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 void cma_add_one(struct ib_device *device);
  69 static void cma_remove_one(struct ib_device *device);
  70
  71 static struct ib_client cma_client = {
  72         .name   = "cma",
  73         .add    = cma_add_one,
  74         .remove = cma_remove_one
  75 };
  76
  77 static struct ib_sa_client sa_client;
  78 static struct rdma_addr_client addr_client;
  79 static LIST_HEAD(dev_list);
  80 static LIST_HEAD(listen_any_list);
  81 static DEFINE_MUTEX(lock);
  82 static struct workqueue_struct *cma_wq;
  83 static DEFINE_IDR(sdp_ps);
  84 static DEFINE_IDR(tcp_ps);
  85 static DEFINE_IDR(udp_ps);
  86 static DEFINE_IDR(ipoib_ps);
  87 static DEFINE_IDR(ib_ps);
  88
  89 struct cma_device {
  90         struct list_head        list;
  91         struct ib_device        *device;
  92         struct completion       comp;
  93         atomic_t                refcount;
  94         struct list_head        id_list;
  95 };
  96
  97 struct rdma_bind_list {
  98         struct idr              *ps;
  99         struct hlist_head       owners;
 100         unsigned short          port;
 101 };
 102
 103 enum {
 104         CMA_OPTION_AFONLY,
 105 };
 106
 107 /*
 108  * Device removal can occur at anytime, so we need extra handling to
 109  * serialize notifying the user of device removal with other callbacks.
 110  * We do this by disabling removal notification while a callback is in process,
 111  * and reporting it after the callback completes.
 112  */
 113 struct rdma_id_private {
 114         struct rdma_cm_id       id;
 115
 116         struct rdma_bind_list   *bind_list;
 117         struct hlist_node       node;
 118         struct list_head        list; /* listen_any_list or cma_device.list */
 119         struct list_head        listen_list; /* per device listens */
 120         struct cma_device       *cma_dev;
 121         struct list_head        mc_list;
 122
 123         int                     internal_id;
 124         enum rdma_cm_state      state;
 125         spinlock_t              lock;
 126         struct mutex            qp_mutex;
 127
 128         struct completion       comp;
 129         atomic_t                refcount;
 130         struct mutex            handler_mutex;
 131
 132         int                     backlog;
 133         int                     timeout_ms;
 134         struct ib_sa_query      *query;
 135         int                     query_id;
 136         union {
 137                 struct ib_cm_id *ib;
 138                 struct iw_cm_id *iw;
 139         } cm_id;
 140
 141         u32                     seq_num;
 142         u32                     qkey;
 143         u32                     qp_num;
 144         pid_t                   owner;
 145         u32                     options;
 146         u8                      srq;
 147         u8                      tos;
 148         u8                      reuseaddr;
 149         u8                      afonly;
 150 };
 151
 152 struct cma_multicast {
 153         struct rdma_id_private *id_priv;
 154         union {
 155                 struct ib_sa_multicast *ib;
 156         } multicast;
 157         struct list_head        list;
 158         void                    *context;
 159         struct sockaddr_storage addr;
 160         struct kref             mcref;
 161 };
 162
 163 struct cma_work {
 164         struct work_struct      work;
 165         struct rdma_id_private  *id;
 166         enum rdma_cm_state      old_state;
 167         enum rdma_cm_state      new_state;
 168         struct rdma_cm_event    event;
 169 };
 170
 171 struct cma_ndev_work {
 172         struct work_struct      work;
 173         struct rdma_id_private  *id;
 174         struct rdma_cm_event    event;
 175 };
 176
 177 struct iboe_mcast_work {
 178         struct work_struct       work;
 179         struct rdma_id_private  *id;
 180         struct cma_multicast    *mc;
 181 };
 182
 183 union cma_ip_addr {
 184         struct in6_addr ip6;
 185         struct {
 186                 __be32 pad[3];
 187                 __be32 addr;
 188         } ip4;
 189 };
 190
 191 struct cma_hdr {
 192         u8 cma_version;
 193         u8 ip_version;  /* IP version: 7:4 */
 194         __be16 port;
 195         union cma_ip_addr src_addr;
 196         union cma_ip_addr dst_addr;
 197 };
 198
 199 struct sdp_hh {
 200         u8 bsdh[16];
 201         u8 sdp_version; /* Major version: 7:4 */
 202         u8 ip_version;  /* IP version: 7:4 */
 203         u8 sdp_specific1[10];
 204         __be16 port;
 205         __be16 sdp_specific2;
 206         union cma_ip_addr src_addr;
 207         union cma_ip_addr dst_addr;
 208 };
 209
 210 struct sdp_hah {
 211         u8 bsdh[16];
 212         u8 sdp_version;
 213 };
 214
 215 #define CMA_VERSION 0x00
 216 #define SDP_MAJ_VERSION 0x2
 217
 218 static int cma_comp(struct rdma_id_private *id_priv, enum rdma_cm_state comp)
 219 {
 220         unsigned long flags;
 221         int ret;
 222
 223         spin_lock_irqsave(&id_priv->lock, flags);
 224         ret = (id_priv->state == comp);
 225         spin_unlock_irqrestore(&id_priv->lock, flags);
 226         return ret;
 227 }
 228
 229 static int cma_comp_exch(struct rdma_id_private *id_priv,
 230                          enum rdma_cm_state comp, enum rdma_cm_state exch)
 231 {
 232         unsigned long flags;
 233         int ret;
 234
 235         spin_lock_irqsave(&id_priv->lock, flags);
 236         if ((ret = (id_priv->state == comp)))
 237                 id_priv->state = exch;
 238         spin_unlock_irqrestore(&id_priv->lock, flags);
 239         return ret;
 240 }
 241
 242 static enum rdma_cm_state cma_exch(struct rdma_id_private *id_priv,
 243                                    enum rdma_cm_state exch)
 244 {
 245         unsigned long flags;
 246         enum rdma_cm_state old;
 247
 248         spin_lock_irqsave(&id_priv->lock, flags);
 249         old = id_priv->state;
 250         id_priv->state = exch;
 251         spin_unlock_irqrestore(&id_priv->lock, flags);
 252         return old;
 253 }
 254
 255 static inline u8 cma_get_ip_ver(struct cma_hdr *hdr)
 256 {
 257         return hdr->ip_version >> 4;
 258 }
 259
 260 static inline void cma_set_ip_ver(struct cma_hdr *hdr, u8 ip_ver)
 261 {
 262         hdr->ip_version = (ip_ver << 4) | (hdr->ip_version & 0xF);
 263 }
 264
 265 static inline u8 sdp_get_majv(u8 sdp_version)
 266 {
 267         return sdp_version >> 4;
 268 }
 269
 270 static inline u8 sdp_get_ip_ver(struct sdp_hh *hh)
 271 {
 272         return hh->ip_version >> 4;
 273 }
 274
 275 static inline void sdp_set_ip_ver(struct sdp_hh *hh, u8 ip_ver)
 276 {
 277         hh->ip_version = (ip_ver << 4) | (hh->ip_version & 0xF);
 278 }
 279
 280 static void cma_attach_to_dev(struct rdma_id_private *id_priv,
 281                               struct cma_device *cma_dev)
 282 {
 283         atomic_inc(&cma_dev->refcount);
 284         id_priv->cma_dev = cma_dev;
 285         id_priv->id.device = cma_dev->device;
 286         id_priv->id.route.addr.dev_addr.transport =
 287                 rdma_node_get_transport(cma_dev->device->node_type);
 288         list_add_tail(&id_priv->list, &cma_dev->id_list);
 289 }
 290
 291 static inline void cma_deref_dev(struct cma_device *cma_dev)
 292 {
 293         if (atomic_dec_and_test(&cma_dev->refcount))
 294                 complete(&cma_dev->comp);
 295 }
 296
 297 static inline void release_mc(struct kref *kref)
 298 {
 299         struct cma_multicast *mc = container_of(kref, struct cma_multicast, mcref);
 300
 301         kfree(mc->multicast.ib);
 302         kfree(mc);
 303 }
 304
 305 static void cma_release_dev(struct rdma_id_private *id_priv)
 306 {
 307         mutex_lock(&lock);
 308         list_del(&id_priv->list);
 309         cma_deref_dev(id_priv->cma_dev);
 310         id_priv->cma_dev = NULL;
 311         mutex_unlock(&lock);
 312 }
 313
 314 static inline struct sockaddr *cma_src_addr(struct rdma_id_private *id_priv)
 315 {
 316         return (struct sockaddr *) &id_priv->id.route.addr.src_addr;
 317 }
 318
 319 static inline struct sockaddr *cma_dst_addr(struct rdma_id_private *id_priv)
 320 {
 321         return (struct sockaddr *) &id_priv->id.route.addr.dst_addr;
 322 }
 323
 324 static inline unsigned short cma_family(struct rdma_id_private *id_priv)
 325 {
 326         return id_priv->id.route.addr.src_addr.ss_family;
 327 }
 328
 329 static int cma_set_qkey(struct rdma_id_private *id_priv)
 330 {
 331         struct ib_sa_mcmember_rec rec;
 332         int ret = 0;
 333
 334         if (id_priv->qkey)
 335                 return 0;
 336
 337         switch (id_priv->id.ps) {
 338         case RDMA_PS_UDP:
 339                 id_priv->qkey = RDMA_UDP_QKEY;
 340                 break;
 341         case RDMA_PS_IPOIB:
 342                 ib_addr_get_mgid(&id_priv->id.route.addr.dev_addr, &rec.mgid);
 343                 ret = ib_sa_get_mcmember_rec(id_priv->id.device,
 344                                              id_priv->id.port_num, &rec.mgid,
 345                                              &rec);
 346                 if (!ret)
 347                         id_priv->qkey = be32_to_cpu(rec.qkey);
 348                 break;
 349         default:
 350                 break;
 351         }
 352         return ret;
 353 }
 354
 355 static int find_gid_port(struct ib_device *device, union ib_gid *gid, u8 port_num)
 356 {
 357         int i;
 358         int err;
 359         struct ib_port_attr props;
 360         union ib_gid tmp;
 361
 362         err = ib_query_port(device, port_num, &props);
 363         if (err)
 364                 return err;
 365
 366         for (i = 0; i < props.gid_tbl_len; ++i) {
 367                 err = ib_query_gid(device, port_num, i, &tmp);
 368                 if (err)
 369                         return err;
 370                 if (!memcmp(&tmp, gid, sizeof tmp))
 371                         return 0;
 372         }
 373
 374         return -EADDRNOTAVAIL;
 375 }
 376
 377 static void cma_translate_ib(struct sockaddr_ib *sib, struct rdma_dev_addr *dev_addr)
 378 {
 379         dev_addr->dev_type = ARPHRD_INFINIBAND;
 380         rdma_addr_set_sgid(dev_addr, (union ib_gid *) &sib->sib_addr);
 381         ib_addr_set_pkey(dev_addr, ntohs(sib->sib_pkey));
 382 }
 383
 384 static int cma_translate_addr(struct sockaddr *addr, struct rdma_dev_addr *dev_addr)
 385 {
 386         int ret;
 387
 388         if (addr->sa_family != AF_IB) {
 389                 ret = rdma_translate_ip(addr, dev_addr);
 390         } else {
 391                 cma_translate_ib((struct sockaddr_ib *) addr, dev_addr);
 392                 ret = 0;
 393         }
 394
 395         return ret;
 396 }
 397
 398 static int cma_acquire_dev(struct rdma_id_private *id_priv)
 399 {
 400         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
 401         struct cma_device *cma_dev;
 402         union ib_gid gid, iboe_gid;
 403         int ret = -ENODEV;
 404         u8 port;
 405         enum rdma_link_layer dev_ll = dev_addr->dev_type == ARPHRD_INFINIBAND ?
 406                 IB_LINK_LAYER_INFINIBAND : IB_LINK_LAYER_ETHERNET;
 407
 408         if (dev_ll != IB_LINK_LAYER_INFINIBAND &&
 409             id_priv->id.ps == RDMA_PS_IPOIB)
 410                 return -EINVAL;
 411
 412         mutex_lock(&lock);
 413         iboe_addr_get_sgid(dev_addr, &iboe_gid);
 414         memcpy(&gid, dev_addr->src_dev_addr +
 415                rdma_addr_gid_offset(dev_addr), sizeof gid);
 416         list_for_each_entry(cma_dev, &dev_list, list) {
 417                 for (port = 1; port <= cma_dev->device->phys_port_cnt; ++port) {
 418                         if (rdma_port_get_link_layer(cma_dev->device, port) == dev_ll) {
 419                                 if (rdma_node_get_transport(cma_dev->device->node_type) == RDMA_TRANSPORT_IB &&
 420                                     rdma_port_get_link_layer(cma_dev->device, port) == IB_LINK_LAYER_ETHERNET)
 421                                         ret = find_gid_port(cma_dev->device, &iboe_gid, port);
 422                                 else
 423                                         ret = find_gid_port(cma_dev->device, &gid, port);
 424
 425                                 if (!ret) {
 426                                         id_priv->id.port_num = port;
 427                                         goto out;
 428                                 }
 429                         }
 430                 }
 431         }
 432
 433 out:
 434         if (!ret)
 435                 cma_attach_to_dev(id_priv, cma_dev);
 436
 437         mutex_unlock(&lock);
 438         return ret;
 439 }
 440
 441 /*
 442  * Select the source IB device and address to reach the destination IB address.
