drivers/infiniband/core/cma.c

   1 /*
   2  * Copyright (c) 2005 Voltaire Inc.  All rights reserved.
   3  * Copyright (c) 2002-2005, Network Appliance, Inc. All rights reserved.
   4  * Copyright (c) 1999-2005, Mellanox Technologies, Inc. All rights reserved.
   5  * Copyright (c) 2005-2006 Intel Corporation.  All rights reserved.
   6  *
   7  * This software is available to you under a choice of one of two
   8  * licenses.  You may choose to be licensed under the terms of the GNU
   9  * General Public License (GPL) Version 2, available from the file
  10  * COPYING in the main directory of this source tree, or the
  11  * OpenIB.org BSD license below:
  12  *
  13  *     Redistribution and use in source and binary forms, with or
  14  *     without modification, are permitted provided that the following
  15  *     conditions are met:
  16  *
  17  *      - Redistributions of source code must retain the above
  18  *        copyright notice, this list of conditions and the following
  19  *        disclaimer.
  20  *
  21  *      - Redistributions in binary form must reproduce the above
  22  *        copyright notice, this list of conditions and the following
  23  *        disclaimer in the documentation and/or other materials
  24  *        provided with the distribution.
  25  *
  26  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
  27  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  28  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
  29  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
  30  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
  31  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
  32  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  33  * SOFTWARE.
  34  */
  35
  36 #include <linux/completion.h>
  37 #include <linux/in.h>
  38 #include <linux/in6.h>
  39 #include <linux/mutex.h>
  40 #include <linux/random.h>
  41 #include <linux/idr.h>
  42 #include <linux/inetdevice.h>
  43 #include <linux/slab.h>
  44 #include <linux/module.h>
  45 #include <net/route.h>
  46
  47 #include <net/tcp.h>
  48 #include <net/ipv6.h>
  49
  50 #include <rdma/rdma_cm.h>
  51 #include <rdma/rdma_cm_ib.h>
  52 #include <rdma/rdma_netlink.h>
  53 #include <rdma/ib.h>
  54 #include <rdma/ib_cache.h>
  55 #include <rdma/ib_cm.h>
  56 #include <rdma/ib_sa.h>
  57 #include <rdma/iw_cm.h>
  58
  59 MODULE_AUTHOR("Sean Hefty");
  60 MODULE_DESCRIPTION("Generic RDMA CM Agent");
  61 MODULE_LICENSE("Dual BSD/GPL");
  62
  63 #define CMA_CM_RESPONSE_TIMEOUT 20
  64 #define CMA_MAX_CM_RETRIES 15
  65 #define CMA_CM_MRA_SETTING (IB_CM_MRA_FLAG_DELAY | 24)
  66 #define CMA_IBOE_PACKET_LIFETIME 18
  67
  68 static const char * const cma_events[] = {
  69         [RDMA_CM_EVENT_ADDR_RESOLVED]    = "address resolved",
  70         [RDMA_CM_EVENT_ADDR_ERROR]       = "address error",
  71         [RDMA_CM_EVENT_ROUTE_RESOLVED]   = "route resolved ",
  72         [RDMA_CM_EVENT_ROUTE_ERROR]      = "route error",
  73         [RDMA_CM_EVENT_CONNECT_REQUEST]  = "connect request",
  74         [RDMA_CM_EVENT_CONNECT_RESPONSE] = "connect response",
  75         [RDMA_CM_EVENT_CONNECT_ERROR]    = "connect error",
  76         [RDMA_CM_EVENT_UNREACHABLE]      = "unreachable",
  77         [RDMA_CM_EVENT_REJECTED]         = "rejected",
  78         [RDMA_CM_EVENT_ESTABLISHED]      = "established",
  79         [RDMA_CM_EVENT_DISCONNECTED]     = "disconnected",
  80         [RDMA_CM_EVENT_DEVICE_REMOVAL]   = "device removal",
  81         [RDMA_CM_EVENT_MULTICAST_JOIN]   = "multicast join",
  82         [RDMA_CM_EVENT_MULTICAST_ERROR]  = "multicast error",
  83         [RDMA_CM_EVENT_ADDR_CHANGE]      = "address change",
  84         [RDMA_CM_EVENT_TIMEWAIT_EXIT]    = "timewait exit",
  85 };
  86
  87 const char *rdma_event_msg(enum rdma_cm_event_type event)
  88 {
  89         size_t index = event;
  90
  91         return (index < ARRAY_SIZE(cma_events) && cma_events[index]) ?
  92                         cma_events[index] : "unrecognized event";
  93 }
  94 EXPORT_SYMBOL(rdma_event_msg);
  95
  96 static void cma_add_one(struct ib_device *device);
  97 static void cma_remove_one(struct ib_device *device);
  98
  99 static struct ib_client cma_client = {
 100         .name   = "cma",
 101         .add    = cma_add_one,
 102         .remove = cma_remove_one
 103 };
 104
 105 static struct ib_sa_client sa_client;
 106 static struct rdma_addr_client addr_client;
 107 static LIST_HEAD(dev_list);
 108 static LIST_HEAD(listen_any_list);
 109 static DEFINE_MUTEX(lock);
 110 static struct workqueue_struct *cma_wq;
 111 static DEFINE_IDR(tcp_ps);
 112 static DEFINE_IDR(udp_ps);
 113 static DEFINE_IDR(ipoib_ps);
 114 static DEFINE_IDR(ib_ps);
 115
 116 struct cma_device {
 117         struct list_head        list;
 118         struct ib_device        *device;
 119         struct completion       comp;
 120         atomic_t                refcount;
 121         struct list_head        id_list;
 122 };
 123
 124 struct rdma_bind_list {
 125         struct idr              *ps;
 126         struct hlist_head       owners;
 127         unsigned short          port;
 128 };
 129
 130 enum {
 131         CMA_OPTION_AFONLY,
 132 };
 133
 134 /*
 135  * Device removal can occur at anytime, so we need extra handling to
 136  * serialize notifying the user of device removal with other callbacks.
 137  * We do this by disabling removal notification while a callback is in process,
 138  * and reporting it after the callback completes.
 139  */
 140 struct rdma_id_private {
 141         struct rdma_cm_id       id;
 142
 143         struct rdma_bind_list   *bind_list;
 144         struct hlist_node       node;
 145         struct list_head        list; /* listen_any_list or cma_device.list */
 146         struct list_head        listen_list; /* per device listens */
 147         struct cma_device       *cma_dev;
 148         struct list_head        mc_list;
 149
 150         int                     internal_id;
 151         enum rdma_cm_state      state;
 152         spinlock_t              lock;
 153         struct mutex            qp_mutex;
 154
 155         struct completion       comp;
 156         atomic_t                refcount;
 157         struct mutex            handler_mutex;
 158
 159         int                     backlog;
 160         int                     timeout_ms;
 161         struct ib_sa_query      *query;
 162         int                     query_id;
 163         union {
 164                 struct ib_cm_id *ib;
 165                 struct iw_cm_id *iw;
 166         } cm_id;
 167
 168         u32                     seq_num;
 169         u32                     qkey;
 170         u32                     qp_num;
 171         pid_t                   owner;
 172         u32                     options;
 173         u8                      srq;
 174         u8                      tos;
 175         u8                      reuseaddr;
 176         u8                      afonly;
 177 };
 178
 179 struct cma_multicast {
 180         struct rdma_id_private *id_priv;
 181         union {
 182                 struct ib_sa_multicast *ib;
 183         } multicast;
 184         struct list_head        list;
 185         void                    *context;
 186         struct sockaddr_storage addr;
 187         struct kref             mcref;
 188 };
 189
 190 struct cma_work {
 191         struct work_struct      work;
 192         struct rdma_id_private  *id;
 193         enum rdma_cm_state      old_state;
 194         enum rdma_cm_state      new_state;
 195         struct rdma_cm_event    event;
 196 };
 197
 198 struct cma_ndev_work {
 199         struct work_struct      work;
 200         struct rdma_id_private  *id;
 201         struct rdma_cm_event    event;
 202 };
 203
 204 struct iboe_mcast_work {
 205         struct work_struct       work;
 206         struct rdma_id_private  *id;
 207         struct cma_multicast    *mc;
 208 };
 209
 210 union cma_ip_addr {
 211         struct in6_addr ip6;
 212         struct {
 213                 __be32 pad[3];
 214                 __be32 addr;
 215         } ip4;
 216 };
 217
 218 struct cma_hdr {
 219         u8 cma_version;
 220         u8 ip_version;  /* IP version: 7:4 */
 221         __be16 port;
 222         union cma_ip_addr src_addr;
 223         union cma_ip_addr dst_addr;
 224 };
 225
 226 #define CMA_VERSION 0x00
 227
 228 static int cma_comp(struct rdma_id_private *id_priv, enum rdma_cm_state comp)
 229 {
 230         unsigned long flags;
 231         int ret;
 232
 233         spin_lock_irqsave(&id_priv->lock, flags);
 234         ret = (id_priv->state == comp);
 235         spin_unlock_irqrestore(&id_priv->lock, flags);
 236         return ret;
 237 }
 238
 239 static int cma_comp_exch(struct rdma_id_private *id_priv,
 240                          enum rdma_cm_state comp, enum rdma_cm_state exch)
 241 {
 242         unsigned long flags;
 243         int ret;
 244
 245         spin_lock_irqsave(&id_priv->lock, flags);
 246         if ((ret = (id_priv->state == comp)))
 247                 id_priv->state = exch;
 248         spin_unlock_irqrestore(&id_priv->lock, flags);
 249         return ret;
 250 }
 251
 252 static enum rdma_cm_state cma_exch(struct rdma_id_private *id_priv,
 253                                    enum rdma_cm_state exch)
 254 {
 255         unsigned long flags;
 256         enum rdma_cm_state old;
 257
 258         spin_lock_irqsave(&id_priv->lock, flags);
 259         old = id_priv->state;
 260         id_priv->state = exch;
 261         spin_unlock_irqrestore(&id_priv->lock, flags);
 262         return old;
 263 }
 264
 265 static inline u8 cma_get_ip_ver(struct cma_hdr *hdr)
 266 {
 267         return hdr->ip_version >> 4;
 268 }
 269
 270 static inline void cma_set_ip_ver(struct cma_hdr *hdr, u8 ip_ver)
 271 {
 272         hdr->ip_version = (ip_ver << 4) | (hdr->ip_version & 0xF);
 273 }
 274
 275 static void cma_attach_to_dev(struct rdma_id_private *id_priv,
 276                               struct cma_device *cma_dev)
 277 {
 278         atomic_inc(&cma_dev->refcount);
 279         id_priv->cma_dev = cma_dev;
 280         id_priv->id.device = cma_dev->device;
 281         id_priv->id.route.addr.dev_addr.transport =
 282                 rdma_node_get_transport(cma_dev->device->node_type);
 283         list_add_tail(&id_priv->list, &cma_dev->id_list);
 284 }
 285
 286 static inline void cma_deref_dev(struct cma_device *cma_dev)
 287 {
 288         if (atomic_dec_and_test(&cma_dev->refcount))
 289                 complete(&cma_dev->comp);
 290 }
 291
 292 static inline void release_mc(struct kref *kref)
 293 {
 294         struct cma_multicast *mc = container_of(kref, struct cma_multicast, mcref);
 295
 296         kfree(mc->multicast.ib);
 297         kfree(mc);
 298 }
 299
 300 static void cma_release_dev(struct rdma_id_private *id_priv)
 301 {
 302         mutex_lock(&lock);
 303         list_del(&id_priv->list);
 304         cma_deref_dev(id_priv->cma_dev);
 305         id_priv->cma_dev = NULL;
 306         mutex_unlock(&lock);
 307 }
 308
 309 static inline struct sockaddr *cma_src_addr(struct rdma_id_private *id_priv)
 310 {
 311         return (struct sockaddr *) &id_priv->id.route.addr.src_addr;
 312 }
 313
 314 static inline struct sockaddr *cma_dst_addr(struct rdma_id_private *id_priv)
 315 {
 316         return (struct sockaddr *) &id_priv->id.route.addr.dst_addr;
 317 }
 318
 319 static inline unsigned short cma_family(struct rdma_id_private *id_priv)
 320 {
 321         return id_priv->id.route.addr.src_addr.ss_family;
 322 }
 323
 324 static int cma_set_qkey(struct rdma_id_private *id_priv, u32 qkey)
 325 {
 326         struct ib_sa_mcmember_rec rec;
 327         int ret = 0;
 328
 329         if (id_priv->qkey) {
 330                 if (qkey && id_priv->qkey != qkey)
 331                         return -EINVAL;
 332                 return 0;
 333         }
 334
 335         if (qkey) {
 336                 id_priv->qkey = qkey;
 337                 return 0;
 338         }
 339
 340         switch (id_priv->id.ps) {
 341         case RDMA_PS_UDP:
 342         case RDMA_PS_IB:
 343                 id_priv->qkey = RDMA_UDP_QKEY;
 344                 break;
 345         case RDMA_PS_IPOIB:
 346                 ib_addr_get_mgid(&id_priv->id.route.addr.dev_addr, &rec.mgid);
 347                 ret = ib_sa_get_mcmember_rec(id_priv->id.device,
 348                                              id_priv->id.port_num, &rec.mgid,
 349                                              &rec);
 350                 if (!ret)
 351                         id_priv->qkey = be32_to_cpu(rec.qkey);
 352                 break;
 353         default:
 354                 break;
 355         }
 356         return ret;
 357 }
 358
 359 static void cma_translate_ib(struct sockaddr_ib *sib, struct rdma_dev_addr *dev_addr)
 360 {
 361         dev_addr->dev_type = ARPHRD_INFINIBAND;
 362         rdma_addr_set_sgid(dev_addr, (union ib_gid *) &sib->sib_addr);
 363         ib_addr_set_pkey(dev_addr, ntohs(sib->sib_pkey));
 364 }
 365
 366 static int cma_translate_addr(struct sockaddr *addr, struct rdma_dev_addr *dev_addr)
 367 {
 368         int ret;
 369
 370         if (addr->sa_family != AF_IB) {
 371                 ret = rdma_translate_ip(addr, dev_addr, NULL);
 372         } else {
 373                 cma_translate_ib((struct sockaddr_ib *) addr, dev_addr);
 374                 ret = 0;
 375         }
 376
 377         return ret;
 378 }
 379
 380 static inline int cma_validate_port(struct ib_device *device, u8 port,
 381                                       union ib_gid *gid, int dev_type)
 382 {
 383         u8 found_port;
 384         int ret = -ENODEV;
 385
 386         if ((dev_type == ARPHRD_INFINIBAND) && !rdma_protocol_ib(device, port))
 387                 return ret;
 388
 389         if ((dev_type != ARPHRD_INFINIBAND) && rdma_protocol_ib(device, port))
 390                 return ret;
 391
 392         ret = ib_find_cached_gid(device, gid, &found_port, NULL);
 393         if (port != found_port)
 394                 return -ENODEV;
 395
 396         return ret;
 397 }
 398
 399 static int cma_acquire_dev(struct rdma_id_private *id_priv,
 400                            struct rdma_id_private *listen_id_priv)
 401 {
 402         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
 403         struct cma_device *cma_dev;
 404         union ib_gid gid, iboe_gid, *gidp;
 405         int ret = -ENODEV;
 406         u8 port;
 407
 408         if (dev_addr->dev_type != ARPHRD_INFINIBAND &&
 409             id_priv->id.ps == RDMA_PS_IPOIB)
 410                 return -EINVAL;
 411
 412         mutex_lock(&lock);
 413         rdma_ip2gid((struct sockaddr *)&id_priv->id.route.addr.src_addr,
 414                     &iboe_gid);
 415
 416         memcpy(&gid, dev_addr->src_dev_addr +
 417                rdma_addr_gid_offset(dev_addr), sizeof gid);
 418
 419         if (listen_id_priv) {
 420                 cma_dev = listen_id_priv->cma_dev;
 421                 port = listen_id_priv->id.port_num;
 422                 gidp = rdma_protocol_roce(cma_dev->device, port) ?
 423                        &iboe_gid : &gid;
 424
 425                 ret = cma_validate_port(cma_dev->device, port, gidp,
 426                                         dev_addr->dev_type);
 427                 if (!ret) {
 428                         id_priv->id.port_num = port;
 429                         goto out;
 430                 }
 431         }
 432
 433         list_for_each_entry(cma_dev, &dev_list, list) {
 434                 for (port = 1; port <= cma_dev->device->phys_port_cnt; ++port) {
 435                         if (listen_id_priv &&
 436                             listen_id_priv->cma_dev == cma_dev &&
 437                             listen_id_priv->id.port_num == port)
 438                                 continue;
 439
 440                         gidp = rdma_protocol_roce(cma_dev->device, port) ?
 441                                &iboe_gid : &gid;
 442
 443                         ret = cma_validate_port(cma_dev->device, port, gidp,
 444                                                 dev_addr->dev_type);
 445                         if (!ret) {
 446                                 id_priv->id.port_num = port;
 447                                 goto out;
 448                         }
 449                 }
 450         }
 451
 452 out:
 453         if (!ret)
 454                 cma_attach_to_dev(id_priv, cma_dev);
 455
 456         mutex_unlock(&lock);
 457         return ret;
 458 }
 459
 460 /*
 461  * Select the source IB device and address to reach the destination IB address.
