LXR linux/net/rds/tcp

   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
  41int rds_tcp_keepalive(struct socket *sock)
  42{
  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 */
  46        int keepalive = 1;
  47        int ret = 0;
  48
  49        ret = kernel_setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE,
  50                                (char *)&keepalive, sizeof(keepalive));
  51        if (ret < 0)
  52                goto bail;
  53
  54        ret = kernel_setsockopt(sock, IPPROTO_TCP, TCP_KEEPCNT,
  55                                (char *)&keepcnt, sizeof(keepcnt));
  56        if (ret < 0)
  57                goto bail;
  58
  59        ret = kernel_setsockopt(sock, IPPROTO_TCP, TCP_KEEPIDLE,
  60                                (char *)&keepidle, sizeof(keepidle));
  61        if (ret < 0)
  62                goto bail;
  63
  64        /* KEEPINTVL is the interval between successive probes. We follow
  65         * the model in xs_tcp_finish_connecting() and re-use keepidle.
  66         */
  67        ret = kernel_setsockopt(sock, IPPROTO_TCP, TCP_KEEPINTVL,
  68                                (char *)&keepidle, sizeof(keepidle));
  69bail:
  70        return ret;
  71}
  72
  73int rds_tcp_accept_one(struct socket *sock)
  74{
  75        struct socket *new_sock = NULL;
  76        struct rds_connection *conn;
  77        int ret;
  78        struct inet_sock *inet;
  79        struct rds_tcp_connection *rs_tcp = NULL;
  80        int conn_state;
  81        struct sock *nsk;
  82
  83        ret = sock_create_kern(sock_net(sock->sk), sock->sk->sk_family,
  84                               sock->sk->sk_type, sock->sk->sk_protocol,
  85                               &new_sock);
  86        if (ret)
  87                goto out;
  88
  89        new_sock->type = sock->type;
  90        new_sock->ops = sock->ops;
  91        ret = sock->ops->accept(sock, new_sock, O_NONBLOCK);
  92        if (ret < 0)
  93                goto out;
  94
  95        ret = rds_tcp_keepalive(new_sock);
  96        if (ret < 0)
  97                goto out;
  98
  99        rds_tcp_tune(new_sock);
 100
 101        inet = inet_sk(new_sock->sk);
 102
 103        rdsdebug("accepted tcp %pI4:%u -> %pI4:%u\n",
 104                 &inet->inet_saddr, ntohs(inet->inet_sport),
 105                 &inet->inet_daddr, ntohs(inet->inet_dport));
 106
 107        conn = rds_conn_create(sock_net(sock->sk),
 108                               inet->inet_saddr, inet->inet_daddr,
 109                               &rds_tcp_transport, GFP_KERNEL);
 110        if (IS_ERR(conn)) {
 111                ret = PTR_ERR(conn);
 112                goto out;
 113        }
 114        /* An incoming SYN request came in, and TCP just accepted it.
 115         *
 116         * If the client reboots, this conn will need to be cleaned up.
 117         * rds_tcp_state_change() will do that cleanup
 118         */
 119        rs_tcp = (struct rds_tcp_connection *)conn->c_transport_data;
 120        rds_conn_transition(conn, RDS_CONN_DOWN, RDS_CONN_CONNECTING);
 121        mutex_lock(&rs_tcp->t_conn_lock);
 122        conn_state = rds_conn_state(conn);
 123        if (conn_state != RDS_CONN_CONNECTING && conn_state != RDS_CONN_UP)
 124                goto rst_nsk;
 125        if (rs_tcp->t_sock) {
 126                /* Need to resolve a duelling SYN between peers.
 127                 * We have an outstanding SYN to this peer, which may
 128                 * potentially have transitioned to the RDS_CONN_UP state,
 129                 * so we must quiesce any send threads before resetting
 130                 * c_transport_data.
