kernel/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/net_namespace.h>
  48 #include <net/netns/generic.h>
  49 #include <net/tcp.h>
  50 #include <net/ipv6.h>
  51 #include <net/ip_fib.h>
  52 #include <net/ip6_route.h>
  53
  54 #include <rdma/rdma_cm.h>
  55 #include <rdma/rdma_cm_ib.h>
  56 #include <rdma/rdma_netlink.h>
  57 #include <rdma/ib.h>
  58 #include <rdma/ib_cache.h>
  59 #include <rdma/ib_cm.h>
  60 #include <rdma/ib_sa.h>
  61 #include <rdma/iw_cm.h>
  62
  63 MODULE_AUTHOR("Sean Hefty");
  64 MODULE_DESCRIPTION("Generic RDMA CM Agent");
  65 MODULE_LICENSE("Dual BSD/GPL");
  66
  67 #define CMA_CM_RESPONSE_TIMEOUT 20
  68 #define CMA_MAX_CM_RETRIES 15
  69 #define CMA_CM_MRA_SETTING (IB_CM_MRA_FLAG_DELAY | 24)
  70 #define CMA_IBOE_PACKET_LIFETIME 18
  71
  72 static const char * const cma_events[] = {
  73         [RDMA_CM_EVENT_ADDR_RESOLVED]    = "address resolved",
  74         [RDMA_CM_EVENT_ADDR_ERROR]       = "address error",
  75         [RDMA_CM_EVENT_ROUTE_RESOLVED]   = "route resolved ",
  76         [RDMA_CM_EVENT_ROUTE_ERROR]      = "route error",
  77         [RDMA_CM_EVENT_CONNECT_REQUEST]  = "connect request",
  78         [RDMA_CM_EVENT_CONNECT_RESPONSE] = "connect response",
  79         [RDMA_CM_EVENT_CONNECT_ERROR]    = "connect error",
  80         [RDMA_CM_EVENT_UNREACHABLE]      = "unreachable",
  81         [RDMA_CM_EVENT_REJECTED]         = "rejected",
  82         [RDMA_CM_EVENT_ESTABLISHED]      = "established",
  83         [RDMA_CM_EVENT_DISCONNECTED]     = "disconnected",
  84         [RDMA_CM_EVENT_DEVICE_REMOVAL]   = "device removal",
  85         [RDMA_CM_EVENT_MULTICAST_JOIN]   = "multicast join",
  86         [RDMA_CM_EVENT_MULTICAST_ERROR]  = "multicast error",
  87         [RDMA_CM_EVENT_ADDR_CHANGE]      = "address change",
  88         [RDMA_CM_EVENT_TIMEWAIT_EXIT]    = "timewait exit",
  89 };
  90
  91 const char *__attribute_const__ rdma_event_msg(enum rdma_cm_event_type event)
  92 {
  93         size_t index = event;
  94
  95         return (index < ARRAY_SIZE(cma_events) && cma_events[index]) ?
  96                         cma_events[index] : "unrecognized event";
  97 }
  98 EXPORT_SYMBOL(rdma_event_msg);
  99
 100 static void cma_add_one(struct ib_device *device);
 101 static void cma_remove_one(struct ib_device *device, void *client_data);
 102
 103 static struct ib_client cma_client = {
 104         .name   = "cma",
 105         .add    = cma_add_one,
 106         .remove = cma_remove_one
 107 };
 108
 109 static struct ib_sa_client sa_client;
 110 static struct rdma_addr_client addr_client;
 111 static LIST_HEAD(dev_list);
 112 static LIST_HEAD(listen_any_list);
 113 static DEFINE_MUTEX(lock);
 114 static struct workqueue_struct *cma_wq;
 115 static int cma_pernet_id;
 116
 117 struct cma_pernet {
 118         struct idr tcp_ps;
 119         struct idr udp_ps;
 120         struct idr ipoib_ps;
 121         struct idr ib_ps;
 122 };
 123
 124 static struct cma_pernet *cma_pernet(struct net *net)
 125 {
 126         return net_generic(net, cma_pernet_id);
 127 }
 128
 129 static struct idr *cma_pernet_idr(struct net *net, enum rdma_port_space ps)
 130 {
 131         struct cma_pernet *pernet = cma_pernet(net);
 132
 133         switch (ps) {
 134         case RDMA_PS_TCP:
 135                 return &pernet->tcp_ps;
 136         case RDMA_PS_UDP:
 137                 return &pernet->udp_ps;
 138         case RDMA_PS_IPOIB:
 139                 return &pernet->ipoib_ps;
 140         case RDMA_PS_IB:
 141                 return &pernet->ib_ps;
 142         default:
 143                 return NULL;
 144         }
 145 }
 146
 147 struct cma_device {
 148         struct list_head        list;
 149         struct ib_device        *device;
 150         struct completion       comp;
 151         atomic_t                refcount;
 152         struct list_head        id_list;
 153 };
 154
 155 struct rdma_bind_list {
 156         enum rdma_port_space    ps;
 157         struct hlist_head       owners;
 158         unsigned short          port;
 159 };
 160
 161 static int cma_ps_alloc(struct net *net, enum rdma_port_space ps,
 162                         struct rdma_bind_list *bind_list, int snum)
 163 {
 164         struct idr *idr = cma_pernet_idr(net, ps);
 165
 166         return idr_alloc(idr, bind_list, snum, snum + 1, GFP_KERNEL);
 167 }
 168
 169 static struct rdma_bind_list *cma_ps_find(struct net *net,
 170                                           enum rdma_port_space ps, int snum)
 171 {
 172         struct idr *idr = cma_pernet_idr(net, ps);
 173
 174         return idr_find(idr, snum);
 175 }
 176
 177 static void cma_ps_remove(struct net *net, enum rdma_port_space ps, int snum)
 178 {
 179         struct idr *idr = cma_pernet_idr(net, ps);
 180
 181         idr_remove(idr, snum);
 182 }
 183
 184 enum {
 185         CMA_OPTION_AFONLY,
 186 };
 187
 188 /*
 189  * Device removal can occur at anytime, so we need extra handling to
 190  * serialize notifying the user of device removal with other callbacks.
 191  * We do this by disabling removal notification while a callback is in process,
 192  * and reporting it after the callback completes.
 193  */
 194 struct rdma_id_private {
 195         struct rdma_cm_id       id;
 196
 197         struct rdma_bind_list   *bind_list;
 198         struct hlist_node       node;
 199         struct list_head        list; /* listen_any_list or cma_device.list */
 200         struct list_head        listen_list; /* per device listens */
 201         struct cma_device       *cma_dev;
 202         struct list_head        mc_list;
 203
 204         int                     internal_id;
 205         enum rdma_cm_state      state;
 206         spinlock_t              lock;
 207         struct mutex            qp_mutex;
 208
 209         struct completion       comp;
 210         atomic_t                refcount;
 211         struct mutex            handler_mutex;
 212
 213         int                     backlog;
 214         int                     timeout_ms;
 215         struct ib_sa_query      *query;
 216         int                     query_id;
 217         union {
 218                 struct ib_cm_id *ib;
 219                 struct iw_cm_id *iw;
 220         } cm_id;
 221
 222         u32                     seq_num;
 223         u32                     qkey;
 224         u32                     qp_num;
 225         pid_t                   owner;
 226         u32                     options;
 227         u8                      srq;
 228         u8                      tos;
 229         u8                      reuseaddr;
 230         u8                      afonly;
 231 };
 232
 233 struct cma_multicast {
 234         struct rdma_id_private *id_priv;
 235         union {
 236                 struct ib_sa_multicast *ib;
 237         } multicast;
 238         struct list_head        list;
 239         void                    *context;
 240         struct sockaddr_storage addr;
 241         struct kref             mcref;
 242 };
 243
 244 struct cma_work {
 245         struct work_struct      work;
 246         struct rdma_id_private  *id;
 247         enum rdma_cm_state      old_state;
 248         enum rdma_cm_state      new_state;
 249         struct rdma_cm_event    event;
 250 };
 251
 252 struct cma_ndev_work {
 253         struct work_struct      work;
 254         struct rdma_id_private  *id;
 255         struct rdma_cm_event    event;
 256 };
 257
 258 struct iboe_mcast_work {
 259         struct work_struct       work;
 260         struct rdma_id_private  *id;
 261         struct cma_multicast    *mc;
 262 };
 263
 264 union cma_ip_addr {
 265         struct in6_addr ip6;
 266         struct {
 267                 __be32 pad[3];
 268                 __be32 addr;
 269         } ip4;
 270 };
 271
 272 struct cma_hdr {
 273         u8 cma_version;
 274         u8 ip_version;  /* IP version: 7:4 */
 275         __be16 port;
 276         union cma_ip_addr src_addr;
 277         union cma_ip_addr dst_addr;
 278 };
 279
 280 #define CMA_VERSION 0x00
 281
 282 struct cma_req_info {
 283         struct ib_device *device;
 284         int port;
 285         union ib_gid local_gid;
 286         __be64 service_id;
 287         u16 pkey;
 288         bool has_gid:1;
 289 };
 290
 291 static int cma_comp(struct rdma_id_private *id_priv, enum rdma_cm_state comp)
 292 {
 293         unsigned long flags;
 294         int ret;
 295
 296         spin_lock_irqsave(&id_priv->lock, flags);
 297         ret = (id_priv->state == comp);
 298         spin_unlock_irqrestore(&id_priv->lock, flags);
 299         return ret;
 300 }
 301
 302 static int cma_comp_exch(struct rdma_id_private *id_priv,
 303                          enum rdma_cm_state comp, enum rdma_cm_state exch)
 304 {
 305         unsigned long flags;
 306         int ret;
 307
 308         spin_lock_irqsave(&id_priv->lock, flags);
 309         if ((ret = (id_priv->state == comp)))
 310                 id_priv->state = exch;
 311         spin_unlock_irqrestore(&id_priv->lock, flags);
 312         return ret;
 313 }
 314
 315 static enum rdma_cm_state cma_exch(struct rdma_id_private *id_priv,
 316                                    enum rdma_cm_state exch)
 317 {
 318         unsigned long flags;
 319         enum rdma_cm_state old;
 320
 321         spin_lock_irqsave(&id_priv->lock, flags);
 322         old = id_priv->state;
 323         id_priv->state = exch;
 324         spin_unlock_irqrestore(&id_priv->lock, flags);
 325         return old;
 326 }
 327
 328 static inline u8 cma_get_ip_ver(const struct cma_hdr *hdr)
 329 {
 330         return hdr->ip_version >> 4;
 331 }
 332
 333 static inline void cma_set_ip_ver(struct cma_hdr *hdr, u8 ip_ver)
 334 {
 335         hdr->ip_version = (ip_ver << 4) | (hdr->ip_version & 0xF);
 336 }
 337
 338 static void cma_attach_to_dev(struct rdma_id_private *id_priv,
 339                               struct cma_device *cma_dev)
 340 {
 341         atomic_inc(&cma_dev->refcount);
 342         id_priv->cma_dev = cma_dev;
 343         id_priv->id.device = cma_dev->device;
 344         id_priv->id.route.addr.dev_addr.transport =
 345                 rdma_node_get_transport(cma_dev->device->node_type);
 346         list_add_tail(&id_priv->list, &cma_dev->id_list);
 347 }
 348
 349 static inline void cma_deref_dev(struct cma_device *cma_dev)
 350 {
 351         if (atomic_dec_and_test(&cma_dev->refcount))
 352                 complete(&cma_dev->comp);
 353 }
 354
 355 static inline void release_mc(struct kref *kref)
 356 {
 357         struct cma_multicast *mc = container_of(kref, struct cma_multicast, mcref);
 358
 359         kfree(mc->multicast.ib);
 360         kfree(mc);
 361 }
 362
 363 static void cma_release_dev(struct rdma_id_private *id_priv)
 364 {
 365         mutex_lock(&lock);
 366         list_del(&id_priv->list);
 367         cma_deref_dev(id_priv->cma_dev);
 368         id_priv->cma_dev = NULL;
 369         mutex_unlock(&lock);
 370 }
 371
 372 static inline struct sockaddr *cma_src_addr(struct rdma_id_private *id_priv)
 373 {
 374         return (struct sockaddr *) &id_priv->id.route.addr.src_addr;
 375 }
 376
 377 static inline struct sockaddr *cma_dst_addr(struct rdma_id_private *id_priv)
 378 {
 379         return (struct sockaddr *) &id_priv->id.route.addr.dst_addr;
 380 }
 381
 382 static inline unsigned short cma_family(struct rdma_id_private *id_priv)
 383 {
 384         return id_priv->id.route.addr.src_addr.ss_family;
 385 }
 386
 387 static int cma_set_qkey(struct rdma_id_private *id_priv, u32 qkey)
 388 {
 389         struct ib_sa_mcmember_rec rec;
 390         int ret = 0;
 391
 392         if (id_priv->qkey) {
 393                 if (qkey && id_priv->qkey != qkey)
 394                         return -EINVAL;
 395                 return 0;
 396         }
 397
 398         if (qkey) {
 399                 id_priv->qkey = qkey;
 400                 return 0;
 401         }
 402
 403         switch (id_priv->id.ps) {
 404         case RDMA_PS_UDP:
 405         case RDMA_PS_IB:
 406                 id_priv->qkey = RDMA_UDP_QKEY;
 407                 break;
 408         case RDMA_PS_IPOIB:
 409                 ib_addr_get_mgid(&id_priv->id.route.addr.dev_addr, &rec.mgid);
 410                 ret = ib_sa_get_mcmember_rec(id_priv->id.device,
 411                                              id_priv->id.port_num, &rec.mgid,
 412                                              &rec);
 413                 if (!ret)
 414                         id_priv->qkey = be32_to_cpu(rec.qkey);
 415                 break;
 416         default:
 417                 break;
 418         }
 419         return ret;
 420 }
 421
 422 static void cma_translate_ib(struct sockaddr_ib *sib, struct rdma_dev_addr *dev_addr)
 423 {
 424         dev_addr->dev_type = ARPHRD_INFINIBAND;
 425         rdma_addr_set_sgid(dev_addr, (union ib_gid *) &sib->sib_addr);
 426         ib_addr_set_pkey(dev_addr, ntohs(sib->sib_pkey));
 427 }
 428
 429 static int cma_translate_addr(struct sockaddr *addr, struct rdma_dev_addr *dev_addr)
 430 {
 431         int ret;
 432
 433         if (addr->sa_family != AF_IB) {
 434                 ret = rdma_translate_ip(addr, dev_addr, NULL);
 435         } else {
 436                 cma_translate_ib((struct sockaddr_ib *) addr, dev_addr);
 437                 ret = 0;
 438         }
 439
 440         return ret;
 441 }
 442
 443 static inline int cma_validate_port(struct ib_device *device, u8 port,
 444                                       union ib_gid *gid, int dev_type,
 445                                       int bound_if_index)
 446 {
 447         int ret = -ENODEV;
 448         struct net_device *ndev = NULL;
 449
 450         if ((dev_type == ARPHRD_INFINIBAND) && !rdma_protocol_ib(device, port))
 451                 return ret;
 452
 453         if ((dev_type != ARPHRD_INFINIBAND) && rdma_protocol_ib(device, port))
 454                 return ret;
 455
 456         if (dev_type == ARPHRD_ETHER && rdma_protocol_roce(device, port))
 457                 ndev = dev_get_by_index(&init_net, bound_if_index);
 458
 459         ret = ib_find_cached_gid_by_port(device, gid, port, ndev, NULL);
 460
 461         if (ndev)
 462                 dev_put(ndev);
 463
 464         return ret;
 465 }
 466
 467 static int cma_acquire_dev(struct rdma_id_private *id_priv,
 468                            struct rdma_id_private *listen_id_priv)
 469 {
 470         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
 471         struct cma_device *cma_dev;
 472         union ib_gid gid, iboe_gid, *gidp;
 473         int ret = -ENODEV;
 474         u8 port;
 475
 476         if (dev_addr->dev_type != ARPHRD_INFINIBAND &&
 477             id_priv->id.ps == RDMA_PS_IPOIB)
 478                 return -EINVAL;
 479
 480         mutex_lock(&lock);
 481         rdma_ip2gid((struct sockaddr *)&id_priv->id.route.addr.src_addr,
 482                     &iboe_gid);
 483
 484         memcpy(&gid, dev_addr->src_dev_addr +
 485                rdma_addr_gid_offset(dev_addr), sizeof gid);
 486
 487         if (listen_id_priv) {
 488                 cma_dev = listen_id_priv->cma_dev;
 489                 port = listen_id_priv->id.port_num;
 490                 gidp = rdma_protocol_roce(cma_dev->device, port) ?
 491                        &iboe_gid : &gid;
 492
 493                 ret = cma_validate_port(cma_dev->device, port, gidp,
 494                                         dev_addr->dev_type,
 495                                         dev_addr->bound_dev_if);
 496                 if (!ret) {
 497                         id_priv->id.port_num = port;
 498                         goto out;
 499                 }
 500         }
 501
 502         list_for_each_entry(cma_dev, &dev_list, list) {
 503                 for (port = 1; port <= cma_dev->device->phys_port_cnt; ++port) {
 504                         if (listen_id_priv &&
 505                             listen_id_priv->cma_dev == cma_dev &&
 506                             listen_id_priv->id.port_num == port)
 507                                 continue;
 508
 509                         gidp = rdma_protocol_roce(cma_dev->device, port) ?
 510                                &iboe_gid : &gid;
 511
 512                         ret = cma_validate_port(cma_dev->device, port, gidp,
 513                                                 dev_addr->dev_type,
 514                                                 dev_addr->bound_dev_if);
 515                         if (!ret) {
 516                                 id_priv->id.port_num = port;
 517                                 goto out;
 518                         }
 519                 }
 520         }
 521
 522 out:
 523         if (!ret)
 524                 cma_attach_to_dev(id_priv, cma_dev);
 525
 526         mutex_unlock(&lock);
 527         return ret;
 528 }
 529
 530 /*
 531  * Select the source IB device and address to reach the destination IB address.