 462  */
 463 static int cma_resolve_ib_dev(struct rdma_id_private *id_priv)
 464 {
 465         struct cma_device *cma_dev, *cur_dev;
 466         struct sockaddr_ib *addr;
 467         union ib_gid gid, sgid, *dgid;
 468         u16 pkey, index;
 469         u8 p;
 470         int i;
 471
 472         cma_dev = NULL;
 473         addr = (struct sockaddr_ib *) cma_dst_addr(id_priv);
 474         dgid = (union ib_gid *) &addr->sib_addr;
 475         pkey = ntohs(addr->sib_pkey);
 476
 477         list_for_each_entry(cur_dev, &dev_list, list) {
 478                 for (p = 1; p <= cur_dev->device->phys_port_cnt; ++p) {
 479                         if (!rdma_cap_af_ib(cur_dev->device, p))
 480                                 continue;
 481
 482                         if (ib_find_cached_pkey(cur_dev->device, p, pkey, &index))
 483                                 continue;
 484
 485                         for (i = 0; !ib_get_cached_gid(cur_dev->device, p, i, &gid); i++) {
 486                                 if (!memcmp(&gid, dgid, sizeof(gid))) {
 487                                         cma_dev = cur_dev;
 488                                         sgid = gid;
 489                                         id_priv->id.port_num = p;
 490                                         goto found;
 491                                 }
 492
 493                                 if (!cma_dev && (gid.global.subnet_prefix ==
 494                                                  dgid->global.subnet_prefix)) {
 495                                         cma_dev = cur_dev;
 496                                         sgid = gid;
 497                                         id_priv->id.port_num = p;
 498                                 }
 499                         }
 500                 }
 501         }
 502
 503         if (!cma_dev)
 504                 return -ENODEV;
 505
 506 found:
 507         cma_attach_to_dev(id_priv, cma_dev);
 508         addr = (struct sockaddr_ib *) cma_src_addr(id_priv);
 509         memcpy(&addr->sib_addr, &sgid, sizeof sgid);
 510         cma_translate_ib(addr, &id_priv->id.route.addr.dev_addr);
 511         return 0;
 512 }
 513
 514 static void cma_deref_id(struct rdma_id_private *id_priv)
 515 {
 516         if (atomic_dec_and_test(&id_priv->refcount))
 517                 complete(&id_priv->comp);
 518 }
 519
 520 static int cma_disable_callback(struct rdma_id_private *id_priv,
 521                                 enum rdma_cm_state state)
 522 {
 523         mutex_lock(&id_priv->handler_mutex);
 524         if (id_priv->state != state) {
 525                 mutex_unlock(&id_priv->handler_mutex);
 526                 return -EINVAL;
 527         }
 528         return 0;
 529 }
 530
 531 struct rdma_cm_id *rdma_create_id(rdma_cm_event_handler event_handler,
 532                                   void *context, enum rdma_port_space ps,
 533                                   enum ib_qp_type qp_type)
 534 {
 535         struct rdma_id_private *id_priv;
 536
 537         id_priv = kzalloc(sizeof *id_priv, GFP_KERNEL);
 538         if (!id_priv)
 539                 return ERR_PTR(-ENOMEM);
 540
 541         id_priv->owner = task_pid_nr(current);
 542         id_priv->state = RDMA_CM_IDLE;
 543         id_priv->id.context = context;
 544         id_priv->id.event_handler = event_handler;
 545         id_priv->id.ps = ps;
 546         id_priv->id.qp_type = qp_type;
 547         spin_lock_init(&id_priv->lock);
 548         mutex_init(&id_priv->qp_mutex);
 549         init_completion(&id_priv->comp);
 550         atomic_set(&id_priv->refcount, 1);
 551         mutex_init(&id_priv->handler_mutex);
 552         INIT_LIST_HEAD(&id_priv->listen_list);
 553         INIT_LIST_HEAD(&id_priv->mc_list);
 554         get_random_bytes(&id_priv->seq_num, sizeof id_priv->seq_num);
 555
 556         return &id_priv->id;
 557 }
 558 EXPORT_SYMBOL(rdma_create_id);
 559
 560 static int cma_init_ud_qp(struct rdma_id_private *id_priv, struct ib_qp *qp)
 561 {
 562         struct ib_qp_attr qp_attr;
 563         int qp_attr_mask, ret;
 564
 565         qp_attr.qp_state = IB_QPS_INIT;
 566         ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
 567         if (ret)
 568                 return ret;
 569
 570         ret = ib_modify_qp(qp, &qp_attr, qp_attr_mask);
 571         if (ret)
 572                 return ret;
 573
 574         qp_attr.qp_state = IB_QPS_RTR;
 575         ret = ib_modify_qp(qp, &qp_attr, IB_QP_STATE);
 576         if (ret)
 577                 return ret;
 578
 579         qp_attr.qp_state = IB_QPS_RTS;
 580         qp_attr.sq_psn = 0;
 581         ret = ib_modify_qp(qp, &qp_attr, IB_QP_STATE | IB_QP_SQ_PSN);
 582
 583         return ret;
 584 }
 585
 586 static int cma_init_conn_qp(struct rdma_id_private *id_priv, struct ib_qp *qp)
 587 {
 588         struct ib_qp_attr qp_attr;
 589         int qp_attr_mask, ret;
 590
 591         qp_attr.qp_state = IB_QPS_INIT;
 592         ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
 593         if (ret)
 594                 return ret;
 595
 596         return ib_modify_qp(qp, &qp_attr, qp_attr_mask);
 597 }
 598
 599 int rdma_create_qp(struct rdma_cm_id *id, struct ib_pd *pd,
 600                    struct ib_qp_init_attr *qp_init_attr)
 601 {
 602         struct rdma_id_private *id_priv;
 603         struct ib_qp *qp;
 604         int ret;
 605
 606         id_priv = container_of(id, struct rdma_id_private, id);
 607         if (id->device != pd->device)
 608                 return -EINVAL;
 609
 610         qp = ib_create_qp(pd, qp_init_attr);
 611         if (IS_ERR(qp))
 612                 return PTR_ERR(qp);
 613
 614         if (id->qp_type == IB_QPT_UD)
 615                 ret = cma_init_ud_qp(id_priv, qp);
 616         else
 617                 ret = cma_init_conn_qp(id_priv, qp);
 618         if (ret)
 619                 goto err;
 620
 621         id->qp = qp;
 622         id_priv->qp_num = qp->qp_num;
 623         id_priv->srq = (qp->srq != NULL);
 624         return 0;
 625 err:
 626         ib_destroy_qp(qp);
 627         return ret;
 628 }
 629 EXPORT_SYMBOL(rdma_create_qp);
 630
 631 void rdma_destroy_qp(struct rdma_cm_id *id)
 632 {
 633         struct rdma_id_private *id_priv;
 634
 635         id_priv = container_of(id, struct rdma_id_private, id);
 636         mutex_lock(&id_priv->qp_mutex);
 637         ib_destroy_qp(id_priv->id.qp);
 638         id_priv->id.qp = NULL;
 639         mutex_unlock(&id_priv->qp_mutex);
 640 }
 641 EXPORT_SYMBOL(rdma_destroy_qp);
 642
 643 static int cma_modify_qp_rtr(struct rdma_id_private *id_priv,
 644                              struct rdma_conn_param *conn_param)
 645 {
 646         struct ib_qp_attr qp_attr;
 647         int qp_attr_mask, ret;
 648         union ib_gid sgid;
 649
 650         mutex_lock(&id_priv->qp_mutex);
 651         if (!id_priv->id.qp) {
 652                 ret = 0;
 653                 goto out;
 654         }
 655
 656         /* Need to update QP attributes from default values. */
 657         qp_attr.qp_state = IB_QPS_INIT;
 658         ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
 659         if (ret)
 660                 goto out;
 661
 662         ret = ib_modify_qp(id_priv->id.qp, &qp_attr, qp_attr_mask);
 663         if (ret)
 664                 goto out;
 665
 666         qp_attr.qp_state = IB_QPS_RTR;
 667         ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
 668         if (ret)
 669                 goto out;
 670
 671         ret = ib_query_gid(id_priv->id.device, id_priv->id.port_num,
 672                            qp_attr.ah_attr.grh.sgid_index, &sgid);
 673         if (ret)
 674                 goto out;
 675
 676         BUG_ON(id_priv->cma_dev->device != id_priv->id.device);
 677
 678         if (rdma_protocol_roce(id_priv->id.device, id_priv->id.port_num)) {
 679                 ret = rdma_addr_find_smac_by_sgid(&sgid, qp_attr.smac, NULL);
 680
 681                 if (ret)
 682                         goto out;
 683         }
 684         if (conn_param)
 685                 qp_attr.max_dest_rd_atomic = conn_param->responder_resources;
 686         ret = ib_modify_qp(id_priv->id.qp, &qp_attr, qp_attr_mask);
 687 out:
 688         mutex_unlock(&id_priv->qp_mutex);
 689         return ret;
 690 }
 691
 692 static int cma_modify_qp_rts(struct rdma_id_private *id_priv,
 693                              struct rdma_conn_param *conn_param)
 694 {
 695         struct ib_qp_attr qp_attr;
 696         int qp_attr_mask, ret;
 697
 698         mutex_lock(&id_priv->qp_mutex);
 699         if (!id_priv->id.qp) {
 700                 ret = 0;
 701                 goto out;
 702         }
 703
 704         qp_attr.qp_state = IB_QPS_RTS;
 705         ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
 706         if (ret)
 707                 goto out;
 708
 709         if (conn_param)
 710                 qp_attr.max_rd_atomic = conn_param->initiator_depth;
 711         ret = ib_modify_qp(id_priv->id.qp, &qp_attr, qp_attr_mask);
 712 out:
 713         mutex_unlock(&id_priv->qp_mutex);
 714         return ret;
 715 }
 716
 717 static int cma_modify_qp_err(struct rdma_id_private *id_priv)
 718 {
 719         struct ib_qp_attr qp_attr;
 720         int ret;
 721
 722         mutex_lock(&id_priv->qp_mutex);
 723         if (!id_priv->id.qp) {
 724                 ret = 0;
 725                 goto out;
 726         }
 727
 728         qp_attr.qp_state = IB_QPS_ERR;
 729         ret = ib_modify_qp(id_priv->id.qp, &qp_attr, IB_QP_STATE);
 730 out:
 731         mutex_unlock(&id_priv->qp_mutex);
 732         return ret;
 733 }
 734
 735 static int cma_ib_init_qp_attr(struct rdma_id_private *id_priv,
 736                                struct ib_qp_attr *qp_attr, int *qp_attr_mask)
 737 {
 738         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
 739         int ret;
 740         u16 pkey;
 741
 742         if (rdma_cap_eth_ah(id_priv->id.device, id_priv->id.port_num))
 743                 pkey = 0xffff;
 744         else
 745                 pkey = ib_addr_get_pkey(dev_addr);
 746
 747         ret = ib_find_cached_pkey(id_priv->id.device, id_priv->id.port_num,
 748                                   pkey, &qp_attr->pkey_index);
 749         if (ret)
 750                 return ret;
 751
 752         qp_attr->port_num = id_priv->id.port_num;
 753         *qp_attr_mask = IB_QP_STATE | IB_QP_PKEY_INDEX | IB_QP_PORT;
 754
 755         if (id_priv->id.qp_type == IB_QPT_UD) {
 756                 ret = cma_set_qkey(id_priv, 0);
 757                 if (ret)
 758                         return ret;
 759
 760                 qp_attr->qkey = id_priv->qkey;
 761                 *qp_attr_mask |= IB_QP_QKEY;
 762         } else {
 763                 qp_attr->qp_access_flags = 0;
 764                 *qp_attr_mask |= IB_QP_ACCESS_FLAGS;
 765         }
 766         return 0;
 767 }
 768
 769 int rdma_init_qp_attr(struct rdma_cm_id *id, struct ib_qp_attr *qp_attr,
 770                        int *qp_attr_mask)
 771 {
 772         struct rdma_id_private *id_priv;
 773         int ret = 0;
 774
 775         id_priv = container_of(id, struct rdma_id_private, id);
 776         if (rdma_cap_ib_cm(id->device, id->port_num)) {
 777                 if (!id_priv->cm_id.ib || (id_priv->id.qp_type == IB_QPT_UD))
 778                         ret = cma_ib_init_qp_attr(id_priv, qp_attr, qp_attr_mask);
 779                 else
 780                         ret = ib_cm_init_qp_attr(id_priv->cm_id.ib, qp_attr,
 781                                                  qp_attr_mask);
 782
 783                 if (qp_attr->qp_state == IB_QPS_RTR)
 784                         qp_attr->rq_psn = id_priv->seq_num;
 785         } else if (rdma_cap_iw_cm(id->device, id->port_num)) {
 786                 if (!id_priv->cm_id.iw) {
 787                         qp_attr->qp_access_flags = 0;
 788                         *qp_attr_mask = IB_QP_STATE | IB_QP_ACCESS_FLAGS;
 789                 } else
 790                         ret = iw_cm_init_qp_attr(id_priv->cm_id.iw, qp_attr,
 791                                                  qp_attr_mask);
 792         } else
 793                 ret = -ENOSYS;
 794
 795         return ret;
 796 }
 797 EXPORT_SYMBOL(rdma_init_qp_attr);
 798
 799 static inline int cma_zero_addr(struct sockaddr *addr)
 800 {
 801         switch (addr->sa_family) {
 802         case AF_INET:
 803                 return ipv4_is_zeronet(((struct sockaddr_in *)addr)->sin_addr.s_addr);
 804         case AF_INET6:
 805                 return ipv6_addr_any(&((struct sockaddr_in6 *) addr)->sin6_addr);
 806         case AF_IB:
 807                 return ib_addr_any(&((struct sockaddr_ib *) addr)->sib_addr);
 808         default:
 809                 return 0;
 810         }
 811 }
 812
 813 static inline int cma_loopback_addr(struct sockaddr *addr)
 814 {
 815         switch (addr->sa_family) {
 816         case AF_INET:
 817                 return ipv4_is_loopback(((struct sockaddr_in *) addr)->sin_addr.s_addr);
 818         case AF_INET6:
 819                 return ipv6_addr_loopback(&((struct sockaddr_in6 *) addr)->sin6_addr);
 820         case AF_IB:
 821                 return ib_addr_loopback(&((struct sockaddr_ib *) addr)->sib_addr);
 822         default:
 823                 return 0;
 824         }
 825 }
 826
 827 static inline int cma_any_addr(struct sockaddr *addr)
 828 {
 829         return cma_zero_addr(addr) || cma_loopback_addr(addr);
 830 }
 831
 832 static int cma_addr_cmp(struct sockaddr *src, struct sockaddr *dst)
 833 {
 834         if (src->sa_family != dst->sa_family)
 835                 return -1;
 836
 837         switch (src->sa_family) {
 838         case AF_INET:
 839                 return ((struct sockaddr_in *) src)->sin_addr.s_addr !=
 840                        ((struct sockaddr_in *) dst)->sin_addr.s_addr;
 841         case AF_INET6:
 842                 return ipv6_addr_cmp(&((struct sockaddr_in6 *) src)->sin6_addr,
 843                                      &((struct sockaddr_in6 *) dst)->sin6_addr);
 844         default:
 845                 return ib_addr_cmp(&((struct sockaddr_ib *) src)->sib_addr,
 846                                    &((struct sockaddr_ib *) dst)->sib_addr);
 847         }
 848 }
 849
 850 static __be16 cma_port(struct sockaddr *addr)
 851 {
 852         struct sockaddr_ib *sib;
 853
 854         switch (addr->sa_family) {
 855         case AF_INET:
 856                 return ((struct sockaddr_in *) addr)->sin_port;
 857         case AF_INET6:
 858                 return ((struct sockaddr_in6 *) addr)->sin6_port;
 859         case AF_IB:
 860                 sib = (struct sockaddr_ib *) addr;
 861                 return htons((u16) (be64_to_cpu(sib->sib_sid) &
 862                                     be64_to_cpu(sib->sib_sid_mask)));
 863         default:
 864                 return 0;
 865         }
 866 }
 867
 868 static inline int cma_any_port(struct sockaddr *addr)
 869 {
 870         return !cma_port(addr);
 871 }
 872
 873 static void cma_save_ib_info(struct rdma_cm_id *id, struct rdma_cm_id *listen_id,
 874                              struct ib_sa_path_rec *path)
 875 {
 876         struct sockaddr_ib *listen_ib, *ib;
 877
 878         listen_ib = (struct sockaddr_ib *) &listen_id->route.addr.src_addr;
 879         ib = (struct sockaddr_ib *) &id->route.addr.src_addr;
 880         ib->sib_family = listen_ib->sib_family;
 881         ib->sib_pkey = path->pkey;
 882         ib->sib_flowinfo = path->flow_label;
 883         memcpy(&ib->sib_addr, &path->sgid, 16);
 884         ib->sib_sid = listen_ib->sib_sid;
 885         ib->sib_sid_mask = cpu_to_be64(0xffffffffffffffffULL);
 886         ib->sib_scope_id = listen_ib->sib_scope_id;
 887
 888         ib = (struct sockaddr_ib *) &id->route.addr.dst_addr;
 889         ib->sib_family = listen_ib->sib_family;
 890         ib->sib_pkey = path->pkey;
 891         ib->sib_flowinfo = path->flow_label;
 892         memcpy(&ib->sib_addr, &path->dgid, 16);
 893 }
 894
 895 static __be16 ss_get_port(const struct sockaddr_storage *ss)
 896 {
 897         if (ss->ss_family == AF_INET)
 898                 return ((struct sockaddr_in *)ss)->sin_port;
 899         else if (ss->ss_family == AF_INET6)
 900                 return ((struct sockaddr_in6 *)ss)->sin6_port;
 901         BUG();
 902 }
 903
 904 static void cma_save_ip4_info(struct rdma_cm_id *id, struct rdma_cm_id *listen_id,
 905                               struct cma_hdr *hdr)
 906 {
 907         struct sockaddr_in *ip4;
 908
 909         ip4 = (struct sockaddr_in *) &id->route.addr.src_addr;
 910         ip4->sin_family = AF_INET;
 911         ip4->sin_addr.s_addr = hdr->dst_addr.ip4.addr;
 912         ip4->sin_port = ss_get_port(&listen_id->route.addr.src_addr);
 913
 914         ip4 = (struct sockaddr_in *) &id->route.addr.dst_addr;
 915         ip4->sin_family = AF_INET;
 916         ip4->sin_addr.s_addr = hdr->src_addr.ip4.addr;
 917         ip4->sin_port = hdr->port;
 918 }
 919
 920 static void cma_save_ip6_info(struct rdma_cm_id *id, struct rdma_cm_id *listen_id,
 921                               struct cma_hdr *hdr)
 922 {
 923         struct sockaddr_in6 *ip6;
 924
 925         ip6 = (struct sockaddr_in6 *) &id->route.addr.src_addr;
 926         ip6->sin6_family = AF_INET6;
 927         ip6->sin6_addr = hdr->dst_addr.ip6;
 928         ip6->sin6_port = ss_get_port(&listen_id->route.addr.src_addr);
 929
 930         ip6 = (struct sockaddr_in6 *) &id->route.addr.dst_addr;
 931         ip6->sin6_family = AF_INET6;
 932         ip6->sin6_addr = hdr->src_addr.ip6;
 933         ip6->sin6_port = hdr->port;
 934 }
 935
 936 static int cma_save_net_info(struct rdma_cm_id *id, struct rdma_cm_id *listen_id,
 937                              struct ib_cm_event *ib_event)
 938 {
 939         struct cma_hdr *hdr;
 940
 941         if ((listen_id->route.addr.src_addr.ss_family == AF_IB) &&
 942             (ib_event->event == IB_CM_REQ_RECEIVED)) {
 943                 cma_save_ib_info(id, listen_id, ib_event->param.req_rcvd.primary_path);
 944                 return 0;
 945         }
 946
 947         hdr = ib_event->private_data;
 948         if (hdr->cma_version != CMA_VERSION)
 949                 return -EINVAL;
 950
 951         switch (cma_get_ip_ver(hdr)) {
 952         case 4:
 953                 cma_save_ip4_info(id, listen_id, hdr);
 954                 break;
 955         case 6:
 956                 cma_save_ip6_info(id, listen_id, hdr);
 957                 break;
 958         default:
 959                 return -EINVAL;
 960         }
 961         return 0;
 962 }
 963
 964 static inline int cma_user_data_offset(struct rdma_id_private *id_priv)
 965 {
 966         return cma_family(id_priv) == AF_IB ? 0 : sizeof(struct cma_hdr);
 967 }
 968
 969 static void cma_cancel_route(struct rdma_id_private *id_priv)
 970 {
 971         if (rdma_cap_ib_sa(id_priv->id.device, id_priv->id.port_num)) {
 972                 if (id_priv->query)
 973                         ib_sa_cancel_query(id_priv->query_id, id_priv->query);
 974         }
 975 }
 976
 977 static void cma_cancel_listens(struct rdma_id_private *id_priv)
 978 {
 979         struct rdma_id_private *dev_id_priv;
 980
 981         /*
 982          * Remove from listen_any_list to prevent added devices from spawning
 983          * additional listen requests.