 443  */
 444 static int cma_resolve_ib_dev(struct rdma_id_private *id_priv)
 445 {
 446         struct cma_device *cma_dev, *cur_dev;
 447         struct sockaddr_ib *addr;
 448         union ib_gid gid, sgid, *dgid;
 449         u16 pkey, index;
 450         u8 port, p;
 451         int i;
 452
 453         cma_dev = NULL;
 454         addr = (struct sockaddr_ib *) cma_dst_addr(id_priv);
 455         dgid = (union ib_gid *) &addr->sib_addr;
 456         pkey = ntohs(addr->sib_pkey);
 457
 458         list_for_each_entry(cur_dev, &dev_list, list) {
 459                 if (rdma_node_get_transport(cur_dev->device->node_type) != RDMA_TRANSPORT_IB)
 460                         continue;
 461
 462                 for (p = 1; p <= cur_dev->device->phys_port_cnt; ++p) {
 463                         if (ib_find_cached_pkey(cur_dev->device, p, pkey, &index))
 464                                 continue;
 465
 466                         for (i = 0; !ib_get_cached_gid(cur_dev->device, p, i, &gid); i++) {
 467                                 if (!memcmp(&gid, dgid, sizeof(gid))) {
 468                                         cma_dev = cur_dev;
 469                                         sgid = gid;
 470                                         port = p;
 471                                         goto found;
 472                                 }
 473
 474                                 if (!cma_dev && (gid.global.subnet_prefix ==
 475                                                  dgid->global.subnet_prefix)) {
 476                                         cma_dev = cur_dev;
 477                                         sgid = gid;
 478                                         port = p;
 479                                 }
 480                         }
 481                 }
 482         }
 483
 484         if (!cma_dev)
 485                 return -ENODEV;
 486
 487 found:
 488         cma_attach_to_dev(id_priv, cma_dev);
 489         id_priv->id.port_num = port;
 490         addr = (struct sockaddr_ib *) cma_src_addr(id_priv);
 491         memcpy(&addr->sib_addr, &sgid, sizeof sgid);
 492         cma_translate_ib(addr, &id_priv->id.route.addr.dev_addr);
 493         return 0;
 494 }
 495
 496 static void cma_deref_id(struct rdma_id_private *id_priv)
 497 {
 498         if (atomic_dec_and_test(&id_priv->refcount))
 499                 complete(&id_priv->comp);
 500 }
 501
 502 static int cma_disable_callback(struct rdma_id_private *id_priv,
 503                                 enum rdma_cm_state state)
 504 {
 505         mutex_lock(&id_priv->handler_mutex);
 506         if (id_priv->state != state) {
 507                 mutex_unlock(&id_priv->handler_mutex);
 508                 return -EINVAL;
 509         }
 510         return 0;
 511 }
 512
 513 struct rdma_cm_id *rdma_create_id(rdma_cm_event_handler event_handler,
 514                                   void *context, enum rdma_port_space ps,
 515                                   enum ib_qp_type qp_type)
 516 {
 517         struct rdma_id_private *id_priv;
 518
 519         id_priv = kzalloc(sizeof *id_priv, GFP_KERNEL);
 520         if (!id_priv)
 521                 return ERR_PTR(-ENOMEM);
 522
 523         id_priv->owner = task_pid_nr(current);
 524         id_priv->state = RDMA_CM_IDLE;
 525         id_priv->id.context = context;
 526         id_priv->id.event_handler = event_handler;
 527         id_priv->id.ps = ps;
 528         id_priv->id.qp_type = qp_type;
 529         spin_lock_init(&id_priv->lock);
 530         mutex_init(&id_priv->qp_mutex);
 531         init_completion(&id_priv->comp);
 532         atomic_set(&id_priv->refcount, 1);
 533         mutex_init(&id_priv->handler_mutex);
 534         INIT_LIST_HEAD(&id_priv->listen_list);
 535         INIT_LIST_HEAD(&id_priv->mc_list);
 536         get_random_bytes(&id_priv->seq_num, sizeof id_priv->seq_num);
 537
 538         return &id_priv->id;
 539 }
 540 EXPORT_SYMBOL(rdma_create_id);
 541
 542 static int cma_init_ud_qp(struct rdma_id_private *id_priv, struct ib_qp *qp)
 543 {
 544         struct ib_qp_attr qp_attr;
 545         int qp_attr_mask, ret;
 546
 547         qp_attr.qp_state = IB_QPS_INIT;
 548         ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
 549         if (ret)
 550                 return ret;
 551
 552         ret = ib_modify_qp(qp, &qp_attr, qp_attr_mask);
 553         if (ret)
 554                 return ret;
 555
 556         qp_attr.qp_state = IB_QPS_RTR;
 557         ret = ib_modify_qp(qp, &qp_attr, IB_QP_STATE);
 558         if (ret)
 559                 return ret;
 560
 561         qp_attr.qp_state = IB_QPS_RTS;
 562         qp_attr.sq_psn = 0;
 563         ret = ib_modify_qp(qp, &qp_attr, IB_QP_STATE | IB_QP_SQ_PSN);
 564
 565         return ret;
 566 }
 567
 568 static int cma_init_conn_qp(struct rdma_id_private *id_priv, struct ib_qp *qp)
 569 {
 570         struct ib_qp_attr qp_attr;
 571         int qp_attr_mask, ret;
 572
 573         qp_attr.qp_state = IB_QPS_INIT;
 574         ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
 575         if (ret)
 576                 return ret;
 577
 578         return ib_modify_qp(qp, &qp_attr, qp_attr_mask);
 579 }
 580
 581 int rdma_create_qp(struct rdma_cm_id *id, struct ib_pd *pd,
 582                    struct ib_qp_init_attr *qp_init_attr)
 583 {
 584         struct rdma_id_private *id_priv;
 585         struct ib_qp *qp;
 586         int ret;
 587
 588         id_priv = container_of(id, struct rdma_id_private, id);
 589         if (id->device != pd->device)
 590                 return -EINVAL;
 591
 592         qp = ib_create_qp(pd, qp_init_attr);
 593         if (IS_ERR(qp))
 594                 return PTR_ERR(qp);
 595
 596         if (id->qp_type == IB_QPT_UD)
 597                 ret = cma_init_ud_qp(id_priv, qp);
 598         else
 599                 ret = cma_init_conn_qp(id_priv, qp);
 600         if (ret)
 601                 goto err;
 602
 603         id->qp = qp;
 604         id_priv->qp_num = qp->qp_num;
 605         id_priv->srq = (qp->srq != NULL);
 606         return 0;
 607 err:
 608         ib_destroy_qp(qp);
 609         return ret;
 610 }
 611 EXPORT_SYMBOL(rdma_create_qp);
 612
 613 void rdma_destroy_qp(struct rdma_cm_id *id)
 614 {
 615         struct rdma_id_private *id_priv;
 616
 617         id_priv = container_of(id, struct rdma_id_private, id);
 618         mutex_lock(&id_priv->qp_mutex);
 619         ib_destroy_qp(id_priv->id.qp);
 620         id_priv->id.qp = NULL;
 621         mutex_unlock(&id_priv->qp_mutex);
 622 }
 623 EXPORT_SYMBOL(rdma_destroy_qp);
 624
 625 static int cma_modify_qp_rtr(struct rdma_id_private *id_priv,
 626                              struct rdma_conn_param *conn_param)
 627 {
 628         struct ib_qp_attr qp_attr;
 629         int qp_attr_mask, ret;
 630
 631         mutex_lock(&id_priv->qp_mutex);
 632         if (!id_priv->id.qp) {
 633                 ret = 0;
 634                 goto out;
 635         }
 636
 637         /* Need to update QP attributes from default values. */
 638         qp_attr.qp_state = IB_QPS_INIT;
 639         ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
 640         if (ret)
 641                 goto out;
 642
 643         ret = ib_modify_qp(id_priv->id.qp, &qp_attr, qp_attr_mask);
 644         if (ret)
 645                 goto out;
 646
 647         qp_attr.qp_state = IB_QPS_RTR;
 648         ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
 649         if (ret)
 650                 goto out;
 651
 652         if (conn_param)
 653                 qp_attr.max_dest_rd_atomic = conn_param->responder_resources;
 654         ret = ib_modify_qp(id_priv->id.qp, &qp_attr, qp_attr_mask);
 655 out:
 656         mutex_unlock(&id_priv->qp_mutex);
 657         return ret;
 658 }
 659
 660 static int cma_modify_qp_rts(struct rdma_id_private *id_priv,
 661                              struct rdma_conn_param *conn_param)
 662 {
 663         struct ib_qp_attr qp_attr;
 664         int qp_attr_mask, ret;
 665
 666         mutex_lock(&id_priv->qp_mutex);
 667         if (!id_priv->id.qp) {
 668                 ret = 0;
 669                 goto out;
 670         }
 671
 672         qp_attr.qp_state = IB_QPS_RTS;
 673         ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
 674         if (ret)
 675                 goto out;
 676
 677         if (conn_param)
 678                 qp_attr.max_rd_atomic = conn_param->initiator_depth;
 679         ret = ib_modify_qp(id_priv->id.qp, &qp_attr, qp_attr_mask);
 680 out:
 681         mutex_unlock(&id_priv->qp_mutex);
 682         return ret;
 683 }
 684
 685 static int cma_modify_qp_err(struct rdma_id_private *id_priv)
 686 {
 687         struct ib_qp_attr qp_attr;
 688         int ret;
 689
 690         mutex_lock(&id_priv->qp_mutex);
 691         if (!id_priv->id.qp) {
 692                 ret = 0;
 693                 goto out;
 694         }
 695
 696         qp_attr.qp_state = IB_QPS_ERR;
 697         ret = ib_modify_qp(id_priv->id.qp, &qp_attr, IB_QP_STATE);
 698 out:
 699         mutex_unlock(&id_priv->qp_mutex);
 700         return ret;
 701 }
 702
 703 static int cma_ib_init_qp_attr(struct rdma_id_private *id_priv,
 704                                struct ib_qp_attr *qp_attr, int *qp_attr_mask)
 705 {
 706         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
 707         int ret;
 708         u16 pkey;
 709
 710         if (rdma_port_get_link_layer(id_priv->id.device, id_priv->id.port_num) ==
 711             IB_LINK_LAYER_INFINIBAND)
 712                 pkey = ib_addr_get_pkey(dev_addr);
 713         else
 714                 pkey = 0xffff;
 715
 716         ret = ib_find_cached_pkey(id_priv->id.device, id_priv->id.port_num,
 717                                   pkey, &qp_attr->pkey_index);
 718         if (ret)
 719                 return ret;
 720
 721         qp_attr->port_num = id_priv->id.port_num;
 722         *qp_attr_mask = IB_QP_STATE | IB_QP_PKEY_INDEX | IB_QP_PORT;
 723
 724         if (id_priv->id.qp_type == IB_QPT_UD) {
 725                 ret = cma_set_qkey(id_priv);
 726                 if (ret)
 727                         return ret;
 728
 729                 qp_attr->qkey = id_priv->qkey;
 730                 *qp_attr_mask |= IB_QP_QKEY;
 731         } else {
 732                 qp_attr->qp_access_flags = 0;
 733                 *qp_attr_mask |= IB_QP_ACCESS_FLAGS;
 734         }
 735         return 0;
 736 }
 737
 738 int rdma_init_qp_attr(struct rdma_cm_id *id, struct ib_qp_attr *qp_attr,
 739                        int *qp_attr_mask)
 740 {
 741         struct rdma_id_private *id_priv;
 742         int ret = 0;
 743
 744         id_priv = container_of(id, struct rdma_id_private, id);
 745         switch (rdma_node_get_transport(id_priv->id.device->node_type)) {
 746         case RDMA_TRANSPORT_IB:
 747                 if (!id_priv->cm_id.ib || (id_priv->id.qp_type == IB_QPT_UD))
 748                         ret = cma_ib_init_qp_attr(id_priv, qp_attr, qp_attr_mask);
 749                 else
 750                         ret = ib_cm_init_qp_attr(id_priv->cm_id.ib, qp_attr,
 751                                                  qp_attr_mask);
 752                 if (qp_attr->qp_state == IB_QPS_RTR)
 753                         qp_attr->rq_psn = id_priv->seq_num;
 754                 break;
 755         case RDMA_TRANSPORT_IWARP:
 756                 if (!id_priv->cm_id.iw) {
 757                         qp_attr->qp_access_flags = 0;
 758                         *qp_attr_mask = IB_QP_STATE | IB_QP_ACCESS_FLAGS;
 759                 } else
 760                         ret = iw_cm_init_qp_attr(id_priv->cm_id.iw, qp_attr,
 761                                                  qp_attr_mask);
 762                 break;
 763         default:
 764                 ret = -ENOSYS;
 765                 break;
 766         }
 767
 768         return ret;
 769 }
 770 EXPORT_SYMBOL(rdma_init_qp_attr);
 771
 772 static inline int cma_zero_addr(struct sockaddr *addr)
 773 {
 774         switch (addr->sa_family) {
 775         case AF_INET:
 776                 return ipv4_is_zeronet(((struct sockaddr_in *)addr)->sin_addr.s_addr);
 777         case AF_INET6:
 778                 return ipv6_addr_any(&((struct sockaddr_in6 *) addr)->sin6_addr);
 779         case AF_IB:
 780                 return ib_addr_any(&((struct sockaddr_ib *) addr)->sib_addr);
 781         default:
 782                 return 0;
 783         }
 784 }
 785
 786 static inline int cma_loopback_addr(struct sockaddr *addr)
 787 {
 788         switch (addr->sa_family) {
 789         case AF_INET:
 790                 return ipv4_is_loopback(((struct sockaddr_in *) addr)->sin_addr.s_addr);
 791         case AF_INET6:
 792                 return ipv6_addr_loopback(&((struct sockaddr_in6 *) addr)->sin6_addr);
 793         case AF_IB:
 794                 return ib_addr_loopback(&((struct sockaddr_ib *) addr)->sib_addr);
 795         default:
 796                 return 0;
 797         }
 798 }
 799
 800 static inline int cma_any_addr(struct sockaddr *addr)
 801 {
 802         return cma_zero_addr(addr) || cma_loopback_addr(addr);
 803 }
 804
 805 static int cma_addr_cmp(struct sockaddr *src, struct sockaddr *dst)
 806 {
 807         if (src->sa_family != dst->sa_family)
 808                 return -1;
 809
 810         switch (src->sa_family) {
 811         case AF_INET:
 812                 return ((struct sockaddr_in *) src)->sin_addr.s_addr !=
 813                        ((struct sockaddr_in *) dst)->sin_addr.s_addr;
 814         case AF_INET6:
 815                 return ipv6_addr_cmp(&((struct sockaddr_in6 *) src)->sin6_addr,
 816                                      &((struct sockaddr_in6 *) dst)->sin6_addr);
 817         default:
 818                 return ib_addr_cmp(&((struct sockaddr_ib *) src)->sib_addr,
 819                                    &((struct sockaddr_ib *) dst)->sib_addr);
 820         }
 821 }
 822
 823 static __be16 cma_port(struct sockaddr *addr)
 824 {
 825         struct sockaddr_ib *sib;
 826
 827         switch (addr->sa_family) {
 828         case AF_INET:
 829                 return ((struct sockaddr_in *) addr)->sin_port;
 830         case AF_INET6:
 831                 return ((struct sockaddr_in6 *) addr)->sin6_port;
 832         case AF_IB:
 833                 sib = (struct sockaddr_ib *) addr;
 834                 return htons((u16) (be64_to_cpu(sib->sib_sid) &
 835                                     be64_to_cpu(sib->sib_sid_mask)));
 836         default:
 837                 return 0;
 838         }
 839 }
 840
 841 static inline int cma_any_port(struct sockaddr *addr)
 842 {
 843         return !cma_port(addr);
 844 }
 845
 846 static int cma_get_net_info(void *hdr, enum rdma_port_space ps,
 847                             u8 *ip_ver, __be16 *port,
 848                             union cma_ip_addr **src, union cma_ip_addr **dst)
 849 {
 850         switch (ps) {
 851         case RDMA_PS_SDP:
 852                 if (sdp_get_majv(((struct sdp_hh *) hdr)->sdp_version) !=
 853                     SDP_MAJ_VERSION)
 854                         return -EINVAL;
 855
 856                 *ip_ver = sdp_get_ip_ver(hdr);
 857                 *port   = ((struct sdp_hh *) hdr)->port;
 858                 *src    = &((struct sdp_hh *) hdr)->src_addr;
 859                 *dst    = &((struct sdp_hh *) hdr)->dst_addr;
 860                 break;
 861         default:
 862                 if (((struct cma_hdr *) hdr)->cma_version != CMA_VERSION)
 863                         return -EINVAL;
 864
 865                 *ip_ver = cma_get_ip_ver(hdr);
 866                 *port   = ((struct cma_hdr *) hdr)->port;
 867                 *src    = &((struct cma_hdr *) hdr)->src_addr;
 868                 *dst    = &((struct cma_hdr *) hdr)->dst_addr;
 869                 break;
 870         }
 871
 872         if (*ip_ver != 4 && *ip_ver != 6)
 873                 return -EINVAL;
 874         return 0;
 875 }
 876
 877 static void cma_save_net_info(struct rdma_addr *addr,
 878                               struct rdma_addr *listen_addr,
 879                               u8 ip_ver, __be16 port,
 880                               union cma_ip_addr *src, union cma_ip_addr *dst)
 881 {
 882         struct sockaddr_in *listen4, *ip4;
 883         struct sockaddr_in6 *listen6, *ip6;
 884
 885         switch (ip_ver) {
 886         case 4:
 887                 listen4 = (struct sockaddr_in *) &listen_addr->src_addr;
 888                 ip4 = (struct sockaddr_in *) &addr->src_addr;
 889                 ip4->sin_family = listen4->sin_family;
 890                 ip4->sin_addr.s_addr = dst->ip4.addr;
 891                 ip4->sin_port = listen4->sin_port;
 892
 893                 ip4 = (struct sockaddr_in *) &addr->dst_addr;
 894                 ip4->sin_family = listen4->sin_family;
 895                 ip4->sin_addr.s_addr = src->ip4.addr;
 896                 ip4->sin_port = port;
 897                 break;
 898         case 6:
 899                 listen6 = (struct sockaddr_in6 *) &listen_addr->src_addr;
 900                 ip6 = (struct sockaddr_in6 *) &addr->src_addr;
 901                 ip6->sin6_family = listen6->sin6_family;
 902                 ip6->sin6_addr = dst->ip6;
 903                 ip6->sin6_port = listen6->sin6_port;
 904
 905                 ip6 = (struct sockaddr_in6 *) &addr->dst_addr;
 906                 ip6->sin6_family = listen6->sin6_family;
 907                 ip6->sin6_addr = src->ip6;
 908                 ip6->sin6_port = port;
 909                 break;
 910         default:
 911                 break;
 912         }
 913 }
 914
 915 static inline int cma_user_data_offset(enum rdma_port_space ps)
 916 {
 917         switch (ps) {
 918         case RDMA_PS_SDP:
 919                 return 0;
 920         default:
 921                 return sizeof(struct cma_hdr);
 922         }
 923 }
 924
 925 static void cma_cancel_route(struct rdma_id_private *id_priv)
 926 {
 927         switch (rdma_port_get_link_layer(id_priv->id.device, id_priv->id.port_num)) {
 928         case IB_LINK_LAYER_INFINIBAND:
 929                 if (id_priv->query)
 930                         ib_sa_cancel_query(id_priv->query_id, id_priv->query);
 931                 break;
 932         default:
 933                 break;
 934         }
 935 }
 936
 937 static void cma_cancel_listens(struct rdma_id_private *id_priv)
 938 {
 939         struct rdma_id_private *dev_id_priv;
 940
 941         /*
 942          * Remove from listen_any_list to prevent added devices from spawning
 943          * additional listen requests.
 944          */
 945         mutex_lock(&lock);
 946         list_del(&id_priv->list);
 947
 948         while (!list_empty(&id_priv->listen_list)) {
 949                 dev_id_priv = list_entry(id_priv->listen_list.next,
 950                                          struct rdma_id_private, listen_list);
 951                 /* sync with device removal to avoid duplicate destruction */
 952                 list_del_init(&dev_id_priv->list);
 953                 list_del(&dev_id_priv->listen_list);
 954                 mutex_unlock(&lock);
 955
 956                 rdma_destroy_id(&dev_id_priv->id);
 957                 mutex_lock(&lock);
 958         }
 959         mutex_unlock(&lock);
 960 }
 961
 962 static void cma_cancel_operation(struct rdma_id_private *id_priv,
 963                                  enum rdma_cm_state state)
 964 {
 965         switch (state) {
 966         case RDMA_CM_ADDR_QUERY:
 967                 rdma_addr_cancel(&id_priv->id.route.addr.dev_addr);
 968                 break;
 969         case RDMA_CM_ROUTE_QUERY:
 970                 cma_cancel_route(id_priv);
 971                 break;
 972         case RDMA_CM_LISTEN:
 973                 if (cma_any_addr(cma_src_addr(id_priv)) && !id_priv->cma_dev)
 974                         cma_cancel_listens(id_priv);
 975                 break;
 976         default:
 977                 break;
 978         }
 979 }
 980
 981 static void cma_release_port(struct rdma_id_private *id_priv)
 982 {
 983         struct rdma_bind_list *bind_list = id_priv->bind_list;
 984
 985         if (!bind_list)
 986                 return;
 987
 988         mutex_lock(&lock);
 989         hlist_del(&id_priv->node);
 990         if (hlist_empty(&bind_list->owners)) {
 991                 idr_remove(bind_list->ps, bind_list->port);
 992                 kfree(bind_list);
 993         }
 994         mutex_unlock(&lock);
 995 }
 996
 997 static void cma_leave_mc_groups(struct rdma_id_private *id_priv)
 998 {
 999         struct cma_multicast *mc;
1000
1001         while (!list_empty(&id_priv->mc_list)) {
1002                 mc = container_of(id_priv->mc_list.next,
1003                                   struct cma_multicast, list);
1004                 list_del(&mc->list);
1005                 switch (rdma_port_get_link_layer(id_priv->cma_dev->device, id_priv->id.port_num)) {
1006                 case IB_LINK_LAYER_INFINIBAND:
1007                         ib_sa_free_multicast(mc->multicast.ib);
1008                         kfree(mc);
1009                         break;
1010                 case IB_LINK_LAYER_ETHERNET:
1011                         kref_put(&mc->mcref, release_mc);
1012                         break;
1013                 default:
1014                         break;
1015                 }
1016         }
1017 }
1018
1019 void rdma_destroy_id(struct rdma_cm_id *id)
1020 {
1021         struct rdma_id_private *id_priv;
1022         enum rdma_cm_state state;
1023
1024         id_priv = container_of(id, struct rdma_id_private, id);
1025         state = cma_exch(id_priv, RDMA_CM_DESTROYING);
1026         cma_cancel_operation(id_priv, state);
1027
1028         /*
1029          * Wait for any active callback to finish.  New callbacks will find
1030          * the id_priv state set to destroying and abort.