 532  */
 533 static int cma_resolve_ib_dev(struct rdma_id_private *id_priv)
 534 {
 535         struct cma_device *cma_dev, *cur_dev;
 536         struct sockaddr_ib *addr;
 537         union ib_gid gid, sgid, *dgid;
 538         u16 pkey, index;
 539         u8 p;
 540         int i;
 541
 542         cma_dev = NULL;
 543         addr = (struct sockaddr_ib *) cma_dst_addr(id_priv);
 544         dgid = (union ib_gid *) &addr->sib_addr;
 545         pkey = ntohs(addr->sib_pkey);
 546
 547         list_for_each_entry(cur_dev, &dev_list, list) {
 548                 for (p = 1; p <= cur_dev->device->phys_port_cnt; ++p) {
 549                         if (!rdma_cap_af_ib(cur_dev->device, p))
 550                                 continue;
 551
 552                         if (ib_find_cached_pkey(cur_dev->device, p, pkey, &index))
 553                                 continue;
 554
 555                         for (i = 0; !ib_get_cached_gid(cur_dev->device, p, i,
 556                                                        &gid, NULL);
 557                              i++) {
 558                                 if (!memcmp(&gid, dgid, sizeof(gid))) {
 559                                         cma_dev = cur_dev;
 560                                         sgid = gid;
 561                                         id_priv->id.port_num = p;
 562                                         goto found;
 563                                 }
 564
 565                                 if (!cma_dev && (gid.global.subnet_prefix ==
 566                                                  dgid->global.subnet_prefix)) {
 567                                         cma_dev = cur_dev;
 568                                         sgid = gid;
 569                                         id_priv->id.port_num = p;
 570                                 }
 571                         }
 572                 }
 573         }
 574
 575         if (!cma_dev)
 576                 return -ENODEV;
 577
 578 found:
 579         cma_attach_to_dev(id_priv, cma_dev);
 580         addr = (struct sockaddr_ib *) cma_src_addr(id_priv);
 581         memcpy(&addr->sib_addr, &sgid, sizeof sgid);
 582         cma_translate_ib(addr, &id_priv->id.route.addr.dev_addr);
 583         return 0;
 584 }
 585
 586 static void cma_deref_id(struct rdma_id_private *id_priv)
 587 {
 588         if (atomic_dec_and_test(&id_priv->refcount))
 589                 complete(&id_priv->comp);
 590 }
 591
 592 static int cma_disable_callback(struct rdma_id_private *id_priv,
 593                                 enum rdma_cm_state state)
 594 {
 595         mutex_lock(&id_priv->handler_mutex);
 596         if (id_priv->state != state) {
 597                 mutex_unlock(&id_priv->handler_mutex);
 598                 return -EINVAL;
 599         }
 600         return 0;
 601 }
 602
 603 struct rdma_cm_id *rdma_create_id(struct net *net,
 604                                   rdma_cm_event_handler event_handler,
 605                                   void *context, enum rdma_port_space ps,
 606                                   enum ib_qp_type qp_type)
 607 {
 608         struct rdma_id_private *id_priv;
 609
 610         id_priv = kzalloc(sizeof *id_priv, GFP_KERNEL);
 611         if (!id_priv)
 612                 return ERR_PTR(-ENOMEM);
 613
 614         id_priv->owner = task_pid_nr(current);
 615         id_priv->state = RDMA_CM_IDLE;
 616         id_priv->id.context = context;
 617         id_priv->id.event_handler = event_handler;
 618         id_priv->id.ps = ps;
 619         id_priv->id.qp_type = qp_type;
 620         spin_lock_init(&id_priv->lock);
 621         mutex_init(&id_priv->qp_mutex);
 622         init_completion(&id_priv->comp);
 623         atomic_set(&id_priv->refcount, 1);
 624         mutex_init(&id_priv->handler_mutex);
 625         INIT_LIST_HEAD(&id_priv->listen_list);
 626         INIT_LIST_HEAD(&id_priv->mc_list);
 627         get_random_bytes(&id_priv->seq_num, sizeof id_priv->seq_num);
 628         id_priv->id.route.addr.dev_addr.net = get_net(net);
 629
 630         return &id_priv->id;
 631 }
 632 EXPORT_SYMBOL(rdma_create_id);
 633
 634 static int cma_init_ud_qp(struct rdma_id_private *id_priv, struct ib_qp *qp)
 635 {
 636         struct ib_qp_attr qp_attr;
 637         int qp_attr_mask, ret;
 638
 639         qp_attr.qp_state = IB_QPS_INIT;
 640         ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
 641         if (ret)
 642                 return ret;
 643
 644         ret = ib_modify_qp(qp, &qp_attr, qp_attr_mask);
 645         if (ret)
 646                 return ret;
 647
 648         qp_attr.qp_state = IB_QPS_RTR;
 649         ret = ib_modify_qp(qp, &qp_attr, IB_QP_STATE);
 650         if (ret)
 651                 return ret;
 652
 653         qp_attr.qp_state = IB_QPS_RTS;
 654         qp_attr.sq_psn = 0;
 655         ret = ib_modify_qp(qp, &qp_attr, IB_QP_STATE | IB_QP_SQ_PSN);
 656
 657         return ret;
 658 }
 659
 660 static int cma_init_conn_qp(struct rdma_id_private *id_priv, struct ib_qp *qp)
 661 {
 662         struct ib_qp_attr qp_attr;
 663         int qp_attr_mask, ret;
 664
 665         qp_attr.qp_state = IB_QPS_INIT;
 666         ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
 667         if (ret)
 668                 return ret;
 669
 670         return ib_modify_qp(qp, &qp_attr, qp_attr_mask);
 671 }
 672
 673 int rdma_create_qp(struct rdma_cm_id *id, struct ib_pd *pd,
 674                    struct ib_qp_init_attr *qp_init_attr)
 675 {
 676         struct rdma_id_private *id_priv;
 677         struct ib_qp *qp;
 678         int ret;
 679
 680         id_priv = container_of(id, struct rdma_id_private, id);
 681         if (id->device != pd->device)
 682                 return -EINVAL;
 683
 684         qp = ib_create_qp(pd, qp_init_attr);
 685         if (IS_ERR(qp))
 686                 return PTR_ERR(qp);
 687
 688         if (id->qp_type == IB_QPT_UD)
 689                 ret = cma_init_ud_qp(id_priv, qp);
 690         else
 691                 ret = cma_init_conn_qp(id_priv, qp);
 692         if (ret)
 693                 goto err;
 694
 695         id->qp = qp;
 696         id_priv->qp_num = qp->qp_num;
 697         id_priv->srq = (qp->srq != NULL);
 698         return 0;
 699 err:
 700         ib_destroy_qp(qp);
 701         return ret;
 702 }
 703 EXPORT_SYMBOL(rdma_create_qp);
 704
 705 void rdma_destroy_qp(struct rdma_cm_id *id)
 706 {
 707         struct rdma_id_private *id_priv;
 708
 709         id_priv = container_of(id, struct rdma_id_private, id);
 710         mutex_lock(&id_priv->qp_mutex);
 711         ib_destroy_qp(id_priv->id.qp);
 712         id_priv->id.qp = NULL;
 713         mutex_unlock(&id_priv->qp_mutex);
 714 }
 715 EXPORT_SYMBOL(rdma_destroy_qp);
 716
 717 static int cma_modify_qp_rtr(struct rdma_id_private *id_priv,
 718                              struct rdma_conn_param *conn_param)
 719 {
 720         struct ib_qp_attr qp_attr;
 721         int qp_attr_mask, ret;
 722         union ib_gid sgid;
 723
 724         mutex_lock(&id_priv->qp_mutex);
 725         if (!id_priv->id.qp) {
 726                 ret = 0;
 727                 goto out;
 728         }
 729
 730         /* Need to update QP attributes from default values. */
 731         qp_attr.qp_state = IB_QPS_INIT;
 732         ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
 733         if (ret)
 734                 goto out;
 735
 736         ret = ib_modify_qp(id_priv->id.qp, &qp_attr, qp_attr_mask);
 737         if (ret)
 738                 goto out;
 739
 740         qp_attr.qp_state = IB_QPS_RTR;
 741         ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
 742         if (ret)
 743                 goto out;
 744
 745         ret = ib_query_gid(id_priv->id.device, id_priv->id.port_num,
 746                            qp_attr.ah_attr.grh.sgid_index, &sgid, NULL);
 747         if (ret)
 748                 goto out;
 749
 750         BUG_ON(id_priv->cma_dev->device != id_priv->id.device);
 751
 752         if (conn_param)
 753                 qp_attr.max_dest_rd_atomic = conn_param->responder_resources;
 754         ret = ib_modify_qp(id_priv->id.qp, &qp_attr, qp_attr_mask);
 755 out:
 756         mutex_unlock(&id_priv->qp_mutex);
 757         return ret;
 758 }
 759
 760 static int cma_modify_qp_rts(struct rdma_id_private *id_priv,
 761                              struct rdma_conn_param *conn_param)
 762 {
 763         struct ib_qp_attr qp_attr;
 764         int qp_attr_mask, ret;
 765
 766         mutex_lock(&id_priv->qp_mutex);
 767         if (!id_priv->id.qp) {
 768                 ret = 0;
 769                 goto out;
 770         }
 771
 772         qp_attr.qp_state = IB_QPS_RTS;
 773         ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
 774         if (ret)
 775                 goto out;
 776
 777         if (conn_param)
 778                 qp_attr.max_rd_atomic = conn_param->initiator_depth;
 779         ret = ib_modify_qp(id_priv->id.qp, &qp_attr, qp_attr_mask);
 780 out:
 781         mutex_unlock(&id_priv->qp_mutex);
 782         return ret;
 783 }
 784
 785 static int cma_modify_qp_err(struct rdma_id_private *id_priv)
 786 {
 787         struct ib_qp_attr qp_attr;
 788         int ret;
 789
 790         mutex_lock(&id_priv->qp_mutex);
 791         if (!id_priv->id.qp) {
 792                 ret = 0;
 793                 goto out;
 794         }
 795
 796         qp_attr.qp_state = IB_QPS_ERR;
 797         ret = ib_modify_qp(id_priv->id.qp, &qp_attr, IB_QP_STATE);
 798 out:
 799         mutex_unlock(&id_priv->qp_mutex);
 800         return ret;
 801 }
 802
 803 static int cma_ib_init_qp_attr(struct rdma_id_private *id_priv,
 804                                struct ib_qp_attr *qp_attr, int *qp_attr_mask)
 805 {
 806         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
 807         int ret;
 808         u16 pkey;
 809
 810         if (rdma_cap_eth_ah(id_priv->id.device, id_priv->id.port_num))
 811                 pkey = 0xffff;
 812         else
 813                 pkey = ib_addr_get_pkey(dev_addr);
 814
 815         ret = ib_find_cached_pkey(id_priv->id.device, id_priv->id.port_num,
 816                                   pkey, &qp_attr->pkey_index);
 817         if (ret)
 818                 return ret;
 819
 820         qp_attr->port_num = id_priv->id.port_num;
 821         *qp_attr_mask = IB_QP_STATE | IB_QP_PKEY_INDEX | IB_QP_PORT;
 822
 823         if (id_priv->id.qp_type == IB_QPT_UD) {
 824                 ret = cma_set_qkey(id_priv, 0);
 825                 if (ret)
 826                         return ret;
 827
 828                 qp_attr->qkey = id_priv->qkey;
 829                 *qp_attr_mask |= IB_QP_QKEY;
 830         } else {
 831                 qp_attr->qp_access_flags = 0;
 832                 *qp_attr_mask |= IB_QP_ACCESS_FLAGS;
 833         }
 834         return 0;
 835 }
 836
 837 int rdma_init_qp_attr(struct rdma_cm_id *id, struct ib_qp_attr *qp_attr,
 838                        int *qp_attr_mask)
 839 {
 840         struct rdma_id_private *id_priv;
 841         int ret = 0;
 842
 843         id_priv = container_of(id, struct rdma_id_private, id);
 844         if (rdma_cap_ib_cm(id->device, id->port_num)) {
 845                 if (!id_priv->cm_id.ib || (id_priv->id.qp_type == IB_QPT_UD))
 846                         ret = cma_ib_init_qp_attr(id_priv, qp_attr, qp_attr_mask);
 847                 else
 848                         ret = ib_cm_init_qp_attr(id_priv->cm_id.ib, qp_attr,
 849                                                  qp_attr_mask);
 850
 851                 if (qp_attr->qp_state == IB_QPS_RTR)
 852                         qp_attr->rq_psn = id_priv->seq_num;
 853         } else if (rdma_cap_iw_cm(id->device, id->port_num)) {
 854                 if (!id_priv->cm_id.iw) {
 855                         qp_attr->qp_access_flags = 0;
 856                         *qp_attr_mask = IB_QP_STATE | IB_QP_ACCESS_FLAGS;
 857                 } else
 858                         ret = iw_cm_init_qp_attr(id_priv->cm_id.iw, qp_attr,
 859                                                  qp_attr_mask);
 860         } else
 861                 ret = -ENOSYS;
 862
 863         return ret;
 864 }
 865 EXPORT_SYMBOL(rdma_init_qp_attr);
 866
 867 static inline int cma_zero_addr(struct sockaddr *addr)
 868 {
 869         switch (addr->sa_family) {
 870         case AF_INET:
 871                 return ipv4_is_zeronet(((struct sockaddr_in *)addr)->sin_addr.s_addr);
 872         case AF_INET6:
 873                 return ipv6_addr_any(&((struct sockaddr_in6 *) addr)->sin6_addr);
 874         case AF_IB:
 875                 return ib_addr_any(&((struct sockaddr_ib *) addr)->sib_addr);
 876         default:
 877                 return 0;
 878         }
 879 }
 880
 881 static inline int cma_loopback_addr(struct sockaddr *addr)
 882 {
 883         switch (addr->sa_family) {
 884         case AF_INET:
 885                 return ipv4_is_loopback(((struct sockaddr_in *) addr)->sin_addr.s_addr);
 886         case AF_INET6:
 887                 return ipv6_addr_loopback(&((struct sockaddr_in6 *) addr)->sin6_addr);
 888         case AF_IB:
 889                 return ib_addr_loopback(&((struct sockaddr_ib *) addr)->sib_addr);
 890         default:
 891                 return 0;
 892         }
 893 }
 894
 895 static inline int cma_any_addr(struct sockaddr *addr)
 896 {
 897         return cma_zero_addr(addr) || cma_loopback_addr(addr);
 898 }
 899
 900 static int cma_addr_cmp(struct sockaddr *src, struct sockaddr *dst)
 901 {
 902         if (src->sa_family != dst->sa_family)
 903                 return -1;
 904
 905         switch (src->sa_family) {
 906         case AF_INET:
 907                 return ((struct sockaddr_in *) src)->sin_addr.s_addr !=
 908                        ((struct sockaddr_in *) dst)->sin_addr.s_addr;
 909         case AF_INET6:
 910                 return ipv6_addr_cmp(&((struct sockaddr_in6 *) src)->sin6_addr,
 911                                      &((struct sockaddr_in6 *) dst)->sin6_addr);
 912         default:
 913                 return ib_addr_cmp(&((struct sockaddr_ib *) src)->sib_addr,
 914                                    &((struct sockaddr_ib *) dst)->sib_addr);
 915         }
 916 }
 917
 918 static __be16 cma_port(struct sockaddr *addr)
 919 {
 920         struct sockaddr_ib *sib;
 921
 922         switch (addr->sa_family) {
 923         case AF_INET:
 924                 return ((struct sockaddr_in *) addr)->sin_port;
 925         case AF_INET6:
 926                 return ((struct sockaddr_in6 *) addr)->sin6_port;
 927         case AF_IB:
 928                 sib = (struct sockaddr_ib *) addr;
 929                 return htons((u16) (be64_to_cpu(sib->sib_sid) &
 930                                     be64_to_cpu(sib->sib_sid_mask)));
 931         default:
 932                 return 0;
 933         }
 934 }
 935
 936 static inline int cma_any_port(struct sockaddr *addr)
 937 {
 938         return !cma_port(addr);
 939 }
 940
 941 static void cma_save_ib_info(struct sockaddr *src_addr,
 942                              struct sockaddr *dst_addr,
 943                              struct rdma_cm_id *listen_id,
 944                              struct ib_sa_path_rec *path)
 945 {
 946         struct sockaddr_ib *listen_ib, *ib;
 947
 948         listen_ib = (struct sockaddr_ib *) &listen_id->route.addr.src_addr;
 949         if (src_addr) {
 950                 ib = (struct sockaddr_ib *)src_addr;
 951                 ib->sib_family = AF_IB;
 952                 if (path) {
 953                         ib->sib_pkey = path->pkey;
 954                         ib->sib_flowinfo = path->flow_label;
 955                         memcpy(&ib->sib_addr, &path->sgid, 16);
 956                         ib->sib_sid = path->service_id;
 957                         ib->sib_scope_id = 0;
 958                 } else {
 959                         ib->sib_pkey = listen_ib->sib_pkey;
 960                         ib->sib_flowinfo = listen_ib->sib_flowinfo;
 961                         ib->sib_addr = listen_ib->sib_addr;
 962                         ib->sib_sid = listen_ib->sib_sid;
 963                         ib->sib_scope_id = listen_ib->sib_scope_id;
 964                 }
 965                 ib->sib_sid_mask = cpu_to_be64(0xffffffffffffffffULL);
 966         }
 967         if (dst_addr) {
 968                 ib = (struct sockaddr_ib *)dst_addr;
 969                 ib->sib_family = AF_IB;
 970                 if (path) {
 971                         ib->sib_pkey = path->pkey;
 972                         ib->sib_flowinfo = path->flow_label;
 973                         memcpy(&ib->sib_addr, &path->dgid, 16);
 974                 }
 975         }
 976 }
 977
 978 static void cma_save_ip4_info(struct sockaddr *src_addr,
 979                               struct sockaddr *dst_addr,
 980                               struct cma_hdr *hdr,
 981                               __be16 local_port)
 982 {
 983         struct sockaddr_in *ip4;
 984
 985         if (src_addr) {
 986                 ip4 = (struct sockaddr_in *)src_addr;
 987                 ip4->sin_family = AF_INET;
 988                 ip4->sin_addr.s_addr = hdr->dst_addr.ip4.addr;
 989                 ip4->sin_port = local_port;
 990         }
 991
 992         if (dst_addr) {
 993                 ip4 = (struct sockaddr_in *)dst_addr;
 994                 ip4->sin_family = AF_INET;
 995                 ip4->sin_addr.s_addr = hdr->src_addr.ip4.addr;
 996                 ip4->sin_port = hdr->port;
 997         }
 998 }
 999
1000 static void cma_save_ip6_info(struct sockaddr *src_addr,
1001                               struct sockaddr *dst_addr,
1002                               struct cma_hdr *hdr,
1003                               __be16 local_port)
1004 {
1005         struct sockaddr_in6 *ip6;
1006
1007         if (src_addr) {
1008                 ip6 = (struct sockaddr_in6 *)src_addr;
