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 #include <net/ip_fib.h>
  50 #include <net/ip6_route.h>
  51
  52 #include <rdma/rdma_cm.h>
  53 #include <rdma/rdma_cm_ib.h>
  54 #include <rdma/rdma_netlink.h>
  55 #include <rdma/ib.h>
  56 #include <rdma/ib_cache.h>
  57 #include <rdma/ib_cm.h>
  58 #include <rdma/ib_sa.h>
  59 #include <rdma/iw_cm.h>
  60
  61 MODULE_AUTHOR("Sean Hefty");
  62 MODULE_DESCRIPTION("Generic RDMA CM Agent");
  63 MODULE_LICENSE("Dual BSD/GPL");
  64
  65 #define CMA_CM_RESPONSE_TIMEOUT 20
  66 #define CMA_MAX_CM_RETRIES 15
  67 #define CMA_CM_MRA_SETTING (IB_CM_MRA_FLAG_DELAY | 24)
  68 #define CMA_IBOE_PACKET_LIFETIME 18
  69
  70 static const char * const cma_events[] = {
  71         [RDMA_CM_EVENT_ADDR_RESOLVED]    = "address resolved",
  72         [RDMA_CM_EVENT_ADDR_ERROR]       = "address error",
  73         [RDMA_CM_EVENT_ROUTE_RESOLVED]   = "route resolved ",
  74         [RDMA_CM_EVENT_ROUTE_ERROR]      = "route error",
  75         [RDMA_CM_EVENT_CONNECT_REQUEST]  = "connect request",
  76         [RDMA_CM_EVENT_CONNECT_RESPONSE] = "connect response",
  77         [RDMA_CM_EVENT_CONNECT_ERROR]    = "connect error",
  78         [RDMA_CM_EVENT_UNREACHABLE]      = "unreachable",
  79         [RDMA_CM_EVENT_REJECTED]         = "rejected",
  80         [RDMA_CM_EVENT_ESTABLISHED]      = "established",
  81         [RDMA_CM_EVENT_DISCONNECTED]     = "disconnected",
  82         [RDMA_CM_EVENT_DEVICE_REMOVAL]   = "device removal",
  83         [RDMA_CM_EVENT_MULTICAST_JOIN]   = "multicast join",
  84         [RDMA_CM_EVENT_MULTICAST_ERROR]  = "multicast error",
  85         [RDMA_CM_EVENT_ADDR_CHANGE]      = "address change",
  86         [RDMA_CM_EVENT_TIMEWAIT_EXIT]    = "timewait exit",
  87 };
  88
  89 const char *rdma_event_msg(enum rdma_cm_event_type event)
  90 {
  91         size_t index = event;
  92
  93         return (index < ARRAY_SIZE(cma_events) && cma_events[index]) ?
  94                         cma_events[index] : "unrecognized event";
  95 }
  96 EXPORT_SYMBOL(rdma_event_msg);
  97
  98 static void cma_add_one(struct ib_device *device);
  99 static void cma_remove_one(struct ib_device *device, void *client_data);
 100
 101 static struct ib_client cma_client = {
 102         .name   = "cma",
 103         .add    = cma_add_one,
 104         .remove = cma_remove_one
 105 };
 106
 107 static struct ib_sa_client sa_client;
 108 static struct rdma_addr_client addr_client;
 109 static LIST_HEAD(dev_list);
 110 static LIST_HEAD(listen_any_list);
 111 static DEFINE_MUTEX(lock);
 112 static struct workqueue_struct *cma_wq;
 113 static DEFINE_IDR(tcp_ps);
 114 static DEFINE_IDR(udp_ps);
 115 static DEFINE_IDR(ipoib_ps);
 116 static DEFINE_IDR(ib_ps);
 117
 118 static struct idr *cma_idr(enum rdma_port_space ps)
 119 {
 120         switch (ps) {
 121         case RDMA_PS_TCP:
 122                 return &tcp_ps;
 123         case RDMA_PS_UDP:
 124                 return &udp_ps;
 125         case RDMA_PS_IPOIB:
 126                 return &ipoib_ps;
 127         case RDMA_PS_IB:
 128                 return &ib_ps;
 129         default:
 130                 return NULL;
 131         }
 132 }
 133
 134 struct cma_device {
 135         struct list_head        list;
 136         struct ib_device        *device;
 137         struct completion       comp;
 138         atomic_t                refcount;
 139         struct list_head        id_list;
 140 };
 141
 142 struct rdma_bind_list {
 143         enum rdma_port_space    ps;
 144         struct hlist_head       owners;
 145         unsigned short          port;
 146 };
 147
 148 static int cma_ps_alloc(enum rdma_port_space ps,
 149                         struct rdma_bind_list *bind_list, int snum)
 150 {
 151         struct idr *idr = cma_idr(ps);
 152
 153         return idr_alloc(idr, bind_list, snum, snum + 1, GFP_KERNEL);
 154 }
 155
 156 static struct rdma_bind_list *cma_ps_find(enum rdma_port_space ps, int snum)
 157 {
 158         struct idr *idr = cma_idr(ps);
 159
 160         return idr_find(idr, snum);
 161 }
 162
 163 static void cma_ps_remove(enum rdma_port_space ps, int snum)
 164 {
 165         struct idr *idr = cma_idr(ps);
 166
 167         idr_remove(idr, snum);
 168 }
 169
 170 enum {
 171         CMA_OPTION_AFONLY,
 172 };
 173
 174 /*
 175  * Device removal can occur at anytime, so we need extra handling to
 176  * serialize notifying the user of device removal with other callbacks.
 177  * We do this by disabling removal notification while a callback is in process,
 178  * and reporting it after the callback completes.
 179  */
 180 struct rdma_id_private {
 181         struct rdma_cm_id       id;
 182
 183         struct rdma_bind_list   *bind_list;
 184         struct hlist_node       node;
 185         struct list_head        list; /* listen_any_list or cma_device.list */
 186         struct list_head        listen_list; /* per device listens */
 187         struct cma_device       *cma_dev;
 188         struct list_head        mc_list;
 189
 190         int                     internal_id;
 191         enum rdma_cm_state      state;
 192         spinlock_t              lock;
 193         struct mutex            qp_mutex;
 194
 195         struct completion       comp;
 196         atomic_t                refcount;
 197         struct mutex            handler_mutex;
 198
 199         int                     backlog;
 200         int                     timeout_ms;
 201         struct ib_sa_query      *query;
 202         int                     query_id;
 203         union {
 204                 struct ib_cm_id *ib;
 205                 struct iw_cm_id *iw;
 206         } cm_id;
 207
 208         u32                     seq_num;
 209         u32                     qkey;
 210         u32                     qp_num;
 211         pid_t                   owner;
 212         u32                     options;
 213         u8                      srq;
 214         u8                      tos;
 215         u8                      reuseaddr;
 216         u8                      afonly;
 217 };
 218
 219 struct cma_multicast {
 220         struct rdma_id_private *id_priv;
 221         union {
 222                 struct ib_sa_multicast *ib;
 223         } multicast;
 224         struct list_head        list;
 225         void                    *context;
 226         struct sockaddr_storage addr;
 227         struct kref             mcref;
 228 };
 229
 230 struct cma_work {
 231         struct work_struct      work;
 232         struct rdma_id_private  *id;
 233         enum rdma_cm_state      old_state;
 234         enum rdma_cm_state      new_state;
 235         struct rdma_cm_event    event;
 236 };
 237
 238 struct cma_ndev_work {
 239         struct work_struct      work;
 240         struct rdma_id_private  *id;
 241         struct rdma_cm_event    event;
 242 };
 243
 244 struct iboe_mcast_work {
 245         struct work_struct       work;
 246         struct rdma_id_private  *id;
 247         struct cma_multicast    *mc;
 248 };
 249
 250 union cma_ip_addr {
 251         struct in6_addr ip6;
 252         struct {
 253                 __be32 pad[3];
 254                 __be32 addr;
 255         } ip4;
 256 };
 257
 258 struct cma_hdr {
 259         u8 cma_version;
 260         u8 ip_version;  /* IP version: 7:4 */
 261         __be16 port;
 262         union cma_ip_addr src_addr;
 263         union cma_ip_addr dst_addr;
 264 };
 265
 266 #define CMA_VERSION 0x00
 267
 268 struct cma_req_info {
 269         struct ib_device *device;
 270         int port;
 271         union ib_gid local_gid;
 272         __be64 service_id;
 273         u16 pkey;
 274         bool has_gid:1;
 275 };
 276
 277 static int cma_comp(struct rdma_id_private *id_priv, enum rdma_cm_state comp)
 278 {
 279         unsigned long flags;
 280         int ret;
 281
 282         spin_lock_irqsave(&id_priv->lock, flags);
 283         ret = (id_priv->state == comp);
 284         spin_unlock_irqrestore(&id_priv->lock, flags);
 285         return ret;
 286 }
 287
 288 static int cma_comp_exch(struct rdma_id_private *id_priv,
 289                          enum rdma_cm_state comp, enum rdma_cm_state exch)
 290 {
 291         unsigned long flags;
 292         int ret;
 293
 294         spin_lock_irqsave(&id_priv->lock, flags);
 295         if ((ret = (id_priv->state == comp)))
 296                 id_priv->state = exch;
 297         spin_unlock_irqrestore(&id_priv->lock, flags);
 298         return ret;
 299 }
 300
 301 static enum rdma_cm_state cma_exch(struct rdma_id_private *id_priv,
 302                                    enum rdma_cm_state exch)
 303 {
 304         unsigned long flags;
 305         enum rdma_cm_state old;
 306
 307         spin_lock_irqsave(&id_priv->lock, flags);
 308         old = id_priv->state;
 309         id_priv->state = exch;
 310         spin_unlock_irqrestore(&id_priv->lock, flags);
 311         return old;
 312 }
 313
 314 static inline u8 cma_get_ip_ver(const struct cma_hdr *hdr)
 315 {
 316         return hdr->ip_version >> 4;
 317 }
 318
 319 static inline void cma_set_ip_ver(struct cma_hdr *hdr, u8 ip_ver)
 320 {
 321         hdr->ip_version = (ip_ver << 4) | (hdr->ip_version & 0xF);
 322 }
 323
 324 static void cma_attach_to_dev(struct rdma_id_private *id_priv,
 325                               struct cma_device *cma_dev)
 326 {
 327         atomic_inc(&cma_dev->refcount);
 328         id_priv->cma_dev = cma_dev;
 329         id_priv->id.device = cma_dev->device;
 330         id_priv->id.route.addr.dev_addr.transport =
 331                 rdma_node_get_transport(cma_dev->device->node_type);
 332         list_add_tail(&id_priv->list, &cma_dev->id_list);
 333 }
 334
 335 static inline void cma_deref_dev(struct cma_device *cma_dev)
 336 {
 337         if (atomic_dec_and_test(&cma_dev->refcount))
 338                 complete(&cma_dev->comp);
 339 }
 340
 341 static inline void release_mc(struct kref *kref)
 342 {
 343         struct cma_multicast *mc = container_of(kref, struct cma_multicast, mcref);
 344
 345         kfree(mc->multicast.ib);
 346         kfree(mc);
 347 }
 348
 349 static void cma_release_dev(struct rdma_id_private *id_priv)
 350 {
 351         mutex_lock(&lock);
 352         list_del(&id_priv->list);
 353         cma_deref_dev(id_priv->cma_dev);
 354         id_priv->cma_dev = NULL;
 355         mutex_unlock(&lock);
 356 }
 357
 358 static inline struct sockaddr *cma_src_addr(struct rdma_id_private *id_priv)
 359 {
 360         return (struct sockaddr *) &id_priv->id.route.addr.src_addr;
 361 }
 362
 363 static inline struct sockaddr *cma_dst_addr(struct rdma_id_private *id_priv)
 364 {
 365         return (struct sockaddr *) &id_priv->id.route.addr.dst_addr;
 366 }
 367
 368 static inline unsigned short cma_family(struct rdma_id_private *id_priv)
 369 {
 370         return id_priv->id.route.addr.src_addr.ss_family;
 371 }
 372
 373 static int cma_set_qkey(struct rdma_id_private *id_priv, u32 qkey)
 374 {
 375         struct ib_sa_mcmember_rec rec;
 376         int ret = 0;
 377
 378         if (id_priv->qkey) {
 379                 if (qkey && id_priv->qkey != qkey)
 380                         return -EINVAL;
 381                 return 0;
 382         }
 383
 384         if (qkey) {
 385                 id_priv->qkey = qkey;
 386                 return 0;
 387         }
 388
 389         switch (id_priv->id.ps) {
 390         case RDMA_PS_UDP:
 391         case RDMA_PS_IB:
 392                 id_priv->qkey = RDMA_UDP_QKEY;
 393                 break;
 394         case RDMA_PS_IPOIB:
 395                 ib_addr_get_mgid(&id_priv->id.route.addr.dev_addr, &rec.mgid);
 396                 ret = ib_sa_get_mcmember_rec(id_priv->id.device,
 397                                              id_priv->id.port_num, &rec.mgid,
 398                                              &rec);
 399                 if (!ret)
 400                         id_priv->qkey = be32_to_cpu(rec.qkey);
 401                 break;
 402         default:
 403                 break;
 404         }
 405         return ret;
 406 }
 407
 408 static void cma_translate_ib(struct sockaddr_ib *sib, struct rdma_dev_addr *dev_addr)
 409 {
 410         dev_addr->dev_type = ARPHRD_INFINIBAND;
 411         rdma_addr_set_sgid(dev_addr, (union ib_gid *) &sib->sib_addr);
 412         ib_addr_set_pkey(dev_addr, ntohs(sib->sib_pkey));
 413 }
 414
 415 static int cma_translate_addr(struct sockaddr *addr, struct rdma_dev_addr *dev_addr)
 416 {
 417         int ret;
 418
 419         if (addr->sa_family != AF_IB) {
 420                 ret = rdma_translate_ip(addr, dev_addr, NULL);
 421         } else {
 422                 cma_translate_ib((struct sockaddr_ib *) addr, dev_addr);
 423                 ret = 0;
 424         }
 425
 426         return ret;
 427 }
 428
 429 static inline int cma_validate_port(struct ib_device *device, u8 port,
 430                                       union ib_gid *gid, int dev_type)
 431 {
 432         u8 found_port;
 433         int ret = -ENODEV;
 434
 435         if ((dev_type == ARPHRD_INFINIBAND) && !rdma_protocol_ib(device, port))
 436                 return ret;
 437
 438         if ((dev_type != ARPHRD_INFINIBAND) && rdma_protocol_ib(device, port))
 439                 return ret;
 440
 441         ret = ib_find_cached_gid(device, gid, &found_port, NULL);
 442         if (port != found_port)
 443                 return -ENODEV;
 444
 445         return ret;
 446 }
 447
 448 static int cma_acquire_dev(struct rdma_id_private *id_priv,
 449                            struct rdma_id_private *listen_id_priv)
 450 {
 451         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
 452         struct cma_device *cma_dev;
 453         union ib_gid gid, iboe_gid, *gidp;
 454         int ret = -ENODEV;
 455         u8 port;
 456
 457         if (dev_addr->dev_type != ARPHRD_INFINIBAND &&
 458             id_priv->id.ps == RDMA_PS_IPOIB)
 459                 return -EINVAL;
 460
 461         mutex_lock(&lock);
 462         rdma_ip2gid((struct sockaddr *)&id_priv->id.route.addr.src_addr,
 463                     &iboe_gid);
 464
 465         memcpy(&gid, dev_addr->src_dev_addr +
 466                rdma_addr_gid_offset(dev_addr), sizeof gid);
 467
 468         if (listen_id_priv) {
 469                 cma_dev = listen_id_priv->cma_dev;
 470                 port = listen_id_priv->id.port_num;
 471                 gidp = rdma_protocol_roce(cma_dev->device, port) ?
 472                        &iboe_gid : &gid;
 473
 474                 ret = cma_validate_port(cma_dev->device, port, gidp,
 475                                         dev_addr->dev_type);
 476                 if (!ret) {
 477                         id_priv->id.port_num = port;
 478                         goto out;
 479                 }
 480         }
 481
 482         list_for_each_entry(cma_dev, &dev_list, list) {
 483                 for (port = 1; port <= cma_dev->device->phys_port_cnt; ++port) {
 484                         if (listen_id_priv &&
 485                             listen_id_priv->cma_dev == cma_dev &&
 486                             listen_id_priv->id.port_num == port)
 487                                 continue;
 488
 489                         gidp = rdma_protocol_roce(cma_dev->device, port) ?
 490                                &iboe_gid : &gid;
 491
 492                         ret = cma_validate_port(cma_dev->device, port, gidp,
 493                                                 dev_addr->dev_type);
 494                         if (!ret) {
 495                                 id_priv->id.port_num = port;
 496                                 goto out;
 497                         }
 498                 }
 499         }
 500
 501 out:
 502         if (!ret)
 503                 cma_attach_to_dev(id_priv, cma_dev);
 504
 505         mutex_unlock(&lock);
 506         return ret;
 507 }
 508
 509 /*
 510  * Select the source IB device and address to reach the destination IB address.
