net/rds/tcp.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/slab.h>
  35 #include <linux/in.h>
  36 #include <linux/module.h>
  37 #include <net/tcp.h>
  38
  39 #include "rds.h"
  40 #include "tcp.h"
  41
  42 /* only for info exporting */
  43 static DEFINE_SPINLOCK(rds_tcp_tc_list_lock);
  44 static LIST_HEAD(rds_tcp_tc_list);
  45 static unsigned int rds_tcp_tc_count;
  46
  47 /* Track rds_tcp_connection structs so they can be cleaned up */
  48 static DEFINE_SPINLOCK(rds_tcp_conn_lock);
  49 static LIST_HEAD(rds_tcp_conn_list);
  50
  51 static struct kmem_cache *rds_tcp_conn_slab;
  52
  53 #define RDS_TCP_DEFAULT_BUFSIZE (128 * 1024)
  54
  55 /* doing it this way avoids calling tcp_sk() */
  56 void rds_tcp_nonagle(struct socket *sock)
  57 {
  58         mm_segment_t oldfs = get_fs();
  59         int val = 1;
  60
  61         set_fs(KERNEL_DS);
  62         sock->ops->setsockopt(sock, SOL_TCP, TCP_NODELAY, (char __user *)&val,
  63                               sizeof(val));
  64         set_fs(oldfs);
  65 }
  66
  67 void rds_tcp_tune(struct socket *sock)
  68 {
  69         struct sock *sk = sock->sk;
  70
  71         rds_tcp_nonagle(sock);
  72
  73         /*
  74          * We're trying to saturate gigabit with the default,
  75          * see svc_sock_setbufsize().
  76          */
  77         lock_sock(sk);
  78         sk->sk_sndbuf = RDS_TCP_DEFAULT_BUFSIZE;
  79         sk->sk_rcvbuf = RDS_TCP_DEFAULT_BUFSIZE;
  80         sk->sk_userlocks |= SOCK_SNDBUF_LOCK|SOCK_RCVBUF_LOCK;
  81         release_sock(sk);
  82 }
  83
  84 u32 rds_tcp_snd_nxt(struct rds_tcp_connection *tc)
  85 {
  86         return tcp_sk(tc->t_sock->sk)->snd_nxt;
  87 }
  88
  89 u32 rds_tcp_snd_una(struct rds_tcp_connection *tc)
  90 {
  91         return tcp_sk(tc->t_sock->sk)->snd_una;
  92 }
  93
  94 void rds_tcp_restore_callbacks(struct socket *sock,
  95                                struct rds_tcp_connection *tc)
  96 {
  97         rdsdebug("restoring sock %p callbacks from tc %p\n", sock, tc);
  98         write_lock_bh(&sock->sk->sk_callback_lock);
  99
 100         /* done under the callback_lock to serialize with write_space */
 101         spin_lock(&rds_tcp_tc_list_lock);
 102         list_del_init(&tc->t_list_item);
 103         rds_tcp_tc_count--;
 104         spin_unlock(&rds_tcp_tc_list_lock);
 105
 106         tc->t_sock = NULL;
 107
 108         sock->sk->sk_write_space = tc->t_orig_write_space;
 109         sock->sk->sk_data_ready = tc->t_orig_data_ready;
 110         sock->sk->sk_state_change = tc->t_orig_state_change;
 111         sock->sk->sk_user_data = NULL;
 112
 113         write_unlock_bh(&sock->sk->sk_callback_lock);
 114 }
 115
 116 /*
 117  * This is the only path that sets tc->t_sock.  Send and receive trust that
 118  * it is set.  The RDS_CONN_CONNECTED bit protects those paths from being
 119  * called while it isn't set.