1009                 ip6->sin6_family = AF_INET6;
1010                 ip6->sin6_addr = hdr->dst_addr.ip6;
1011                 ip6->sin6_port = local_port;
1012         }
1013
1014         if (dst_addr) {
1015                 ip6 = (struct sockaddr_in6 *)dst_addr;
1016                 ip6->sin6_family = AF_INET6;
1017                 ip6->sin6_addr = hdr->src_addr.ip6;
1018                 ip6->sin6_port = hdr->port;
1019         }
1020 }
1021
1022 static u16 cma_port_from_service_id(__be64 service_id)
1023 {
1024         return (u16)be64_to_cpu(service_id);
1025 }
1026
1027 static int cma_save_ip_info(struct sockaddr *src_addr,
1028                             struct sockaddr *dst_addr,
1029                             struct ib_cm_event *ib_event,
1030                             __be64 service_id)
1031 {
1032         struct cma_hdr *hdr;
1033         __be16 port;
1034
1035         hdr = ib_event->private_data;
1036         if (hdr->cma_version != CMA_VERSION)
1037                 return -EINVAL;
1038
1039         port = htons(cma_port_from_service_id(service_id));
1040
1041         switch (cma_get_ip_ver(hdr)) {
1042         case 4:
1043                 cma_save_ip4_info(src_addr, dst_addr, hdr, port);
1044                 break;
1045         case 6:
1046                 cma_save_ip6_info(src_addr, dst_addr, hdr, port);
1047                 break;
1048         default:
1049                 return -EAFNOSUPPORT;
1050         }
1051
1052         return 0;
1053 }
1054
1055 static int cma_save_net_info(struct sockaddr *src_addr,
1056                              struct sockaddr *dst_addr,
1057                              struct rdma_cm_id *listen_id,
1058                              struct ib_cm_event *ib_event,
1059                              sa_family_t sa_family, __be64 service_id)
1060 {
1061         if (sa_family == AF_IB) {
1062                 if (ib_event->event == IB_CM_REQ_RECEIVED)
1063                         cma_save_ib_info(src_addr, dst_addr, listen_id,
1064                                          ib_event->param.req_rcvd.primary_path);
1065                 else if (ib_event->event == IB_CM_SIDR_REQ_RECEIVED)
1066                         cma_save_ib_info(src_addr, dst_addr, listen_id, NULL);
1067                 return 0;
1068         }
1069
1070         return cma_save_ip_info(src_addr, dst_addr, ib_event, service_id);
1071 }
1072
1073 static int cma_save_req_info(const struct ib_cm_event *ib_event,
1074                              struct cma_req_info *req)
1075 {
1076         const struct ib_cm_req_event_param *req_param =
1077                 &ib_event->param.req_rcvd;
1078         const struct ib_cm_sidr_req_event_param *sidr_param =
1079                 &ib_event->param.sidr_req_rcvd;
1080
1081         switch (ib_event->event) {
1082         case IB_CM_REQ_RECEIVED:
1083                 req->device     = req_param->listen_id->device;
1084                 req->port       = req_param->port;
1085                 memcpy(&req->local_gid, &req_param->primary_path->sgid,
1086                        sizeof(req->local_gid));
1087                 req->has_gid    = true;
1088                 req->service_id = req_param->primary_path->service_id;
1089                 req->pkey       = be16_to_cpu(req_param->primary_path->pkey);
1090                 break;
1091         case IB_CM_SIDR_REQ_RECEIVED:
1092                 req->device     = sidr_param->listen_id->device;
1093                 req->port       = sidr_param->port;
1094                 req->has_gid    = false;
1095                 req->service_id = sidr_param->service_id;
1096                 req->pkey       = sidr_param->pkey;
1097                 break;
1098         default:
1099                 return -EINVAL;
1100         }
1101
1102         return 0;
1103 }
1104
1105 static bool validate_ipv4_net_dev(struct net_device *net_dev,
1106                                   const struct sockaddr_in *dst_addr,
1107                                   const struct sockaddr_in *src_addr)
1108 {
1109         __be32 daddr = dst_addr->sin_addr.s_addr,
1110                saddr = src_addr->sin_addr.s_addr;
1111         struct fib_result res;
1112         struct flowi4 fl4;
1113         int err;
1114         bool ret;
1115
1116         if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1117             ipv4_is_lbcast(daddr) || ipv4_is_zeronet(saddr) ||
1118             ipv4_is_zeronet(daddr) || ipv4_is_loopback(daddr) ||
1119             ipv4_is_loopback(saddr))
1120                 return false;
1121
1122         memset(&fl4, 0, sizeof(fl4));
1123         fl4.flowi4_iif = net_dev->ifindex;
1124         fl4.daddr = daddr;
1125         fl4.saddr = saddr;
1126
1127         rcu_read_lock();
1128         err = fib_lookup(dev_net(net_dev), &fl4, &res, 0);
1129         ret = err == 0 && FIB_RES_DEV(res) == net_dev;
1130         rcu_read_unlock();
1131
1132         return ret;
1133 }
1134
1135 static bool validate_ipv6_net_dev(struct net_device *net_dev,
1136                                   const struct sockaddr_in6 *dst_addr,
1137                                   const struct sockaddr_in6 *src_addr)
1138 {
1139 #if IS_ENABLED(CONFIG_IPV6)
1140         const int strict = ipv6_addr_type(&dst_addr->sin6_addr) &
1141                            IPV6_ADDR_LINKLOCAL;
1142         struct rt6_info *rt = rt6_lookup(dev_net(net_dev), &dst_addr->sin6_addr,
1143                                          &src_addr->sin6_addr, net_dev->ifindex,
1144                                          strict);
1145         bool ret;
1146
1147         if (!rt)
1148                 return false;
1149
1150         ret = rt->rt6i_idev->dev == net_dev;
1151         ip6_rt_put(rt);
1152
1153         return ret;
1154 #else
1155         return false;
1156 #endif
1157 }
1158
1159 static bool validate_net_dev(struct net_device *net_dev,
1160                              const struct sockaddr *daddr,
1161                              const struct sockaddr *saddr)
1162 {
1163         const struct sockaddr_in *daddr4 = (const struct sockaddr_in *)daddr;
1164         const struct sockaddr_in *saddr4 = (const struct sockaddr_in *)saddr;
1165         const struct sockaddr_in6 *daddr6 = (const struct sockaddr_in6 *)daddr;
1166         const struct sockaddr_in6 *saddr6 = (const struct sockaddr_in6 *)saddr;
1167
1168         switch (daddr->sa_family) {
1169         case AF_INET:
1170                 return saddr->sa_family == AF_INET &&
1171                        validate_ipv4_net_dev(net_dev, daddr4, saddr4);
1172
1173         case AF_INET6:
1174                 return saddr->sa_family == AF_INET6 &&
1175                        validate_ipv6_net_dev(net_dev, daddr6, saddr6);
1176
1177         default:
1178                 return false;
1179         }
1180 }
1181
1182 static struct net_device *cma_get_net_dev(struct ib_cm_event *ib_event,
1183                                           const struct cma_req_info *req)
1184 {
1185         struct sockaddr_storage listen_addr_storage, src_addr_storage;
1186         struct sockaddr *listen_addr = (struct sockaddr *)&listen_addr_storage,
1187                         *src_addr = (struct sockaddr *)&src_addr_storage;
1188         struct net_device *net_dev;
1189         const union ib_gid *gid = req->has_gid ? &req->local_gid : NULL;
1190         int err;
1191
1192         err = cma_save_ip_info(listen_addr, src_addr, ib_event,
1193                                req->service_id);
1194         if (err)
1195                 return ERR_PTR(err);
1196
1197         net_dev = ib_get_net_dev_by_params(req->device, req->port, req->pkey,
1198                                            gid, listen_addr);
1199         if (!net_dev)
1200                 return ERR_PTR(-ENODEV);
1201
1202         if (!validate_net_dev(net_dev, listen_addr, src_addr)) {
1203                 dev_put(net_dev);
1204                 return ERR_PTR(-EHOSTUNREACH);
1205         }
1206
1207         return net_dev;
1208 }
1209
1210 static enum rdma_port_space rdma_ps_from_service_id(__be64 service_id)
1211 {
1212         return (be64_to_cpu(service_id) >> 16) & 0xffff;
1213 }
1214
1215 static bool cma_match_private_data(struct rdma_id_private *id_priv,
1216                                    const struct cma_hdr *hdr)
1217 {
1218         struct sockaddr *addr = cma_src_addr(id_priv);
1219         __be32 ip4_addr;
1220         struct in6_addr ip6_addr;
1221
1222         if (cma_any_addr(addr) && !id_priv->afonly)
1223                 return true;
1224
1225         switch (addr->sa_family) {
1226         case AF_INET:
1227                 ip4_addr = ((struct sockaddr_in *)addr)->sin_addr.s_addr;
1228                 if (cma_get_ip_ver(hdr) != 4)
1229                         return false;
1230                 if (!cma_any_addr(addr) &&
1231                     hdr->dst_addr.ip4.addr != ip4_addr)
1232                         return false;
1233                 break;
1234         case AF_INET6:
1235                 ip6_addr = ((struct sockaddr_in6 *)addr)->sin6_addr;
1236                 if (cma_get_ip_ver(hdr) != 6)
1237                         return false;
1238                 if (!cma_any_addr(addr) &&
1239                     memcmp(&hdr->dst_addr.ip6, &ip6_addr, sizeof(ip6_addr)))
1240                         return false;
1241                 break;
1242         case AF_IB:
1243                 return true;
1244         default:
1245                 return false;
1246         }
1247
1248         return true;
1249 }
1250
1251 static bool cma_protocol_roce_dev_port(struct ib_device *device, int port_num)
1252 {
1253         enum rdma_link_layer ll = rdma_port_get_link_layer(device, port_num);
1254         enum rdma_transport_type transport =
1255                 rdma_node_get_transport(device->node_type);
1256
1257         return ll == IB_LINK_LAYER_ETHERNET && transport == RDMA_TRANSPORT_IB;
1258 }
1259
1260 static bool cma_protocol_roce(const struct rdma_cm_id *id)
1261 {
1262         struct ib_device *device = id->device;
1263         const int port_num = id->port_num ?: rdma_start_port(device);
1264
1265         return cma_protocol_roce_dev_port(device, port_num);
1266 }
1267
1268 static bool cma_match_net_dev(const struct rdma_cm_id *id,
1269                               const struct net_device *net_dev,
1270                               u8 port_num)
1271 {
1272         const struct rdma_addr *addr = &id->route.addr;
1273
1274         if (!net_dev)
1275                 /* This request is an AF_IB request or a RoCE request */
1276                 return (!id->port_num || id->port_num == port_num) &&
1277                        (addr->src_addr.ss_family == AF_IB ||
1278                         cma_protocol_roce_dev_port(id->device, port_num));
1279
1280         return !addr->dev_addr.bound_dev_if ||
1281                (net_eq(dev_net(net_dev), addr->dev_addr.net) &&
1282                 addr->dev_addr.bound_dev_if == net_dev->ifindex);
1283 }
1284
1285 static struct rdma_id_private *cma_find_listener(
1286                 const struct rdma_bind_list *bind_list,
1287                 const struct ib_cm_id *cm_id,
1288                 const struct ib_cm_event *ib_event,
1289                 const struct cma_req_info *req,
1290                 const struct net_device *net_dev)
1291 {
1292         struct rdma_id_private *id_priv, *id_priv_dev;
1293
1294         if (!bind_list)
1295                 return ERR_PTR(-EINVAL);
1296
1297         hlist_for_each_entry(id_priv, &bind_list->owners, node) {
1298                 if (cma_match_private_data(id_priv, ib_event->private_data)) {
1299                         if (id_priv->id.device == cm_id->device &&
1300                             cma_match_net_dev(&id_priv->id, net_dev, req->port))
1301                                 return id_priv;
1302                         list_for_each_entry(id_priv_dev,
1303                                             &id_priv->listen_list,
1304                                             listen_list) {
1305                                 if (id_priv_dev->id.device == cm_id->device &&
1306                                     cma_match_net_dev(&id_priv_dev->id, net_dev, req->port))
1307                                         return id_priv_dev;
1308                         }
1309                 }
1310         }
1311
1312         return ERR_PTR(-EINVAL);
1313 }
1314
1315 static struct rdma_id_private *cma_id_from_event(struct ib_cm_id *cm_id,
1316                                                  struct ib_cm_event *ib_event,
1317                                                  struct net_device **net_dev)
1318 {
1319         struct cma_req_info req;
1320         struct rdma_bind_list *bind_list;
1321         struct rdma_id_private *id_priv;
1322         int err;
1323
1324         err = cma_save_req_info(ib_event, &req);
1325         if (err)
1326                 return ERR_PTR(err);
1327
1328         *net_dev = cma_get_net_dev(ib_event, &req);
1329         if (IS_ERR(*net_dev)) {
1330                 if (PTR_ERR(*net_dev) == -EAFNOSUPPORT) {
1331                         /* Assuming the protocol is AF_IB */
1332                         *net_dev = NULL;
1333                 } else if (cma_protocol_roce_dev_port(req.device, req.port)) {
1334                         /* TODO find the net dev matching the request parameters
1335                          * through the RoCE GID table */
1336                         *net_dev = NULL;
1337                 } else {
1338                         return ERR_CAST(*net_dev);
1339                 }
1340         }
1341
1342         bind_list = cma_ps_find(*net_dev ? dev_net(*net_dev) : &init_net,
1343                                 rdma_ps_from_service_id(req.service_id),
1344                                 cma_port_from_service_id(req.service_id));
1345         id_priv = cma_find_listener(bind_list, cm_id, ib_event, &req, *net_dev);
1346         if (IS_ERR(id_priv) && *net_dev) {
1347                 dev_put(*net_dev);
1348                 *net_dev = NULL;
1349         }
1350
1351         return id_priv;
1352 }
1353
1354 static inline int cma_user_data_offset(struct rdma_id_private *id_priv)
1355 {
1356         return cma_family(id_priv) == AF_IB ? 0 : sizeof(struct cma_hdr);
1357 }
1358
1359 static void cma_cancel_route(struct rdma_id_private *id_priv)
1360 {
1361         if (rdma_cap_ib_sa(id_priv->id.device, id_priv->id.port_num)) {
1362                 if (id_priv->query)
1363                         ib_sa_cancel_query(id_priv->query_id, id_priv->query);
1364         }
1365 }
1366
1367 static void cma_cancel_listens(struct rdma_id_private *id_priv)
1368 {
1369         struct rdma_id_private *dev_id_priv;
1370
1371         /*
1372          * Remove from listen_any_list to prevent added devices from spawning
1373          * additional listen requests.
1374          */
1375         mutex_lock(&lock);
1376         list_del(&id_priv->list);
1377
1378         while (!list_empty(&id_priv->listen_list)) {
1379                 dev_id_priv = list_entry(id_priv->listen_list.next,
1380                                          struct rdma_id_private, listen_list);
1381                 /* sync with device removal to avoid duplicate destruction */
1382                 list_del_init(&dev_id_priv->list);
1383                 list_del(&dev_id_priv->listen_list);
1384                 mutex_unlock(&lock);
1385
1386                 rdma_destroy_id(&dev_id_priv->id);
1387                 mutex_lock(&lock);
1388         }
1389         mutex_unlock(&lock);
1390 }
1391
1392 static void cma_cancel_operation(struct rdma_id_private *id_priv,
1393                                  enum rdma_cm_state state)
1394 {
1395         switch (state) {
1396         case RDMA_CM_ADDR_QUERY:
1397                 rdma_addr_cancel(&id_priv->id.route.addr.dev_addr);
1398                 break;
1399         case RDMA_CM_ROUTE_QUERY:
1400                 cma_cancel_route(id_priv);
1401                 break;
1402         case RDMA_CM_LISTEN:
1403                 if (cma_any_addr(cma_src_addr(id_priv)) && !id_priv->cma_dev)
1404                         cma_cancel_listens(id_priv);
1405                 break;
1406         default:
1407                 break;
1408         }
1409 }
1410
1411 static void cma_release_port(struct rdma_id_private *id_priv)
1412 {
1413         struct rdma_bind_list *bind_list = id_priv->bind_list;
1414         struct net *net = id_priv->id.route.addr.dev_addr.net;
1415
1416         if (!bind_list)
1417                 return;
1418
1419         mutex_lock(&lock);
1420         hlist_del(&id_priv->node);
1421         if (hlist_empty(&bind_list->owners)) {
1422                 cma_ps_remove(net, bind_list->ps, bind_list->port);
1423                 kfree(bind_list);
1424         }
1425         mutex_unlock(&lock);
1426 }
1427
1428 static void cma_leave_mc_groups(struct rdma_id_private *id_priv)
1429 {
1430         struct cma_multicast *mc;
1431
1432         while (!list_empty(&id_priv->mc_list)) {
1433                 mc = container_of(id_priv->mc_list.next,
1434                                   struct cma_multicast, list);
1435                 list_del(&mc->list);
1436                 if (rdma_cap_ib_mcast(id_priv->cma_dev->device,
1437                                       id_priv->id.port_num)) {
1438                         ib_sa_free_multicast(mc->multicast.ib);
1439                         kfree(mc);
1440                 } else
1441                         kref_put(&mc->mcref, release_mc);
1442         }
1443 }
1444
1445 void rdma_destroy_id(struct rdma_cm_id *id)
1446 {
1447         struct rdma_id_private *id_priv;
1448         enum rdma_cm_state state;
1449
1450         id_priv = container_of(id, struct rdma_id_private, id);
1451         state = cma_exch(id_priv, RDMA_CM_DESTROYING);
1452         cma_cancel_operation(id_priv, state);
1453
1454         /*
1455          * Wait for any active callback to finish.  New callbacks will find
1456          * the id_priv state set to destroying and abort.