 511  */
 512 static int cma_resolve_ib_dev(struct rdma_id_private *id_priv)
 513 {
 514         struct cma_device *cma_dev, *cur_dev;
 515         struct sockaddr_ib *addr;
 516         union ib_gid gid, sgid, *dgid;
 517         u16 pkey, index;
 518         u8 p;
 519         int i;
 520
 521         cma_dev = NULL;
 522         addr = (struct sockaddr_ib *) cma_dst_addr(id_priv);
 523         dgid = (union ib_gid *) &addr->sib_addr;
 524         pkey = ntohs(addr->sib_pkey);
 525
 526         list_for_each_entry(cur_dev, &dev_list, list) {
 527                 for (p = 1; p <= cur_dev->device->phys_port_cnt; ++p) {
 528                         if (!rdma_cap_af_ib(cur_dev->device, p))
 529                                 continue;
 530
 531                         if (ib_find_cached_pkey(cur_dev->device, p, pkey, &index))
 532                                 continue;
 533
 534                         for (i = 0; !ib_get_cached_gid(cur_dev->device, p, i, &gid); i++) {
 535                                 if (!memcmp(&gid, dgid, sizeof(gid))) {
 536                                         cma_dev = cur_dev;
 537                                         sgid = gid;
 538                                         id_priv->id.port_num = p;
 539                                         goto found;
 540                                 }
 541
 542                                 if (!cma_dev && (gid.global.subnet_prefix ==
 543                                                  dgid->global.subnet_prefix)) {
 544                                         cma_dev = cur_dev;
 545                                         sgid = gid;
 546                                         id_priv->id.port_num = p;
 547                                 }
 548                         }
 549                 }
 550         }
 551
 552         if (!cma_dev)
 553                 return -ENODEV;
 554
 555 found:
 556         cma_attach_to_dev(id_priv, cma_dev);
 557         addr = (struct sockaddr_ib *) cma_src_addr(id_priv);
 558         memcpy(&addr->sib_addr, &sgid, sizeof sgid);
 559         cma_translate_ib(addr, &id_priv->id.route.addr.dev_addr);
 560         return 0;
 561 }
 562
 563 static void cma_deref_id(struct rdma_id_private *id_priv)
 564 {
 565         if (atomic_dec_and_test(&id_priv->refcount))
 566                 complete(&id_priv->comp);
 567 }
 568
 569 static int cma_disable_callback(struct rdma_id_private *id_priv,
 570                                 enum rdma_cm_state state)
 571 {
 572         mutex_lock(&id_priv->handler_mutex);
 573         if (id_priv->state != state) {
 574                 mutex_unlock(&id_priv->handler_mutex);
 575                 return -EINVAL;
 576         }
 577         return 0;
 578 }
 579
 580 struct rdma_cm_id *rdma_create_id(rdma_cm_event_handler event_handler,
 581                                   void *context, enum rdma_port_space ps,
 582                                   enum ib_qp_type qp_type)
 583 {
 584         struct rdma_id_private *id_priv;
 585
 586         id_priv = kzalloc(sizeof *id_priv, GFP_KERNEL);
 587         if (!id_priv)
 588                 return ERR_PTR(-ENOMEM);
 589
 590         id_priv->owner = task_pid_nr(current);
 591         id_priv->state = RDMA_CM_IDLE;
 592         id_priv->id.context = context;
 593         id_priv->id.event_handler = event_handler;
 594         id_priv->id.ps = ps;
 595         id_priv->id.qp_type = qp_type;
 596         spin_lock_init(&id_priv->lock);
 597         mutex_init(&id_priv->qp_mutex);
 598         init_completion(&id_priv->comp);
 599         atomic_set(&id_priv->refcount, 1);
 600         mutex_init(&id_priv->handler_mutex);
 601         INIT_LIST_HEAD(&id_priv->listen_list);
 602         INIT_LIST_HEAD(&id_priv->mc_list);
 603         get_random_bytes(&id_priv->seq_num, sizeof id_priv->seq_num);
 604
 605         return &id_priv->id;
 606 }
 607 EXPORT_SYMBOL(rdma_create_id);
 608
 609 static int cma_init_ud_qp(struct rdma_id_private *id_priv, struct ib_qp *qp)
 610 {
 611         struct ib_qp_attr qp_attr;
 612         int qp_attr_mask, ret;
 613
 614         qp_attr.qp_state = IB_QPS_INIT;
 615         ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
 616         if (ret)
 617                 return ret;
 618
 619         ret = ib_modify_qp(qp, &qp_attr, qp_attr_mask);
 620         if (ret)
 621                 return ret;
 622
 623         qp_attr.qp_state = IB_QPS_RTR;
 624         ret = ib_modify_qp(qp, &qp_attr, IB_QP_STATE);
 625         if (ret)
 626                 return ret;
 627
 628         qp_attr.qp_state = IB_QPS_RTS;
 629         qp_attr.sq_psn = 0;
 630         ret = ib_modify_qp(qp, &qp_attr, IB_QP_STATE | IB_QP_SQ_PSN);
 631
 632         return ret;
 633 }
 634
 635 static int cma_init_conn_qp(struct rdma_id_private *id_priv, struct ib_qp *qp)
 636 {
 637         struct ib_qp_attr qp_attr;
 638         int qp_attr_mask, ret;
 639
 640         qp_attr.qp_state = IB_QPS_INIT;
 641         ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
 642         if (ret)
 643                 return ret;
 644
 645         return ib_modify_qp(qp, &qp_attr, qp_attr_mask);
 646 }
 647
 648 int rdma_create_qp(struct rdma_cm_id *id, struct ib_pd *pd,
 649                    struct ib_qp_init_attr *qp_init_attr)
 650 {
 651         struct rdma_id_private *id_priv;
 652         struct ib_qp *qp;
 653         int ret;
 654
 655         id_priv = container_of(id, struct rdma_id_private, id);
 656         if (id->device != pd->device)
 657                 return -EINVAL;
 658
 659         qp = ib_create_qp(pd, qp_init_attr);
 660         if (IS_ERR(qp))
 661                 return PTR_ERR(qp);
 662
 663         if (id->qp_type == IB_QPT_UD)
 664                 ret = cma_init_ud_qp(id_priv, qp);
 665         else
 666                 ret = cma_init_conn_qp(id_priv, qp);
 667         if (ret)
 668                 goto err;
 669
 670         id->qp = qp;
 671         id_priv->qp_num = qp->qp_num;
 672         id_priv->srq = (qp->srq != NULL);
 673         return 0;
 674 err:
 675         ib_destroy_qp(qp);
 676         return ret;
 677 }
 678 EXPORT_SYMBOL(rdma_create_qp);
 679
 680 void rdma_destroy_qp(struct rdma_cm_id *id)
 681 {
 682         struct rdma_id_private *id_priv;
 683
 684         id_priv = container_of(id, struct rdma_id_private, id);
 685         mutex_lock(&id_priv->qp_mutex);
 686         ib_destroy_qp(id_priv->id.qp);
 687         id_priv->id.qp = NULL;
 688         mutex_unlock(&id_priv->qp_mutex);
 689 }
 690 EXPORT_SYMBOL(rdma_destroy_qp);
 691
 692 static int cma_modify_qp_rtr(struct rdma_id_private *id_priv,
 693                              struct rdma_conn_param *conn_param)
 694 {
 695         struct ib_qp_attr qp_attr;
 696         int qp_attr_mask, ret;
 697         union ib_gid sgid;
 698
 699         mutex_lock(&id_priv->qp_mutex);
 700         if (!id_priv->id.qp) {
 701                 ret = 0;
 702                 goto out;
 703         }
 704
 705         /* Need to update QP attributes from default values. */
 706         qp_attr.qp_state = IB_QPS_INIT;
 707         ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
 708         if (ret)
 709                 goto out;
 710
 711         ret = ib_modify_qp(id_priv->id.qp, &qp_attr, qp_attr_mask);
 712         if (ret)
 713                 goto out;
 714
 715         qp_attr.qp_state = IB_QPS_RTR;
 716         ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
 717         if (ret)
 718                 goto out;
 719
 720         ret = ib_query_gid(id_priv->id.device, id_priv->id.port_num,
 721                            qp_attr.ah_attr.grh.sgid_index, &sgid);
 722         if (ret)
 723                 goto out;
 724
 725         BUG_ON(id_priv->cma_dev->device != id_priv->id.device);
 726
 727         if (rdma_protocol_roce(id_priv->id.device, id_priv->id.port_num)) {
 728                 ret = rdma_addr_find_smac_by_sgid(&sgid, qp_attr.smac, NULL);
 729
 730                 if (ret)
 731                         goto out;
 732         }
 733         if (conn_param)
 734                 qp_attr.max_dest_rd_atomic = conn_param->responder_resources;
 735         ret = ib_modify_qp(id_priv->id.qp, &qp_attr, qp_attr_mask);
 736 out:
 737         mutex_unlock(&id_priv->qp_mutex);
 738         return ret;
 739 }
 740
 741 static int cma_modify_qp_rts(struct rdma_id_private *id_priv,
 742                              struct rdma_conn_param *conn_param)
 743 {
 744         struct ib_qp_attr qp_attr;
 745         int qp_attr_mask, ret;
 746
 747         mutex_lock(&id_priv->qp_mutex);
 748         if (!id_priv->id.qp) {
 749                 ret = 0;
 750                 goto out;
 751         }
 752
 753         qp_attr.qp_state = IB_QPS_RTS;
 754         ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
 755         if (ret)
 756                 goto out;
 757
 758         if (conn_param)
 759                 qp_attr.max_rd_atomic = conn_param->initiator_depth;
 760         ret = ib_modify_qp(id_priv->id.qp, &qp_attr, qp_attr_mask);
 761 out:
 762         mutex_unlock(&id_priv->qp_mutex);
 763         return ret;
 764 }
 765
 766 static int cma_modify_qp_err(struct rdma_id_private *id_priv)
 767 {
 768         struct ib_qp_attr qp_attr;
 769         int ret;
 770
 771         mutex_lock(&id_priv->qp_mutex);
 772         if (!id_priv->id.qp) {
 773                 ret = 0;
 774                 goto out;
 775         }
 776
 777         qp_attr.qp_state = IB_QPS_ERR;
 778         ret = ib_modify_qp(id_priv->id.qp, &qp_attr, IB_QP_STATE);
 779 out:
 780         mutex_unlock(&id_priv->qp_mutex);
 781         return ret;
 782 }
 783
 784 static int cma_ib_init_qp_attr(struct rdma_id_private *id_priv,
 785                                struct ib_qp_attr *qp_attr, int *qp_attr_mask)
 786 {
 787         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
 788         int ret;
 789         u16 pkey;
 790
 791         if (rdma_cap_eth_ah(id_priv->id.device, id_priv->id.port_num))
 792                 pkey = 0xffff;
 793         else
 794                 pkey = ib_addr_get_pkey(dev_addr);
 795
 796         ret = ib_find_cached_pkey(id_priv->id.device, id_priv->id.port_num,
 797                                   pkey, &qp_attr->pkey_index);
 798         if (ret)
 799                 return ret;
 800
 801         qp_attr->port_num = id_priv->id.port_num;
 802         *qp_attr_mask = IB_QP_STATE | IB_QP_PKEY_INDEX | IB_QP_PORT;
 803
 804         if (id_priv->id.qp_type == IB_QPT_UD) {
 805                 ret = cma_set_qkey(id_priv, 0);
 806                 if (ret)
 807                         return ret;
 808
 809                 qp_attr->qkey = id_priv->qkey;
 810                 *qp_attr_mask |= IB_QP_QKEY;
 811         } else {
 812                 qp_attr->qp_access_flags = 0;
 813                 *qp_attr_mask |= IB_QP_ACCESS_FLAGS;
 814         }
 815         return 0;
 816 }
 817
 818 int rdma_init_qp_attr(struct rdma_cm_id *id, struct ib_qp_attr *qp_attr,
 819                        int *qp_attr_mask)
 820 {
 821         struct rdma_id_private *id_priv;
 822         int ret = 0;
 823
 824         id_priv = container_of(id, struct rdma_id_private, id);
 825         if (rdma_cap_ib_cm(id->device, id->port_num)) {
 826                 if (!id_priv->cm_id.ib || (id_priv->id.qp_type == IB_QPT_UD))
 827                         ret = cma_ib_init_qp_attr(id_priv, qp_attr, qp_attr_mask);
 828                 else
 829                         ret = ib_cm_init_qp_attr(id_priv->cm_id.ib, qp_attr,
 830                                                  qp_attr_mask);
 831
 832                 if (qp_attr->qp_state == IB_QPS_RTR)
 833                         qp_attr->rq_psn = id_priv->seq_num;
 834         } else if (rdma_cap_iw_cm(id->device, id->port_num)) {
 835                 if (!id_priv->cm_id.iw) {
 836                         qp_attr->qp_access_flags = 0;
 837                         *qp_attr_mask = IB_QP_STATE | IB_QP_ACCESS_FLAGS;
 838                 } else
 839                         ret = iw_cm_init_qp_attr(id_priv->cm_id.iw, qp_attr,
 840                                                  qp_attr_mask);
 841         } else
 842                 ret = -ENOSYS;
 843
 844         return ret;
 845 }
 846 EXPORT_SYMBOL(rdma_init_qp_attr);
 847
 848 static inline int cma_zero_addr(struct sockaddr *addr)
 849 {
 850         switch (addr->sa_family) {
 851         case AF_INET:
 852                 return ipv4_is_zeronet(((struct sockaddr_in *)addr)->sin_addr.s_addr);
 853         case AF_INET6:
 854                 return ipv6_addr_any(&((struct sockaddr_in6 *) addr)->sin6_addr);
 855         case AF_IB:
 856                 return ib_addr_any(&((struct sockaddr_ib *) addr)->sib_addr);
 857         default:
 858                 return 0;
 859         }
 860 }
 861
 862 static inline int cma_loopback_addr(struct sockaddr *addr)
 863 {
 864         switch (addr->sa_family) {
 865         case AF_INET:
 866                 return ipv4_is_loopback(((struct sockaddr_in *) addr)->sin_addr.s_addr);
 867         case AF_INET6:
 868                 return ipv6_addr_loopback(&((struct sockaddr_in6 *) addr)->sin6_addr);
 869         case AF_IB:
 870                 return ib_addr_loopback(&((struct sockaddr_ib *) addr)->sib_addr);
 871         default:
 872                 return 0;
 873         }
 874 }
 875
 876 static inline int cma_any_addr(struct sockaddr *addr)
 877 {
 878         return cma_zero_addr(addr) || cma_loopback_addr(addr);
 879 }
 880
 881 static int cma_addr_cmp(struct sockaddr *src, struct sockaddr *dst)
 882 {
 883         if (src->sa_family != dst->sa_family)
 884                 return -1;
 885
 886         switch (src->sa_family) {
 887         case AF_INET:
 888                 return ((struct sockaddr_in *) src)->sin_addr.s_addr !=
 889                        ((struct sockaddr_in *) dst)->sin_addr.s_addr;
 890         case AF_INET6:
 891                 return ipv6_addr_cmp(&((struct sockaddr_in6 *) src)->sin6_addr,
 892                                      &((struct sockaddr_in6 *) dst)->sin6_addr);
 893         default:
 894                 return ib_addr_cmp(&((struct sockaddr_ib *) src)->sib_addr,
 895                                    &((struct sockaddr_ib *) dst)->sib_addr);
 896         }
 897 }
 898
 899 static __be16 cma_port(struct sockaddr *addr)
 900 {
 901         struct sockaddr_ib *sib;
 902
 903         switch (addr->sa_family) {
 904         case AF_INET:
 905                 return ((struct sockaddr_in *) addr)->sin_port;
 906         case AF_INET6:
 907                 return ((struct sockaddr_in6 *) addr)->sin6_port;
 908         case AF_IB:
 909                 sib = (struct sockaddr_ib *) addr;
 910                 return htons((u16) (be64_to_cpu(sib->sib_sid) &
 911                                     be64_to_cpu(sib->sib_sid_mask)));
 912         default:
 913                 return 0;
 914         }
 915 }
 916
 917 static inline int cma_any_port(struct sockaddr *addr)
 918 {
 919         return !cma_port(addr);
 920 }
 921
 922 static void cma_save_ib_info(struct sockaddr *src_addr,
 923                              struct sockaddr *dst_addr,
 924                              struct rdma_cm_id *listen_id,
 925                              struct ib_sa_path_rec *path)
 926 {
 927         struct sockaddr_ib *listen_ib, *ib;
 928
 929         listen_ib = (struct sockaddr_ib *) &listen_id->route.addr.src_addr;
 930         if (src_addr) {
 931                 ib = (struct sockaddr_ib *)src_addr;
 932                 ib->sib_family = AF_IB;
 933                 if (path) {
 934                         ib->sib_pkey = path->pkey;
 935                         ib->sib_flowinfo = path->flow_label;
 936                         memcpy(&ib->sib_addr, &path->sgid, 16);
 937                         ib->sib_sid = path->service_id;
 938                         ib->sib_scope_id = 0;
 939                 } else {
 940                         ib->sib_pkey = listen_ib->sib_pkey;
 941                         ib->sib_flowinfo = listen_ib->sib_flowinfo;
 942                         ib->sib_addr = listen_ib->sib_addr;
 943                         ib->sib_sid = listen_ib->sib_sid;
 944                         ib->sib_scope_id = listen_ib->sib_scope_id;
 945                 }
 946                 ib->sib_sid_mask = cpu_to_be64(0xffffffffffffffffULL);
 947         }
 948         if (dst_addr) {
 949                 ib = (struct sockaddr_ib *)dst_addr;
 950                 ib->sib_family = AF_IB;
 951                 if (path) {
 952                         ib->sib_pkey = path->pkey;
 953                         ib->sib_flowinfo = path->flow_label;
 954                         memcpy(&ib->sib_addr, &path->dgid, 16);
 955                 }
 956         }
 957 }
 958
 959 static void cma_save_ip4_info(struct sockaddr *src_addr,
 960                               struct sockaddr *dst_addr,
 961                               struct cma_hdr *hdr,
 962                               __be16 local_port)
 963 {
 964         struct sockaddr_in *ip4;
 965
 966         if (src_addr) {
 967                 ip4 = (struct sockaddr_in *)src_addr;
 968                 ip4->sin_family = AF_INET;
 969                 ip4->sin_addr.s_addr = hdr->dst_addr.ip4.addr;
 970                 ip4->sin_port = local_port;
 971         }
 972
 973         if (dst_addr) {
 974                 ip4 = (struct sockaddr_in *)dst_addr;
 975                 ip4->sin_family = AF_INET;
 976                 ip4->sin_addr.s_addr = hdr->src_addr.ip4.addr;
 977                 ip4->sin_port = hdr->port;
 978         }
 979 }
 980
 981 static void cma_save_ip6_info(struct sockaddr *src_addr,
 982                               struct sockaddr *dst_addr,
 983                               struct cma_hdr *hdr,
 984                               __be16 local_port)
 985 {
 986         struct sockaddr_in6 *ip6;
 987
 988         if (src_addr) {
 989                 ip6 = (struct sockaddr_in6 *)src_addr;
 990                 ip6->sin6_family = AF_INET6;
 991                 ip6->sin6_addr = hdr->dst_addr.ip6;
 992                 ip6->sin6_port = local_port;
 993         }
 994
 995         if (dst_addr) {
 996                 ip6 = (struct sockaddr_in6 *)dst_addr;
 997                 ip6->sin6_family = AF_INET6;
 998                 ip6->sin6_addr = hdr->src_addr.ip6;
 999                 ip6->sin6_port = hdr->port;
1000         }
1001 }
1002
1003 static u16 cma_port_from_service_id(__be64 service_id)
1004 {
1005         return (u16)be64_to_cpu(service_id);
1006 }
1007
1008 static int cma_save_ip_info(struct sockaddr *src_addr,
1009                             struct sockaddr *dst_addr,
1010                             struct ib_cm_event *ib_event,
1011                             __be64 service_id)
1012 {
1013         struct cma_hdr *hdr;
1014         __be16 port;
1015
1016         hdr = ib_event->private_data;
1017         if (hdr->cma_version != CMA_VERSION)
1018                 return -EINVAL;
1019
1020         port = htons(cma_port_from_service_id(service_id));
1021
1022         switch (cma_get_ip_ver(hdr)) {
1023         case 4:
1024                 cma_save_ip4_info(src_addr, dst_addr, hdr, port);
1025                 break;
1026         case 6:
1027                 cma_save_ip6_info(src_addr, dst_addr, hdr, port);
1028                 break;
1029         default:
1030                 return -EAFNOSUPPORT;
1031         }
1032
1033         return 0;
1034 }
1035
1036 static int cma_save_net_info(struct sockaddr *src_addr,
1037                              struct sockaddr *dst_addr,
1038                              struct rdma_cm_id *listen_id,
1039                              struct ib_cm_event *ib_event,
1040                              sa_family_t sa_family, __be64 service_id)
1041 {
1042         if (sa_family == AF_IB) {
1043                 if (ib_event->event == IB_CM_REQ_RECEIVED)
1044                         cma_save_ib_info(src_addr, dst_addr, listen_id,
1045                                          ib_event->param.req_rcvd.primary_path);
1046                 else if (ib_event->event == IB_CM_SIDR_REQ_RECEIVED)
1047                         cma_save_ib_info(src_addr, dst_addr, listen_id, NULL);
1048                 return 0;
1049         }
1050
1051         return cma_save_ip_info(src_addr, dst_addr, ib_event, service_id);
1052 }
1053
1054 static int cma_save_req_info(const struct ib_cm_event *ib_event,
1055                              struct cma_req_info *req)
1056 {
1057         const struct ib_cm_req_event_param *req_param =
1058                 &ib_event->param.req_rcvd;
1059         const struct ib_cm_sidr_req_event_param *sidr_param =
1060                 &ib_event->param.sidr_req_rcvd;
1061
1062         switch (ib_event->event) {
1063         case IB_CM_REQ_RECEIVED:
1064                 req->device     = req_param->listen_id->device;
1065                 req->port       = req_param->port;
1066                 memcpy(&req->local_gid, &req_param->primary_path->sgid,
1067                        sizeof(req->local_gid));
1068                 req->has_gid    = true;
1069                 req->service_id = req_param->primary_path->service_id;
1070                 req->pkey       = req_param->bth_pkey;
