xref: /linux/net/rds/tcp_connect.c (revision d257f9bf06129613de539ea71ecea60848b662cd)
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_conn_path *cp;
44 	struct rds_tcp_connection *tc;
45 
46 	read_lock_bh(&sk->sk_callback_lock);
47 	cp = sk->sk_user_data;
48 	if (!cp) {
49 		state_change = sk->sk_state_change;
50 		goto out;
51 	}
52 	tc = cp->cp_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 		/* Force the peer to reconnect so that we have the
64 		 * TCP ports going from <smaller-ip>.<transient> to
65 		 * <larger-ip>.<RDS_TCP_PORT>. We avoid marking the
66 		 * RDS connection as RDS_CONN_UP until the reconnect,
67 		 * to avoid RDS datagram loss.
68 		 */
69 		if (!IS_CANONICAL(cp->cp_conn->c_laddr, cp->cp_conn->c_faddr) &&
70 		    rds_conn_path_transition(cp, RDS_CONN_CONNECTING,
71 					     RDS_CONN_ERROR)) {
72 			rds_conn_path_drop(cp);
73 		} else {
74 			rds_connect_path_complete(cp, RDS_CONN_CONNECTING);
75 		}
76 		break;
77 	case TCP_CLOSE_WAIT:
78 	case TCP_CLOSE:
79 		rds_conn_path_drop(cp);
80 	default:
81 		break;
82 	}
83 out:
84 	read_unlock_bh(&sk->sk_callback_lock);
85 	state_change(sk);
86 }
87 
88 int rds_tcp_conn_path_connect(struct rds_conn_path *cp)
89 {
90 	struct socket *sock = NULL;
91 	struct sockaddr_in src, dest;
92 	int ret;
93 	struct rds_connection *conn = cp->cp_conn;
94 	struct rds_tcp_connection *tc = cp->cp_transport_data;
95 
96 	/* for multipath rds,we only trigger the connection after
97 	 * the handshake probe has determined the number of paths.
98 	 */
99 	if (cp->cp_index > 0 && cp->cp_conn->c_npaths < 2)
100 		return -EAGAIN;
101 
102 	mutex_lock(&tc->t_conn_path_lock);
103 
104 	if (rds_conn_path_up(cp)) {
105 		mutex_unlock(&tc->t_conn_path_lock);
106 		return 0;
107 	}
108 	ret = sock_create_kern(rds_conn_net(conn), PF_INET,
109 			       SOCK_STREAM, IPPROTO_TCP, &sock);
110 	if (ret < 0)
111 		goto out;
112 
113 	rds_tcp_tune(sock);
114 
115 	src.sin_family = AF_INET;
116 	src.sin_addr.s_addr = (__force u32)conn->c_laddr;
117 	src.sin_port = (__force u16)htons(0);
118 
119 	ret = sock->ops->bind(sock, (struct sockaddr *)&src, sizeof(src));
120 	if (ret) {
121 		rdsdebug("bind failed with %d at address %pI4\n",
122 			 ret, &conn->c_laddr);
123 		goto out;
124 	}
125 
126 	dest.sin_family = AF_INET;
127 	dest.sin_addr.s_addr = (__force u32)conn->c_faddr;
128 	dest.sin_port = (__force u16)htons(RDS_TCP_PORT);
129 
130 	/*
131 	 * once we call connect() we can start getting callbacks and they
132 	 * own the socket
133 	 */
134 	rds_tcp_set_callbacks(sock, cp);
135 	ret = sock->ops->connect(sock, (struct sockaddr *)&dest, sizeof(dest),
136 				 O_NONBLOCK);
137 
138 	rdsdebug("connect to address %pI4 returned %d\n", &conn->c_faddr, ret);
139 	if (ret == -EINPROGRESS)
140 		ret = 0;
141 	if (ret == 0) {
142 		rds_tcp_keepalive(sock);
143 		sock = NULL;
144 	} else {
145 		rds_tcp_restore_callbacks(sock, cp->cp_transport_data);
146 	}
147 
148 out:
149 	mutex_unlock(&tc->t_conn_path_lock);
150 	if (sock)
151 		sock_release(sock);
152 	return ret;
153 }
154 
155 /*
156  * Before killing the tcp socket this needs to serialize with callbacks.  The
157  * caller has already grabbed the sending sem so we're serialized with other
158  * senders.
159  *
160  * TCP calls the callbacks with the sock lock so we hold it while we reset the
161  * callbacks to those set by TCP.  Our callbacks won't execute again once we
162  * hold the sock lock.
163  */
164 void rds_tcp_conn_path_shutdown(struct rds_conn_path *cp)
165 {
166 	struct rds_tcp_connection *tc = cp->cp_transport_data;
167 	struct socket *sock = tc->t_sock;
168 
169 	rdsdebug("shutting down conn %p tc %p sock %p\n",
170 		 cp->cp_conn, tc, sock);
171 
172 	if (sock) {
173 		if (cp->cp_conn->c_destroy_in_prog)
174 			rds_tcp_set_linger(sock);
175 		sock->ops->shutdown(sock, RCV_SHUTDOWN | SEND_SHUTDOWN);
176 		lock_sock(sock->sk);
177 		rds_tcp_restore_callbacks(sock, tc); /* tc->tc_sock = NULL */
178 
179 		release_sock(sock->sk);
180 		sock_release(sock);
181 	}
182 
183 	if (tc->t_tinc) {
184 		rds_inc_put(&tc->t_tinc->ti_inc);
185 		tc->t_tinc = NULL;
186 	}
187 	tc->t_tinc_hdr_rem = sizeof(struct rds_header);
188 	tc->t_tinc_data_rem = 0;
189 }
190