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 #include <net/net_namespace.h>
  39 #include <net/netns/generic.h>
  40
  41 #include "rds_single_path.h"
  42 #include "rds.h"
  43 #include "tcp.h"
  44
  45 /* only for info exporting */
  46 static DEFINE_SPINLOCK(rds_tcp_tc_list_lock);
  47 static LIST_HEAD(rds_tcp_tc_list);
  48 static unsigned int rds_tcp_tc_count;
  49
  50 /* Track rds_tcp_connection structs so they can be cleaned up */
  51 static DEFINE_SPINLOCK(rds_tcp_conn_lock);
  52 static LIST_HEAD(rds_tcp_conn_list);
  53
  54 static struct kmem_cache *rds_tcp_conn_slab;
  55
  56 static int rds_tcp_skbuf_handler(struct ctl_table *ctl, int write,
  57                                  void __user *buffer, size_t *lenp,
  58                                  loff_t *fpos);
  59
  60 static int rds_tcp_min_sndbuf = SOCK_MIN_SNDBUF;
  61 static int rds_tcp_min_rcvbuf = SOCK_MIN_RCVBUF;
  62
  63 static struct ctl_table rds_tcp_sysctl_table[] = {
  64 #define RDS_TCP_SNDBUF  0
  65         {
  66                 .procname       = "rds_tcp_sndbuf",
  67                 /* data is per-net pointer */
  68                 .maxlen         = sizeof(int),
  69                 .mode           = 0644,
  70                 .proc_handler   = rds_tcp_skbuf_handler,
  71                 .extra1         = &rds_tcp_min_sndbuf,
  72         },
  73 #define RDS_TCP_RCVBUF  1
  74         {
  75                 .procname       = "rds_tcp_rcvbuf",
  76                 /* data is per-net pointer */
  77                 .maxlen         = sizeof(int),
  78                 .mode           = 0644,
  79                 .proc_handler   = rds_tcp_skbuf_handler,
  80                 .extra1         = &rds_tcp_min_rcvbuf,
  81         },
  82         { }
  83 };
  84
  85 /* doing it this way avoids calling tcp_sk() */
  86 void rds_tcp_nonagle(struct socket *sock)
  87 {
  88         mm_segment_t oldfs = get_fs();
  89         int val = 1;
  90
  91         set_fs(KERNEL_DS);
  92         sock->ops->setsockopt(sock, SOL_TCP, TCP_NODELAY, (char __user *)&val,
  93                               sizeof(val));
  94         set_fs(oldfs);
  95 }
  96
  97 u32 rds_tcp_snd_nxt(struct rds_tcp_connection *tc)
  98 {
  99         return tcp_sk(tc->t_sock->sk)->snd_nxt;
 100 }
 101
 102 u32 rds_tcp_snd_una(struct rds_tcp_connection *tc)
 103 {
 104         return tcp_sk(tc->t_sock->sk)->snd_una;
 105 }
 106
 107 void rds_tcp_restore_callbacks(struct socket *sock,
 108                                struct rds_tcp_connection *tc)
 109 {
 110         rdsdebug("restoring sock %p callbacks from tc %p\n", sock, tc);
 111         write_lock_bh(&sock->sk->sk_callback_lock);
 112
 113         /* done under the callback_lock to serialize with write_space */
 114         spin_lock(&rds_tcp_tc_list_lock);
 115         list_del_init(&tc->t_list_item);
 116         rds_tcp_tc_count--;
 117         spin_unlock(&rds_tcp_tc_list_lock);
 118
 119         tc->t_sock = NULL;
 120
 121         sock->sk->sk_write_space = tc->t_orig_write_space;
 122         sock->sk->sk_data_ready = tc->t_orig_data_ready;
 123         sock->sk->sk_state_change = tc->t_orig_state_change;
 124         sock->sk->sk_user_data = NULL;
 125
 126         write_unlock_bh(&sock->sk->sk_callback_lock);
 127 }
 128
 129 /*
 130  * rds_tcp_reset_callbacks() switches the to the new sock and
 131  * returns the existing tc->t_sock.
 132  *
 133  * The only functions that set tc->t_sock are rds_tcp_set_callbacks
 134  * and rds_tcp_reset_callbacks.  Send and receive trust that
 135  * it is set.  The absence of RDS_CONN_UP bit protects those paths
 136  * from being called while it isn't set.
