linux/net/rds/ib.c
<<
>>
Prefs
   1/*
   2 * Copyright (c) 2006, 2018 Oracle and/or its affiliates. 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/in.h>
  35#include <linux/if.h>
  36#include <linux/netdevice.h>
  37#include <linux/inetdevice.h>
  38#include <linux/if_arp.h>
  39#include <linux/delay.h>
  40#include <linux/slab.h>
  41#include <linux/module.h>
  42#include <net/addrconf.h>
  43
  44#include "rds_single_path.h"
  45#include "rds.h"
  46#include "ib.h"
  47#include "ib_mr.h"
  48
  49static unsigned int rds_ib_mr_1m_pool_size = RDS_MR_1M_POOL_SIZE;
  50static unsigned int rds_ib_mr_8k_pool_size = RDS_MR_8K_POOL_SIZE;
  51unsigned int rds_ib_retry_count = RDS_IB_DEFAULT_RETRY_COUNT;
  52static atomic_t rds_ib_unloading;
  53
  54module_param(rds_ib_mr_1m_pool_size, int, 0444);
  55MODULE_PARM_DESC(rds_ib_mr_1m_pool_size, " Max number of 1M mr per HCA");
  56module_param(rds_ib_mr_8k_pool_size, int, 0444);
  57MODULE_PARM_DESC(rds_ib_mr_8k_pool_size, " Max number of 8K mr per HCA");
  58module_param(rds_ib_retry_count, int, 0444);
  59MODULE_PARM_DESC(rds_ib_retry_count, " Number of hw retries before reporting an error");
  60
  61/*
  62 * we have a clumsy combination of RCU and a rwsem protecting this list
  63 * because it is used both in the get_mr fast path and while blocking in
  64 * the FMR flushing path.
  65 */
  66DECLARE_RWSEM(rds_ib_devices_lock);
  67struct list_head rds_ib_devices;
  68
  69/* NOTE: if also grabbing ibdev lock, grab this first */
  70DEFINE_SPINLOCK(ib_nodev_conns_lock);
  71LIST_HEAD(ib_nodev_conns);
  72
  73static void rds_ib_nodev_connect(void)
  74{
  75        struct rds_ib_connection *ic;
  76
  77        spin_lock(&ib_nodev_conns_lock);
  78        list_for_each_entry(ic, &ib_nodev_conns, ib_node)
  79                rds_conn_connect_if_down(ic->conn);
  80        spin_unlock(&ib_nodev_conns_lock);
  81}
  82
  83static void rds_ib_dev_shutdown(struct rds_ib_device *rds_ibdev)
  84{
  85        struct rds_ib_connection *ic;
  86        unsigned long flags;
  87
  88        spin_lock_irqsave(&rds_ibdev->spinlock, flags);
  89        list_for_each_entry(ic, &rds_ibdev->conn_list, ib_node)
  90                rds_conn_path_drop(&ic->conn->c_path[0], true);
  91        spin_unlock_irqrestore(&rds_ibdev->spinlock, flags);
  92}
  93
  94/*
  95 * rds_ib_destroy_mr_pool() blocks on a few things and mrs drop references
  96 * from interrupt context so we push freing off into a work struct in krdsd.