1031          */
1032         mutex_lock(&id_priv->handler_mutex);
1033         mutex_unlock(&id_priv->handler_mutex);
1034
1035         if (id_priv->cma_dev) {
1036                 switch (rdma_node_get_transport(id_priv->id.device->node_type)) {
1037                 case RDMA_TRANSPORT_IB:
1038                         if (id_priv->cm_id.ib)
1039                                 ib_destroy_cm_id(id_priv->cm_id.ib);
1040                         break;
1041                 case RDMA_TRANSPORT_IWARP:
1042                         if (id_priv->cm_id.iw)
1043                                 iw_destroy_cm_id(id_priv->cm_id.iw);
1044                         break;
1045                 default:
1046                         break;
1047                 }
1048                 cma_leave_mc_groups(id_priv);
1049                 cma_release_dev(id_priv);
1050         }
1051
1052         cma_release_port(id_priv);
1053         cma_deref_id(id_priv);
1054         wait_for_completion(&id_priv->comp);
1055
1056         if (id_priv->internal_id)
1057                 cma_deref_id(id_priv->id.context);
1058
1059         kfree(id_priv->id.route.path_rec);
1060         kfree(id_priv);
1061 }
1062 EXPORT_SYMBOL(rdma_destroy_id);
1063
1064 static int cma_rep_recv(struct rdma_id_private *id_priv)
1065 {
1066         int ret;
1067
1068         ret = cma_modify_qp_rtr(id_priv, NULL);
1069         if (ret)
1070                 goto reject;
1071
1072         ret = cma_modify_qp_rts(id_priv, NULL);
1073         if (ret)
1074                 goto reject;
1075
1076         ret = ib_send_cm_rtu(id_priv->cm_id.ib, NULL, 0);
1077         if (ret)
1078                 goto reject;
1079
1080         return 0;
1081 reject:
1082         cma_modify_qp_err(id_priv);
1083         ib_send_cm_rej(id_priv->cm_id.ib, IB_CM_REJ_CONSUMER_DEFINED,
1084                        NULL, 0, NULL, 0);
1085         return ret;
1086 }
1087
1088 static int cma_verify_rep(struct rdma_id_private *id_priv, void *data)
1089 {
1090         if (id_priv->id.ps == RDMA_PS_SDP &&
1091             sdp_get_majv(((struct sdp_hah *) data)->sdp_version) !=
1092             SDP_MAJ_VERSION)
1093                 return -EINVAL;
1094
1095         return 0;
1096 }
1097
1098 static void cma_set_rep_event_data(struct rdma_cm_event *event,
1099                                    struct ib_cm_rep_event_param *rep_data,
1100                                    void *private_data)
1101 {
1102         event->param.conn.private_data = private_data;
1103         event->param.conn.private_data_len = IB_CM_REP_PRIVATE_DATA_SIZE;
1104         event->param.conn.responder_resources = rep_data->responder_resources;
1105         event->param.conn.initiator_depth = rep_data->initiator_depth;
1106         event->param.conn.flow_control = rep_data->flow_control;
1107         event->param.conn.rnr_retry_count = rep_data->rnr_retry_count;
1108         event->param.conn.srq = rep_data->srq;
1109         event->param.conn.qp_num = rep_data->remote_qpn;
1110 }
1111
1112 static int cma_ib_handler(struct ib_cm_id *cm_id, struct ib_cm_event *ib_event)
1113 {
1114         struct rdma_id_private *id_priv = cm_id->context;
1115         struct rdma_cm_event event;
1116         int ret = 0;
1117
1118         if ((ib_event->event != IB_CM_TIMEWAIT_EXIT &&
1119                 cma_disable_callback(id_priv, RDMA_CM_CONNECT)) ||
1120             (ib_event->event == IB_CM_TIMEWAIT_EXIT &&
1121                 cma_disable_callback(id_priv, RDMA_CM_DISCONNECT)))
1122                 return 0;
1123
1124         memset(&event, 0, sizeof event);
1125         switch (ib_event->event) {
1126         case IB_CM_REQ_ERROR:
1127         case IB_CM_REP_ERROR:
1128                 event.event = RDMA_CM_EVENT_UNREACHABLE;
1129                 event.status = -ETIMEDOUT;
1130                 break;
1131         case IB_CM_REP_RECEIVED:
1132                 event.status = cma_verify_rep(id_priv, ib_event->private_data);
1133                 if (event.status)
1134                         event.event = RDMA_CM_EVENT_CONNECT_ERROR;
1135                 else if (id_priv->id.qp && id_priv->id.ps != RDMA_PS_SDP) {
1136                         event.status = cma_rep_recv(id_priv);
1137                         event.event = event.status ? RDMA_CM_EVENT_CONNECT_ERROR :
1138                                                      RDMA_CM_EVENT_ESTABLISHED;
1139                 } else
1140                         event.event = RDMA_CM_EVENT_CONNECT_RESPONSE;
1141                 cma_set_rep_event_data(&event, &ib_event->param.rep_rcvd,
1142                                        ib_event->private_data);
1143                 break;
1144         case IB_CM_RTU_RECEIVED:
1145         case IB_CM_USER_ESTABLISHED:
1146                 event.event = RDMA_CM_EVENT_ESTABLISHED;
1147                 break;
1148         case IB_CM_DREQ_ERROR:
1149                 event.status = -ETIMEDOUT; /* fall through */
1150         case IB_CM_DREQ_RECEIVED:
1151         case IB_CM_DREP_RECEIVED:
1152                 if (!cma_comp_exch(id_priv, RDMA_CM_CONNECT,
1153                                    RDMA_CM_DISCONNECT))
1154                         goto out;
1155                 event.event = RDMA_CM_EVENT_DISCONNECTED;
1156                 break;
1157         case IB_CM_TIMEWAIT_EXIT:
1158                 event.event = RDMA_CM_EVENT_TIMEWAIT_EXIT;
1159                 break;
1160         case IB_CM_MRA_RECEIVED:
1161                 /* ignore event */
1162                 goto out;
1163         case IB_CM_REJ_RECEIVED:
1164                 cma_modify_qp_err(id_priv);
1165                 event.status = ib_event->param.rej_rcvd.reason;
1166                 event.event = RDMA_CM_EVENT_REJECTED;
1167                 event.param.conn.private_data = ib_event->private_data;
1168                 event.param.conn.private_data_len = IB_CM_REJ_PRIVATE_DATA_SIZE;
1169                 break;
1170         default:
1171                 printk(KERN_ERR "RDMA CMA: unexpected IB CM event: %d\n",
1172                        ib_event->event);
1173                 goto out;
1174         }
1175
1176         ret = id_priv->id.event_handler(&id_priv->id, &event);
1177         if (ret) {
1178                 /* Destroy the CM ID by returning a non-zero value. */
1179                 id_priv->cm_id.ib = NULL;
1180                 cma_exch(id_priv, RDMA_CM_DESTROYING);
1181                 mutex_unlock(&id_priv->handler_mutex);
1182                 rdma_destroy_id(&id_priv->id);
1183                 return ret;
1184         }
1185 out:
1186         mutex_unlock(&id_priv->handler_mutex);
1187         return ret;
1188 }
1189
1190 static struct rdma_id_private *cma_new_conn_id(struct rdma_cm_id *listen_id,
1191                                                struct ib_cm_event *ib_event)
1192 {
1193         struct rdma_id_private *id_priv;
1194         struct rdma_cm_id *id;
1195         struct rdma_route *rt;
1196         union cma_ip_addr *src, *dst;
1197         __be16 port;
1198         u8 ip_ver;
1199         int ret;
1200
1201         if (cma_get_net_info(ib_event->private_data, listen_id->ps,
1202                              &ip_ver, &port, &src, &dst))
1203                 return NULL;
1204
1205         id = rdma_create_id(listen_id->event_handler, listen_id->context,
1206                             listen_id->ps, ib_event->param.req_rcvd.qp_type);
1207         if (IS_ERR(id))
1208                 return NULL;
1209
1210         id_priv = container_of(id, struct rdma_id_private, id);
1211         cma_save_net_info(&id->route.addr, &listen_id->route.addr,
1212                           ip_ver, port, src, dst);
1213
1214         rt = &id->route;
1215         rt->num_paths = ib_event->param.req_rcvd.alternate_path ? 2 : 1;
1216         rt->path_rec = kmalloc(sizeof *rt->path_rec * rt->num_paths,
1217                                GFP_KERNEL);
1218         if (!rt->path_rec)
1219                 goto err;
1220
1221         rt->path_rec[0] = *ib_event->param.req_rcvd.primary_path;
1222         if (rt->num_paths == 2)
1223                 rt->path_rec[1] = *ib_event->param.req_rcvd.alternate_path;
1224
1225         if (cma_any_addr(cma_src_addr(id_priv))) {
1226                 rt->addr.dev_addr.dev_type = ARPHRD_INFINIBAND;
1227                 rdma_addr_set_sgid(&rt->addr.dev_addr, &rt->path_rec[0].sgid);
1228                 ib_addr_set_pkey(&rt->addr.dev_addr, be16_to_cpu(rt->path_rec[0].pkey));
1229         } else {
1230                 ret = cma_translate_addr(cma_src_addr(id_priv), &rt->addr.dev_addr);
1231                 if (ret)
1232                         goto err;
1233         }
1234         rdma_addr_set_dgid(&rt->addr.dev_addr, &rt->path_rec[0].dgid);
1235
1236         id_priv->state = RDMA_CM_CONNECT;
1237         return id_priv;
1238
1239 err:
1240         rdma_destroy_id(id);
1241         return NULL;
1242 }
1243
1244 static struct rdma_id_private *cma_new_udp_id(struct rdma_cm_id *listen_id,
1245                                               struct ib_cm_event *ib_event)
1246 {
1247         struct rdma_id_private *id_priv;
1248         struct rdma_cm_id *id;
1249         union cma_ip_addr *src, *dst;
1250         __be16 port;
1251         u8 ip_ver;
1252         int ret;
1253
1254         id = rdma_create_id(listen_id->event_handler, listen_id->context,
1255                             listen_id->ps, IB_QPT_UD);
1256         if (IS_ERR(id))
1257                 return NULL;
1258
1259         id_priv = container_of(id, struct rdma_id_private, id);
1260         if (cma_get_net_info(ib_event->private_data, listen_id->ps,
1261                              &ip_ver, &port, &src, &dst))
1262                 goto err;
1263
1264         cma_save_net_info(&id->route.addr, &listen_id->route.addr,
1265                           ip_ver, port, src, dst);
1266
1267         if (!cma_any_addr((struct sockaddr *) &id->route.addr.src_addr)) {
1268                 ret = cma_translate_addr(cma_src_addr(id_priv), &id->route.addr.dev_addr);
1269                 if (ret)
1270                         goto err;
1271         }
1272
1273         id_priv->state = RDMA_CM_CONNECT;
1274         return id_priv;
1275 err:
1276         rdma_destroy_id(id);
1277         return NULL;
1278 }
1279
1280 static void cma_set_req_event_data(struct rdma_cm_event *event,
1281                                    struct ib_cm_req_event_param *req_data,
1282                                    void *private_data, int offset)
1283 {
1284         event->param.conn.private_data = private_data + offset;
1285         event->param.conn.private_data_len = IB_CM_REQ_PRIVATE_DATA_SIZE - offset;
1286         event->param.conn.responder_resources = req_data->responder_resources;
1287         event->param.conn.initiator_depth = req_data->initiator_depth;
1288         event->param.conn.flow_control = req_data->flow_control;
1289         event->param.conn.retry_count = req_data->retry_count;
1290         event->param.conn.rnr_retry_count = req_data->rnr_retry_count;
1291         event->param.conn.srq = req_data->srq;
1292         event->param.conn.qp_num = req_data->remote_qpn;
1293 }
1294
1295 static int cma_check_req_qp_type(struct rdma_cm_id *id, struct ib_cm_event *ib_event)
1296 {
1297         return (((ib_event->event == IB_CM_REQ_RECEIVED) &&
1298                  (ib_event->param.req_rcvd.qp_type == id->qp_type)) ||
1299                 ((ib_event->event == IB_CM_SIDR_REQ_RECEIVED) &&
1300                  (id->qp_type == IB_QPT_UD)) ||
1301                 (!id->qp_type));
1302 }
1303
1304 static int cma_req_handler(struct ib_cm_id *cm_id, struct ib_cm_event *ib_event)
1305 {
1306         struct rdma_id_private *listen_id, *conn_id;
1307         struct rdma_cm_event event;
1308         int offset, ret;
1309
1310         listen_id = cm_id->context;
1311         if (!cma_check_req_qp_type(&listen_id->id, ib_event))
1312                 return -EINVAL;
1313
1314         if (cma_disable_callback(listen_id, RDMA_CM_LISTEN))
1315                 return -ECONNABORTED;
1316
1317         memset(&event, 0, sizeof event);
1318         offset = cma_user_data_offset(listen_id->id.ps);
1319         event.event = RDMA_CM_EVENT_CONNECT_REQUEST;
1320         if (ib_event->event == IB_CM_SIDR_REQ_RECEIVED) {
1321                 conn_id = cma_new_udp_id(&listen_id->id, ib_event);
1322                 event.param.ud.private_data = ib_event->private_data + offset;
1323                 event.param.ud.private_data_len =
1324                                 IB_CM_SIDR_REQ_PRIVATE_DATA_SIZE - offset;
1325         } else {
1326                 conn_id = cma_new_conn_id(&listen_id->id, ib_event);
1327                 cma_set_req_event_data(&event, &ib_event->param.req_rcvd,
1328                                        ib_event->private_data, offset);
1329         }
1330         if (!conn_id) {
1331                 ret = -ENOMEM;
1332                 goto err1;
1333         }
1334
1335         mutex_lock_nested(&conn_id->handler_mutex, SINGLE_DEPTH_NESTING);
1336         ret = cma_acquire_dev(conn_id);
1337         if (ret)
1338                 goto err2;
1339
1340         conn_id->cm_id.ib = cm_id;
1341         cm_id->context = conn_id;
1342         cm_id->cm_handler = cma_ib_handler;
1343
1344         /*
1345          * Protect against the user destroying conn_id from another thread
1346          * until we're done accessing it.
1347          */
1348         atomic_inc(&conn_id->refcount);
1349         ret = conn_id->id.event_handler(&conn_id->id, &event);
1350         if (ret)
1351                 goto err3;
1352
1353         /*
1354          * Acquire mutex to prevent user executing rdma_destroy_id()
1355          * while we're accessing the cm_id.
1356          */
1357         mutex_lock(&lock);
1358         if (cma_comp(conn_id, RDMA_CM_CONNECT) && (conn_id->id.qp_type != IB_QPT_UD))
1359                 ib_send_cm_mra(cm_id, CMA_CM_MRA_SETTING, NULL, 0);
1360         mutex_unlock(&lock);
1361         mutex_unlock(&conn_id->handler_mutex);
1362         mutex_unlock(&listen_id->handler_mutex);
1363         cma_deref_id(conn_id);
1364         return 0;
1365
1366 err3:
1367         cma_deref_id(conn_id);
1368         /* Destroy the CM ID by returning a non-zero value. */
1369         conn_id->cm_id.ib = NULL;
1370 err2:
1371         cma_exch(conn_id, RDMA_CM_DESTROYING);
1372         mutex_unlock(&conn_id->handler_mutex);
1373 err1:
1374         mutex_unlock(&listen_id->handler_mutex);
1375         if (conn_id)
1376                 rdma_destroy_id(&conn_id->id);
1377         return ret;
1378 }
1379
1380 static __be64 cma_get_service_id(enum rdma_port_space ps, struct sockaddr *addr)
1381 {
1382         if (addr->sa_family == AF_IB)
1383                 return ((struct sockaddr_ib *) addr)->sib_sid;
1384
1385         return cpu_to_be64(((u64)ps << 16) + be16_to_cpu(cma_port(addr)));
1386 }
1387
1388 static void cma_set_compare_data(enum rdma_port_space ps, struct sockaddr *addr,
1389                                  struct ib_cm_compare_data *compare)
1390 {
1391         struct cma_hdr *cma_data, *cma_mask;
1392         struct sdp_hh *sdp_data, *sdp_mask;
1393         __be32 ip4_addr;
1394         struct in6_addr ip6_addr;
1395
1396         memset(compare, 0, sizeof *compare);
1397         cma_data = (void *) compare->data;
1398         cma_mask = (void *) compare->mask;
1399         sdp_data = (void *) compare->data;
1400         sdp_mask = (void *) compare->mask;
1401
1402         switch (addr->sa_family) {
1403         case AF_INET:
1404                 ip4_addr = ((struct sockaddr_in *) addr)->sin_addr.s_addr;
1405                 if (ps == RDMA_PS_SDP) {
1406                         sdp_set_ip_ver(sdp_data, 4);
1407                         sdp_set_ip_ver(sdp_mask, 0xF);
1408                         sdp_data->dst_addr.ip4.addr = ip4_addr;
1409                         sdp_mask->dst_addr.ip4.addr = htonl(~0);
1410                 } else {
1411                         cma_set_ip_ver(cma_data, 4);
1412                         cma_set_ip_ver(cma_mask, 0xF);
1413                         if (!cma_any_addr(addr)) {
1414                                 cma_data->dst_addr.ip4.addr = ip4_addr;
1415                                 cma_mask->dst_addr.ip4.addr = htonl(~0);
1416                         }
1417                 }
1418                 break;
1419         case AF_INET6:
1420                 ip6_addr = ((struct sockaddr_in6 *) addr)->sin6_addr;
1421                 if (ps == RDMA_PS_SDP) {
1422                         sdp_set_ip_ver(sdp_data, 6);
1423                         sdp_set_ip_ver(sdp_mask, 0xF);
1424                         sdp_data->dst_addr.ip6 = ip6_addr;
1425                         memset(&sdp_mask->dst_addr.ip6, 0xFF,
1426                                sizeof sdp_mask->dst_addr.ip6);
1427                 } else {
1428                         cma_set_ip_ver(cma_data, 6);
1429                         cma_set_ip_ver(cma_mask, 0xF);
1430                         if (!cma_any_addr(addr)) {
1431                                 cma_data->dst_addr.ip6 = ip6_addr;
1432                                 memset(&cma_mask->dst_addr.ip6, 0xFF,
1433                                        sizeof cma_mask->dst_addr.ip6);
1434                         }
1435                 }
1436                 break;
1437         default:
1438                 break;
1439         }
1440 }
1441
1442 static int cma_iw_handler(struct iw_cm_id *iw_id, struct iw_cm_event *iw_event)
1443 {
1444         struct rdma_id_private *id_priv = iw_id->context;
1445         struct rdma_cm_event event;
1446         struct sockaddr_in *sin;
1447         int ret = 0;
1448
1449         if (cma_disable_callback(id_priv, RDMA_CM_CONNECT))
1450                 return 0;
1451
1452         memset(&event, 0, sizeof event);
1453         switch (iw_event->event) {
1454         case IW_CM_EVENT_CLOSE:
1455                 event.event = RDMA_CM_EVENT_DISCONNECTED;
1456                 break;
1457         case IW_CM_EVENT_CONNECT_REPLY:
1458                 sin = (struct sockaddr_in *) cma_src_addr(id_priv);
1459                 *sin = iw_event->local_addr;
1460                 sin = (struct sockaddr_in *) cma_dst_addr(id_priv);
1461                 *sin = iw_event->remote_addr;
1462                 switch (iw_event->status) {
1463                 case 0:
1464                         event.event = RDMA_CM_EVENT_ESTABLISHED;
1465                         event.param.conn.initiator_depth = iw_event->ird;
1466                         event.param.conn.responder_resources = iw_event->ord;
1467                         break;
1468                 case -ECONNRESET:
1469                 case -ECONNREFUSED:
1470                         event.event = RDMA_CM_EVENT_REJECTED;
1471                         break;
1472                 case -ETIMEDOUT:
1473                         event.event = RDMA_CM_EVENT_UNREACHABLE;
1474                         break;
1475                 default:
1476                         event.event = RDMA_CM_EVENT_CONNECT_ERROR;
1477                         break;
1478                 }
1479                 break;
1480         case IW_CM_EVENT_ESTABLISHED:
1481                 event.event = RDMA_CM_EVENT_ESTABLISHED;
1482                 event.param.conn.initiator_depth = iw_event->ird;
1483                 event.param.conn.responder_resources = iw_event->ord;
1484                 break;
1485         default:
1486                 BUG_ON(1);
1487         }
1488
1489         event.status = iw_event->status;
1490         event.param.conn.private_data = iw_event->private_data;
1491         event.param.conn.private_data_len = iw_event->private_data_len;
1492         ret = id_priv->id.event_handler(&id_priv->id, &event);
1493         if (ret) {
1494                 /* Destroy the CM ID by returning a non-zero value. */
1495                 id_priv->cm_id.iw = NULL;
1496                 cma_exch(id_priv, RDMA_CM_DESTROYING);
1497                 mutex_unlock(&id_priv->handler_mutex);
1498                 rdma_destroy_id(&id_priv->id);
1499                 return ret;
1500         }
1501
1502         mutex_unlock(&id_priv->handler_mutex);
1503         return ret;
1504 }
1505
1506 static int iw_conn_req_handler(struct iw_cm_id *cm_id,
1507                                struct iw_cm_event *iw_event)
1508 {
1509         struct rdma_cm_id *new_cm_id;
1510         struct rdma_id_private *listen_id, *conn_id;
1511         struct sockaddr_in *sin;
1512         struct net_device *dev = NULL;
1513         struct rdma_cm_event event;
1514         int ret;
1515         struct ib_device_attr attr;
1516
1517         listen_id = cm_id->context;
1518         if (cma_disable_callback(listen_id, RDMA_CM_LISTEN))
1519                 return -ECONNABORTED;
1520
1521         /* Create a new RDMA id for the new IW CM ID */
1522         new_cm_id = rdma_create_id(listen_id->id.event_handler,
1523                                    listen_id->id.context,
1524                                    RDMA_PS_TCP, IB_QPT_RC);
1525         if (IS_ERR(new_cm_id)) {
1526                 ret = -ENOMEM;
1527                 goto out;
1528         }
1529         conn_id = container_of(new_cm_id, struct rdma_id_private, id);
1530         mutex_lock_nested(&conn_id->handler_mutex, SINGLE_DEPTH_NESTING);
1531         conn_id->state = RDMA_CM_CONNECT;
1532
1533         dev = ip_dev_find(&init_net, iw_event->local_addr.sin_addr.s_addr);
1534         if (!dev) {
1535                 ret = -EADDRNOTAVAIL;
1536                 mutex_unlock(&conn_id->handler_mutex);
1537                 rdma_destroy_id(new_cm_id);
1538                 goto out;
1539         }
1540         ret = rdma_copy_addr(&conn_id->id.route.addr.dev_addr, dev, NULL);
1541         if (ret) {
1542                 mutex_unlock(&conn_id->handler_mutex);
1543                 rdma_destroy_id(new_cm_id);
1544                 goto out;
1545         }
1546
1547         ret = cma_acquire_dev(conn_id);
1548         if (ret) {
1549                 mutex_unlock(&conn_id->handler_mutex);
1550                 rdma_destroy_id(new_cm_id);
1551                 goto out;
1552         }
1553
1554         conn_id->cm_id.iw = cm_id;
1555         cm_id->context = conn_id;
1556         cm_id->cm_handler = cma_iw_handler;
1557
1558         sin = (struct sockaddr_in *) cma_src_addr(conn_id);
1559         *sin = iw_event->local_addr;
1560         sin = (struct sockaddr_in *) cma_dst_addr(conn_id);
1561         *sin = iw_event->remote_addr;
1562
1563         ret = ib_query_device(conn_id->id.device, &attr);
1564         if (ret) {
1565                 mutex_unlock(&conn_id->handler_mutex);
1566                 rdma_destroy_id(new_cm_id);
1567                 goto out;
1568         }
1569
1570         memset(&event, 0, sizeof event);
1571         event.event = RDMA_CM_EVENT_CONNECT_REQUEST;
1572         event.param.conn.private_data = iw_event->private_data;
1573         event.param.conn.private_data_len = iw_event->private_data_len;
1574         event.param.conn.initiator_depth = iw_event->ird;
1575         event.param.conn.responder_resources = iw_event->ord;
1576
1577         /*
1578          * Protect against the user destroying conn_id from another thread
1579          * until we're done accessing it.