 137  */
 138 void rds_tcp_reset_callbacks(struct socket *sock,
 139                              struct rds_connection *conn)
 140 {
 141         struct rds_tcp_connection *tc = conn->c_transport_data;
 142         struct socket *osock = tc->t_sock;
 143
 144         if (!osock)
 145                 goto newsock;
 146
 147         /* Need to resolve a duelling SYN between peers.
 148          * We have an outstanding SYN to this peer, which may
 149          * potentially have transitioned to the RDS_CONN_UP state,
 150          * so we must quiesce any send threads before resetting
 151          * c_transport_data. We quiesce these threads by setting
 152          * c_state to something other than RDS_CONN_UP, and then
 153          * waiting for any existing threads in rds_send_xmit to
 154          * complete release_in_xmit(). (Subsequent threads entering
 155          * rds_send_xmit() will bail on !rds_conn_up().
 156          *
 157          * However an incoming syn-ack at this point would end up
 158          * marking the conn as RDS_CONN_UP, and would again permit
 159          * rds_send_xmi() threads through, so ideally we would
 160          * synchronize on RDS_CONN_UP after lock_sock(), but cannot
 161          * do that: waiting on !RDS_IN_XMIT after lock_sock() may
 162          * end up deadlocking with tcp_sendmsg(), and the RDS_IN_XMIT
 163          * would not get set. As a result, we set c_state to
 164          * RDS_CONN_RESETTTING, to ensure that rds_tcp_state_change
 165          * cannot mark rds_conn_path_up() in the window before lock_sock()
 166          */
 167         atomic_set(&conn->c_state, RDS_CONN_RESETTING);
 168         wait_event(conn->c_waitq, !test_bit(RDS_IN_XMIT, &conn->c_flags));
 169         lock_sock(osock->sk);
 170         /* reset receive side state for rds_tcp_data_recv() for osock  */
 171         if (tc->t_tinc) {
 172                 rds_inc_put(&tc->t_tinc->ti_inc);
 173                 tc->t_tinc = NULL;
 174         }
 175         tc->t_tinc_hdr_rem = sizeof(struct rds_header);
 176         tc->t_tinc_data_rem = 0;
 177         tc->t_sock = NULL;
 178
 179         write_lock_bh(&osock->sk->sk_callback_lock);
 180
 181         osock->sk->sk_user_data = NULL;
 182         osock->sk->sk_data_ready = tc->t_orig_data_ready;
 183         osock->sk->sk_write_space = tc->t_orig_write_space;
 184         osock->sk->sk_state_change = tc->t_orig_state_change;
 185         write_unlock_bh(&osock->sk->sk_callback_lock);
 186         release_sock(osock->sk);
 187         sock_release(osock);
 188 newsock:
 189         rds_send_path_reset(&conn->c_path[0]);
 190         lock_sock(sock->sk);
 191         write_lock_bh(&sock->sk->sk_callback_lock);
 192         tc->t_sock = sock;
 193         sock->sk->sk_user_data = conn;
 194         sock->sk->sk_data_ready = rds_tcp_data_ready;
 195         sock->sk->sk_write_space = rds_tcp_write_space;
 196         sock->sk->sk_state_change = rds_tcp_state_change;
 197
 198         write_unlock_bh(&sock->sk->sk_callback_lock);
 199         release_sock(sock->sk);
 200 }
 201
 202 /* Add tc to rds_tcp_tc_list and set tc->t_sock. See comments
 203  * above rds_tcp_reset_callbacks for notes about synchronization
 204  * with data path
 205  */
 206 void rds_tcp_set_callbacks(struct socket *sock, struct rds_connection *conn)
 207 {
 208         struct rds_tcp_connection *tc = conn->c_transport_data;
 209
 210         rdsdebug("setting sock %p callbacks to tc %p\n", sock, tc);
 211         write_lock_bh(&sock->sk->sk_callback_lock);
 212
 213         /* done under the callback_lock to serialize with write_space */
 214         spin_lock(&rds_tcp_tc_list_lock);
 215         list_add_tail(&tc->t_list_item, &rds_tcp_tc_list);
 216         rds_tcp_tc_count++;
 217         spin_unlock(&rds_tcp_tc_list_lock);
 218