  97 */
  98static void rds_ib_dev_free(struct work_struct *work)
  99{
 100        struct rds_ib_ipaddr *i_ipaddr, *i_next;
 101        struct rds_ib_device *rds_ibdev = container_of(work,
 102                                        struct rds_ib_device, free_work);
 103
 104        if (rds_ibdev->mr_8k_pool)
 105                rds_ib_destroy_mr_pool(rds_ibdev->mr_8k_pool);
 106        if (rds_ibdev->mr_1m_pool)
 107                rds_ib_destroy_mr_pool(rds_ibdev->mr_1m_pool);
 108        if (rds_ibdev->pd)
 109                ib_dealloc_pd(rds_ibdev->pd);
 110
 111        list_for_each_entry_safe(i_ipaddr, i_next, &rds_ibdev->ipaddr_list, list) {
 112                list_del(&i_ipaddr->list);
 113                kfree(i_ipaddr);
 114        }
 115
 116        kfree(rds_ibdev->vector_load);
 117
 118        kfree(rds_ibdev);
 119}
 120
 121void rds_ib_dev_put(struct rds_ib_device *rds_ibdev)
 122{
 123        BUG_ON(refcount_read(&rds_ibdev->refcount) == 0);
 124        if (refcount_dec_and_test(&rds_ibdev->refcount))
 125                queue_work(rds_wq, &rds_ibdev->free_work);
 126}
 127
 128static void rds_ib_add_one(struct ib_device *device)
 129{
 130        struct rds_ib_device *rds_ibdev;
 131        bool has_fr, has_fmr;
 132
 133        /* Only handle IB (no iWARP) devices */
 134        if (device->node_type != RDMA_NODE_IB_CA)
 135                return;
 136
 137        rds_ibdev = kzalloc_node(sizeof(struct rds_ib_device), GFP_KERNEL,
 138                                 ibdev_to_node(device));
 139        if (!rds_ibdev)
 140                return;
 141
 142        spin_lock_init(&rds_ibdev->spinlock);
 143        refcount_set(&rds_ibdev->refcount, 1);
 144        INIT_WORK(&rds_ibdev->free_work, rds_ib_dev_free);
 145
 146        INIT_LIST_HEAD(&rds_ibdev->ipaddr_list);
 147        INIT_LIST_HEAD(&rds_ibdev->conn_list);
 148
 149        rds_ibdev->max_wrs = device->attrs.max_qp_wr;
 150        rds_ibdev->max_sge = min(device->attrs.max_send_sge, RDS_IB_MAX_SGE);
 151
 152        has_fr = (device->attrs.device_cap_flags &
 153                  IB_DEVICE_MEM_MGT_EXTENSIONS);
 154        has_fmr = (device->ops.alloc_fmr && device->ops.dealloc_fmr &&
 155                   device->ops.map_phys_fmr && device->ops.unmap_fmr);
 156        rds_ibdev->use_fastreg = (has_fr && !has_fmr);
 157
 158        rds_ibdev->fmr_max_remaps = device->attrs.max_map_per_fmr?: 32;
 159        rds_ibdev->max_1m_mrs = device->attrs.max_mr ?
 160                min_t(unsigned int, (device->attrs.max_mr / 2),
 161                      rds_ib_mr_1m_pool_size) : rds_ib_mr_1m_pool_size;
 162
 163        rds_ibdev->max_8k_mrs = device->attrs.max_mr ?
 164                min_t(unsigned int, ((device->attrs.max_mr / 2) * RDS_MR_8K_SCALE),
 165                      rds_ib_mr_8k_pool_size) : rds_ib_mr_8k_pool_size;
 166
 167        rds_ibdev->max_initiator_depth = device->attrs.max_qp_init_rd_atom;
 168        rds_ibdev->max_responder_resources = device->attrs.max_qp_rd_atom;
 169
 170        rds_ibdev->vector_load = kcalloc(device->num_comp_vectors,
 171                                         sizeof(int),
 172                                         GFP_KERNEL);
 173        if (!rds_ibdev->vector_load) {
 174                pr_err("RDS/IB: %s failed to allocate vector memory\n",
 175                        __func__);
 176                goto put_dev;
 177        }
 178
 179        rds_ibdev->dev = device;
 180        rds_ibdev->pd = ib_alloc_pd(device, 0);
 181        if (IS_ERR(rds_ibdev->pd)) {
 182                rds_ibdev->pd = NULL;
 183                goto put_dev;
 184        }
 185
 186        rds_ibdev->mr_1m_pool =