1071                 break;
1072         case IB_CM_SIDR_REQ_RECEIVED:
1073                 req->device     = sidr_param->listen_id->device;
1074                 req->port       = sidr_param->port;
1075                 req->has_gid    = false;
1076                 req->service_id = sidr_param->service_id;
1077                 req->pkey       = sidr_param->bth_pkey;
1078                 break;
1079         default:
1080                 return -EINVAL;
1081         }
1082
1083         return 0;
1084 }
1085
1086 static bool validate_ipv4_net_dev(struct net_device *net_dev,
1087                                   const struct sockaddr_in *dst_addr,
1088                                   const struct sockaddr_in *src_addr)
1089 {
1090         __be32 daddr = dst_addr->sin_addr.s_addr,
1091                saddr = src_addr->sin_addr.s_addr;
1092         struct fib_result res;
1093         struct flowi4 fl4;
1094         int err;
1095         bool ret;
1096
1097         if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1098             ipv4_is_lbcast(daddr) || ipv4_is_zeronet(saddr) ||
1099             ipv4_is_zeronet(daddr) || ipv4_is_loopback(daddr) ||
1100             ipv4_is_loopback(saddr))
1101                 return false;
1102
1103         memset(&fl4, 0, sizeof(fl4));
1104         fl4.flowi4_iif = net_dev->ifindex;
1105         fl4.daddr = daddr;
1106         fl4.saddr = saddr;
1107
1108         rcu_read_lock();
1109         err = fib_lookup(dev_net(net_dev), &fl4, &res, 0);
1110         if (err)
1111                 return false;
1112
1113         ret = FIB_RES_DEV(res) == net_dev;
1114         rcu_read_unlock();
1115
1116         return ret;
1117 }
1118
1119 static bool validate_ipv6_net_dev(struct net_device *net_dev,
1120                                   const struct sockaddr_in6 *dst_addr,
1121                                   const struct sockaddr_in6 *src_addr)
1122 {
1123 #if IS_ENABLED(CONFIG_IPV6)
1124         const int strict = ipv6_addr_type(&dst_addr->sin6_addr) &
1125                            IPV6_ADDR_LINKLOCAL;
1126         struct rt6_info *rt = rt6_lookup(dev_net(net_dev), &dst_addr->sin6_addr,
1127                                          &src_addr->sin6_addr, net_dev->ifindex,
1128                                          strict);
1129         bool ret;
1130
1131         if (!rt)
1132                 return false;
1133
1134         ret = rt->rt6i_idev->dev == net_dev;
1135         ip6_rt_put(rt);
1136
1137         return ret;
1138 #else
1139         return false;
1140 #endif
1141 }
1142
1143 static bool validate_net_dev(struct net_device *net_dev,
1144                              const struct sockaddr *daddr,
1145                              const struct sockaddr *saddr)
1146 {
1147         const struct sockaddr_in *daddr4 = (const struct sockaddr_in *)daddr;
1148         const struct sockaddr_in *saddr4 = (const struct sockaddr_in *)saddr;
1149         const struct sockaddr_in6 *daddr6 = (const struct sockaddr_in6 *)daddr;
1150         const struct sockaddr_in6 *saddr6 = (const struct sockaddr_in6 *)saddr;
1151
1152         switch (daddr->sa_family) {
1153         case AF_INET:
1154                 return saddr->sa_family == AF_INET &&
1155                        validate_ipv4_net_dev(net_dev, daddr4, saddr4);
1156
1157         case AF_INET6:
1158                 return saddr->sa_family == AF_INET6 &&
1159                        validate_ipv6_net_dev(net_dev, daddr6, saddr6);
1160
1161         default:
1162                 return false;
1163         }
1164 }
1165
1166 static struct net_device *cma_get_net_dev(struct ib_cm_event *ib_event,
1167                                           const struct cma_req_info *req)
1168 {
1169         struct sockaddr_storage listen_addr_storage, src_addr_storage;
1170         struct sockaddr *listen_addr = (struct sockaddr *)&listen_addr_storage,
1171                         *src_addr = (struct sockaddr *)&src_addr_storage;
1172         struct net_device *net_dev;
1173         const union ib_gid *gid = req->has_gid ? &req->local_gid : NULL;
1174         int err;
1175
1176         err = cma_save_ip_info(listen_addr, src_addr, ib_event,
1177                                req->service_id);
1178         if (err)
1179                 return ERR_PTR(err);
1180
1181         net_dev = ib_get_net_dev_by_params(req->device, req->port, req->pkey,
1182                                            gid, listen_addr);
1183         if (!net_dev)
1184                 return ERR_PTR(-ENODEV);
1185
1186         if (!validate_net_dev(net_dev, listen_addr, src_addr)) {
1187                 dev_put(net_dev);
1188                 return ERR_PTR(-EHOSTUNREACH);
1189         }
1190
1191         return net_dev;
1192 }
1193
1194 static enum rdma_port_space rdma_ps_from_service_id(__be64 service_id)
1195 {
1196         return (be64_to_cpu(service_id) >> 16) & 0xffff;
1197 }
1198
1199 static bool cma_match_private_data(struct rdma_id_private *id_priv,
1200                                    const struct cma_hdr *hdr)
1201 {
1202         struct sockaddr *addr = cma_src_addr(id_priv);
1203         __be32 ip4_addr;
1204         struct in6_addr ip6_addr;
1205
1206         if (cma_any_addr(addr) && !id_priv->afonly)
1207                 return true;
1208
1209         switch (addr->sa_family) {
1210         case AF_INET:
1211                 ip4_addr = ((struct sockaddr_in *)addr)->sin_addr.s_addr;
1212                 if (cma_get_ip_ver(hdr) != 4)
1213                         return false;
1214                 if (!cma_any_addr(addr) &&
1215                     hdr->dst_addr.ip4.addr != ip4_addr)
1216                         return false;
1217                 break;
1218         case AF_INET6:
1219                 ip6_addr = ((struct sockaddr_in6 *)addr)->sin6_addr;
1220                 if (cma_get_ip_ver(hdr) != 6)
1221                         return false;
1222                 if (!cma_any_addr(addr) &&
1223                     memcmp(&hdr->dst_addr.ip6, &ip6_addr, sizeof(ip6_addr)))
1224                         return false;
1225                 break;
1226         case AF_IB:
1227                 return true;
1228         default:
1229                 return false;
1230         }
1231
1232         return true;
1233 }
1234
1235 static bool cma_match_net_dev(const struct rdma_id_private *id_priv,
1236                               const struct net_device *net_dev)
1237 {
1238         const struct rdma_addr *addr = &id_priv->id.route.addr;
1239
1240         if (!net_dev)
1241                 /* This request is an AF_IB request */
1242                 return addr->src_addr.ss_family == AF_IB;
1243
1244         return !addr->dev_addr.bound_dev_if ||
1245                (net_eq(dev_net(net_dev), &init_net) &&
1246                 addr->dev_addr.bound_dev_if == net_dev->ifindex);
1247 }
1248
1249 static struct rdma_id_private *cma_find_listener(
1250                 const struct rdma_bind_list *bind_list,
1251                 const struct ib_cm_id *cm_id,
1252                 const struct ib_cm_event *ib_event,
1253                 const struct cma_req_info *req,
1254                 const struct net_device *net_dev)
1255 {
1256         struct rdma_id_private *id_priv, *id_priv_dev;
1257
1258         if (!bind_list)
1259                 return ERR_PTR(-EINVAL);
1260
1261         hlist_for_each_entry(id_priv, &bind_list->owners, node) {
1262                 if (cma_match_private_data(id_priv, ib_event->private_data)) {
1263                         if (id_priv->id.device == cm_id->device &&
1264                             cma_match_net_dev(id_priv, net_dev))
1265                                 return id_priv;
1266                         list_for_each_entry(id_priv_dev,
1267                                             &id_priv->listen_list,
1268                                             listen_list) {
1269                                 if (id_priv_dev->id.device == cm_id->device &&
1270                                     cma_match_net_dev(id_priv_dev, net_dev))
1271                                         return id_priv_dev;
1272                         }
1273                 }
1274         }
1275
1276         return ERR_PTR(-EINVAL);
1277 }
1278
1279 static struct rdma_id_private *cma_id_from_event(struct ib_cm_id *cm_id,
1280                                                  struct ib_cm_event *ib_event,
1281                                                  struct net_device **net_dev)
1282 {
1283         struct cma_req_info req;
1284         struct rdma_bind_list *bind_list;
1285         struct rdma_id_private *id_priv;
1286         int err;
1287
1288         err = cma_save_req_info(ib_event, &req);
1289         if (err)
1290                 return ERR_PTR(err);
1291
1292         *net_dev = cma_get_net_dev(ib_event, &req);
1293         if (IS_ERR(*net_dev)) {
1294                 if (PTR_ERR(*net_dev) == -EAFNOSUPPORT) {
1295                         /* Assuming the protocol is AF_IB */
1296                         *net_dev = NULL;
1297                 } else {
1298                         return ERR_CAST(*net_dev);
1299                 }
1300         }
1301
1302         bind_list = cma_ps_find(rdma_ps_from_service_id(req.service_id),
1303                                 cma_port_from_service_id(req.service_id));
1304         id_priv = cma_find_listener(bind_list, cm_id, ib_event, &req, *net_dev);
1305         if (IS_ERR(id_priv)) {
1306                 dev_put(*net_dev);
1307                 *net_dev = NULL;
1308         }
1309
1310         return id_priv;
1311 }
1312
1313 static inline int cma_user_data_offset(struct rdma_id_private *id_priv)
1314 {
1315         return cma_family(id_priv) == AF_IB ? 0 : sizeof(struct cma_hdr);
1316 }
1317
1318 static void cma_cancel_route(struct rdma_id_private *id_priv)
1319 {
1320         if (rdma_cap_ib_sa(id_priv->id.device, id_priv->id.port_num)) {
1321                 if (id_priv->query)
1322                         ib_sa_cancel_query(id_priv->query_id, id_priv->query);
1323         }
1324 }
1325
1326 static void cma_cancel_listens(struct rdma_id_private *id_priv)
1327 {
1328         struct rdma_id_private *dev_id_priv;
1329
1330         /*
1331          * Remove from listen_any_list to prevent added devices from spawning
1332          * additional listen requests.
1333          */
1334         mutex_lock(&lock);
1335         list_del(&id_priv->list);
1336
1337         while (!list_empty(&id_priv->listen_list)) {
1338                 dev_id_priv = list_entry(id_priv->listen_list.next,
1339                                          struct rdma_id_private, listen_list);
1340                 /* sync with device removal to avoid duplicate destruction */
1341                 list_del_init(&dev_id_priv->list);
1342                 list_del(&dev_id_priv->listen_list);
1343                 mutex_unlock(&lock);
1344
1345                 rdma_destroy_id(&dev_id_priv->id);
1346                 mutex_lock(&lock);
1347         }
1348         mutex_unlock(&lock);
1349 }
1350
1351 static void cma_cancel_operation(struct rdma_id_private *id_priv,
1352                                  enum rdma_cm_state state)
1353 {
1354         switch (state) {
1355         case RDMA_CM_ADDR_QUERY:
1356                 rdma_addr_cancel(&id_priv->id.route.addr.dev_addr);
1357                 break;
1358         case RDMA_CM_ROUTE_QUERY:
1359                 cma_cancel_route(id_priv);
1360                 break;
1361         case RDMA_CM_LISTEN:
1362                 if (cma_any_addr(cma_src_addr(id_priv)) && !id_priv->cma_dev)
1363                         cma_cancel_listens(id_priv);
1364                 break;
1365         default:
1366                 break;
1367         }
1368 }
1369
1370 static void cma_release_port(struct rdma_id_private *id_priv)
1371 {
1372         struct rdma_bind_list *bind_list = id_priv->bind_list;
1373
1374         if (!bind_list)
1375                 return;
1376
1377         mutex_lock(&lock);
1378         hlist_del(&id_priv->node);
1379         if (hlist_empty(&bind_list->owners)) {
1380                 cma_ps_remove(bind_list->ps, bind_list->port);
1381                 kfree(bind_list);
1382         }
1383         mutex_unlock(&lock);
1384 }
1385
1386 static void cma_leave_mc_groups(struct rdma_id_private *id_priv)
1387 {
1388         struct cma_multicast *mc;
1389
1390         while (!list_empty(&id_priv->mc_list)) {
1391                 mc = container_of(id_priv->mc_list.next,
1392                                   struct cma_multicast, list);
1393                 list_del(&mc->list);
1394                 if (rdma_cap_ib_mcast(id_priv->cma_dev->device,
1395                                       id_priv->id.port_num)) {
1396                         ib_sa_free_multicast(mc->multicast.ib);
1397                         kfree(mc);
1398                 } else
1399                         kref_put(&mc->mcref, release_mc);
1400         }
1401 }
1402
1403 void rdma_destroy_id(struct rdma_cm_id *id)
1404 {
1405         struct rdma_id_private *id_priv;
1406         enum rdma_cm_state state;
1407
1408         id_priv = container_of(id, struct rdma_id_private, id);
1409         state = cma_exch(id_priv, RDMA_CM_DESTROYING);
1410         cma_cancel_operation(id_priv, state);
1411
1412         /*
1413          * Wait for any active callback to finish.  New callbacks will find
1414          * the id_priv state set to destroying and abort.
1415          */
1416         mutex_lock(&id_priv->handler_mutex);
1417         mutex_unlock(&id_priv->handler_mutex);
1418
1419         if (id_priv->cma_dev) {
1420                 if (rdma_cap_ib_cm(id_priv->id.device, 1)) {
1421                         if (id_priv->cm_id.ib)
1422                                 ib_destroy_cm_id(id_priv->cm_id.ib);
1423                 } else if (rdma_cap_iw_cm(id_priv->id.device, 1)) {
1424                         if (id_priv->cm_id.iw)
1425                                 iw_destroy_cm_id(id_priv->cm_id.iw);
1426                 }
1427                 cma_leave_mc_groups(id_priv);
1428                 cma_release_dev(id_priv);
1429         }
1430
1431         cma_release_port(id_priv);
1432         cma_deref_id(id_priv);
1433         wait_for_completion(&id_priv->comp);
1434
1435         if (id_priv->internal_id)
1436                 cma_deref_id(id_priv->id.context);
1437
1438         kfree(id_priv->id.route.path_rec);
1439         kfree(id_priv);
1440 }
1441 EXPORT_SYMBOL(rdma_destroy_id);
1442
1443 static int cma_rep_recv(struct rdma_id_private *id_priv)
1444 {
1445         int ret;
1446
1447         ret = cma_modify_qp_rtr(id_priv, NULL);
1448         if (ret)
1449                 goto reject;
1450
1451         ret = cma_modify_qp_rts(id_priv, NULL);
1452         if (ret)
1453                 goto reject;
1454
1455         ret = ib_send_cm_rtu(id_priv->cm_id.ib, NULL, 0);
1456         if (ret)
1457                 goto reject;
1458
1459         return 0;
1460 reject:
1461         cma_modify_qp_err(id_priv);
1462         ib_send_cm_rej(id_priv->cm_id.ib, IB_CM_REJ_CONSUMER_DEFINED,
1463                        NULL, 0, NULL, 0);
1464         return ret;
1465 }
1466
1467 static void cma_set_rep_event_data(struct rdma_cm_event *event,
1468                                    struct ib_cm_rep_event_param *rep_data,
1469                                    void *private_data)
1470 {
1471         event->param.conn.private_data = private_data;
1472         event->param.conn.private_data_len = IB_CM_REP_PRIVATE_DATA_SIZE;
1473         event->param.conn.responder_resources = rep_data->responder_resources;
1474         event->param.conn.initiator_depth = rep_data->initiator_depth;
1475         event->param.conn.flow_control = rep_data->flow_control;
1476         event->param.conn.rnr_retry_count = rep_data->rnr_retry_count;
1477         event->param.conn.srq = rep_data->srq;
1478         event->param.conn.qp_num = rep_data->remote_qpn;
1479 }
1480
1481 static int cma_ib_handler(struct ib_cm_id *cm_id, struct ib_cm_event *ib_event)
1482 {
1483         struct rdma_id_private *id_priv = cm_id->context;
1484         struct rdma_cm_event event;
1485         int ret = 0;
1486
1487         if ((ib_event->event != IB_CM_TIMEWAIT_EXIT &&
1488                 cma_disable_callback(id_priv, RDMA_CM_CONNECT)) ||
1489             (ib_event->event == IB_CM_TIMEWAIT_EXIT &&
1490                 cma_disable_callback(id_priv, RDMA_CM_DISCONNECT)))
1491                 return 0;
1492
1493         memset(&event, 0, sizeof event);
1494         switch (ib_event->event) {
1495         case IB_CM_REQ_ERROR:
1496         case IB_CM_REP_ERROR:
1497                 event.event = RDMA_CM_EVENT_UNREACHABLE;
1498                 event.status = -ETIMEDOUT;
1499                 break;
1500         case IB_CM_REP_RECEIVED:
1501                 if (id_priv->id.qp) {
1502                         event.status = cma_rep_recv(id_priv);
1503                         event.event = event.status ? RDMA_CM_EVENT_CONNECT_ERROR :
1504                                                      RDMA_CM_EVENT_ESTABLISHED;
1505                 } else {
1506                         event.event = RDMA_CM_EVENT_CONNECT_RESPONSE;
1507                 }
1508                 cma_set_rep_event_data(&event, &ib_event->param.rep_rcvd,
1509                                        ib_event->private_data);
1510                 break;
1511         case IB_CM_RTU_RECEIVED:
1512         case IB_CM_USER_ESTABLISHED:
1513                 event.event = RDMA_CM_EVENT_ESTABLISHED;
1514                 break;
1515         case IB_CM_DREQ_ERROR:
1516                 event.status = -ETIMEDOUT; /* fall through */
1517         case IB_CM_DREQ_RECEIVED:
1518         case IB_CM_DREP_RECEIVED:
1519                 if (!cma_comp_exch(id_priv, RDMA_CM_CONNECT,
1520                                    RDMA_CM_DISCONNECT))
1521                         goto out;
1522                 event.event = RDMA_CM_EVENT_DISCONNECTED;
1523                 break;
1524         case IB_CM_TIMEWAIT_EXIT:
1525                 event.event = RDMA_CM_EVENT_TIMEWAIT_EXIT;
1526                 break;
1527         case IB_CM_MRA_RECEIVED:
1528                 /* ignore event */
1529                 goto out;
1530         case IB_CM_REJ_RECEIVED:
1531                 cma_modify_qp_err(id_priv);
1532                 event.status = ib_event->param.rej_rcvd.reason;
1533                 event.event = RDMA_CM_EVENT_REJECTED;
1534                 event.param.conn.private_data = ib_event->private_data;
1535                 event.param.conn.private_data_len = IB_CM_REJ_PRIVATE_DATA_SIZE;
1536                 break;
1537         default:
1538                 printk(KERN_ERR "RDMA CMA: unexpected IB CM event: %d\n",
1539                        ib_event->event);
1540                 goto out;
1541         }
1542
1543         ret = id_priv->id.event_handler(&id_priv->id, &event);
1544         if (ret) {
1545                 /* Destroy the CM ID by returning a non-zero value. */
1546                 id_priv->cm_id.ib = NULL;
1547                 cma_exch(id_priv, RDMA_CM_DESTROYING);
1548                 mutex_unlock(&id_priv->handler_mutex);
1549                 rdma_destroy_id(&id_priv->id);
1550                 return ret;
1551         }
1552 out:
1553         mutex_unlock(&id_priv->handler_mutex);
1554         return ret;
1555 }
1556
1557 static struct rdma_id_private *cma_new_conn_id(struct rdma_cm_id *listen_id,
1558                                                struct ib_cm_event *ib_event,
1559                                                struct net_device *net_dev)
1560 {
1561         struct rdma_id_private *id_priv;
1562         struct rdma_cm_id *id;
1563         struct rdma_route *rt;
1564         const sa_family_t ss_family = listen_id->route.addr.src_addr.ss_family;
1565         const __be64 service_id =
1566                       ib_event->param.req_rcvd.primary_path->service_id;
1567         int ret;
1568
1569         id = rdma_create_id(listen_id->event_handler, listen_id->context,
1570                             listen_id->ps, ib_event->param.req_rcvd.qp_type);
1571         if (IS_ERR(id))
1572                 return NULL;
1573
1574         id_priv = container_of(id, struct rdma_id_private, id);
1575         if (cma_save_net_info((struct sockaddr *)&id->route.addr.src_addr,
1576                               (struct sockaddr *)&id->route.addr.dst_addr,
1577                               listen_id, ib_event, ss_family, service_id))
1578                 goto err;
1579
1580         rt = &id->route;
1581         rt->num_paths = ib_event->param.req_rcvd.alternate_path ? 2 : 1;
1582         rt->path_rec = kmalloc(sizeof *rt->path_rec * rt->num_paths,
1583                                GFP_KERNEL);
1584         if (!rt->path_rec)
1585                 goto err;
1586
1587         rt->path_rec[0] = *ib_event->param.req_rcvd.primary_path;
1588         if (rt->num_paths == 2)
1589                 rt->path_rec[1] = *ib_event->param.req_rcvd.alternate_path;
1590
1591         if (net_dev) {
1592                 ret = rdma_copy_addr(&rt->addr.dev_addr, net_dev, NULL);
1593                 if (ret)
1594                         goto err;
1595         } else {
1596                 /* An AF_IB connection */
1597                 WARN_ON_ONCE(ss_family != AF_IB);