 219         /* accepted sockets need our listen data ready undone */
 220         if (sock->sk->sk_data_ready == rds_tcp_listen_data_ready)
 221                 sock->sk->sk_data_ready = sock->sk->sk_user_data;
 222
 223         tc->t_sock = sock;
 224         tc->conn = conn;
 225         tc->t_orig_data_ready = sock->sk->sk_data_ready;
 226         tc->t_orig_write_space = sock->sk->sk_write_space;
 227         tc->t_orig_state_change = sock->sk->sk_state_change;
 228
 229         sock->sk->sk_user_data = conn;
 230         sock->sk->sk_data_ready = rds_tcp_data_ready;
 231         sock->sk->sk_write_space = rds_tcp_write_space;
 232         sock->sk->sk_state_change = rds_tcp_state_change;
 233
 234         write_unlock_bh(&sock->sk->sk_callback_lock);
 235 }
 236
 237 static void rds_tcp_tc_info(struct socket *sock, unsigned int len,
 238                             struct rds_info_iterator *iter,
 239                             struct rds_info_lengths *lens)
 240 {
 241         struct rds_info_tcp_socket tsinfo;
 242         struct rds_tcp_connection *tc;
 243         unsigned long flags;
 244         struct sockaddr_in sin;
 245         int sinlen;
 246
 247         spin_lock_irqsave(&rds_tcp_tc_list_lock, flags);
 248
 249         if (len / sizeof(tsinfo) < rds_tcp_tc_count)
 250                 goto out;
 251
 252         list_for_each_entry(tc, &rds_tcp_tc_list, t_list_item) {
 253
 254                 sock->ops->getname(sock, (struct sockaddr *)&sin, &sinlen, 0);
 255                 tsinfo.local_addr = sin.sin_addr.s_addr;
 256                 tsinfo.local_port = sin.sin_port;
 257                 sock->ops->getname(sock, (struct sockaddr *)&sin, &sinlen, 1);
 258                 tsinfo.peer_addr = sin.sin_addr.s_addr;
 259                 tsinfo.peer_port = sin.sin_port;
 260
 261                 tsinfo.hdr_rem = tc->t_tinc_hdr_rem;
 262                 tsinfo.data_rem = tc->t_tinc_data_rem;
 263                 tsinfo.last_sent_nxt = tc->t_last_sent_nxt;
 264                 tsinfo.last_expected_una = tc->t_last_expected_una;
 265                 tsinfo.last_seen_una = tc->t_last_seen_una;
 266
 267                 rds_info_copy(iter, &tsinfo, sizeof(tsinfo));
 268         }
 269
 270 out:
 271         lens->nr = rds_tcp_tc_count;
 272         lens->each = sizeof(tsinfo);
 273
 274         spin_unlock_irqrestore(&rds_tcp_tc_list_lock, flags);
 275 }
 276
 277 static int rds_tcp_laddr_check(struct net *net, __be32 addr)
 278 {
 279         if (inet_addr_type(net, addr) == RTN_LOCAL)
 280                 return 0;
 281         return -EADDRNOTAVAIL;
 282 }
 283
 284 static int rds_tcp_conn_alloc(struct rds_connection *conn, gfp_t gfp)
 285 {
 286         struct rds_tcp_connection *tc;
 287
 288         tc = kmem_cache_alloc(rds_tcp_conn_slab, gfp);
 289         if (!tc)
 290                 return -ENOMEM;
 291
 292         mutex_init(&tc->t_conn_lock);
 293         tc->t_sock = NULL;
 294         tc->t_tinc = NULL;
 295         tc->t_tinc_hdr_rem = sizeof(struct rds_header);
 296         tc->t_tinc_data_rem = 0;
 297
 298         conn->c_transport_data = tc;
 299
 300         spin_lock_irq(&rds_tcp_conn_lock);
 301         list_add_tail(&tc->t_tcp_node, &rds_tcp_conn_list);
 302         spin_unlock_irq(&rds_tcp_conn_lock);
 303
 304         rdsdebug("alloced tc %p\n", conn->c_transport_data);
 305         return 0;
 306 }
 307
 308 static void rds_tcp_conn_free(void *arg)
 309 {
 310         struct rds_tcp_connection *tc = arg;
 311         unsigned long flags;
 312         rdsdebug("freeing tc %p\n", tc);
 313
 314         spin_lock_irqsave(&rds_tcp_conn_lock, flags);
 315         list_del(&tc->t_tcp_node);
 316         spin_unlock_irqrestore(&rds_tcp_conn_lock, flags);
 317
 318         kmem_cache_free(rds_tcp_conn_slab, tc);
 319 }
 320
 321 static void rds_tcp_destroy_conns(void)