 984          */
 985         mutex_lock(&lock);
 986         list_del(&id_priv->list);
 987
 988         while (!list_empty(&id_priv->listen_list)) {
 989                 dev_id_priv = list_entry(id_priv->listen_list.next,
 990                                          struct rdma_id_private, listen_list);
 991                 /* sync with device removal to avoid duplicate destruction */
 992                 list_del_init(&dev_id_priv->list);
 993                 list_del(&dev_id_priv->listen_list);
 994                 mutex_unlock(&lock);
 995
 996                 rdma_destroy_id(&dev_id_priv->id);
 997                 mutex_lock(&lock);
 998         }
 999         mutex_unlock(&lock);
1000 }
1001
1002 static void cma_cancel_operation(struct rdma_id_private *id_priv,
1003                                  enum rdma_cm_state state)
1004 {
1005         switch (state) {
1006         case RDMA_CM_ADDR_QUERY:
1007                 rdma_addr_cancel(&id_priv->id.route.addr.dev_addr);
1008                 break;
1009         case RDMA_CM_ROUTE_QUERY:
1010                 cma_cancel_route(id_priv);
1011                 break;
1012         case RDMA_CM_LISTEN:
1013                 if (cma_any_addr(cma_src_addr(id_priv)) && !id_priv->cma_dev)
1014                         cma_cancel_listens(id_priv);
1015                 break;
1016         default:
1017                 break;
1018         }
1019 }
1020
1021 static void cma_release_port(struct rdma_id_private *id_priv)
1022 {
1023         struct rdma_bind_list *bind_list = id_priv->bind_list;
1024
1025         if (!bind_list)
1026                 return;
1027
1028         mutex_lock(&lock);
1029         hlist_del(&id_priv->node);
1030         if (hlist_empty(&bind_list->owners)) {
1031                 idr_remove(bind_list->ps, bind_list->port);
1032                 kfree(bind_list);
1033         }
1034         mutex_unlock(&lock);
1035 }
1036
1037 static void cma_leave_mc_groups(struct rdma_id_private *id_priv)
1038 {
1039         struct cma_multicast *mc;
1040
1041         while (!list_empty(&id_priv->mc_list)) {
1042                 mc = container_of(id_priv->mc_list.next,
1043                                   struct cma_multicast, list);
1044                 list_del(&mc->list);
1045                 if (rdma_cap_ib_mcast(id_priv->cma_dev->device,
1046                                       id_priv->id.port_num)) {
1047                         ib_sa_free_multicast(mc->multicast.ib);
1048                         kfree(mc);
1049                 } else
1050                         kref_put(&mc->mcref, release_mc);
1051         }
1052 }
1053
1054 void rdma_destroy_id(struct rdma_cm_id *id)
1055 {
1056         struct rdma_id_private *id_priv;
1057         enum rdma_cm_state state;
1058
1059         id_priv = container_of(id, struct rdma_id_private, id);
1060         state = cma_exch(id_priv, RDMA_CM_DESTROYING);
1061         cma_cancel_operation(id_priv, state);
1062
1063         /*
1064          * Wait for any active callback to finish.  New callbacks will find
1065          * the id_priv state set to destroying and abort.
1066          */
1067         mutex_lock(&id_priv->handler_mutex);
1068         mutex_unlock(&id_priv->handler_mutex);
1069
1070         if (id_priv->cma_dev) {
1071                 if (rdma_cap_ib_cm(id_priv->id.device, 1)) {
1072                         if (id_priv->cm_id.ib)
1073                                 ib_destroy_cm_id(id_priv->cm_id.ib);
1074                 } else if (rdma_cap_iw_cm(id_priv->id.device, 1)) {
1075                         if (id_priv->cm_id.iw)
1076                                 iw_destroy_cm_id(id_priv->cm_id.iw);
1077                 }
1078                 cma_leave_mc_groups(id_priv);
1079                 cma_release_dev(id_priv);
1080         }
1081
1082         cma_release_port(id_priv);
1083         cma_deref_id(id_priv);
1084         wait_for_completion(&id_priv->comp);
1085
1086         if (id_priv->internal_id)
1087                 cma_deref_id(id_priv->id.context);
1088
1089         kfree(id_priv->id.route.path_rec);
1090         kfree(id_priv);
1091 }
1092 EXPORT_SYMBOL(rdma_destroy_id);
1093
1094 static int cma_rep_recv(struct rdma_id_private *id_priv)
1095 {
1096         int ret;
1097
1098         ret = cma_modify_qp_rtr(id_priv, NULL);
1099         if (ret)
1100                 goto reject;
1101
1102         ret = cma_modify_qp_rts(id_priv, NULL);
1103         if (ret)
1104                 goto reject;
1105
1106         ret = ib_send_cm_rtu(id_priv->cm_id.ib, NULL, 0);
1107         if (ret)
1108                 goto reject;
1109
1110         return 0;
1111 reject:
1112         cma_modify_qp_err(id_priv);
1113         ib_send_cm_rej(id_priv->cm_id.ib, IB_CM_REJ_CONSUMER_DEFINED,
1114                        NULL, 0, NULL, 0);
1115         return ret;
1116 }
1117
1118 static void cma_set_rep_event_data(struct rdma_cm_event *event,
1119                                    struct ib_cm_rep_event_param *rep_data,
1120                                    void *private_data)
1121 {
1122         event->param.conn.private_data = private_data;
1123         event->param.conn.private_data_len = IB_CM_REP_PRIVATE_DATA_SIZE;
1124         event->param.conn.responder_resources = rep_data->responder_resources;
1125         event->param.conn.initiator_depth = rep_data->initiator_depth;
1126         event->param.conn.flow_control = rep_data->flow_control;
1127         event->param.conn.rnr_retry_count = rep_data->rnr_retry_count;
1128         event->param.conn.srq = rep_data->srq;
1129         event->param.conn.qp_num = rep_data->remote_qpn;
1130 }
1131
1132 static int cma_ib_handler(struct ib_cm_id *cm_id, struct ib_cm_event *ib_event)
1133 {
1134         struct rdma_id_private *id_priv = cm_id->context;
1135         struct rdma_cm_event event;
1136         int ret = 0;
1137
1138         if ((ib_event->event != IB_CM_TIMEWAIT_EXIT &&
1139                 cma_disable_callback(id_priv, RDMA_CM_CONNECT)) ||
1140             (ib_event->event == IB_CM_TIMEWAIT_EXIT &&
1141                 cma_disable_callback(id_priv, RDMA_CM_DISCONNECT)))
1142                 return 0;
1143
1144         memset(&event, 0, sizeof event);
1145         switch (ib_event->event) {
1146         case IB_CM_REQ_ERROR:
1147         case IB_CM_REP_ERROR:
1148                 event.event = RDMA_CM_EVENT_UNREACHABLE;
1149                 event.status = -ETIMEDOUT;
1150                 break;
1151         case IB_CM_REP_RECEIVED:
1152                 if (id_priv->id.qp) {
1153                         event.status = cma_rep_recv(id_priv);
1154                         event.event = event.status ? RDMA_CM_EVENT_CONNECT_ERROR :
1155                                                      RDMA_CM_EVENT_ESTABLISHED;
1156                 } else {
1157                         event.event = RDMA_CM_EVENT_CONNECT_RESPONSE;
1158                 }
1159                 cma_set_rep_event_data(&event, &ib_event->param.rep_rcvd,
1160                                        ib_event->private_data);
1161                 break;
1162         case IB_CM_RTU_RECEIVED:
1163         case IB_CM_USER_ESTABLISHED:
1164                 event.event = RDMA_CM_EVENT_ESTABLISHED;
1165                 break;
1166         case IB_CM_DREQ_ERROR:
1167                 event.status = -ETIMEDOUT; /* fall through */
1168         case IB_CM_DREQ_RECEIVED:
1169         case IB_CM_DREP_RECEIVED:
1170                 if (!cma_comp_exch(id_priv, RDMA_CM_CONNECT,
1171                                    RDMA_CM_DISCONNECT))
1172                         goto out;
1173                 event.event = RDMA_CM_EVENT_DISCONNECTED;
1174                 break;
1175         case IB_CM_TIMEWAIT_EXIT:
1176                 event.event = RDMA_CM_EVENT_TIMEWAIT_EXIT;
1177                 break;
1178         case IB_CM_MRA_RECEIVED:
1179                 /* ignore event */
1180                 goto out;
1181         case IB_CM_REJ_RECEIVED:
1182                 cma_modify_qp_err(id_priv);
1183                 event.status = ib_event->param.rej_rcvd.reason;
1184                 event.event = RDMA_CM_EVENT_REJECTED;
1185                 event.param.conn.private_data = ib_event->private_data;
1186                 event.param.conn.private_data_len = IB_CM_REJ_PRIVATE_DATA_SIZE;
1187                 break;
1188         default:
1189                 printk(KERN_ERR "RDMA CMA: unexpected IB CM event: %d\n",
1190                        ib_event->event);
1191                 goto out;
1192         }
1193
1194         ret = id_priv->id.event_handler(&id_priv->id, &event);
1195         if (ret) {
1196                 /* Destroy the CM ID by returning a non-zero value. */
1197                 id_priv->cm_id.ib = NULL;
1198                 cma_exch(id_priv, RDMA_CM_DESTROYING);
1199                 mutex_unlock(&id_priv->handler_mutex);
1200                 rdma_destroy_id(&id_priv->id);
1201                 return ret;
1202         }
1203 out:
1204         mutex_unlock(&id_priv->handler_mutex);
1205         return ret;
1206 }
1207
1208 static struct rdma_id_private *cma_new_conn_id(struct rdma_cm_id *listen_id,
1209                                                struct ib_cm_event *ib_event)
1210 {
1211         struct rdma_id_private *id_priv;
1212         struct rdma_cm_id *id;
1213         struct rdma_route *rt;
1214         int ret;
1215
1216         id = rdma_create_id(listen_id->event_handler, listen_id->context,
1217                             listen_id->ps, ib_event->param.req_rcvd.qp_type);
1218         if (IS_ERR(id))
1219                 return NULL;
1220
1221         id_priv = container_of(id, struct rdma_id_private, id);
1222         if (cma_save_net_info(id, listen_id, ib_event))
1223                 goto err;
1224
1225         rt = &id->route;
1226         rt->num_paths = ib_event->param.req_rcvd.alternate_path ? 2 : 1;
1227         rt->path_rec = kmalloc(sizeof *rt->path_rec * rt->num_paths,
1228                                GFP_KERNEL);
1229         if (!rt->path_rec)
1230                 goto err;
1231
1232         rt->path_rec[0] = *ib_event->param.req_rcvd.primary_path;
1233         if (rt->num_paths == 2)
1234                 rt->path_rec[1] = *ib_event->param.req_rcvd.alternate_path;
1235
1236         if (cma_any_addr(cma_src_addr(id_priv))) {
1237                 rt->addr.dev_addr.dev_type = ARPHRD_INFINIBAND;
1238                 rdma_addr_set_sgid(&rt->addr.dev_addr, &rt->path_rec[0].sgid);
1239                 ib_addr_set_pkey(&rt->addr.dev_addr, be16_to_cpu(rt->path_rec[0].pkey));
1240         } else {
1241                 ret = cma_translate_addr(cma_src_addr(id_priv), &rt->addr.dev_addr);
1242                 if (ret)
1243                         goto err;
1244         }
1245         rdma_addr_set_dgid(&rt->addr.dev_addr, &rt->path_rec[0].dgid);
1246
1247         id_priv->state = RDMA_CM_CONNECT;
1248         return id_priv;
1249
1250 err:
1251         rdma_destroy_id(id);
1252         return NULL;
1253 }
1254
1255 static struct rdma_id_private *cma_new_udp_id(struct rdma_cm_id *listen_id,
1256                                               struct ib_cm_event *ib_event)
1257 {
1258         struct rdma_id_private *id_priv;
1259         struct rdma_cm_id *id;
1260         int ret;
1261
1262         id = rdma_create_id(listen_id->event_handler, listen_id->context,
1263                             listen_id->ps, IB_QPT_UD);
1264         if (IS_ERR(id))
1265                 return NULL;
1266
1267         id_priv = container_of(id, struct rdma_id_private, id);
1268         if (cma_save_net_info(id, listen_id, ib_event))
1269                 goto err;
1270
1271         if (!cma_any_addr((struct sockaddr *) &id->route.addr.src_addr)) {
1272                 ret = cma_translate_addr(cma_src_addr(id_priv), &id->route.addr.dev_addr);
1273                 if (ret)
1274                         goto err;
1275         }
1276
1277         id_priv->state = RDMA_CM_CONNECT;
1278         return id_priv;
1279 err:
1280         rdma_destroy_id(id);
1281         return NULL;
1282 }
1283
1284 static void cma_set_req_event_data(struct rdma_cm_event *event,
1285                                    struct ib_cm_req_event_param *req_data,
1286                                    void *private_data, int offset)
1287 {
1288         event->param.conn.private_data = private_data + offset;
1289         event->param.conn.private_data_len = IB_CM_REQ_PRIVATE_DATA_SIZE - offset;
1290         event->param.conn.responder_resources = req_data->responder_resources;
1291         event->param.conn.initiator_depth = req_data->initiator_depth;
1292         event->param.conn.flow_control = req_data->flow_control;
1293         event->param.conn.retry_count = req_data->retry_count;
1294         event->param.conn.rnr_retry_count = req_data->rnr_retry_count;
1295         event->param.conn.srq = req_data->srq;
1296         event->param.conn.qp_num = req_data->remote_qpn;
1297 }
1298
1299 static int cma_check_req_qp_type(struct rdma_cm_id *id, struct ib_cm_event *ib_event)
1300 {
1301         return (((ib_event->event == IB_CM_REQ_RECEIVED) &&
1302                  (ib_event->param.req_rcvd.qp_type == id->qp_type)) ||
1303                 ((ib_event->event == IB_CM_SIDR_REQ_RECEIVED) &&
1304                  (id->qp_type == IB_QPT_UD)) ||
1305                 (!id->qp_type));
1306 }
1307
1308 static int cma_req_handler(struct ib_cm_id *cm_id, struct ib_cm_event *ib_event)
1309 {
1310         struct rdma_id_private *listen_id, *conn_id;
1311         struct rdma_cm_event event;
1312         int offset, ret;
1313
1314         listen_id = cm_id->context;
1315         if (!cma_check_req_qp_type(&listen_id->id, ib_event))
1316                 return -EINVAL;
1317
1318         if (cma_disable_callback(listen_id, RDMA_CM_LISTEN))
1319                 return -ECONNABORTED;
1320
1321         memset(&event, 0, sizeof event);
1322         offset = cma_user_data_offset(listen_id);
1323         event.event = RDMA_CM_EVENT_CONNECT_REQUEST;
1324         if (ib_event->event == IB_CM_SIDR_REQ_RECEIVED) {
1325                 conn_id = cma_new_udp_id(&listen_id->id, ib_event);
1326                 event.param.ud.private_data = ib_event->private_data + offset;
1327                 event.param.ud.private_data_len =
1328                                 IB_CM_SIDR_REQ_PRIVATE_DATA_SIZE - offset;
1329         } else {
1330                 conn_id = cma_new_conn_id(&listen_id->id, ib_event);
1331                 cma_set_req_event_data(&event, &ib_event->param.req_rcvd,
1332                                        ib_event->private_data, offset);
1333         }
1334         if (!conn_id) {
1335                 ret = -ENOMEM;
1336                 goto err1;
1337         }
1338
1339         mutex_lock_nested(&conn_id->handler_mutex, SINGLE_DEPTH_NESTING);
1340         ret = cma_acquire_dev(conn_id, listen_id);
1341         if (ret)
1342                 goto err2;
1343
1344         conn_id->cm_id.ib = cm_id;
1345         cm_id->context = conn_id;
1346         cm_id->cm_handler = cma_ib_handler;
1347
1348         /*
1349          * Protect against the user destroying conn_id from another thread
1350          * until we're done accessing it.
1351          */
1352         atomic_inc(&conn_id->refcount);
1353         ret = conn_id->id.event_handler(&conn_id->id, &event);
1354         if (ret)
1355                 goto err3;
1356         /*
1357          * Acquire mutex to prevent user executing rdma_destroy_id()
1358          * while we're accessing the cm_id.