1598
1599                 cma_translate_ib((struct sockaddr_ib *)cma_src_addr(id_priv),
1600                                  &rt->addr.dev_addr);
1601         }
1602         rdma_addr_set_dgid(&rt->addr.dev_addr, &rt->path_rec[0].dgid);
1603
1604         id_priv->state = RDMA_CM_CONNECT;
1605         return id_priv;
1606
1607 err:
1608         rdma_destroy_id(id);
1609         return NULL;
1610 }
1611
1612 static struct rdma_id_private *cma_new_udp_id(struct rdma_cm_id *listen_id,
1613                                               struct ib_cm_event *ib_event,
1614                                               struct net_device *net_dev)
1615 {
1616         struct rdma_id_private *id_priv;
1617         struct rdma_cm_id *id;
1618         const sa_family_t ss_family = listen_id->route.addr.src_addr.ss_family;
1619         int ret;
1620
1621         id = rdma_create_id(listen_id->event_handler, listen_id->context,
1622                             listen_id->ps, IB_QPT_UD);
1623         if (IS_ERR(id))
1624                 return NULL;
1625
1626         id_priv = container_of(id, struct rdma_id_private, id);
1627         if (cma_save_net_info((struct sockaddr *)&id->route.addr.src_addr,
1628                               (struct sockaddr *)&id->route.addr.dst_addr,
1629                               listen_id, ib_event, ss_family,
1630                               ib_event->param.sidr_req_rcvd.service_id))
1631                 goto err;
1632
1633         if (net_dev) {
1634                 ret = rdma_copy_addr(&id->route.addr.dev_addr, net_dev, NULL);
1635                 if (ret)
1636                         goto err;
1637         } else {
1638                 /* An AF_IB connection */
1639                 WARN_ON_ONCE(ss_family != AF_IB);
1640
1641                 if (!cma_any_addr(cma_src_addr(id_priv)))
1642                         cma_translate_ib((struct sockaddr_ib *)
1643                                                 cma_src_addr(id_priv),
1644                                          &id->route.addr.dev_addr);
1645         }
1646
1647         id_priv->state = RDMA_CM_CONNECT;
1648         return id_priv;
1649 err:
1650         rdma_destroy_id(id);
1651         return NULL;
1652 }
1653
1654 static void cma_set_req_event_data(struct rdma_cm_event *event,
1655                                    struct ib_cm_req_event_param *req_data,
1656                                    void *private_data, int offset)
1657 {
1658         event->param.conn.private_data = private_data + offset;
1659         event->param.conn.private_data_len = IB_CM_REQ_PRIVATE_DATA_SIZE - offset;
1660         event->param.conn.responder_resources = req_data->responder_resources;
1661         event->param.conn.initiator_depth = req_data->initiator_depth;
1662         event->param.conn.flow_control = req_data->flow_control;
1663         event->param.conn.retry_count = req_data->retry_count;
1664         event->param.conn.rnr_retry_count = req_data->rnr_retry_count;
1665         event->param.conn.srq = req_data->srq;
1666         event->param.conn.qp_num = req_data->remote_qpn;
1667 }
1668
1669 static int cma_check_req_qp_type(struct rdma_cm_id *id, struct ib_cm_event *ib_event)
1670 {
1671         return (((ib_event->event == IB_CM_REQ_RECEIVED) &&
1672                  (ib_event->param.req_rcvd.qp_type == id->qp_type)) ||
1673                 ((ib_event->event == IB_CM_SIDR_REQ_RECEIVED) &&
1674                  (id->qp_type == IB_QPT_UD)) ||
1675                 (!id->qp_type));
1676 }
1677
1678 static int cma_req_handler(struct ib_cm_id *cm_id, struct ib_cm_event *ib_event)
1679 {
1680         struct rdma_id_private *listen_id, *conn_id;
1681         struct rdma_cm_event event;
1682         struct net_device *net_dev;
1683         int offset, ret;
1684
1685         listen_id = cma_id_from_event(cm_id, ib_event, &net_dev);
1686         if (IS_ERR(listen_id))
1687                 return PTR_ERR(listen_id);
1688
1689         if (!cma_check_req_qp_type(&listen_id->id, ib_event)) {
1690                 ret = -EINVAL;
1691                 goto net_dev_put;
1692         }
1693
1694         if (cma_disable_callback(listen_id, RDMA_CM_LISTEN)) {
1695                 ret = -ECONNABORTED;
1696                 goto net_dev_put;
1697         }
1698
1699         memset(&event, 0, sizeof event);
1700         offset = cma_user_data_offset(listen_id);
1701         event.event = RDMA_CM_EVENT_CONNECT_REQUEST;
1702         if (ib_event->event == IB_CM_SIDR_REQ_RECEIVED) {
1703                 conn_id = cma_new_udp_id(&listen_id->id, ib_event, net_dev);
1704                 event.param.ud.private_data = ib_event->private_data + offset;
1705                 event.param.ud.private_data_len =
1706                                 IB_CM_SIDR_REQ_PRIVATE_DATA_SIZE - offset;
1707         } else {
1708                 conn_id = cma_new_conn_id(&listen_id->id, ib_event, net_dev);
1709                 cma_set_req_event_data(&event, &ib_event->param.req_rcvd,
1710                                        ib_event->private_data, offset);
1711         }
1712         if (!conn_id) {
1713                 ret = -ENOMEM;
1714                 goto err1;
1715         }
1716
1717         mutex_lock_nested(&conn_id->handler_mutex, SINGLE_DEPTH_NESTING);
1718         ret = cma_acquire_dev(conn_id, listen_id);
1719         if (ret)
1720                 goto err2;
1721
1722         conn_id->cm_id.ib = cm_id;
1723         cm_id->context = conn_id;
1724         cm_id->cm_handler = cma_ib_handler;
1725
1726         /*
1727          * Protect against the user destroying conn_id from another thread
1728          * until we're done accessing it.
1729          */
1730         atomic_inc(&conn_id->refcount);
1731         ret = conn_id->id.event_handler(&conn_id->id, &event);
1732         if (ret)
1733                 goto err3;
1734         /*
1735          * Acquire mutex to prevent user executing rdma_destroy_id()
1736          * while we're accessing the cm_id.
1737          */
1738         mutex_lock(&lock);
1739         if (cma_comp(conn_id, RDMA_CM_CONNECT) &&
1740             (conn_id->id.qp_type != IB_QPT_UD))
1741                 ib_send_cm_mra(cm_id, CMA_CM_MRA_SETTING, NULL, 0);
1742         mutex_unlock(&lock);
1743         mutex_unlock(&conn_id->handler_mutex);
1744         mutex_unlock(&listen_id->handler_mutex);
1745         cma_deref_id(conn_id);
1746         if (net_dev)
1747                 dev_put(net_dev);
1748         return 0;
1749
1750 err3:
1751         cma_deref_id(conn_id);
1752         /* Destroy the CM ID by returning a non-zero value. */
1753         conn_id->cm_id.ib = NULL;
1754 err2:
1755         cma_exch(conn_id, RDMA_CM_DESTROYING);
1756         mutex_unlock(&conn_id->handler_mutex);
1757 err1:
1758         mutex_unlock(&listen_id->handler_mutex);
1759         if (conn_id)
1760                 rdma_destroy_id(&conn_id->id);
1761
1762 net_dev_put:
1763         if (net_dev)
1764                 dev_put(net_dev);
1765
1766         return ret;
1767 }
1768
1769 __be64 rdma_get_service_id(struct rdma_cm_id *id, struct sockaddr *addr)
1770 {
1771         if (addr->sa_family == AF_IB)
1772                 return ((struct sockaddr_ib *) addr)->sib_sid;
1773
1774         return cpu_to_be64(((u64)id->ps << 16) + be16_to_cpu(cma_port(addr)));
1775 }
1776 EXPORT_SYMBOL(rdma_get_service_id);
1777
1778 static int cma_iw_handler(struct iw_cm_id *iw_id, struct iw_cm_event *iw_event)
1779 {
1780         struct rdma_id_private *id_priv = iw_id->context;
1781         struct rdma_cm_event event;
1782         int ret = 0;
1783         struct sockaddr *laddr = (struct sockaddr *)&iw_event->local_addr;
1784         struct sockaddr *raddr = (struct sockaddr *)&iw_event->remote_addr;
1785
1786         if (cma_disable_callback(id_priv, RDMA_CM_CONNECT))
1787                 return 0;
1788
1789         memset(&event, 0, sizeof event);
1790         switch (iw_event->event) {
1791         case IW_CM_EVENT_CLOSE:
1792                 event.event = RDMA_CM_EVENT_DISCONNECTED;
1793                 break;
1794         case IW_CM_EVENT_CONNECT_REPLY:
1795                 memcpy(cma_src_addr(id_priv), laddr,
1796                        rdma_addr_size(laddr));
1797                 memcpy(cma_dst_addr(id_priv), raddr,
1798                        rdma_addr_size(raddr));
1799                 switch (iw_event->status) {
1800                 case 0:
1801                         event.event = RDMA_CM_EVENT_ESTABLISHED;
1802                         event.param.conn.initiator_depth = iw_event->ird;
1803                         event.param.conn.responder_resources = iw_event->ord;
1804                         break;
1805                 case -ECONNRESET:
1806                 case -ECONNREFUSED:
1807                         event.event = RDMA_CM_EVENT_REJECTED;
1808                         break;
1809                 case -ETIMEDOUT:
1810                         event.event = RDMA_CM_EVENT_UNREACHABLE;
1811                         break;
1812                 default:
1813                         event.event = RDMA_CM_EVENT_CONNECT_ERROR;
1814                         break;
1815                 }
1816                 break;
1817         case IW_CM_EVENT_ESTABLISHED:
1818                 event.event = RDMA_CM_EVENT_ESTABLISHED;
1819                 event.param.conn.initiator_depth = iw_event->ird;
1820                 event.param.conn.responder_resources = iw_event->ord;
1821                 break;
1822         default:
1823                 BUG_ON(1);
1824         }
1825
1826         event.status = iw_event->status;
1827         event.param.conn.private_data = iw_event->private_data;
1828         event.param.conn.private_data_len = iw_event->private_data_len;
1829         ret = id_priv->id.event_handler(&id_priv->id, &event);
1830         if (ret) {
1831                 /* Destroy the CM ID by returning a non-zero value. */
1832                 id_priv->cm_id.iw = NULL;
1833                 cma_exch(id_priv, RDMA_CM_DESTROYING);
1834                 mutex_unlock(&id_priv->handler_mutex);
1835                 rdma_destroy_id(&id_priv->id);
1836                 return ret;
1837         }
1838
1839         mutex_unlock(&id_priv->handler_mutex);
1840         return ret;
1841 }
1842
1843 static int iw_conn_req_handler(struct iw_cm_id *cm_id,
1844                                struct iw_cm_event *iw_event)
1845 {
1846         struct rdma_cm_id *new_cm_id;
1847         struct rdma_id_private *listen_id, *conn_id;
1848         struct rdma_cm_event event;
1849         int ret;
1850         struct ib_device_attr attr;
1851         struct sockaddr *laddr = (struct sockaddr *)&iw_event->local_addr;
1852         struct sockaddr *raddr = (struct sockaddr *)&iw_event->remote_addr;
1853
1854         listen_id = cm_id->context;
1855         if (cma_disable_callback(listen_id, RDMA_CM_LISTEN))
1856                 return -ECONNABORTED;
1857
1858         /* Create a new RDMA id for the new IW CM ID */
1859         new_cm_id = rdma_create_id(listen_id->id.event_handler,
1860                                    listen_id->id.context,
1861                                    RDMA_PS_TCP, IB_QPT_RC);
1862         if (IS_ERR(new_cm_id)) {
1863                 ret = -ENOMEM;
1864                 goto out;
1865         }
1866         conn_id = container_of(new_cm_id, struct rdma_id_private, id);
1867         mutex_lock_nested(&conn_id->handler_mutex, SINGLE_DEPTH_NESTING);
1868         conn_id->state = RDMA_CM_CONNECT;
1869
1870         ret = rdma_translate_ip(laddr, &conn_id->id.route.addr.dev_addr, NULL);
1871         if (ret) {
1872                 mutex_unlock(&conn_id->handler_mutex);
1873                 rdma_destroy_id(new_cm_id);
1874                 goto out;
1875         }
1876
1877         ret = cma_acquire_dev(conn_id, listen_id);
1878         if (ret) {
1879                 mutex_unlock(&conn_id->handler_mutex);
1880                 rdma_destroy_id(new_cm_id);
1881                 goto out;
1882         }
1883
1884         conn_id->cm_id.iw = cm_id;
1885         cm_id->context = conn_id;
1886         cm_id->cm_handler = cma_iw_handler;
1887
1888         memcpy(cma_src_addr(conn_id), laddr, rdma_addr_size(laddr));
1889         memcpy(cma_dst_addr(conn_id), raddr, rdma_addr_size(raddr));
1890
1891         ret = ib_query_device(conn_id->id.device, &attr);
1892         if (ret) {
1893                 mutex_unlock(&conn_id->handler_mutex);
1894                 rdma_destroy_id(new_cm_id);
1895                 goto out;
1896         }
1897
1898         memset(&event, 0, sizeof event);
1899         event.event = RDMA_CM_EVENT_CONNECT_REQUEST;
1900         event.param.conn.private_data = iw_event->private_data;
1901         event.param.conn.private_data_len = iw_event->private_data_len;
1902         event.param.conn.initiator_depth = iw_event->ird;
1903         event.param.conn.responder_resources = iw_event->ord;
1904
1905         /*
1906          * Protect against the user destroying conn_id from another thread
1907          * until we're done accessing it.
1908          */
1909         atomic_inc(&conn_id->refcount);
1910         ret = conn_id->id.event_handler(&conn_id->id, &event);
1911         if (ret) {
1912                 /* User wants to destroy the CM ID */
1913                 conn_id->cm_id.iw = NULL;
1914                 cma_exch(conn_id, RDMA_CM_DESTROYING);
1915                 mutex_unlock(&conn_id->handler_mutex);
1916                 cma_deref_id(conn_id);
1917                 rdma_destroy_id(&conn_id->id);
1918                 goto out;
1919         }
1920
1921         mutex_unlock(&conn_id->handler_mutex);
1922         cma_deref_id(conn_id);
1923
1924 out:
1925         mutex_unlock(&listen_id->handler_mutex);
1926         return ret;
1927 }
1928
1929 static int cma_ib_listen(struct rdma_id_private *id_priv)
1930 {
1931         struct sockaddr *addr;
1932         struct ib_cm_id *id;
1933         __be64 svc_id;
1934
1935         addr = cma_src_addr(id_priv);
1936         svc_id = rdma_get_service_id(&id_priv->id, addr);
1937         id = ib_cm_insert_listen(id_priv->id.device, cma_req_handler, svc_id);
1938         if (IS_ERR(id))
1939                 return PTR_ERR(id);
1940         id_priv->cm_id.ib = id;
1941
1942         return 0;
1943 }
1944
1945 static int cma_iw_listen(struct rdma_id_private *id_priv, int backlog)
1946 {
1947         int ret;
1948         struct iw_cm_id *id;
1949
1950         id = iw_create_cm_id(id_priv->id.device,
1951                              iw_conn_req_handler,
1952                              id_priv);
1953         if (IS_ERR(id))
1954                 return PTR_ERR(id);
1955
1956         id->tos = id_priv->tos;
1957         id_priv->cm_id.iw = id;
1958
1959         memcpy(&id_priv->cm_id.iw->local_addr, cma_src_addr(id_priv),
1960                rdma_addr_size(cma_src_addr(id_priv)));
1961
1962         ret = iw_cm_listen(id_priv->cm_id.iw, backlog);
1963
1964         if (ret) {
1965                 iw_destroy_cm_id(id_priv->cm_id.iw);
1966                 id_priv->cm_id.iw = NULL;
1967         }
1968
1969         return ret;
1970 }
1971
1972 static int cma_listen_handler(struct rdma_cm_id *id,
1973                               struct rdma_cm_event *event)
1974 {
1975         struct rdma_id_private *id_priv = id->context;
1976
1977         id->context = id_priv->id.context;
1978         id->event_handler = id_priv->id.event_handler;
1979         return id_priv->id.event_handler(id, event);
1980 }
1981
1982 static void cma_listen_on_dev(struct rdma_id_private *id_priv,
1983                               struct cma_device *cma_dev)
1984 {
1985         struct rdma_id_private *dev_id_priv;
1986         struct rdma_cm_id *id;
1987         int ret;
1988
1989         if (cma_family(id_priv) == AF_IB && !rdma_cap_ib_cm(cma_dev->device, 1))
1990                 return;
1991
1992         id = rdma_create_id(cma_listen_handler, id_priv, id_priv->id.ps,
1993                             id_priv->id.qp_type);
1994         if (IS_ERR(id))
1995                 return;
1996
1997         dev_id_priv = container_of(id, struct rdma_id_private, id);
1998
1999         dev_id_priv->state = RDMA_CM_ADDR_BOUND;
2000         memcpy(cma_src_addr(dev_id_priv), cma_src_addr(id_priv),
2001                rdma_addr_size(cma_src_addr(id_priv)));
2002
2003         cma_attach_to_dev(dev_id_priv, cma_dev);
2004         list_add_tail(&dev_id_priv->listen_list, &id_priv->listen_list);
2005         atomic_inc(&id_priv->refcount);
2006         dev_id_priv->internal_id = 1;
2007         dev_id_priv->afonly = id_priv->afonly;
2008
2009         ret = rdma_listen(id, id_priv->backlog);
2010         if (ret)
2011                 printk(KERN_WARNING "RDMA CMA: cma_listen_on_dev, error %d, "
2012                        "listening on device %s\n", ret, cma_dev->device->name);
2013 }
2014
2015 static void cma_listen_on_all(struct rdma_id_private *id_priv)
2016 {
2017         struct cma_device *cma_dev;
2018
2019         mutex_lock(&lock);
2020         list_add_tail(&id_priv->list, &listen_any_list);
2021         list_for_each_entry(cma_dev, &dev_list, list)
2022                 cma_listen_on_dev(id_priv, cma_dev);
2023         mutex_unlock(&lock);
2024 }
2025
2026 void rdma_set_service_type(struct rdma_cm_id *id, int tos)
2027 {
2028         struct rdma_id_private *id_priv;
2029
2030         id_priv = container_of(id, struct rdma_id_private, id);
2031         id_priv->tos = (u8) tos;
2032 }
2033 EXPORT_SYMBOL(rdma_set_service_type);
2034
2035 static void cma_query_handler(int status, struct ib_sa_path_rec *path_rec,
2036                               void *context)
2037 {
2038         struct cma_work *work = context;
2039         struct rdma_route *route;
2040
2041         route = &work->id->id.route;
2042
2043         if (!status) {
2044                 route->num_paths = 1;
2045                 *route->path_rec = *path_rec;
2046         } else {
2047                 work->old_state = RDMA_CM_ROUTE_QUERY;
2048                 work->new_state = RDMA_CM_ADDR_RESOLVED;
2049                 work->event.event = RDMA_CM_EVENT_ROUTE_ERROR;
2050                 work->event.status = status;
2051         }
2052
2053         queue_work(cma_wq, &work->work);
2054 }
2055
2056 static int cma_query_ib_route(struct rdma_id_private *id_priv, int timeout_ms,
2057                               struct cma_work *work)
2058 {
2059         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
2060         struct ib_sa_path_rec path_rec;
2061         ib_sa_comp_mask comp_mask;
2062         struct sockaddr_in6 *sin6;
2063         struct sockaddr_ib *sib;
2064
2065         memset(&path_rec, 0, sizeof path_rec);
2066         rdma_addr_get_sgid(dev_addr, &path_rec.sgid);
2067         rdma_addr_get_dgid(dev_addr, &path_rec.dgid);
2068         path_rec.pkey = cpu_to_be16(ib_addr_get_pkey(dev_addr));
2069         path_rec.numb_path = 1;
2070         path_rec.reversible = 1;
2071         path_rec.service_id = rdma_get_service_id(&id_priv->id, cma_dst_addr(id_priv));
2072
2073         comp_mask = IB_SA_PATH_REC_DGID | IB_SA_PATH_REC_SGID |
2074                     IB_SA_PATH_REC_PKEY | IB_SA_PATH_REC_NUMB_PATH |
2075                     IB_SA_PATH_REC_REVERSIBLE | IB_SA_PATH_REC_SERVICE_ID;
2076
2077         switch (cma_family(id_priv)) {
2078         case AF_INET:
2079                 path_rec.qos_class = cpu_to_be16((u16) id_priv->tos);
2080                 comp_mask |= IB_SA_PATH_REC_QOS_CLASS;
2081                 break;
2082         case AF_INET6:
2083                 sin6 = (struct sockaddr_in6 *) cma_src_addr(id_priv);
2084                 path_rec.traffic_class = (u8) (be32_to_cpu(sin6->sin6_flowinfo) >> 20);
2085                 comp_mask |= IB_SA_PATH_REC_TRAFFIC_CLASS;
2086                 break;
2087         case AF_IB:
2088                 sib = (struct sockaddr_ib *) cma_src_addr(id_priv);
2089                 path_rec.traffic_class = (u8) (be32_to_cpu(sib->sib_flowinfo) >> 20);
2090                 comp_mask |= IB_SA_PATH_REC_TRAFFIC_CLASS;
2091                 break;
2092         }
2093
2094         id_priv->query_id = ib_sa_path_rec_get(&sa_client, id_priv->id.device,
2095                                                id_priv->id.port_num, &path_rec,
2096                                                comp_mask, timeout_ms,
2097                                                GFP_KERNEL, cma_query_handler,
2098                                                work, &id_priv->query);
2099
2100         return (id_priv->query_id < 0) ? id_priv->query_id : 0;
2101 }
2102
2103 static void cma_work_handler(struct work_struct *_work)
2104 {
2105         struct cma_work *work = container_of(_work, struct cma_work, work);
2106         struct rdma_id_private *id_priv = work->id;
2107         int destroy = 0;
2108
2109         mutex_lock(&id_priv->handler_mutex);
2110         if (!cma_comp_exch(id_priv, work->old_state, work->new_state))
2111                 goto out;
2112
2113         if (id_priv->id.event_handler(&id_priv->id, &work->event)) {
2114                 cma_exch(id_priv, RDMA_CM_DESTROYING);
2115                 destroy = 1;
2116         }
2117 out:
2118         mutex_unlock(&id_priv->handler_mutex);
2119         cma_deref_id(id_priv);
2120         if (destroy)