 322 {
 323         struct rds_tcp_connection *tc, *_tc;
 324         LIST_HEAD(tmp_list);
 325
 326         /* avoid calling conn_destroy with irqs off */
 327         spin_lock_irq(&rds_tcp_conn_lock);
 328         list_splice(&rds_tcp_conn_list, &tmp_list);
 329         INIT_LIST_HEAD(&rds_tcp_conn_list);
 330         spin_unlock_irq(&rds_tcp_conn_lock);
 331
 332         list_for_each_entry_safe(tc, _tc, &tmp_list, t_tcp_node)
 333                 rds_conn_destroy(tc->conn);
 334 }
 335
 336 static void rds_tcp_exit(void);
 337
 338 struct rds_transport rds_tcp_transport = {
 339         .laddr_check            = rds_tcp_laddr_check,
 340         .xmit_path_prepare      = rds_tcp_xmit_path_prepare,
 341         .xmit_path_complete     = rds_tcp_xmit_path_complete,
 342         .xmit                   = rds_tcp_xmit,
 343         .recv                   = rds_tcp_recv,
 344         .conn_alloc             = rds_tcp_conn_alloc,
 345         .conn_free              = rds_tcp_conn_free,
 346         .conn_connect           = rds_tcp_conn_connect,
 347         .conn_path_shutdown     = rds_tcp_conn_path_shutdown,
 348         .inc_copy_to_user       = rds_tcp_inc_copy_to_user,
 349         .inc_free               = rds_tcp_inc_free,
 350         .stats_info_copy        = rds_tcp_stats_info_copy,
 351         .exit                   = rds_tcp_exit,
 352         .t_owner                = THIS_MODULE,
 353         .t_name                 = "tcp",
 354         .t_type                 = RDS_TRANS_TCP,
 355         .t_prefer_loopback      = 1,
 356 };
 357
 358 static int rds_tcp_netid;
 359
 360 /* per-network namespace private data for this module */
 361 struct rds_tcp_net {
 362         struct socket *rds_tcp_listen_sock;
 363         struct work_struct rds_tcp_accept_w;
 364         struct ctl_table_header *rds_tcp_sysctl;
 365         struct ctl_table *ctl_table;
 366         int sndbuf_size;
 367         int rcvbuf_size;
 368 };
 369
 370 /* All module specific customizations to the RDS-TCP socket should be done in
 371  * rds_tcp_tune() and applied after socket creation.
 372  */
 373 void rds_tcp_tune(struct socket *sock)
 374 {
 375         struct sock *sk = sock->sk;
 376         struct net *net = sock_net(sk);
 377         struct rds_tcp_net *rtn = net_generic(net, rds_tcp_netid);
 378
 379         rds_tcp_nonagle(sock);
 380         lock_sock(sk);
 381         if (rtn->sndbuf_size > 0) {
 382                 sk->sk_sndbuf = rtn->sndbuf_size;
 383                 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
 384         }
 385         if (rtn->rcvbuf_size > 0) {
 386                 sk->sk_sndbuf = rtn->rcvbuf_size;
 387                 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
 388         }
 389         release_sock(sk);
 390 }
 391
 392 static void rds_tcp_accept_worker(struct work_struct *work)
 393 {
 394         struct rds_tcp_net *rtn = container_of(work,
 395                                                struct rds_tcp_net,
 396                                                rds_tcp_accept_w);
 397
 398         while (rds_tcp_accept_one(rtn->rds_tcp_listen_sock) == 0)
 399                 cond_resched();
 400 }
 401
 402 void rds_tcp_accept_work(struct sock *sk)
 403 {
 404         struct net *net = sock_net(sk);
 405         struct rds_tcp_net *rtn = net_generic(net, rds_tcp_netid);
 406
 407         queue_work(rds_wq, &rtn->rds_tcp_accept_w);
 408 }
 409
 410 static __net_init int rds_tcp_init_net(struct net *net)
 411 {
 412         struct rds_tcp_net *rtn = net_generic(net, rds_tcp_netid);
 413         struct ctl_table *tbl;
 414         int err = 0;
 415
 416         memset(rtn, 0, sizeof(*rtn));
 417
 418         /* {snd, rcv}buf_size default to 0, which implies we let the
 419          * stack pick the value, and permit auto-tuning of buffer size.