1580          */
1581         atomic_inc(&conn_id->refcount);
1582         ret = conn_id->id.event_handler(&conn_id->id, &event);
1583         if (ret) {
1584                 /* User wants to destroy the CM ID */
1585                 conn_id->cm_id.iw = NULL;
1586                 cma_exch(conn_id, RDMA_CM_DESTROYING);
1587                 mutex_unlock(&conn_id->handler_mutex);
1588                 cma_deref_id(conn_id);
1589                 rdma_destroy_id(&conn_id->id);
1590                 goto out;
1591         }
1592
1593         mutex_unlock(&conn_id->handler_mutex);
1594         cma_deref_id(conn_id);
1595
1596 out:
1597         if (dev)
1598                 dev_put(dev);
1599         mutex_unlock(&listen_id->handler_mutex);
1600         return ret;
1601 }
1602
1603 static int cma_ib_listen(struct rdma_id_private *id_priv)
1604 {
1605         struct ib_cm_compare_data compare_data;
1606         struct sockaddr *addr;
1607         struct ib_cm_id *id;
1608         __be64 svc_id;
1609         int ret;
1610
1611         id = ib_create_cm_id(id_priv->id.device, cma_req_handler, id_priv);
1612         if (IS_ERR(id))
1613                 return PTR_ERR(id);
1614
1615         id_priv->cm_id.ib = id;
1616
1617         addr = cma_src_addr(id_priv);
1618         svc_id = cma_get_service_id(id_priv->id.ps, addr);
1619         if (cma_any_addr(addr) && !id_priv->afonly)
1620                 ret = ib_cm_listen(id_priv->cm_id.ib, svc_id, 0, NULL);
1621         else {
1622                 cma_set_compare_data(id_priv->id.ps, addr, &compare_data);
1623                 ret = ib_cm_listen(id_priv->cm_id.ib, svc_id, 0, &compare_data);
1624         }
1625
1626         if (ret) {
1627                 ib_destroy_cm_id(id_priv->cm_id.ib);
1628                 id_priv->cm_id.ib = NULL;
1629         }
1630
1631         return ret;
1632 }
1633
1634 static int cma_iw_listen(struct rdma_id_private *id_priv, int backlog)
1635 {
1636         int ret;
1637         struct sockaddr_in *sin;
1638         struct iw_cm_id *id;
1639
1640         id = iw_create_cm_id(id_priv->id.device,
1641                              iw_conn_req_handler,
1642                              id_priv);
1643         if (IS_ERR(id))
1644                 return PTR_ERR(id);
1645
1646         id_priv->cm_id.iw = id;
1647
1648         sin = (struct sockaddr_in *) cma_src_addr(id_priv);
1649         id_priv->cm_id.iw->local_addr = *sin;
1650
1651         ret = iw_cm_listen(id_priv->cm_id.iw, backlog);
1652
1653         if (ret) {
1654                 iw_destroy_cm_id(id_priv->cm_id.iw);
1655                 id_priv->cm_id.iw = NULL;
1656         }
1657
1658         return ret;
1659 }
1660
1661 static int cma_listen_handler(struct rdma_cm_id *id,
1662                               struct rdma_cm_event *event)
1663 {
1664         struct rdma_id_private *id_priv = id->context;
1665
1666         id->context = id_priv->id.context;
1667         id->event_handler = id_priv->id.event_handler;
1668         return id_priv->id.event_handler(id, event);
1669 }
1670
1671 static void cma_listen_on_dev(struct rdma_id_private *id_priv,
1672                               struct cma_device *cma_dev)
1673 {
1674         struct rdma_id_private *dev_id_priv;
1675         struct rdma_cm_id *id;
1676         int ret;
1677
1678         id = rdma_create_id(cma_listen_handler, id_priv, id_priv->id.ps,
1679                             id_priv->id.qp_type);
1680         if (IS_ERR(id))
1681                 return;
1682
1683         dev_id_priv = container_of(id, struct rdma_id_private, id);
1684
1685         dev_id_priv->state = RDMA_CM_ADDR_BOUND;
1686         memcpy(cma_src_addr(dev_id_priv), cma_src_addr(id_priv),
1687                rdma_addr_size(cma_src_addr(id_priv)));
1688
1689         cma_attach_to_dev(dev_id_priv, cma_dev);
1690         list_add_tail(&dev_id_priv->listen_list, &id_priv->listen_list);
1691         atomic_inc(&id_priv->refcount);
1692         dev_id_priv->internal_id = 1;
1693         dev_id_priv->afonly = id_priv->afonly;
1694
1695         ret = rdma_listen(id, id_priv->backlog);
1696         if (ret)
1697                 printk(KERN_WARNING "RDMA CMA: cma_listen_on_dev, error %d, "
1698                        "listening on device %s\n", ret, cma_dev->device->name);
1699 }
1700
1701 static void cma_listen_on_all(struct rdma_id_private *id_priv)
1702 {
1703         struct cma_device *cma_dev;
1704
1705         mutex_lock(&lock);
1706         list_add_tail(&id_priv->list, &listen_any_list);
1707         list_for_each_entry(cma_dev, &dev_list, list)
1708                 cma_listen_on_dev(id_priv, cma_dev);
1709         mutex_unlock(&lock);
1710 }
1711
1712 void rdma_set_service_type(struct rdma_cm_id *id, int tos)
1713 {
1714         struct rdma_id_private *id_priv;
1715
1716         id_priv = container_of(id, struct rdma_id_private, id);
1717         id_priv->tos = (u8) tos;
1718 }
1719 EXPORT_SYMBOL(rdma_set_service_type);
1720
1721 static void cma_query_handler(int status, struct ib_sa_path_rec *path_rec,
1722                               void *context)
1723 {
1724         struct cma_work *work = context;
1725         struct rdma_route *route;
1726
1727         route = &work->id->id.route;
1728
1729         if (!status) {
1730                 route->num_paths = 1;
1731                 *route->path_rec = *path_rec;
1732         } else {
1733                 work->old_state = RDMA_CM_ROUTE_QUERY;
1734                 work->new_state = RDMA_CM_ADDR_RESOLVED;
1735                 work->event.event = RDMA_CM_EVENT_ROUTE_ERROR;
1736                 work->event.status = status;
1737         }
1738
1739         queue_work(cma_wq, &work->work);
1740 }
1741
1742 static int cma_query_ib_route(struct rdma_id_private *id_priv, int timeout_ms,
1743                               struct cma_work *work)
1744 {
1745         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
1746         struct ib_sa_path_rec path_rec;
1747         ib_sa_comp_mask comp_mask;
1748         struct sockaddr_in6 *sin6;
1749         struct sockaddr_ib *sib;
1750
1751         memset(&path_rec, 0, sizeof path_rec);
1752         rdma_addr_get_sgid(dev_addr, &path_rec.sgid);
1753         rdma_addr_get_dgid(dev_addr, &path_rec.dgid);
1754         path_rec.pkey = cpu_to_be16(ib_addr_get_pkey(dev_addr));
1755         path_rec.numb_path = 1;
1756         path_rec.reversible = 1;
1757         path_rec.service_id = cma_get_service_id(id_priv->id.ps, cma_dst_addr(id_priv));
1758
1759         comp_mask = IB_SA_PATH_REC_DGID | IB_SA_PATH_REC_SGID |
1760                     IB_SA_PATH_REC_PKEY | IB_SA_PATH_REC_NUMB_PATH |
1761                     IB_SA_PATH_REC_REVERSIBLE | IB_SA_PATH_REC_SERVICE_ID;
1762
1763         switch (cma_family(id_priv)) {
1764         case AF_INET:
1765                 path_rec.qos_class = cpu_to_be16((u16) id_priv->tos);
1766                 comp_mask |= IB_SA_PATH_REC_QOS_CLASS;
1767                 break;
1768         case AF_INET6:
1769                 sin6 = (struct sockaddr_in6 *) cma_src_addr(id_priv);
1770                 path_rec.traffic_class = (u8) (be32_to_cpu(sin6->sin6_flowinfo) >> 20);
1771                 comp_mask |= IB_SA_PATH_REC_TRAFFIC_CLASS;
1772                 break;
1773         case AF_IB:
1774                 sib = (struct sockaddr_ib *) cma_src_addr(id_priv);
1775                 path_rec.traffic_class = (u8) (be32_to_cpu(sib->sib_flowinfo) >> 20);
1776                 comp_mask |= IB_SA_PATH_REC_TRAFFIC_CLASS;
1777                 break;
1778         }
1779
1780         id_priv->query_id = ib_sa_path_rec_get(&sa_client, id_priv->id.device,
1781                                                id_priv->id.port_num, &path_rec,
1782                                                comp_mask, timeout_ms,
1783                                                GFP_KERNEL, cma_query_handler,
1784                                                work, &id_priv->query);
1785
1786         return (id_priv->query_id < 0) ? id_priv->query_id : 0;
1787 }
1788
1789 static void cma_work_handler(struct work_struct *_work)
1790 {
1791         struct cma_work *work = container_of(_work, struct cma_work, work);
1792         struct rdma_id_private *id_priv = work->id;
1793         int destroy = 0;
1794
1795         mutex_lock(&id_priv->handler_mutex);
1796         if (!cma_comp_exch(id_priv, work->old_state, work->new_state))
1797                 goto out;
1798
1799         if (id_priv->id.event_handler(&id_priv->id, &work->event)) {
1800                 cma_exch(id_priv, RDMA_CM_DESTROYING);
1801                 destroy = 1;
1802         }
1803 out:
1804         mutex_unlock(&id_priv->handler_mutex);
1805         cma_deref_id(id_priv);
1806         if (destroy)
1807                 rdma_destroy_id(&id_priv->id);
1808         kfree(work);
1809 }
1810
1811 static void cma_ndev_work_handler(struct work_struct *_work)
1812 {
1813         struct cma_ndev_work *work = container_of(_work, struct cma_ndev_work, work);
1814         struct rdma_id_private *id_priv = work->id;
1815         int destroy = 0;
1816
1817         mutex_lock(&id_priv->handler_mutex);
1818         if (id_priv->state == RDMA_CM_DESTROYING ||
1819             id_priv->state == RDMA_CM_DEVICE_REMOVAL)
1820                 goto out;
1821
1822         if (id_priv->id.event_handler(&id_priv->id, &work->event)) {
1823                 cma_exch(id_priv, RDMA_CM_DESTROYING);
1824                 destroy = 1;
1825         }
1826
1827 out:
1828         mutex_unlock(&id_priv->handler_mutex);
1829         cma_deref_id(id_priv);
1830         if (destroy)
1831                 rdma_destroy_id(&id_priv->id);
1832         kfree(work);
1833 }
1834
1835 static int cma_resolve_ib_route(struct rdma_id_private *id_priv, int timeout_ms)
1836 {
1837         struct rdma_route *route = &id_priv->id.route;
1838         struct cma_work *work;
1839         int ret;
1840
1841         work = kzalloc(sizeof *work, GFP_KERNEL);
1842         if (!work)
1843                 return -ENOMEM;
1844
1845         work->id = id_priv;
1846         INIT_WORK(&work->work, cma_work_handler);
1847         work->old_state = RDMA_CM_ROUTE_QUERY;
1848         work->new_state = RDMA_CM_ROUTE_RESOLVED;
1849         work->event.event = RDMA_CM_EVENT_ROUTE_RESOLVED;
1850
1851         route->path_rec = kmalloc(sizeof *route->path_rec, GFP_KERNEL);
1852         if (!route->path_rec) {
1853                 ret = -ENOMEM;
1854                 goto err1;
1855         }
1856
1857         ret = cma_query_ib_route(id_priv, timeout_ms, work);
1858         if (ret)
1859                 goto err2;
1860
1861         return 0;
1862 err2:
1863         kfree(route->path_rec);
1864         route->path_rec = NULL;
1865 err1:
1866         kfree(work);
1867         return ret;
1868 }
1869
1870 int rdma_set_ib_paths(struct rdma_cm_id *id,
1871                       struct ib_sa_path_rec *path_rec, int num_paths)
1872 {
1873         struct rdma_id_private *id_priv;
1874         int ret;
1875
1876         id_priv = container_of(id, struct rdma_id_private, id);
1877         if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_RESOLVED,
1878                            RDMA_CM_ROUTE_RESOLVED))
1879                 return -EINVAL;
1880
1881         id->route.path_rec = kmemdup(path_rec, sizeof *path_rec * num_paths,
1882                                      GFP_KERNEL);
1883         if (!id->route.path_rec) {
1884                 ret = -ENOMEM;
1885                 goto err;
1886         }
1887
1888         id->route.num_paths = num_paths;
1889         return 0;
1890 err:
1891         cma_comp_exch(id_priv, RDMA_CM_ROUTE_RESOLVED, RDMA_CM_ADDR_RESOLVED);
1892         return ret;
1893 }
1894 EXPORT_SYMBOL(rdma_set_ib_paths);
1895
1896 static int cma_resolve_iw_route(struct rdma_id_private *id_priv, int timeout_ms)
1897 {
1898         struct cma_work *work;
1899
1900         work = kzalloc(sizeof *work, GFP_KERNEL);
1901         if (!work)
1902                 return -ENOMEM;
1903
1904         work->id = id_priv;
1905         INIT_WORK(&work->work, cma_work_handler);
1906         work->old_state = RDMA_CM_ROUTE_QUERY;
1907         work->new_state = RDMA_CM_ROUTE_RESOLVED;
1908         work->event.event = RDMA_CM_EVENT_ROUTE_RESOLVED;
1909         queue_work(cma_wq, &work->work);
1910         return 0;
1911 }
1912
1913 static int cma_resolve_iboe_route(struct rdma_id_private *id_priv)
1914 {
1915         struct rdma_route *route = &id_priv->id.route;
1916         struct rdma_addr *addr = &route->addr;
1917         struct cma_work *work;
1918         int ret;
1919         struct net_device *ndev = NULL;
1920         u16 vid;
1921
1922         work = kzalloc(sizeof *work, GFP_KERNEL);
1923         if (!work)
1924                 return -ENOMEM;
1925
1926         work->id = id_priv;
1927         INIT_WORK(&work->work, cma_work_handler);
1928
1929         route->path_rec = kzalloc(sizeof *route->path_rec, GFP_KERNEL);
1930         if (!route->path_rec) {
1931                 ret = -ENOMEM;
1932                 goto err1;
1933         }
1934
1935         route->num_paths = 1;
1936
1937         if (addr->dev_addr.bound_dev_if)
1938                 ndev = dev_get_by_index(&init_net, addr->dev_addr.bound_dev_if);
1939         if (!ndev) {
1940                 ret = -ENODEV;
1941                 goto err2;
1942         }
1943
1944         vid = rdma_vlan_dev_vlan_id(ndev);
1945
1946         iboe_mac_vlan_to_ll(&route->path_rec->sgid, addr->dev_addr.src_dev_addr, vid);
1947         iboe_mac_vlan_to_ll(&route->path_rec->dgid, addr->dev_addr.dst_dev_addr, vid);
1948
1949         route->path_rec->hop_limit = 1;
1950         route->path_rec->reversible = 1;
1951         route->path_rec->pkey = cpu_to_be16(0xffff);
1952         route->path_rec->mtu_selector = IB_SA_EQ;
1953         route->path_rec->sl = netdev_get_prio_tc_map(
1954                         ndev->priv_flags & IFF_802_1Q_VLAN ?
