2 * Copyright (c) 2006 Oracle. All rights reserved.
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:
10 * Redistribution and use in source and binary forms, with or
11 * without modification, are permitted provided that the following
14 * - Redistributions of source code must retain the above
15 * copyright notice, this list of conditions and the following
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.
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
33 #include <linux/kernel.h>
34 #include <linux/gfp.h>
41 int rds_tcp_keepalive(struct socket *sock)
43 /* values below based on xs_udp_default_timeout */
44 int keepidle = 5; /* send a probe 'keepidle' secs after last data */
45 int keepcnt = 5; /* number of unack'ed probes before declaring dead */
49 ret = kernel_setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE,
50 (char *)&keepalive, sizeof(keepalive));
54 ret = kernel_setsockopt(sock, IPPROTO_TCP, TCP_KEEPCNT,
55 (char *)&keepcnt, sizeof(keepcnt));
59 ret = kernel_setsockopt(sock, IPPROTO_TCP, TCP_KEEPIDLE,
60 (char *)&keepidle, sizeof(keepidle));
64 /* KEEPINTVL is the interval between successive probes. We follow
65 * the model in xs_tcp_finish_connecting() and re-use keepidle.
67 ret = kernel_setsockopt(sock, IPPROTO_TCP, TCP_KEEPINTVL,
68 (char *)&keepidle, sizeof(keepidle));
73 /* rds_tcp_accept_one_path(): if accepting on cp_index > 0, make sure the
74 * client's ipaddr < server's ipaddr. Otherwise, close the accepted
75 * socket and force a reconneect from smaller -> larger ip addr. The reason
76 * we special case cp_index 0 is to allow the rds probe ping itself to itself
77 * get through efficiently.
78 * Since reconnects are only initiated from the node with the numerically
79 * smaller ip address, we recycle conns in RDS_CONN_ERROR on the passive side
80 * by moving them to CONNECTING in this function.
82 struct rds_tcp_connection *rds_tcp_accept_one_path(struct rds_connection *conn)
85 bool peer_is_smaller = (conn->c_faddr < conn->c_laddr);
86 int npaths = conn->c_npaths;
89 struct rds_conn_path *cp = &conn->c_path[0];
92 ret = rds_conn_path_transition(cp, RDS_CONN_DOWN,
95 rds_conn_path_transition(cp, RDS_CONN_ERROR,
97 return cp->cp_transport_data;
100 /* for mprds, paths with cp_index > 0 MUST be initiated by the peer
101 * with the smaller address.
103 if (!peer_is_smaller)
106 for (i = 1; i < npaths; i++) {
107 struct rds_conn_path *cp = &conn->c_path[i];
109 if (rds_conn_path_transition(cp, RDS_CONN_DOWN,
110 RDS_CONN_CONNECTING) ||
111 rds_conn_path_transition(cp, RDS_CONN_ERROR,
112 RDS_CONN_CONNECTING)) {
113 return cp->cp_transport_data;
119 int rds_tcp_accept_one(struct socket *sock)
121 struct socket *new_sock = NULL;
122 struct rds_connection *conn;
124 struct inet_sock *inet;
125 struct rds_tcp_connection *rs_tcp = NULL;
127 struct rds_conn_path *cp;
129 if (!sock) /* module unload or netns delete in progress */
132 ret = sock_create_kern(sock_net(sock->sk), sock->sk->sk_family,
133 sock->sk->sk_type, sock->sk->sk_protocol,
138 new_sock->type = sock->type;
139 new_sock->ops = sock->ops;
140 ret = sock->ops->accept(sock, new_sock, O_NONBLOCK);
144 ret = rds_tcp_keepalive(new_sock);
148 rds_tcp_tune(new_sock);
150 inet = inet_sk(new_sock->sk);
152 rdsdebug("accepted tcp %pI4:%u -> %pI4:%u\n",
153 &inet->inet_saddr, ntohs(inet->inet_sport),
154 &inet->inet_daddr, ntohs(inet->inet_dport));
156 conn = rds_conn_create(sock_net(sock->sk),
157 inet->inet_saddr, inet->inet_daddr,
158 &rds_tcp_transport, GFP_KERNEL);
163 /* An incoming SYN request came in, and TCP just accepted it.