 420          */
 421         if (net == &init_net) {
 422                 tbl = rds_tcp_sysctl_table;
 423         } else {
 424                 tbl = kmemdup(rds_tcp_sysctl_table,
 425                               sizeof(rds_tcp_sysctl_table), GFP_KERNEL);
 426                 if (!tbl) {
 427                         pr_warn("could not set allocate syctl table\n");
 428                         return -ENOMEM;
 429                 }
 430                 rtn->ctl_table = tbl;
 431         }
 432         tbl[RDS_TCP_SNDBUF].data = &rtn->sndbuf_size;
 433         tbl[RDS_TCP_RCVBUF].data = &rtn->rcvbuf_size;
 434         rtn->rds_tcp_sysctl = register_net_sysctl(net, "net/rds/tcp", tbl);
 435         if (!rtn->rds_tcp_sysctl) {
 436                 pr_warn("could not register sysctl\n");
 437                 err = -ENOMEM;
 438                 goto fail;
 439         }
 440         rtn->rds_tcp_listen_sock = rds_tcp_listen_init(net);
 441         if (!rtn->rds_tcp_listen_sock) {
 442                 pr_warn("could not set up listen sock\n");
 443                 unregister_net_sysctl_table(rtn->rds_tcp_sysctl);
 444                 rtn->rds_tcp_sysctl = NULL;
 445                 err = -EAFNOSUPPORT;
 446                 goto fail;
 447         }
 448         INIT_WORK(&rtn->rds_tcp_accept_w, rds_tcp_accept_worker);
 449         return 0;
 450
 451 fail:
 452         if (net != &init_net)
 453                 kfree(tbl);
 454         return err;
 455 }
 456
 457 static void __net_exit rds_tcp_exit_net(struct net *net)
 458 {
 459         struct rds_tcp_net *rtn = net_generic(net, rds_tcp_netid);
 460
 461         if (rtn->rds_tcp_sysctl)
 462                 unregister_net_sysctl_table(rtn->rds_tcp_sysctl);
 463
 464         if (net != &init_net && rtn->ctl_table)
 465                 kfree(rtn->ctl_table);
 466
 467         /* If rds_tcp_exit_net() is called as a result of netns deletion,
 468          * the rds_tcp_kill_sock() device notifier would already have cleaned
 469          * up the listen socket, thus there is no work to do in this function.
 470          *
 471          * If rds_tcp_exit_net() is called as a result of module unload,
 472          * i.e., due to rds_tcp_exit() -> unregister_pernet_subsys(), then
 473          * we do need to clean up the listen socket here.
 474          */
 475         if (rtn->rds_tcp_listen_sock) {
 476                 rds_tcp_listen_stop(rtn->rds_tcp_listen_sock);
 477                 rtn->rds_tcp_listen_sock = NULL;
 478                 flush_work(&rtn->rds_tcp_accept_w);
 479         }
 480 }
 481
 482 static struct pernet_operations rds_tcp_net_ops = {
 483         .init = rds_tcp_init_net,
 484         .exit = rds_tcp_exit_net,
 485         .id = &rds_tcp_netid,
 486         .size = sizeof(struct rds_tcp_net),
 487 };
 488
 489 static void rds_tcp_kill_sock(struct net *net)
 490 {
 491         struct rds_tcp_connection *tc, *_tc;
 492         struct sock *sk;
 493         LIST_HEAD(tmp_list);
 494         struct rds_tcp_net *rtn = net_generic(net, rds_tcp_netid);
 495
 496         rds_tcp_listen_stop(rtn->rds_tcp_listen_sock);
 497         rtn->rds_tcp_listen_sock = NULL;
 498         flush_work(&rtn->rds_tcp_accept_w);
 499         spin_lock_irq(&rds_tcp_conn_lock);
 500         list_for_each_entry_safe(tc, _tc, &rds_tcp_conn_list, t_tcp_node) {
 501                 struct net *c_net = read_pnet(&tc->conn->c_net);
 502
 503                 if (net != c_net || !tc->t_sock)
 504                         continue;
 505                 list_move_tail(&tc->t_tcp_node, &tmp_list);
 506         }
 507         spin_unlock_irq(&rds_tcp_conn_lock);
 508         list_for_each_entry_safe(tc, _tc, &tmp_list, t_tcp_node) {
 509                 sk = tc->t_sock->sk;
 510                 sk->sk_prot->disconnect(sk, 0);
 511                 tcp_done(sk);
 512                 rds_conn_destroy(tc->conn);
 513         }
 514 }
 515
 516 static int rds_tcp_dev_event(struct notifier_block *this,
 517                              unsigned long event, void *ptr)
 518 {
 519         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
 520
 521         /* rds-tcp registers as a pernet subys, so the ->exit will only
 522          * get invoked after network acitivity has quiesced. We need to
 523          * clean up all sockets  to quiesce network activity, and use
 524          * the unregistration of the per-net loopback device as a trigger
 525          * to start that cleanup.