2121                 rdma_destroy_id(&id_priv->id);
2122         kfree(work);
2123 }
2124
2125 static void cma_ndev_work_handler(struct work_struct *_work)
2126 {
2127         struct cma_ndev_work *work = container_of(_work, struct cma_ndev_work, work);
2128         struct rdma_id_private *id_priv = work->id;
2129         int destroy = 0;
2130
2131         mutex_lock(&id_priv->handler_mutex);
2132         if (id_priv->state == RDMA_CM_DESTROYING ||
2133             id_priv->state == RDMA_CM_DEVICE_REMOVAL)
2134                 goto out;
2135
2136         if (id_priv->id.event_handler(&id_priv->id, &work->event)) {
2137                 cma_exch(id_priv, RDMA_CM_DESTROYING);
2138                 destroy = 1;
2139         }
2140
2141 out:
2142         mutex_unlock(&id_priv->handler_mutex);
2143         cma_deref_id(id_priv);
2144         if (destroy)
2145                 rdma_destroy_id(&id_priv->id);
2146         kfree(work);
2147 }
2148
2149 static int cma_resolve_ib_route(struct rdma_id_private *id_priv, int timeout_ms)
2150 {
2151         struct rdma_route *route = &id_priv->id.route;
2152         struct cma_work *work;
2153         int ret;
2154
2155         work = kzalloc(sizeof *work, GFP_KERNEL);
2156         if (!work)
2157                 return -ENOMEM;
2158
2159         work->id = id_priv;
2160         INIT_WORK(&work->work, cma_work_handler);
2161         work->old_state = RDMA_CM_ROUTE_QUERY;
2162         work->new_state = RDMA_CM_ROUTE_RESOLVED;
2163         work->event.event = RDMA_CM_EVENT_ROUTE_RESOLVED;
2164
2165         route->path_rec = kmalloc(sizeof *route->path_rec, GFP_KERNEL);
2166         if (!route->path_rec) {
2167                 ret = -ENOMEM;
2168                 goto err1;
2169         }
2170
2171         ret = cma_query_ib_route(id_priv, timeout_ms, work);
2172         if (ret)
2173                 goto err2;
2174
2175         return 0;
2176 err2:
2177         kfree(route->path_rec);
2178         route->path_rec = NULL;
2179 err1:
2180         kfree(work);
2181         return ret;
2182 }
2183
2184 int rdma_set_ib_paths(struct rdma_cm_id *id,
2185                       struct ib_sa_path_rec *path_rec, int num_paths)
2186 {
2187         struct rdma_id_private *id_priv;
2188         int ret;
2189
2190         id_priv = container_of(id, struct rdma_id_private, id);
2191         if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_RESOLVED,
2192                            RDMA_CM_ROUTE_RESOLVED))
2193                 return -EINVAL;
2194
2195         id->route.path_rec = kmemdup(path_rec, sizeof *path_rec * num_paths,
2196                                      GFP_KERNEL);
2197         if (!id->route.path_rec) {
2198                 ret = -ENOMEM;
2199                 goto err;
2200         }
2201
2202         id->route.num_paths = num_paths;
2203         return 0;
2204 err:
2205         cma_comp_exch(id_priv, RDMA_CM_ROUTE_RESOLVED, RDMA_CM_ADDR_RESOLVED);
2206         return ret;
2207 }
2208 EXPORT_SYMBOL(rdma_set_ib_paths);
2209
2210 static int cma_resolve_iw_route(struct rdma_id_private *id_priv, int timeout_ms)
2211 {
2212         struct cma_work *work;
2213
2214         work = kzalloc(sizeof *work, GFP_KERNEL);
2215         if (!work)
2216                 return -ENOMEM;
2217
2218         work->id = id_priv;
2219         INIT_WORK(&work->work, cma_work_handler);
2220         work->old_state = RDMA_CM_ROUTE_QUERY;
2221         work->new_state = RDMA_CM_ROUTE_RESOLVED;
2222         work->event.event = RDMA_CM_EVENT_ROUTE_RESOLVED;
2223         queue_work(cma_wq, &work->work);
2224         return 0;
2225 }
2226
2227 static int iboe_tos_to_sl(struct net_device *ndev, int tos)
2228 {
2229         int prio;
2230         struct net_device *dev;
2231
2232         prio = rt_tos2priority(tos);
2233         dev = ndev->priv_flags & IFF_802_1Q_VLAN ?
2234                 vlan_dev_real_dev(ndev) : ndev;
2235
2236         if (dev->num_tc)
2237                 return netdev_get_prio_tc_map(dev, prio);
2238
2239 #if IS_ENABLED(CONFIG_VLAN_8021Q)
2240         if (ndev->priv_flags & IFF_802_1Q_VLAN)
2241                 return (vlan_dev_get_egress_qos_mask(ndev, prio) &
2242                         VLAN_PRIO_MASK) >> VLAN_PRIO_SHIFT;
2243 #endif
2244         return 0;
2245 }
2246
2247 static int cma_resolve_iboe_route(struct rdma_id_private *id_priv)
2248 {
2249         struct rdma_route *route = &id_priv->id.route;
2250         struct rdma_addr *addr = &route->addr;
2251         struct cma_work *work;
2252         int ret;
2253         struct net_device *ndev = NULL;
2254
2255
2256         work = kzalloc(sizeof *work, GFP_KERNEL);
2257         if (!work)
2258                 return -ENOMEM;
2259
2260         work->id = id_priv;
2261         INIT_WORK(&work->work, cma_work_handler);
2262
2263         route->path_rec = kzalloc(sizeof *route->path_rec, GFP_KERNEL);
2264         if (!route->path_rec) {
2265                 ret = -ENOMEM;
2266                 goto err1;
2267         }
2268
2269         route->num_paths = 1;
2270
2271         if (addr->dev_addr.bound_dev_if)
2272                 ndev = dev_get_by_index(&init_net, addr->dev_addr.bound_dev_if);
2273         if (!ndev) {
2274                 ret = -ENODEV;
2275                 goto err2;
2276         }
2277
2278         route->path_rec->vlan_id = rdma_vlan_dev_vlan_id(ndev);
2279         memcpy(route->path_rec->dmac, addr->dev_addr.dst_dev_addr, ETH_ALEN);
2280         memcpy(route->path_rec->smac, ndev->dev_addr, ndev->addr_len);
2281
2282         rdma_ip2gid((struct sockaddr *)&id_priv->id.route.addr.src_addr,
2283                     &route->path_rec->sgid);
2284         rdma_ip2gid((struct sockaddr *)&id_priv->id.route.addr.dst_addr,
2285                     &route->path_rec->dgid);
2286
2287         route->path_rec->hop_limit = 1;
2288         route->path_rec->reversible = 1;
2289         route->path_rec->pkey = cpu_to_be16(0xffff);
2290         route->path_rec->mtu_selector = IB_SA_EQ;
2291         route->path_rec->sl = iboe_tos_to_sl(ndev, id_priv->tos);
2292         route->path_rec->mtu = iboe_get_mtu(ndev->mtu);
2293         route->path_rec->rate_selector = IB_SA_EQ;
2294         route->path_rec->rate = iboe_get_rate(ndev);
2295         dev_put(ndev);
2296         route->path_rec->packet_life_time_selector = IB_SA_EQ;
2297         route->path_rec->packet_life_time = CMA_IBOE_PACKET_LIFETIME;
2298         if (!route->path_rec->mtu) {
2299                 ret = -EINVAL;
2300                 goto err2;
2301         }
2302
2303         work->old_state = RDMA_CM_ROUTE_QUERY;
2304         work->new_state = RDMA_CM_ROUTE_RESOLVED;
2305         work->event.event = RDMA_CM_EVENT_ROUTE_RESOLVED;
2306         work->event.status = 0;
2307
2308         queue_work(cma_wq, &work->work);
2309
2310         return 0;
2311
2312 err2:
2313         kfree(route->path_rec);
2314         route->path_rec = NULL;
2315 err1:
2316         kfree(work);
2317         return ret;
2318 }
2319
2320 int rdma_resolve_route(struct rdma_cm_id *id, int timeout_ms)
2321 {
2322         struct rdma_id_private *id_priv;
2323         int ret;
2324
2325         id_priv = container_of(id, struct rdma_id_private, id);
2326         if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_RESOLVED, RDMA_CM_ROUTE_QUERY))
2327                 return -EINVAL;
2328
2329         atomic_inc(&id_priv->refcount);
2330         if (rdma_cap_ib_sa(id->device, id->port_num))
2331                 ret = cma_resolve_ib_route(id_priv, timeout_ms);
2332         else if (rdma_protocol_roce(id->device, id->port_num))
2333                 ret = cma_resolve_iboe_route(id_priv);
2334         else if (rdma_protocol_iwarp(id->device, id->port_num))
2335                 ret = cma_resolve_iw_route(id_priv, timeout_ms);
2336         else
2337                 ret = -ENOSYS;
2338
2339         if (ret)
2340                 goto err;
2341
2342         return 0;
2343 err:
2344         cma_comp_exch(id_priv, RDMA_CM_ROUTE_QUERY, RDMA_CM_ADDR_RESOLVED);
2345         cma_deref_id(id_priv);
2346         return ret;
2347 }
2348 EXPORT_SYMBOL(rdma_resolve_route);
2349
2350 static void cma_set_loopback(struct sockaddr *addr)
2351 {
2352         switch (addr->sa_family) {
2353         case AF_INET:
2354                 ((struct sockaddr_in *) addr)->sin_addr.s_addr = htonl(INADDR_LOOPBACK);
2355                 break;
2356         case AF_INET6:
2357                 ipv6_addr_set(&((struct sockaddr_in6 *) addr)->sin6_addr,
2358                               0, 0, 0, htonl(1));
2359                 break;
2360         default:
2361                 ib_addr_set(&((struct sockaddr_ib *) addr)->sib_addr,
2362                             0, 0, 0, htonl(1));
2363                 break;
2364         }
2365 }
2366
2367 static int cma_bind_loopback(struct rdma_id_private *id_priv)
2368 {
2369         struct cma_device *cma_dev, *cur_dev;
2370         struct ib_port_attr port_attr;
2371         union ib_gid gid;
2372         u16 pkey;
2373         int ret;
2374         u8 p;
2375
2376         cma_dev = NULL;
2377         mutex_lock(&lock);
2378         list_for_each_entry(cur_dev, &dev_list, list) {
2379                 if (cma_family(id_priv) == AF_IB &&
2380                     !rdma_cap_ib_cm(cur_dev->device, 1))
2381                         continue;
2382
2383                 if (!cma_dev)
2384                         cma_dev = cur_dev;
2385
2386                 for (p = 1; p <= cur_dev->device->phys_port_cnt; ++p) {
2387                         if (!ib_query_port(cur_dev->device, p, &port_attr) &&
2388                             port_attr.state == IB_PORT_ACTIVE) {
2389                                 cma_dev = cur_dev;
2390                                 goto port_found;
2391                         }
2392                 }
2393         }
2394
2395         if (!cma_dev) {
2396                 ret = -ENODEV;
2397                 goto out;
2398         }
2399
2400         p = 1;
2401
2402 port_found:
2403         ret = ib_get_cached_gid(cma_dev->device, p, 0, &gid);
2404         if (ret)
2405                 goto out;
2406
2407         ret = ib_get_cached_pkey(cma_dev->device, p, 0, &pkey);
2408         if (ret)
2409                 goto out;
2410
2411         id_priv->id.route.addr.dev_addr.dev_type =
2412                 (rdma_protocol_ib(cma_dev->device, p)) ?
2413                 ARPHRD_INFINIBAND : ARPHRD_ETHER;
2414
2415         rdma_addr_set_sgid(&id_priv->id.route.addr.dev_addr, &gid);
2416         ib_addr_set_pkey(&id_priv->id.route.addr.dev_addr, pkey);
2417         id_priv->id.port_num = p;
2418         cma_attach_to_dev(id_priv, cma_dev);
2419         cma_set_loopback(cma_src_addr(id_priv));
2420 out:
2421         mutex_unlock(&lock);
2422         return ret;
2423 }
2424
2425 static void addr_handler(int status, struct sockaddr *src_addr,
2426                          struct rdma_dev_addr *dev_addr, void *context)
2427 {
2428         struct rdma_id_private *id_priv = context;
2429         struct rdma_cm_event event;
2430
2431         memset(&event, 0, sizeof event);
2432         mutex_lock(&id_priv->handler_mutex);
2433         if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_QUERY,
2434                            RDMA_CM_ADDR_RESOLVED))
2435                 goto out;
2436
2437         memcpy(cma_src_addr(id_priv), src_addr, rdma_addr_size(src_addr));
2438         if (!status && !id_priv->cma_dev)
2439                 status = cma_acquire_dev(id_priv, NULL);
2440
2441         if (status) {
2442                 if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_RESOLVED,
2443                                    RDMA_CM_ADDR_BOUND))
2444                         goto out;
2445                 event.event = RDMA_CM_EVENT_ADDR_ERROR;
2446                 event.status = status;
2447         } else
2448                 event.event = RDMA_CM_EVENT_ADDR_RESOLVED;
2449
2450         if (id_priv->id.event_handler(&id_priv->id, &event)) {
2451                 cma_exch(id_priv, RDMA_CM_DESTROYING);
2452                 mutex_unlock(&id_priv->handler_mutex);
2453                 cma_deref_id(id_priv);
2454                 rdma_destroy_id(&id_priv->id);
2455                 return;
2456         }
2457 out:
2458         mutex_unlock(&id_priv->handler_mutex);
2459         cma_deref_id(id_priv);
2460 }
2461
2462 static int cma_resolve_loopback(struct rdma_id_private *id_priv)
2463 {
2464         struct cma_work *work;
2465         union ib_gid gid;
2466         int ret;
2467
2468         work = kzalloc(sizeof *work, GFP_KERNEL);
2469         if (!work)
2470                 return -ENOMEM;
2471
2472         if (!id_priv->cma_dev) {
2473                 ret = cma_bind_loopback(id_priv);
2474                 if (ret)
2475                         goto err;
2476         }
2477
2478         rdma_addr_get_sgid(&id_priv->id.route.addr.dev_addr, &gid);
2479         rdma_addr_set_dgid(&id_priv->id.route.addr.dev_addr, &gid);
2480
2481         work->id = id_priv;
2482         INIT_WORK(&work->work, cma_work_handler);
2483         work->old_state = RDMA_CM_ADDR_QUERY;
2484         work->new_state = RDMA_CM_ADDR_RESOLVED;
2485         work->event.event = RDMA_CM_EVENT_ADDR_RESOLVED;
2486         queue_work(cma_wq, &work->work);
2487         return 0;
2488 err:
2489         kfree(work);
2490         return ret;
2491 }
2492
2493 static int cma_resolve_ib_addr(struct rdma_id_private *id_priv)
2494 {
2495         struct cma_work *work;
2496         int ret;
2497
2498         work = kzalloc(sizeof *work, GFP_KERNEL);
2499         if (!work)
2500                 return -ENOMEM;
2501
2502         if (!id_priv->cma_dev) {
2503                 ret = cma_resolve_ib_dev(id_priv);
2504                 if (ret)
2505                         goto err;
2506         }
2507
2508         rdma_addr_set_dgid(&id_priv->id.route.addr.dev_addr, (union ib_gid *)
2509                 &(((struct sockaddr_ib *) &id_priv->id.route.addr.dst_addr)->sib_addr));
2510
2511         work->id = id_priv;
2512         INIT_WORK(&work->work, cma_work_handler);
2513         work->old_state = RDMA_CM_ADDR_QUERY;
2514         work->new_state = RDMA_CM_ADDR_RESOLVED;
2515         work->event.event = RDMA_CM_EVENT_ADDR_RESOLVED;
2516         queue_work(cma_wq, &work->work);
2517         return 0;
2518 err:
2519         kfree(work);
2520         return ret;
2521 }
2522
2523 static int cma_bind_addr(struct rdma_cm_id *id, struct sockaddr *src_addr,
2524                          struct sockaddr *dst_addr)
2525 {
2526         if (!src_addr || !src_addr->sa_family) {
2527                 src_addr = (struct sockaddr *) &id->route.addr.src_addr;
2528                 src_addr->sa_family = dst_addr->sa_family;
2529                 if (dst_addr->sa_family == AF_INET6) {
2530                         struct sockaddr_in6 *src_addr6 = (struct sockaddr_in6 *) src_addr;
2531                         struct sockaddr_in6 *dst_addr6 = (struct sockaddr_in6 *) dst_addr;
2532                         src_addr6->sin6_scope_id = dst_addr6->sin6_scope_id;
2533                         if (ipv6_addr_type(&dst_addr6->sin6_addr) & IPV6_ADDR_LINKLOCAL)
2534                                 id->route.addr.dev_addr.bound_dev_if = dst_addr6->sin6_scope_id;
2535                 } else if (dst_addr->sa_family == AF_IB) {
2536                         ((struct sockaddr_ib *) src_addr)->sib_pkey =
2537                                 ((struct sockaddr_ib *) dst_addr)->sib_pkey;
2538                 }
2539         }
2540         return rdma_bind_addr(id, src_addr);
2541 }
2542
2543 int rdma_resolve_addr(struct rdma_cm_id *id, struct sockaddr *src_addr,
2544                       struct sockaddr *dst_addr, int timeout_ms)
2545 {
2546         struct rdma_id_private *id_priv;
2547         int ret;
2548
2549         id_priv = container_of(id, struct rdma_id_private, id);
2550         if (id_priv->state == RDMA_CM_IDLE) {
2551                 ret = cma_bind_addr(id, src_addr, dst_addr);
2552                 if (ret)
2553                         return ret;
2554         }
2555
2556         if (cma_family(id_priv) != dst_addr->sa_family)
2557                 return -EINVAL;
2558
2559         if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_BOUND, RDMA_CM_ADDR_QUERY))
2560                 return -EINVAL;
2561
2562         atomic_inc(&id_priv->refcount);
2563         memcpy(cma_dst_addr(id_priv), dst_addr, rdma_addr_size(dst_addr));
2564         if (cma_any_addr(dst_addr)) {
2565                 ret = cma_resolve_loopback(id_priv);
2566         } else {
2567                 if (dst_addr->sa_family == AF_IB) {
2568                         ret = cma_resolve_ib_addr(id_priv);
2569                 } else {
2570                         ret = rdma_resolve_ip(&addr_client, cma_src_addr(id_priv),
2571                                               dst_addr, &id->route.addr.dev_addr,
2572                                               timeout_ms, addr_handler, id_priv);
2573                 }
2574         }
2575         if (ret)
2576                 goto err;
2577
2578         return 0;
2579 err:
2580         cma_comp_exch(id_priv, RDMA_CM_ADDR_QUERY, RDMA_CM_ADDR_BOUND);
2581         cma_deref_id(id_priv);
2582         return ret;
2583 }
2584 EXPORT_SYMBOL(rdma_resolve_addr);
2585
2586 int rdma_set_reuseaddr(struct rdma_cm_id *id, int reuse)
2587 {
2588         struct rdma_id_private *id_priv;
2589         unsigned long flags;
2590         int ret;
2591
2592         id_priv = container_of(id, struct rdma_id_private, id);
2593         spin_lock_irqsave(&id_priv->lock, flags);
2594         if (reuse || id_priv->state == RDMA_CM_IDLE) {
2595                 id_priv->reuseaddr = reuse;
2596                 ret = 0;
2597         } else {
2598                 ret = -EINVAL;
2599         }
2600         spin_unlock_irqrestore(&id_priv->lock, flags);
2601         return ret;
2602 }
2603 EXPORT_SYMBOL(rdma_set_reuseaddr);
2604
2605 int rdma_set_afonly(struct rdma_cm_id *id, int afonly)
2606 {
2607         struct rdma_id_private *id_priv;
2608         unsigned long flags;
2609         int ret;
2610
2611         id_priv = container_of(id, struct rdma_id_private, id);
2612         spin_lock_irqsave(&id_priv->lock, flags);
2613         if (id_priv->state == RDMA_CM_IDLE || id_priv->state == RDMA_CM_ADDR_BOUND) {
2614                 id_priv->options |= (1 << CMA_OPTION_AFONLY);
2615                 id_priv->afonly = afonly;
2616                 ret = 0;
2617         } else {
2618                 ret = -EINVAL;
2619         }
2620         spin_unlock_irqrestore(&id_priv->lock, flags);
2621         return ret;
2622 }
2623 EXPORT_SYMBOL(rdma_set_afonly);
2624
2625 static void cma_bind_port(struct rdma_bind_list *bind_list,
2626                           struct rdma_id_private *id_priv)
2627 {
2628         struct sockaddr *addr;
2629         struct sockaddr_ib *sib;
2630         u64 sid, mask;
2631         __be16 port;
2632
2633         addr = cma_src_addr(id_priv);
2634         port = htons(bind_list->port);
2635
2636         switch (addr->sa_family) {
2637         case AF_INET:
2638                 ((struct sockaddr_in *) addr)->sin_port = port;
2639                 break;
2640         case AF_INET6:
2641                 ((struct sockaddr_in6 *) addr)->sin6_port = port;
2642                 break;
2643         case AF_IB:
2644                 sib = (struct sockaddr_ib *) addr;
2645                 sid = be64_to_cpu(sib->sib_sid);
2646                 mask = be64_to_cpu(sib->sib_sid_mask);
2647                 sib->sib_sid = cpu_to_be64((sid & mask) | (u64) ntohs(port));
2648                 sib->sib_sid_mask = cpu_to_be64(~0ULL);
2649                 break;
2650         }
2651         id_priv->bind_list = bind_list;
2652         hlist_add_head(&id_priv->node, &bind_list->owners);
2653 }
2654
2655 static int cma_alloc_port(enum rdma_port_space ps,
2656                           struct rdma_id_private *id_priv, unsigned short snum)
2657 {
2658         struct rdma_bind_list *bind_list;
2659         int ret;
2660
2661         bind_list = kzalloc(sizeof *bind_list, GFP_KERNEL);
2662         if (!bind_list)
2663                 return -ENOMEM;
2664
2665         ret = cma_ps_alloc(ps, bind_list, snum);
2666         if (ret < 0)
2667                 goto err;
2668
2669         bind_list->ps = ps;
2670         bind_list->port = (unsigned short)ret;
2671         cma_bind_port(bind_list, id_priv);
2672         return 0;
2673 err:
2674         kfree(bind_list);
2675         return ret == -ENOSPC ? -EADDRNOTAVAIL : ret;
2676 }
2677
2678 static int cma_alloc_any_port(enum rdma_port_space ps,
2679                               struct rdma_id_private *id_priv)
2680 {
2681         static unsigned int last_used_port;
2682         int low, high, remaining;
2683         unsigned int rover;
2684
2685         inet_get_local_port_range(&init_net, &low, &high);
2686         remaining = (high - low) + 1;
2687         rover = prandom_u32() % remaining + low;
2688 retry:
2689         if (last_used_port != rover &&
2690             !cma_ps_find(ps, (unsigned short)rover)) {
2691                 int ret = cma_alloc_port(ps, id_priv, rover);
2692                 /*
2693                  * Remember previously used port number in order to avoid
2694                  * re-using same port immediately after it is closed.