1359          */
1360         mutex_lock(&lock);
1361         if (cma_comp(conn_id, RDMA_CM_CONNECT) &&
1362             (conn_id->id.qp_type != IB_QPT_UD))
1363                 ib_send_cm_mra(cm_id, CMA_CM_MRA_SETTING, NULL, 0);
1364         mutex_unlock(&lock);
1365         mutex_unlock(&conn_id->handler_mutex);
1366         mutex_unlock(&listen_id->handler_mutex);
1367         cma_deref_id(conn_id);
1368         return 0;
1369
1370 err3:
1371         cma_deref_id(conn_id);
1372         /* Destroy the CM ID by returning a non-zero value. */
1373         conn_id->cm_id.ib = NULL;
1374 err2:
1375         cma_exch(conn_id, RDMA_CM_DESTROYING);
1376         mutex_unlock(&conn_id->handler_mutex);
1377 err1:
1378         mutex_unlock(&listen_id->handler_mutex);
1379         if (conn_id)
1380                 rdma_destroy_id(&conn_id->id);
1381         return ret;
1382 }
1383
1384 __be64 rdma_get_service_id(struct rdma_cm_id *id, struct sockaddr *addr)
1385 {
1386         if (addr->sa_family == AF_IB)
1387                 return ((struct sockaddr_ib *) addr)->sib_sid;
1388
1389         return cpu_to_be64(((u64)id->ps << 16) + be16_to_cpu(cma_port(addr)));
1390 }
1391 EXPORT_SYMBOL(rdma_get_service_id);
1392
1393 static void cma_set_compare_data(enum rdma_port_space ps, struct sockaddr *addr,
1394                                  struct ib_cm_compare_data *compare)
1395 {
1396         struct cma_hdr *cma_data, *cma_mask;
1397         __be32 ip4_addr;
1398         struct in6_addr ip6_addr;
1399
1400         memset(compare, 0, sizeof *compare);
1401         cma_data = (void *) compare->data;
1402         cma_mask = (void *) compare->mask;
1403
1404         switch (addr->sa_family) {
1405         case AF_INET:
1406                 ip4_addr = ((struct sockaddr_in *) addr)->sin_addr.s_addr;
1407                 cma_set_ip_ver(cma_data, 4);
1408                 cma_set_ip_ver(cma_mask, 0xF);
1409                 if (!cma_any_addr(addr)) {
1410                         cma_data->dst_addr.ip4.addr = ip4_addr;
1411                         cma_mask->dst_addr.ip4.addr = htonl(~0);
1412                 }
1413                 break;
1414         case AF_INET6:
1415                 ip6_addr = ((struct sockaddr_in6 *) addr)->sin6_addr;
1416                 cma_set_ip_ver(cma_data, 6);
1417                 cma_set_ip_ver(cma_mask, 0xF);
1418                 if (!cma_any_addr(addr)) {
1419                         cma_data->dst_addr.ip6 = ip6_addr;
1420                         memset(&cma_mask->dst_addr.ip6, 0xFF,
1421                                sizeof cma_mask->dst_addr.ip6);
1422                 }
1423                 break;
1424         default:
1425                 break;
1426         }
1427 }
1428
1429 static int cma_iw_handler(struct iw_cm_id *iw_id, struct iw_cm_event *iw_event)
1430 {
1431         struct rdma_id_private *id_priv = iw_id->context;
1432         struct rdma_cm_event event;
1433         int ret = 0;
1434         struct sockaddr *laddr = (struct sockaddr *)&iw_event->local_addr;
1435         struct sockaddr *raddr = (struct sockaddr *)&iw_event->remote_addr;
1436
1437         if (cma_disable_callback(id_priv, RDMA_CM_CONNECT))
1438                 return 0;
1439
1440         memset(&event, 0, sizeof event);
1441         switch (iw_event->event) {
1442         case IW_CM_EVENT_CLOSE:
1443                 event.event = RDMA_CM_EVENT_DISCONNECTED;
1444                 break;
1445         case IW_CM_EVENT_CONNECT_REPLY:
1446                 memcpy(cma_src_addr(id_priv), laddr,
1447                        rdma_addr_size(laddr));
1448                 memcpy(cma_dst_addr(id_priv), raddr,
1449                        rdma_addr_size(raddr));
1450                 switch (iw_event->status) {
1451                 case 0:
1452                         event.event = RDMA_CM_EVENT_ESTABLISHED;
1453                         event.param.conn.initiator_depth = iw_event->ird;
1454                         event.param.conn.responder_resources = iw_event->ord;
1455                         break;
1456                 case -ECONNRESET:
1457                 case -ECONNREFUSED:
1458                         event.event = RDMA_CM_EVENT_REJECTED;
1459                         break;
1460                 case -ETIMEDOUT:
1461                         event.event = RDMA_CM_EVENT_UNREACHABLE;
1462                         break;
1463                 default:
1464                         event.event = RDMA_CM_EVENT_CONNECT_ERROR;
1465                         break;
1466                 }
1467                 break;
1468         case IW_CM_EVENT_ESTABLISHED:
1469                 event.event = RDMA_CM_EVENT_ESTABLISHED;
1470                 event.param.conn.initiator_depth = iw_event->ird;
1471                 event.param.conn.responder_resources = iw_event->ord;
1472                 break;
1473         default:
1474                 BUG_ON(1);
1475         }
1476
1477         event.status = iw_event->status;
1478         event.param.conn.private_data = iw_event->private_data;
1479         event.param.conn.private_data_len = iw_event->private_data_len;
1480         ret = id_priv->id.event_handler(&id_priv->id, &event);
1481         if (ret) {
1482                 /* Destroy the CM ID by returning a non-zero value. */
1483                 id_priv->cm_id.iw = NULL;
1484                 cma_exch(id_priv, RDMA_CM_DESTROYING);
1485                 mutex_unlock(&id_priv->handler_mutex);
1486                 rdma_destroy_id(&id_priv->id);
1487                 return ret;
1488         }
1489
1490         mutex_unlock(&id_priv->handler_mutex);
1491         return ret;
1492 }
1493
1494 static int iw_conn_req_handler(struct iw_cm_id *cm_id,
1495                                struct iw_cm_event *iw_event)
1496 {
1497         struct rdma_cm_id *new_cm_id;
1498         struct rdma_id_private *listen_id, *conn_id;
1499         struct rdma_cm_event event;
1500         int ret;
1501         struct ib_device_attr attr;
1502         struct sockaddr *laddr = (struct sockaddr *)&iw_event->local_addr;
1503         struct sockaddr *raddr = (struct sockaddr *)&iw_event->remote_addr;
1504
1505         listen_id = cm_id->context;
1506         if (cma_disable_callback(listen_id, RDMA_CM_LISTEN))
1507                 return -ECONNABORTED;
1508
1509         /* Create a new RDMA id for the new IW CM ID */
1510         new_cm_id = rdma_create_id(listen_id->id.event_handler,
1511                                    listen_id->id.context,
1512                                    RDMA_PS_TCP, IB_QPT_RC);
1513         if (IS_ERR(new_cm_id)) {
1514                 ret = -ENOMEM;
1515                 goto out;
1516         }
1517         conn_id = container_of(new_cm_id, struct rdma_id_private, id);
1518         mutex_lock_nested(&conn_id->handler_mutex, SINGLE_DEPTH_NESTING);
1519         conn_id->state = RDMA_CM_CONNECT;
1520
1521         ret = rdma_translate_ip(laddr, &conn_id->id.route.addr.dev_addr, NULL);
1522         if (ret) {
1523                 mutex_unlock(&conn_id->handler_mutex);
1524                 rdma_destroy_id(new_cm_id);
1525                 goto out;
1526         }
1527
1528         ret = cma_acquire_dev(conn_id, listen_id);
1529         if (ret) {
1530                 mutex_unlock(&conn_id->handler_mutex);
1531                 rdma_destroy_id(new_cm_id);
1532                 goto out;
1533         }
1534
1535         conn_id->cm_id.iw = cm_id;
1536         cm_id->context = conn_id;
1537         cm_id->cm_handler = cma_iw_handler;
1538
1539         memcpy(cma_src_addr(conn_id), laddr, rdma_addr_size(laddr));
1540         memcpy(cma_dst_addr(conn_id), raddr, rdma_addr_size(raddr));
1541
1542         ret = ib_query_device(conn_id->id.device, &attr);
1543         if (ret) {
1544                 mutex_unlock(&conn_id->handler_mutex);
1545                 rdma_destroy_id(new_cm_id);
1546                 goto out;
1547         }
1548
1549         memset(&event, 0, sizeof event);
1550         event.event = RDMA_CM_EVENT_CONNECT_REQUEST;
1551         event.param.conn.private_data = iw_event->private_data;
1552         event.param.conn.private_data_len = iw_event->private_data_len;
1553         event.param.conn.initiator_depth = iw_event->ird;
1554         event.param.conn.responder_resources = iw_event->ord;
1555
1556         /*
1557          * Protect against the user destroying conn_id from another thread
1558          * until we're done accessing it.
1559          */
1560         atomic_inc(&conn_id->refcount);
1561         ret = conn_id->id.event_handler(&conn_id->id, &event);
1562         if (ret) {
1563                 /* User wants to destroy the CM ID */
1564                 conn_id->cm_id.iw = NULL;
1565                 cma_exch(conn_id, RDMA_CM_DESTROYING);
1566                 mutex_unlock(&conn_id->handler_mutex);
1567                 cma_deref_id(conn_id);
1568                 rdma_destroy_id(&conn_id->id);
1569                 goto out;
1570         }
1571
1572         mutex_unlock(&conn_id->handler_mutex);
1573         cma_deref_id(conn_id);
1574
1575 out:
1576         mutex_unlock(&listen_id->handler_mutex);
1577         return ret;
1578 }
1579
1580 static int cma_ib_listen(struct rdma_id_private *id_priv)
1581 {
1582         struct ib_cm_compare_data compare_data;
1583         struct sockaddr *addr;
1584         struct ib_cm_id *id;
1585         __be64 svc_id;
1586         int ret;
1587
1588         id = ib_create_cm_id(id_priv->id.device, cma_req_handler, id_priv);
1589         if (IS_ERR(id))
1590                 return PTR_ERR(id);
1591
1592         id_priv->cm_id.ib = id;
1593
1594         addr = cma_src_addr(id_priv);
1595         svc_id = rdma_get_service_id(&id_priv->id, addr);
1596         if (cma_any_addr(addr) && !id_priv->afonly)
1597                 ret = ib_cm_listen(id_priv->cm_id.ib, svc_id, 0, NULL);
1598         else {
1599                 cma_set_compare_data(id_priv->id.ps, addr, &compare_data);
1600                 ret = ib_cm_listen(id_priv->cm_id.ib, svc_id, 0, &compare_data);
1601         }
1602
1603         if (ret) {
1604                 ib_destroy_cm_id(id_priv->cm_id.ib);
1605                 id_priv->cm_id.ib = NULL;
1606         }
1607
1608         return ret;
1609 }
1610
1611 static int cma_iw_listen(struct rdma_id_private *id_priv, int backlog)
1612 {
1613         int ret;
1614         struct iw_cm_id *id;
1615
1616         id = iw_create_cm_id(id_priv->id.device,
1617                              iw_conn_req_handler,
1618                              id_priv);
1619         if (IS_ERR(id))
1620                 return PTR_ERR(id);
1621
1622         id_priv->cm_id.iw = id;
1623
1624         memcpy(&id_priv->cm_id.iw->local_addr, cma_src_addr(id_priv),
1625                rdma_addr_size(cma_src_addr(id_priv)));
1626
1627         ret = iw_cm_listen(id_priv->cm_id.iw, backlog);
1628
1629         if (ret) {
1630                 iw_destroy_cm_id(id_priv->cm_id.iw);
1631                 id_priv->cm_id.iw = NULL;
1632         }
1633
1634         return ret;
1635 }
1636
1637 static int cma_listen_handler(struct rdma_cm_id *id,
1638                               struct rdma_cm_event *event)
1639 {
1640         struct rdma_id_private *id_priv = id->context;
1641
1642         id->context = id_priv->id.context;
1643         id->event_handler = id_priv->id.event_handler;
1644         return id_priv->id.event_handler(id, event);
1645 }
1646
1647 static void cma_listen_on_dev(struct rdma_id_private *id_priv,
1648                               struct cma_device *cma_dev)
1649 {
1650         struct rdma_id_private *dev_id_priv;
1651         struct rdma_cm_id *id;
1652         int ret;
1653
1654         if (cma_family(id_priv) == AF_IB && !rdma_cap_ib_cm(cma_dev->device, 1))
1655                 return;
1656
1657         id = rdma_create_id(cma_listen_handler, id_priv, id_priv->id.ps,
1658                             id_priv->id.qp_type);
1659         if (IS_ERR(id))
1660                 return;
1661
1662         dev_id_priv = container_of(id, struct rdma_id_private, id);
1663
1664         dev_id_priv->state = RDMA_CM_ADDR_BOUND;
1665         memcpy(cma_src_addr(dev_id_priv), cma_src_addr(id_priv),
1666                rdma_addr_size(cma_src_addr(id_priv)));
1667
1668         cma_attach_to_dev(dev_id_priv, cma_dev);
1669         list_add_tail(&dev_id_priv->listen_list, &id_priv->listen_list);
1670         atomic_inc(&id_priv->refcount);
1671         dev_id_priv->internal_id = 1;
1672         dev_id_priv->afonly = id_priv->afonly;
1673
1674         ret = rdma_listen(id, id_priv->backlog);
1675         if (ret)
1676                 printk(KERN_WARNING "RDMA CMA: cma_listen_on_dev, error %d, "
1677                        "listening on device %s\n", ret, cma_dev->device->name);
1678 }
1679
1680 static void cma_listen_on_all(struct rdma_id_private *id_priv)
1681 {
1682         struct cma_device *cma_dev;
1683
1684         mutex_lock(&lock);
1685         list_add_tail(&id_priv->list, &listen_any_list);
1686         list_for_each_entry(cma_dev, &dev_list, list)
1687                 cma_listen_on_dev(id_priv, cma_dev);
1688         mutex_unlock(&lock);
1689 }
1690
1691 void rdma_set_service_type(struct rdma_cm_id *id, int tos)
1692 {
1693         struct rdma_id_private *id_priv;
1694
1695         id_priv = container_of(id, struct rdma_id_private, id);
1696         id_priv->tos = (u8) tos;
1697 }
1698 EXPORT_SYMBOL(rdma_set_service_type);
1699
1700 static void cma_query_handler(int status, struct ib_sa_path_rec *path_rec,
1701                               void *context)
1702 {
1703         struct cma_work *work = context;
1704         struct rdma_route *route;
1705
1706         route = &work->id->id.route;
1707
1708         if (!status) {
1709                 route->num_paths = 1;
1710                 *route->path_rec = *path_rec;
1711         } else {
1712                 work->old_state = RDMA_CM_ROUTE_QUERY;
1713                 work->new_state = RDMA_CM_ADDR_RESOLVED;
1714                 work->event.event = RDMA_CM_EVENT_ROUTE_ERROR;
1715                 work->event.status = status;
1716         }
1717
1718         queue_work(cma_wq, &work->work);
1719 }
1720
1721 static int cma_query_ib_route(struct rdma_id_private *id_priv, int timeout_ms,
1722                               struct cma_work *work)
1723 {
1724         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
1725         struct ib_sa_path_rec path_rec;
1726         ib_sa_comp_mask comp_mask;
1727         struct sockaddr_in6 *sin6;
1728         struct sockaddr_ib *sib;
1729
1730         memset(&path_rec, 0, sizeof path_rec);
1731         rdma_addr_get_sgid(dev_addr, &path_rec.sgid);
1732         rdma_addr_get_dgid(dev_addr, &path_rec.dgid);
1733         path_rec.pkey = cpu_to_be16(ib_addr_get_pkey(dev_addr));
1734         path_rec.numb_path = 1;
1735         path_rec.reversible = 1;
1736         path_rec.service_id = rdma_get_service_id(&id_priv->id, cma_dst_addr(id_priv));
1737
1738         comp_mask = IB_SA_PATH_REC_DGID | IB_SA_PATH_REC_SGID |
1739                     IB_SA_PATH_REC_PKEY | IB_SA_PATH_REC_NUMB_PATH |
1740                     IB_SA_PATH_REC_REVERSIBLE | IB_SA_PATH_REC_SERVICE_ID;
1741
1742         switch (cma_family(id_priv)) {
1743         case AF_INET:
1744                 path_rec.qos_class = cpu_to_be16((u16) id_priv->tos);
1745                 comp_mask |= IB_SA_PATH_REC_QOS_CLASS;
1746                 break;
1747         case AF_INET6:
1748                 sin6 = (struct sockaddr_in6 *) cma_src_addr(id_priv);
1749                 path_rec.traffic_class = (u8) (be32_to_cpu(sin6->sin6_flowinfo) >> 20);
1750                 comp_mask |= IB_SA_PATH_REC_TRAFFIC_CLASS;
1751                 break;
1752         case AF_IB:
1753                 sib = (struct sockaddr_ib *) cma_src_addr(id_priv);
1754                 path_rec.traffic_class = (u8) (be32_to_cpu(sib->sib_flowinfo) >> 20);
1755                 comp_mask |= IB_SA_PATH_REC_TRAFFIC_CLASS;
1756                 break;
1757         }
1758
1759         id_priv->query_id = ib_sa_path_rec_get(&sa_client, id_priv->id.device,
1760                                                id_priv->id.port_num, &path_rec,
1761                                                comp_mask, timeout_ms,
1762                                                GFP_KERNEL, cma_query_handler,
1763                                                work, &id_priv->query);
1764
1765         return (id_priv->query_id < 0) ? id_priv->query_id : 0;
1766 }
1767
1768 static void cma_work_handler(struct work_struct *_work)
1769 {
1770         struct cma_work *work = container_of(_work, struct cma_work, work);
1771         struct rdma_id_private *id_priv = work->id;
1772         int destroy = 0;
1773
1774         mutex_lock(&id_priv->handler_mutex);
1775         if (!cma_comp_exch(id_priv, work->old_state, work->new_state))
1776                 goto out;
1777
1778         if (id_priv->id.event_handler(&id_priv->id, &work->event)) {
1779                 cma_exch(id_priv, RDMA_CM_DESTROYING);
1780                 destroy = 1;
1781         }
1782 out:
1783         mutex_unlock(&id_priv->handler_mutex);
1784         cma_deref_id(id_priv);
1785         if (destroy)
1786                 rdma_destroy_id(&id_priv->id);
1787         kfree(work);
1788 }
1789
1790 static void cma_ndev_work_handler(struct work_struct *_work)
1791 {
1792         struct cma_ndev_work *work = container_of(_work, struct cma_ndev_work, work);
1793         struct rdma_id_private *id_priv = work->id;
1794         int destroy = 0;
1795
1796         mutex_lock(&id_priv->handler_mutex);
1797         if (id_priv->state == RDMA_CM_DESTROYING ||
1798             id_priv->state == RDMA_CM_DEVICE_REMOVAL)
1799                 goto out;
1800
1801         if (id_priv->id.event_handler(&id_priv->id, &work->event)) {
1802                 cma_exch(id_priv, RDMA_CM_DESTROYING);
1803                 destroy = 1;
1804         }
1805
1806 out:
1807         mutex_unlock(&id_priv->handler_mutex);
1808         cma_deref_id(id_priv);
1809         if (destroy)
1810                 rdma_destroy_id(&id_priv->id);
1811         kfree(work);
1812 }
1813
1814 static int cma_resolve_ib_route(struct rdma_id_private *id_priv, int timeout_ms)
1815 {
1816         struct rdma_route *route = &id_priv->id.route;
1817         struct cma_work *work;
1818         int ret;
1819
1820         work = kzalloc(sizeof *work, GFP_KERNEL);
1821         if (!work)
1822                 return -ENOMEM;
1823
1824         work->id = id_priv;
1825         INIT_WORK(&work->work, cma_work_handler);
1826         work->old_state = RDMA_CM_ROUTE_QUERY;
1827         work->new_state = RDMA_CM_ROUTE_RESOLVED;
1828         work->event.event = RDMA_CM_EVENT_ROUTE_RESOLVED;
1829
1830         route->path_rec = kmalloc(sizeof *route->path_rec, GFP_KERNEL);
1831         if (!route->path_rec) {
1832                 ret = -ENOMEM;
1833                 goto err1;
1834         }
1835
1836         ret = cma_query_ib_route(id_priv, timeout_ms, work);
1837         if (ret)
1838                 goto err2;
1839
1840         return 0;
1841 err2:
1842         kfree(route->path_rec);
1843         route->path_rec = NULL;
1844 err1:
1845         kfree(work);
1846         return ret;
1847 }
1848
1849 int rdma_set_ib_paths(struct rdma_cm_id *id,
1850                       struct ib_sa_path_rec *path_rec, int num_paths)
1851 {
1852         struct rdma_id_private *id_priv;
1853         int ret;
1854
1855         id_priv = container_of(id, struct rdma_id_private, id);
1856         if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_RESOLVED,
1857                            RDMA_CM_ROUTE_RESOLVED))
1858                 return -EINVAL;
1859
1860         id->route.path_rec = kmemdup(path_rec, sizeof *path_rec * num_paths,
1861                                      GFP_KERNEL);
1862         if (!id->route.path_rec) {
1863                 ret = -ENOMEM;
1864                 goto err;
1865         }
1866
1867         id->route.num_paths = num_paths;
1868         return 0;
1869 err:
1870         cma_comp_exch(id_priv, RDMA_CM_ROUTE_RESOLVED, RDMA_CM_ADDR_RESOLVED);
1871         return ret;
1872 }
1873 EXPORT_SYMBOL(rdma_set_ib_paths);
1874
1875 static int cma_resolve_iw_route(struct rdma_id_private *id_priv, int timeout_ms)
1876 {
1877         struct cma_work *work;
1878
1879         work = kzalloc(sizeof *work, GFP_KERNEL);
1880         if (!work)
1881                 return -ENOMEM;
1882
1883         work->id = id_priv;
1884         INIT_WORK(&work->work, cma_work_handler);
1885         work->old_state = RDMA_CM_ROUTE_QUERY;
1886         work->new_state = RDMA_CM_ROUTE_RESOLVED;
1887         work->event.event = RDMA_CM_EVENT_ROUTE_RESOLVED;
1888         queue_work(cma_wq, &work->work);
1889         return 0;
1890 }
1891
1892 static int iboe_tos_to_sl(struct net_device *ndev, int tos)
1893 {
1894         int prio;
1895         struct net_device *dev;
1896
1897         prio = rt_tos2priority(tos);
1898         dev = ndev->priv_flags & IFF_802_1Q_VLAN ?