1457          */
1458         mutex_lock(&id_priv->handler_mutex);
1459         mutex_unlock(&id_priv->handler_mutex);
1460
1461         if (id_priv->cma_dev) {
1462                 if (rdma_cap_ib_cm(id_priv->id.device, 1)) {
1463                         if (id_priv->cm_id.ib)
1464                                 ib_destroy_cm_id(id_priv->cm_id.ib);
1465                 } else if (rdma_cap_iw_cm(id_priv->id.device, 1)) {
1466                         if (id_priv->cm_id.iw)
1467                                 iw_destroy_cm_id(id_priv->cm_id.iw);
1468                 }
1469                 cma_leave_mc_groups(id_priv);
1470                 cma_release_dev(id_priv);
1471         }
1472
1473         cma_release_port(id_priv);
1474         cma_deref_id(id_priv);
1475         wait_for_completion(&id_priv->comp);
1476
1477         if (id_priv->internal_id)
1478                 cma_deref_id(id_priv->id.context);
1479
1480         kfree(id_priv->id.route.path_rec);
1481         put_net(id_priv->id.route.addr.dev_addr.net);
1482         kfree(id_priv);
1483 }
1484 EXPORT_SYMBOL(rdma_destroy_id);
1485
1486 static int cma_rep_recv(struct rdma_id_private *id_priv)
1487 {
1488         int ret;
1489
1490         ret = cma_modify_qp_rtr(id_priv, NULL);
1491         if (ret)
1492                 goto reject;
1493
1494         ret = cma_modify_qp_rts(id_priv, NULL);
1495         if (ret)
1496                 goto reject;
1497
1498         ret = ib_send_cm_rtu(id_priv->cm_id.ib, NULL, 0);
1499         if (ret)
1500                 goto reject;
1501
1502         return 0;
1503 reject:
1504         cma_modify_qp_err(id_priv);
1505         ib_send_cm_rej(id_priv->cm_id.ib, IB_CM_REJ_CONSUMER_DEFINED,
1506                        NULL, 0, NULL, 0);
1507         return ret;
1508 }
1509
1510 static void cma_set_rep_event_data(struct rdma_cm_event *event,
1511                                    struct ib_cm_rep_event_param *rep_data,
1512                                    void *private_data)
1513 {
1514         event->param.conn.private_data = private_data;
1515         event->param.conn.private_data_len = IB_CM_REP_PRIVATE_DATA_SIZE;
1516         event->param.conn.responder_resources = rep_data->responder_resources;
1517         event->param.conn.initiator_depth = rep_data->initiator_depth;
1518         event->param.conn.flow_control = rep_data->flow_control;
1519         event->param.conn.rnr_retry_count = rep_data->rnr_retry_count;
1520         event->param.conn.srq = rep_data->srq;
1521         event->param.conn.qp_num = rep_data->remote_qpn;
1522 }
1523
1524 static int cma_ib_handler(struct ib_cm_id *cm_id, struct ib_cm_event *ib_event)
1525 {
1526         struct rdma_id_private *id_priv = cm_id->context;
1527         struct rdma_cm_event event;
1528         int ret = 0;
1529
1530         if ((ib_event->event != IB_CM_TIMEWAIT_EXIT &&
1531                 cma_disable_callback(id_priv, RDMA_CM_CONNECT)) ||
1532             (ib_event->event == IB_CM_TIMEWAIT_EXIT &&
1533                 cma_disable_callback(id_priv, RDMA_CM_DISCONNECT)))
1534                 return 0;
1535
1536         memset(&event, 0, sizeof event);
1537         switch (ib_event->event) {
1538         case IB_CM_REQ_ERROR:
1539         case IB_CM_REP_ERROR:
1540                 event.event = RDMA_CM_EVENT_UNREACHABLE;
1541                 event.status = -ETIMEDOUT;
1542                 break;
1543         case IB_CM_REP_RECEIVED:
1544                 if (id_priv->id.qp) {
1545                         event.status = cma_rep_recv(id_priv);
1546                         event.event = event.status ? RDMA_CM_EVENT_CONNECT_ERROR :
1547                                                      RDMA_CM_EVENT_ESTABLISHED;
1548                 } else {
1549                         event.event = RDMA_CM_EVENT_CONNECT_RESPONSE;
1550                 }
1551                 cma_set_rep_event_data(&event, &ib_event->param.rep_rcvd,
1552                                        ib_event->private_data);
1553                 break;
1554         case IB_CM_RTU_RECEIVED:
1555         case IB_CM_USER_ESTABLISHED:
1556                 event.event = RDMA_CM_EVENT_ESTABLISHED;
1557                 break;
1558         case IB_CM_DREQ_ERROR:
1559                 event.status = -ETIMEDOUT; /* fall through */
1560         case IB_CM_DREQ_RECEIVED:
1561         case IB_CM_DREP_RECEIVED:
1562                 if (!cma_comp_exch(id_priv, RDMA_CM_CONNECT,
1563                                    RDMA_CM_DISCONNECT))
1564                         goto out;
1565                 event.event = RDMA_CM_EVENT_DISCONNECTED;
1566                 break;
1567         case IB_CM_TIMEWAIT_EXIT:
1568                 event.event = RDMA_CM_EVENT_TIMEWAIT_EXIT;
1569                 break;
1570         case IB_CM_MRA_RECEIVED:
1571                 /* ignore event */
1572                 goto out;
1573         case IB_CM_REJ_RECEIVED:
1574                 cma_modify_qp_err(id_priv);
1575                 event.status = ib_event->param.rej_rcvd.reason;
1576                 event.event = RDMA_CM_EVENT_REJECTED;
1577                 event.param.conn.private_data = ib_event->private_data;
1578                 event.param.conn.private_data_len = IB_CM_REJ_PRIVATE_DATA_SIZE;
1579                 break;
1580         default:
1581                 printk(KERN_ERR "RDMA CMA: unexpected IB CM event: %d\n",
1582                        ib_event->event);
1583                 goto out;
1584         }
1585
1586         ret = id_priv->id.event_handler(&id_priv->id, &event);
1587         if (ret) {
1588                 /* Destroy the CM ID by returning a non-zero value. */
1589                 id_priv->cm_id.ib = NULL;
1590                 cma_exch(id_priv, RDMA_CM_DESTROYING);
1591                 mutex_unlock(&id_priv->handler_mutex);
1592                 rdma_destroy_id(&id_priv->id);
1593                 return ret;
1594         }
1595 out:
1596         mutex_unlock(&id_priv->handler_mutex);
1597         return ret;
1598 }
1599
1600 static struct rdma_id_private *cma_new_conn_id(struct rdma_cm_id *listen_id,
1601                                                struct ib_cm_event *ib_event,
1602                                                struct net_device *net_dev)
1603 {
1604         struct rdma_id_private *id_priv;
1605         struct rdma_cm_id *id;
1606         struct rdma_route *rt;
1607         const sa_family_t ss_family = listen_id->route.addr.src_addr.ss_family;
1608         const __be64 service_id =
1609                       ib_event->param.req_rcvd.primary_path->service_id;
1610         int ret;
1611
1612         id = rdma_create_id(listen_id->route.addr.dev_addr.net,
1613                             listen_id->event_handler, listen_id->context,
1614                             listen_id->ps, ib_event->param.req_rcvd.qp_type);
1615         if (IS_ERR(id))
1616                 return NULL;
1617
1618         id_priv = container_of(id, struct rdma_id_private, id);
1619         if (cma_save_net_info((struct sockaddr *)&id->route.addr.src_addr,
1620                               (struct sockaddr *)&id->route.addr.dst_addr,
1621                               listen_id, ib_event, ss_family, service_id))
1622                 goto err;
1623
1624         rt = &id->route;
1625         rt->num_paths = ib_event->param.req_rcvd.alternate_path ? 2 : 1;
1626         rt->path_rec = kmalloc(sizeof *rt->path_rec * rt->num_paths,
1627                                GFP_KERNEL);
1628         if (!rt->path_rec)
1629                 goto err;
1630
1631         rt->path_rec[0] = *ib_event->param.req_rcvd.primary_path;
1632         if (rt->num_paths == 2)
1633                 rt->path_rec[1] = *ib_event->param.req_rcvd.alternate_path;
1634
1635         if (net_dev) {
1636                 ret = rdma_copy_addr(&rt->addr.dev_addr, net_dev, NULL);
1637                 if (ret)
1638                         goto err;
1639         } else {
1640                 if (!cma_protocol_roce(listen_id) &&
1641                     cma_any_addr(cma_src_addr(id_priv))) {
1642                         rt->addr.dev_addr.dev_type = ARPHRD_INFINIBAND;
1643                         rdma_addr_set_sgid(&rt->addr.dev_addr, &rt->path_rec[0].sgid);
1644                         ib_addr_set_pkey(&rt->addr.dev_addr, be16_to_cpu(rt->path_rec[0].pkey));
1645                 } else if (!cma_any_addr(cma_src_addr(id_priv))) {
1646                         ret = cma_translate_addr(cma_src_addr(id_priv), &rt->addr.dev_addr);
1647                         if (ret)
1648                                 goto err;
1649                 }
1650         }
1651         rdma_addr_set_dgid(&rt->addr.dev_addr, &rt->path_rec[0].dgid);
1652
1653         id_priv->state = RDMA_CM_CONNECT;
1654         return id_priv;
1655
1656 err:
1657         rdma_destroy_id(id);
1658         return NULL;
1659 }
1660
1661 static struct rdma_id_private *cma_new_udp_id(struct rdma_cm_id *listen_id,
1662                                               struct ib_cm_event *ib_event,
1663                                               struct net_device *net_dev)
1664 {
1665         struct rdma_id_private *id_priv;
1666         struct rdma_cm_id *id;
1667         const sa_family_t ss_family = listen_id->route.addr.src_addr.ss_family;
1668         struct net *net = listen_id->route.addr.dev_addr.net;
1669         int ret;
1670
1671         id = rdma_create_id(net, listen_id->event_handler, listen_id->context,
1672                             listen_id->ps, IB_QPT_UD);
1673         if (IS_ERR(id))
1674                 return NULL;
1675
1676         id_priv = container_of(id, struct rdma_id_private, id);
1677         if (cma_save_net_info((struct sockaddr *)&id->route.addr.src_addr,
1678                               (struct sockaddr *)&id->route.addr.dst_addr,
1679                               listen_id, ib_event, ss_family,
1680                               ib_event->param.sidr_req_rcvd.service_id))
1681                 goto err;
1682
1683         if (net_dev) {
1684                 ret = rdma_copy_addr(&id->route.addr.dev_addr, net_dev, NULL);
1685                 if (ret)
1686                         goto err;
1687         } else {
1688                 if (!cma_any_addr(cma_src_addr(id_priv))) {
1689                         ret = cma_translate_addr(cma_src_addr(id_priv),
1690                                                  &id->route.addr.dev_addr);
1691                         if (ret)
1692                                 goto err;
1693                 }
1694         }
1695
1696         id_priv->state = RDMA_CM_CONNECT;
1697         return id_priv;
1698 err:
1699         rdma_destroy_id(id);
1700         return NULL;
1701 }
1702
1703 static void cma_set_req_event_data(struct rdma_cm_event *event,
1704                                    struct ib_cm_req_event_param *req_data,
1705                                    void *private_data, int offset)
1706 {
1707         event->param.conn.private_data = private_data + offset;
1708         event->param.conn.private_data_len = IB_CM_REQ_PRIVATE_DATA_SIZE - offset;
1709         event->param.conn.responder_resources = req_data->responder_resources;
1710         event->param.conn.initiator_depth = req_data->initiator_depth;
1711         event->param.conn.flow_control = req_data->flow_control;
1712         event->param.conn.retry_count = req_data->retry_count;
1713         event->param.conn.rnr_retry_count = req_data->rnr_retry_count;
1714         event->param.conn.srq = req_data->srq;
1715         event->param.conn.qp_num = req_data->remote_qpn;
1716 }
1717
1718 static int cma_check_req_qp_type(struct rdma_cm_id *id, struct ib_cm_event *ib_event)
1719 {
1720         return (((ib_event->event == IB_CM_REQ_RECEIVED) &&
1721                  (ib_event->param.req_rcvd.qp_type == id->qp_type)) ||
1722                 ((ib_event->event == IB_CM_SIDR_REQ_RECEIVED) &&
1723                  (id->qp_type == IB_QPT_UD)) ||
1724                 (!id->qp_type));
1725 }
1726
1727 static int cma_req_handler(struct ib_cm_id *cm_id, struct ib_cm_event *ib_event)
1728 {
1729         struct rdma_id_private *listen_id, *conn_id;
1730         struct rdma_cm_event event;
1731         struct net_device *net_dev;
1732         int offset, ret;
1733
1734         listen_id = cma_id_from_event(cm_id, ib_event, &net_dev);
1735         if (IS_ERR(listen_id))
1736                 return PTR_ERR(listen_id);
1737
1738         if (!cma_check_req_qp_type(&listen_id->id, ib_event)) {
1739                 ret = -EINVAL;
1740                 goto net_dev_put;
1741         }
1742
1743         if (cma_disable_callback(listen_id, RDMA_CM_LISTEN)) {
1744                 ret = -ECONNABORTED;
1745                 goto net_dev_put;
1746         }
1747
1748         memset(&event, 0, sizeof event);
1749         offset = cma_user_data_offset(listen_id);
1750         event.event = RDMA_CM_EVENT_CONNECT_REQUEST;
1751         if (ib_event->event == IB_CM_SIDR_REQ_RECEIVED) {
1752                 conn_id = cma_new_udp_id(&listen_id->id, ib_event, net_dev);
1753                 event.param.ud.private_data = ib_event->private_data + offset;
1754                 event.param.ud.private_data_len =
1755                                 IB_CM_SIDR_REQ_PRIVATE_DATA_SIZE - offset;
1756         } else {
1757                 conn_id = cma_new_conn_id(&listen_id->id, ib_event, net_dev);
1758                 cma_set_req_event_data(&event, &ib_event->param.req_rcvd,
1759                                        ib_event->private_data, offset);
1760         }
1761         if (!conn_id) {
1762                 ret = -ENOMEM;
1763                 goto err1;
1764         }
1765
1766         mutex_lock_nested(&conn_id->handler_mutex, SINGLE_DEPTH_NESTING);
1767         ret = cma_acquire_dev(conn_id, listen_id);
1768         if (ret)
1769                 goto err2;
1770
1771         conn_id->cm_id.ib = cm_id;
1772         cm_id->context = conn_id;
1773         cm_id->cm_handler = cma_ib_handler;
1774
1775         /*
1776          * Protect against the user destroying conn_id from another thread
1777          * until we're done accessing it.
1778          */
1779         atomic_inc(&conn_id->refcount);
1780         ret = conn_id->id.event_handler(&conn_id->id, &event);
1781         if (ret)
1782                 goto err3;
1783         /*
1784          * Acquire mutex to prevent user executing rdma_destroy_id()
1785          * while we're accessing the cm_id.
1786          */
1787         mutex_lock(&lock);
1788         if (cma_comp(conn_id, RDMA_CM_CONNECT) &&
1789             (conn_id->id.qp_type != IB_QPT_UD))
1790                 ib_send_cm_mra(cm_id, CMA_CM_MRA_SETTING, NULL, 0);
1791         mutex_unlock(&lock);
1792         mutex_unlock(&conn_id->handler_mutex);
1793         mutex_unlock(&listen_id->handler_mutex);
1794         cma_deref_id(conn_id);
1795         if (net_dev)
1796                 dev_put(net_dev);
1797         return 0;
1798
1799 err3:
1800         cma_deref_id(conn_id);
1801         /* Destroy the CM ID by returning a non-zero value. */
1802         conn_id->cm_id.ib = NULL;
1803 err2:
1804         cma_exch(conn_id, RDMA_CM_DESTROYING);
1805         mutex_unlock(&conn_id->handler_mutex);
1806 err1:
1807         mutex_unlock(&listen_id->handler_mutex);
1808         if (conn_id)
1809                 rdma_destroy_id(&conn_id->id);
1810
1811 net_dev_put:
1812         if (net_dev)
1813                 dev_put(net_dev);
1814
1815         return ret;
1816 }
1817
1818 __be64 rdma_get_service_id(struct rdma_cm_id *id, struct sockaddr *addr)
1819 {
1820         if (addr->sa_family == AF_IB)
1821                 return ((struct sockaddr_ib *) addr)->sib_sid;
1822
1823         return cpu_to_be64(((u64)id->ps << 16) + be16_to_cpu(cma_port(addr)));
1824 }
1825 EXPORT_SYMBOL(rdma_get_service_id);
1826
1827 static int cma_iw_handler(struct iw_cm_id *iw_id, struct iw_cm_event *iw_event)
1828 {
1829         struct rdma_id_private *id_priv = iw_id->context;
1830         struct rdma_cm_event event;
1831         int ret = 0;
1832         struct sockaddr *laddr = (struct sockaddr *)&iw_event->local_addr;
1833         struct sockaddr *raddr = (struct sockaddr *)&iw_event->remote_addr;
1834
1835         if (cma_disable_callback(id_priv, RDMA_CM_CONNECT))
1836                 return 0;
1837
1838         memset(&event, 0, sizeof event);
1839         switch (iw_event->event) {
1840         case IW_CM_EVENT_CLOSE:
1841                 event.event = RDMA_CM_EVENT_DISCONNECTED;
1842                 break;
1843         case IW_CM_EVENT_CONNECT_REPLY:
1844                 memcpy(cma_src_addr(id_priv), laddr,
1845                        rdma_addr_size(laddr));
1846                 memcpy(cma_dst_addr(id_priv), raddr,
1847                        rdma_addr_size(raddr));
1848                 switch (iw_event->status) {
1849                 case 0:
1850                         event.event = RDMA_CM_EVENT_ESTABLISHED;
1851                         event.param.conn.initiator_depth = iw_event->ird;
1852                         event.param.conn.responder_resources = iw_event->ord;
1853                         break;
1854                 case -ECONNRESET:
1855                 case -ECONNREFUSED:
1856                         event.event = RDMA_CM_EVENT_REJECTED;
1857                         break;
1858                 case -ETIMEDOUT:
1859                         event.event = RDMA_CM_EVENT_UNREACHABLE;
1860                         break;
1861                 default:
1862                         event.event = RDMA_CM_EVENT_CONNECT_ERROR;
1863                         break;
1864                 }
1865                 break;
1866         case IW_CM_EVENT_ESTABLISHED:
1867                 event.event = RDMA_CM_EVENT_ESTABLISHED;
1868                 event.param.conn.initiator_depth = iw_event->ird;
1869                 event.param.conn.responder_resources = iw_event->ord;
1870                 break;
1871         default:
1872                 BUG_ON(1);
1873         }
1874
1875         event.status = iw_event->status;
1876         event.param.conn.private_data = iw_event->private_data;
1877         event.param.conn.private_data_len = iw_event->private_data_len;
1878         ret = id_priv->id.event_handler(&id_priv->id, &event);
1879         if (ret) {
1880                 /* Destroy the CM ID by returning a non-zero value. */
1881                 id_priv->cm_id.iw = NULL;
1882                 cma_exch(id_priv, RDMA_CM_DESTROYING);
1883                 mutex_unlock(&id_priv->handler_mutex);
1884                 rdma_destroy_id(&id_priv->id);
1885                 return ret;
1886         }
1887
1888         mutex_unlock(&id_priv->handler_mutex);
1889         return ret;
1890 }
1891
1892 static int iw_conn_req_handler(struct iw_cm_id *cm_id,
1893                                struct iw_cm_event *iw_event)
1894 {
1895         struct rdma_cm_id *new_cm_id;
1896         struct rdma_id_private *listen_id, *conn_id;
1897         struct rdma_cm_event event;
1898         int ret;
1899         struct ib_device_attr attr;
1900         struct sockaddr *laddr = (struct sockaddr *)&iw_event->local_addr;
1901         struct sockaddr *raddr = (struct sockaddr *)&iw_event->remote_addr;
1902
1903         listen_id = cm_id->context;
1904         if (cma_disable_callback(listen_id, RDMA_CM_LISTEN))
1905                 return -ECONNABORTED;
1906
1907         /* Create a new RDMA id for the new IW CM ID */
1908         new_cm_id = rdma_create_id(listen_id->id.route.addr.dev_addr.net,
1909                                    listen_id->id.event_handler,
1910                                    listen_id->id.context,
1911                                    RDMA_PS_TCP, IB_QPT_RC);
1912         if (IS_ERR(new_cm_id)) {
1913                 ret = -ENOMEM;
1914                 goto out;
1915         }
1916         conn_id = container_of(new_cm_id, struct rdma_id_private, id);
1917         mutex_lock_nested(&conn_id->handler_mutex, SINGLE_DEPTH_NESTING);
1918         conn_id->state = RDMA_CM_CONNECT;
1919
1920         ret = rdma_translate_ip(laddr, &conn_id->id.route.addr.dev_addr, NULL);
1921         if (ret) {
1922                 mutex_unlock(&conn_id->handler_mutex);
1923                 rdma_destroy_id(new_cm_id);
1924                 goto out;
1925         }
1926
1927         ret = cma_acquire_dev(conn_id, listen_id);
1928         if (ret) {
1929                 mutex_unlock(&conn_id->handler_mutex);
1930                 rdma_destroy_id(new_cm_id);
1931                 goto out;
1932         }
1933
1934         conn_id->cm_id.iw = cm_id;
1935         cm_id->context = conn_id;
1936         cm_id->cm_handler = cma_iw_handler;
1937
1938         memcpy(cma_src_addr(conn_id), laddr, rdma_addr_size(laddr));
1939         memcpy(cma_dst_addr(conn_id), raddr, rdma_addr_size(raddr));
1940
1941         ret = ib_query_device(conn_id->id.device, &attr);
1942         if (ret) {
1943                 mutex_unlock(&conn_id->handler_mutex);
1944                 rdma_destroy_id(new_cm_id);
1945                 goto out;
1946         }
1947
1948         memset(&event, 0, sizeof event);
1949         event.event = RDMA_CM_EVENT_CONNECT_REQUEST;
1950         event.param.conn.private_data = iw_event->private_data;
1951         event.param.conn.private_data_len = iw_event->private_data_len;
1952         event.param.conn.initiator_depth = iw_event->ird;
1953         event.param.conn.responder_resources = iw_event->ord;
1954
1955         /*
1956          * Protect against the user destroying conn_id from another thread
1957          * until we're done accessing it.