2695                  */
2696                 if (!ret)
2697                         last_used_port = rover;
2698                 if (ret != -EADDRNOTAVAIL)
2699                         return ret;
2700         }
2701         if (--remaining) {
2702                 rover++;
2703                 if ((rover < low) || (rover > high))
2704                         rover = low;
2705                 goto retry;
2706         }
2707         return -EADDRNOTAVAIL;
2708 }
2709
2710 /*
2711  * Check that the requested port is available.  This is called when trying to
2712  * bind to a specific port, or when trying to listen on a bound port.  In
2713  * the latter case, the provided id_priv may already be on the bind_list, but
2714  * we still need to check that it's okay to start listening.
2715  */
2716 static int cma_check_port(struct rdma_bind_list *bind_list,
2717                           struct rdma_id_private *id_priv, uint8_t reuseaddr)
2718 {
2719         struct rdma_id_private *cur_id;
2720         struct sockaddr *addr, *cur_addr;
2721
2722         addr = cma_src_addr(id_priv);
2723         hlist_for_each_entry(cur_id, &bind_list->owners, node) {
2724                 if (id_priv == cur_id)
2725                         continue;
2726
2727                 if ((cur_id->state != RDMA_CM_LISTEN) && reuseaddr &&
2728                     cur_id->reuseaddr)
2729                         continue;
2730
2731                 cur_addr = cma_src_addr(cur_id);
2732                 if (id_priv->afonly && cur_id->afonly &&
2733                     (addr->sa_family != cur_addr->sa_family))
2734                         continue;
2735
2736                 if (cma_any_addr(addr) || cma_any_addr(cur_addr))
2737                         return -EADDRNOTAVAIL;
2738
2739                 if (!cma_addr_cmp(addr, cur_addr))
2740                         return -EADDRINUSE;
2741         }
2742         return 0;
2743 }
2744
2745 static int cma_use_port(enum rdma_port_space ps,
2746                         struct rdma_id_private *id_priv)
2747 {
2748         struct rdma_bind_list *bind_list;
2749         unsigned short snum;
2750         int ret;
2751
2752         snum = ntohs(cma_port(cma_src_addr(id_priv)));
2753         if (snum < PROT_SOCK && !capable(CAP_NET_BIND_SERVICE))
2754                 return -EACCES;
2755
2756         bind_list = cma_ps_find(ps, snum);
2757         if (!bind_list) {
2758                 ret = cma_alloc_port(ps, id_priv, snum);
2759         } else {
2760                 ret = cma_check_port(bind_list, id_priv, id_priv->reuseaddr);
2761                 if (!ret)
2762                         cma_bind_port(bind_list, id_priv);
2763         }
2764         return ret;
2765 }
2766
2767 static int cma_bind_listen(struct rdma_id_private *id_priv)
2768 {
2769         struct rdma_bind_list *bind_list = id_priv->bind_list;
2770         int ret = 0;
2771
2772         mutex_lock(&lock);
2773         if (bind_list->owners.first->next)
2774                 ret = cma_check_port(bind_list, id_priv, 0);
2775         mutex_unlock(&lock);
2776         return ret;
2777 }
2778
2779 static enum rdma_port_space cma_select_inet_ps(
2780                 struct rdma_id_private *id_priv)
2781 {
2782         switch (id_priv->id.ps) {
2783         case RDMA_PS_TCP:
2784         case RDMA_PS_UDP:
2785         case RDMA_PS_IPOIB:
2786         case RDMA_PS_IB:
2787                 return id_priv->id.ps;
2788         default:
2789
2790                 return 0;
2791         }
2792 }
2793
2794 static enum rdma_port_space cma_select_ib_ps(struct rdma_id_private *id_priv)
2795 {
2796         enum rdma_port_space ps = 0;
2797         struct sockaddr_ib *sib;
2798         u64 sid_ps, mask, sid;
2799
2800         sib = (struct sockaddr_ib *) cma_src_addr(id_priv);
2801         mask = be64_to_cpu(sib->sib_sid_mask) & RDMA_IB_IP_PS_MASK;
2802         sid = be64_to_cpu(sib->sib_sid) & mask;
2803
2804         if ((id_priv->id.ps == RDMA_PS_IB) && (sid == (RDMA_IB_IP_PS_IB & mask))) {
2805                 sid_ps = RDMA_IB_IP_PS_IB;
2806                 ps = RDMA_PS_IB;
2807         } else if (((id_priv->id.ps == RDMA_PS_IB) || (id_priv->id.ps == RDMA_PS_TCP)) &&
2808                    (sid == (RDMA_IB_IP_PS_TCP & mask))) {
2809                 sid_ps = RDMA_IB_IP_PS_TCP;
2810                 ps = RDMA_PS_TCP;
2811         } else if (((id_priv->id.ps == RDMA_PS_IB) || (id_priv->id.ps == RDMA_PS_UDP)) &&
2812                    (sid == (RDMA_IB_IP_PS_UDP & mask))) {
2813                 sid_ps = RDMA_IB_IP_PS_UDP;
2814                 ps = RDMA_PS_UDP;
2815         }
2816
2817         if (ps) {
2818                 sib->sib_sid = cpu_to_be64(sid_ps | ntohs(cma_port((struct sockaddr *) sib)));
2819                 sib->sib_sid_mask = cpu_to_be64(RDMA_IB_IP_PS_MASK |
2820                                                 be64_to_cpu(sib->sib_sid_mask));
2821         }
2822         return ps;
2823 }
2824
2825 static int cma_get_port(struct rdma_id_private *id_priv)
2826 {
2827         enum rdma_port_space ps;
2828         int ret;
2829
2830         if (cma_family(id_priv) != AF_IB)
2831                 ps = cma_select_inet_ps(id_priv);
2832         else
2833                 ps = cma_select_ib_ps(id_priv);
2834         if (!ps)
2835                 return -EPROTONOSUPPORT;
2836
2837         mutex_lock(&lock);
2838         if (cma_any_port(cma_src_addr(id_priv)))
2839                 ret = cma_alloc_any_port(ps, id_priv);
2840         else
2841                 ret = cma_use_port(ps, id_priv);
2842         mutex_unlock(&lock);
2843
2844         return ret;
2845 }
2846
2847 static int cma_check_linklocal(struct rdma_dev_addr *dev_addr,
2848                                struct sockaddr *addr)
2849 {
2850 #if IS_ENABLED(CONFIG_IPV6)
2851         struct sockaddr_in6 *sin6;
2852
2853         if (addr->sa_family != AF_INET6)
2854                 return 0;
2855
2856         sin6 = (struct sockaddr_in6 *) addr;
2857
2858         if (!(ipv6_addr_type(&sin6->sin6_addr) & IPV6_ADDR_LINKLOCAL))
2859                 return 0;
2860
2861         if (!sin6->sin6_scope_id)
2862                         return -EINVAL;
2863
2864         dev_addr->bound_dev_if = sin6->sin6_scope_id;
2865 #endif
2866         return 0;
2867 }
2868
2869 int rdma_listen(struct rdma_cm_id *id, int backlog)
2870 {
2871         struct rdma_id_private *id_priv;
2872         int ret;
2873
2874         id_priv = container_of(id, struct rdma_id_private, id);
2875         if (id_priv->state == RDMA_CM_IDLE) {
2876                 id->route.addr.src_addr.ss_family = AF_INET;
2877                 ret = rdma_bind_addr(id, cma_src_addr(id_priv));
2878                 if (ret)
2879                         return ret;
2880         }
2881
2882         if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_BOUND, RDMA_CM_LISTEN))
2883                 return -EINVAL;
2884
2885         if (id_priv->reuseaddr) {
2886                 ret = cma_bind_listen(id_priv);
2887                 if (ret)
2888                         goto err;
2889         }
2890
2891         id_priv->backlog = backlog;
2892         if (id->device) {
2893                 if (rdma_cap_ib_cm(id->device, 1)) {
2894                         ret = cma_ib_listen(id_priv);
2895                         if (ret)
2896                                 goto err;
2897                 } else if (rdma_cap_iw_cm(id->device, 1)) {
2898                         ret = cma_iw_listen(id_priv, backlog);
2899                         if (ret)
2900                                 goto err;
2901                 } else {
2902                         ret = -ENOSYS;
2903                         goto err;
2904                 }
2905         } else
2906                 cma_listen_on_all(id_priv);
2907
2908         return 0;
2909 err:
2910         id_priv->backlog = 0;
2911         cma_comp_exch(id_priv, RDMA_CM_LISTEN, RDMA_CM_ADDR_BOUND);
2912         return ret;
2913 }
2914 EXPORT_SYMBOL(rdma_listen);
2915
2916 int rdma_bind_addr(struct rdma_cm_id *id, struct sockaddr *addr)
2917 {
2918         struct rdma_id_private *id_priv;
2919         int ret;
2920
2921         if (addr->sa_family != AF_INET && addr->sa_family != AF_INET6 &&
2922             addr->sa_family != AF_IB)
2923                 return -EAFNOSUPPORT;
2924
2925         id_priv = container_of(id, struct rdma_id_private, id);
2926         if (!cma_comp_exch(id_priv, RDMA_CM_IDLE, RDMA_CM_ADDR_BOUND))
2927                 return -EINVAL;
2928
2929         ret = cma_check_linklocal(&id->route.addr.dev_addr, addr);
2930         if (ret)
2931                 goto err1;
2932
2933         memcpy(cma_src_addr(id_priv), addr, rdma_addr_size(addr));
2934         if (!cma_any_addr(addr)) {
2935                 ret = cma_translate_addr(addr, &id->route.addr.dev_addr);
2936                 if (ret)
2937                         goto err1;
2938
2939                 ret = cma_acquire_dev(id_priv, NULL);
2940                 if (ret)
2941                         goto err1;
2942         }
2943
2944         if (!(id_priv->options & (1 << CMA_OPTION_AFONLY))) {
2945                 if (addr->sa_family == AF_INET)
2946                         id_priv->afonly = 1;
2947 #if IS_ENABLED(CONFIG_IPV6)
2948                 else if (addr->sa_family == AF_INET6)
2949                         id_priv->afonly = init_net.ipv6.sysctl.bindv6only;
2950 #endif
2951         }
2952         ret = cma_get_port(id_priv);
2953         if (ret)
2954                 goto err2;
2955
2956         return 0;
2957 err2:
2958         if (id_priv->cma_dev)
2959                 cma_release_dev(id_priv);
2960 err1:
2961         cma_comp_exch(id_priv, RDMA_CM_ADDR_BOUND, RDMA_CM_IDLE);
2962         return ret;
2963 }
2964 EXPORT_SYMBOL(rdma_bind_addr);
2965
2966 static int cma_format_hdr(void *hdr, struct rdma_id_private *id_priv)
2967 {
2968         struct cma_hdr *cma_hdr;
2969
2970         cma_hdr = hdr;
2971         cma_hdr->cma_version = CMA_VERSION;
2972         if (cma_family(id_priv) == AF_INET) {
2973                 struct sockaddr_in *src4, *dst4;
2974
2975                 src4 = (struct sockaddr_in *) cma_src_addr(id_priv);
2976                 dst4 = (struct sockaddr_in *) cma_dst_addr(id_priv);
2977
2978                 cma_set_ip_ver(cma_hdr, 4);
2979                 cma_hdr->src_addr.ip4.addr = src4->sin_addr.s_addr;
2980                 cma_hdr->dst_addr.ip4.addr = dst4->sin_addr.s_addr;
2981                 cma_hdr->port = src4->sin_port;
2982         } else if (cma_family(id_priv) == AF_INET6) {
2983                 struct sockaddr_in6 *src6, *dst6;
2984
2985                 src6 = (struct sockaddr_in6 *) cma_src_addr(id_priv);
2986                 dst6 = (struct sockaddr_in6 *) cma_dst_addr(id_priv);
2987
2988                 cma_set_ip_ver(cma_hdr, 6);
2989                 cma_hdr->src_addr.ip6 = src6->sin6_addr;
2990                 cma_hdr->dst_addr.ip6 = dst6->sin6_addr;
2991                 cma_hdr->port = src6->sin6_port;
2992         }
2993         return 0;
2994 }
2995
2996 static int cma_sidr_rep_handler(struct ib_cm_id *cm_id,
2997                                 struct ib_cm_event *ib_event)
2998 {
2999         struct rdma_id_private *id_priv = cm_id->context;
3000         struct rdma_cm_event event;
3001         struct ib_cm_sidr_rep_event_param *rep = &ib_event->param.sidr_rep_rcvd;
3002         int ret = 0;
3003
3004         if (cma_disable_callback(id_priv, RDMA_CM_CONNECT))
3005                 return 0;
3006
3007         memset(&event, 0, sizeof event);
3008         switch (ib_event->event) {
3009         case IB_CM_SIDR_REQ_ERROR:
3010                 event.event = RDMA_CM_EVENT_UNREACHABLE;
3011                 event.status = -ETIMEDOUT;
3012                 break;
3013         case IB_CM_SIDR_REP_RECEIVED:
3014                 event.param.ud.private_data = ib_event->private_data;
3015                 event.param.ud.private_data_len = IB_CM_SIDR_REP_PRIVATE_DATA_SIZE;
3016                 if (rep->status != IB_SIDR_SUCCESS) {
3017                         event.event = RDMA_CM_EVENT_UNREACHABLE;
3018                         event.status = ib_event->param.sidr_rep_rcvd.status;
3019                         break;
3020                 }
3021                 ret = cma_set_qkey(id_priv, rep->qkey);
3022                 if (ret) {
3023                         event.event = RDMA_CM_EVENT_ADDR_ERROR;
3024                         event.status = ret;
3025                         break;
3026                 }
3027                 ib_init_ah_from_path(id_priv->id.device, id_priv->id.port_num,
3028                                      id_priv->id.route.path_rec,
3029                                      &event.param.ud.ah_attr);
3030                 event.param.ud.qp_num = rep->qpn;
3031                 event.param.ud.qkey = rep->qkey;
3032                 event.event = RDMA_CM_EVENT_ESTABLISHED;
3033                 event.status = 0;
3034                 break;
3035         default:
3036                 printk(KERN_ERR "RDMA CMA: unexpected IB CM event: %d\n",
3037                        ib_event->event);
3038                 goto out;
3039         }
3040
3041         ret = id_priv->id.event_handler(&id_priv->id, &event);
3042         if (ret) {
3043                 /* Destroy the CM ID by returning a non-zero value. */
3044                 id_priv->cm_id.ib = NULL;
3045                 cma_exch(id_priv, RDMA_CM_DESTROYING);
3046                 mutex_unlock(&id_priv->handler_mutex);
3047                 rdma_destroy_id(&id_priv->id);
3048                 return ret;
3049         }
3050 out:
3051         mutex_unlock(&id_priv->handler_mutex);
3052         return ret;
3053 }
3054
3055 static int cma_resolve_ib_udp(struct rdma_id_private *id_priv,
3056                               struct rdma_conn_param *conn_param)
3057 {
3058         struct ib_cm_sidr_req_param req;
3059         struct ib_cm_id *id;
3060         void *private_data;
3061         int offset, ret;
3062
3063         memset(&req, 0, sizeof req);
3064         offset = cma_user_data_offset(id_priv);
3065         req.private_data_len = offset + conn_param->private_data_len;
3066         if (req.private_data_len < conn_param->private_data_len)
3067                 return -EINVAL;
3068
3069         if (req.private_data_len) {
3070                 private_data = kzalloc(req.private_data_len, GFP_ATOMIC);
3071                 if (!private_data)
3072                         return -ENOMEM;
3073         } else {
3074                 private_data = NULL;
3075         }
3076
3077         if (conn_param->private_data && conn_param->private_data_len)
3078                 memcpy(private_data + offset, conn_param->private_data,
3079                        conn_param->private_data_len);
3080
3081         if (private_data) {
3082                 ret = cma_format_hdr(private_data, id_priv);
3083                 if (ret)
3084                         goto out;
3085                 req.private_data = private_data;
3086         }
3087
3088         id = ib_create_cm_id(id_priv->id.device, cma_sidr_rep_handler,
3089                              id_priv);
3090         if (IS_ERR(id)) {
3091                 ret = PTR_ERR(id);
3092                 goto out;
3093         }
3094         id_priv->cm_id.ib = id;
3095
3096         req.path = id_priv->id.route.path_rec;
3097         req.service_id = rdma_get_service_id(&id_priv->id, cma_dst_addr(id_priv));
3098         req.timeout_ms = 1 << (CMA_CM_RESPONSE_TIMEOUT - 8);
3099         req.max_cm_retries = CMA_MAX_CM_RETRIES;
3100
3101         ret = ib_send_cm_sidr_req(id_priv->cm_id.ib, &req);
3102         if (ret) {
3103                 ib_destroy_cm_id(id_priv->cm_id.ib);
3104                 id_priv->cm_id.ib = NULL;
3105         }
3106 out:
3107         kfree(private_data);
3108         return ret;
3109 }
3110
3111 static int cma_connect_ib(struct rdma_id_private *id_priv,
3112                           struct rdma_conn_param *conn_param)
3113 {
3114         struct ib_cm_req_param req;
3115         struct rdma_route *route;
3116         void *private_data;
3117         struct ib_cm_id *id;
3118         int offset, ret;
3119
3120         memset(&req, 0, sizeof req);
3121         offset = cma_user_data_offset(id_priv);
3122         req.private_data_len = offset + conn_param->private_data_len;
3123         if (req.private_data_len < conn_param->private_data_len)
3124                 return -EINVAL;
3125
3126         if (req.private_data_len) {
3127                 private_data = kzalloc(req.private_data_len, GFP_ATOMIC);
3128                 if (!private_data)
3129                         return -ENOMEM;
3130         } else {
3131                 private_data = NULL;
3132         }
3133
3134         if (conn_param->private_data && conn_param->private_data_len)
3135                 memcpy(private_data + offset, conn_param->private_data,
3136                        conn_param->private_data_len);
3137
3138         id = ib_create_cm_id(id_priv->id.device, cma_ib_handler, id_priv);
3139         if (IS_ERR(id)) {
3140                 ret = PTR_ERR(id);
3141                 goto out;
3142         }
3143         id_priv->cm_id.ib = id;
3144
3145         route = &id_priv->id.route;
3146         if (private_data) {