 120  */
 121 void rds_tcp_set_callbacks(struct socket *sock, struct rds_connection *conn)
 122 {
 123         struct rds_tcp_connection *tc = conn->c_transport_data;
 124
 125         rdsdebug("setting sock %p callbacks to tc %p\n", sock, tc);
 126         write_lock_bh(&sock->sk->sk_callback_lock);
 127
 128         /* done under the callback_lock to serialize with write_space */
 129         spin_lock(&rds_tcp_tc_list_lock);
 130         list_add_tail(&tc->t_list_item, &rds_tcp_tc_list);
 131         rds_tcp_tc_count++;
 132         spin_unlock(&rds_tcp_tc_list_lock);
 133
 134         /* accepted sockets need our listen data ready undone */
 135         if (sock->sk->sk_data_ready == rds_tcp_listen_data_ready)
 136                 sock->sk->sk_data_ready = sock->sk->sk_user_data;
 137
 138         tc->t_sock = sock;
 139         tc->conn = conn;
 140         tc->t_orig_data_ready = sock->sk->sk_data_ready;
 141         tc->t_orig_write_space = sock->sk->sk_write_space;
 142         tc->t_orig_state_change = sock->sk->sk_state_change;
 143
 144         sock->sk->sk_user_data = conn;
 145         sock->sk->sk_data_ready = rds_tcp_data_ready;
 146         sock->sk->sk_write_space = rds_tcp_write_space;
 147         sock->sk->sk_state_change = rds_tcp_state_change;
 148
 149         write_unlock_bh(&sock->sk->sk_callback_lock);
 150 }
 151
 152 static void rds_tcp_tc_info(struct socket *sock, unsigned int len,
 153                             struct rds_info_iterator *iter,
 154                             struct rds_info_lengths *lens)
 155 {
 156         struct rds_info_tcp_socket tsinfo;
 157         struct rds_tcp_connection *tc;
 158         unsigned long flags;
 159         struct sockaddr_in sin;
 160         int sinlen;
 161
 162         spin_lock_irqsave(&rds_tcp_tc_list_lock, flags);
 163
 164         if (len / sizeof(tsinfo) < rds_tcp_tc_count)
 165                 goto out;
 166
 167         list_for_each_entry(tc, &rds_tcp_tc_list, t_list_item) {
 168
 169                 sock->ops->getname(sock, (struct sockaddr *)&sin, &sinlen, 0);
 170                 tsinfo.local_addr = sin.sin_addr.s_addr;
 171                 tsinfo.local_port = sin.sin_port;
 172                 sock->ops->getname(sock, (struct sockaddr *)&sin, &sinlen, 1);
 173                 tsinfo.peer_addr = sin.sin_addr.s_addr;
 174                 tsinfo.peer_port = sin.sin_port;
 175
 176                 tsinfo.hdr_rem = tc->t_tinc_hdr_rem;
 177                 tsinfo.data_rem = tc->t_tinc_data_rem;
 178                 tsinfo.last_sent_nxt = tc->t_last_sent_nxt;
 179                 tsinfo.last_expected_una = tc->t_last_expected_una;
 180                 tsinfo.last_seen_una = tc->t_last_seen_una;
 181
 182                 rds_info_copy(iter, &tsinfo, sizeof(tsinfo));
 183         }
 184
 185 out:
 186         lens->nr = rds_tcp_tc_count;
 187         lens->each = sizeof(tsinfo);
 188
 189         spin_unlock_irqrestore(&rds_tcp_tc_list_lock, flags);
 190 }
 191
 192 static int rds_tcp_laddr_check(__be32 addr)
 193 {
 194         if (inet_addr_type(&init_net, addr) == RTN_LOCAL)
 195                 return 0;
 196         return -EADDRNOTAVAIL;
 197 }
 198
 199 static int rds_tcp_conn_alloc(struct rds_connection *conn, gfp_t gfp)
 200 {
 201         struct rds_tcp_connection *tc;
 202
 203         tc = kmem_cache_alloc(rds_tcp_conn_slab, gfp);
 204         if (!tc)
 205                 return -ENOMEM;
 206
 207         tc->t_sock = NULL;
 208         tc->t_tinc = NULL;
 209         tc->t_tinc_hdr_rem = sizeof(struct rds_header);
 210         tc->t_tinc_data_rem = 0;
 211
 212         conn->c_transport_data = tc;
 213
 214         spin_lock_irq(&rds_tcp_conn_lock);
 215         list_add_tail(&tc->t_tcp_node, &rds_tcp_conn_list);
 216         spin_unlock_irq(&rds_tcp_conn_lock);
 217