 526          */
 527         if (event == NETDEV_UNREGISTER_FINAL &&
 528             dev->ifindex == LOOPBACK_IFINDEX)
 529                 rds_tcp_kill_sock(dev_net(dev));
 530
 531         return NOTIFY_DONE;
 532 }
 533
 534 static struct notifier_block rds_tcp_dev_notifier = {
 535         .notifier_call        = rds_tcp_dev_event,
 536         .priority = -10, /* must be called after other network notifiers */
 537 };
 538
 539 /* when sysctl is used to modify some kernel socket parameters,this
 540  * function  resets the RDS connections in that netns  so that we can
 541  * restart with new parameters.  The assumption is that such reset
 542  * events are few and far-between.
 543  */
 544 static void rds_tcp_sysctl_reset(struct net *net)
 545 {
 546         struct rds_tcp_connection *tc, *_tc;
 547
 548         spin_lock_irq(&rds_tcp_conn_lock);
 549         list_for_each_entry_safe(tc, _tc, &rds_tcp_conn_list, t_tcp_node) {
 550                 struct net *c_net = read_pnet(&tc->conn->c_net);
 551
 552                 if (net != c_net || !tc->t_sock)
 553                         continue;
 554
 555                 rds_conn_drop(tc->conn); /* reconnect with new parameters */
 556         }
 557         spin_unlock_irq(&rds_tcp_conn_lock);
 558 }
 559
 560 static int rds_tcp_skbuf_handler(struct ctl_table *ctl, int write,
 561                                  void __user *buffer, size_t *lenp,
 562                                  loff_t *fpos)
 563 {
 564         struct net *net = current->nsproxy->net_ns;
 565         int err;
 566
 567         err = proc_dointvec_minmax(ctl, write, buffer, lenp, fpos);
 568         if (err < 0) {
 569                 pr_warn("Invalid input. Must be >= %d\n",
 570                         *(int *)(ctl->extra1));
 571                 return err;
 572         }
 573         if (write)
 574                 rds_tcp_sysctl_reset(net);
 575         return 0;
 576 }
 577
 578 static void rds_tcp_exit(void)
 579 {
 580         rds_info_deregister_func(RDS_INFO_TCP_SOCKETS, rds_tcp_tc_info);
 581         unregister_pernet_subsys(&rds_tcp_net_ops);
 582         if (unregister_netdevice_notifier(&rds_tcp_dev_notifier))
 583                 pr_warn("could not unregister rds_tcp_dev_notifier\n");
 584         rds_tcp_destroy_conns();
 585         rds_trans_unregister(&rds_tcp_transport);
 586         rds_tcp_recv_exit();
 587         kmem_cache_destroy(rds_tcp_conn_slab);
 588 }
 589 module_exit(rds_tcp_exit);
 590
 591 static int rds_tcp_init(void)
 592 {
 593         int ret;
 594
 595         rds_tcp_conn_slab = kmem_cache_create("rds_tcp_connection",
 596                                               sizeof(struct rds_tcp_connection),
 597                                               0, 0, NULL);
 598         if (!rds_tcp_conn_slab) {
 599                 ret = -ENOMEM;
 600                 goto out;
 601         }
 602
 603         ret = register_netdevice_notifier(&rds_tcp_dev_notifier);
 604         if (ret) {
 605                 pr_warn("could not register rds_tcp_dev_notifier\n");
 606                 goto out;
 607         }
 608
 609         ret = register_pernet_subsys(&rds_tcp_net_ops);
 610         if (ret)
 611                 goto out_slab;
 612
 613         ret = rds_tcp_recv_init();
 614         if (ret)
 615                 goto out_slab;
 616
 617         ret = rds_trans_register(&rds_tcp_transport);
 618         if (ret)
 619                 goto out_recv;
 620
 621         rds_info_register_func(RDS_INFO_TCP_SOCKETS, rds_tcp_tc_info);
 622
 623         goto out;
 624
 625 out_recv:
 626         rds_tcp_recv_exit();
 627 out_slab:
 628         unregister_pernet_subsys(&rds_tcp_net_ops);
 629         kmem_cache_destroy(rds_tcp_conn_slab);
 630 out:
 631         return ret;
 632 }
 633 module_init(rds_tcp_init);
 634
 635 MODULE_AUTHOR("Oracle Corporation <rds-devel@oss.oracle.com>");
 636 MODULE_DESCRIPTION("RDS: TCP transport");
 637 MODULE_LICENSE("Dual BSD/GPL");
 638