 131                 */
 132                wait_event(conn->c_waitq,
 133                           !test_bit(RDS_IN_XMIT, &conn->c_flags));
 134                if (ntohl(inet->inet_saddr) < ntohl(inet->inet_daddr)) {
 135                        goto rst_nsk;
 136                } else if (rs_tcp->t_sock) {
 137                        rds_tcp_restore_callbacks(rs_tcp->t_sock, rs_tcp);
 138                        conn->c_outgoing = 0;
 139                }
 140        }
 141        rds_tcp_set_callbacks(new_sock, conn);
 142        rds_connect_complete(conn); /* marks RDS_CONN_UP */
 143        new_sock = NULL;
 144        ret = 0;
 145        goto out;
 146rst_nsk:
 147        /* reset the newly returned accept sock and bail */
 148        nsk = new_sock->sk;
 149        rds_tcp_stats_inc(s_tcp_listen_closed_stale);
 150        nsk->sk_user_data = NULL;
 151        nsk->sk_prot->disconnect(nsk, 0);
 152        tcp_done(nsk);
 153        new_sock = NULL;
 154        ret = 0;
 155out:
 156        if (rs_tcp)
 157                mutex_unlock(&rs_tcp->t_conn_lock);
 158        if (new_sock)
 159                sock_release(new_sock);
 160        return ret;
 161}
 162
 163void rds_tcp_listen_data_ready(struct sock *sk)
 164{
 165        void (*ready)(struct sock *sk);
 166
 167        rdsdebug("listen data ready sk %p\n", sk);
 168
 169        read_lock(&sk->sk_callback_lock);
 170        ready = sk->sk_user_data;
 171        if (!ready) { /* check for teardown race */
 172                ready = sk->sk_data_ready;
 173                goto out;
 174        }
 175
 176        /*
 177         * ->sk_data_ready is also called for a newly established child socket
 178         * before it has been accepted and the accepter has set up their
 179         * data_ready.. we only want to queue listen work for our listening
 180         * socket
 181         */
 182        if (sk->sk_state == TCP_LISTEN)
 183                rds_tcp_accept_work(sk);
 184
 185out:
 186        read_unlock(&sk->sk_callback_lock);
 187        ready(sk);
 188}
 189
 190struct socket *rds_tcp_listen_init(struct net *net)
 191{
 192        struct sockaddr_in sin;
 193        struct socket *sock = NULL;
 194        int ret;
 195
 196        ret = sock_create_kern(net, PF_INET, SOCK_STREAM, IPPROTO_TCP, &sock);
 197        if (ret < 0)
 198                goto out;
 199
 200        sock->sk->sk_reuse = SK_CAN_REUSE;
 201        rds_tcp_nonagle(sock);
 202
 203        write_lock_bh(&sock->sk->sk_callback_lock);
 204        sock->sk->sk_user_data = sock->sk->sk_data_ready;
 205        sock->sk->sk_data_ready = rds_tcp_listen_data_ready;
 206        write_unlock_bh(&sock->sk->sk_callback_lock);
 207
 208        sin.sin_family = PF_INET;
 209        sin.sin_addr.s_addr = (__force u32)htonl(INADDR_ANY);
 210        sin.sin_port = (__force u16)htons(RDS_TCP_PORT);
 211
 212        ret = sock->ops->bind(sock, (struct sockaddr *)&sin, sizeof(sin));
 213        if (ret < 0)
 214                goto out;
 215
 216        ret = sock->ops->listen(sock, 64);
 217        if (ret < 0)
 218                goto out;
 219
 220        return sock;
 221out:
 222        if (sock)
 223                sock_release(sock);
 224        return NULL;
 225}
 226
 227void rds_tcp_listen_stop(struct socket *sock)
 228{
 229        struct sock *sk;
 230
 231        if (!sock)
 232                return;
 233
 234        sk = sock->sk;
 235
 236        /* serialize with and prevent further callbacks */
 237        lock_sock(sk);
 238        write_lock_bh(&sk->sk_callback_lock);
 239        if (sk->sk_user_data) {
 240                sk->sk_data_ready = sk->sk_user_data;
 241                sk->sk_user_data = NULL;
 242        }
 243        write_unlock_bh(&sk->sk_callback_lock);
 244        release_sock(sk);
 245
 246        /* wait for accepts to stop and close the socket */
 247        flush_workqueue(rds_wq);
 248        sock_release(sock);
 249}
 250