1955                                 vlan_dev_real_dev(ndev) : ndev,
1956                         rt_tos2priority(id_priv->tos));
1957
1958         route->path_rec->mtu = iboe_get_mtu(ndev->mtu);
1959         route->path_rec->rate_selector = IB_SA_EQ;
1960         route->path_rec->rate = iboe_get_rate(ndev);
1961         dev_put(ndev);
1962         route->path_rec->packet_life_time_selector = IB_SA_EQ;
1963         route->path_rec->packet_life_time = CMA_IBOE_PACKET_LIFETIME;
1964         if (!route->path_rec->mtu) {
1965                 ret = -EINVAL;
1966                 goto err2;
1967         }
1968
1969         work->old_state = RDMA_CM_ROUTE_QUERY;
1970         work->new_state = RDMA_CM_ROUTE_RESOLVED;
1971         work->event.event = RDMA_CM_EVENT_ROUTE_RESOLVED;
1972         work->event.status = 0;
1973
1974         queue_work(cma_wq, &work->work);
1975
1976         return 0;
1977
1978 err2:
1979         kfree(route->path_rec);
1980         route->path_rec = NULL;
1981 err1:
1982         kfree(work);
1983         return ret;
1984 }
1985
1986 int rdma_resolve_route(struct rdma_cm_id *id, int timeout_ms)
1987 {
1988         struct rdma_id_private *id_priv;
1989         int ret;
1990
1991         id_priv = container_of(id, struct rdma_id_private, id);
1992         if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_RESOLVED, RDMA_CM_ROUTE_QUERY))
1993                 return -EINVAL;
1994
1995         atomic_inc(&id_priv->refcount);
1996         switch (rdma_node_get_transport(id->device->node_type)) {
1997         case RDMA_TRANSPORT_IB:
1998                 switch (rdma_port_get_link_layer(id->device, id->port_num)) {
1999                 case IB_LINK_LAYER_INFINIBAND:
2000                         ret = cma_resolve_ib_route(id_priv, timeout_ms);
2001                         break;
2002                 case IB_LINK_LAYER_ETHERNET:
2003                         ret = cma_resolve_iboe_route(id_priv);
2004                         break;
2005                 default:
2006                         ret = -ENOSYS;
2007                 }
2008                 break;
2009         case RDMA_TRANSPORT_IWARP:
2010                 ret = cma_resolve_iw_route(id_priv, timeout_ms);
2011                 break;
2012         default:
2013                 ret = -ENOSYS;
2014                 break;
2015         }
2016         if (ret)
2017                 goto err;
2018
2019         return 0;
2020 err:
2021         cma_comp_exch(id_priv, RDMA_CM_ROUTE_QUERY, RDMA_CM_ADDR_RESOLVED);
2022         cma_deref_id(id_priv);
2023         return ret;
2024 }
2025 EXPORT_SYMBOL(rdma_resolve_route);
2026
2027 static void cma_set_loopback(struct sockaddr *addr)
2028 {
2029         switch (addr->sa_family) {
2030         case AF_INET:
2031                 ((struct sockaddr_in *) addr)->sin_addr.s_addr = htonl(INADDR_LOOPBACK);
2032                 break;
2033         case AF_INET6:
2034                 ipv6_addr_set(&((struct sockaddr_in6 *) addr)->sin6_addr,
2035                               0, 0, 0, htonl(1));
2036                 break;
2037         default:
2038                 ib_addr_set(&((struct sockaddr_ib *) addr)->sib_addr,
2039                             0, 0, 0, htonl(1));
2040                 break;
2041         }
2042 }
2043
2044 static int cma_bind_loopback(struct rdma_id_private *id_priv)
2045 {
2046         struct cma_device *cma_dev, *cur_dev;
2047         struct ib_port_attr port_attr;
2048         union ib_gid gid;
2049         u16 pkey;
2050         int ret;
2051         u8 p;
2052
2053         cma_dev = NULL;
2054         mutex_lock(&lock);
2055         list_for_each_entry(cur_dev, &dev_list, list) {
2056                 if (cma_family(id_priv) == AF_IB &&
2057                     rdma_node_get_transport(cur_dev->device->node_type) != RDMA_TRANSPORT_IB)
2058                         continue;
2059
2060                 if (!cma_dev)
2061                         cma_dev = cur_dev;
2062
2063                 for (p = 1; p <= cur_dev->device->phys_port_cnt; ++p) {
2064                         if (!ib_query_port(cur_dev->device, p, &port_attr) &&
2065                             port_attr.state == IB_PORT_ACTIVE) {
2066                                 cma_dev = cur_dev;
2067                                 goto port_found;
2068                         }
2069                 }
2070         }
2071
2072         if (!cma_dev) {
2073                 ret = -ENODEV;
2074                 goto out;
2075         }
2076
2077         p = 1;
2078
2079 port_found:
2080         ret = ib_get_cached_gid(cma_dev->device, p, 0, &gid);
2081         if (ret)
2082                 goto out;
2083
2084         ret = ib_get_cached_pkey(cma_dev->device, p, 0, &pkey);
2085         if (ret)
2086                 goto out;
2087
2088         id_priv->id.route.addr.dev_addr.dev_type =
2089                 (rdma_port_get_link_layer(cma_dev->device, p) == IB_LINK_LAYER_INFINIBAND) ?
2090                 ARPHRD_INFINIBAND : ARPHRD_ETHER;
2091
2092         rdma_addr_set_sgid(&id_priv->id.route.addr.dev_addr, &gid);
2093         ib_addr_set_pkey(&id_priv->id.route.addr.dev_addr, pkey);
2094         id_priv->id.port_num = p;
2095         cma_attach_to_dev(id_priv, cma_dev);
2096         cma_set_loopback(cma_src_addr(id_priv));
2097 out:
2098         mutex_unlock(&lock);
2099         return ret;
2100 }
2101
2102 static void addr_handler(int status, struct sockaddr *src_addr,
2103                          struct rdma_dev_addr *dev_addr, void *context)
2104 {
2105         struct rdma_id_private *id_priv = context;
2106         struct rdma_cm_event event;
2107
2108         memset(&event, 0, sizeof event);
2109         mutex_lock(&id_priv->handler_mutex);
2110         if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_QUERY,
2111                            RDMA_CM_ADDR_RESOLVED))
2112                 goto out;
2113
2114         if (!status && !id_priv->cma_dev)
2115                 status = cma_acquire_dev(id_priv);
2116
2117         if (status) {
2118                 if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_RESOLVED,
2119                                    RDMA_CM_ADDR_BOUND))
2120                         goto out;
2121                 event.event = RDMA_CM_EVENT_ADDR_ERROR;
2122                 event.status = status;
2123         } else {
2124                 memcpy(cma_src_addr(id_priv), src_addr, rdma_addr_size(src_addr));
2125                 event.event = RDMA_CM_EVENT_ADDR_RESOLVED;
2126         }
2127
2128         if (id_priv->id.event_handler(&id_priv->id, &event)) {
2129                 cma_exch(id_priv, RDMA_CM_DESTROYING);
2130                 mutex_unlock(&id_priv->handler_mutex);
2131                 cma_deref_id(id_priv);
2132                 rdma_destroy_id(&id_priv->id);
2133                 return;
2134         }
2135 out:
2136         mutex_unlock(&id_priv->handler_mutex);
2137         cma_deref_id(id_priv);
2138 }
2139
2140 static int cma_resolve_loopback(struct rdma_id_private *id_priv)
2141 {
2142         struct cma_work *work;
2143         union ib_gid gid;
2144         int ret;
2145
2146         work = kzalloc(sizeof *work, GFP_KERNEL);
2147         if (!work)
2148                 return -ENOMEM;
2149
2150         if (!id_priv->cma_dev) {
2151                 ret = cma_bind_loopback(id_priv);
2152                 if (ret)
2153                         goto err;
2154         }
2155
2156         rdma_addr_get_sgid(&id_priv->id.route.addr.dev_addr, &gid);
2157         rdma_addr_set_dgid(&id_priv->id.route.addr.dev_addr, &gid);
2158
2159         work->id = id_priv;
2160         INIT_WORK(&work->work, cma_work_handler);
2161         work->old_state = RDMA_CM_ADDR_QUERY;
2162         work->new_state = RDMA_CM_ADDR_RESOLVED;
2163         work->event.event = RDMA_CM_EVENT_ADDR_RESOLVED;
2164         queue_work(cma_wq, &work->work);
2165         return 0;
2166 err:
2167         kfree(work);
2168         return ret;
2169 }
2170
2171 static int cma_resolve_ib_addr(struct rdma_id_private *id_priv)
2172 {
2173         struct cma_work *work;
2174         int ret;
2175
2176         work = kzalloc(sizeof *work, GFP_KERNEL);
2177         if (!work)
2178                 return -ENOMEM;
2179
2180         if (!id_priv->cma_dev) {
2181                 ret = cma_resolve_ib_dev(id_priv);
2182                 if (ret)
2183                         goto err;
2184         }
2185
2186         rdma_addr_set_dgid(&id_priv->id.route.addr.dev_addr, (union ib_gid *)
2187                 &(((struct sockaddr_ib *) &id_priv->id.route.addr.dst_addr)->sib_addr));
2188
2189         work->id = id_priv;
2190         INIT_WORK(&work->work, cma_work_handler);
2191         work->old_state = RDMA_CM_ADDR_QUERY;
2192         work->new_state = RDMA_CM_ADDR_RESOLVED;
2193         work->event.event = RDMA_CM_EVENT_ADDR_RESOLVED;
2194         queue_work(cma_wq, &work->work);
2195         return 0;
2196 err:
2197         kfree(work);
2198         return ret;
2199 }
2200
2201 static int cma_bind_addr(struct rdma_cm_id *id, struct sockaddr *src_addr,
2202                          struct sockaddr *dst_addr)
2203 {
2204         if (!src_addr || !src_addr->sa_family) {
2205                 src_addr = (struct sockaddr *) &id->route.addr.src_addr;
2206                 src_addr->sa_family = dst_addr->sa_family;
2207                 if (dst_addr->sa_family == AF_INET6) {
2208                         ((struct sockaddr_in6 *) src_addr)->sin6_scope_id =
2209                                 ((struct sockaddr_in6 *) dst_addr)->sin6_scope_id;
2210                 } else if (dst_addr->sa_family == AF_IB) {
2211                         ((struct sockaddr_ib *) src_addr)->sib_pkey =
2212                                 ((struct sockaddr_ib *) dst_addr)->sib_pkey;
2213                 }
2214         }
2215         return rdma_bind_addr(id, src_addr);
2216 }
2217
2218 int rdma_resolve_addr(struct rdma_cm_id *id, struct sockaddr *src_addr,
2219                       struct sockaddr *dst_addr, int timeout_ms)
2220 {
2221         struct rdma_id_private *id_priv;
2222         int ret;
2223
2224         id_priv = container_of(id, struct rdma_id_private, id);
2225         if (id_priv->state == RDMA_CM_IDLE) {
2226                 ret = cma_bind_addr(id, src_addr, dst_addr);
2227                 if (ret)
2228                         return ret;
2229         }
2230
2231         if (cma_family(id_priv) != dst_addr->sa_family)
2232                 return -EINVAL;
2233
2234         if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_BOUND, RDMA_CM_ADDR_QUERY))
2235                 return -EINVAL;
2236
2237         atomic_inc(&id_priv->refcount);
2238         memcpy(cma_dst_addr(id_priv), dst_addr, rdma_addr_size(dst_addr));
2239         if (cma_any_addr(dst_addr)) {
2240                 ret = cma_resolve_loopback(id_priv);
2241         } else {
2242                 if (dst_addr->sa_family == AF_IB) {
2243                         ret = cma_resolve_ib_addr(id_priv);
2244                 } else {
2245                         ret = rdma_resolve_ip(&addr_client, cma_src_addr(id_priv),
2246                                               dst_addr, &id->route.addr.dev_addr,
2247                                               timeout_ms, addr_handler, id_priv);
2248                 }
2249         }
2250         if (ret)
2251                 goto err;
2252
2253         return 0;
2254 err:
2255         cma_comp_exch(id_priv, RDMA_CM_ADDR_QUERY, RDMA_CM_ADDR_BOUND);
2256         cma_deref_id(id_priv);
2257         return ret;
2258 }
2259 EXPORT_SYMBOL(rdma_resolve_addr);
2260
2261 int rdma_set_reuseaddr(struct rdma_cm_id *id, int reuse)
2262 {
2263         struct rdma_id_private *id_priv;
2264         unsigned long flags;
2265         int ret;
2266
2267         id_priv = container_of(id, struct rdma_id_private, id);
2268         spin_lock_irqsave(&id_priv->lock, flags);
2269         if (reuse || id_priv->state == RDMA_CM_IDLE) {
2270                 id_priv->reuseaddr = reuse;
2271                 ret = 0;
2272         } else {
2273                 ret = -EINVAL;
2274         }
2275         spin_unlock_irqrestore(&id_priv->lock, flags);
2276         return ret;
2277 }
2278 EXPORT_SYMBOL(rdma_set_reuseaddr);
2279
2280 int rdma_set_afonly(struct rdma_cm_id *id, int afonly)
2281 {
2282         struct rdma_id_private *id_priv;
2283         unsigned long flags;
2284         int ret;
2285
2286         id_priv = container_of(id, struct rdma_id_private, id);
2287         spin_lock_irqsave(&id_priv->lock, flags);
2288         if (id_priv->state == RDMA_CM_IDLE || id_priv->state == RDMA_CM_ADDR_BOUND) {
2289                 id_priv->options |= (1 << CMA_OPTION_AFONLY);
2290                 id_priv->afonly = afonly;
2291                 ret = 0;
2292         } else {
2293                 ret = -EINVAL;
2294         }
2295         spin_unlock_irqrestore(&id_priv->lock, flags);
2296         return ret;
2297 }
2298 EXPORT_SYMBOL(rdma_set_afonly);
2299
2300 static void cma_bind_port(struct rdma_bind_list *bind_list,
2301                           struct rdma_id_private *id_priv)
2302 {
2303         struct sockaddr *addr;
2304         struct sockaddr_ib *sib;
2305         u64 sid, mask;
2306         __be16 port;
2307
2308         addr = cma_src_addr(id_priv);
2309         port = htons(bind_list->port);
2310
2311         switch (addr->sa_family) {
2312         case AF_INET:
2313                 ((struct sockaddr_in *) addr)->sin_port = port;
2314                 break;
2315         case AF_INET6:
2316                 ((struct sockaddr_in6 *) addr)->sin6_port = port;
2317                 break;
2318         case AF_IB:
2319                 sib = (struct sockaddr_ib *) addr;
2320                 sid = be64_to_cpu(sib->sib_sid);
2321                 mask = be64_to_cpu(sib->sib_sid_mask);
2322                 sib->sib_sid = cpu_to_be64((sid & mask) | (u64) ntohs(port));
2323                 sib->sib_sid_mask = cpu_to_be64(~0ULL);
2324                 break;
2325         }
2326         id_priv->bind_list = bind_list;
2327         hlist_add_head(&id_priv->node, &bind_list->owners);
2328 }
2329
2330 static int cma_alloc_port(struct idr *ps, struct rdma_id_private *id_priv,
2331                           unsigned short snum)
2332 {
2333         struct rdma_bind_list *bind_list;
2334         int ret;
2335
2336         bind_list = kzalloc(sizeof *bind_list, GFP_KERNEL);
2337         if (!bind_list)
2338                 return -ENOMEM;
2339
2340         ret = idr_alloc(ps, bind_list, snum, snum + 1, GFP_KERNEL);
2341         if (ret < 0)
2342                 goto err;
2343
2344         bind_list->ps = ps;
2345         bind_list->port = (unsigned short)ret;
2346         cma_bind_port(bind_list, id_priv);
2347         return 0;
2348 err:
2349         kfree(bind_list);
2350         return ret == -ENOSPC ? -EADDRNOTAVAIL : ret;
2351 }
2352
2353 static int cma_alloc_any_port(struct idr *ps, struct rdma_id_private *id_priv)
2354 {
2355         static unsigned int last_used_port;
2356         int low, high, remaining;
2357         unsigned int rover;
2358
2359         inet_get_local_port_range(&low, &high);
2360         remaining = (high - low) + 1;
2361         rover = net_random() % remaining + low;
2362 retry:
2363         if (last_used_port != rover &&
2364             !idr_find(ps, (unsigned short) rover)) {
2365                 int ret = cma_alloc_port(ps, id_priv, rover);
2366                 /*
2367                  * Remember previously used port number in order to avoid
2368                  * re-using same port immediately after it is closed.
2369                  */
2370                 if (!ret)
2371                         last_used_port = rover;
2372                 if (ret != -EADDRNOTAVAIL)
2373                         return ret;
2374         }
2375         if (--remaining) {
2376                 rover++;
2377                 if ((rover < low) || (rover > high))
2378                         rover = low;
2379                 goto retry;
2380         }
2381         return -EADDRNOTAVAIL;
2382 }
2383
2384 /*
2385  * Check that the requested port is available.  This is called when trying to
2386  * bind to a specific port, or when trying to listen on a bound port.  In
2387  * the latter case, the provided id_priv may already be on the bind_list, but
2388  * we still need to check that it's okay to start listening.