 187                rds_ib_create_mr_pool(rds_ibdev, RDS_IB_MR_1M_POOL);
 188        if (IS_ERR(rds_ibdev->mr_1m_pool)) {
 189                rds_ibdev->mr_1m_pool = NULL;
 190                goto put_dev;
 191        }
 192
 193        rds_ibdev->mr_8k_pool =
 194                rds_ib_create_mr_pool(rds_ibdev, RDS_IB_MR_8K_POOL);
 195        if (IS_ERR(rds_ibdev->mr_8k_pool)) {
 196                rds_ibdev->mr_8k_pool = NULL;
 197                goto put_dev;
 198        }
 199
 200        rdsdebug("RDS/IB: max_mr = %d, max_wrs = %d, max_sge = %d, fmr_max_remaps = %d, max_1m_mrs = %d, max_8k_mrs = %d\n",
 201                 device->attrs.max_fmr, rds_ibdev->max_wrs, rds_ibdev->max_sge,
 202                 rds_ibdev->fmr_max_remaps, rds_ibdev->max_1m_mrs,
 203                 rds_ibdev->max_8k_mrs);
 204
 205        pr_info("RDS/IB: %s: %s supported and preferred\n",
 206                device->name,
 207                rds_ibdev->use_fastreg ? "FRMR" : "FMR");
 208
 209        down_write(&rds_ib_devices_lock);
 210        list_add_tail_rcu(&rds_ibdev->list, &rds_ib_devices);
 211        up_write(&rds_ib_devices_lock);
 212        refcount_inc(&rds_ibdev->refcount);
 213
 214        ib_set_client_data(device, &rds_ib_client, rds_ibdev);
 215        refcount_inc(&rds_ibdev->refcount);
 216
 217        rds_ib_nodev_connect();
 218
 219put_dev:
 220        rds_ib_dev_put(rds_ibdev);
 221}
 222
 223/*
 224 * New connections use this to find the device to associate with the
 225 * connection.  It's not in the fast path so we're not concerned about the
 226 * performance of the IB call.  (As of this writing, it uses an interrupt
 227 * blocking spinlock to serialize walking a per-device list of all registered
 228 * clients.)
 229 *
 230 * RCU is used to handle incoming connections racing with device teardown.
 231 * Rather than use a lock to serialize removal from the client_data and
 232 * getting a new reference, we use an RCU grace period.  The destruction
 233 * path removes the device from client_data and then waits for all RCU
 234 * readers to finish.
 235 *
 236 * A new connection can get NULL from this if its arriving on a
 237 * device that is in the process of being removed.
 238 */
 239struct rds_ib_device *rds_ib_get_client_data(struct ib_device *device)
 240{
 241        struct rds_ib_device *rds_ibdev;
 242
 243        rcu_read_lock();
 244        rds_ibdev = ib_get_client_data(device, &rds_ib_client);
 245        if (rds_ibdev)
 246                refcount_inc(&rds_ibdev->refcount);
 247        rcu_read_unlock();
 248        return rds_ibdev;
 249}
 250
 251/*
 252 * The IB stack is letting us know that a device is going away.  This can
 253 * happen if the underlying HCA driver is removed or if PCI hotplug is removing
 254 * the pci function, for example.
 255 *
 256 * This can be called at any time and can be racing with any other RDS path.
 257 */
 258static void rds_ib_remove_one(struct ib_device *device, void *client_data)
 259{
 260        struct rds_ib_device *rds_ibdev = client_data;
 261
 262        if (!rds_ibdev)
 263                return;
 264
 265        rds_ib_dev_shutdown(rds_ibdev);
 266
 267        /* stop connection attempts from getting a reference to this device. */
 268        ib_set_client_data(device, &rds_ib_client, NULL);
 269
 270        down_write(&rds_ib_devices_lock);
 271        list_del_rcu(&rds_ibdev->list);
 272        up_write(&rds_ib_devices_lock);
 273
 274        /*
 275         * This synchronize rcu is waiting for readers of both the ib
 276         * client data and the devices list to finish before we drop
 277         * both of those references.