1899                 vlan_dev_real_dev(ndev) : ndev;
1900
1901         if (dev->num_tc)
1902                 return netdev_get_prio_tc_map(dev, prio);
1903
1904 #if IS_ENABLED(CONFIG_VLAN_8021Q)
1905         if (ndev->priv_flags & IFF_802_1Q_VLAN)
1906                 return (vlan_dev_get_egress_qos_mask(ndev, prio) &
1907                         VLAN_PRIO_MASK) >> VLAN_PRIO_SHIFT;
1908 #endif
1909         return 0;
1910 }
1911
1912 static int cma_resolve_iboe_route(struct rdma_id_private *id_priv)
1913 {
1914         struct rdma_route *route = &id_priv->id.route;
1915         struct rdma_addr *addr = &route->addr;
1916         struct cma_work *work;
1917         int ret;
1918         struct net_device *ndev = NULL;
1919
1920
1921         work = kzalloc(sizeof *work, GFP_KERNEL);
1922         if (!work)
1923                 return -ENOMEM;
1924
1925         work->id = id_priv;
1926         INIT_WORK(&work->work, cma_work_handler);
1927
1928         route->path_rec = kzalloc(sizeof *route->path_rec, GFP_KERNEL);
1929         if (!route->path_rec) {
1930                 ret = -ENOMEM;
1931                 goto err1;
1932         }
1933
1934         route->num_paths = 1;
1935
1936         if (addr->dev_addr.bound_dev_if)
1937                 ndev = dev_get_by_index(&init_net, addr->dev_addr.bound_dev_if);
1938         if (!ndev) {
1939                 ret = -ENODEV;
1940                 goto err2;
1941         }
1942
1943         route->path_rec->vlan_id = rdma_vlan_dev_vlan_id(ndev);
1944         memcpy(route->path_rec->dmac, addr->dev_addr.dst_dev_addr, ETH_ALEN);
1945         memcpy(route->path_rec->smac, ndev->dev_addr, ndev->addr_len);
1946
1947         rdma_ip2gid((struct sockaddr *)&id_priv->id.route.addr.src_addr,
1948                     &route->path_rec->sgid);
1949         rdma_ip2gid((struct sockaddr *)&id_priv->id.route.addr.dst_addr,
1950                     &route->path_rec->dgid);
1951
1952         route->path_rec->hop_limit = 1;
1953         route->path_rec->reversible = 1;
1954         route->path_rec->pkey = cpu_to_be16(0xffff);
1955         route->path_rec->mtu_selector = IB_SA_EQ;
1956         route->path_rec->sl = iboe_tos_to_sl(ndev, id_priv->tos);
1957         route->path_rec->mtu = iboe_get_mtu(ndev->mtu);
1958         route->path_rec->rate_selector = IB_SA_EQ;
1959         route->path_rec->rate = iboe_get_rate(ndev);
1960         dev_put(ndev);
1961         route->path_rec->packet_life_time_selector = IB_SA_EQ;
1962         route->path_rec->packet_life_time = CMA_IBOE_PACKET_LIFETIME;
1963         if (!route->path_rec->mtu) {
1964                 ret = -EINVAL;
1965                 goto err2;
1966         }
1967
1968         work->old_state = RDMA_CM_ROUTE_QUERY;
1969         work->new_state = RDMA_CM_ROUTE_RESOLVED;
1970         work->event.event = RDMA_CM_EVENT_ROUTE_RESOLVED;
1971         work->event.status = 0;
1972
1973         queue_work(cma_wq, &work->work);
1974
1975         return 0;
1976
1977 err2:
1978         kfree(route->path_rec);
1979         route->path_rec = NULL;
1980 err1:
1981         kfree(work);
1982         return ret;
1983 }
1984
1985 int rdma_resolve_route(struct rdma_cm_id *id, int timeout_ms)
1986 {
1987         struct rdma_id_private *id_priv;
1988         int ret;
1989
1990         id_priv = container_of(id, struct rdma_id_private, id);
1991         if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_RESOLVED, RDMA_CM_ROUTE_QUERY))
1992                 return -EINVAL;
1993
1994         atomic_inc(&id_priv->refcount);
1995         if (rdma_cap_ib_sa(id->device, id->port_num))
1996                 ret = cma_resolve_ib_route(id_priv, timeout_ms);
1997         else if (rdma_protocol_roce(id->device, id->port_num))
1998                 ret = cma_resolve_iboe_route(id_priv);
1999         else if (rdma_protocol_iwarp(id->device, id->port_num))
2000                 ret = cma_resolve_iw_route(id_priv, timeout_ms);
2001         else
2002                 ret = -ENOSYS;
2003
2004         if (ret)
2005                 goto err;
2006
2007         return 0;
2008 err:
2009         cma_comp_exch(id_priv, RDMA_CM_ROUTE_QUERY, RDMA_CM_ADDR_RESOLVED);
2010         cma_deref_id(id_priv);
2011         return ret;
2012 }
2013 EXPORT_SYMBOL(rdma_resolve_route);
2014
2015 static void cma_set_loopback(struct sockaddr *addr)
2016 {
2017         switch (addr->sa_family) {
2018         case AF_INET:
2019                 ((struct sockaddr_in *) addr)->sin_addr.s_addr = htonl(INADDR_LOOPBACK);
2020                 break;
2021         case AF_INET6:
2022                 ipv6_addr_set(&((struct sockaddr_in6 *) addr)->sin6_addr,
2023                               0, 0, 0, htonl(1));
2024                 break;
2025         default:
2026                 ib_addr_set(&((struct sockaddr_ib *) addr)->sib_addr,
2027                             0, 0, 0, htonl(1));
2028                 break;
2029         }
2030 }
2031
2032 static int cma_bind_loopback(struct rdma_id_private *id_priv)
2033 {
2034         struct cma_device *cma_dev, *cur_dev;
2035         struct ib_port_attr port_attr;
2036         union ib_gid gid;
2037         u16 pkey;
2038         int ret;
2039         u8 p;
2040
2041         cma_dev = NULL;
2042         mutex_lock(&lock);
2043         list_for_each_entry(cur_dev, &dev_list, list) {
2044                 if (cma_family(id_priv) == AF_IB &&
2045                     !rdma_cap_ib_cm(cur_dev->device, 1))
2046                         continue;
2047
2048                 if (!cma_dev)
2049                         cma_dev = cur_dev;
2050
2051                 for (p = 1; p <= cur_dev->device->phys_port_cnt; ++p) {
2052                         if (!ib_query_port(cur_dev->device, p, &port_attr) &&
2053                             port_attr.state == IB_PORT_ACTIVE) {
2054                                 cma_dev = cur_dev;
2055                                 goto port_found;
2056                         }
2057                 }
2058         }
2059
2060         if (!cma_dev) {
2061                 ret = -ENODEV;
2062                 goto out;
2063         }
2064
2065         p = 1;
2066
2067 port_found:
2068         ret = ib_get_cached_gid(cma_dev->device, p, 0, &gid);
2069         if (ret)
2070                 goto out;
2071
2072         ret = ib_get_cached_pkey(cma_dev->device, p, 0, &pkey);
2073         if (ret)
2074                 goto out;
2075
2076         id_priv->id.route.addr.dev_addr.dev_type =
2077                 (rdma_protocol_ib(cma_dev->device, p)) ?
2078                 ARPHRD_INFINIBAND : ARPHRD_ETHER;
2079
2080         rdma_addr_set_sgid(&id_priv->id.route.addr.dev_addr, &gid);
2081         ib_addr_set_pkey(&id_priv->id.route.addr.dev_addr, pkey);
2082         id_priv->id.port_num = p;
2083         cma_attach_to_dev(id_priv, cma_dev);
2084         cma_set_loopback(cma_src_addr(id_priv));
2085 out:
2086         mutex_unlock(&lock);
2087         return ret;
2088 }
2089
2090 static void addr_handler(int status, struct sockaddr *src_addr,
2091                          struct rdma_dev_addr *dev_addr, void *context)
2092 {
2093         struct rdma_id_private *id_priv = context;
2094         struct rdma_cm_event event;
2095
2096         memset(&event, 0, sizeof event);
2097         mutex_lock(&id_priv->handler_mutex);
2098         if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_QUERY,
2099                            RDMA_CM_ADDR_RESOLVED))
2100                 goto out;
2101
2102         memcpy(cma_src_addr(id_priv), src_addr, rdma_addr_size(src_addr));
2103         if (!status && !id_priv->cma_dev)
2104                 status = cma_acquire_dev(id_priv, NULL);
2105
2106         if (status) {
2107                 if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_RESOLVED,
2108                                    RDMA_CM_ADDR_BOUND))
2109                         goto out;
2110                 event.event = RDMA_CM_EVENT_ADDR_ERROR;
2111                 event.status = status;
2112         } else
2113                 event.event = RDMA_CM_EVENT_ADDR_RESOLVED;
2114
2115         if (id_priv->id.event_handler(&id_priv->id, &event)) {
2116                 cma_exch(id_priv, RDMA_CM_DESTROYING);
2117                 mutex_unlock(&id_priv->handler_mutex);
2118                 cma_deref_id(id_priv);
2119                 rdma_destroy_id(&id_priv->id);
2120                 return;
2121         }
2122 out:
2123         mutex_unlock(&id_priv->handler_mutex);
2124         cma_deref_id(id_priv);
2125 }
2126
2127 static int cma_resolve_loopback(struct rdma_id_private *id_priv)
2128 {
2129         struct cma_work *work;
2130         union ib_gid gid;
2131         int ret;
2132
2133         work = kzalloc(sizeof *work, GFP_KERNEL);
2134         if (!work)
2135                 return -ENOMEM;
2136
2137         if (!id_priv->cma_dev) {
2138                 ret = cma_bind_loopback(id_priv);
2139                 if (ret)
2140                         goto err;
2141         }
2142
2143         rdma_addr_get_sgid(&id_priv->id.route.addr.dev_addr, &gid);
2144         rdma_addr_set_dgid(&id_priv->id.route.addr.dev_addr, &gid);
2145
2146         work->id = id_priv;
2147         INIT_WORK(&work->work, cma_work_handler);
2148         work->old_state = RDMA_CM_ADDR_QUERY;
2149         work->new_state = RDMA_CM_ADDR_RESOLVED;
2150         work->event.event = RDMA_CM_EVENT_ADDR_RESOLVED;
2151         queue_work(cma_wq, &work->work);
2152         return 0;
2153 err:
2154         kfree(work);
2155         return ret;
2156 }
2157
2158 static int cma_resolve_ib_addr(struct rdma_id_private *id_priv)
2159 {
2160         struct cma_work *work;
2161         int ret;
2162
2163         work = kzalloc(sizeof *work, GFP_KERNEL);
2164         if (!work)
2165                 return -ENOMEM;
2166
2167         if (!id_priv->cma_dev) {
2168                 ret = cma_resolve_ib_dev(id_priv);
2169                 if (ret)
2170                         goto err;
2171         }
2172
2173         rdma_addr_set_dgid(&id_priv->id.route.addr.dev_addr, (union ib_gid *)
2174                 &(((struct sockaddr_ib *) &id_priv->id.route.addr.dst_addr)->sib_addr));
2175
2176         work->id = id_priv;
2177         INIT_WORK(&work->work, cma_work_handler);
2178         work->old_state = RDMA_CM_ADDR_QUERY;
2179         work->new_state = RDMA_CM_ADDR_RESOLVED;
2180         work->event.event = RDMA_CM_EVENT_ADDR_RESOLVED;
2181         queue_work(cma_wq, &work->work);
2182         return 0;
2183 err:
2184         kfree(work);
2185         return ret;
2186 }
2187
2188 static int cma_bind_addr(struct rdma_cm_id *id, struct sockaddr *src_addr,
2189                          struct sockaddr *dst_addr)
2190 {
2191         if (!src_addr || !src_addr->sa_family) {
2192                 src_addr = (struct sockaddr *) &id->route.addr.src_addr;
2193                 src_addr->sa_family = dst_addr->sa_family;
2194                 if (dst_addr->sa_family == AF_INET6) {
2195                         ((struct sockaddr_in6 *) src_addr)->sin6_scope_id =
2196                                 ((struct sockaddr_in6 *) dst_addr)->sin6_scope_id;
2197                 } else if (dst_addr->sa_family == AF_IB) {
2198                         ((struct sockaddr_ib *) src_addr)->sib_pkey =
2199                                 ((struct sockaddr_ib *) dst_addr)->sib_pkey;
2200                 }
2201         }
2202         return rdma_bind_addr(id, src_addr);
2203 }
2204
2205 int rdma_resolve_addr(struct rdma_cm_id *id, struct sockaddr *src_addr,
2206                       struct sockaddr *dst_addr, int timeout_ms)
2207 {
2208         struct rdma_id_private *id_priv;
2209         int ret;
2210
2211         id_priv = container_of(id, struct rdma_id_private, id);
2212         if (id_priv->state == RDMA_CM_IDLE) {
2213                 ret = cma_bind_addr(id, src_addr, dst_addr);
2214                 if (ret)
2215                         return ret;
2216         }
2217
2218         if (cma_family(id_priv) != dst_addr->sa_family)
2219                 return -EINVAL;
2220
2221         if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_BOUND, RDMA_CM_ADDR_QUERY))
2222                 return -EINVAL;
2223
2224         atomic_inc(&id_priv->refcount);
2225         memcpy(cma_dst_addr(id_priv), dst_addr, rdma_addr_size(dst_addr));
2226         if (cma_any_addr(dst_addr)) {
2227                 ret = cma_resolve_loopback(id_priv);
2228         } else {
2229                 if (dst_addr->sa_family == AF_IB) {
2230                         ret = cma_resolve_ib_addr(id_priv);
2231                 } else {
2232                         ret = rdma_resolve_ip(&addr_client, cma_src_addr(id_priv),
2233                                               dst_addr, &id->route.addr.dev_addr,
2234                                               timeout_ms, addr_handler, id_priv);
2235                 }
2236         }
2237         if (ret)
2238                 goto err;
2239
2240         return 0;
2241 err:
2242         cma_comp_exch(id_priv, RDMA_CM_ADDR_QUERY, RDMA_CM_ADDR_BOUND);
2243         cma_deref_id(id_priv);
2244         return ret;
2245 }
2246 EXPORT_SYMBOL(rdma_resolve_addr);
2247
2248 int rdma_set_reuseaddr(struct rdma_cm_id *id, int reuse)
2249 {
2250         struct rdma_id_private *id_priv;
2251         unsigned long flags;
2252         int ret;
2253
2254         id_priv = container_of(id, struct rdma_id_private, id);
2255         spin_lock_irqsave(&id_priv->lock, flags);
2256         if (reuse || id_priv->state == RDMA_CM_IDLE) {
2257                 id_priv->reuseaddr = reuse;
2258                 ret = 0;
2259         } else {
2260                 ret = -EINVAL;
2261         }
2262         spin_unlock_irqrestore(&id_priv->lock, flags);
2263         return ret;
2264 }
2265 EXPORT_SYMBOL(rdma_set_reuseaddr);
2266
2267 int rdma_set_afonly(struct rdma_cm_id *id, int afonly)
2268 {
2269         struct rdma_id_private *id_priv;
2270         unsigned long flags;
2271         int ret;
2272
2273         id_priv = container_of(id, struct rdma_id_private, id);
2274         spin_lock_irqsave(&id_priv->lock, flags);
2275         if (id_priv->state == RDMA_CM_IDLE || id_priv->state == RDMA_CM_ADDR_BOUND) {
2276                 id_priv->options |= (1 << CMA_OPTION_AFONLY);
2277                 id_priv->afonly = afonly;
2278                 ret = 0;
2279         } else {
2280                 ret = -EINVAL;
2281         }
2282         spin_unlock_irqrestore(&id_priv->lock, flags);
2283         return ret;
2284 }
2285 EXPORT_SYMBOL(rdma_set_afonly);
2286
2287 static void cma_bind_port(struct rdma_bind_list *bind_list,
2288                           struct rdma_id_private *id_priv)
2289 {
2290         struct sockaddr *addr;
2291         struct sockaddr_ib *sib;
2292         u64 sid, mask;
2293         __be16 port;
2294
2295         addr = cma_src_addr(id_priv);
2296         port = htons(bind_list->port);
2297
2298         switch (addr->sa_family) {
2299         case AF_INET:
2300                 ((struct sockaddr_in *) addr)->sin_port = port;
2301                 break;
2302         case AF_INET6:
2303                 ((struct sockaddr_in6 *) addr)->sin6_port = port;
2304                 break;
2305         case AF_IB:
2306                 sib = (struct sockaddr_ib *) addr;
2307                 sid = be64_to_cpu(sib->sib_sid);
2308                 mask = be64_to_cpu(sib->sib_sid_mask);
2309                 sib->sib_sid = cpu_to_be64((sid & mask) | (u64) ntohs(port));
2310                 sib->sib_sid_mask = cpu_to_be64(~0ULL);
2311                 break;
2312         }
2313         id_priv->bind_list = bind_list;
2314         hlist_add_head(&id_priv->node, &bind_list->owners);
2315 }
2316
2317 static int cma_alloc_port(struct idr *ps, struct rdma_id_private *id_priv,
2318                           unsigned short snum)
2319 {
2320         struct rdma_bind_list *bind_list;
2321         int ret;
2322
2323         bind_list = kzalloc(sizeof *bind_list, GFP_KERNEL);
2324         if (!bind_list)
2325                 return -ENOMEM;
2326
2327         ret = idr_alloc(ps, bind_list, snum, snum + 1, GFP_KERNEL);
2328         if (ret < 0)
2329                 goto err;
2330
2331         bind_list->ps = ps;
2332         bind_list->port = (unsigned short)ret;
2333         cma_bind_port(bind_list, id_priv);
2334         return 0;
2335 err:
2336         kfree(bind_list);
2337         return ret == -ENOSPC ? -EADDRNOTAVAIL : ret;
2338 }
2339
2340 static int cma_alloc_any_port(struct idr *ps, struct rdma_id_private *id_priv)
2341 {
2342         static unsigned int last_used_port;
2343         int low, high, remaining;
2344         unsigned int rover;
2345
2346         inet_get_local_port_range(&init_net, &low, &high);
2347         remaining = (high - low) + 1;
2348         rover = prandom_u32() % remaining + low;
2349 retry:
2350         if (last_used_port != rover &&
2351             !idr_find(ps, (unsigned short) rover)) {
2352                 int ret = cma_alloc_port(ps, id_priv, rover);
2353                 /*
2354                  * Remember previously used port number in order to avoid
2355                  * re-using same port immediately after it is closed.