1958          */
1959         atomic_inc(&conn_id->refcount);
1960         ret = conn_id->id.event_handler(&conn_id->id, &event);
1961         if (ret) {
1962                 /* User wants to destroy the CM ID */
1963                 conn_id->cm_id.iw = NULL;
1964                 cma_exch(conn_id, RDMA_CM_DESTROYING);
1965                 mutex_unlock(&conn_id->handler_mutex);
1966                 cma_deref_id(conn_id);
1967                 rdma_destroy_id(&conn_id->id);
1968                 goto out;
1969         }
1970
1971         mutex_unlock(&conn_id->handler_mutex);
1972         cma_deref_id(conn_id);
1973
1974 out:
1975         mutex_unlock(&listen_id->handler_mutex);
1976         return ret;
1977 }
1978
1979 static int cma_ib_listen(struct rdma_id_private *id_priv)
1980 {
1981         struct sockaddr *addr;
1982         struct ib_cm_id *id;
1983         __be64 svc_id;
1984
1985         addr = cma_src_addr(id_priv);
1986         svc_id = rdma_get_service_id(&id_priv->id, addr);
1987         id = ib_cm_insert_listen(id_priv->id.device, cma_req_handler, svc_id);
1988         if (IS_ERR(id))
1989                 return PTR_ERR(id);
1990         id_priv->cm_id.ib = id;
1991
1992         return 0;
1993 }
1994
1995 static int cma_iw_listen(struct rdma_id_private *id_priv, int backlog)
1996 {
1997         int ret;
1998         struct iw_cm_id *id;
1999
2000         id = iw_create_cm_id(id_priv->id.device,
2001                              iw_conn_req_handler,
2002                              id_priv);
2003         if (IS_ERR(id))
2004                 return PTR_ERR(id);
2005
2006         id->tos = id_priv->tos;
2007         id_priv->cm_id.iw = id;
2008
2009         memcpy(&id_priv->cm_id.iw->local_addr, cma_src_addr(id_priv),
2010                rdma_addr_size(cma_src_addr(id_priv)));
2011
2012         ret = iw_cm_listen(id_priv->cm_id.iw, backlog);
2013
2014         if (ret) {
2015                 iw_destroy_cm_id(id_priv->cm_id.iw);
2016                 id_priv->cm_id.iw = NULL;
2017         }
2018
2019         return ret;
2020 }
2021
2022 static int cma_listen_handler(struct rdma_cm_id *id,
2023                               struct rdma_cm_event *event)
2024 {
2025         struct rdma_id_private *id_priv = id->context;
2026
2027         id->context = id_priv->id.context;
2028         id->event_handler = id_priv->id.event_handler;
2029         return id_priv->id.event_handler(id, event);
2030 }
2031
2032 static void cma_listen_on_dev(struct rdma_id_private *id_priv,
2033                               struct cma_device *cma_dev)
2034 {
2035         struct rdma_id_private *dev_id_priv;
2036         struct rdma_cm_id *id;
2037         struct net *net = id_priv->id.route.addr.dev_addr.net;
2038         int ret;
2039
2040         if (cma_family(id_priv) == AF_IB && !rdma_cap_ib_cm(cma_dev->device, 1))
2041                 return;
2042
2043         id = rdma_create_id(net, cma_listen_handler, id_priv, id_priv->id.ps,
2044                             id_priv->id.qp_type);
2045         if (IS_ERR(id))
2046                 return;
2047
2048         dev_id_priv = container_of(id, struct rdma_id_private, id);
2049
2050         dev_id_priv->state = RDMA_CM_ADDR_BOUND;
2051         memcpy(cma_src_addr(dev_id_priv), cma_src_addr(id_priv),
2052                rdma_addr_size(cma_src_addr(id_priv)));
2053
2054         cma_attach_to_dev(dev_id_priv, cma_dev);
2055         list_add_tail(&dev_id_priv->listen_list, &id_priv->listen_list);
2056         atomic_inc(&id_priv->refcount);
2057         dev_id_priv->internal_id = 1;
2058         dev_id_priv->afonly = id_priv->afonly;
2059
2060         ret = rdma_listen(id, id_priv->backlog);
2061         if (ret)
2062                 printk(KERN_WARNING "RDMA CMA: cma_listen_on_dev, error %d, "
2063                        "listening on device %s\n", ret, cma_dev->device->name);
2064 }
2065
2066 static void cma_listen_on_all(struct rdma_id_private *id_priv)
2067 {
2068         struct cma_device *cma_dev;
2069
2070         mutex_lock(&lock);
2071         list_add_tail(&id_priv->list, &listen_any_list);
2072         list_for_each_entry(cma_dev, &dev_list, list)
2073                 cma_listen_on_dev(id_priv, cma_dev);
2074         mutex_unlock(&lock);
2075 }
2076
2077 void rdma_set_service_type(struct rdma_cm_id *id, int tos)
2078 {
2079         struct rdma_id_private *id_priv;
2080
2081         id_priv = container_of(id, struct rdma_id_private, id);
2082         id_priv->tos = (u8) tos;
2083 }
2084 EXPORT_SYMBOL(rdma_set_service_type);
2085
2086 static void cma_query_handler(int status, struct ib_sa_path_rec *path_rec,
2087                               void *context)
2088 {
2089         struct cma_work *work = context;
2090         struct rdma_route *route;
2091
2092         route = &work->id->id.route;
2093
2094         if (!status) {
2095                 route->num_paths = 1;
2096                 *route->path_rec = *path_rec;
2097         } else {
2098                 work->old_state = RDMA_CM_ROUTE_QUERY;
2099                 work->new_state = RDMA_CM_ADDR_RESOLVED;
2100                 work->event.event = RDMA_CM_EVENT_ROUTE_ERROR;
2101                 work->event.status = status;
2102         }
2103
2104         queue_work(cma_wq, &work->work);
2105 }
2106
2107 static int cma_query_ib_route(struct rdma_id_private *id_priv, int timeout_ms,
2108                               struct cma_work *work)
2109 {
2110         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
2111         struct ib_sa_path_rec path_rec;
2112         ib_sa_comp_mask comp_mask;
2113         struct sockaddr_in6 *sin6;
2114         struct sockaddr_ib *sib;
2115
2116         memset(&path_rec, 0, sizeof path_rec);
2117         rdma_addr_get_sgid(dev_addr, &path_rec.sgid);
2118         rdma_addr_get_dgid(dev_addr, &path_rec.dgid);
2119         path_rec.pkey = cpu_to_be16(ib_addr_get_pkey(dev_addr));
2120         path_rec.numb_path = 1;
2121         path_rec.reversible = 1;
2122         path_rec.service_id = rdma_get_service_id(&id_priv->id, cma_dst_addr(id_priv));
2123
2124         comp_mask = IB_SA_PATH_REC_DGID | IB_SA_PATH_REC_SGID |
2125                     IB_SA_PATH_REC_PKEY | IB_SA_PATH_REC_NUMB_PATH |
2126                     IB_SA_PATH_REC_REVERSIBLE | IB_SA_PATH_REC_SERVICE_ID;
2127
2128         switch (cma_family(id_priv)) {
2129         case AF_INET:
2130                 path_rec.qos_class = cpu_to_be16((u16) id_priv->tos);
2131                 comp_mask |= IB_SA_PATH_REC_QOS_CLASS;
2132                 break;
2133         case AF_INET6:
2134                 sin6 = (struct sockaddr_in6 *) cma_src_addr(id_priv);
2135                 path_rec.traffic_class = (u8) (be32_to_cpu(sin6->sin6_flowinfo) >> 20);
2136                 comp_mask |= IB_SA_PATH_REC_TRAFFIC_CLASS;
2137                 break;
2138         case AF_IB:
2139                 sib = (struct sockaddr_ib *) cma_src_addr(id_priv);
2140                 path_rec.traffic_class = (u8) (be32_to_cpu(sib->sib_flowinfo) >> 20);
2141                 comp_mask |= IB_SA_PATH_REC_TRAFFIC_CLASS;
2142                 break;
2143         }
2144
2145         id_priv->query_id = ib_sa_path_rec_get(&sa_client, id_priv->id.device,
2146                                                id_priv->id.port_num, &path_rec,
2147                                                comp_mask, timeout_ms,
2148                                                GFP_KERNEL, cma_query_handler,
2149                                                work, &id_priv->query);
2150
2151         return (id_priv->query_id < 0) ? id_priv->query_id : 0;
2152 }
2153
2154 static void cma_work_handler(struct work_struct *_work)
2155 {
2156         struct cma_work *work = container_of(_work, struct cma_work, work);
2157         struct rdma_id_private *id_priv = work->id;
2158         int destroy = 0;
2159
2160         mutex_lock(&id_priv->handler_mutex);
2161         if (!cma_comp_exch(id_priv, work->old_state, work->new_state))
2162                 goto out;
2163
2164         if (id_priv->id.event_handler(&id_priv->id, &work->event)) {
2165                 cma_exch(id_priv, RDMA_CM_DESTROYING);
2166                 destroy = 1;
2167         }
2168 out:
2169         mutex_unlock(&id_priv->handler_mutex);
2170         cma_deref_id(id_priv);
2171         if (destroy)
2172                 rdma_destroy_id(&id_priv->id);
2173         kfree(work);
2174 }
2175
2176 static void cma_ndev_work_handler(struct work_struct *_work)
2177 {
2178         struct cma_ndev_work *work = container_of(_work, struct cma_ndev_work, work);
2179         struct rdma_id_private *id_priv = work->id;
2180         int destroy = 0;
2181
2182         mutex_lock(&id_priv->handler_mutex);
2183         if (id_priv->state == RDMA_CM_DESTROYING ||
2184             id_priv->state == RDMA_CM_DEVICE_REMOVAL)
2185                 goto out;
2186
2187         if (id_priv->id.event_handler(&id_priv->id, &work->event)) {
2188                 cma_exch(id_priv, RDMA_CM_DESTROYING);
2189                 destroy = 1;
2190         }
2191
2192 out:
2193         mutex_unlock(&id_priv->handler_mutex);
2194         cma_deref_id(id_priv);
2195         if (destroy)
2196                 rdma_destroy_id(&id_priv->id);
2197         kfree(work);
2198 }
2199
2200 static int cma_resolve_ib_route(struct rdma_id_private *id_priv, int timeout_ms)
2201 {
2202         struct rdma_route *route = &id_priv->id.route;
2203         struct cma_work *work;
2204         int ret;
2205
2206         work = kzalloc(sizeof *work, GFP_KERNEL);
2207         if (!work)
2208                 return -ENOMEM;
2209
2210         work->id = id_priv;
2211         INIT_WORK(&work->work, cma_work_handler);
2212         work->old_state = RDMA_CM_ROUTE_QUERY;
2213         work->new_state = RDMA_CM_ROUTE_RESOLVED;
2214         work->event.event = RDMA_CM_EVENT_ROUTE_RESOLVED;
2215
2216         route->path_rec = kmalloc(sizeof *route->path_rec, GFP_KERNEL);
2217         if (!route->path_rec) {
2218                 ret = -ENOMEM;
2219                 goto err1;
2220         }
2221
2222         ret = cma_query_ib_route(id_priv, timeout_ms, work);
2223         if (ret)
2224                 goto err2;
2225
2226         return 0;
2227 err2:
2228         kfree(route->path_rec);
2229         route->path_rec = NULL;
2230 err1:
2231         kfree(work);
2232         return ret;
2233 }
2234
2235 int rdma_set_ib_paths(struct rdma_cm_id *id,
2236                       struct ib_sa_path_rec *path_rec, int num_paths)
2237 {
2238         struct rdma_id_private *id_priv;
2239         int ret;
2240
2241         id_priv = container_of(id, struct rdma_id_private, id);
2242         if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_RESOLVED,
2243                            RDMA_CM_ROUTE_RESOLVED))
2244                 return -EINVAL;
2245
2246         id->route.path_rec = kmemdup(path_rec, sizeof *path_rec * num_paths,
2247                                      GFP_KERNEL);
2248         if (!id->route.path_rec) {
2249                 ret = -ENOMEM;
2250                 goto err;
2251         }
2252
2253         id->route.num_paths = num_paths;
2254         return 0;
2255 err:
2256         cma_comp_exch(id_priv, RDMA_CM_ROUTE_RESOLVED, RDMA_CM_ADDR_RESOLVED);
2257         return ret;
2258 }
2259 EXPORT_SYMBOL(rdma_set_ib_paths);
2260
2261 static int cma_resolve_iw_route(struct rdma_id_private *id_priv, int timeout_ms)
2262 {
2263         struct cma_work *work;
2264
2265         work = kzalloc(sizeof *work, GFP_KERNEL);
2266         if (!work)
2267                 return -ENOMEM;
2268
2269         work->id = id_priv;
2270         INIT_WORK(&work->work, cma_work_handler);
2271         work->old_state = RDMA_CM_ROUTE_QUERY;
2272         work->new_state = RDMA_CM_ROUTE_RESOLVED;
2273         work->event.event = RDMA_CM_EVENT_ROUTE_RESOLVED;
2274         queue_work(cma_wq, &work->work);
2275         return 0;
2276 }
2277
2278 static int iboe_tos_to_sl(struct net_device *ndev, int tos)
2279 {
2280         int prio;
2281         struct net_device *dev;
2282
2283         prio = rt_tos2priority(tos);
2284         dev = ndev->priv_flags & IFF_802_1Q_VLAN ?
2285                 vlan_dev_real_dev(ndev) : ndev;
2286
2287         if (dev->num_tc)
2288                 return netdev_get_prio_tc_map(dev, prio);
2289
2290 #if IS_ENABLED(CONFIG_VLAN_8021Q)
2291         if (ndev->priv_flags & IFF_802_1Q_VLAN)
2292                 return (vlan_dev_get_egress_qos_mask(ndev, prio) &
2293                         VLAN_PRIO_MASK) >> VLAN_PRIO_SHIFT;
2294 #endif
2295         return 0;
2296 }
2297
2298 static int cma_resolve_iboe_route(struct rdma_id_private *id_priv)
2299 {
2300         struct rdma_route *route = &id_priv->id.route;
2301         struct rdma_addr *addr = &route->addr;
2302         struct cma_work *work;
2303         int ret;
2304         struct net_device *ndev = NULL;
2305
2306
2307         work = kzalloc(sizeof *work, GFP_KERNEL);
2308         if (!work)
2309                 return -ENOMEM;
2310
2311         work->id = id_priv;
2312         INIT_WORK(&work->work, cma_work_handler);
2313
2314         route->path_rec = kzalloc(sizeof *route->path_rec, GFP_KERNEL);
2315         if (!route->path_rec) {
2316                 ret = -ENOMEM;
2317                 goto err1;
2318         }
2319
2320         route->num_paths = 1;
2321
2322         if (addr->dev_addr.bound_dev_if) {
2323                 ndev = dev_get_by_index(&init_net, addr->dev_addr.bound_dev_if);
2324                 route->path_rec->net = &init_net;
2325                 route->path_rec->ifindex = addr->dev_addr.bound_dev_if;
2326         }
2327         if (!ndev) {
2328                 ret = -ENODEV;
2329                 goto err2;
2330         }
2331
2332         memcpy(route->path_rec->dmac, addr->dev_addr.dst_dev_addr, ETH_ALEN);
2333
2334         rdma_ip2gid((struct sockaddr *)&id_priv->id.route.addr.src_addr,
2335                     &route->path_rec->sgid);
2336         rdma_ip2gid((struct sockaddr *)&id_priv->id.route.addr.dst_addr,
2337                     &route->path_rec->dgid);
2338
2339         route->path_rec->hop_limit = 1;
2340         route->path_rec->reversible = 1;
2341         route->path_rec->pkey = cpu_to_be16(0xffff);
2342         route->path_rec->mtu_selector = IB_SA_EQ;
2343         route->path_rec->sl = iboe_tos_to_sl(ndev, id_priv->tos);
2344         route->path_rec->mtu = iboe_get_mtu(ndev->mtu);
2345         route->path_rec->rate_selector = IB_SA_EQ;
2346         route->path_rec->rate = iboe_get_rate(ndev);
2347         dev_put(ndev);
2348         route->path_rec->packet_life_time_selector = IB_SA_EQ;
2349         route->path_rec->packet_life_time = CMA_IBOE_PACKET_LIFETIME;
2350         if (!route->path_rec->mtu) {
2351                 ret = -EINVAL;
2352                 goto err2;
2353         }
2354
2355         work->old_state = RDMA_CM_ROUTE_QUERY;
2356         work->new_state = RDMA_CM_ROUTE_RESOLVED;
2357         work->event.event = RDMA_CM_EVENT_ROUTE_RESOLVED;
2358         work->event.status = 0;
2359
2360         queue_work(cma_wq, &work->work);
2361
2362         return 0;
2363
2364 err2:
2365         kfree(route->path_rec);
2366         route->path_rec = NULL;
2367 err1:
2368         kfree(work);
2369         return ret;
2370 }
2371
2372 int rdma_resolve_route(struct rdma_cm_id *id, int timeout_ms)
2373 {
2374         struct rdma_id_private *id_priv;
2375         int ret;
2376
2377         id_priv = container_of(id, struct rdma_id_private, id);
2378         if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_RESOLVED, RDMA_CM_ROUTE_QUERY))
2379                 return -EINVAL;
2380
2381         atomic_inc(&id_priv->refcount);
2382         if (rdma_cap_ib_sa(id->device, id->port_num))
2383                 ret = cma_resolve_ib_route(id_priv, timeout_ms);
2384         else if (rdma_protocol_roce(id->device, id->port_num))
2385                 ret = cma_resolve_iboe_route(id_priv);
2386         else if (rdma_protocol_iwarp(id->device, id->port_num))
2387                 ret = cma_resolve_iw_route(id_priv, timeout_ms);
2388         else
2389                 ret = -ENOSYS;
2390
2391         if (ret)
2392                 goto err;
2393
2394         return 0;
2395 err:
2396         cma_comp_exch(id_priv, RDMA_CM_ROUTE_QUERY, RDMA_CM_ADDR_RESOLVED);
2397         cma_deref_id(id_priv);
2398         return ret;
2399 }
2400 EXPORT_SYMBOL(rdma_resolve_route);
2401
2402 static void cma_set_loopback(struct sockaddr *addr)
2403 {
2404         switch (addr->sa_family) {
2405         case AF_INET:
2406                 ((struct sockaddr_in *) addr)->sin_addr.s_addr = htonl(INADDR_LOOPBACK);
2407                 break;
2408         case AF_INET6:
2409                 ipv6_addr_set(&((struct sockaddr_in6 *) addr)->sin6_addr,
2410                               0, 0, 0, htonl(1));
2411                 break;
2412         default:
2413                 ib_addr_set(&((struct sockaddr_ib *) addr)->sib_addr,
2414                             0, 0, 0, htonl(1));
2415                 break;
2416         }
2417 }
2418
2419 static int cma_bind_loopback(struct rdma_id_private *id_priv)
2420 {
2421         struct cma_device *cma_dev, *cur_dev;
2422         struct ib_port_attr port_attr;
2423         union ib_gid gid;
2424         u16 pkey;
2425         int ret;
2426         u8 p;
2427
2428         cma_dev = NULL;
2429         mutex_lock(&lock);
2430         list_for_each_entry(cur_dev, &dev_list, list) {
2431                 if (cma_family(id_priv) == AF_IB &&
2432                     !rdma_cap_ib_cm(cur_dev->device, 1))
2433                         continue;
2434
2435                 if (!cma_dev)
2436                         cma_dev = cur_dev;
2437
2438                 for (p = 1; p <= cur_dev->device->phys_port_cnt; ++p) {
2439                         if (!ib_query_port(cur_dev->device, p, &port_attr) &&
2440                             port_attr.state == IB_PORT_ACTIVE) {
2441                                 cma_dev = cur_dev;
2442                                 goto port_found;
2443                         }
2444                 }
2445         }
2446
2447         if (!cma_dev) {
2448                 ret = -ENODEV;
2449                 goto out;
2450         }
2451
2452         p = 1;
2453
2454 port_found:
2455         ret = ib_get_cached_gid(cma_dev->device, p, 0, &gid, NULL);
2456         if (ret)
2457                 goto out;
2458
2459         ret = ib_get_cached_pkey(cma_dev->device, p, 0, &pkey);
2460         if (ret)
2461                 goto out;
2462
2463         id_priv->id.route.addr.dev_addr.dev_type =
2464                 (rdma_protocol_ib(cma_dev->device, p)) ?