 218         rdsdebug("alloced tc %p\n", conn->c_transport_data);
 219         return 0;
 220 }
 221
 222 static void rds_tcp_conn_free(void *arg)
 223 {
 224         struct rds_tcp_connection *tc = arg;
 225         unsigned long flags;
 226         rdsdebug("freeing tc %p\n", tc);
 227
 228         spin_lock_irqsave(&rds_tcp_conn_lock, flags);
 229         list_del(&tc->t_tcp_node);
 230         spin_unlock_irqrestore(&rds_tcp_conn_lock, flags);
 231
 232         kmem_cache_free(rds_tcp_conn_slab, tc);
 233 }
 234
 235 static void rds_tcp_destroy_conns(void)
 236 {
 237         struct rds_tcp_connection *tc, *_tc;
 238         LIST_HEAD(tmp_list);
 239
 240         /* avoid calling conn_destroy with irqs off */
 241         spin_lock_irq(&rds_tcp_conn_lock);
 242         list_splice(&rds_tcp_conn_list, &tmp_list);
 243         INIT_LIST_HEAD(&rds_tcp_conn_list);
 244         spin_unlock_irq(&rds_tcp_conn_lock);
 245
 246         list_for_each_entry_safe(tc, _tc, &tmp_list, t_tcp_node) {
 247                 if (tc->conn->c_passive)
 248                         rds_conn_destroy(tc->conn->c_passive);
 249                 rds_conn_destroy(tc->conn);
 250         }
 251 }
 252
 253 static void rds_tcp_exit(void)
 254 {
 255         rds_info_deregister_func(RDS_INFO_TCP_SOCKETS, rds_tcp_tc_info);
 256         rds_tcp_listen_stop();
 257         rds_tcp_destroy_conns();
 258         rds_trans_unregister(&rds_tcp_transport);
 259         rds_tcp_recv_exit();
 260         kmem_cache_destroy(rds_tcp_conn_slab);
 261 }
 262 module_exit(rds_tcp_exit);
 263
 264 struct rds_transport rds_tcp_transport = {
 265         .laddr_check            = rds_tcp_laddr_check,
 266         .xmit_prepare           = rds_tcp_xmit_prepare,
 267         .xmit_complete          = rds_tcp_xmit_complete,
 268         .xmit                   = rds_tcp_xmit,
 269         .recv                   = rds_tcp_recv,
 270         .conn_alloc             = rds_tcp_conn_alloc,
 271         .conn_free              = rds_tcp_conn_free,
 272         .conn_connect           = rds_tcp_conn_connect,
 273         .conn_shutdown          = rds_tcp_conn_shutdown,
 274         .inc_copy_to_user       = rds_tcp_inc_copy_to_user,
 275         .inc_free               = rds_tcp_inc_free,
 276         .stats_info_copy        = rds_tcp_stats_info_copy,
 277         .exit                   = rds_tcp_exit,
 278         .t_owner                = THIS_MODULE,
 279         .t_name                 = "tcp",
 280         .t_type                 = RDS_TRANS_TCP,
 281         .t_prefer_loopback      = 1,
 282 };
 283
 284 static int rds_tcp_init(void)
 285 {
 286         int ret;
 287
 288         rds_tcp_conn_slab = kmem_cache_create("rds_tcp_connection",
 289                                               sizeof(struct rds_tcp_connection),
 290                                               0, 0, NULL);
 291         if (!rds_tcp_conn_slab) {
 292                 ret = -ENOMEM;
 293                 goto out;
 294         }
 295
 296         ret = rds_tcp_recv_init();
 297         if (ret)
 298                 goto out_slab;
 299
 300         ret = rds_trans_register(&rds_tcp_transport);
 301         if (ret)
 302                 goto out_recv;
 303
 304         ret = rds_tcp_listen_init();
 305         if (ret)
 306                 goto out_register;
 307
 308         rds_info_register_func(RDS_INFO_TCP_SOCKETS, rds_tcp_tc_info);
 309
 310         goto out;
 311
 312 out_register:
 313         rds_trans_unregister(&rds_tcp_transport);
 314 out_recv:
 315         rds_tcp_recv_exit();
 316 out_slab:
 317         kmem_cache_destroy(rds_tcp_conn_slab);
 318 out:
 319         return ret;
 320 }
 321 module_init(rds_tcp_init);
 322
 323 MODULE_AUTHOR("Oracle Corporation <rds-devel@oss.oracle.com>");
 324 MODULE_DESCRIPTION("RDS: TCP transport");
 325 MODULE_LICENSE("Dual BSD/GPL");
 326