2356                  */
2357                 if (!ret)
2358                         last_used_port = rover;
2359                 if (ret != -EADDRNOTAVAIL)
2360                         return ret;
2361         }
2362         if (--remaining) {
2363                 rover++;
2364                 if ((rover < low) || (rover > high))
2365                         rover = low;
2366                 goto retry;
2367         }
2368         return -EADDRNOTAVAIL;
2369 }
2370
2371 /*
2372  * Check that the requested port is available.  This is called when trying to
2373  * bind to a specific port, or when trying to listen on a bound port.  In
2374  * the latter case, the provided id_priv may already be on the bind_list, but
2375  * we still need to check that it's okay to start listening.
2376  */
2377 static int cma_check_port(struct rdma_bind_list *bind_list,
2378                           struct rdma_id_private *id_priv, uint8_t reuseaddr)
2379 {
2380         struct rdma_id_private *cur_id;
2381         struct sockaddr *addr, *cur_addr;
2382
2383         addr = cma_src_addr(id_priv);
2384         hlist_for_each_entry(cur_id, &bind_list->owners, node) {
2385                 if (id_priv == cur_id)
2386                         continue;
2387
2388                 if ((cur_id->state != RDMA_CM_LISTEN) && reuseaddr &&
2389                     cur_id->reuseaddr)
2390                         continue;
2391
2392                 cur_addr = cma_src_addr(cur_id);
2393                 if (id_priv->afonly && cur_id->afonly &&
2394                     (addr->sa_family != cur_addr->sa_family))
2395                         continue;
2396
2397                 if (cma_any_addr(addr) || cma_any_addr(cur_addr))
2398                         return -EADDRNOTAVAIL;
2399
2400                 if (!cma_addr_cmp(addr, cur_addr))
2401                         return -EADDRINUSE;
2402         }
2403         return 0;
2404 }
2405
2406 static int cma_use_port(struct idr *ps, struct rdma_id_private *id_priv)
2407 {
2408         struct rdma_bind_list *bind_list;
2409         unsigned short snum;
2410         int ret;
2411
2412         snum = ntohs(cma_port(cma_src_addr(id_priv)));
2413         if (snum < PROT_SOCK && !capable(CAP_NET_BIND_SERVICE))
2414                 return -EACCES;
2415
2416         bind_list = idr_find(ps, snum);
2417         if (!bind_list) {
2418                 ret = cma_alloc_port(ps, id_priv, snum);
2419         } else {
2420                 ret = cma_check_port(bind_list, id_priv, id_priv->reuseaddr);
2421                 if (!ret)
2422                         cma_bind_port(bind_list, id_priv);
2423         }
2424         return ret;
2425 }
2426
2427 static int cma_bind_listen(struct rdma_id_private *id_priv)
2428 {
2429         struct rdma_bind_list *bind_list = id_priv->bind_list;
2430         int ret = 0;
2431
2432         mutex_lock(&lock);
2433         if (bind_list->owners.first->next)
2434                 ret = cma_check_port(bind_list, id_priv, 0);
2435         mutex_unlock(&lock);
2436         return ret;
2437 }
2438
2439 static struct idr *cma_select_inet_ps(struct rdma_id_private *id_priv)
2440 {
2441         switch (id_priv->id.ps) {
2442         case RDMA_PS_TCP:
2443                 return &tcp_ps;
2444         case RDMA_PS_UDP:
2445                 return &udp_ps;
2446         case RDMA_PS_IPOIB:
2447                 return &ipoib_ps;
2448         case RDMA_PS_IB:
2449                 return &ib_ps;
2450         default:
2451                 return NULL;
2452         }
2453 }
2454
2455 static struct idr *cma_select_ib_ps(struct rdma_id_private *id_priv)
2456 {
2457         struct idr *ps = NULL;
2458         struct sockaddr_ib *sib;
2459         u64 sid_ps, mask, sid;
2460
2461         sib = (struct sockaddr_ib *) cma_src_addr(id_priv);
2462         mask = be64_to_cpu(sib->sib_sid_mask) & RDMA_IB_IP_PS_MASK;
2463         sid = be64_to_cpu(sib->sib_sid) & mask;
2464
2465         if ((id_priv->id.ps == RDMA_PS_IB) && (sid == (RDMA_IB_IP_PS_IB & mask))) {
2466                 sid_ps = RDMA_IB_IP_PS_IB;
2467                 ps = &ib_ps;
2468         } else if (((id_priv->id.ps == RDMA_PS_IB) || (id_priv->id.ps == RDMA_PS_TCP)) &&
2469                    (sid == (RDMA_IB_IP_PS_TCP & mask))) {
2470                 sid_ps = RDMA_IB_IP_PS_TCP;
2471                 ps = &tcp_ps;
2472         } else if (((id_priv->id.ps == RDMA_PS_IB) || (id_priv->id.ps == RDMA_PS_UDP)) &&
2473                    (sid == (RDMA_IB_IP_PS_UDP & mask))) {
2474                 sid_ps = RDMA_IB_IP_PS_UDP;
2475                 ps = &udp_ps;
2476         }
2477
2478         if (ps) {
2479                 sib->sib_sid = cpu_to_be64(sid_ps | ntohs(cma_port((struct sockaddr *) sib)));
2480                 sib->sib_sid_mask = cpu_to_be64(RDMA_IB_IP_PS_MASK |
2481                                                 be64_to_cpu(sib->sib_sid_mask));
2482         }
2483         return ps;
2484 }
2485
2486 static int cma_get_port(struct rdma_id_private *id_priv)
2487 {
2488         struct idr *ps;
2489         int ret;
2490
2491         if (cma_family(id_priv) != AF_IB)
2492                 ps = cma_select_inet_ps(id_priv);
2493         else
2494                 ps = cma_select_ib_ps(id_priv);
2495         if (!ps)
2496                 return -EPROTONOSUPPORT;
2497
2498         mutex_lock(&lock);
2499         if (cma_any_port(cma_src_addr(id_priv)))
2500                 ret = cma_alloc_any_port(ps, id_priv);
2501         else
2502                 ret = cma_use_port(ps, id_priv);
2503         mutex_unlock(&lock);
2504
2505         return ret;
2506 }
2507
2508 static int cma_check_linklocal(struct rdma_dev_addr *dev_addr,
2509                                struct sockaddr *addr)
2510 {
2511 #if IS_ENABLED(CONFIG_IPV6)
2512         struct sockaddr_in6 *sin6;
2513
2514         if (addr->sa_family != AF_INET6)
2515                 return 0;
2516
2517         sin6 = (struct sockaddr_in6 *) addr;
2518
2519         if (!(ipv6_addr_type(&sin6->sin6_addr) & IPV6_ADDR_LINKLOCAL))
2520                 return 0;
2521
2522         if (!sin6->sin6_scope_id)
2523                         return -EINVAL;
2524
2525         dev_addr->bound_dev_if = sin6->sin6_scope_id;
2526 #endif
2527         return 0;
2528 }
2529
2530 int rdma_listen(struct rdma_cm_id *id, int backlog)
2531 {
2532         struct rdma_id_private *id_priv;
2533         int ret;
2534
2535         id_priv = container_of(id, struct rdma_id_private, id);
2536         if (id_priv->state == RDMA_CM_IDLE) {
2537                 id->route.addr.src_addr.ss_family = AF_INET;
2538                 ret = rdma_bind_addr(id, cma_src_addr(id_priv));
2539                 if (ret)
2540                         return ret;
2541         }
2542
2543         if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_BOUND, RDMA_CM_LISTEN))
2544                 return -EINVAL;
2545
2546         if (id_priv->reuseaddr) {
2547                 ret = cma_bind_listen(id_priv);
2548                 if (ret)
2549                         goto err;
2550         }
2551
2552         id_priv->backlog = backlog;
2553         if (id->device) {
2554                 if (rdma_cap_ib_cm(id->device, 1)) {
2555                         ret = cma_ib_listen(id_priv);
2556                         if (ret)
2557                                 goto err;
2558                 } else if (rdma_cap_iw_cm(id->device, 1)) {
2559                         ret = cma_iw_listen(id_priv, backlog);
2560                         if (ret)
2561                                 goto err;
2562                 } else {
2563                         ret = -ENOSYS;
2564                         goto err;
2565                 }
2566         } else
2567                 cma_listen_on_all(id_priv);
2568
2569         return 0;
2570 err:
2571         id_priv->backlog = 0;
2572         cma_comp_exch(id_priv, RDMA_CM_LISTEN, RDMA_CM_ADDR_BOUND);
2573         return ret;
2574 }
2575 EXPORT_SYMBOL(rdma_listen);
2576
2577 int rdma_bind_addr(struct rdma_cm_id *id, struct sockaddr *addr)
2578 {
2579         struct rdma_id_private *id_priv;
2580         int ret;
2581
2582         if (addr->sa_family != AF_INET && addr->sa_family != AF_INET6 &&
2583             addr->sa_family != AF_IB)
2584                 return -EAFNOSUPPORT;
2585
2586         id_priv = container_of(id, struct rdma_id_private, id);
2587         if (!cma_comp_exch(id_priv, RDMA_CM_IDLE, RDMA_CM_ADDR_BOUND))
2588                 return -EINVAL;
2589
2590         ret = cma_check_linklocal(&id->route.addr.dev_addr, addr);
2591         if (ret)
2592                 goto err1;
2593
2594         memcpy(cma_src_addr(id_priv), addr, rdma_addr_size(addr));
2595         if (!cma_any_addr(addr)) {
2596                 ret = cma_translate_addr(addr, &id->route.addr.dev_addr);
2597                 if (ret)
2598                         goto err1;
2599
2600                 ret = cma_acquire_dev(id_priv, NULL);
2601                 if (ret)
2602                         goto err1;
2603         }
2604
2605         if (!(id_priv->options & (1 << CMA_OPTION_AFONLY))) {
2606                 if (addr->sa_family == AF_INET)
2607                         id_priv->afonly = 1;
2608 #if IS_ENABLED(CONFIG_IPV6)
2609                 else if (addr->sa_family == AF_INET6)
2610                         id_priv->afonly = init_net.ipv6.sysctl.bindv6only;
2611 #endif
2612         }
2613         ret = cma_get_port(id_priv);
2614         if (ret)
2615                 goto err2;
2616
2617         return 0;
2618 err2:
2619         if (id_priv->cma_dev)
2620                 cma_release_dev(id_priv);
2621 err1:
2622         cma_comp_exch(id_priv, RDMA_CM_ADDR_BOUND, RDMA_CM_IDLE);
2623         return ret;
2624 }
2625 EXPORT_SYMBOL(rdma_bind_addr);
2626
2627 static int cma_format_hdr(void *hdr, struct rdma_id_private *id_priv)
2628 {
2629         struct cma_hdr *cma_hdr;
2630
2631         cma_hdr = hdr;
2632         cma_hdr->cma_version = CMA_VERSION;
2633         if (cma_family(id_priv) == AF_INET) {
2634                 struct sockaddr_in *src4, *dst4;
2635
2636                 src4 = (struct sockaddr_in *) cma_src_addr(id_priv);
2637                 dst4 = (struct sockaddr_in *) cma_dst_addr(id_priv);
2638
2639                 cma_set_ip_ver(cma_hdr, 4);
2640                 cma_hdr->src_addr.ip4.addr = src4->sin_addr.s_addr;
2641                 cma_hdr->dst_addr.ip4.addr = dst4->sin_addr.s_addr;
2642                 cma_hdr->port = src4->sin_port;
2643         } else if (cma_family(id_priv) == AF_INET6) {
2644                 struct sockaddr_in6 *src6, *dst6;
2645
2646                 src6 = (struct sockaddr_in6 *) cma_src_addr(id_priv);
2647                 dst6 = (struct sockaddr_in6 *) cma_dst_addr(id_priv);
2648
2649                 cma_set_ip_ver(cma_hdr, 6);
2650                 cma_hdr->src_addr.ip6 = src6->sin6_addr;
2651                 cma_hdr->dst_addr.ip6 = dst6->sin6_addr;
2652                 cma_hdr->port = src6->sin6_port;
2653         }
2654         return 0;
2655 }
2656
2657 static int cma_sidr_rep_handler(struct ib_cm_id *cm_id,
2658                                 struct ib_cm_event *ib_event)
2659 {
2660         struct rdma_id_private *id_priv = cm_id->context;
2661         struct rdma_cm_event event;
2662         struct ib_cm_sidr_rep_event_param *rep = &ib_event->param.sidr_rep_rcvd;
2663         int ret = 0;
2664
2665         if (cma_disable_callback(id_priv, RDMA_CM_CONNECT))
2666                 return 0;
2667
2668         memset(&event, 0, sizeof event);
2669         switch (ib_event->event) {
2670         case IB_CM_SIDR_REQ_ERROR:
2671                 event.event = RDMA_CM_EVENT_UNREACHABLE;
2672                 event.status = -ETIMEDOUT;
2673                 break;
2674         case IB_CM_SIDR_REP_RECEIVED:
2675                 event.param.ud.private_data = ib_event->private_data;
2676                 event.param.ud.private_data_len = IB_CM_SIDR_REP_PRIVATE_DATA_SIZE;
2677                 if (rep->status != IB_SIDR_SUCCESS) {
2678                         event.event = RDMA_CM_EVENT_UNREACHABLE;
2679                         event.status = ib_event->param.sidr_rep_rcvd.status;
2680                         break;
2681                 }
2682                 ret = cma_set_qkey(id_priv, rep->qkey);
2683                 if (ret) {
2684                         event.event = RDMA_CM_EVENT_ADDR_ERROR;
2685                         event.status = ret;
2686                         break;
2687                 }
2688                 ib_init_ah_from_path(id_priv->id.device, id_priv->id.port_num,
2689                                      id_priv->id.route.path_rec,
2690                                      &event.param.ud.ah_attr);
2691                 event.param.ud.qp_num = rep->qpn;
2692                 event.param.ud.qkey = rep->qkey;
2693                 event.event = RDMA_CM_EVENT_ESTABLISHED;
2694                 event.status = 0;
2695                 break;
2696         default:
2697                 printk(KERN_ERR "RDMA CMA: unexpected IB CM event: %d\n",
2698                        ib_event->event);
2699                 goto out;
2700         }
2701
2702         ret = id_priv->id.event_handler(&id_priv->id, &event);
2703         if (ret) {
2704                 /* Destroy the CM ID by returning a non-zero value. */
2705                 id_priv->cm_id.ib = NULL;
2706                 cma_exch(id_priv, RDMA_CM_DESTROYING);
2707                 mutex_unlock(&id_priv->handler_mutex);
2708                 rdma_destroy_id(&id_priv->id);
2709                 return ret;
2710         }
2711 out:
2712         mutex_unlock(&id_priv->handler_mutex);
2713         return ret;
2714 }
2715
2716 static int cma_resolve_ib_udp(struct rdma_id_private *id_priv,
2717                               struct rdma_conn_param *conn_param)
2718 {
2719         struct ib_cm_sidr_req_param req;
2720         struct ib_cm_id *id;
2721         void *private_data;
2722         int offset, ret;
2723
2724         memset(&req, 0, sizeof req);
2725         offset = cma_user_data_offset(id_priv);
2726         req.private_data_len = offset + conn_param->private_data_len;
2727         if (req.private_data_len < conn_param->private_data_len)
2728                 return -EINVAL;
2729
2730         if (req.private_data_len) {
2731                 private_data = kzalloc(req.private_data_len, GFP_ATOMIC);
2732                 if (!private_data)
2733                         return -ENOMEM;
2734         } else {
2735                 private_data = NULL;
2736         }
2737
2738         if (conn_param->private_data && conn_param->private_data_len)
2739                 memcpy(private_data + offset, conn_param->private_data,
2740                        conn_param->private_data_len);
2741
2742         if (private_data) {
2743                 ret = cma_format_hdr(private_data, id_priv);
2744                 if (ret)
2745                         goto out;
2746                 req.private_data = private_data;
2747         }
2748
2749         id = ib_create_cm_id(id_priv->id.device, cma_sidr_rep_handler,
2750                              id_priv);
2751         if (IS_ERR(id)) {
2752                 ret = PTR_ERR(id);
2753                 goto out;
2754         }
2755         id_priv->cm_id.ib = id;
2756
2757         req.path = id_priv->id.route.path_rec;
2758         req.service_id = rdma_get_service_id(&id_priv->id, cma_dst_addr(id_priv));
2759         req.timeout_ms = 1 << (CMA_CM_RESPONSE_TIMEOUT - 8);
2760         req.max_cm_retries = CMA_MAX_CM_RETRIES;
2761
2762         ret = ib_send_cm_sidr_req(id_priv->cm_id.ib, &req);
2763         if (ret) {
2764                 ib_destroy_cm_id(id_priv->cm_id.ib);
2765                 id_priv->cm_id.ib = NULL;
2766         }
2767 out:
2768         kfree(private_data);
2769         return ret;
2770 }
2771
2772 static int cma_connect_ib(struct rdma_id_private *id_priv,
2773                           struct rdma_conn_param *conn_param)
2774 {
2775         struct ib_cm_req_param req;
2776         struct rdma_route *route;
2777         void *private_data;
2778         struct ib_cm_id *id;
2779         int offset, ret;
2780
2781         memset(&req, 0, sizeof req);
2782         offset = cma_user_data_offset(id_priv);
2783         req.private_data_len = offset + conn_param->private_data_len;
2784         if (req.private_data_len < conn_param->private_data_len)
2785                 return -EINVAL;
2786
2787         if (req.private_data_len) {
2788                 private_data = kzalloc(req.private_data_len, GFP_ATOMIC);
2789                 if (!private_data)
2790                         return -ENOMEM;
2791         } else {
2792                 private_data = NULL;
2793         }
2794
2795         if (conn_param->private_data && conn_param->private_data_len)
2796                 memcpy(private_data + offset, conn_param->private_data,
2797                        conn_param->private_data_len);
2798
2799         id = ib_create_cm_id(id_priv->id.device, cma_ib_handler, id_priv);
2800         if (IS_ERR(id)) {
2801                 ret = PTR_ERR(id);
2802                 goto out;
2803         }
2804         id_priv->cm_id.ib = id;
2805
2806         route = &id_priv->id.route;
2807         if (private_data) {