3147                 ret = cma_format_hdr(private_data, id_priv);
3148                 if (ret)
3149                         goto out;
3150                 req.private_data = private_data;
3151         }
3152
3153         req.primary_path = &route->path_rec[0];
3154         if (route->num_paths == 2)
3155                 req.alternate_path = &route->path_rec[1];
3156
3157         req.service_id = rdma_get_service_id(&id_priv->id, cma_dst_addr(id_priv));
3158         req.qp_num = id_priv->qp_num;
3159         req.qp_type = id_priv->id.qp_type;
3160         req.starting_psn = id_priv->seq_num;
3161         req.responder_resources = conn_param->responder_resources;
3162         req.initiator_depth = conn_param->initiator_depth;
3163         req.flow_control = conn_param->flow_control;
3164         req.retry_count = min_t(u8, 7, conn_param->retry_count);
3165         req.rnr_retry_count = min_t(u8, 7, conn_param->rnr_retry_count);
3166         req.remote_cm_response_timeout = CMA_CM_RESPONSE_TIMEOUT;
3167         req.local_cm_response_timeout = CMA_CM_RESPONSE_TIMEOUT;
3168         req.max_cm_retries = CMA_MAX_CM_RETRIES;
3169         req.srq = id_priv->srq ? 1 : 0;
3170
3171         ret = ib_send_cm_req(id_priv->cm_id.ib, &req);
3172 out:
3173         if (ret && !IS_ERR(id)) {
3174                 ib_destroy_cm_id(id);
3175                 id_priv->cm_id.ib = NULL;
3176         }
3177
3178         kfree(private_data);
3179         return ret;
3180 }
3181
3182 static int cma_connect_iw(struct rdma_id_private *id_priv,
3183                           struct rdma_conn_param *conn_param)
3184 {
3185         struct iw_cm_id *cm_id;
3186         int ret;
3187         struct iw_cm_conn_param iw_param;
3188
3189         cm_id = iw_create_cm_id(id_priv->id.device, cma_iw_handler, id_priv);
3190         if (IS_ERR(cm_id))
3191                 return PTR_ERR(cm_id);
3192
3193         cm_id->tos = id_priv->tos;
3194         id_priv->cm_id.iw = cm_id;
3195
3196         memcpy(&cm_id->local_addr, cma_src_addr(id_priv),
3197                rdma_addr_size(cma_src_addr(id_priv)));
3198         memcpy(&cm_id->remote_addr, cma_dst_addr(id_priv),
3199                rdma_addr_size(cma_dst_addr(id_priv)));
3200
3201         ret = cma_modify_qp_rtr(id_priv, conn_param);
3202         if (ret)
3203                 goto out;
3204
3205         if (conn_param) {
3206                 iw_param.ord = conn_param->initiator_depth;
3207                 iw_param.ird = conn_param->responder_resources;
3208                 iw_param.private_data = conn_param->private_data;
3209                 iw_param.private_data_len = conn_param->private_data_len;
3210                 iw_param.qpn = id_priv->id.qp ? id_priv->qp_num : conn_param->qp_num;
3211         } else {
3212                 memset(&iw_param, 0, sizeof iw_param);
3213                 iw_param.qpn = id_priv->qp_num;
3214         }
3215         ret = iw_cm_connect(cm_id, &iw_param);
3216 out:
3217         if (ret) {
3218                 iw_destroy_cm_id(cm_id);
3219                 id_priv->cm_id.iw = NULL;
3220         }
3221         return ret;
3222 }
3223
3224 int rdma_connect(struct rdma_cm_id *id, struct rdma_conn_param *conn_param)
3225 {
3226         struct rdma_id_private *id_priv;
3227         int ret;
3228
3229         id_priv = container_of(id, struct rdma_id_private, id);
3230         if (!cma_comp_exch(id_priv, RDMA_CM_ROUTE_RESOLVED, RDMA_CM_CONNECT))
3231                 return -EINVAL;
3232
3233         if (!id->qp) {
3234                 id_priv->qp_num = conn_param->qp_num;
3235                 id_priv->srq = conn_param->srq;
3236         }
3237
3238         if (rdma_cap_ib_cm(id->device, id->port_num)) {
3239                 if (id->qp_type == IB_QPT_UD)
3240                         ret = cma_resolve_ib_udp(id_priv, conn_param);
3241                 else
3242                         ret = cma_connect_ib(id_priv, conn_param);
3243         } else if (rdma_cap_iw_cm(id->device, id->port_num))
3244                 ret = cma_connect_iw(id_priv, conn_param);
3245         else
3246                 ret = -ENOSYS;
3247         if (ret)
3248                 goto err;
3249
3250         return 0;
3251 err:
3252         cma_comp_exch(id_priv, RDMA_CM_CONNECT, RDMA_CM_ROUTE_RESOLVED);
3253         return ret;
3254 }
3255 EXPORT_SYMBOL(rdma_connect);
3256
3257 static int cma_accept_ib(struct rdma_id_private *id_priv,
3258                          struct rdma_conn_param *conn_param)
3259 {
3260         struct ib_cm_rep_param rep;
3261         int ret;
3262
3263         ret = cma_modify_qp_rtr(id_priv, conn_param);
3264         if (ret)
3265                 goto out;
3266
3267         ret = cma_modify_qp_rts(id_priv, conn_param);
3268         if (ret)
3269                 goto out;
3270
3271         memset(&rep, 0, sizeof rep);
3272         rep.qp_num = id_priv->qp_num;
3273         rep.starting_psn = id_priv->seq_num;
3274         rep.private_data = conn_param->private_data;
3275         rep.private_data_len = conn_param->private_data_len;
3276         rep.responder_resources = conn_param->responder_resources;
3277         rep.initiator_depth = conn_param->initiator_depth;
3278         rep.failover_accepted = 0;
3279         rep.flow_control = conn_param->flow_control;
3280         rep.rnr_retry_count = min_t(u8, 7, conn_param->rnr_retry_count);
3281         rep.srq = id_priv->srq ? 1 : 0;
3282
3283         ret = ib_send_cm_rep(id_priv->cm_id.ib, &rep);
3284 out:
3285         return ret;
3286 }
3287
3288 static int cma_accept_iw(struct rdma_id_private *id_priv,
3289                   struct rdma_conn_param *conn_param)
3290 {
3291         struct iw_cm_conn_param iw_param;
3292         int ret;
3293
3294         ret = cma_modify_qp_rtr(id_priv, conn_param);
3295         if (ret)
3296                 return ret;
3297
3298         iw_param.ord = conn_param->initiator_depth;
3299         iw_param.ird = conn_param->responder_resources;
3300         iw_param.private_data = conn_param->private_data;
3301         iw_param.private_data_len = conn_param->private_data_len;
3302         if (id_priv->id.qp) {
3303                 iw_param.qpn = id_priv->qp_num;
3304         } else
3305                 iw_param.qpn = conn_param->qp_num;
3306
3307         return iw_cm_accept(id_priv->cm_id.iw, &iw_param);
3308 }
3309
3310 static int cma_send_sidr_rep(struct rdma_id_private *id_priv,
3311                              enum ib_cm_sidr_status status, u32 qkey,
3312                              const void *private_data, int private_data_len)
3313 {
3314         struct ib_cm_sidr_rep_param rep;
3315         int ret;
3316
3317         memset(&rep, 0, sizeof rep);
3318         rep.status = status;
3319         if (status == IB_SIDR_SUCCESS) {
3320                 ret = cma_set_qkey(id_priv, qkey);
3321                 if (ret)
3322                         return ret;
3323                 rep.qp_num = id_priv->qp_num;
3324                 rep.qkey = id_priv->qkey;
3325         }
3326         rep.private_data = private_data;
3327         rep.private_data_len = private_data_len;
3328
3329         return ib_send_cm_sidr_rep(id_priv->cm_id.ib, &rep);
3330 }
3331
3332 int rdma_accept(struct rdma_cm_id *id, struct rdma_conn_param *conn_param)
3333 {
3334         struct rdma_id_private *id_priv;
3335         int ret;
3336
3337         id_priv = container_of(id, struct rdma_id_private, id);
3338
3339         id_priv->owner = task_pid_nr(current);
3340
3341         if (!cma_comp(id_priv, RDMA_CM_CONNECT))
3342                 return -EINVAL;
3343
3344         if (!id->qp && conn_param) {
3345                 id_priv->qp_num = conn_param->qp_num;
3346                 id_priv->srq = conn_param->srq;
3347         }
3348
3349         if (rdma_cap_ib_cm(id->device, id->port_num)) {
3350                 if (id->qp_type == IB_QPT_UD) {
3351                         if (conn_param)
3352                                 ret = cma_send_sidr_rep(id_priv, IB_SIDR_SUCCESS,
3353                                                         conn_param->qkey,
3354                                                         conn_param->private_data,
3355                                                         conn_param->private_data_len);
3356                         else
3357                                 ret = cma_send_sidr_rep(id_priv, IB_SIDR_SUCCESS,
3358                                                         0, NULL, 0);
3359                 } else {
3360                         if (conn_param)
3361                                 ret = cma_accept_ib(id_priv, conn_param);
3362                         else
3363                                 ret = cma_rep_recv(id_priv);
3364                 }
3365         } else if (rdma_cap_iw_cm(id->device, id->port_num))
3366                 ret = cma_accept_iw(id_priv, conn_param);
3367         else
3368                 ret = -ENOSYS;
3369
3370         if (ret)
3371                 goto reject;
3372
3373         return 0;
3374 reject:
3375         cma_modify_qp_err(id_priv);
3376         rdma_reject(id, NULL, 0);
3377         return ret;
3378 }
3379 EXPORT_SYMBOL(rdma_accept);
3380
3381 int rdma_notify(struct rdma_cm_id *id, enum ib_event_type event)
3382 {
3383         struct rdma_id_private *id_priv;
3384         int ret;
3385
3386         id_priv = container_of(id, struct rdma_id_private, id);
3387         if (!id_priv->cm_id.ib)
3388                 return -EINVAL;
3389
3390         switch (id->device->node_type) {
3391         case RDMA_NODE_IB_CA:
3392                 ret = ib_cm_notify(id_priv->cm_id.ib, event);
3393                 break;
3394         default:
3395                 ret = 0;
3396                 break;
3397         }
3398         return ret;
3399 }
3400 EXPORT_SYMBOL(rdma_notify);
3401
3402 int rdma_reject(struct rdma_cm_id *id, const void *private_data,
3403                 u8 private_data_len)
3404 {
3405         struct rdma_id_private *id_priv;
3406         int ret;
3407
3408         id_priv = container_of(id, struct rdma_id_private, id);
3409         if (!id_priv->cm_id.ib)
3410                 return -EINVAL;
3411
3412         if (rdma_cap_ib_cm(id->device, id->port_num)) {
3413                 if (id->qp_type == IB_QPT_UD)
3414                         ret = cma_send_sidr_rep(id_priv, IB_SIDR_REJECT, 0,
3415                                                 private_data, private_data_len);
3416                 else
3417                         ret = ib_send_cm_rej(id_priv->cm_id.ib,
3418                                              IB_CM_REJ_CONSUMER_DEFINED, NULL,
3419                                              0, private_data, private_data_len);
3420         } else if (rdma_cap_iw_cm(id->device, id->port_num)) {
3421                 ret = iw_cm_reject(id_priv->cm_id.iw,
3422                                    private_data, private_data_len);
3423         } else
3424                 ret = -ENOSYS;
3425
3426         return ret;
3427 }
3428 EXPORT_SYMBOL(rdma_reject);
3429
3430 int rdma_disconnect(struct rdma_cm_id *id)
3431 {
3432         struct rdma_id_private *id_priv;
3433         int ret;
3434
3435         id_priv = container_of(id, struct rdma_id_private, id);
3436         if (!id_priv->cm_id.ib)
3437                 return -EINVAL;
3438
3439         if (rdma_cap_ib_cm(id->device, id->port_num)) {
3440                 ret = cma_modify_qp_err(id_priv);
3441                 if (ret)
3442                         goto out;
3443                 /* Initiate or respond to a disconnect. */
3444                 if (ib_send_cm_dreq(id_priv->cm_id.ib, NULL, 0))
3445                         ib_send_cm_drep(id_priv->cm_id.ib, NULL, 0);
3446         } else if (rdma_cap_iw_cm(id->device, id->port_num)) {
3447                 ret = iw_cm_disconnect(id_priv->cm_id.iw, 0);
3448         } else
3449                 ret = -EINVAL;
3450
3451 out:
3452         return ret;
3453 }
3454 EXPORT_SYMBOL(rdma_disconnect);
3455
3456 static int cma_ib_mc_handler(int status, struct ib_sa_multicast *multicast)
3457 {
3458         struct rdma_id_private *id_priv;
3459         struct cma_multicast *mc = multicast->context;
3460         struct rdma_cm_event event;
3461         int ret;
3462
3463         id_priv = mc->id_priv;
3464         if (cma_disable_callback(id_priv, RDMA_CM_ADDR_BOUND) &&
3465             cma_disable_callback(id_priv, RDMA_CM_ADDR_RESOLVED))
3466                 return 0;
3467
3468         if (!status)
3469                 status = cma_set_qkey(id_priv, be32_to_cpu(multicast->rec.qkey));
3470         mutex_lock(&id_priv->qp_mutex);
3471         if (!status && id_priv->id.qp)
3472                 status = ib_attach_mcast(id_priv->id.qp, &multicast->rec.mgid,
3473                                          be16_to_cpu(multicast->rec.mlid));
3474         mutex_unlock(&id_priv->qp_mutex);
3475
3476         memset(&event, 0, sizeof event);
3477         event.status = status;
3478         event.param.ud.private_data = mc->context;
3479         if (!status) {
3480                 event.event = RDMA_CM_EVENT_MULTICAST_JOIN;
3481                 ib_init_ah_from_mcmember(id_priv->id.device,
3482                                          id_priv->id.port_num, &multicast->rec,
3483                                          &event.param.ud.ah_attr);
3484                 event.param.ud.qp_num = 0xFFFFFF;
3485                 event.param.ud.qkey = be32_to_cpu(multicast->rec.qkey);
3486         } else
3487                 event.event = RDMA_CM_EVENT_MULTICAST_ERROR;
3488
3489         ret = id_priv->id.event_handler(&id_priv->id, &event);
3490         if (ret) {
3491                 cma_exch(id_priv, RDMA_CM_DESTROYING);
3492                 mutex_unlock(&id_priv->handler_mutex);
3493                 rdma_destroy_id(&id_priv->id);
3494                 return 0;
3495         }
3496
3497         mutex_unlock(&id_priv->handler_mutex);
3498         return 0;
3499 }
3500
3501 static void cma_set_mgid(struct rdma_id_private *id_priv,
3502                          struct sockaddr *addr, union ib_gid *mgid)
3503 {
3504         unsigned char mc_map[MAX_ADDR_LEN];
3505         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
3506         struct sockaddr_in *sin = (struct sockaddr_in *) addr;
3507         struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *) addr;
3508
3509         if (cma_any_addr(addr)) {
3510                 memset(mgid, 0, sizeof *mgid);
3511         } else if ((addr->sa_family == AF_INET6) &&
3512                    ((be32_to_cpu(sin6->sin6_addr.s6_addr32[0]) & 0xFFF0FFFF) ==
3513                                                                  0xFF10A01B)) {
3514                 /* IPv6 address is an SA assigned MGID. */
3515                 memcpy(mgid, &sin6->sin6_addr, sizeof *mgid);
3516         } else if (addr->sa_family == AF_IB) {
3517                 memcpy(mgid, &((struct sockaddr_ib *) addr)->sib_addr, sizeof *mgid);
3518         } else if ((addr->sa_family == AF_INET6)) {
3519                 ipv6_ib_mc_map(&sin6->sin6_addr, dev_addr->broadcast, mc_map);
3520                 if (id_priv->id.ps == RDMA_PS_UDP)
3521                         mc_map[7] = 0x01;       /* Use RDMA CM signature */
3522                 *mgid = *(union ib_gid *) (mc_map + 4);
3523         } else {
3524                 ip_ib_mc_map(sin->sin_addr.s_addr, dev_addr->broadcast, mc_map);
3525                 if (id_priv->id.ps == RDMA_PS_UDP)
3526                         mc_map[7] = 0x01;       /* Use RDMA CM signature */
3527                 *mgid = *(union ib_gid *) (mc_map + 4);
3528         }
3529 }
3530
3531 static int cma_join_ib_multicast(struct rdma_id_private *id_priv,
3532                                  struct cma_multicast *mc)
3533 {
3534         struct ib_sa_mcmember_rec rec;
3535         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
3536         ib_sa_comp_mask comp_mask;
3537         int ret;
3538
3539         ib_addr_get_mgid(dev_addr, &rec.mgid);
3540         ret = ib_sa_get_mcmember_rec(id_priv->id.device, id_priv->id.port_num,
3541                                      &rec.mgid, &rec);
3542         if (ret)
3543                 return ret;
3544
3545         ret = cma_set_qkey(id_priv, 0);
3546         if (ret)
3547                 return ret;
3548
3549         cma_set_mgid(id_priv, (struct sockaddr *) &mc->addr, &rec.mgid);
3550         rec.qkey = cpu_to_be32(id_priv->qkey);
3551         rdma_addr_get_sgid(dev_addr, &rec.port_gid);
3552         rec.pkey = cpu_to_be16(ib_addr_get_pkey(dev_addr));
3553         rec.join_state = 1;
3554
3555         comp_mask = IB_SA_MCMEMBER_REC_MGID | IB_SA_MCMEMBER_REC_PORT_GID |
3556                     IB_SA_MCMEMBER_REC_PKEY | IB_SA_MCMEMBER_REC_JOIN_STATE |
3557                     IB_SA_MCMEMBER_REC_QKEY | IB_SA_MCMEMBER_REC_SL |
3558                     IB_SA_MCMEMBER_REC_FLOW_LABEL |
3559                     IB_SA_MCMEMBER_REC_TRAFFIC_CLASS;
3560
3561         if (id_priv->id.ps == RDMA_PS_IPOIB)
3562                 comp_mask |= IB_SA_MCMEMBER_REC_RATE |
3563                              IB_SA_MCMEMBER_REC_RATE_SELECTOR |
3564                              IB_SA_MCMEMBER_REC_MTU_SELECTOR |
3565                              IB_SA_MCMEMBER_REC_MTU |