2389  */
2390 static int cma_check_port(struct rdma_bind_list *bind_list,
2391                           struct rdma_id_private *id_priv, uint8_t reuseaddr)
2392 {
2393         struct rdma_id_private *cur_id;
2394         struct sockaddr *addr, *cur_addr;
2395
2396         addr = cma_src_addr(id_priv);
2397         hlist_for_each_entry(cur_id, &bind_list->owners, node) {
2398                 if (id_priv == cur_id)
2399                         continue;
2400
2401                 if ((cur_id->state != RDMA_CM_LISTEN) && reuseaddr &&
2402                     cur_id->reuseaddr)
2403                         continue;
2404
2405                 cur_addr = cma_src_addr(cur_id);
2406                 if (id_priv->afonly && cur_id->afonly &&
2407                     (addr->sa_family != cur_addr->sa_family))
2408                         continue;
2409
2410                 if (cma_any_addr(addr) || cma_any_addr(cur_addr))
2411                         return -EADDRNOTAVAIL;
2412
2413                 if (!cma_addr_cmp(addr, cur_addr))
2414                         return -EADDRINUSE;
2415         }
2416         return 0;
2417 }
2418
2419 static int cma_use_port(struct idr *ps, struct rdma_id_private *id_priv)
2420 {
2421         struct rdma_bind_list *bind_list;
2422         unsigned short snum;
2423         int ret;
2424
2425         snum = ntohs(cma_port(cma_src_addr(id_priv)));
2426         if (snum < PROT_SOCK && !capable(CAP_NET_BIND_SERVICE))
2427                 return -EACCES;
2428
2429         bind_list = idr_find(ps, snum);
2430         if (!bind_list) {
2431                 ret = cma_alloc_port(ps, id_priv, snum);
2432         } else {
2433                 ret = cma_check_port(bind_list, id_priv, id_priv->reuseaddr);
2434                 if (!ret)
2435                         cma_bind_port(bind_list, id_priv);
2436         }
2437         return ret;
2438 }
2439
2440 static int cma_bind_listen(struct rdma_id_private *id_priv)
2441 {
2442         struct rdma_bind_list *bind_list = id_priv->bind_list;
2443         int ret = 0;
2444
2445         mutex_lock(&lock);
2446         if (bind_list->owners.first->next)
2447                 ret = cma_check_port(bind_list, id_priv, 0);
2448         mutex_unlock(&lock);
2449         return ret;
2450 }
2451
2452 static struct idr *cma_select_inet_ps(struct rdma_id_private *id_priv)
2453 {
2454         switch (id_priv->id.ps) {
2455         case RDMA_PS_SDP:
2456                 return &sdp_ps;
2457         case RDMA_PS_TCP:
2458                 return &tcp_ps;
2459         case RDMA_PS_UDP:
2460                 return &udp_ps;
2461         case RDMA_PS_IPOIB:
2462                 return &ipoib_ps;
2463         case RDMA_PS_IB:
2464                 return &ib_ps;
2465         default:
2466                 return NULL;
2467         }
2468 }
2469
2470 static struct idr *cma_select_ib_ps(struct rdma_id_private *id_priv)
2471 {
2472         struct idr *ps = NULL;
2473         struct sockaddr_ib *sib;
2474         u64 sid_ps, mask, sid;
2475
2476         sib = (struct sockaddr_ib *) cma_src_addr(id_priv);
2477         mask = be64_to_cpu(sib->sib_sid_mask) & RDMA_IB_IP_PS_MASK;
2478         sid = be64_to_cpu(sib->sib_sid) & mask;
2479
2480         if ((id_priv->id.ps == RDMA_PS_IB) && (sid == (RDMA_IB_IP_PS_IB & mask))) {
2481                 sid_ps = RDMA_IB_IP_PS_IB;
2482                 ps = &ib_ps;
2483         } else if (((id_priv->id.ps == RDMA_PS_IB) || (id_priv->id.ps == RDMA_PS_TCP)) &&
2484                    (sid == (RDMA_IB_IP_PS_TCP & mask))) {
2485                 sid_ps = RDMA_IB_IP_PS_TCP;
2486                 ps = &tcp_ps;
2487         } else if (((id_priv->id.ps == RDMA_PS_IB) || (id_priv->id.ps == RDMA_PS_UDP)) &&
2488                    (sid == (RDMA_IB_IP_PS_UDP & mask))) {
2489                 sid_ps = RDMA_IB_IP_PS_UDP;
2490                 ps = &udp_ps;
2491         }
2492
2493         if (ps) {
2494                 sib->sib_sid = cpu_to_be64(sid_ps | ntohs(cma_port((struct sockaddr *) sib)));
2495                 sib->sib_sid_mask = cpu_to_be64(RDMA_IB_IP_PS_MASK |
2496                                                 be64_to_cpu(sib->sib_sid_mask));
2497         }
2498         return ps;
2499 }
2500
2501 static int cma_get_port(struct rdma_id_private *id_priv)
2502 {
2503         struct idr *ps;
2504         int ret;
2505
2506         if (cma_family(id_priv) != AF_IB)
2507                 ps = cma_select_inet_ps(id_priv);
2508         else
2509                 ps = cma_select_ib_ps(id_priv);
2510         if (!ps)
2511                 return -EPROTONOSUPPORT;
2512
2513         mutex_lock(&lock);
2514         if (cma_any_port(cma_src_addr(id_priv)))
2515                 ret = cma_alloc_any_port(ps, id_priv);
2516         else
2517                 ret = cma_use_port(ps, id_priv);
2518         mutex_unlock(&lock);
2519
2520         return ret;
2521 }
2522
2523 static int cma_check_linklocal(struct rdma_dev_addr *dev_addr,
2524                                struct sockaddr *addr)
2525 {
2526 #if IS_ENABLED(CONFIG_IPV6)
2527         struct sockaddr_in6 *sin6;
2528
2529         if (addr->sa_family != AF_INET6)
2530                 return 0;
2531
2532         sin6 = (struct sockaddr_in6 *) addr;
2533         if ((ipv6_addr_type(&sin6->sin6_addr) & IPV6_ADDR_LINKLOCAL) &&
2534             !sin6->sin6_scope_id)
2535                         return -EINVAL;
2536
2537         dev_addr->bound_dev_if = sin6->sin6_scope_id;
2538 #endif
2539         return 0;
2540 }
2541
2542 int rdma_listen(struct rdma_cm_id *id, int backlog)
2543 {
2544         struct rdma_id_private *id_priv;
2545         int ret;
2546
2547         id_priv = container_of(id, struct rdma_id_private, id);
2548         if (id_priv->state == RDMA_CM_IDLE) {
2549                 id->route.addr.src_addr.ss_family = AF_INET;
2550                 ret = rdma_bind_addr(id, cma_src_addr(id_priv));
2551                 if (ret)
2552                         return ret;
2553         }
2554
2555         if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_BOUND, RDMA_CM_LISTEN))
2556                 return -EINVAL;
2557
2558         if (id_priv->reuseaddr) {
2559                 ret = cma_bind_listen(id_priv);
2560                 if (ret)
2561                         goto err;
2562         }
2563
2564         id_priv->backlog = backlog;
2565         if (id->device) {
2566                 switch (rdma_node_get_transport(id->device->node_type)) {
2567                 case RDMA_TRANSPORT_IB:
2568                         ret = cma_ib_listen(id_priv);
2569                         if (ret)
2570                                 goto err;
2571                         break;
2572                 case RDMA_TRANSPORT_IWARP:
2573                         ret = cma_iw_listen(id_priv, backlog);
2574                         if (ret)
2575                                 goto err;
2576                         break;
2577                 default:
2578                         ret = -ENOSYS;
2579                         goto err;
2580                 }
2581         } else
2582                 cma_listen_on_all(id_priv);
2583
2584         return 0;
2585 err:
2586         id_priv->backlog = 0;
2587         cma_comp_exch(id_priv, RDMA_CM_LISTEN, RDMA_CM_ADDR_BOUND);
2588         return ret;
2589 }
2590 EXPORT_SYMBOL(rdma_listen);
2591
2592 int rdma_bind_addr(struct rdma_cm_id *id, struct sockaddr *addr)
2593 {
2594         struct rdma_id_private *id_priv;
2595         int ret;
2596
2597         if (addr->sa_family != AF_INET && addr->sa_family != AF_INET6 &&
2598             addr->sa_family != AF_IB)
2599                 return -EAFNOSUPPORT;
2600
2601         id_priv = container_of(id, struct rdma_id_private, id);
2602         if (!cma_comp_exch(id_priv, RDMA_CM_IDLE, RDMA_CM_ADDR_BOUND))
2603                 return -EINVAL;
2604
2605         ret = cma_check_linklocal(&id->route.addr.dev_addr, addr);
2606         if (ret)
2607                 goto err1;
2608
2609         if (!cma_any_addr(addr)) {
2610                 ret = cma_translate_addr(addr, &id->route.addr.dev_addr);
2611                 if (ret)
2612                         goto err1;
2613
2614                 ret = cma_acquire_dev(id_priv);
2615                 if (ret)
2616                         goto err1;
2617         }
2618
2619         memcpy(cma_src_addr(id_priv), addr, rdma_addr_size(addr));
2620         if (!(id_priv->options & (1 << CMA_OPTION_AFONLY))) {
2621                 if (addr->sa_family == AF_INET)
2622                         id_priv->afonly = 1;
2623 #if IS_ENABLED(CONFIG_IPV6)
2624                 else if (addr->sa_family == AF_INET6)
2625                         id_priv->afonly = init_net.ipv6.sysctl.bindv6only;
2626 #endif
2627         }
2628         ret = cma_get_port(id_priv);
2629         if (ret)
2630                 goto err2;
2631
2632         return 0;
2633 err2:
2634         if (id_priv->cma_dev)
2635                 cma_release_dev(id_priv);
2636 err1:
2637         cma_comp_exch(id_priv, RDMA_CM_ADDR_BOUND, RDMA_CM_IDLE);
2638         return ret;
2639 }
2640 EXPORT_SYMBOL(rdma_bind_addr);
2641
2642 static int cma_format_hdr(void *hdr, struct rdma_id_private *id_priv)
2643 {
2644         struct cma_hdr *cma_hdr;
2645         struct sdp_hh *sdp_hdr;
2646
2647         if (cma_family(id_priv) == AF_INET) {
2648                 struct sockaddr_in *src4, *dst4;
2649
2650                 src4 = (struct sockaddr_in *) cma_src_addr(id_priv);
2651                 dst4 = (struct sockaddr_in *) cma_dst_addr(id_priv);
2652
2653                 switch (id_priv->id.ps) {
2654                 case RDMA_PS_SDP:
2655                         sdp_hdr = hdr;
2656                         if (sdp_get_majv(sdp_hdr->sdp_version) != SDP_MAJ_VERSION)
2657                                 return -EINVAL;
2658                         sdp_set_ip_ver(sdp_hdr, 4);
2659                         sdp_hdr->src_addr.ip4.addr = src4->sin_addr.s_addr;
2660                         sdp_hdr->dst_addr.ip4.addr = dst4->sin_addr.s_addr;
2661                         sdp_hdr->port = src4->sin_port;
2662                         break;
2663                 default:
2664                         cma_hdr = hdr;
2665                         cma_hdr->cma_version = CMA_VERSION;
2666                         cma_set_ip_ver(cma_hdr, 4);
2667                         cma_hdr->src_addr.ip4.addr = src4->sin_addr.s_addr;
2668                         cma_hdr->dst_addr.ip4.addr = dst4->sin_addr.s_addr;
2669                         cma_hdr->port = src4->sin_port;
2670                         break;
2671                 }
2672         } else {
2673                 struct sockaddr_in6 *src6, *dst6;
2674
2675                 src6 = (struct sockaddr_in6 *) cma_src_addr(id_priv);
2676                 dst6 = (struct sockaddr_in6 *) cma_dst_addr(id_priv);
2677
2678                 switch (id_priv->id.ps) {
2679                 case RDMA_PS_SDP:
2680                         sdp_hdr = hdr;
2681                         if (sdp_get_majv(sdp_hdr->sdp_version) != SDP_MAJ_VERSION)
2682                                 return -EINVAL;
2683                         sdp_set_ip_ver(sdp_hdr, 6);
2684                         sdp_hdr->src_addr.ip6 = src6->sin6_addr;
2685                         sdp_hdr->dst_addr.ip6 = dst6->sin6_addr;
2686                         sdp_hdr->port = src6->sin6_port;
2687                         break;
2688                 default:
2689                         cma_hdr = hdr;
2690                         cma_hdr->cma_version = CMA_VERSION;
2691                         cma_set_ip_ver(cma_hdr, 6);
2692                         cma_hdr->src_addr.ip6 = src6->sin6_addr;
2693                         cma_hdr->dst_addr.ip6 = dst6->sin6_addr;
2694                         cma_hdr->port = src6->sin6_port;
2695                         break;
2696                 }
2697         }
2698         return 0;
2699 }
2700
2701 static int cma_sidr_rep_handler(struct ib_cm_id *cm_id,
2702                                 struct ib_cm_event *ib_event)
2703 {
2704         struct rdma_id_private *id_priv = cm_id->context;
2705         struct rdma_cm_event event;
2706         struct ib_cm_sidr_rep_event_param *rep = &ib_event->param.sidr_rep_rcvd;
2707         int ret = 0;
2708
2709         if (cma_disable_callback(id_priv, RDMA_CM_CONNECT))
2710                 return 0;
2711
2712         memset(&event, 0, sizeof event);
2713         switch (ib_event->event) {
2714         case IB_CM_SIDR_REQ_ERROR:
2715                 event.event = RDMA_CM_EVENT_UNREACHABLE;
2716                 event.status = -ETIMEDOUT;
2717                 break;
2718         case IB_CM_SIDR_REP_RECEIVED:
2719                 event.param.ud.private_data = ib_event->private_data;
2720                 event.param.ud.private_data_len = IB_CM_SIDR_REP_PRIVATE_DATA_SIZE;
2721                 if (rep->status != IB_SIDR_SUCCESS) {
2722                         event.event = RDMA_CM_EVENT_UNREACHABLE;
2723                         event.status = ib_event->param.sidr_rep_rcvd.status;
2724                         break;
2725                 }
2726                 ret = cma_set_qkey(id_priv);
2727                 if (ret) {
2728                         event.event = RDMA_CM_EVENT_ADDR_ERROR;
2729                         event.status = -EINVAL;
2730                         break;
2731                 }
2732                 if (id_priv->qkey != rep->qkey) {
2733                         event.event = RDMA_CM_EVENT_UNREACHABLE;
2734                         event.status = -EINVAL;
2735                         break;
2736                 }
2737                 ib_init_ah_from_path(id_priv->id.device, id_priv->id.port_num,
2738                                      id_priv->id.route.path_rec,
2739                                      &event.param.ud.ah_attr);
2740                 event.param.ud.qp_num = rep->qpn;
2741                 event.param.ud.qkey = rep->qkey;
2742                 event.event = RDMA_CM_EVENT_ESTABLISHED;
2743                 event.status = 0;
2744                 break;
2745         default:
2746                 printk(KERN_ERR "RDMA CMA: unexpected IB CM event: %d\n",
2747                        ib_event->event);
2748                 goto out;
2749         }
2750
2751         ret = id_priv->id.event_handler(&id_priv->id, &event);
2752         if (ret) {
2753                 /* Destroy the CM ID by returning a non-zero value. */
2754                 id_priv->cm_id.ib = NULL;
2755                 cma_exch(id_priv, RDMA_CM_DESTROYING);
2756                 mutex_unlock(&id_priv->handler_mutex);
2757                 rdma_destroy_id(&id_priv->id);
2758                 return ret;
2759         }
2760 out:
2761         mutex_unlock(&id_priv->handler_mutex);
2762         return ret;
2763 }
2764
2765 static int cma_resolve_ib_udp(struct rdma_id_private *id_priv,
2766                               struct rdma_conn_param *conn_param)
2767 {
2768         struct ib_cm_sidr_req_param req;
2769         struct ib_cm_id *id;
2770         int ret;
2771
2772         req.private_data_len = sizeof(struct cma_hdr) +
2773                                conn_param->private_data_len;
2774         if (req.private_data_len < conn_param->private_data_len)
2775                 return -EINVAL;
2776
2777         req.private_data = kzalloc(req.private_data_len, GFP_ATOMIC);
2778         if (!req.private_data)
2779                 return -ENOMEM;
2780
2781         if (conn_param->private_data && conn_param->private_data_len)
2782                 memcpy((void *) req.private_data + sizeof(struct cma_hdr),
2783                        conn_param->private_data, conn_param->private_data_len);
2784
2785         ret = cma_format_hdr((void *) req.private_data, id_priv);
2786         if (ret)
2787                 goto out;
2788
2789         id = ib_create_cm_id(id_priv->id.device, cma_sidr_rep_handler,
2790                              id_priv);
2791         if (IS_ERR(id)) {
2792                 ret = PTR_ERR(id);
2793                 goto out;
2794         }
2795         id_priv->cm_id.ib = id;
2796
2797         req.path = id_priv->id.route.path_rec;
2798         req.service_id = cma_get_service_id(id_priv->id.ps, cma_dst_addr(id_priv));
2799         req.timeout_ms = 1 << (CMA_CM_RESPONSE_TIMEOUT - 8);
2800         req.max_cm_retries = CMA_MAX_CM_RETRIES;
2801
2802         ret = ib_send_cm_sidr_req(id_priv->cm_id.ib, &req);
2803         if (ret) {
2804                 ib_destroy_cm_id(id_priv->cm_id.ib);
2805                 id_priv->cm_id.ib = NULL;
2806         }
2807 out:
2808         kfree(req.private_data);
2809         return ret;
2810 }
2811
2812 static int cma_connect_ib(struct rdma_id_private *id_priv,
2813                           struct rdma_conn_param *conn_param)
2814 {
2815         struct ib_cm_req_param req;
2816         struct rdma_route *route;
2817         void *private_data;
2818         struct ib_cm_id *id;
2819         int offset, ret;
2820
2821         memset(&req, 0, sizeof req);
2822         offset = cma_user_data_offset(id_priv->id.ps);
2823         req.private_data_len = offset + conn_param->private_data_len;
2824         if (req.private_data_len < conn_param->private_data_len)
2825                 return -EINVAL;
2826
2827         private_data = kzalloc(req.private_data_len, GFP_ATOMIC);
2828         if (!private_data)
2829                 return -ENOMEM;
2830
2831         if (conn_param->private_data && conn_param->private_data_len)
2832                 memcpy(private_data + offset, conn_param->private_data,
2833                        conn_param->private_data_len);
2834
2835         id = ib_create_cm_id(id_priv->id.device, cma_ib_handler, id_priv);
2836         if (IS_ERR(id)) {
2837                 ret = PTR_ERR(id);
2838                 goto out;
2839         }
2840         id_priv->cm_id.ib = id;
2841
2842         route = &id_priv->id.route;
2843         ret = cma_format_hdr(private_data, id_priv);
2844         if (ret)
2845                 goto out;
2846         req.private_data = private_data;
2847
2848         req.primary_path = &route->path_rec[0];