2808                 ret = cma_format_hdr(private_data, id_priv);
2809                 if (ret)
2810                         goto out;
2811                 req.private_data = private_data;
2812         }
2813
2814         req.primary_path = &route->path_rec[0];
2815         if (route->num_paths == 2)
2816                 req.alternate_path = &route->path_rec[1];
2817
2818         req.service_id = rdma_get_service_id(&id_priv->id, cma_dst_addr(id_priv));
2819         req.qp_num = id_priv->qp_num;
2820         req.qp_type = id_priv->id.qp_type;
2821         req.starting_psn = id_priv->seq_num;
2822         req.responder_resources = conn_param->responder_resources;
2823         req.initiator_depth = conn_param->initiator_depth;
2824         req.flow_control = conn_param->flow_control;
2825         req.retry_count = min_t(u8, 7, conn_param->retry_count);
2826         req.rnr_retry_count = min_t(u8, 7, conn_param->rnr_retry_count);
2827         req.remote_cm_response_timeout = CMA_CM_RESPONSE_TIMEOUT;
2828         req.local_cm_response_timeout = CMA_CM_RESPONSE_TIMEOUT;
2829         req.max_cm_retries = CMA_MAX_CM_RETRIES;
2830         req.srq = id_priv->srq ? 1 : 0;
2831
2832         ret = ib_send_cm_req(id_priv->cm_id.ib, &req);
2833 out:
2834         if (ret && !IS_ERR(id)) {
2835                 ib_destroy_cm_id(id);
2836                 id_priv->cm_id.ib = NULL;
2837         }
2838
2839         kfree(private_data);
2840         return ret;
2841 }
2842
2843 static int cma_connect_iw(struct rdma_id_private *id_priv,
2844                           struct rdma_conn_param *conn_param)
2845 {
2846         struct iw_cm_id *cm_id;
2847         int ret;
2848         struct iw_cm_conn_param iw_param;
2849
2850         cm_id = iw_create_cm_id(id_priv->id.device, cma_iw_handler, id_priv);
2851         if (IS_ERR(cm_id))
2852                 return PTR_ERR(cm_id);
2853
2854         id_priv->cm_id.iw = cm_id;
2855
2856         memcpy(&cm_id->local_addr, cma_src_addr(id_priv),
2857                rdma_addr_size(cma_src_addr(id_priv)));
2858         memcpy(&cm_id->remote_addr, cma_dst_addr(id_priv),
2859                rdma_addr_size(cma_dst_addr(id_priv)));
2860
2861         ret = cma_modify_qp_rtr(id_priv, conn_param);
2862         if (ret)
2863                 goto out;
2864
2865         if (conn_param) {
2866                 iw_param.ord = conn_param->initiator_depth;
2867                 iw_param.ird = conn_param->responder_resources;
2868                 iw_param.private_data = conn_param->private_data;
2869                 iw_param.private_data_len = conn_param->private_data_len;
2870                 iw_param.qpn = id_priv->id.qp ? id_priv->qp_num : conn_param->qp_num;
2871         } else {
2872                 memset(&iw_param, 0, sizeof iw_param);
2873                 iw_param.qpn = id_priv->qp_num;
2874         }
2875         ret = iw_cm_connect(cm_id, &iw_param);
2876 out:
2877         if (ret) {
2878                 iw_destroy_cm_id(cm_id);
2879                 id_priv->cm_id.iw = NULL;
2880         }
2881         return ret;
2882 }
2883
2884 int rdma_connect(struct rdma_cm_id *id, struct rdma_conn_param *conn_param)
2885 {
2886         struct rdma_id_private *id_priv;
2887         int ret;
2888
2889         id_priv = container_of(id, struct rdma_id_private, id);
2890         if (!cma_comp_exch(id_priv, RDMA_CM_ROUTE_RESOLVED, RDMA_CM_CONNECT))
2891                 return -EINVAL;
2892
2893         if (!id->qp) {
2894                 id_priv->qp_num = conn_param->qp_num;
2895                 id_priv->srq = conn_param->srq;
2896         }
2897
2898         if (rdma_cap_ib_cm(id->device, id->port_num)) {
2899                 if (id->qp_type == IB_QPT_UD)
2900                         ret = cma_resolve_ib_udp(id_priv, conn_param);
2901                 else
2902                         ret = cma_connect_ib(id_priv, conn_param);
2903         } else if (rdma_cap_iw_cm(id->device, id->port_num))
2904                 ret = cma_connect_iw(id_priv, conn_param);
2905         else
2906                 ret = -ENOSYS;
2907         if (ret)
2908                 goto err;
2909
2910         return 0;
2911 err:
2912         cma_comp_exch(id_priv, RDMA_CM_CONNECT, RDMA_CM_ROUTE_RESOLVED);
2913         return ret;
2914 }
2915 EXPORT_SYMBOL(rdma_connect);
2916
2917 static int cma_accept_ib(struct rdma_id_private *id_priv,
2918                          struct rdma_conn_param *conn_param)
2919 {
2920         struct ib_cm_rep_param rep;
2921         int ret;
2922
2923         ret = cma_modify_qp_rtr(id_priv, conn_param);
2924         if (ret)
2925                 goto out;
2926
2927         ret = cma_modify_qp_rts(id_priv, conn_param);
2928         if (ret)
2929                 goto out;
2930
2931         memset(&rep, 0, sizeof rep);
2932         rep.qp_num = id_priv->qp_num;
2933         rep.starting_psn = id_priv->seq_num;
2934         rep.private_data = conn_param->private_data;
2935         rep.private_data_len = conn_param->private_data_len;
2936         rep.responder_resources = conn_param->responder_resources;
2937         rep.initiator_depth = conn_param->initiator_depth;
2938         rep.failover_accepted = 0;
2939         rep.flow_control = conn_param->flow_control;
2940         rep.rnr_retry_count = min_t(u8, 7, conn_param->rnr_retry_count);
2941         rep.srq = id_priv->srq ? 1 : 0;
2942
2943         ret = ib_send_cm_rep(id_priv->cm_id.ib, &rep);
2944 out:
2945         return ret;
2946 }
2947
2948 static int cma_accept_iw(struct rdma_id_private *id_priv,
2949                   struct rdma_conn_param *conn_param)
2950 {
2951         struct iw_cm_conn_param iw_param;
2952         int ret;
2953
2954         ret = cma_modify_qp_rtr(id_priv, conn_param);
2955         if (ret)
2956                 return ret;
2957
2958         iw_param.ord = conn_param->initiator_depth;
2959         iw_param.ird = conn_param->responder_resources;
2960         iw_param.private_data = conn_param->private_data;
2961         iw_param.private_data_len = conn_param->private_data_len;
2962         if (id_priv->id.qp) {
2963                 iw_param.qpn = id_priv->qp_num;
2964         } else
2965                 iw_param.qpn = conn_param->qp_num;
2966
2967         return iw_cm_accept(id_priv->cm_id.iw, &iw_param);
2968 }
2969
2970 static int cma_send_sidr_rep(struct rdma_id_private *id_priv,
2971                              enum ib_cm_sidr_status status, u32 qkey,
2972                              const void *private_data, int private_data_len)
2973 {
2974         struct ib_cm_sidr_rep_param rep;
2975         int ret;
2976
2977         memset(&rep, 0, sizeof rep);
2978         rep.status = status;
2979         if (status == IB_SIDR_SUCCESS) {
2980                 ret = cma_set_qkey(id_priv, qkey);
2981                 if (ret)
2982                         return ret;
2983                 rep.qp_num = id_priv->qp_num;
2984                 rep.qkey = id_priv->qkey;
2985         }
2986         rep.private_data = private_data;
2987         rep.private_data_len = private_data_len;
2988
2989         return ib_send_cm_sidr_rep(id_priv->cm_id.ib, &rep);
2990 }
2991
2992 int rdma_accept(struct rdma_cm_id *id, struct rdma_conn_param *conn_param)
2993 {
2994         struct rdma_id_private *id_priv;
2995         int ret;
2996
2997         id_priv = container_of(id, struct rdma_id_private, id);
2998
2999         id_priv->owner = task_pid_nr(current);
3000
3001         if (!cma_comp(id_priv, RDMA_CM_CONNECT))
3002                 return -EINVAL;
3003
3004         if (!id->qp && conn_param) {
3005                 id_priv->qp_num = conn_param->qp_num;
3006                 id_priv->srq = conn_param->srq;
3007         }
3008
3009         if (rdma_cap_ib_cm(id->device, id->port_num)) {
3010                 if (id->qp_type == IB_QPT_UD) {
3011                         if (conn_param)
3012                                 ret = cma_send_sidr_rep(id_priv, IB_SIDR_SUCCESS,
3013                                                         conn_param->qkey,
3014                                                         conn_param->private_data,
3015                                                         conn_param->private_data_len);
3016                         else
3017                                 ret = cma_send_sidr_rep(id_priv, IB_SIDR_SUCCESS,
3018                                                         0, NULL, 0);
3019                 } else {
3020                         if (conn_param)
3021                                 ret = cma_accept_ib(id_priv, conn_param);
3022                         else
3023                                 ret = cma_rep_recv(id_priv);
3024                 }
3025         } else if (rdma_cap_iw_cm(id->device, id->port_num))
3026                 ret = cma_accept_iw(id_priv, conn_param);
3027         else
3028                 ret = -ENOSYS;
3029
3030         if (ret)
3031                 goto reject;
3032
3033         return 0;
3034 reject:
3035         cma_modify_qp_err(id_priv);
3036         rdma_reject(id, NULL, 0);
3037         return ret;
3038 }
3039 EXPORT_SYMBOL(rdma_accept);
3040
3041 int rdma_notify(struct rdma_cm_id *id, enum ib_event_type event)
3042 {
3043         struct rdma_id_private *id_priv;
3044         int ret;
3045
3046         id_priv = container_of(id, struct rdma_id_private, id);
3047         if (!id_priv->cm_id.ib)
3048                 return -EINVAL;
3049
3050         switch (id->device->node_type) {
3051         case RDMA_NODE_IB_CA:
3052                 ret = ib_cm_notify(id_priv->cm_id.ib, event);
3053                 break;
3054         default:
3055                 ret = 0;
3056                 break;
3057         }
3058         return ret;
3059 }
3060 EXPORT_SYMBOL(rdma_notify);
3061
3062 int rdma_reject(struct rdma_cm_id *id, const void *private_data,
3063                 u8 private_data_len)
3064 {
3065         struct rdma_id_private *id_priv;
3066         int ret;
3067
3068         id_priv = container_of(id, struct rdma_id_private, id);
3069         if (!id_priv->cm_id.ib)
3070                 return -EINVAL;
3071
3072         if (rdma_cap_ib_cm(id->device, id->port_num)) {
3073                 if (id->qp_type == IB_QPT_UD)
3074                         ret = cma_send_sidr_rep(id_priv, IB_SIDR_REJECT, 0,
3075                                                 private_data, private_data_len);
3076                 else
3077                         ret = ib_send_cm_rej(id_priv->cm_id.ib,
3078                                              IB_CM_REJ_CONSUMER_DEFINED, NULL,
3079                                              0, private_data, private_data_len);
3080         } else if (rdma_cap_iw_cm(id->device, id->port_num)) {
3081                 ret = iw_cm_reject(id_priv->cm_id.iw,
3082                                    private_data, private_data_len);
3083         } else
3084                 ret = -ENOSYS;
3085
3086         return ret;
3087 }
3088 EXPORT_SYMBOL(rdma_reject);
3089
3090 int rdma_disconnect(struct rdma_cm_id *id)
3091 {
3092         struct rdma_id_private *id_priv;
3093         int ret;
3094
3095         id_priv = container_of(id, struct rdma_id_private, id);
3096         if (!id_priv->cm_id.ib)
3097                 return -EINVAL;
3098
3099         if (rdma_cap_ib_cm(id->device, id->port_num)) {
3100                 ret = cma_modify_qp_err(id_priv);
3101                 if (ret)
3102                         goto out;
3103                 /* Initiate or respond to a disconnect. */
3104                 if (ib_send_cm_dreq(id_priv->cm_id.ib, NULL, 0))
3105                         ib_send_cm_drep(id_priv->cm_id.ib, NULL, 0);
3106         } else if (rdma_cap_iw_cm(id->device, id->port_num)) {
3107                 ret = iw_cm_disconnect(id_priv->cm_id.iw, 0);
3108         } else
3109                 ret = -EINVAL;
3110
3111 out:
3112         return ret;
3113 }
3114 EXPORT_SYMBOL(rdma_disconnect);
3115
3116 static int cma_ib_mc_handler(int status, struct ib_sa_multicast *multicast)
3117 {
3118         struct rdma_id_private *id_priv;
3119         struct cma_multicast *mc = multicast->context;
3120         struct rdma_cm_event event;
3121         int ret;
3122
3123         id_priv = mc->id_priv;
3124         if (cma_disable_callback(id_priv, RDMA_CM_ADDR_BOUND) &&
3125             cma_disable_callback(id_priv, RDMA_CM_ADDR_RESOLVED))
3126                 return 0;
3127
3128         if (!status)
3129                 status = cma_set_qkey(id_priv, be32_to_cpu(multicast->rec.qkey));
3130         mutex_lock(&id_priv->qp_mutex);
3131         if (!status && id_priv->id.qp)
3132                 status = ib_attach_mcast(id_priv->id.qp, &multicast->rec.mgid,
3133                                          be16_to_cpu(multicast->rec.mlid));
3134         mutex_unlock(&id_priv->qp_mutex);
3135
3136         memset(&event, 0, sizeof event);
3137         event.status = status;
3138         event.param.ud.private_data = mc->context;
3139         if (!status) {
3140                 event.event = RDMA_CM_EVENT_MULTICAST_JOIN;
3141                 ib_init_ah_from_mcmember(id_priv->id.device,
3142                                          id_priv->id.port_num, &multicast->rec,
3143                                          &event.param.ud.ah_attr);
3144                 event.param.ud.qp_num = 0xFFFFFF;
3145                 event.param.ud.qkey = be32_to_cpu(multicast->rec.qkey);
3146         } else
3147                 event.event = RDMA_CM_EVENT_MULTICAST_ERROR;
3148
3149         ret = id_priv->id.event_handler(&id_priv->id, &event);
3150         if (ret) {
3151                 cma_exch(id_priv, RDMA_CM_DESTROYING);
3152                 mutex_unlock(&id_priv->handler_mutex);
3153                 rdma_destroy_id(&id_priv->id);
3154                 return 0;
3155         }
3156
3157         mutex_unlock(&id_priv->handler_mutex);
3158         return 0;
3159 }
3160
3161 static void cma_set_mgid(struct rdma_id_private *id_priv,
3162                          struct sockaddr *addr, union ib_gid *mgid)
3163 {
3164         unsigned char mc_map[MAX_ADDR_LEN];
3165         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
3166         struct sockaddr_in *sin = (struct sockaddr_in *) addr;
3167         struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *) addr;
3168
3169         if (cma_any_addr(addr)) {
3170                 memset(mgid, 0, sizeof *mgid);
3171         } else if ((addr->sa_family == AF_INET6) &&
3172                    ((be32_to_cpu(sin6->sin6_addr.s6_addr32[0]) & 0xFFF0FFFF) ==
3173                                                                  0xFF10A01B)) {
3174                 /* IPv6 address is an SA assigned MGID. */
3175                 memcpy(mgid, &sin6->sin6_addr, sizeof *mgid);
3176         } else if (addr->sa_family == AF_IB) {
3177                 memcpy(mgid, &((struct sockaddr_ib *) addr)->sib_addr, sizeof *mgid);
3178         } else if ((addr->sa_family == AF_INET6)) {
3179                 ipv6_ib_mc_map(&sin6->sin6_addr, dev_addr->broadcast, mc_map);
3180                 if (id_priv->id.ps == RDMA_PS_UDP)
3181                         mc_map[7] = 0x01;       /* Use RDMA CM signature */
3182                 *mgid = *(union ib_gid *) (mc_map + 4);
3183         } else {
3184                 ip_ib_mc_map(sin->sin_addr.s_addr, dev_addr->broadcast, mc_map);
3185                 if (id_priv->id.ps == RDMA_PS_UDP)
3186                         mc_map[7] = 0x01;       /* Use RDMA CM signature */
3187                 *mgid = *(union ib_gid *) (mc_map + 4);
3188         }
3189 }
3190
3191 static int cma_join_ib_multicast(struct rdma_id_private *id_priv,
3192                                  struct cma_multicast *mc)
3193 {
3194         struct ib_sa_mcmember_rec rec;
3195         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
3196         ib_sa_comp_mask comp_mask;
3197         int ret;
3198
3199         ib_addr_get_mgid(dev_addr, &rec.mgid);
3200         ret = ib_sa_get_mcmember_rec(id_priv->id.device, id_priv->id.port_num,
3201                                      &rec.mgid, &rec);
3202         if (ret)
3203                 return ret;
3204
3205         ret = cma_set_qkey(id_priv, 0);
3206         if (ret)
3207                 return ret;
3208
3209         cma_set_mgid(id_priv, (struct sockaddr *) &mc->addr, &rec.mgid);
3210         rec.qkey = cpu_to_be32(id_priv->qkey);
3211         rdma_addr_get_sgid(dev_addr, &rec.port_gid);
3212         rec.pkey = cpu_to_be16(ib_addr_get_pkey(dev_addr));
3213         rec.join_state = 1;
3214
3215         comp_mask = IB_SA_MCMEMBER_REC_MGID | IB_SA_MCMEMBER_REC_PORT_GID |
3216                     IB_SA_MCMEMBER_REC_PKEY | IB_SA_MCMEMBER_REC_JOIN_STATE |
3217                     IB_SA_MCMEMBER_REC_QKEY | IB_SA_MCMEMBER_REC_SL |
3218                     IB_SA_MCMEMBER_REC_FLOW_LABEL |
3219                     IB_SA_MCMEMBER_REC_TRAFFIC_CLASS;
3220
3221         if (id_priv->id.ps == RDMA_PS_IPOIB)
3222                 comp_mask |= IB_SA_MCMEMBER_REC_RATE |
3223                              IB_SA_MCMEMBER_REC_RATE_SELECTOR |
3224                              IB_SA_MCMEMBER_REC_MTU_SELECTOR |
3225                              IB_SA_MCMEMBER_REC_MTU |
3226                              IB_SA_MCMEMBER_REC_HOP_LIMIT;
3227
