = inet->inet_dport } } };
+
+                       /* If this fails, retransmit mechanism of transport layer will
+                        * keep trying until route appears or the connection times
+                        * itself out.
+                        */
+                       security_sk_classify_flow(sk, &fl);
+                       if (ip_route_output_flow(sock_net(sk), &rt, &fl, sk, 0))
+                               goto no_route;
+               }
+               sk_setup_caps(sk, &rt->u.dst);
+       }
+       skb_dst_set(skb, dst_clone(&rt->u.dst));
+
+       /* Queue the packet to IP for output */
+       rc = ip_queue_xmit(skb, 0);
+
+error:
+       /* Update stats */
+       if (rc >= 0) {
+               lsa->tx_packets++;
+               lsa->tx_bytes += len;
+               rc = len;
+       } else {
+               lsa->tx_errors++;
+       }
+
+       return rc;
+
+no_route:
+       IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTNOROUTES);
+       kfree_skb(skb);
+       return -EHOSTUNREACH;
+}
+
+static int l2tp_ip_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
+                          size_t len, int noblock, int flags, int *addr_len)
+{
+       struct inet_sock *inet = inet_sk(sk);
+       struct l2tp_ip_sock *lsk = l2tp_ip_sk(sk);
+       size_t copied = 0;
+       int err = -EOPNOTSUPP;
+       struct sockaddr_in *sin = (struct sockaddr_in *)msg->msg_name;
+       struct sk_buff *skb;
+
+       if (flags & MSG_OOB)
+               goto out;
+
+       if (addr_len)
+               *addr_len = sizeof(*sin);
+
+       skb = skb_recv_datagram(sk, flags, noblock, &err);
+       if (!skb)
+               goto out;
+
+       copied = skb->len;
+       if (len < copied) {
+               msg->msg_flags |= MSG_TRUNC;
+               copied = len;
+       }
+
+       err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
+       if (err)
+               goto done;
+
+       sock_recv_timestamp(msg, sk, skb);
+
+       /* Copy the address. */
+       if (sin) {
+               sin->sin_family = AF_INET;
+               sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
+               sin->sin_port = 0;
+               memset(&sin->sin_zero, 0, sizeof(sin->sin_zero));
+       }
+       if (inet->cmsg_flags)
+               ip_cmsg_recv(msg, skb);
+       if (flags & MSG_TRUNC)
+               copied = skb->len;
+done:
+       skb_free_datagram(sk, skb);
+out:
+       if (err) {
+               lsk->rx_errors++;
+               return err;
+       }
+
+       lsk->rx_packets++;
+       lsk->rx_bytes += copied;
+
+       return copied;
+}
+
+struct proto l2tp_ip_prot = {
+       .name              = "L2TP/IP",
+       .owner             = THIS_MODULE,
+       .init              = l2tp_ip_open,
+       .close             = l2tp_ip_close,
+       .bind              = l2tp_ip_bind,
+       .connect           = l2tp_ip_connect,
+       .disconnect        = udp_disconnect,
+       .ioctl             = udp_ioctl,
+       .destroy           = l2tp_ip_destroy_sock,
+       .setsockopt        = ip_setsockopt,
+       .getsockopt        = ip_getsockopt,
+       .sendmsg           = l2tp_ip_sendmsg,
+       .recvmsg           = l2tp_ip_recvmsg,
+       .backlog_rcv       = l2tp_ip_backlog_recv,
+       .hash              = inet_hash,
+       .unhash            = inet_unhash,
+       .obj_size          = sizeof(struct l2tp_ip_sock),
+#ifdef CONFIG_COMPAT
+       .compat_setsockopt = compat_ip_setsockopt,
+       .compat_getsockopt = compat_ip_getsockopt,
+#endif
+};
+
+static const struct proto_ops l2tp_ip_ops = {
+       .family            = PF_INET,
+       .owner             = THIS_MODULE,
+       .release           = inet_release,
+       .bind              = inet_bind,
+       .connect           = inet_dgram_connect,
+       .socketpair        = sock_no_socketpair,
+       .accept            = sock_no_accept,
+       .getname           = l2tp_ip_getname,
+       .poll              = datagram_poll,
+       .ioctl             = inet_ioctl,
+       .listen            = sock_no_listen,
+       .shutdown          = inet_shutdown,
+       .setsockopt        = sock_common_setsockopt,
+       .getsockopt        = sock_common_getsockopt,
+       .sendmsg           = inet_sendmsg,
+       .recvmsg           = sock_common_recvmsg,
+       .mmap              = sock_no_mmap,
+       .sendpage          = sock_no_sendpage,
+#ifdef CONFIG_COMPAT
+       .compat_setsockopt = compat_sock_common_setsockopt,
+       .compat_getsockopt = compat_sock_common_getsockopt,
+#endif
+};
+
+static struct inet_protosw l2tp_ip_protosw = {
+       .type           = SOCK_DGRAM,
+       .protocol       = IPPROTO_L2TP,
+       .prot           = &l2tp_ip_prot,
+       .ops            = &l2tp_ip_ops,
+       .no_check       = 0,
+};
+
+static struct net_protocol l2tp_ip_protocol __read_mostly = {
+       .handler        = l2tp_ip_recv,
+};
+
+static int __init l2tp_ip_init(void)
+{
+       int err;
+
+       printk(KERN_INFO "L2TP IP encapsulation support (L2TPv3)\n");
+
+       err = proto_register(&l2tp_ip_prot, 1);
+       if (err != 0)
+               goto out;
+
+       err = inet_add_protocol(&l2tp_ip_protocol, IPPROTO_L2TP);
+       if (err)
+               goto out1;
+
+       inet_register_protosw(&l2tp_ip_protosw);
+       return 0;
+
+out1:
+       proto_unregister(&l2tp_ip_prot);
+out:
+       return err;
+}
+
+static void __exit l2tp_ip_exit(void)
+{
+       inet_unregister_protosw(&l2tp_ip_protosw);
+       inet_del_protocol(&l2tp_ip_protocol, IPPROTO_L2TP);
+       proto_unregister(&l2tp_ip_prot);
+}
+
+module_init(l2tp_ip_init);
+module_exit(l2tp_ip_exit);
+
+MODULE_LICENSE("GPL");
+MODULE_AUTHOR("James Chapman <jchapman@katalix.com>");
+MODULE_DESCRIPTION("L2TP over IP");
+MODULE_VERSION("1.0");
+MODULE_ALIAS_NET_PF_PROTO_TYPE(PF_INET, SOCK_DGRAM, IPPROTO_L2TP);
index e5b5312..63fc62b 100644 (file)
@@ -305,6 +305,7 @@ static int pppol2tp_sendmsg(struct kiocb *iocb, struct socket *sock, struct msgh
        struct l2tp_session *session;
        struct l2tp_tunnel *tunnel;
        struct pppol2tp_session *ps;
+       int uhlen;
 