2849         if (route->num_paths == 2)
2850                 req.alternate_path = &route->path_rec[1];
2851
2852         req.service_id = cma_get_service_id(id_priv->id.ps, cma_dst_addr(id_priv));
2853         req.qp_num = id_priv->qp_num;
2854         req.qp_type = id_priv->id.qp_type;
2855         req.starting_psn = id_priv->seq_num;
2856         req.responder_resources = conn_param->responder_resources;
2857         req.initiator_depth = conn_param->initiator_depth;
2858         req.flow_control = conn_param->flow_control;
2859         req.retry_count = min_t(u8, 7, conn_param->retry_count);
2860         req.rnr_retry_count = min_t(u8, 7, conn_param->rnr_retry_count);
2861         req.remote_cm_response_timeout = CMA_CM_RESPONSE_TIMEOUT;
2862         req.local_cm_response_timeout = CMA_CM_RESPONSE_TIMEOUT;
2863         req.max_cm_retries = CMA_MAX_CM_RETRIES;
2864         req.srq = id_priv->srq ? 1 : 0;
2865
2866         ret = ib_send_cm_req(id_priv->cm_id.ib, &req);
2867 out:
2868         if (ret && !IS_ERR(id)) {
2869                 ib_destroy_cm_id(id);
2870                 id_priv->cm_id.ib = NULL;
2871         }
2872
2873         kfree(private_data);
2874         return ret;
2875 }
2876
2877 static int cma_connect_iw(struct rdma_id_private *id_priv,
2878                           struct rdma_conn_param *conn_param)
2879 {
2880         struct iw_cm_id *cm_id;
2881         struct sockaddr_in* sin;
2882         int ret;
2883         struct iw_cm_conn_param iw_param;
2884
2885         cm_id = iw_create_cm_id(id_priv->id.device, cma_iw_handler, id_priv);
2886         if (IS_ERR(cm_id))
2887                 return PTR_ERR(cm_id);
2888
2889         id_priv->cm_id.iw = cm_id;
2890
2891         sin = (struct sockaddr_in *) cma_src_addr(id_priv);
2892         cm_id->local_addr = *sin;
2893
2894         sin = (struct sockaddr_in *) cma_dst_addr(id_priv);
2895         cm_id->remote_addr = *sin;
2896
2897         ret = cma_modify_qp_rtr(id_priv, conn_param);
2898         if (ret)
2899                 goto out;
2900
2901         if (conn_param) {
2902                 iw_param.ord = conn_param->initiator_depth;
2903                 iw_param.ird = conn_param->responder_resources;
2904                 iw_param.private_data = conn_param->private_data;
2905                 iw_param.private_data_len = conn_param->private_data_len;
2906                 iw_param.qpn = id_priv->id.qp ? id_priv->qp_num : conn_param->qp_num;
2907         } else {
2908                 memset(&iw_param, 0, sizeof iw_param);
2909                 iw_param.qpn = id_priv->qp_num;
2910         }
2911         ret = iw_cm_connect(cm_id, &iw_param);
2912 out:
2913         if (ret) {
2914                 iw_destroy_cm_id(cm_id);
2915                 id_priv->cm_id.iw = NULL;
2916         }
2917         return ret;
2918 }
2919
2920 int rdma_connect(struct rdma_cm_id *id, struct rdma_conn_param *conn_param)
2921 {
2922         struct rdma_id_private *id_priv;
2923         int ret;
2924
2925         id_priv = container_of(id, struct rdma_id_private, id);
2926         if (!cma_comp_exch(id_priv, RDMA_CM_ROUTE_RESOLVED, RDMA_CM_CONNECT))
2927                 return -EINVAL;
2928
2929         if (!id->qp) {
2930                 id_priv->qp_num = conn_param->qp_num;
2931                 id_priv->srq = conn_param->srq;
2932         }
2933
2934         switch (rdma_node_get_transport(id->device->node_type)) {
2935         case RDMA_TRANSPORT_IB:
2936                 if (id->qp_type == IB_QPT_UD)
2937                         ret = cma_resolve_ib_udp(id_priv, conn_param);
2938                 else
2939                         ret = cma_connect_ib(id_priv, conn_param);
2940                 break;
2941         case RDMA_TRANSPORT_IWARP:
2942                 ret = cma_connect_iw(id_priv, conn_param);
2943                 break;
2944         default:
2945                 ret = -ENOSYS;
2946                 break;
2947         }
2948         if (ret)
2949                 goto err;
2950
2951         return 0;
2952 err:
2953         cma_comp_exch(id_priv, RDMA_CM_CONNECT, RDMA_CM_ROUTE_RESOLVED);
2954         return ret;
2955 }
2956 EXPORT_SYMBOL(rdma_connect);
2957
2958 static int cma_accept_ib(struct rdma_id_private *id_priv,
2959                          struct rdma_conn_param *conn_param)
2960 {
2961         struct ib_cm_rep_param rep;
2962         int ret;
2963
2964         ret = cma_modify_qp_rtr(id_priv, conn_param);
2965         if (ret)
2966                 goto out;
2967
2968         ret = cma_modify_qp_rts(id_priv, conn_param);
2969         if (ret)
2970                 goto out;
2971
2972         memset(&rep, 0, sizeof rep);
2973         rep.qp_num = id_priv->qp_num;
2974         rep.starting_psn = id_priv->seq_num;
2975         rep.private_data = conn_param->private_data;
2976         rep.private_data_len = conn_param->private_data_len;
2977         rep.responder_resources = conn_param->responder_resources;
2978         rep.initiator_depth = conn_param->initiator_depth;
2979         rep.failover_accepted = 0;
2980         rep.flow_control = conn_param->flow_control;
2981         rep.rnr_retry_count = min_t(u8, 7, conn_param->rnr_retry_count);
2982         rep.srq = id_priv->srq ? 1 : 0;
2983
2984         ret = ib_send_cm_rep(id_priv->cm_id.ib, &rep);
2985 out:
2986         return ret;
2987 }
2988
2989 static int cma_accept_iw(struct rdma_id_private *id_priv,
2990                   struct rdma_conn_param *conn_param)
2991 {
2992         struct iw_cm_conn_param iw_param;
2993         int ret;
2994
2995         ret = cma_modify_qp_rtr(id_priv, conn_param);
2996         if (ret)
2997                 return ret;
2998
2999         iw_param.ord = conn_param->initiator_depth;
3000         iw_param.ird = conn_param->responder_resources;
3001         iw_param.private_data = conn_param->private_data;
3002         iw_param.private_data_len = conn_param->private_data_len;
3003         if (id_priv->id.qp) {
3004                 iw_param.qpn = id_priv->qp_num;
3005         } else
3006                 iw_param.qpn = conn_param->qp_num;
3007
3008         return iw_cm_accept(id_priv->cm_id.iw, &iw_param);
3009 }
3010
3011 static int cma_send_sidr_rep(struct rdma_id_private *id_priv,
3012                              enum ib_cm_sidr_status status,
3013                              const void *private_data, int private_data_len)
3014 {
3015         struct ib_cm_sidr_rep_param rep;
3016         int ret;
3017
3018         memset(&rep, 0, sizeof rep);
3019         rep.status = status;
3020         if (status == IB_SIDR_SUCCESS) {
3021                 ret = cma_set_qkey(id_priv);
3022                 if (ret)
3023                         return ret;
3024                 rep.qp_num = id_priv->qp_num;
3025                 rep.qkey = id_priv->qkey;
3026         }
3027         rep.private_data = private_data;
3028         rep.private_data_len = private_data_len;
3029
3030         return ib_send_cm_sidr_rep(id_priv->cm_id.ib, &rep);
3031 }
3032
3033 int rdma_accept(struct rdma_cm_id *id, struct rdma_conn_param *conn_param)
3034 {
3035         struct rdma_id_private *id_priv;
3036         int ret;
3037
3038         id_priv = container_of(id, struct rdma_id_private, id);
3039
3040         id_priv->owner = task_pid_nr(current);
3041
3042         if (!cma_comp(id_priv, RDMA_CM_CONNECT))
3043                 return -EINVAL;
3044
3045         if (!id->qp && conn_param) {
3046                 id_priv->qp_num = conn_param->qp_num;
3047                 id_priv->srq = conn_param->srq;
3048         }
3049
3050         switch (rdma_node_get_transport(id->device->node_type)) {
3051         case RDMA_TRANSPORT_IB:
3052                 if (id->qp_type == IB_QPT_UD) {
3053                         if (conn_param)
3054                                 ret = cma_send_sidr_rep(id_priv, IB_SIDR_SUCCESS,
3055                                                         conn_param->private_data,
3056                                                         conn_param->private_data_len);
3057                         else
3058                                 ret = cma_send_sidr_rep(id_priv, IB_SIDR_SUCCESS,
3059                                                         NULL, 0);
3060                 } else {
3061                         if (conn_param)
3062                                 ret = cma_accept_ib(id_priv, conn_param);
3063                         else
3064                                 ret = cma_rep_recv(id_priv);
3065                 }
3066                 break;
3067         case RDMA_TRANSPORT_IWARP:
3068                 ret = cma_accept_iw(id_priv, conn_param);
3069                 break;
3070         default:
3071                 ret = -ENOSYS;
3072                 break;
3073         }
3074
3075         if (ret)
3076                 goto reject;
3077
3078         return 0;
3079 reject:
3080         cma_modify_qp_err(id_priv);
3081         rdma_reject(id, NULL, 0);
3082         return ret;
3083 }
3084 EXPORT_SYMBOL(rdma_accept);
3085
3086 int rdma_notify(struct rdma_cm_id *id, enum ib_event_type event)
3087 {
3088         struct rdma_id_private *id_priv;
3089         int ret;
3090
3091         id_priv = container_of(id, struct rdma_id_private, id);
3092         if (!id_priv->cm_id.ib)
3093                 return -EINVAL;
3094
3095         switch (id->device->node_type) {
3096         case RDMA_NODE_IB_CA:
3097                 ret = ib_cm_notify(id_priv->cm_id.ib, event);
3098                 break;
3099         default:
3100                 ret = 0;
3101                 break;
3102         }
3103         return ret;
3104 }
3105 EXPORT_SYMBOL(rdma_notify);
3106
3107 int rdma_reject(struct rdma_cm_id *id, const void *private_data,
3108                 u8 private_data_len)
3109 {
3110         struct rdma_id_private *id_priv;
3111         int ret;
3112
3113         id_priv = container_of(id, struct rdma_id_private, id);
3114         if (!id_priv->cm_id.ib)
3115                 return -EINVAL;
3116
3117         switch (rdma_node_get_transport(id->device->node_type)) {
3118         case RDMA_TRANSPORT_IB:
3119                 if (id->qp_type == IB_QPT_UD)
3120                         ret = cma_send_sidr_rep(id_priv, IB_SIDR_REJECT,
3121                                                 private_data, private_data_len);
3122                 else
3123                         ret = ib_send_cm_rej(id_priv->cm_id.ib,
3124                                              IB_CM_REJ_CONSUMER_DEFINED, NULL,
3125                                              0, private_data, private_data_len);
3126                 break;
3127         case RDMA_TRANSPORT_IWARP:
3128                 ret = iw_cm_reject(id_priv->cm_id.iw,
3129                                    private_data, private_data_len);
3130                 break;
3131         default:
3132                 ret = -ENOSYS;
3133                 break;
3134         }
3135         return ret;
3136 }
3137 EXPORT_SYMBOL(rdma_reject);
3138
3139 int rdma_disconnect(struct rdma_cm_id *id)
3140 {
3141         struct rdma_id_private *id_priv;
3142         int ret;
3143
3144         id_priv = container_of(id, struct rdma_id_private, id);
3145         if (!id_priv->cm_id.ib)
3146                 return -EINVAL;
3147
3148         switch (rdma_node_get_transport(id->device->node_type)) {
3149         case RDMA_TRANSPORT_IB:
3150                 ret = cma_modify_qp_err(id_priv);
3151                 if (ret)
3152                         goto out;
3153                 /* Initiate or respond to a disconnect. */
3154                 if (ib_send_cm_dreq(id_priv->cm_id.ib, NULL, 0))
3155                         ib_send_cm_drep(id_priv->cm_id.ib, NULL, 0);
3156                 break;
3157         case RDMA_TRANSPORT_IWARP:
3158                 ret = iw_cm_disconnect(id_priv->cm_id.iw, 0);
3159                 break;
3160         default:
3161                 ret = -EINVAL;
3162                 break;
3163         }
3164 out:
3165         return ret;
3166 }
3167 EXPORT_SYMBOL(rdma_disconnect);
3168
3169 static int cma_ib_mc_handler(int status, struct ib_sa_multicast *multicast)
3170 {
3171         struct rdma_id_private *id_priv;
3172         struct cma_multicast *mc = multicast->context;
3173         struct rdma_cm_event event;
3174         int ret;
3175
3176         id_priv = mc->id_priv;
3177         if (cma_disable_callback(id_priv, RDMA_CM_ADDR_BOUND) &&
3178             cma_disable_callback(id_priv, RDMA_CM_ADDR_RESOLVED))
3179                 return 0;
3180
3181         mutex_lock(&id_priv->qp_mutex);
3182         if (!status && id_priv->id.qp)
3183                 status = ib_attach_mcast(id_priv->id.qp, &multicast->rec.mgid,
3184                                          be16_to_cpu(multicast->rec.mlid));
3185         mutex_unlock(&id_priv->qp_mutex);
3186
3187         memset(&event, 0, sizeof event);
3188         event.status = status;
3189         event.param.ud.private_data = mc->context;
3190         if (!status) {
3191                 event.event = RDMA_CM_EVENT_MULTICAST_JOIN;
3192                 ib_init_ah_from_mcmember(id_priv->id.device,
3193                                          id_priv->id.port_num, &multicast->rec,
3194                                          &event.param.ud.ah_attr);
3195                 event.param.ud.qp_num = 0xFFFFFF;
3196                 event.param.ud.qkey = be32_to_cpu(multicast->rec.qkey);
3197         } else
3198                 event.event = RDMA_CM_EVENT_MULTICAST_ERROR;
3199
3200         ret = id_priv->id.event_handler(&id_priv->id, &event);
3201         if (ret) {
3202                 cma_exch(id_priv, RDMA_CM_DESTROYING);
3203                 mutex_unlock(&id_priv->handler_mutex);
3204                 rdma_destroy_id(&id_priv->id);
3205                 return 0;
3206         }
3207
3208         mutex_unlock(&id_priv->handler_mutex);
3209         return 0;
3210 }
3211
3212 static void cma_set_mgid(struct rdma_id_private *id_priv,
3213                          struct sockaddr *addr, union ib_gid *mgid)
3214 {
3215         unsigned char mc_map[MAX_ADDR_LEN];
3216         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
3217         struct sockaddr_in *sin = (struct sockaddr_in *) addr;
3218         struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *) addr;
3219
3220         if (cma_any_addr(addr)) {
3221                 memset(mgid, 0, sizeof *mgid);
3222         } else if ((addr->sa_family == AF_INET6) &&
3223                    ((be32_to_cpu(sin6->sin6_addr.s6_addr32[0]) & 0xFFF0FFFF) ==
3224                                                                  0xFF10A01B)) {
3225                 /* IPv6 address is an SA assigned MGID. */
3226                 memcpy(mgid, &sin6->sin6_addr, sizeof *mgid);
3227         } else if ((addr->sa_family == AF_INET6)) {
3228                 ipv6_ib_mc_map(&sin6->sin6_addr, dev_addr->broadcast, mc_map);
3229                 if (id_priv->id.ps == RDMA_PS_UDP)
3230                         mc_map[7] = 0x01;       /* Use RDMA CM signature */
3231                 *mgid = *(union ib_gid *) (mc_map + 4);
3232         } else {
3233                 ip_ib_mc_map(sin->sin_addr.s_addr, dev_addr->broadcast, mc_map);
3234                 if (id_priv->id.ps == RDMA_PS_UDP)
3235                         mc_map[7] = 0x01;       /* Use RDMA CM signature */
3236                 *mgid = *(union ib_gid *) (mc_map + 4);
3237         }
3238 }
3239
3240 static int cma_join_ib_multicast(struct rdma_id_private *id_priv,
3241                                  struct cma_multicast *mc)
3242 {
3243         struct ib_sa_mcmember_rec rec;
3244         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
3245         ib_sa_comp_mask comp_mask;
3246         int ret;
3247
3248         ib_addr_get_mgid(dev_addr, &rec.mgid);
3249         ret = ib_sa_get_mcmember_rec(id_priv->id.device, id_priv->id.port_num,
3250                                      &rec.mgid, &rec);
3251         if (ret)
3252                 return ret;
3253
3254         cma_set_mgid(id_priv, (struct sockaddr *) &mc->addr, &rec.mgid);
3255         if (id_priv->id.ps == RDMA_PS_UDP)
3256                 rec.qkey = cpu_to_be32(RDMA_UDP_QKEY);
3257         rdma_addr_get_sgid(dev_addr, &rec.port_gid);
3258         rec.pkey = cpu_to_be16(ib_addr_get_pkey(dev_addr));
3259         rec.join_state = 1;
3260
3261         comp_mask = IB_SA_MCMEMBER_REC_MGID | IB_SA_MCMEMBER_REC_PORT_GID |
3262                     IB_SA_MCMEMBER_REC_PKEY | IB_SA_MCMEMBER_REC_JOIN_STATE |
3263                     IB_SA_MCMEMBER_REC_QKEY | IB_SA_MCMEMBER_REC_SL |
3264                     IB_SA_MCMEMBER_REC_FLOW_LABEL |
3265                     IB_SA_MCMEMBER_REC_TRAFFIC_CLASS;
3266
3267         if (id_priv->id.ps == RDMA_PS_IPOIB)
3268                 comp_mask |= IB_SA_MCMEMBER_REC_RATE |
3269                              IB_SA_MCMEMBER_REC_RATE_SELECTOR |
3270                              IB_SA_MCMEMBER_REC_MTU_SELECTOR |
3271                              IB_SA_MCMEMBER_REC_MTU |
3272                              IB_SA_MCMEMBER_REC_HOP_LIMIT;
3273
3274         mc->multicast.ib = ib_sa_join_multicast(&sa_client, id_priv->id.device,
3275                                                 id_priv->id.port_num, &rec,
3276                                                 comp_mask, GFP_KERNEL,
3277                                                 cma_ib_mc_handler, mc);
3278         return PTR_RET(mc->multicast.ib);
3279 }
3280
3281 static void iboe_mcast_work_handler(struct work_struct *work)
3282 {
3283         struct iboe_mcast_work *mw = container_of(work, struct iboe_mcast_work, work);
3284         struct cma_multicast *mc = mw->mc;
3285         struct ib_sa_multicast *m = mc->multicast.ib;
3286
3287         mc->multicast.ib->context = mc;
3288         cma_ib_mc_handler(0, m);
3289         kref_put(&mc->mcref, release_mc);
3290         kfree(mw);
3291 }
3292
3293 static void cma_iboe_set_mgid(struct sockaddr *addr, union ib_gid *mgid)
3294 {
3295         struct sockaddr_in *sin = (struct sockaddr_in *)addr;