2465                 ARPHRD_INFINIBAND : ARPHRD_ETHER;
2466
2467         rdma_addr_set_sgid(&id_priv->id.route.addr.dev_addr, &gid);
2468         ib_addr_set_pkey(&id_priv->id.route.addr.dev_addr, pkey);
2469         id_priv->id.port_num = p;
2470         cma_attach_to_dev(id_priv, cma_dev);
2471         cma_set_loopback(cma_src_addr(id_priv));
2472 out:
2473         mutex_unlock(&lock);
2474         return ret;
2475 }
2476
2477 static void addr_handler(int status, struct sockaddr *src_addr,
2478                          struct rdma_dev_addr *dev_addr, void *context)
2479 {
2480         struct rdma_id_private *id_priv = context;
2481         struct rdma_cm_event event;
2482
2483         memset(&event, 0, sizeof event);
2484         mutex_lock(&id_priv->handler_mutex);
2485         if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_QUERY,
2486                            RDMA_CM_ADDR_RESOLVED))
2487                 goto out;
2488
2489         memcpy(cma_src_addr(id_priv), src_addr, rdma_addr_size(src_addr));
2490         if (!status && !id_priv->cma_dev)
2491                 status = cma_acquire_dev(id_priv, NULL);
2492
2493         if (status) {
2494                 if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_RESOLVED,
2495                                    RDMA_CM_ADDR_BOUND))
2496                         goto out;
2497                 event.event = RDMA_CM_EVENT_ADDR_ERROR;
2498                 event.status = status;
2499         } else
2500                 event.event = RDMA_CM_EVENT_ADDR_RESOLVED;
2501
2502         if (id_priv->id.event_handler(&id_priv->id, &event)) {
2503                 cma_exch(id_priv, RDMA_CM_DESTROYING);
2504                 mutex_unlock(&id_priv->handler_mutex);
2505                 cma_deref_id(id_priv);
2506                 rdma_destroy_id(&id_priv->id);
2507                 return;
2508         }
2509 out:
2510         mutex_unlock(&id_priv->handler_mutex);
2511         cma_deref_id(id_priv);
2512 }
2513
2514 static int cma_resolve_loopback(struct rdma_id_private *id_priv)
2515 {
2516         struct cma_work *work;
2517         union ib_gid gid;
2518         int ret;
2519
2520         work = kzalloc(sizeof *work, GFP_KERNEL);
2521         if (!work)
2522                 return -ENOMEM;
2523
2524         if (!id_priv->cma_dev) {
2525                 ret = cma_bind_loopback(id_priv);
2526                 if (ret)
2527                         goto err;
2528         }
2529
2530         rdma_addr_get_sgid(&id_priv->id.route.addr.dev_addr, &gid);
2531         rdma_addr_set_dgid(&id_priv->id.route.addr.dev_addr, &gid);
2532
2533         work->id = id_priv;
2534         INIT_WORK(&work->work, cma_work_handler);
2535         work->old_state = RDMA_CM_ADDR_QUERY;
2536         work->new_state = RDMA_CM_ADDR_RESOLVED;
2537         work->event.event = RDMA_CM_EVENT_ADDR_RESOLVED;
2538         queue_work(cma_wq, &work->work);
2539         return 0;
2540 err:
2541         kfree(work);
2542         return ret;
2543 }
2544
2545 static int cma_resolve_ib_addr(struct rdma_id_private *id_priv)
2546 {
2547         struct cma_work *work;
2548         int ret;
2549
2550         work = kzalloc(sizeof *work, GFP_KERNEL);
2551         if (!work)
2552                 return -ENOMEM;
2553
2554         if (!id_priv->cma_dev) {
2555                 ret = cma_resolve_ib_dev(id_priv);
2556                 if (ret)
2557                         goto err;
2558         }
2559
2560         rdma_addr_set_dgid(&id_priv->id.route.addr.dev_addr, (union ib_gid *)
2561                 &(((struct sockaddr_ib *) &id_priv->id.route.addr.dst_addr)->sib_addr));
2562
2563         work->id = id_priv;
2564         INIT_WORK(&work->work, cma_work_handler);
2565         work->old_state = RDMA_CM_ADDR_QUERY;
2566         work->new_state = RDMA_CM_ADDR_RESOLVED;
2567         work->event.event = RDMA_CM_EVENT_ADDR_RESOLVED;
2568         queue_work(cma_wq, &work->work);
2569         return 0;
2570 err:
2571         kfree(work);
2572         return ret;
2573 }
2574
2575 static int cma_bind_addr(struct rdma_cm_id *id, struct sockaddr *src_addr,
2576                          struct sockaddr *dst_addr)
2577 {
2578         if (!src_addr || !src_addr->sa_family) {
2579                 src_addr = (struct sockaddr *) &id->route.addr.src_addr;
2580                 src_addr->sa_family = dst_addr->sa_family;
2581                 if (dst_addr->sa_family == AF_INET6) {
2582                         struct sockaddr_in6 *src_addr6 = (struct sockaddr_in6 *) src_addr;
2583                         struct sockaddr_in6 *dst_addr6 = (struct sockaddr_in6 *) dst_addr;
2584                         src_addr6->sin6_scope_id = dst_addr6->sin6_scope_id;
2585                         if (ipv6_addr_type(&dst_addr6->sin6_addr) & IPV6_ADDR_LINKLOCAL)
2586                                 id->route.addr.dev_addr.bound_dev_if = dst_addr6->sin6_scope_id;
2587                 } else if (dst_addr->sa_family == AF_IB) {
2588                         ((struct sockaddr_ib *) src_addr)->sib_pkey =
2589                                 ((struct sockaddr_ib *) dst_addr)->sib_pkey;
2590                 }
2591         }
2592         return rdma_bind_addr(id, src_addr);
2593 }
2594
2595 int rdma_resolve_addr(struct rdma_cm_id *id, struct sockaddr *src_addr,
2596                       struct sockaddr *dst_addr, int timeout_ms)
2597 {
2598         struct rdma_id_private *id_priv;
2599         int ret;
2600
2601         id_priv = container_of(id, struct rdma_id_private, id);
2602         if (id_priv->state == RDMA_CM_IDLE) {
2603                 ret = cma_bind_addr(id, src_addr, dst_addr);
2604                 if (ret)
2605                         return ret;
2606         }
2607
2608         if (cma_family(id_priv) != dst_addr->sa_family)
2609                 return -EINVAL;
2610
2611         if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_BOUND, RDMA_CM_ADDR_QUERY))
2612                 return -EINVAL;
2613
2614         atomic_inc(&id_priv->refcount);
2615         memcpy(cma_dst_addr(id_priv), dst_addr, rdma_addr_size(dst_addr));
2616         if (cma_any_addr(dst_addr)) {
2617                 ret = cma_resolve_loopback(id_priv);
2618         } else {
2619                 if (dst_addr->sa_family == AF_IB) {
2620                         ret = cma_resolve_ib_addr(id_priv);
2621                 } else {
2622                         ret = rdma_resolve_ip(&addr_client, cma_src_addr(id_priv),
2623                                               dst_addr, &id->route.addr.dev_addr,
2624                                               timeout_ms, addr_handler, id_priv);
2625                 }
2626         }
2627         if (ret)
2628                 goto err;
2629
2630         return 0;
2631 err:
2632         cma_comp_exch(id_priv, RDMA_CM_ADDR_QUERY, RDMA_CM_ADDR_BOUND);
2633         cma_deref_id(id_priv);
2634         return ret;
2635 }
2636 EXPORT_SYMBOL(rdma_resolve_addr);
2637
2638 int rdma_set_reuseaddr(struct rdma_cm_id *id, int reuse)
2639 {
2640         struct rdma_id_private *id_priv;
2641         unsigned long flags;
2642         int ret;
2643
2644         id_priv = container_of(id, struct rdma_id_private, id);
2645         spin_lock_irqsave(&id_priv->lock, flags);
2646         if (reuse || id_priv->state == RDMA_CM_IDLE) {
2647                 id_priv->reuseaddr = reuse;
2648                 ret = 0;
2649         } else {
2650                 ret = -EINVAL;
2651         }
2652         spin_unlock_irqrestore(&id_priv->lock, flags);
2653         return ret;
2654 }
2655 EXPORT_SYMBOL(rdma_set_reuseaddr);
2656
2657 int rdma_set_afonly(struct rdma_cm_id *id, int afonly)
2658 {
2659         struct rdma_id_private *id_priv;
2660         unsigned long flags;
2661         int ret;
2662
2663         id_priv = container_of(id, struct rdma_id_private, id);
2664         spin_lock_irqsave(&id_priv->lock, flags);
2665         if (id_priv->state == RDMA_CM_IDLE || id_priv->state == RDMA_CM_ADDR_BOUND) {
2666                 id_priv->options |= (1 << CMA_OPTION_AFONLY);
2667                 id_priv->afonly = afonly;
2668                 ret = 0;
2669         } else {
2670                 ret = -EINVAL;
2671         }
2672         spin_unlock_irqrestore(&id_priv->lock, flags);
2673         return ret;
2674 }
2675 EXPORT_SYMBOL(rdma_set_afonly);
2676
2677 static void cma_bind_port(struct rdma_bind_list *bind_list,
2678                           struct rdma_id_private *id_priv)
2679 {
2680         struct sockaddr *addr;
2681         struct sockaddr_ib *sib;
2682         u64 sid, mask;
2683         __be16 port;
2684
2685         addr = cma_src_addr(id_priv);
2686         port = htons(bind_list->port);
2687
2688         switch (addr->sa_family) {
2689         case AF_INET:
2690                 ((struct sockaddr_in *) addr)->sin_port = port;
2691                 break;
2692         case AF_INET6:
2693                 ((struct sockaddr_in6 *) addr)->sin6_port = port;
2694                 break;
2695         case AF_IB:
2696                 sib = (struct sockaddr_ib *) addr;
2697                 sid = be64_to_cpu(sib->sib_sid);
2698                 mask = be64_to_cpu(sib->sib_sid_mask);
2699                 sib->sib_sid = cpu_to_be64((sid & mask) | (u64) ntohs(port));
2700                 sib->sib_sid_mask = cpu_to_be64(~0ULL);
2701                 break;
2702         }
2703         id_priv->bind_list = bind_list;
2704         hlist_add_head(&id_priv->node, &bind_list->owners);
2705 }
2706
2707 static int cma_alloc_port(enum rdma_port_space ps,
2708                           struct rdma_id_private *id_priv, unsigned short snum)
2709 {
2710         struct rdma_bind_list *bind_list;
2711         int ret;
2712
2713         bind_list = kzalloc(sizeof *bind_list, GFP_KERNEL);
2714         if (!bind_list)
2715                 return -ENOMEM;
2716
2717         ret = cma_ps_alloc(id_priv->id.route.addr.dev_addr.net, ps, bind_list,
2718                            snum);
2719         if (ret < 0)
2720                 goto err;
2721
2722         bind_list->ps = ps;
2723         bind_list->port = (unsigned short)ret;
2724         cma_bind_port(bind_list, id_priv);
2725         return 0;
2726 err:
2727         kfree(bind_list);
2728         return ret == -ENOSPC ? -EADDRNOTAVAIL : ret;
2729 }
2730
2731 static int cma_alloc_any_port(enum rdma_port_space ps,
2732                               struct rdma_id_private *id_priv)
2733 {
2734         static unsigned int last_used_port;
2735         int low, high, remaining;
2736         unsigned int rover;
2737         struct net *net = id_priv->id.route.addr.dev_addr.net;
2738
2739         inet_get_local_port_range(net, &low, &high);
2740         remaining = (high - low) + 1;
2741         rover = prandom_u32() % remaining + low;
2742 retry:
2743         if (last_used_port != rover &&
2744             !cma_ps_find(net, ps, (unsigned short)rover)) {
2745                 int ret = cma_alloc_port(ps, id_priv, rover);
2746                 /*
2747                  * Remember previously used port number in order to avoid
2748                  * re-using same port immediately after it is closed.
2749                  */
2750                 if (!ret)
2751                         last_used_port = rover;
2752                 if (ret != -EADDRNOTAVAIL)
2753                         return ret;
2754         }
2755         if (--remaining) {
2756                 rover++;
2757                 if ((rover < low) || (rover > high))
2758                         rover = low;
2759                 goto retry;
2760         }
2761         return -EADDRNOTAVAIL;
2762 }
2763
2764 /*
2765  * Check that the requested port is available.  This is called when trying to
2766  * bind to a specific port, or when trying to listen on a bound port.  In
2767  * the latter case, the provided id_priv may already be on the bind_list, but
2768  * we still need to check that it's okay to start listening.
2769  */
2770 static int cma_check_port(struct rdma_bind_list *bind_list,
2771                           struct rdma_id_private *id_priv, uint8_t reuseaddr)
2772 {
2773         struct rdma_id_private *cur_id;
2774         struct sockaddr *addr, *cur_addr;
2775
2776         addr = cma_src_addr(id_priv);
2777         hlist_for_each_entry(cur_id, &bind_list->owners, node) {
2778                 if (id_priv == cur_id)
2779                         continue;
2780
2781                 if ((cur_id->state != RDMA_CM_LISTEN) && reuseaddr &&
2782                     cur_id->reuseaddr)
2783                         continue;
2784
2785                 cur_addr = cma_src_addr(cur_id);
2786                 if (id_priv->afonly && cur_id->afonly &&
2787                     (addr->sa_family != cur_addr->sa_family))
2788                         continue;
2789
2790                 if (cma_any_addr(addr) || cma_any_addr(cur_addr))
2791                         return -EADDRNOTAVAIL;
2792
2793                 if (!cma_addr_cmp(addr, cur_addr))
2794                         return -EADDRINUSE;
2795         }
2796         return 0;
2797 }
2798
2799 static int cma_use_port(enum rdma_port_space ps,
2800                         struct rdma_id_private *id_priv)
2801 {
2802         struct rdma_bind_list *bind_list;
2803         unsigned short snum;
2804         int ret;
2805
2806         snum = ntohs(cma_port(cma_src_addr(id_priv)));
2807         if (snum < PROT_SOCK && !capable(CAP_NET_BIND_SERVICE))
2808                 return -EACCES;
2809
2810         bind_list = cma_ps_find(id_priv->id.route.addr.dev_addr.net, ps, snum);
2811         if (!bind_list) {
2812                 ret = cma_alloc_port(ps, id_priv, snum);
2813         } else {
2814                 ret = cma_check_port(bind_list, id_priv, id_priv->reuseaddr);
2815                 if (!ret)
2816                         cma_bind_port(bind_list, id_priv);
2817         }
2818         return ret;
2819 }
2820
2821 static int cma_bind_listen(struct rdma_id_private *id_priv)
2822 {
2823         struct rdma_bind_list *bind_list = id_priv->bind_list;
2824         int ret = 0;
2825
2826         mutex_lock(&lock);
2827         if (bind_list->owners.first->next)
2828                 ret = cma_check_port(bind_list, id_priv, 0);
2829         mutex_unlock(&lock);
2830         return ret;
2831 }
2832
2833 static enum rdma_port_space cma_select_inet_ps(
2834                 struct rdma_id_private *id_priv)
2835 {
2836         switch (id_priv->id.ps) {
2837         case RDMA_PS_TCP:
2838         case RDMA_PS_UDP:
2839         case RDMA_PS_IPOIB:
2840         case RDMA_PS_IB:
2841                 return id_priv->id.ps;
2842         default:
2843
2844                 return 0;
2845         }
2846 }
2847
2848 static enum rdma_port_space cma_select_ib_ps(struct rdma_id_private *id_priv)
2849 {
2850         enum rdma_port_space ps = 0;
2851         struct sockaddr_ib *sib;
2852         u64 sid_ps, mask, sid;
2853
2854         sib = (struct sockaddr_ib *) cma_src_addr(id_priv);
2855         mask = be64_to_cpu(sib->sib_sid_mask) & RDMA_IB_IP_PS_MASK;
2856         sid = be64_to_cpu(sib->sib_sid) & mask;
2857
2858         if ((id_priv->id.ps == RDMA_PS_IB) && (sid == (RDMA_IB_IP_PS_IB & mask))) {
2859                 sid_ps = RDMA_IB_IP_PS_IB;
2860                 ps = RDMA_PS_IB;
2861         } else if (((id_priv->id.ps == RDMA_PS_IB) || (id_priv->id.ps == RDMA_PS_TCP)) &&
2862                    (sid == (RDMA_IB_IP_PS_TCP & mask))) {
2863                 sid_ps = RDMA_IB_IP_PS_TCP;
2864                 ps = RDMA_PS_TCP;
2865         } else if (((id_priv->id.ps == RDMA_PS_IB) || (id_priv->id.ps == RDMA_PS_UDP)) &&
2866                    (sid == (RDMA_IB_IP_PS_UDP & mask))) {
2867                 sid_ps = RDMA_IB_IP_PS_UDP;
2868                 ps = RDMA_PS_UDP;
2869         }
2870
2871         if (ps) {
2872                 sib->sib_sid = cpu_to_be64(sid_ps | ntohs(cma_port((struct sockaddr *) sib)));
2873                 sib->sib_sid_mask = cpu_to_be64(RDMA_IB_IP_PS_MASK |
2874                                                 be64_to_cpu(sib->sib_sid_mask));
2875         }
2876         return ps;
2877 }
2878
2879 static int cma_get_port(struct rdma_id_private *id_priv)
2880 {
2881         enum rdma_port_space ps;
2882         int ret;
2883
2884         if (cma_family(id_priv) != AF_IB)
2885                 ps = cma_select_inet_ps(id_priv);
2886         else
2887                 ps = cma_select_ib_ps(id_priv);
2888         if (!ps)
2889                 return -EPROTONOSUPPORT;
2890
2891         mutex_lock(&lock);
2892         if (cma_any_port(cma_src_addr(id_priv)))
2893                 ret = cma_alloc_any_port(ps, id_priv);
2894         else
2895                 ret = cma_use_port(ps, id_priv);
2896         mutex_unlock(&lock);
2897
2898         return ret;
2899 }
2900
2901 static int cma_check_linklocal(struct rdma_dev_addr *dev_addr,
2902                                struct sockaddr *addr)
2903 {
2904 #if IS_ENABLED(CONFIG_IPV6)
2905         struct sockaddr_in6 *sin6;
2906
2907         if (addr->sa_family != AF_INET6)
2908                 return 0;
2909
2910         sin6 = (struct sockaddr_in6 *) addr;
2911
2912         if (!(ipv6_addr_type(&sin6->sin6_addr) & IPV6_ADDR_LINKLOCAL))
2913                 return 0;
2914
2915         if (!sin6->sin6_scope_id)
2916                         return -EINVAL;
2917
2918         dev_addr->bound_dev_if = sin6->sin6_scope_id;
2919 #endif
2920         return 0;
2921 }
2922
2923 int rdma_listen(struct rdma_cm_id *id, int backlog)
2924 {
2925         struct rdma_id_private *id_priv;
2926         int ret;
2927
2928         id_priv = container_of(id, struct rdma_id_private, id);
2929         if (id_priv->state == RDMA_CM_IDLE) {
2930                 id->route.addr.src_addr.ss_family = AF_INET;
2931                 ret = rdma_bind_addr(id, cma_src_addr(id_priv));
2932                 if (ret)
2933                         return ret;
2934         }
2935
2936         if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_BOUND, RDMA_CM_LISTEN))
2937                 return -EINVAL;
2938
2939         if (id_priv->reuseaddr) {
2940                 ret = cma_bind_listen(id_priv);
2941                 if (ret)
2942                         goto err;
2943         }
2944
2945         id_priv->backlog = backlog;
2946         if (id->device) {
2947                 if (rdma_cap_ib_cm(id->device, 1)) {