3566                              IB_SA_MCMEMBER_REC_HOP_LIMIT;
3567
3568         mc->multicast.ib = ib_sa_join_multicast(&sa_client, id_priv->id.device,
3569                                                 id_priv->id.port_num, &rec,
3570                                                 comp_mask, GFP_KERNEL,
3571                                                 cma_ib_mc_handler, mc);
3572         return PTR_ERR_OR_ZERO(mc->multicast.ib);
3573 }
3574
3575 static void iboe_mcast_work_handler(struct work_struct *work)
3576 {
3577         struct iboe_mcast_work *mw = container_of(work, struct iboe_mcast_work, work);
3578         struct cma_multicast *mc = mw->mc;
3579         struct ib_sa_multicast *m = mc->multicast.ib;
3580
3581         mc->multicast.ib->context = mc;
3582         cma_ib_mc_handler(0, m);
3583         kref_put(&mc->mcref, release_mc);
3584         kfree(mw);
3585 }
3586
3587 static void cma_iboe_set_mgid(struct sockaddr *addr, union ib_gid *mgid)
3588 {
3589         struct sockaddr_in *sin = (struct sockaddr_in *)addr;
3590         struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)addr;
3591
3592         if (cma_any_addr(addr)) {
3593                 memset(mgid, 0, sizeof *mgid);
3594         } else if (addr->sa_family == AF_INET6) {
3595                 memcpy(mgid, &sin6->sin6_addr, sizeof *mgid);
3596         } else {
3597                 mgid->raw[0] = 0xff;
3598                 mgid->raw[1] = 0x0e;
3599                 mgid->raw[2] = 0;
3600                 mgid->raw[3] = 0;
3601                 mgid->raw[4] = 0;
3602                 mgid->raw[5] = 0;
3603                 mgid->raw[6] = 0;
3604                 mgid->raw[7] = 0;
3605                 mgid->raw[8] = 0;
3606                 mgid->raw[9] = 0;
3607                 mgid->raw[10] = 0xff;
3608                 mgid->raw[11] = 0xff;
3609                 *(__be32 *)(&mgid->raw[12]) = sin->sin_addr.s_addr;
3610         }
3611 }
3612
3613 static int cma_iboe_join_multicast(struct rdma_id_private *id_priv,
3614                                    struct cma_multicast *mc)
3615 {
3616         struct iboe_mcast_work *work;
3617         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
3618         int err;
3619         struct sockaddr *addr = (struct sockaddr *)&mc->addr;
3620         struct net_device *ndev = NULL;
3621
3622         if (cma_zero_addr((struct sockaddr *)&mc->addr))
3623                 return -EINVAL;
3624
3625         work = kzalloc(sizeof *work, GFP_KERNEL);
3626         if (!work)
3627                 return -ENOMEM;
3628
3629         mc->multicast.ib = kzalloc(sizeof(struct ib_sa_multicast), GFP_KERNEL);
3630         if (!mc->multicast.ib) {
3631                 err = -ENOMEM;
3632                 goto out1;
3633         }
3634
3635         cma_iboe_set_mgid(addr, &mc->multicast.ib->rec.mgid);
3636
3637         mc->multicast.ib->rec.pkey = cpu_to_be16(0xffff);
3638         if (id_priv->id.ps == RDMA_PS_UDP)
3639                 mc->multicast.ib->rec.qkey = cpu_to_be32(RDMA_UDP_QKEY);
3640
3641         if (dev_addr->bound_dev_if)
3642                 ndev = dev_get_by_index(&init_net, dev_addr->bound_dev_if);
3643         if (!ndev) {
3644                 err = -ENODEV;
3645                 goto out2;
3646         }
3647         mc->multicast.ib->rec.rate = iboe_get_rate(ndev);
3648         mc->multicast.ib->rec.hop_limit = 1;
3649         mc->multicast.ib->rec.mtu = iboe_get_mtu(ndev->mtu);
3650         dev_put(ndev);
3651         if (!mc->multicast.ib->rec.mtu) {
3652                 err = -EINVAL;
3653                 goto out2;
3654         }
3655         rdma_ip2gid((struct sockaddr *)&id_priv->id.route.addr.src_addr,
3656                     &mc->multicast.ib->rec.port_gid);
3657         work->id = id_priv;
3658         work->mc = mc;
3659         INIT_WORK(&work->work, iboe_mcast_work_handler);
3660         kref_get(&mc->mcref);
3661         queue_work(cma_wq, &work->work);
3662
3663         return 0;
3664
3665 out2:
3666         kfree(mc->multicast.ib);
3667 out1:
3668         kfree(work);
3669         return err;
3670 }
3671
3672 int rdma_join_multicast(struct rdma_cm_id *id, struct sockaddr *addr,
3673                         void *context)
3674 {
3675         struct rdma_id_private *id_priv;
3676         struct cma_multicast *mc;
3677         int ret;
3678
3679         id_priv = container_of(id, struct rdma_id_private, id);
3680         if (!cma_comp(id_priv, RDMA_CM_ADDR_BOUND) &&
3681             !cma_comp(id_priv, RDMA_CM_ADDR_RESOLVED))
3682                 return -EINVAL;
3683
3684         mc = kmalloc(sizeof *mc, GFP_KERNEL);
3685         if (!mc)
3686                 return -ENOMEM;
3687
3688         memcpy(&mc->addr, addr, rdma_addr_size(addr));
3689         mc->context = context;
3690         mc->id_priv = id_priv;
3691
3692         spin_lock(&id_priv->lock);
3693         list_add(&mc->list, &id_priv->mc_list);
3694         spin_unlock(&id_priv->lock);
3695
3696         if (rdma_protocol_roce(id->device, id->port_num)) {
3697                 kref_init(&mc->mcref);
3698                 ret = cma_iboe_join_multicast(id_priv, mc);
3699         } else if (rdma_cap_ib_mcast(id->device, id->port_num))
3700                 ret = cma_join_ib_multicast(id_priv, mc);
3701         else
3702                 ret = -ENOSYS;
3703
3704         if (ret) {
3705                 spin_lock_irq(&id_priv->lock);
3706                 list_del(&mc->list);
3707                 spin_unlock_irq(&id_priv->lock);
3708                 kfree(mc);
3709         }
3710         return ret;
3711 }
3712 EXPORT_SYMBOL(rdma_join_multicast);
3713
3714 void rdma_leave_multicast(struct rdma_cm_id *id, struct sockaddr *addr)
3715 {
3716         struct rdma_id_private *id_priv;
3717         struct cma_multicast *mc;
3718
3719         id_priv = container_of(id, struct rdma_id_private, id);
3720         spin_lock_irq(&id_priv->lock);
3721         list_for_each_entry(mc, &id_priv->mc_list, list) {
3722                 if (!memcmp(&mc->addr, addr, rdma_addr_size(addr))) {
3723                         list_del(&mc->list);
3724                         spin_unlock_irq(&id_priv->lock);
3725
3726                         if (id->qp)
3727                                 ib_detach_mcast(id->qp,
3728                                                 &mc->multicast.ib->rec.mgid,
3729                                                 be16_to_cpu(mc->multicast.ib->rec.mlid));
3730
3731                         BUG_ON(id_priv->cma_dev->device != id->device);
3732
3733                         if (rdma_cap_ib_mcast(id->device, id->port_num)) {
3734                                 ib_sa_free_multicast(mc->multicast.ib);
3735                                 kfree(mc);
3736                         } else if (rdma_protocol_roce(id->device, id->port_num))
3737                                 kref_put(&mc->mcref, release_mc);
3738
3739                         return;
3740                 }
3741         }
3742         spin_unlock_irq(&id_priv->lock);
3743 }
3744 EXPORT_SYMBOL(rdma_leave_multicast);
3745
3746 static int cma_netdev_change(struct net_device *ndev, struct rdma_id_private *id_priv)
3747 {
3748         struct rdma_dev_addr *dev_addr;
3749         struct cma_ndev_work *work;
3750
3751         dev_addr = &id_priv->id.route.addr.dev_addr;
3752
3753         if ((dev_addr->bound_dev_if == ndev->ifindex) &&
3754             memcmp(dev_addr->src_dev_addr, ndev->dev_addr, ndev->addr_len)) {
3755                 printk(KERN_INFO "RDMA CM addr change for ndev %s used by id %p\n",
3756                        ndev->name, &id_priv->id);
3757                 work = kzalloc(sizeof *work, GFP_KERNEL);
3758                 if (!work)
3759                         return -ENOMEM;
3760
3761                 INIT_WORK(&work->work, cma_ndev_work_handler);
3762                 work->id = id_priv;
3763                 work->event.event = RDMA_CM_EVENT_ADDR_CHANGE;
3764                 atomic_inc(&id_priv->refcount);
3765                 queue_work(cma_wq, &work->work);
3766         }
3767
3768         return 0;
3769 }
3770
3771 static int cma_netdev_callback(struct notifier_block *self, unsigned long event,
3772                                void *ptr)
3773 {
3774         struct net_device *ndev = netdev_notifier_info_to_dev(ptr);
3775         struct cma_device *cma_dev;
3776         struct rdma_id_private *id_priv;
3777         int ret = NOTIFY_DONE;
3778
3779         if (dev_net(ndev) != &init_net)
3780                 return NOTIFY_DONE;
3781
3782         if (event != NETDEV_BONDING_FAILOVER)
3783                 return NOTIFY_DONE;
3784
3785         if (!(ndev->flags & IFF_MASTER) || !(ndev->priv_flags & IFF_BONDING))
3786                 return NOTIFY_DONE;
3787
3788         mutex_lock(&lock);
3789         list_for_each_entry(cma_dev, &dev_list, list)
3790                 list_for_each_entry(id_priv, &cma_dev->id_list, list) {
3791                         ret = cma_netdev_change(ndev, id_priv);
3792                         if (ret)
3793                                 goto out;
3794                 }
3795
3796 out:
3797         mutex_unlock(&lock);
3798         return ret;
3799 }
3800
3801 static struct notifier_block cma_nb = {
3802         .notifier_call = cma_netdev_callback
3803 };
3804
3805 static void cma_add_one(struct ib_device *device)
3806 {
3807         struct cma_device *cma_dev;
3808         struct rdma_id_private *id_priv;
3809
3810         cma_dev = kmalloc(sizeof *cma_dev, GFP_KERNEL);
3811         if (!cma_dev)
3812                 return;
3813
3814         cma_dev->device = device;
3815
3816         init_completion(&cma_dev->comp);
3817         atomic_set(&cma_dev->refcount, 1);
3818         INIT_LIST_HEAD(&cma_dev->id_list);
3819         ib_set_client_data(device, &cma_client, cma_dev);
3820
3821         mutex_lock(&lock);
3822         list_add_tail(&cma_dev->list, &dev_list);
3823         list_for_each_entry(id_priv, &listen_any_list, list)
3824                 cma_listen_on_dev(id_priv, cma_dev);
3825         mutex_unlock(&lock);
3826 }
3827
3828 static int cma_remove_id_dev(struct rdma_id_private *id_priv)
3829 {
3830         struct rdma_cm_event event;
3831         enum rdma_cm_state state;
3832         int ret = 0;
3833
3834         /* Record that we want to remove the device */
3835         state = cma_exch(id_priv, RDMA_CM_DEVICE_REMOVAL);
3836         if (state == RDMA_CM_DESTROYING)
3837                 return 0;
3838
3839         cma_cancel_operation(id_priv, state);
3840         mutex_lock(&id_priv->handler_mutex);
3841
3842         /* Check for destruction from another callback. */
3843         if (!cma_comp(id_priv, RDMA_CM_DEVICE_REMOVAL))
3844                 goto out;
3845
3846         memset(&event, 0, sizeof event);
3847         event.event = RDMA_CM_EVENT_DEVICE_REMOVAL;
3848         ret = id_priv->id.event_handler(&id_priv->id, &event);
3849 out:
3850         mutex_unlock(&id_priv->handler_mutex);
3851         return ret;
3852 }
3853
3854 static void cma_process_remove(struct cma_device *cma_dev)
3855 {
3856         struct rdma_id_private *id_priv;
3857         int ret;
3858
3859         mutex_lock(&lock);
3860         while (!list_empty(&cma_dev->id_list)) {
3861                 id_priv = list_entry(cma_dev->id_list.next,
3862                                      struct rdma_id_private, list);
3863
3864                 list_del(&id_priv->listen_list);
3865                 list_del_init(&id_priv->list);
3866                 atomic_inc(&id_priv->refcount);
3867                 mutex_unlock(&lock);
3868
3869                 ret = id_priv->internal_id ? 1 : cma_remove_id_dev(id_priv);
3870                 cma_deref_id(id_priv);
3871                 if (ret)
3872                         rdma_destroy_id(&id_priv->id);
3873
3874                 mutex_lock(&lock);
3875         }
3876         mutex_unlock(&lock);
3877
3878         cma_deref_dev(cma_dev);
3879         wait_for_completion(&cma_dev->comp);
3880 }
3881
3882 static void cma_remove_one(struct ib_device *device, void *client_data)
3883 {
3884         struct cma_device *cma_dev = client_data;
3885
3886         if (!cma_dev)
3887                 return;
3888
3889         mutex_lock(&lock);
3890         list_del(&cma_dev->list);
3891         mutex_unlock(&lock);
3892
3893         cma_process_remove(cma_dev);
3894         kfree(cma_dev);
3895 }
3896
3897 static int cma_get_id_stats(struct sk_buff *skb, struct netlink_callback *cb)
3898 {
3899         struct nlmsghdr *nlh;
3900         struct rdma_cm_id_stats *id_stats;
3901         struct rdma_id_private *id_priv;
3902         struct rdma_cm_id *id = NULL;
3903         struct cma_device *cma_dev;
3904         int i_dev = 0, i_id = 0;
3905
3906         /*
3907          * We export all of the IDs as a sequence of messages.  Each
3908          * ID gets its own netlink message.
3909          */
3910         mutex_lock(&lock);
3911
3912         list_for_each_entry(cma_dev, &dev_list, list) {
3913                 if (i_dev < cb->args[0]) {
3914                         i_dev++;
3915                         continue;
3916                 }
3917
3918                 i_id = 0;
3919                 list_for_each_entry(id_priv, &cma_dev->id_list, list) {
3920                         if (i_id < cb->args[1]) {
3921                                 i_id++;
3922                                 continue;
3923                         }
3924
3925                         id_stats = ibnl_put_msg(skb, &nlh, cb->nlh->nlmsg_seq,
3926                                                 sizeof *id_stats, RDMA_NL_RDMA_CM,
3927                                                 RDMA_NL_RDMA_CM_ID_STATS,
3928                                                 NLM_F_MULTI);
3929                         if (!id_stats)
3930                                 goto out;
3931
3932                         memset(id_stats, 0, sizeof *id_stats);
3933                         id = &id_priv->id;
3934                         id_stats->node_type = id->route.addr.dev_addr.dev_type;
3935                         id_stats->port_num = id->port_num;
3936                         id_stats->bound_dev_if =
3937                                 id->route.addr.dev_addr.bound_dev_if;
3938
3939                         if (ibnl_put_attr(skb, nlh,
3940                                           rdma_addr_size(cma_src_addr(id_priv)),
3941                                           cma_src_addr(id_priv),
3942                                           RDMA_NL_RDMA_CM_ATTR_SRC_ADDR))
3943                                 goto out;
3944                         if (ibnl_put_attr(skb, nlh,
3945                                           rdma_addr_size(cma_src_addr(id_priv)),
3946                                           cma_dst_addr(id_priv),
3947                                           RDMA_NL_RDMA_CM_ATTR_DST_ADDR))
3948                                 goto out;
3949
3950                         id_stats->pid           = id_priv->owner;
3951                         id_stats->port_space    = id->ps;
3952                         id_stats->cm_state      = id_priv->state;
3953                         id_stats->qp_num        = id_priv->qp_num;
3954                         id_stats->qp_type       = id->qp_type;
3955
3956                         i_id++;
3957                 }
3958
3959                 cb->args[1] = 0;
3960                 i_dev++;
3961         }
3962
3963 out:
3964         mutex_unlock(&lock);
3965         cb->args[0] = i_dev;
3966         cb->args[1] = i_id;
3967
3968         return skb->len;
3969 }
3970
3971 static const struct ibnl_client_cbs cma_cb_table[] = {
3972         [RDMA_NL_RDMA_CM_ID_STATS] = { .dump = cma_get_id_stats,
3973                                        .module = THIS_MODULE },
3974 };
3975
3976 static int __init cma_init(void)
3977 {
3978         int ret;
3979
3980         cma_wq = create_singlethread_workqueue("rdma_cm");
3981         if (!cma_wq)
3982                 return -ENOMEM;
3983
3984         ib_sa_register_client(&sa_client);
3985         rdma_addr_register_client(&addr_client);
3986         register_netdevice_notifier(&cma_nb);
3987
3988         ret = ib_register_client(&cma_client);
3989         if (ret)
3990                 goto err;
3991
3992         if (ibnl_add_client(RDMA_NL_RDMA_CM, RDMA_NL_RDMA_CM_NUM_OPS, cma_cb_table))
3993                 printk(KERN_WARNING "RDMA CMA: failed to add netlink callback\n");
3994
3995         return 0;
3996
3997 err:
3998         unregister_netdevice_notifier(&cma_nb);
3999         rdma_addr_unregister_client(&addr_client);
4000         ib_sa_unregister_client(&sa_client);
4001         destroy_workqueue(cma_wq);
4002         return ret;
4003 }
4004
4005 static void __exit cma_cleanup(void)
4006 {
4007         ibnl_remove_client(RDMA_NL_RDMA_CM);
4008         ib_unregister_client(&cma_client);
4009         unregister_netdevice_notifier(&cma_nb);
4010         rdma_addr_unregister_client(&addr_client);
4011         ib_sa_unregister_client(&sa_client);
4012         destroy_workqueue(cma_wq);
4013         idr_destroy(&tcp_ps);
4014         idr_destroy(&udp_ps);
4015         idr_destroy(&ipoib_ps);
4016         idr_destroy(&ib_ps);
4017 }
4018
4019 module_init(cma_init);
4020 module_exit(cma_cleanup);