3228         mc->multicast.ib = ib_sa_join_multicast(&sa_client, id_priv->id.device,
3229                                                 id_priv->id.port_num, &rec,
3230                                                 comp_mask, GFP_KERNEL,
3231                                                 cma_ib_mc_handler, mc);
3232         return PTR_ERR_OR_ZERO(mc->multicast.ib);
3233 }
3234
3235 static void iboe_mcast_work_handler(struct work_struct *work)
3236 {
3237         struct iboe_mcast_work *mw = container_of(work, struct iboe_mcast_work, work);
3238         struct cma_multicast *mc = mw->mc;
3239         struct ib_sa_multicast *m = mc->multicast.ib;
3240
3241         mc->multicast.ib->context = mc;
3242         cma_ib_mc_handler(0, m);
3243         kref_put(&mc->mcref, release_mc);
3244         kfree(mw);
3245 }
3246
3247 static void cma_iboe_set_mgid(struct sockaddr *addr, union ib_gid *mgid)
3248 {
3249         struct sockaddr_in *sin = (struct sockaddr_in *)addr;
3250         struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)addr;
3251
3252         if (cma_any_addr(addr)) {
3253                 memset(mgid, 0, sizeof *mgid);
3254         } else if (addr->sa_family == AF_INET6) {
3255                 memcpy(mgid, &sin6->sin6_addr, sizeof *mgid);
3256         } else {
3257                 mgid->raw[0] = 0xff;
3258                 mgid->raw[1] = 0x0e;
3259                 mgid->raw[2] = 0;
3260                 mgid->raw[3] = 0;
3261                 mgid->raw[4] = 0;
3262                 mgid->raw[5] = 0;
3263                 mgid->raw[6] = 0;
3264                 mgid->raw[7] = 0;
3265                 mgid->raw[8] = 0;
3266                 mgid->raw[9] = 0;
3267                 mgid->raw[10] = 0xff;
3268                 mgid->raw[11] = 0xff;
3269                 *(__be32 *)(&mgid->raw[12]) = sin->sin_addr.s_addr;
3270         }
3271 }
3272
3273 static int cma_iboe_join_multicast(struct rdma_id_private *id_priv,
3274                                    struct cma_multicast *mc)
3275 {
3276         struct iboe_mcast_work *work;
3277         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
3278         int err;
3279         struct sockaddr *addr = (struct sockaddr *)&mc->addr;
3280         struct net_device *ndev = NULL;
3281
3282         if (cma_zero_addr((struct sockaddr *)&mc->addr))
3283                 return -EINVAL;
3284
3285         work = kzalloc(sizeof *work, GFP_KERNEL);
3286         if (!work)
3287                 return -ENOMEM;
3288
3289         mc->multicast.ib = kzalloc(sizeof(struct ib_sa_multicast), GFP_KERNEL);
3290         if (!mc->multicast.ib) {
3291                 err = -ENOMEM;
3292                 goto out1;
3293         }
3294
3295         cma_iboe_set_mgid(addr, &mc->multicast.ib->rec.mgid);
3296
3297         mc->multicast.ib->rec.pkey = cpu_to_be16(0xffff);
3298         if (id_priv->id.ps == RDMA_PS_UDP)
3299                 mc->multicast.ib->rec.qkey = cpu_to_be32(RDMA_UDP_QKEY);
3300
3301         if (dev_addr->bound_dev_if)
3302                 ndev = dev_get_by_index(&init_net, dev_addr->bound_dev_if);
3303         if (!ndev) {
3304                 err = -ENODEV;
3305                 goto out2;
3306         }
3307         mc->multicast.ib->rec.rate = iboe_get_rate(ndev);
3308         mc->multicast.ib->rec.hop_limit = 1;
3309         mc->multicast.ib->rec.mtu = iboe_get_mtu(ndev->mtu);
3310         dev_put(ndev);
3311         if (!mc->multicast.ib->rec.mtu) {
3312                 err = -EINVAL;
3313                 goto out2;
3314         }
3315         rdma_ip2gid((struct sockaddr *)&id_priv->id.route.addr.src_addr,
3316                     &mc->multicast.ib->rec.port_gid);
3317         work->id = id_priv;
3318         work->mc = mc;
3319         INIT_WORK(&work->work, iboe_mcast_work_handler);
3320         kref_get(&mc->mcref);
3321         queue_work(cma_wq, &work->work);
3322
3323         return 0;
3324
3325 out2:
3326         kfree(mc->multicast.ib);
3327 out1:
3328         kfree(work);
3329         return err;
3330 }
3331
3332 int rdma_join_multicast(struct rdma_cm_id *id, struct sockaddr *addr,
3333                         void *context)
3334 {
3335         struct rdma_id_private *id_priv;
3336         struct cma_multicast *mc;
3337         int ret;
3338
3339         id_priv = container_of(id, struct rdma_id_private, id);
3340         if (!cma_comp(id_priv, RDMA_CM_ADDR_BOUND) &&
3341             !cma_comp(id_priv, RDMA_CM_ADDR_RESOLVED))
3342                 return -EINVAL;
3343
3344         mc = kmalloc(sizeof *mc, GFP_KERNEL);
3345         if (!mc)
3346                 return -ENOMEM;
3347
3348         memcpy(&mc->addr, addr, rdma_addr_size(addr));
3349         mc->context = context;
3350         mc->id_priv = id_priv;
3351
3352         spin_lock(&id_priv->lock);
3353         list_add(&mc->list, &id_priv->mc_list);
3354         spin_unlock(&id_priv->lock);
3355
3356         if (rdma_protocol_roce(id->device, id->port_num)) {
3357                 kref_init(&mc->mcref);
3358                 ret = cma_iboe_join_multicast(id_priv, mc);
3359         } else if (rdma_cap_ib_mcast(id->device, id->port_num))
3360                 ret = cma_join_ib_multicast(id_priv, mc);
3361         else
3362                 ret = -ENOSYS;
3363
3364         if (ret) {
3365                 spin_lock_irq(&id_priv->lock);
3366                 list_del(&mc->list);
3367                 spin_unlock_irq(&id_priv->lock);
3368                 kfree(mc);
3369         }
3370         return ret;
3371 }
3372 EXPORT_SYMBOL(rdma_join_multicast);
3373
3374 void rdma_leave_multicast(struct rdma_cm_id *id, struct sockaddr *addr)
3375 {
3376         struct rdma_id_private *id_priv;
3377         struct cma_multicast *mc;
3378
3379         id_priv = container_of(id, struct rdma_id_private, id);
3380         spin_lock_irq(&id_priv->lock);
3381         list_for_each_entry(mc, &id_priv->mc_list, list) {
3382                 if (!memcmp(&mc->addr, addr, rdma_addr_size(addr))) {
3383                         list_del(&mc->list);
3384                         spin_unlock_irq(&id_priv->lock);
3385
3386                         if (id->qp)
3387                                 ib_detach_mcast(id->qp,
3388                                                 &mc->multicast.ib->rec.mgid,
3389                                                 be16_to_cpu(mc->multicast.ib->rec.mlid));
3390
3391                         BUG_ON(id_priv->cma_dev->device != id->device);
3392
3393                         if (rdma_cap_ib_mcast(id->device, id->port_num)) {
3394                                 ib_sa_free_multicast(mc->multicast.ib);
3395                                 kfree(mc);
3396                         } else if (rdma_protocol_roce(id->device, id->port_num))
3397                                 kref_put(&mc->mcref, release_mc);
3398
3399                         return;
3400                 }
3401         }
3402         spin_unlock_irq(&id_priv->lock);
3403 }
3404 EXPORT_SYMBOL(rdma_leave_multicast);
3405
3406 static int cma_netdev_change(struct net_device *ndev, struct rdma_id_private *id_priv)
3407 {
3408         struct rdma_dev_addr *dev_addr;
3409         struct cma_ndev_work *work;
3410
3411         dev_addr = &id_priv->id.route.addr.dev_addr;
3412
3413         if ((dev_addr->bound_dev_if == ndev->ifindex) &&
3414             memcmp(dev_addr->src_dev_addr, ndev->dev_addr, ndev->addr_len)) {
3415                 printk(KERN_INFO "RDMA CM addr change for ndev %s used by id %p\n",
3416                        ndev->name, &id_priv->id);
3417                 work = kzalloc(sizeof *work, GFP_KERNEL);
3418                 if (!work)
3419                         return -ENOMEM;
3420
3421                 INIT_WORK(&work->work, cma_ndev_work_handler);
3422                 work->id = id_priv;
3423                 work->event.event = RDMA_CM_EVENT_ADDR_CHANGE;
3424                 atomic_inc(&id_priv->refcount);
3425                 queue_work(cma_wq, &work->work);
3426         }
3427
3428         return 0;
3429 }
3430
3431 static int cma_netdev_callback(struct notifier_block *self, unsigned long event,
3432                                void *ptr)
3433 {
3434         struct net_device *ndev = netdev_notifier_info_to_dev(ptr);
3435         struct cma_device *cma_dev;
3436         struct rdma_id_private *id_priv;
3437         int ret = NOTIFY_DONE;
3438
3439         if (dev_net(ndev) != &init_net)
3440                 return NOTIFY_DONE;
3441
3442         if (event != NETDEV_BONDING_FAILOVER)
3443                 return NOTIFY_DONE;
3444
3445         if (!(ndev->flags & IFF_MASTER) || !(ndev->priv_flags & IFF_BONDING))
3446                 return NOTIFY_DONE;
3447
3448         mutex_lock(&lock);
3449         list_for_each_entry(cma_dev, &dev_list, list)
3450                 list_for_each_entry(id_priv, &cma_dev->id_list, list) {
3451                         ret = cma_netdev_change(ndev, id_priv);
3452                         if (ret)
3453                                 goto out;
3454                 }
3455
3456 out:
3457         mutex_unlock(&lock);
3458         return ret;
3459 }
3460
3461 static struct notifier_block cma_nb = {
3462         .notifier_call = cma_netdev_callback
3463 };
3464
3465 static void cma_add_one(struct ib_device *device)
3466 {
3467         struct cma_device *cma_dev;
3468         struct rdma_id_private *id_priv;
3469
3470         cma_dev = kmalloc(sizeof *cma_dev, GFP_KERNEL);
3471         if (!cma_dev)
3472                 return;
3473
3474         cma_dev->device = device;
3475
3476         init_completion(&cma_dev->comp);
3477         atomic_set(&cma_dev->refcount, 1);
3478         INIT_LIST_HEAD(&cma_dev->id_list);
3479         ib_set_client_data(device, &cma_client, cma_dev);
3480
3481         mutex_lock(&lock);
3482         list_add_tail(&cma_dev->list, &dev_list);
3483         list_for_each_entry(id_priv, &listen_any_list, list)
3484                 cma_listen_on_dev(id_priv, cma_dev);
3485         mutex_unlock(&lock);
3486 }
3487
3488 static int cma_remove_id_dev(struct rdma_id_private *id_priv)
3489 {
3490         struct rdma_cm_event event;
3491         enum rdma_cm_state state;
3492         int ret = 0;
3493
3494         /* Record that we want to remove the device */
3495         state = cma_exch(id_priv, RDMA_CM_DEVICE_REMOVAL);
3496         if (state == RDMA_CM_DESTROYING)
3497                 return 0;
3498
3499         cma_cancel_operation(id_priv, state);
3500         mutex_lock(&id_priv->handler_mutex);
3501
3502         /* Check for destruction from another callback. */
3503         if (!cma_comp(id_priv, RDMA_CM_DEVICE_REMOVAL))
3504                 goto out;
3505
3506         memset(&event, 0, sizeof event);
3507         event.event = RDMA_CM_EVENT_DEVICE_REMOVAL;
3508         ret = id_priv->id.event_handler(&id_priv->id, &event);
3509 out:
3510         mutex_unlock(&id_priv->handler_mutex);
3511         return ret;
3512 }
3513
3514 static void cma_process_remove(struct cma_device *cma_dev)
3515 {
3516         struct rdma_id_private *id_priv;
3517         int ret;
3518
3519         mutex_lock(&lock);
3520         while (!list_empty(&cma_dev->id_list)) {
3521                 id_priv = list_entry(cma_dev->id_list.next,
3522                                      struct rdma_id_private, list);
3523
3524                 list_del(&id_priv->listen_list);
3525                 list_del_init(&id_priv->list);
3526                 atomic_inc(&id_priv->refcount);
3527                 mutex_unlock(&lock);
3528
3529                 ret = id_priv->internal_id ? 1 : cma_remove_id_dev(id_priv);
3530                 cma_deref_id(id_priv);
3531                 if (ret)
3532                         rdma_destroy_id(&id_priv->id);
3533
3534                 mutex_lock(&lock);
3535         }
3536         mutex_unlock(&lock);
3537
3538         cma_deref_dev(cma_dev);
3539         wait_for_completion(&cma_dev->comp);
3540 }
3541
3542 static void cma_remove_one(struct ib_device *device)
3543 {
3544         struct cma_device *cma_dev;
3545
3546         cma_dev = ib_get_client_data(device, &cma_client);
3547         if (!cma_dev)
3548                 return;
3549
3550         mutex_lock(&lock);
3551         list_del(&cma_dev->list);
3552         mutex_unlock(&lock);
3553
3554         cma_process_remove(cma_dev);
3555         kfree(cma_dev);
3556 }
3557
3558 static int cma_get_id_stats(struct sk_buff *skb, struct netlink_callback *cb)
3559 {
3560         struct nlmsghdr *nlh;
3561         struct rdma_cm_id_stats *id_stats;
3562         struct rdma_id_private *id_priv;
3563         struct rdma_cm_id *id = NULL;
3564         struct cma_device *cma_dev;
3565         int i_dev = 0, i_id = 0;
3566
3567         /*
3568          * We export all of the IDs as a sequence of messages.  Each
3569          * ID gets its own netlink message.
3570          */
3571         mutex_lock(&lock);
3572
3573         list_for_each_entry(cma_dev, &dev_list, list) {
3574                 if (i_dev < cb->args[0]) {
3575                         i_dev++;
3576                         continue;
3577                 }
3578
3579                 i_id = 0;
3580                 list_for_each_entry(id_priv, &cma_dev->id_list, list) {
3581                         if (i_id < cb->args[1]) {
3582                                 i_id++;
3583                                 continue;
3584                         }
3585
3586                         id_stats = ibnl_put_msg(skb, &nlh, cb->nlh->nlmsg_seq,
3587                                                 sizeof *id_stats, RDMA_NL_RDMA_CM,
3588                                                 RDMA_NL_RDMA_CM_ID_STATS,
3589                                                 NLM_F_MULTI);
3590                         if (!id_stats)
3591                                 goto out;
3592
3593                         memset(id_stats, 0, sizeof *id_stats);
3594                         id = &id_priv->id;
3595                         id_stats->node_type = id->route.addr.dev_addr.dev_type;
3596                         id_stats->port_num = id->port_num;
3597                         id_stats->bound_dev_if =
3598                                 id->route.addr.dev_addr.bound_dev_if;
3599
3600                         if (ibnl_put_attr(skb, nlh,
3601                                           rdma_addr_size(cma_src_addr(id_priv)),
3602                                           cma_src_addr(id_priv),
3603                                           RDMA_NL_RDMA_CM_ATTR_SRC_ADDR))
3604                                 goto out;
3605                         if (ibnl_put_attr(skb, nlh,
3606                                           rdma_addr_size(cma_src_addr(id_priv)),
3607                                           cma_dst_addr(id_priv),
3608                                           RDMA_NL_RDMA_CM_ATTR_DST_ADDR))
3609                                 goto out;
3610
3611                         id_stats->pid           = id_priv->owner;
3612                         id_stats->port_space    = id->ps;
3613                         id_stats->cm_state      = id_priv->state;
3614                         id_stats->qp_num        = id_priv->qp_num;
3615                         id_stats->qp_type       = id->qp_type;
3616
3617                         i_id++;
3618                 }
3619
3620                 cb->args[1] = 0;
3621                 i_dev++;
3622         }
3623
3624 out:
3625         mutex_unlock(&lock);
3626         cb->args[0] = i_dev;
3627         cb->args[1] = i_id;
3628
3629         return skb->len;
3630 }
3631
3632 static const struct ibnl_client_cbs cma_cb_table[] = {
3633         [RDMA_NL_RDMA_CM_ID_STATS] = { .dump = cma_get_id_stats,
3634                                        .module = THIS_MODULE },
3635 };
3636
3637 static int __init cma_init(void)
3638 {
3639         int ret;
3640
3641         cma_wq = create_singlethread_workqueue("rdma_cm");
3642         if (!cma_wq)
3643                 return -ENOMEM;
3644
3645         ib_sa_register_client(&sa_client);
3646         rdma_addr_register_client(&addr_client);
3647         register_netdevice_notifier(&cma_nb);
3648
3649         ret = ib_register_client(&cma_client);
3650         if (ret)
3651                 goto err;
3652
3653         if (ibnl_add_client(RDMA_NL_RDMA_CM, RDMA_NL_RDMA_CM_NUM_OPS, cma_cb_table))
3654                 printk(KERN_WARNING "RDMA CMA: failed to add netlink callback\n");
3655
3656         return 0;
3657
3658 err:
3659         unregister_netdevice_notifier(&cma_nb);
3660         rdma_addr_unregister_client(&addr_client);
3661         ib_sa_unregister_client(&sa_client);
3662         destroy_workqueue(cma_wq);
3663         return ret;
3664 }
3665
3666 static void __exit cma_cleanup(void)
3667 {
3668         ibnl_remove_client(RDMA_NL_RDMA_CM);
3669         ib_unregister_client(&cma_client);
3670         unregister_netdevice_notifier(&cma_nb);
3671         rdma_addr_unregister_client(&addr_client);
3672         ib_sa_unregister_client(&sa_client);
3673         destroy_workqueue(cma_wq);
3674         idr_destroy(&tcp_ps);
3675         idr_destroy(&udp_ps);
3676         idr_destroy(&ipoib_ps);
3677         idr_destroy(&ib_ps);
3678 }
3679
3680 module_init(cma_init);
3681 module_exit(cma_cleanup);