165 * If the client reboots, this conn will need to be cleaned up.
166 * rds_tcp_state_change() will do that cleanup
168 rs_tcp = rds_tcp_accept_one_path(conn);
171 mutex_lock(&rs_tcp->t_conn_path_lock);
172 cp = rs_tcp->t_cpath;
173 conn_state = rds_conn_path_state(cp);
174 if (conn_state != RDS_CONN_CONNECTING && conn_state != RDS_CONN_UP &&
175 conn_state != RDS_CONN_ERROR)
177 if (rs_tcp->t_sock) {
178 /* Need to resolve a duelling SYN between peers.
179 * We have an outstanding SYN to this peer, which may
180 * potentially have transitioned to the RDS_CONN_UP state,
181 * so we must quiesce any send threads before resetting
184 if (ntohl(inet->inet_saddr) < ntohl(inet->inet_daddr) ||
188 rds_tcp_reset_callbacks(new_sock, cp);
190 /* rds_connect_path_complete() marks RDS_CONN_UP */
191 rds_connect_path_complete(cp, RDS_CONN_RESETTING);
194 rds_tcp_set_callbacks(new_sock, cp);
195 rds_connect_path_complete(cp, RDS_CONN_CONNECTING);
201 /* reset the newly returned accept sock and bail */
202 kernel_sock_shutdown(new_sock, SHUT_RDWR);
206 mutex_unlock(&rs_tcp->t_conn_path_lock);
208 sock_release(new_sock);
212 void rds_tcp_listen_data_ready(struct sock *sk)
214 void (*ready)(struct sock *sk);
216 rdsdebug("listen data ready sk %p\n", sk);
218 read_lock_bh(&sk->sk_callback_lock);
219 ready = sk->sk_user_data;
220 if (!ready) { /* check for teardown race */
221 ready = sk->sk_data_ready;
226 * ->sk_data_ready is also called for a newly established child socket
227 * before it has been accepted and the accepter has set up their
228 * data_ready.. we only want to queue listen work for our listening
231 if (sk->sk_state == TCP_LISTEN)
232 rds_tcp_accept_work(sk);
234 ready = rds_tcp_listen_sock_def_readable(sock_net(sk));
237 read_unlock_bh(&sk->sk_callback_lock);
241 struct socket *rds_tcp_listen_init(struct net *net)
243 struct sockaddr_in sin;
244 struct socket *sock = NULL;
247 ret = sock_create_kern(net, PF_INET, SOCK_STREAM, IPPROTO_TCP, &sock);
251 sock->sk->sk_reuse = SK_CAN_REUSE;
252 rds_tcp_nonagle(sock);
254 write_lock_bh(&sock->sk->sk_callback_lock);
255 sock->sk->sk_user_data = sock->sk->sk_data_ready;
256 sock->sk->sk_data_ready = rds_tcp_listen_data_ready;
257 write_unlock_bh(&sock->sk->sk_callback_lock);
259 sin.sin_family = PF_INET;
260 sin.sin_addr.s_addr = (__force u32)htonl(INADDR_ANY);
261 sin.sin_port = (__force u16)htons(RDS_TCP_PORT);
263 ret = sock->ops->bind(sock, (struct sockaddr *)&sin, sizeof(sin));
267 ret = sock->ops->listen(sock, 64);
278 void rds_tcp_listen_stop(struct socket *sock)
287 /* serialize with and prevent further callbacks */
289 write_lock_bh(&sk->sk_callback_lock);
290 if (sk->sk_user_data) {
291 sk->sk_data_ready = sk->sk_user_data;
292 sk->sk_user_data = NULL;
294 write_unlock_bh(&sk->sk_callback_lock);
297 /* wait for accepts to stop and close the socket */
298 flush_workqueue(rds_wq);