2948                         ret = cma_ib_listen(id_priv);
2949                         if (ret)
2950                                 goto err;
2951                 } else if (rdma_cap_iw_cm(id->device, 1)) {
2952                         ret = cma_iw_listen(id_priv, backlog);
2953                         if (ret)
2954                                 goto err;
2955                 } else {
2956                         ret = -ENOSYS;
2957                         goto err;
2958                 }
2959         } else
2960                 cma_listen_on_all(id_priv);
2961
2962         return 0;
2963 err:
2964         id_priv->backlog = 0;
2965         cma_comp_exch(id_priv, RDMA_CM_LISTEN, RDMA_CM_ADDR_BOUND);
2966         return ret;
2967 }
2968 EXPORT_SYMBOL(rdma_listen);
2969
2970 int rdma_bind_addr(struct rdma_cm_id *id, struct sockaddr *addr)
2971 {
2972         struct rdma_id_private *id_priv;
2973         int ret;
2974
2975         if (addr->sa_family != AF_INET && addr->sa_family != AF_INET6 &&
2976             addr->sa_family != AF_IB)
2977                 return -EAFNOSUPPORT;
2978
2979         id_priv = container_of(id, struct rdma_id_private, id);
2980         if (!cma_comp_exch(id_priv, RDMA_CM_IDLE, RDMA_CM_ADDR_BOUND))
2981                 return -EINVAL;
2982
2983         ret = cma_check_linklocal(&id->route.addr.dev_addr, addr);
2984         if (ret)
2985                 goto err1;
2986
2987         memcpy(cma_src_addr(id_priv), addr, rdma_addr_size(addr));
2988         if (!cma_any_addr(addr)) {
2989                 ret = cma_translate_addr(addr, &id->route.addr.dev_addr);
2990                 if (ret)
2991                         goto err1;
2992
2993                 ret = cma_acquire_dev(id_priv, NULL);
2994                 if (ret)
2995                         goto err1;
2996         }
2997
2998         if (!(id_priv->options & (1 << CMA_OPTION_AFONLY))) {
2999                 if (addr->sa_family == AF_INET)
3000                         id_priv->afonly = 1;
3001 #if IS_ENABLED(CONFIG_IPV6)
3002                 else if (addr->sa_family == AF_INET6) {
3003                         struct net *net = id_priv->id.route.addr.dev_addr.net;
3004
3005                         id_priv->afonly = net->ipv6.sysctl.bindv6only;
3006                 }
3007 #endif
3008         }
3009         ret = cma_get_port(id_priv);
3010         if (ret)
3011                 goto err2;
3012
3013         return 0;
3014 err2:
3015         if (id_priv->cma_dev)
3016                 cma_release_dev(id_priv);
3017 err1:
3018         cma_comp_exch(id_priv, RDMA_CM_ADDR_BOUND, RDMA_CM_IDLE);
3019         return ret;
3020 }
3021 EXPORT_SYMBOL(rdma_bind_addr);
3022
3023 static int cma_format_hdr(void *hdr, struct rdma_id_private *id_priv)
3024 {
3025         struct cma_hdr *cma_hdr;
3026
3027         cma_hdr = hdr;
3028         cma_hdr->cma_version = CMA_VERSION;
3029         if (cma_family(id_priv) == AF_INET) {
3030                 struct sockaddr_in *src4, *dst4;
3031
3032                 src4 = (struct sockaddr_in *) cma_src_addr(id_priv);
3033                 dst4 = (struct sockaddr_in *) cma_dst_addr(id_priv);
3034
3035                 cma_set_ip_ver(cma_hdr, 4);
3036                 cma_hdr->src_addr.ip4.addr = src4->sin_addr.s_addr;
3037                 cma_hdr->dst_addr.ip4.addr = dst4->sin_addr.s_addr;
3038                 cma_hdr->port = src4->sin_port;
3039         } else if (cma_family(id_priv) == AF_INET6) {
3040                 struct sockaddr_in6 *src6, *dst6;
3041
3042                 src6 = (struct sockaddr_in6 *) cma_src_addr(id_priv);
3043                 dst6 = (struct sockaddr_in6 *) cma_dst_addr(id_priv);
3044
3045                 cma_set_ip_ver(cma_hdr, 6);
3046                 cma_hdr->src_addr.ip6 = src6->sin6_addr;
3047                 cma_hdr->dst_addr.ip6 = dst6->sin6_addr;
3048                 cma_hdr->port = src6->sin6_port;
3049         }
3050         return 0;
3051 }
3052
3053 static int cma_sidr_rep_handler(struct ib_cm_id *cm_id,
3054                                 struct ib_cm_event *ib_event)
3055 {
3056         struct rdma_id_private *id_priv = cm_id->context;
3057         struct rdma_cm_event event;
3058         struct ib_cm_sidr_rep_event_param *rep = &ib_event->param.sidr_rep_rcvd;
3059         int ret = 0;
3060
3061         if (cma_disable_callback(id_priv, RDMA_CM_CONNECT))
3062                 return 0;
3063
3064         memset(&event, 0, sizeof event);
3065         switch (ib_event->event) {
3066         case IB_CM_SIDR_REQ_ERROR:
3067                 event.event = RDMA_CM_EVENT_UNREACHABLE;
3068                 event.status = -ETIMEDOUT;
3069                 break;
3070         case IB_CM_SIDR_REP_RECEIVED:
3071                 event.param.ud.private_data = ib_event->private_data;
3072                 event.param.ud.private_data_len = IB_CM_SIDR_REP_PRIVATE_DATA_SIZE;
3073                 if (rep->status != IB_SIDR_SUCCESS) {
3074                         event.event = RDMA_CM_EVENT_UNREACHABLE;
3075                         event.status = ib_event->param.sidr_rep_rcvd.status;
3076                         break;
3077                 }
3078                 ret = cma_set_qkey(id_priv, rep->qkey);
3079                 if (ret) {
3080                         event.event = RDMA_CM_EVENT_ADDR_ERROR;
3081                         event.status = ret;
3082                         break;
3083                 }
3084                 ib_init_ah_from_path(id_priv->id.device, id_priv->id.port_num,
3085                                      id_priv->id.route.path_rec,
3086                                      &event.param.ud.ah_attr);
3087                 event.param.ud.qp_num = rep->qpn;
3088                 event.param.ud.qkey = rep->qkey;
3089                 event.event = RDMA_CM_EVENT_ESTABLISHED;
3090                 event.status = 0;
3091                 break;
3092         default:
3093                 printk(KERN_ERR "RDMA CMA: unexpected IB CM event: %d\n",
3094                        ib_event->event);
3095                 goto out;
3096         }
3097
3098         ret = id_priv->id.event_handler(&id_priv->id, &event);
3099         if (ret) {
3100                 /* Destroy the CM ID by returning a non-zero value. */
3101                 id_priv->cm_id.ib = NULL;
3102                 cma_exch(id_priv, RDMA_CM_DESTROYING);
3103                 mutex_unlock(&id_priv->handler_mutex);
3104                 rdma_destroy_id(&id_priv->id);
3105                 return ret;
3106         }
3107 out:
3108         mutex_unlock(&id_priv->handler_mutex);
3109         return ret;
3110 }
3111
3112 static int cma_resolve_ib_udp(struct rdma_id_private *id_priv,
3113                               struct rdma_conn_param *conn_param)
3114 {
3115         struct ib_cm_sidr_req_param req;
3116         struct ib_cm_id *id;
3117         void *private_data;
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         if (private_data) {
3139                 ret = cma_format_hdr(private_data, id_priv);
3140                 if (ret)
3141                         goto out;
3142                 req.private_data = private_data;
3143         }
3144
3145         id = ib_create_cm_id(id_priv->id.device, cma_sidr_rep_handler,
3146                              id_priv);
3147         if (IS_ERR(id)) {
3148                 ret = PTR_ERR(id);
3149                 goto out;
3150         }
3151         id_priv->cm_id.ib = id;
3152
3153         req.path = id_priv->id.route.path_rec;
3154         req.service_id = rdma_get_service_id(&id_priv->id, cma_dst_addr(id_priv));
3155         req.timeout_ms = 1 << (CMA_CM_RESPONSE_TIMEOUT - 8);
3156         req.max_cm_retries = CMA_MAX_CM_RETRIES;
3157
3158         ret = ib_send_cm_sidr_req(id_priv->cm_id.ib, &req);
3159         if (ret) {
3160                 ib_destroy_cm_id(id_priv->cm_id.ib);
3161                 id_priv->cm_id.ib = NULL;
3162         }
3163 out:
3164         kfree(private_data);
3165         return ret;
3166 }
3167
3168 static int cma_connect_ib(struct rdma_id_private *id_priv,
3169                           struct rdma_conn_param *conn_param)
3170 {
3171         struct ib_cm_req_param req;
3172         struct rdma_route *route;
3173         void *private_data;
3174         struct ib_cm_id *id;
3175         int offset, ret;
3176
3177         memset(&req, 0, sizeof req);
3178         offset = cma_user_data_offset(id_priv);
3179         req.private_data_len = offset + conn_param->private_data_len;
3180         if (req.private_data_len < conn_param->private_data_len)
3181                 return -EINVAL;
3182
3183         if (req.private_data_len) {
3184                 private_data = kzalloc(req.private_data_len, GFP_ATOMIC);
3185                 if (!private_data)
3186                         return -ENOMEM;
3187         } else {
3188                 private_data = NULL;
3189         }
3190
3191         if (conn_param->private_data && conn_param->private_data_len)
3192                 memcpy(private_data + offset, conn_param->private_data,
3193                        conn_param->private_data_len);
3194
3195         id = ib_create_cm_id(id_priv->id.device, cma_ib_handler, id_priv);
3196         if (IS_ERR(id)) {
3197                 ret = PTR_ERR(id);
3198                 goto out;
3199         }
3200         id_priv->cm_id.ib = id;
3201
3202         route = &id_priv->id.route;
3203         if (private_data) {
3204                 ret = cma_format_hdr(private_data, id_priv);
3205                 if (ret)
3206                         goto out;
3207                 req.private_data = private_data;
3208         }
3209
3210         req.primary_path = &route->path_rec[0];
3211         if (route->num_paths == 2)
3212                 req.alternate_path = &route->path_rec[1];
3213
3214         req.service_id = rdma_get_service_id(&id_priv->id, cma_dst_addr(id_priv));
3215         req.qp_num = id_priv->qp_num;
3216         req.qp_type = id_priv->id.qp_type;
3217         req.starting_psn = id_priv->seq_num;
3218         req.responder_resources = conn_param->responder_resources;
3219         req.initiator_depth = conn_param->initiator_depth;
3220         req.flow_control = conn_param->flow_control;
3221         req.retry_count = min_t(u8, 7, conn_param->retry_count);
3222         req.rnr_retry_count = min_t(u8, 7, conn_param->rnr_retry_count);
3223         req.remote_cm_response_timeout = CMA_CM_RESPONSE_TIMEOUT;
3224         req.local_cm_response_timeout = CMA_CM_RESPONSE_TIMEOUT;
3225         req.max_cm_retries = CMA_MAX_CM_RETRIES;
3226         req.srq = id_priv->srq ? 1 : 0;
3227
3228         ret = ib_send_cm_req(id_priv->cm_id.ib, &req);
3229 out:
3230         if (ret && !IS_ERR(id)) {
3231                 ib_destroy_cm_id(id);
3232                 id_priv->cm_id.ib = NULL;
3233         }
3234
3235         kfree(private_data);
3236         return ret;
3237 }
3238
3239 static int cma_connect_iw(struct rdma_id_private *id_priv,
3240                           struct rdma_conn_param *conn_param)
3241 {
3242         struct iw_cm_id *cm_id;
3243         int ret;
3244         struct iw_cm_conn_param iw_param;
3245
3246         cm_id = iw_create_cm_id(id_priv->id.device, cma_iw_handler, id_priv);
3247         if (IS_ERR(cm_id))
3248                 return PTR_ERR(cm_id);
3249
3250         cm_id->tos = id_priv->tos;
3251         id_priv->cm_id.iw = cm_id;
3252
3253         memcpy(&cm_id->local_addr, cma_src_addr(id_priv),
3254                rdma_addr_size(cma_src_addr(id_priv)));
3255         memcpy(&cm_id->remote_addr, cma_dst_addr(id_priv),
3256                rdma_addr_size(cma_dst_addr(id_priv)));
3257
3258         ret = cma_modify_qp_rtr(id_priv, conn_param);
3259         if (ret)
3260                 goto out;
3261
3262         if (conn_param) {
3263                 iw_param.ord = conn_param->initiator_depth;
3264                 iw_param.ird = conn_param->responder_resources;
3265                 iw_param.private_data = conn_param->private_data;
3266                 iw_param.private_data_len = conn_param->private_data_len;
3267                 iw_param.qpn = id_priv->id.qp ? id_priv->qp_num : conn_param->qp_num;
3268         } else {
3269                 memset(&iw_param, 0, sizeof iw_param);
3270                 iw_param.qpn = id_priv->qp_num;
3271         }
3272         ret = iw_cm_connect(cm_id, &iw_param);
3273 out:
3274         if (ret) {
3275                 iw_destroy_cm_id(cm_id);
3276                 id_priv->cm_id.iw = NULL;
3277         }
3278         return ret;
3279 }
3280
3281 int rdma_connect(struct rdma_cm_id *id, struct rdma_conn_param *conn_param)
3282 {
3283         struct rdma_id_private *id_priv;
3284         int ret;
3285
3286         id_priv = container_of(id, struct rdma_id_private, id);
3287         if (!cma_comp_exch(id_priv, RDMA_CM_ROUTE_RESOLVED, RDMA_CM_CONNECT))
3288                 return -EINVAL;
3289
3290         if (!id->qp) {
3291                 id_priv->qp_num = conn_param->qp_num;
3292                 id_priv->srq = conn_param->srq;
3293         }
3294
3295         if (rdma_cap_ib_cm(id->device, id->port_num)) {
3296                 if (id->qp_type == IB_QPT_UD)
3297                         ret = cma_resolve_ib_udp(id_priv, conn_param);
3298                 else
3299                         ret = cma_connect_ib(id_priv, conn_param);
3300         } else if (rdma_cap_iw_cm(id->device, id->port_num))
3301                 ret = cma_connect_iw(id_priv, conn_param);
3302         else
3303                 ret = -ENOSYS;
3304         if (ret)
3305                 goto err;
3306
3307         return 0;
3308 err:
3309         cma_comp_exch(id_priv, RDMA_CM_CONNECT, RDMA_CM_ROUTE_RESOLVED);
3310         return ret;
3311 }
3312 EXPORT_SYMBOL(rdma_connect);
3313
3314 static int cma_accept_ib(struct rdma_id_private *id_priv,
3315                          struct rdma_conn_param *conn_param)
3316 {
3317         struct ib_cm_rep_param rep;
3318         int ret;
3319
3320         ret = cma_modify_qp_rtr(id_priv, conn_param);
3321         if (ret)
3322                 goto out;
3323
3324         ret = cma_modify_qp_rts(id_priv, conn_param);
3325         if (ret)
3326                 goto out;
3327
3328         memset(&rep, 0, sizeof rep);
3329         rep.qp_num = id_priv->qp_num;
3330         rep.starting_psn = id_priv->seq_num;
3331         rep.private_data = conn_param->private_data;
3332         rep.private_data_len = conn_param->private_data_len;
3333         rep.responder_resources = conn_param->responder_resources;
3334         rep.initiator_depth = conn_param->initiator_depth;
3335         rep.failover_accepted = 0;
3336         rep.flow_control = conn_param->flow_control;
3337         rep.rnr_retry_count = min_t(u8, 7, conn_param->rnr_retry_count);
3338         rep.srq = id_priv->srq ? 1 : 0;
3339
3340         ret = ib_send_cm_rep(id_priv->cm_id.ib, &rep);
3341 out:
3342         return ret;
3343 }
3344
3345 static int cma_accept_iw(struct rdma_id_private *id_priv,
3346                   struct rdma_conn_param *conn_param)
3347 {
3348         struct iw_cm_conn_param iw_param;
3349         int ret;
3350
3351         ret = cma_modify_qp_rtr(id_priv, conn_param);
3352         if (ret)
3353                 return ret;
3354
3355         iw_param.ord = conn_param->initiator_depth;
3356         iw_param.ird = conn_param->responder_resources;
3357         iw_param.private_data = conn_param->private_data;
3358         iw_param.private_data_len = conn_param->private_data_len;
3359         if (id_priv->id.qp) {
3360                 iw_param.qpn = id_priv->qp_num;
3361         } else
3362                 iw_param.qpn = conn_param->qp_num;
3363
3364         return iw_cm_accept(id_priv->cm_id.iw, &iw_param);
3365 }
3366
3367 static int cma_send_sidr_rep(struct rdma_id_private *id_priv,
3368                              enum ib_cm_sidr_status status, u32 qkey,
3369                              const void *private_data, int private_data_len)
3370 {
3371         struct ib_cm_sidr_rep_param rep;
3372         int ret;
3373
3374         memset(&rep, 0, sizeof rep);
3375         rep.status = status;
3376         if (status == IB_SIDR_SUCCESS) {
3377                 ret = cma_set_qkey(id_priv, qkey);
3378                 if (ret)
3379                         return ret;
3380                 rep.qp_num = id_priv->qp_num;
3381                 rep.qkey = id_priv->qkey;
3382         }
3383         rep.private_data = private_data;
3384         rep.private_data_len = private_data_len;
3385
3386         return ib_send_cm_sidr_rep(id_priv->cm_id.ib, &rep);
3387 }
3388
3389 int rdma_accept(struct rdma_cm_id *id, struct rdma_conn_param *conn_param)
3390 {
3391         struct rdma_id_private *id_priv;
3392         int ret;
3393
3394         id_priv = container_of(id, struct rdma_id_private, id);
3395
3396         id_priv->owner = task_pid_nr(current);
3397
3398         if (!cma_comp(id_priv, RDMA_CM_CONNECT))
3399                 return -EINVAL;
3400
3401         if (!id->qp && conn_param) {
3402                 id_priv->qp_num = conn_param->qp_num;
3403                 id_priv->srq = conn_param->srq;
3404         }
3405
3406         if (rdma_cap_ib_cm(id->device, id->port_num)) {
3407                 if (id->qp_type == IB_QPT_UD) {
3408                         if (conn_param)
3409                                 ret = cma_send_sidr_rep(id_priv, IB_SIDR_SUCCESS,
3410                                                         conn_param->qkey,
3411                                                         conn_param->private_data,
3412                                                         conn_param->private_data_len);
3413                         else
3414                                 ret = cma_send_sidr_rep(id_priv, IB_SIDR_SUCCESS,
3415                                                         0, NULL, 0);
3416                 } else {
3417                         if (conn_param)
3418                                 ret = cma_accept_ib(id_priv, conn_param);
3419                         else
3420                                 ret = cma_rep_recv(id_priv);
3421                 }
3422         } else if (rdma_cap_iw_cm(id->device, id->port_num))
3423                 ret = cma_accept_iw(id_priv, conn_param);
3424         else
3425                 ret = -ENOSYS;
3426
3427         if (ret)
3428                 goto reject;