 278         */
 279        synchronize_rcu();
 280        rds_ib_dev_put(rds_ibdev);
 281        rds_ib_dev_put(rds_ibdev);
 282}
 283
 284struct ib_client rds_ib_client = {
 285        .name   = "rds_ib",
 286        .add    = rds_ib_add_one,
 287        .remove = rds_ib_remove_one
 288};
 289
 290static int rds_ib_conn_info_visitor(struct rds_connection *conn,
 291                                    void *buffer)
 292{
 293        struct rds_info_rdma_connection *iinfo = buffer;
 294        struct rds_ib_connection *ic = conn->c_transport_data;
 295
 296        /* We will only ever look at IB transports */
 297        if (conn->c_trans != &rds_ib_transport)
 298                return 0;
 299        if (conn->c_isv6)
 300                return 0;
 301
 302        iinfo->src_addr = conn->c_laddr.s6_addr32[3];
 303        iinfo->dst_addr = conn->c_faddr.s6_addr32[3];
 304        if (ic) {
 305                iinfo->tos = conn->c_tos;
 306                iinfo->sl = ic->i_sl;
 307        }
 308
 309        memset(&iinfo->src_gid, 0, sizeof(iinfo->src_gid));
 310        memset(&iinfo->dst_gid, 0, sizeof(iinfo->dst_gid));
 311        if (rds_conn_state(conn) == RDS_CONN_UP) {
 312                struct rds_ib_device *rds_ibdev;
 313
 314                rdma_read_gids(ic->i_cm_id, (union ib_gid *)&iinfo->src_gid,
 315                               (union ib_gid *)&iinfo->dst_gid);
 316
 317                rds_ibdev = ic->rds_ibdev;
 318                iinfo->max_send_wr = ic->i_send_ring.w_nr;
 319                iinfo->max_recv_wr = ic->i_recv_ring.w_nr;
 320                iinfo->max_send_sge = rds_ibdev->max_sge;
 321                rds_ib_get_mr_info(rds_ibdev, iinfo);
 322                iinfo->cache_allocs = atomic_read(&ic->i_cache_allocs);
 323        }
 324        return 1;
 325}
 326
 327#if IS_ENABLED(CONFIG_IPV6)
 328/* IPv6 version of rds_ib_conn_info_visitor(). */
 329static int rds6_ib_conn_info_visitor(struct rds_connection *conn,
 330                                     void *buffer)
 331{
 332        struct rds6_info_rdma_connection *iinfo6 = buffer;
 333        struct rds_ib_connection *ic = conn->c_transport_data;
 334
 335        /* We will only ever look at IB transports */
 336        if (conn->c_trans != &rds_ib_transport)
 337                return 0;
 338
 339        iinfo6->src_addr = conn->c_laddr;
 340        iinfo6->dst_addr = conn->c_faddr;
 341        if (ic) {
 342                iinfo6->tos = conn->c_tos;
 343                iinfo6->sl = ic->i_sl;
 344        }
 345
 346        memset(&iinfo6->src_gid, 0, sizeof(iinfo6->src_gid));
 347        memset(&iinfo6->dst_gid, 0, sizeof(iinfo6->dst_gid));
 348
 349        if (rds_conn_state(conn) == RDS_CONN_UP) {
 350                struct rds_ib_device *rds_ibdev;
 351
 352                rdma_read_gids(ic->i_cm_id, (union ib_gid *)&iinfo6->src_gid,
 353                               (union ib_gid *)&iinfo6->dst_gid);
 354                rds_ibdev = ic->rds_ibdev;
 355                iinfo6->max_send_wr = ic->i_send_ring.w_nr;
 356                iinfo6->max_recv_wr = ic->i_recv_ring.w_nr;
 357                iinfo6->max_send_sge = rds_ibdev->max_sge;
 358                rds6_ib_get_mr_info(rds_ibdev, iinfo6);
 359                iinfo6->cache_allocs = atomic_read(&ic->i_cache_allocs);
 360        }
 361        return 1;
 362}
 363#endif
 364
 365static void rds_ib_ic_info(struct socket *sock, unsigned int len,
 366                           struct rds_info_iterator *iter,
 367                           struct rds_info_lengths *lens)
 368{
 369        u64 buffer[(sizeof(struct rds_info_rdma_connection) + 7) / 8];
 370
 371        rds_for_each_conn_info(sock, len, iter, lens,
 372                                rds_ib_conn_info_visitor,
 373                                buffer,
 374                                sizeof(struct rds_info_rdma_connection));
 375}
 376
 377#if IS_ENABLED(CONFIG_IPV6)
 378/* IPv6 version of rds_ib_ic_info(). */
 379static void rds6_ib_ic_info(struct socket *sock, unsigned int len,
 380                            struct rds_info_iterator *iter,
 381                            struct rds_info_lengths *lens)
 382{
 383        u64 buffer[(sizeof(struct rds6_info_rdma_connection) + 7) / 8];
 384
 385        rds_for_each_conn_info(sock, len, iter, lens,
 386                               rds6_ib_conn_info_visitor,
 387                               buffer,
 388                               sizeof(struct rds6_info_rdma_connection));
 389}
 390#endif
 391
 392/*
 393 * Early RDS/IB was built to only bind to an address if there is an IPoIB
 394 * device with that address set.
