net/rds/tcp_listen.c

   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/gfp.h>
  35 #include <linux/in.h>
  36 #include <net/tcp.h>
  37
  38 #include "rds.h"
  39 #include "tcp.h"
  40
  41 /*
  42  * cheesy, but simple..
  43  */
  44 static void rds_tcp_accept_worker(struct work_struct *work);
  45 static DECLARE_WORK(rds_tcp_listen_work, rds_tcp_accept_worker);
  46 static struct socket *rds_tcp_listen_sock;
  47
  48 static int rds_tcp_keepalive(struct socket *sock)
  49 {
  50         /* values below based on xs_udp_default_timeout */
  51         int keepidle = 5; /* send a probe 'keepidle' secs after last data */
  52         int keepcnt = 5; /* number of unack'ed probes before declaring dead */
  53         int keepalive = 1;
  54         int ret = 0;
  55
  56         ret = kernel_setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE,
  57                                 (char *)&keepalive, sizeof(keepalive));
  58         if (ret < 0)
  59                 goto bail;
  60
  61         ret = kernel_setsockopt(sock, IPPROTO_TCP, TCP_KEEPCNT,
  62                                 (char *)&keepcnt, sizeof(keepcnt));
  63         if (ret < 0)
  64                 goto bail;
  65
  66         ret = kernel_setsockopt(sock, IPPROTO_TCP, TCP_KEEPIDLE,
  67                                 (char *)&keepidle, sizeof(keepidle));
  68         if (ret < 0)
  69                 goto bail;
  70
  71         /* KEEPINTVL is the interval between successive probes. We follow
  72          * the model in xs_tcp_finish_connecting() and re-use keepidle.
  73          */
  74         ret = kernel_setsockopt(sock, IPPROTO_TCP, TCP_KEEPINTVL,
  75                                 (char *)&keepidle, sizeof(keepidle));
  76 bail:
  77         return ret;
  78 }
  79
  80 static int rds_tcp_accept_one(struct socket *sock)
  81 {
  82         struct socket *new_sock = NULL;
  83         struct rds_connection *conn;
  84         int ret;
  85         struct inet_sock *inet;
  86         struct rds_tcp_connection *rs_tcp;
  87
  88         ret = sock_create_kern(sock_net(sock->sk), sock->sk->sk_family,
  89                                sock->sk->sk_type, sock->sk->sk_protocol,
  90                                &new_sock);
  91         if (ret)
  92                 goto out;
  93
  94         new_sock->type = sock->type;
  95         new_sock->ops = sock->ops;
  96         ret = sock->ops->accept(sock, new_sock, O_NONBLOCK);
  97         if (ret < 0)
  98                 goto out;
  99
 100         ret = rds_tcp_keepalive(new_sock);
 101         if (ret < 0)
 102                 goto out;
 103
 104         rds_tcp_tune(new_sock);
 105
 106         inet = inet_sk(new_sock->sk);
 107
 108         rdsdebug("accepted tcp %pI4:%u -> %pI4:%u\n",
 109                  &inet->inet_saddr, ntohs(inet->inet_sport),
 110                  &inet->inet_daddr, ntohs(inet->inet_dport));
 111
 112         conn = rds_conn_create(sock_net(sock->sk),
 113                                inet->inet_saddr, inet->inet_daddr,
 114                                &rds_tcp_transport, GFP_KERNEL);
 115         if (IS_ERR(conn)) {
 116                 ret = PTR_ERR(conn);
 117                 goto out;
 118         }
 119         /* An incoming SYN request came in, and TCP just accepted it.
 120          * We always create a new conn for listen side of TCP, and do not
 121          * add it to the c_hash_list.
 122          *
 123          * If the client reboots, this conn will need to be cleaned up.