3296         struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)addr;
3297
3298         if (cma_any_addr(addr)) {
3299                 memset(mgid, 0, sizeof *mgid);
3300         } else if (addr->sa_family == AF_INET6) {
3301                 memcpy(mgid, &sin6->sin6_addr, sizeof *mgid);
3302         } else {
3303                 mgid->raw[0] = 0xff;
3304                 mgid->raw[1] = 0x0e;
3305                 mgid->raw[2] = 0;
3306                 mgid->raw[3] = 0;
3307                 mgid->raw[4] = 0;
3308                 mgid->raw[5] = 0;
3309                 mgid->raw[6] = 0;
3310                 mgid->raw[7] = 0;
3311                 mgid->raw[8] = 0;
3312                 mgid->raw[9] = 0;
3313                 mgid->raw[10] = 0xff;
3314                 mgid->raw[11] = 0xff;
3315                 *(__be32 *)(&mgid->raw[12]) = sin->sin_addr.s_addr;
3316         }
3317 }
3318
3319 static int cma_iboe_join_multicast(struct rdma_id_private *id_priv,
3320                                    struct cma_multicast *mc)
3321 {
3322         struct iboe_mcast_work *work;
3323         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
3324         int err;
3325         struct sockaddr *addr = (struct sockaddr *)&mc->addr;
3326         struct net_device *ndev = NULL;
3327
3328         if (cma_zero_addr((struct sockaddr *)&mc->addr))
3329                 return -EINVAL;
3330
3331         work = kzalloc(sizeof *work, GFP_KERNEL);
3332         if (!work)
3333                 return -ENOMEM;
3334
3335         mc->multicast.ib = kzalloc(sizeof(struct ib_sa_multicast), GFP_KERNEL);
3336         if (!mc->multicast.ib) {
3337                 err = -ENOMEM;
3338                 goto out1;
3339         }
3340
3341         cma_iboe_set_mgid(addr, &mc->multicast.ib->rec.mgid);
3342
3343         mc->multicast.ib->rec.pkey = cpu_to_be16(0xffff);
3344         if (id_priv->id.ps == RDMA_PS_UDP)
3345                 mc->multicast.ib->rec.qkey = cpu_to_be32(RDMA_UDP_QKEY);
3346
3347         if (dev_addr->bound_dev_if)
3348                 ndev = dev_get_by_index(&init_net, dev_addr->bound_dev_if);
3349         if (!ndev) {
3350                 err = -ENODEV;
3351                 goto out2;
3352         }
3353         mc->multicast.ib->rec.rate = iboe_get_rate(ndev);
3354         mc->multicast.ib->rec.hop_limit = 1;
3355         mc->multicast.ib->rec.mtu = iboe_get_mtu(ndev->mtu);
3356         dev_put(ndev);
3357         if (!mc->multicast.ib->rec.mtu) {
3358                 err = -EINVAL;
3359                 goto out2;
3360         }
3361         iboe_addr_get_sgid(dev_addr, &mc->multicast.ib->rec.port_gid);
3362         work->id = id_priv;
3363         work->mc = mc;
3364         INIT_WORK(&work->work, iboe_mcast_work_handler);
3365         kref_get(&mc->mcref);
3366         queue_work(cma_wq, &work->work);
3367
3368         return 0;
3369
3370 out2:
3371         kfree(mc->multicast.ib);
3372 out1:
3373         kfree(work);
3374         return err;
3375 }
3376
3377 int rdma_join_multicast(struct rdma_cm_id *id, struct sockaddr *addr,
3378                         void *context)
3379 {
3380         struct rdma_id_private *id_priv;
3381         struct cma_multicast *mc;
3382         int ret;
3383
3384         id_priv = container_of(id, struct rdma_id_private, id);
3385         if (!cma_comp(id_priv, RDMA_CM_ADDR_BOUND) &&
3386             !cma_comp(id_priv, RDMA_CM_ADDR_RESOLVED))
3387                 return -EINVAL;
3388
3389         mc = kmalloc(sizeof *mc, GFP_KERNEL);
3390         if (!mc)
3391                 return -ENOMEM;
3392
3393         memcpy(&mc->addr, addr, rdma_addr_size(addr));
3394         mc->context = context;
3395         mc->id_priv = id_priv;
3396
3397         spin_lock(&id_priv->lock);
3398         list_add(&mc->list, &id_priv->mc_list);
3399         spin_unlock(&id_priv->lock);
3400
3401         switch (rdma_node_get_transport(id->device->node_type)) {
3402         case RDMA_TRANSPORT_IB:
3403                 switch (rdma_port_get_link_layer(id->device, id->port_num)) {
3404                 case IB_LINK_LAYER_INFINIBAND:
3405                         ret = cma_join_ib_multicast(id_priv, mc);
3406                         break;
3407                 case IB_LINK_LAYER_ETHERNET:
3408                         kref_init(&mc->mcref);
3409                         ret = cma_iboe_join_multicast(id_priv, mc);
3410                         break;
3411                 default:
3412                         ret = -EINVAL;
3413                 }
3414                 break;
3415         default:
3416                 ret = -ENOSYS;
3417                 break;
3418         }
3419
3420         if (ret) {
3421                 spin_lock_irq(&id_priv->lock);
3422                 list_del(&mc->list);
3423                 spin_unlock_irq(&id_priv->lock);
3424                 kfree(mc);
3425         }
3426         return ret;
3427 }
3428 EXPORT_SYMBOL(rdma_join_multicast);
3429
3430 void rdma_leave_multicast(struct rdma_cm_id *id, struct sockaddr *addr)
3431 {
3432         struct rdma_id_private *id_priv;
3433         struct cma_multicast *mc;
3434
3435         id_priv = container_of(id, struct rdma_id_private, id);
3436         spin_lock_irq(&id_priv->lock);
3437         list_for_each_entry(mc, &id_priv->mc_list, list) {
3438                 if (!memcmp(&mc->addr, addr, rdma_addr_size(addr))) {
3439                         list_del(&mc->list);
3440                         spin_unlock_irq(&id_priv->lock);
3441
3442                         if (id->qp)
3443                                 ib_detach_mcast(id->qp,
3444                                                 &mc->multicast.ib->rec.mgid,
3445                                                 be16_to_cpu(mc->multicast.ib->rec.mlid));
3446                         if (rdma_node_get_transport(id_priv->cma_dev->device->node_type) == RDMA_TRANSPORT_IB) {
3447                                 switch (rdma_port_get_link_layer(id->device, id->port_num)) {
3448                                 case IB_LINK_LAYER_INFINIBAND:
3449                                         ib_sa_free_multicast(mc->multicast.ib);
3450                                         kfree(mc);
3451                                         break;
3452                                 case IB_LINK_LAYER_ETHERNET:
3453                                         kref_put(&mc->mcref, release_mc);
3454                                         break;
3455                                 default:
3456                                         break;
3457                                 }
3458                         }
3459                         return;
3460                 }
3461         }
3462         spin_unlock_irq(&id_priv->lock);
3463 }
3464 EXPORT_SYMBOL(rdma_leave_multicast);
3465
3466 static int cma_netdev_change(struct net_device *ndev, struct rdma_id_private *id_priv)
3467 {
3468         struct rdma_dev_addr *dev_addr;
3469         struct cma_ndev_work *work;
3470
3471         dev_addr = &id_priv->id.route.addr.dev_addr;
3472
3473         if ((dev_addr->bound_dev_if == ndev->ifindex) &&
3474             memcmp(dev_addr->src_dev_addr, ndev->dev_addr, ndev->addr_len)) {
3475                 printk(KERN_INFO "RDMA CM addr change for ndev %s used by id %p\n",
3476                        ndev->name, &id_priv->id);
3477                 work = kzalloc(sizeof *work, GFP_KERNEL);
3478                 if (!work)
3479                         return -ENOMEM;
3480
3481                 INIT_WORK(&work->work, cma_ndev_work_handler);
3482                 work->id = id_priv;
3483                 work->event.event = RDMA_CM_EVENT_ADDR_CHANGE;
3484                 atomic_inc(&id_priv->refcount);
3485                 queue_work(cma_wq, &work->work);
3486         }
3487
3488         return 0;
3489 }
3490
3491 static int cma_netdev_callback(struct notifier_block *self, unsigned long event,
3492                                void *ptr)
3493 {
3494         struct net_device *ndev = netdev_notifier_info_to_dev(ptr);
3495         struct cma_device *cma_dev;
3496         struct rdma_id_private *id_priv;
3497         int ret = NOTIFY_DONE;
3498
3499         if (dev_net(ndev) != &init_net)
3500                 return NOTIFY_DONE;
3501
3502         if (event != NETDEV_BONDING_FAILOVER)
3503                 return NOTIFY_DONE;
3504
3505         if (!(ndev->flags & IFF_MASTER) || !(ndev->priv_flags & IFF_BONDING))
3506                 return NOTIFY_DONE;
3507
3508         mutex_lock(&lock);
3509         list_for_each_entry(cma_dev, &dev_list, list)
3510                 list_for_each_entry(id_priv, &cma_dev->id_list, list) {
3511                         ret = cma_netdev_change(ndev, id_priv);
3512                         if (ret)
3513                                 goto out;
3514                 }
3515
3516 out:
3517         mutex_unlock(&lock);
3518         return ret;
3519 }
3520
3521 static struct notifier_block cma_nb = {
3522         .notifier_call = cma_netdev_callback
3523 };
3524
3525 static void cma_add_one(struct ib_device *device)
3526 {
3527         struct cma_device *cma_dev;
3528         struct rdma_id_private *id_priv;
3529
3530         cma_dev = kmalloc(sizeof *cma_dev, GFP_KERNEL);
3531         if (!cma_dev)
3532                 return;
3533
3534         cma_dev->device = device;
3535
3536         init_completion(&cma_dev->comp);
3537         atomic_set(&cma_dev->refcount, 1);
3538         INIT_LIST_HEAD(&cma_dev->id_list);
3539         ib_set_client_data(device, &cma_client, cma_dev);
3540
3541         mutex_lock(&lock);
3542         list_add_tail(&cma_dev->list, &dev_list);
3543         list_for_each_entry(id_priv, &listen_any_list, list)
3544                 cma_listen_on_dev(id_priv, cma_dev);
3545         mutex_unlock(&lock);
3546 }
3547
3548 static int cma_remove_id_dev(struct rdma_id_private *id_priv)
3549 {
3550         struct rdma_cm_event event;
3551         enum rdma_cm_state state;
3552         int ret = 0;
3553
3554         /* Record that we want to remove the device */
3555         state = cma_exch(id_priv, RDMA_CM_DEVICE_REMOVAL);
3556         if (state == RDMA_CM_DESTROYING)
3557                 return 0;
3558
3559         cma_cancel_operation(id_priv, state);
3560         mutex_lock(&id_priv->handler_mutex);
3561
3562         /* Check for destruction from another callback. */
3563         if (!cma_comp(id_priv, RDMA_CM_DEVICE_REMOVAL))
3564                 goto out;
3565
3566         memset(&event, 0, sizeof event);
3567         event.event = RDMA_CM_EVENT_DEVICE_REMOVAL;
3568         ret = id_priv->id.event_handler(&id_priv->id, &event);
3569 out:
3570         mutex_unlock(&id_priv->handler_mutex);
3571         return ret;
3572 }
3573
3574 static void cma_process_remove(struct cma_device *cma_dev)
3575 {
3576         struct rdma_id_private *id_priv;
3577         int ret;
3578
3579         mutex_lock(&lock);
3580         while (!list_empty(&cma_dev->id_list)) {
3581                 id_priv = list_entry(cma_dev->id_list.next,
3582                                      struct rdma_id_private, list);
3583
3584                 list_del(&id_priv->listen_list);
3585                 list_del_init(&id_priv->list);
3586                 atomic_inc(&id_priv->refcount);
3587                 mutex_unlock(&lock);
3588
3589                 ret = id_priv->internal_id ? 1 : cma_remove_id_dev(id_priv);
3590                 cma_deref_id(id_priv);
3591                 if (ret)
3592                         rdma_destroy_id(&id_priv->id);
3593
3594                 mutex_lock(&lock);
3595         }
3596         mutex_unlock(&lock);
3597
3598         cma_deref_dev(cma_dev);
3599         wait_for_completion(&cma_dev->comp);
3600 }
3601
3602 static void cma_remove_one(struct ib_device *device)
3603 {
3604         struct cma_device *cma_dev;
3605
3606         cma_dev = ib_get_client_data(device, &cma_client);
3607         if (!cma_dev)
3608                 return;
3609
3610         mutex_lock(&lock);
3611         list_del(&cma_dev->list);
3612         mutex_unlock(&lock);
3613
3614         cma_process_remove(cma_dev);
3615         kfree(cma_dev);
3616 }
3617
3618 static int cma_get_id_stats(struct sk_buff *skb, struct netlink_callback *cb)
3619 {
3620         struct nlmsghdr *nlh;
3621         struct rdma_cm_id_stats *id_stats;
3622         struct rdma_id_private *id_priv;
3623         struct rdma_cm_id *id = NULL;
3624         struct cma_device *cma_dev;
3625         int i_dev = 0, i_id = 0;
3626
3627         /*
3628          * We export all of the IDs as a sequence of messages.  Each
3629          * ID gets its own netlink message.
3630          */
3631         mutex_lock(&lock);
3632
3633         list_for_each_entry(cma_dev, &dev_list, list) {
3634                 if (i_dev < cb->args[0]) {
3635                         i_dev++;
3636                         continue;
3637                 }
3638
3639                 i_id = 0;
3640                 list_for_each_entry(id_priv, &cma_dev->id_list, list) {
3641                         if (i_id < cb->args[1]) {
3642                                 i_id++;
3643                                 continue;
3644                         }
3645
3646                         id_stats = ibnl_put_msg(skb, &nlh, cb->nlh->nlmsg_seq,
3647                                                 sizeof *id_stats, RDMA_NL_RDMA_CM,
3648                                                 RDMA_NL_RDMA_CM_ID_STATS);
3649                         if (!id_stats)
3650                                 goto out;
3651
3652                         memset(id_stats, 0, sizeof *id_stats);
3653                         id = &id_priv->id;
3654                         id_stats->node_type = id->route.addr.dev_addr.dev_type;
3655                         id_stats->port_num = id->port_num;
3656                         id_stats->bound_dev_if =
3657                                 id->route.addr.dev_addr.bound_dev_if;
3658
3659                         if (cma_family(id_priv) == AF_INET) {
3660                                 if (ibnl_put_attr(skb, nlh,
3661                                                   sizeof(struct sockaddr_in),
3662                                                   cma_src_addr(id_priv),
3663                                                   RDMA_NL_RDMA_CM_ATTR_SRC_ADDR)) {
3664                                         goto out;
3665                                 }
3666                                 if (ibnl_put_attr(skb, nlh,
3667                                                   sizeof(struct sockaddr_in),
3668                                                   cma_dst_addr(id_priv),
3669                                                   RDMA_NL_RDMA_CM_ATTR_DST_ADDR)) {
3670                                         goto out;
3671                                 }
3672                         } else if (cma_family(id_priv) == AF_INET6) {
3673                                 if (ibnl_put_attr(skb, nlh,
3674                                                   sizeof(struct sockaddr_in6),
3675                                                   cma_src_addr(id_priv),
3676                                                   RDMA_NL_RDMA_CM_ATTR_SRC_ADDR)) {
3677                                         goto out;
3678                                 }
3679                                 if (ibnl_put_attr(skb, nlh,
3680                                                   sizeof(struct sockaddr_in6),
3681                                                   cma_dst_addr(id_priv),
3682                                                   RDMA_NL_RDMA_CM_ATTR_DST_ADDR)) {
3683                                         goto out;
3684                                 }
3685                         }
3686
3687                         id_stats->pid           = id_priv->owner;
3688                         id_stats->port_space    = id->ps;
3689                         id_stats->cm_state      = id_priv->state;
3690                         id_stats->qp_num        = id_priv->qp_num;
3691                         id_stats->qp_type       = id->qp_type;
3692
3693                         i_id++;
3694                 }
3695
3696                 cb->args[1] = 0;
3697                 i_dev++;
3698         }
3699
3700 out:
3701         mutex_unlock(&lock);
3702         cb->args[0] = i_dev;
3703         cb->args[1] = i_id;
3704
3705         return skb->len;
3706 }
3707
3708 static const struct ibnl_client_cbs cma_cb_table[] = {
3709         [RDMA_NL_RDMA_CM_ID_STATS] = { .dump = cma_get_id_stats,
3710                                        .module = THIS_MODULE },
3711 };
3712
3713 static int __init cma_init(void)
3714 {
3715         int ret;
3716
3717         cma_wq = create_singlethread_workqueue("rdma_cm");
3718         if (!cma_wq)
3719                 return -ENOMEM;
3720
3721         ib_sa_register_client(&sa_client);
3722         rdma_addr_register_client(&addr_client);
3723         register_netdevice_notifier(&cma_nb);
3724
3725         ret = ib_register_client(&cma_client);
3726         if (ret)
3727                 goto err;
3728
3729         if (ibnl_add_client(RDMA_NL_RDMA_CM, RDMA_NL_RDMA_CM_NUM_OPS, cma_cb_table))
3730                 printk(KERN_WARNING "RDMA CMA: failed to add netlink callback\n");
3731
3732         return 0;
3733
3734 err:
3735         unregister_netdevice_notifier(&cma_nb);
3736         rdma_addr_unregister_client(&addr_client);
3737         ib_sa_unregister_client(&sa_client);
3738         destroy_workqueue(cma_wq);
3739         return ret;
3740 }
3741
3742 static void __exit cma_cleanup(void)
3743 {
3744         ibnl_remove_client(RDMA_NL_RDMA_CM);
3745         ib_unregister_client(&cma_client);
3746         unregister_netdevice_notifier(&cma_nb);
3747         rdma_addr_unregister_client(&addr_client);
3748         ib_sa_unregister_client(&sa_client);
3749         destroy_workqueue(cma_wq);
3750         idr_destroy(&sdp_ps);
3751         idr_destroy(&tcp_ps);
3752         idr_destroy(&udp_ps);
3753         idr_destroy(&ipoib_ps);
3754         idr_destroy(&ib_ps);
3755 }
3756
3757 module_init(cma_init);
3758 module_exit(cma_cleanup);