 395 *
 396 * If it were me, I'd advocate for something more flexible.  Sending and
 397 * receiving should be device-agnostic.  Transports would try and maintain
 398 * connections between peers who have messages queued.  Userspace would be
 399 * allowed to influence which paths have priority.  We could call userspace
 400 * asserting this policy "routing".
 401 */
 402static int rds_ib_laddr_check(struct net *net, const struct in6_addr *addr,
 403                              __u32 scope_id)
 404{
 405        int ret;
 406        struct rdma_cm_id *cm_id;
 407#if IS_ENABLED(CONFIG_IPV6)
 408        struct sockaddr_in6 sin6;
 409#endif
 410        struct sockaddr_in sin;
 411        struct sockaddr *sa;
 412        bool isv4;
 413
 414        isv4 = ipv6_addr_v4mapped(addr);
 415        /* Create a CMA ID and try to bind it. This catches both
 416         * IB and iWARP capable NICs.
 417         */
 418        cm_id = rdma_create_id(&init_net, rds_rdma_cm_event_handler,
 419                               NULL, RDMA_PS_TCP, IB_QPT_RC);
 420        if (IS_ERR(cm_id))
 421                return PTR_ERR(cm_id);
 422
 423        if (isv4) {
 424                memset(&sin, 0, sizeof(sin));
 425                sin.sin_family = AF_INET;
 426                sin.sin_addr.s_addr = addr->s6_addr32[3];
 427                sa = (struct sockaddr *)&sin;
 428        } else {
 429#if IS_ENABLED(CONFIG_IPV6)
 430                memset(&sin6, 0, sizeof(sin6));
 431                sin6.sin6_family = AF_INET6;
 432                sin6.sin6_addr = *addr;
 433                sin6.sin6_scope_id = scope_id;
 434                sa = (struct sockaddr *)&sin6;
 435
 436                /* XXX Do a special IPv6 link local address check here.  The
 437                 * reason is that rdma_bind_addr() always succeeds with IPv6
 438                 * link local address regardless it is indeed configured in a
 439                 * system.
 440                 */
 441                if (ipv6_addr_type(addr) & IPV6_ADDR_LINKLOCAL) {
 442                        struct net_device *dev;
 443
 444                        if (scope_id == 0) {
 445                                ret = -EADDRNOTAVAIL;
 446                                goto out;
 447                        }
 448
 449                        /* Use init_net for now as RDS is not network
 450                         * name space aware.
 451                         */
 452                        dev = dev_get_by_index(&init_net, scope_id);
 453                        if (!dev) {
 454                                ret = -EADDRNOTAVAIL;
 455                                goto out;
 456                        }
 457                        if (!ipv6_chk_addr(&init_net, addr, dev, 1)) {
 458                                dev_put(dev);
 459                                ret = -EADDRNOTAVAIL;
 460                                goto out;
 461                        }
 462                        dev_put(dev);
 463                }
 464#else
 465                ret = -EADDRNOTAVAIL;
 466                goto out;
 467#endif
 468        }
 469
 470        /* rdma_bind_addr will only succeed for IB & iWARP devices */
 471        ret = rdma_bind_addr(cm_id, sa);
 472        /* due to this, we will claim to support iWARP devices unless we
 473           check node_type. */
 474        if (ret || !cm_id->device ||
 475            cm_id->device->node_type != RDMA_NODE_IB_CA)
 476                ret = -EADDRNOTAVAIL;
 477
 478        rdsdebug("addr %pI6c%%%u ret %d node type %d\n",
 479                 addr, scope_id, ret,
 480                 cm_id->device ? cm_id->device->node_type : -1);
 481
 482out:
 483        rdma_destroy_id(cm_id);
 484
 485        return ret;
 486}
 487
 488static void rds_ib_unregister_client(void)
 489{
 490        ib_unregister_client(&rds_ib_client);
 491        /* wait for rds_ib_dev_free() to complete */
 492        flush_workqueue(rds_wq);
 493}
 494
 495static void rds_ib_set_unloading(void)
 496{
 497        atomic_set(&rds_ib_unloading, 1);
 498}
 499
 500static bool rds_ib_is_unloading(struct rds_connection *conn)
 501{
 502        struct rds_conn_path *cp = &conn->c_path[0];
 503
 504        return (test_bit(RDS_DESTROY_PENDING, &cp->cp_flags) ||
 505                atomic_read(&rds_ib_unloading) != 0);
 506}
 507
 508void rds_ib_exit(void)
 509{
 510        rds_ib_set_unloading();
 511        synchronize_rcu();
 512        rds_info_deregister_func(RDS_INFO_IB_CONNECTIONS, rds_ib_ic_info);
 513#if IS_ENABLED(CONFIG_IPV6)
 514        rds_info_deregister_func(RDS6_INFO_IB_CONNECTIONS, rds6_ib_ic_info);
 515#endif
 516        rds_ib_unregister_client();
 517        rds_ib_destroy_nodev_conns();
 518        rds_ib_sysctl_exit();
 519        rds_ib_recv_exit();
 520        rds_trans_unregister(&rds_ib_transport);
 521        rds_ib_mr_exit();
 522}
 523
 524static u8 rds_ib_get_tos_map(u8 tos)
 525{
 526        /* 1:1 user to transport map for RDMA transport.
