xref: /openbmc/linux/net/rds/tcp_connect.c (revision eee2fa6ab3225192d6d894c54a6fb02ac9efdff6)
1 /*
2  * Copyright (c) 2006, 2017 Oracle and/or its affiliates. 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 (rds_addr_cmp(&cp->cp_conn->c_laddr,
70 				 &cp->cp_conn->c_faddr) >= 0 &&
71 		    rds_conn_path_transition(cp, RDS_CONN_CONNECTING,
72 					     RDS_CONN_ERROR)) {
73 			rds_conn_path_drop(cp, false);
74 		} else {
75 			rds_connect_path_complete(cp, RDS_CONN_CONNECTING);
76 		}
77 		break;
78 	case TCP_CLOSE_WAIT:
79 	case TCP_CLOSE:
80 		rds_conn_path_drop(cp, false);
81 	default:
82 		break;
83 	}
84 out:
85 	read_unlock_bh(&sk->sk_callback_lock);
86 	state_change(sk);
87 }
88 
89 int rds_tcp_conn_path_connect(struct rds_conn_path *cp)
90 {
91 	struct socket *sock = NULL;
92 	struct sockaddr_in sin;
93 	struct sockaddr *addr;
94 	int addrlen;
95 	int ret;
96 	struct rds_connection *conn = cp->cp_conn;
97 	struct rds_tcp_connection *tc = cp->cp_transport_data;
98 
99 	/* for multipath rds,we only trigger the connection after
100 	 * the handshake probe has determined the number of paths.
101 	 */
102 	if (cp->cp_index > 0 && cp->cp_conn->c_npaths < 2)
103 		return -EAGAIN;
104 
105 	mutex_lock(&tc->t_conn_path_lock);
106 
107 	if (rds_conn_path_up(cp)) {
108 		mutex_unlock(&tc->t_conn_path_lock);
109 		return 0;
110 	}
111 	ret = sock_create_kern(rds_conn_net(conn), PF_INET,
112 			       SOCK_STREAM, IPPROTO_TCP, &sock);
113 	if (ret < 0)
114 		goto out;
115 
116 	rds_tcp_tune(sock);
117 
118 	sin.sin_family = AF_INET;
119 	sin.sin_addr.s_addr = conn->c_laddr.s6_addr32[3];
120 	sin.sin_port = 0;
121 	addr = (struct sockaddr *)&sin;
122 	addrlen = sizeof(sin);
123 
124 	ret = sock->ops->bind(sock, addr, addrlen);
125 	if (ret) {
126 		rdsdebug("bind failed with %d at address %pI6c\n",
127 			 ret, &conn->c_laddr);
128 		goto out;
129 	}
130 
131 	sin.sin_family = AF_INET;
132 	sin.sin_addr.s_addr = conn->c_faddr.s6_addr32[3];
133 	sin.sin_port = htons(RDS_TCP_PORT);
134 	addr = (struct sockaddr *)&sin;
135 	addrlen = sizeof(sin);
136 
137 	/*
138 	 * once we call connect() we can start getting callbacks and they
139 	 * own the socket
140 	 */
141 	rds_tcp_set_callbacks(sock, cp);
142 	ret = sock->ops->connect(sock, addr, addrlen, O_NONBLOCK);
143 
144 	rdsdebug("connect to address %pI6c returned %d\n", &conn->c_faddr, ret);
145 	if (ret == -EINPROGRESS)
146 		ret = 0;
147 	if (ret == 0) {
148 		rds_tcp_keepalive(sock);
149 		sock = NULL;
150 	} else {
151 		rds_tcp_restore_callbacks(sock, cp->cp_transport_data);
152 	}
153 
154 out:
155 	mutex_unlock(&tc->t_conn_path_lock);
156 	if (sock)
157 		sock_release(sock);
158 	return ret;
159 }
160 
161 /*
162  * Before killing the tcp socket this needs to serialize with callbacks.  The
163  * caller has already grabbed the sending sem so we're serialized with other
164  * senders.
165  *
166  * TCP calls the callbacks with the sock lock so we hold it while we reset the
167  * callbacks to those set by TCP.  Our callbacks won't execute again once we
168  * hold the sock lock.
169  */
170 void rds_tcp_conn_path_shutdown(struct rds_conn_path *cp)
171 {
172 	struct rds_tcp_connection *tc = cp->cp_transport_data;
173 	struct socket *sock = tc->t_sock;
174 
175 	rdsdebug("shutting down conn %p tc %p sock %p\n",
176 		 cp->cp_conn, tc, sock);
177 
178 	if (sock) {
179 		if (rds_destroy_pending(cp->cp_conn))
180 			rds_tcp_set_linger(sock);
181 		sock->ops->shutdown(sock, RCV_SHUTDOWN | SEND_SHUTDOWN);
182 		lock_sock(sock->sk);
183 		rds_tcp_restore_callbacks(sock, tc); /* tc->tc_sock = NULL */
184 
185 		release_sock(sock->sk);
186 		sock_release(sock);
187 	}
188 
189 	if (tc->t_tinc) {
190 		rds_inc_put(&tc->t_tinc->ti_inc);
191 		tc->t_tinc = NULL;
192 	}
193 	tc->t_tinc_hdr_rem = sizeof(struct rds_header);
194 	tc->t_tinc_data_rem = 0;
195 }
196