 124          * rds_tcp_state_change() will do that cleanup
 125          */
 126         rs_tcp = (struct rds_tcp_connection *)conn->c_transport_data;
 127         WARN_ON(!rs_tcp || rs_tcp->t_sock);
 128
 129         /*
 130          * see the comment above rds_queue_delayed_reconnect()
 131          */
 132         if (!rds_conn_transition(conn, RDS_CONN_DOWN, RDS_CONN_CONNECTING)) {
 133                 if (rds_conn_state(conn) == RDS_CONN_UP)
 134                         rds_tcp_stats_inc(s_tcp_listen_closed_stale);
 135                 else
 136                         rds_tcp_stats_inc(s_tcp_connect_raced);
 137                 rds_conn_drop(conn);
 138                 ret = 0;
 139                 goto out;
 140         }
 141
 142         rds_tcp_set_callbacks(new_sock, conn);
 143         rds_connect_complete(conn);
 144         new_sock = NULL;
 145         ret = 0;
 146
 147 out:
 148         if (new_sock)
 149                 sock_release(new_sock);
 150         return ret;
 151 }
 152
 153 static void rds_tcp_accept_worker(struct work_struct *work)
 154 {
 155         while (rds_tcp_accept_one(rds_tcp_listen_sock) == 0)
 156                 cond_resched();
 157 }
 158
 159 void rds_tcp_listen_data_ready(struct sock *sk)
 160 {
 161         void (*ready)(struct sock *sk);
 162
 163         rdsdebug("listen data ready sk %p\n", sk);
 164
 165         read_lock(&sk->sk_callback_lock);
 166         ready = sk->sk_user_data;
 167         if (!ready) { /* check for teardown race */
 168                 ready = sk->sk_data_ready;
 169                 goto out;
 170         }
 171
 172         /*
 173          * ->sk_data_ready is also called for a newly established child socket
 174          * before it has been accepted and the accepter has set up their
 175          * data_ready.. we only want to queue listen work for our listening
 176          * socket
 177          */
 178         if (sk->sk_state == TCP_LISTEN)
 179                 queue_work(rds_wq, &rds_tcp_listen_work);
 180
 181 out:
 182         read_unlock(&sk->sk_callback_lock);
 183         ready(sk);
 184 }
 185
 186 int rds_tcp_listen_init(void)
 187 {
 188         struct sockaddr_in sin;
 189         struct socket *sock = NULL;
 190         int ret;
 191
 192         /* MUST call sock_create_kern directly so that we avoid get_net()
 193          * in sk_alloc(). Doing a get_net() will result in cleanup_net()
 194          * never getting invoked, which will leave sock and other things
 195          * in limbo.
 196          */
 197         ret = sock_create_kern(current->nsproxy->net_ns, PF_INET,
 198                                SOCK_STREAM, IPPROTO_TCP, &sock);
 199         if (ret < 0)
 200                 goto out;
 201
 202         sock->sk->sk_reuse = SK_CAN_REUSE;
 203         rds_tcp_nonagle(sock);
 204
 205         write_lock_bh(&sock->sk->sk_callback_lock);
 206         sock->sk->sk_user_data = sock->sk->sk_data_ready;
 207         sock->sk->sk_data_ready = rds_tcp_listen_data_ready;
 208         write_unlock_bh(&sock->sk->sk_callback_lock);
 209
 210         sin.sin_family = PF_INET;
 211         sin.sin_addr.s_addr = (__force u32)htonl(INADDR_ANY);
 212         sin.sin_port = (__force u16)htons(RDS_TCP_PORT);
 213
 214         ret = sock->ops->bind(sock, (struct sockaddr *)&sin, sizeof(sin));
 215         if (ret < 0)
 216                 goto out;
 217
 218         ret = sock->ops->listen(sock, 64);
 219         if (ret < 0)
 220                 goto out;
 221
 222         rds_tcp_listen_sock = sock;
 223         sock = NULL;
 224 out:
 225         if (sock)
 226                 sock_release(sock);
 227         return ret;
 228 }
 229
 230 void rds_tcp_listen_stop(void)
 231 {
 232         struct socket *sock = rds_tcp_listen_sock;
 233         struct sock *sk;
 234
 235         if (!sock)
 236                 return;
 237
 238         sk = sock->sk;
 239
 240         /* serialize with and prevent further callbacks */
 241         lock_sock(sk);
 242         write_lock_bh(&sk->sk_callback_lock);
 243         if (sk->sk_user_data) {
 244                 sk->sk_data_ready = sk->sk_user_data;
 245                 sk->sk_user_data = NULL;
 246         }
 247         write_unlock_bh(&sk->sk_callback_lock);
 248         release_sock(sk);
 249
 250         /* wait for accepts to stop and close the socket */
 251         flush_workqueue(rds_wq);
 252         sock_release(sock);
 253         rds_tcp_listen_sock = NULL;
 254 }