 527         * In future, if custom map is desired, hook can export
 528         * user configurable map.
 529         */
 530        return tos;
 531}
 532
 533struct rds_transport rds_ib_transport = {
 534        .laddr_check            = rds_ib_laddr_check,
 535        .xmit_path_complete     = rds_ib_xmit_path_complete,
 536        .xmit                   = rds_ib_xmit,
 537        .xmit_rdma              = rds_ib_xmit_rdma,
 538        .xmit_atomic            = rds_ib_xmit_atomic,
 539        .recv_path              = rds_ib_recv_path,
 540        .conn_alloc             = rds_ib_conn_alloc,
 541        .conn_free              = rds_ib_conn_free,
 542        .conn_path_connect      = rds_ib_conn_path_connect,
 543        .conn_path_shutdown     = rds_ib_conn_path_shutdown,
 544        .inc_copy_to_user       = rds_ib_inc_copy_to_user,
 545        .inc_free               = rds_ib_inc_free,
 546        .cm_initiate_connect    = rds_ib_cm_initiate_connect,
 547        .cm_handle_connect      = rds_ib_cm_handle_connect,
 548        .cm_connect_complete    = rds_ib_cm_connect_complete,
 549        .stats_info_copy        = rds_ib_stats_info_copy,
 550        .exit                   = rds_ib_exit,
 551        .get_mr                 = rds_ib_get_mr,
 552        .sync_mr                = rds_ib_sync_mr,
 553        .free_mr                = rds_ib_free_mr,
 554        .flush_mrs              = rds_ib_flush_mrs,
 555        .get_tos_map            = rds_ib_get_tos_map,
 556        .t_owner                = THIS_MODULE,
 557        .t_name                 = "infiniband",
 558        .t_unloading            = rds_ib_is_unloading,
 559        .t_type                 = RDS_TRANS_IB
 560};
 561
 562int rds_ib_init(void)
 563{
 564        int ret;
 565
 566        INIT_LIST_HEAD(&rds_ib_devices);
 567
 568        ret = rds_ib_mr_init();
 569        if (ret)
 570                goto out;
 571
 572        ret = ib_register_client(&rds_ib_client);
 573        if (ret)
 574                goto out_mr_exit;
 575
 576        ret = rds_ib_sysctl_init();
 577        if (ret)
 578                goto out_ibreg;
 579
 580        ret = rds_ib_recv_init();
 581        if (ret)
 582                goto out_sysctl;
 583
 584        rds_trans_register(&rds_ib_transport);
 585
 586        rds_info_register_func(RDS_INFO_IB_CONNECTIONS, rds_ib_ic_info);
 587#if IS_ENABLED(CONFIG_IPV6)
 588        rds_info_register_func(RDS6_INFO_IB_CONNECTIONS, rds6_ib_ic_info);
 589#endif
 590
 591        goto out;
 592
 593out_sysctl:
 594        rds_ib_sysctl_exit();
 595out_ibreg:
 596        rds_ib_unregister_client();
 597out_mr_exit:
 598        rds_ib_mr_exit();
 599out:
 600        return ret;
 601}
 602
 603MODULE_LICENSE("GPL");
 604