xref: /linux/net/rds/tcp_connect.c (revision ca55b2fef3a9373fcfc30f82fd26bc7fccbda732)
1 /*
2  * Copyright (c) 2006 Oracle.  All rights reserved.
3  *
4  * This software is available to you under a choice of one of two
5  * licenses.  You may choose to be licensed under the terms of the GNU
6  * General Public License (GPL) Version 2, available from the file
7  * COPYING in the main directory of this source tree, or the
8  * OpenIB.org BSD license below:
9  *
10  *     Redistribution and use in source and binary forms, with or
11  *     without modification, are permitted provided that the following
12  *     conditions are met:
13  *
14  *      - Redistributions of source code must retain the above
15  *        copyright notice, this list of conditions and the following
16  *        disclaimer.
17  *
18  *      - Redistributions in binary form must reproduce the above
19  *        copyright notice, this list of conditions and the following
20  *        disclaimer in the documentation and/or other materials
21  *        provided with the distribution.
22  *
23  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
24  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
25  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
26  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
27  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
28  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
29  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
30  * SOFTWARE.
31  *
32  */
33 #include <linux/kernel.h>
34 #include <linux/in.h>
35 #include <net/tcp.h>
36 
37 #include "rds.h"
38 #include "tcp.h"
39 
40 void rds_tcp_state_change(struct sock *sk)
41 {
42 	void (*state_change)(struct sock *sk);
43 	struct rds_connection *conn;
44 	struct rds_tcp_connection *tc;
45 
46 	read_lock(&sk->sk_callback_lock);
47 	conn = sk->sk_user_data;
48 	if (!conn) {
49 		state_change = sk->sk_state_change;
50 		goto out;
51 	}
52 	tc = conn->c_transport_data;
53 	state_change = tc->t_orig_state_change;
54 
55 	rdsdebug("sock %p state_change to %d\n", tc->t_sock, sk->sk_state);
56 
57 	switch(sk->sk_state) {
58 		/* ignore connecting sockets as they make progress */
59 		case TCP_SYN_SENT:
60 		case TCP_SYN_RECV:
61 			break;
62 		case TCP_ESTABLISHED:
63 			rds_connect_complete(conn);
64 			break;
65 		case TCP_CLOSE_WAIT:
66 		case TCP_CLOSE:
67 			rds_conn_drop(conn);
68 		default:
69 			break;
70 	}
71 out:
72 	read_unlock(&sk->sk_callback_lock);
73 	state_change(sk);
74 }
75 
76 int rds_tcp_conn_connect(struct rds_connection *conn)
77 {
78 	struct socket *sock = NULL;
79 	struct sockaddr_in src, dest;
80 	int ret;
81 
82 	ret = sock_create_kern(rds_conn_net(conn), PF_INET,
83 			       SOCK_STREAM, IPPROTO_TCP, &sock);
84 	if (ret < 0)
85 		goto out;
86 
87 	rds_tcp_tune(sock);
88 
89 	src.sin_family = AF_INET;
90 	src.sin_addr.s_addr = (__force u32)conn->c_laddr;
91 	src.sin_port = (__force u16)htons(0);
92 
93 	ret = sock->ops->bind(sock, (struct sockaddr *)&src, sizeof(src));
94 	if (ret) {
95 		rdsdebug("bind failed with %d at address %pI4\n",
96 			 ret, &conn->c_laddr);
97 		goto out;
98 	}
99 
100 	dest.sin_family = AF_INET;
101 	dest.sin_addr.s_addr = (__force u32)conn->c_faddr;
102 	dest.sin_port = (__force u16)htons(RDS_TCP_PORT);
103 
104 	/*
105 	 * once we call connect() we can start getting callbacks and they
106 	 * own the socket
107 	 */
108 	rds_tcp_set_callbacks(sock, conn);
109 	ret = sock->ops->connect(sock, (struct sockaddr *)&dest, sizeof(dest),
110 				 O_NONBLOCK);
111 
112 	rdsdebug("connect to address %pI4 returned %d\n", &conn->c_faddr, ret);
113 	if (ret == -EINPROGRESS)
114 		ret = 0;
115 	if (ret == 0) {
116 		rds_tcp_keepalive(sock);
117 		sock = NULL;
118 	} else {
119 		rds_tcp_restore_callbacks(sock, conn->c_transport_data);
120 	}
121 
122 out:
123 	if (sock)
124 		sock_release(sock);
125 	return ret;
126 }
127 
128 /*
129  * Before killing the tcp socket this needs to serialize with callbacks.  The
130  * caller has already grabbed the sending sem so we're serialized with other
131  * senders.
132  *
133  * TCP calls the callbacks with the sock lock so we hold it while we reset the
134  * callbacks to those set by TCP.  Our callbacks won't execute again once we
135  * hold the sock lock.
136  */
137 void rds_tcp_conn_shutdown(struct rds_connection *conn)
138 {
139 	struct rds_tcp_connection *tc = conn->c_transport_data;
140 	struct socket *sock = tc->t_sock;
141 
142 	rdsdebug("shutting down conn %p tc %p sock %p\n", conn, tc, sock);
143 
144 	if (sock) {
145 		sock->ops->shutdown(sock, RCV_SHUTDOWN | SEND_SHUTDOWN);
146 		lock_sock(sock->sk);
147 		rds_tcp_restore_callbacks(sock, tc); /* tc->tc_sock = NULL */
148 
149 		release_sock(sock->sk);
150 		sock_release(sock);
151 	}
152 
153 	if (tc->t_tinc) {
154 		rds_inc_put(&tc->t_tinc->ti_inc);
155 		tc->t_tinc = NULL;
156 	}
157 	tc->t_tinc_hdr_rem = sizeof(struct rds_header);
158 	tc->t_tinc_data_rem = 0;
159 }
160