        error = -ENOTCONN;
        if (sock_flag(sk, SOCK_DEAD) || !(sk->sk_state & PPPOX_CONNECTED))
@@ -321,10 +322,12 @@ static int pppol2tp_sendmsg(struct kiocb *iocb, struct socket *sock, struct msgh
        if (tunnel == NULL)
                goto error_put_sess;
 
+       uhlen = (tunnel->encap == L2TP_ENCAPTYPE_UDP) ? sizeof(struct udphdr) : 0;
+
        /* Allocate a socket buffer */
        error = -ENOMEM;
        skb = sock_wmalloc(sk, NET_SKB_PAD + sizeof(struct iphdr) +
-                          sizeof(struct udphdr) + session->hdr_len +
+                          uhlen + session->hdr_len +
                           sizeof(ppph) + total_len,
                           0, GFP_KERNEL);
        if (!skb)
@@ -335,7 +338,7 @@ static int pppol2tp_sendmsg(struct kiocb *iocb, struct socket *sock, struct msgh
        skb_reset_network_header(skb);
        skb_reserve(skb, sizeof(struct iphdr));
        skb_reset_transport_header(skb);
-       skb_reserve(skb, sizeof(struct udphdr));
+       skb_reserve(skb, uhlen);
 
        /* Add PPP header */
        skb->data[0] = ppph[0];