3429
3430         return 0;
3431 reject:
3432         cma_modify_qp_err(id_priv);
3433         rdma_reject(id, NULL, 0);
3434         return ret;
3435 }
3436 EXPORT_SYMBOL(rdma_accept);
3437
3438 int rdma_notify(struct rdma_cm_id *id, enum ib_event_type event)
3439 {
3440         struct rdma_id_private *id_priv;
3441         int ret;
3442
3443         id_priv = container_of(id, struct rdma_id_private, id);
3444         if (!id_priv->cm_id.ib)
3445                 return -EINVAL;
3446
3447         switch (id->device->node_type) {
3448         case RDMA_NODE_IB_CA:
3449                 ret = ib_cm_notify(id_priv->cm_id.ib, event);
3450                 break;
3451         default:
3452                 ret = 0;
3453                 break;
3454         }
3455         return ret;
3456 }
3457 EXPORT_SYMBOL(rdma_notify);
3458
3459 int rdma_reject(struct rdma_cm_id *id, const void *private_data,
3460                 u8 private_data_len)
3461 {
3462         struct rdma_id_private *id_priv;
3463         int ret;
3464
3465         id_priv = container_of(id, struct rdma_id_private, id);
3466         if (!id_priv->cm_id.ib)
3467                 return -EINVAL;
3468
3469         if (rdma_cap_ib_cm(id->device, id->port_num)) {
3470                 if (id->qp_type == IB_QPT_UD)
3471                         ret = cma_send_sidr_rep(id_priv, IB_SIDR_REJECT, 0,
3472                                                 private_data, private_data_len);
3473                 else
3474                         ret = ib_send_cm_rej(id_priv->cm_id.ib,
3475                                              IB_CM_REJ_CONSUMER_DEFINED, NULL,
3476                                              0, private_data, private_data_len);
3477         } else if (rdma_cap_iw_cm(id->device, id->port_num)) {
3478                 ret = iw_cm_reject(id_priv->cm_id.iw,
3479                                    private_data, private_data_len);
3480         } else
3481                 ret = -ENOSYS;
3482
3483         return ret;
3484 }
3485 EXPORT_SYMBOL(rdma_reject);
3486
3487 int rdma_disconnect(struct rdma_cm_id *id)
3488 {
3489         struct rdma_id_private *id_priv;
3490         int ret;
3491
3492         id_priv = container_of(id, struct rdma_id_private, id);
3493         if (!id_priv->cm_id.ib)
3494                 return -EINVAL;
3495
3496         if (rdma_cap_ib_cm(id->device, id->port_num)) {
3497                 ret = cma_modify_qp_err(id_priv);
3498                 if (ret)
3499                         goto out;
3500                 /* Initiate or respond to a disconnect. */
3501                 if (ib_send_cm_dreq(id_priv->cm_id.ib, NULL, 0))
3502                         ib_send_cm_drep(id_priv->cm_id.ib, NULL, 0);
3503         } else if (rdma_cap_iw_cm(id->device, id->port_num)) {
3504                 ret = iw_cm_disconnect(id_priv->cm_id.iw, 0);
3505         } else
3506                 ret = -EINVAL;
3507
3508 out:
3509         return ret;
3510 }
3511 EXPORT_SYMBOL(rdma_disconnect);
3512
3513 static int cma_ib_mc_handler(int status, struct ib_sa_multicast *multicast)
3514 {
3515         struct rdma_id_private *id_priv;
3516         struct cma_multicast *mc = multicast->context;
3517         struct rdma_cm_event event;
3518         int ret;
3519
3520         id_priv = mc->id_priv;
3521         if (cma_disable_callback(id_priv, RDMA_CM_ADDR_BOUND) &&
3522             cma_disable_callback(id_priv, RDMA_CM_ADDR_RESOLVED))
3523                 return 0;
3524
3525         if (!status)
3526                 status = cma_set_qkey(id_priv, be32_to_cpu(multicast->rec.qkey));
3527         mutex_lock(&id_priv->qp_mutex);
3528         if (!status && id_priv->id.qp)
3529                 status = ib_attach_mcast(id_priv->id.qp, &multicast->rec.mgid,
3530                                          be16_to_cpu(multicast->rec.mlid));
3531         mutex_unlock(&id_priv->qp_mutex);
3532
3533         memset(&event, 0, sizeof event);
3534         event.status = status;
3535         event.param.ud.private_data = mc->context;
3536         if (!status) {
3537                 event.event = RDMA_CM_EVENT_MULTICAST_JOIN;
3538                 ib_init_ah_from_mcmember(id_priv->id.device,
3539                                          id_priv->id.port_num, &multicast->rec,
3540                                          &event.param.ud.ah_attr);
3541                 event.param.ud.qp_num = 0xFFFFFF;
3542                 event.param.ud.qkey = be32_to_cpu(multicast->rec.qkey);
3543         } else
3544                 event.event = RDMA_CM_EVENT_MULTICAST_ERROR;
3545
3546         ret = id_priv->id.event_handler(&id_priv->id, &event);
3547         if (ret) {
3548                 cma_exch(id_priv, RDMA_CM_DESTROYING);
3549                 mutex_unlock(&id_priv->handler_mutex);
3550                 rdma_destroy_id(&id_priv->id);
3551                 return 0;
3552         }
3553
3554         mutex_unlock(&id_priv->handler_mutex);
3555         return 0;
3556 }
3557
3558 static void cma_set_mgid(struct rdma_id_private *id_priv,
3559                          struct sockaddr *addr, union ib_gid *mgid)
3560 {
3561         unsigned char mc_map[MAX_ADDR_LEN];
3562         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
3563         struct sockaddr_in *sin = (struct sockaddr_in *) addr;
3564         struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *) addr;
3565
3566         if (cma_any_addr(addr)) {
3567                 memset(mgid, 0, sizeof *mgid);
3568         } else if ((addr->sa_family == AF_INET6) &&
3569                    ((be32_to_cpu(sin6->sin6_addr.s6_addr32[0]) & 0xFFF0FFFF) ==
3570                                                                  0xFF10A01B)) {
3571                 /* IPv6 address is an SA assigned MGID. */
3572                 memcpy(mgid, &sin6->sin6_addr, sizeof *mgid);
3573         } else if (addr->sa_family == AF_IB) {
3574                 memcpy(mgid, &((struct sockaddr_ib *) addr)->sib_addr, sizeof *mgid);
3575         } else if ((addr->sa_family == AF_INET6)) {
3576                 ipv6_ib_mc_map(&sin6->sin6_addr, dev_addr->broadcast, mc_map);
3577                 if (id_priv->id.ps == RDMA_PS_UDP)
3578                         mc_map[7] = 0x01;       /* Use RDMA CM signature */
3579                 *mgid = *(union ib_gid *) (mc_map + 4);
3580         } else {
3581                 ip_ib_mc_map(sin->sin_addr.s_addr, dev_addr->broadcast, mc_map);
3582                 if (id_priv->id.ps == RDMA_PS_UDP)
3583                         mc_map[7] = 0x01;       /* Use RDMA CM signature */
3584                 *mgid = *(union ib_gid *) (mc_map + 4);
3585         }
3586 }
3587
3588 static int cma_join_ib_multicast(struct rdma_id_private *id_priv,
3589                                  struct cma_multicast *mc)
3590 {
3591         struct ib_sa_mcmember_rec rec;
3592         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
3593         ib_sa_comp_mask comp_mask;
3594         int ret;
3595
3596         ib_addr_get_mgid(dev_addr, &rec.mgid);
3597         ret = ib_sa_get_mcmember_rec(id_priv->id.device, id_priv->id.port_num,
3598                                      &rec.mgid, &rec);
3599         if (ret)
3600                 return ret;
3601
3602         ret = cma_set_qkey(id_priv, 0);
3603         if (ret)
3604                 return ret;
3605
3606         cma_set_mgid(id_priv, (struct sockaddr *) &mc->addr, &rec.mgid);
3607         rec.qkey = cpu_to_be32(id_priv->qkey);
3608         rdma_addr_get_sgid(dev_addr, &rec.port_gid);
3609         rec.pkey = cpu_to_be16(ib_addr_get_pkey(dev_addr));
3610         rec.join_state = 1;
3611
3612         comp_mask = IB_SA_MCMEMBER_REC_MGID | IB_SA_MCMEMBER_REC_PORT_GID |
3613                     IB_SA_MCMEMBER_REC_PKEY | IB_SA_MCMEMBER_REC_JOIN_STATE |
3614                     IB_SA_MCMEMBER_REC_QKEY | IB_SA_MCMEMBER_REC_SL |
3615                     IB_SA_MCMEMBER_REC_FLOW_LABEL |
3616                     IB_SA_MCMEMBER_REC_TRAFFIC_CLASS;
3617
3618         if (id_priv->id.ps == RDMA_PS_IPOIB)
3619                 comp_mask |= IB_SA_MCMEMBER_REC_RATE |
3620                              IB_SA_MCMEMBER_REC_RATE_SELECTOR |
3621                              IB_SA_MCMEMBER_REC_MTU_SELECTOR |
3622                              IB_SA_MCMEMBER_REC_MTU |
3623                              IB_SA_MCMEMBER_REC_HOP_LIMIT;
3624
3625         mc->multicast.ib = ib_sa_join_multicast(&sa_client, id_priv->id.device,
3626                                                 id_priv->id.port_num, &rec,
3627                                                 comp_mask, GFP_KERNEL,
3628                                                 cma_ib_mc_handler, mc);
3629         return PTR_ERR_OR_ZERO(mc->multicast.ib);
3630 }
3631
3632 static void iboe_mcast_work_handler(struct work_struct *work)
3633 {
3634         struct iboe_mcast_work *mw = container_of(work, struct iboe_mcast_work, work);
3635         struct cma_multicast *mc = mw->mc;
3636         struct ib_sa_multicast *m = mc->multicast.ib;
3637
3638         mc->multicast.ib->context = mc;
3639         cma_ib_mc_handler(0, m);
3640         kref_put(&mc->mcref, release_mc);
3641         kfree(mw);
3642 }
3643
3644 static void cma_iboe_set_mgid(struct sockaddr *addr, union ib_gid *mgid)
3645 {
3646         struct sockaddr_in *sin = (struct sockaddr_in *)addr;
3647         struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)addr;
3648
3649         if (cma_any_addr(addr)) {
3650                 memset(mgid, 0, sizeof *mgid);
3651         } else if (addr->sa_family == AF_INET6) {
3652                 memcpy(mgid, &sin6->sin6_addr, sizeof *mgid);
3653         } else {
3654                 mgid->raw[0] = 0xff;
3655                 mgid->raw[1] = 0x0e;
3656                 mgid->raw[2] = 0;
3657                 mgid->raw[3] = 0;
3658                 mgid->raw[4] = 0;
3659                 mgid->raw[5] = 0;
3660                 mgid->raw[6] = 0;
3661                 mgid->raw[7] = 0;
3662                 mgid->raw[8] = 0;
3663                 mgid->raw[9] = 0;
3664                 mgid->raw[10] = 0xff;
3665                 mgid->raw[11] = 0xff;
3666                 *(__be32 *)(&mgid->raw[12]) = sin->sin_addr.s_addr;
3667         }
3668 }
3669
3670 static int cma_iboe_join_multicast(struct rdma_id_private *id_priv,
3671                                    struct cma_multicast *mc)
3672 {
3673         struct iboe_mcast_work *work;
3674         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
3675         int err;
3676         struct sockaddr *addr = (struct sockaddr *)&mc->addr;
3677         struct net_device *ndev = NULL;
3678
3679         if (cma_zero_addr((struct sockaddr *)&mc->addr))
3680                 return -EINVAL;
3681
3682         work = kzalloc(sizeof *work, GFP_KERNEL);
3683         if (!work)
3684                 return -ENOMEM;
3685
3686         mc->multicast.ib = kzalloc(sizeof(struct ib_sa_multicast), GFP_KERNEL);
3687         if (!mc->multicast.ib) {
3688                 err = -ENOMEM;
3689                 goto out1;
3690         }
3691
3692         cma_iboe_set_mgid(addr, &mc->multicast.ib->rec.mgid);
3693
3694         mc->multicast.ib->rec.pkey = cpu_to_be16(0xffff);
3695         if (id_priv->id.ps == RDMA_PS_UDP)
3696                 mc->multicast.ib->rec.qkey = cpu_to_be32(RDMA_UDP_QKEY);
3697
3698         if (dev_addr->bound_dev_if)
3699                 ndev = dev_get_by_index(&init_net, dev_addr->bound_dev_if);
3700         if (!ndev) {
3701                 err = -ENODEV;
3702                 goto out2;
3703         }
3704         mc->multicast.ib->rec.rate = iboe_get_rate(ndev);
3705         mc->multicast.ib->rec.hop_limit = 1;
3706         mc->multicast.ib->rec.mtu = iboe_get_mtu(ndev->mtu);
3707         dev_put(ndev);
3708         if (!mc->multicast.ib->rec.mtu) {
3709                 err = -EINVAL;
3710                 goto out2;
3711         }
3712         rdma_ip2gid((struct sockaddr *)&id_priv->id.route.addr.src_addr,
3713                     &mc->multicast.ib->rec.port_gid);
3714         work->id = id_priv;
3715         work->mc = mc;
3716         INIT_WORK(&work->work, iboe_mcast_work_handler);
3717         kref_get(&mc->mcref);
3718         queue_work(cma_wq, &work->work);
3719
3720         return 0;
3721
3722 out2:
3723         kfree(mc->multicast.ib);
3724 out1:
3725         kfree(work);
3726         return err;
3727 }
3728
3729 int rdma_join_multicast(struct rdma_cm_id *id, struct sockaddr *addr,
3730                         void *context)
3731 {
3732         struct rdma_id_private *id_priv;
3733         struct cma_multicast *mc;
3734         int ret;
3735
3736         id_priv = container_of(id, struct rdma_id_private, id);
3737         if (!cma_comp(id_priv, RDMA_CM_ADDR_BOUND) &&
3738             !cma_comp(id_priv, RDMA_CM_ADDR_RESOLVED))
3739                 return -EINVAL;
3740
3741         mc = kmalloc(sizeof *mc, GFP_KERNEL);
3742         if (!mc)
3743                 return -ENOMEM;
3744
3745         memcpy(&mc->addr, addr, rdma_addr_size(addr));
3746         mc->context = context;
3747         mc->id_priv = id_priv;
3748
3749         spin_lock(&id_priv->lock);
3750         list_add(&mc->list, &id_priv->mc_list);
3751         spin_unlock(&id_priv->lock);
3752
3753         if (rdma_protocol_roce(id->device, id->port_num)) {
3754                 kref_init(&mc->mcref);
3755                 ret = cma_iboe_join_multicast(id_priv, mc);
3756         } else if (rdma_cap_ib_mcast(id->device, id->port_num))
3757                 ret = cma_join_ib_multicast(id_priv, mc);
3758         else
3759                 ret = -ENOSYS;
3760
3761         if (ret) {
3762                 spin_lock_irq(&id_priv->lock);
3763                 list_del(&mc->list);
3764                 spin_unlock_irq(&id_priv->lock);
3765                 kfree(mc);
3766         }
3767         return ret;
3768 }
3769 EXPORT_SYMBOL(rdma_join_multicast);
3770
3771 void rdma_leave_multicast(struct rdma_cm_id *id, struct sockaddr *addr)
3772 {
3773         struct rdma_id_private *id_priv;
3774         struct cma_multicast *mc;
3775
3776         id_priv = container_of(id, struct rdma_id_private, id);
3777         spin_lock_irq(&id_priv->lock);
3778         list_for_each_entry(mc, &id_priv->mc_list, list) {
3779                 if (!memcmp(&mc->addr, addr, rdma_addr_size(addr))) {
3780                         list_del(&mc->list);
3781                         spin_unlock_irq(&id_priv->lock);
3782
3783                         if (id->qp)
3784                                 ib_detach_mcast(id->qp,
3785                                                 &mc->multicast.ib->rec.mgid,
3786                                                 be16_to_cpu(mc->multicast.ib->rec.mlid));
3787
3788                         BUG_ON(id_priv->cma_dev->device != id->device);
3789
3790                         if (rdma_cap_ib_mcast(id->device, id->port_num)) {
3791                                 ib_sa_free_multicast(mc->multicast.ib);
3792                                 kfree(mc);
3793                         } else if (rdma_protocol_roce(id->device, id->port_num))
3794                                 kref_put(&mc->mcref, release_mc);
3795
3796                         return;
3797                 }
3798         }
3799         spin_unlock_irq(&id_priv->lock);
3800 }
3801 EXPORT_SYMBOL(rdma_leave_multicast);
3802
3803 static int cma_netdev_change(struct net_device *ndev, struct rdma_id_private *id_priv)
3804 {
3805         struct rdma_dev_addr *dev_addr;
3806         struct cma_ndev_work *work;
3807
3808         dev_addr = &id_priv->id.route.addr.dev_addr;
3809
3810         if ((dev_addr->bound_dev_if == ndev->ifindex) &&
3811             (net_eq(dev_net(ndev), dev_addr->net)) &&
3812             memcmp(dev_addr->src_dev_addr, ndev->dev_addr, ndev->addr_len)) {
3813                 printk(KERN_INFO "RDMA CM addr change for ndev %s used by id %p\n",
3814                        ndev->name, &id_priv->id);
3815                 work = kzalloc(sizeof *work, GFP_KERNEL);
3816                 if (!work)
3817                         return -ENOMEM;
3818
3819                 INIT_WORK(&work->work, cma_ndev_work_handler);
3820                 work->id = id_priv;
3821                 work->event.event = RDMA_CM_EVENT_ADDR_CHANGE;
3822                 atomic_inc(&id_priv->refcount);
3823                 queue_work(cma_wq, &work->work);
3824         }
3825
3826         return 0;
3827 }
3828
3829 static int cma_netdev_callback(struct notifier_block *self, unsigned long event,
3830                                void *ptr)
3831 {
3832         struct net_device *ndev = netdev_notifier_info_to_dev(ptr);
3833         struct cma_device *cma_dev;
3834         struct rdma_id_private *id_priv;
3835         int ret = NOTIFY_DONE;
3836
3837         if (event != NETDEV_BONDING_FAILOVER)
3838                 return NOTIFY_DONE;
3839
3840         if (!(ndev->flags & IFF_MASTER) || !(ndev->priv_flags & IFF_BONDING))
3841                 return NOTIFY_DONE;
3842
3843         mutex_lock(&lock);
3844         list_for_each_entry(cma_dev, &dev_list, list)
3845                 list_for_each_entry(id_priv, &cma_dev->id_list, list) {
3846                         ret = cma_netdev_change(ndev, id_priv);
3847                         if (ret)
3848                                 goto out;
3849                 }
3850
3851 out:
3852         mutex_unlock(&lock);
3853         return ret;
3854 }
3855
3856 static struct notifier_block cma_nb = {
3857         .notifier_call = cma_netdev_callback
3858 };
3859
3860 static void cma_add_one(struct ib_device *device)
3861 {
3862         struct cma_device *cma_dev;
3863         struct rdma_id_private *id_priv;
3864
3865         cma_dev = kmalloc(sizeof *cma_dev, GFP_KERNEL);
3866         if (!cma_dev)
3867                 return;
3868
3869         cma_dev->device = device;
3870
3871         init_completion(&cma_dev->comp);
3872         atomic_set(&cma_dev->refcount, 1);
3873         INIT_LIST_HEAD(&cma_dev->id_list);
3874         ib_set_client_data(device, &cma_client, cma_dev);
3875
3876         mutex_lock(&lock);
3877         list_add_tail(&cma_dev->list, &dev_list);
3878         list_for_each_entry(id_priv, &listen_any_list, list)
3879                 cma_listen_on_dev(id_priv, cma_dev);
3880         mutex_unlock(&lock);
3881 }
3882
3883 static int cma_remove_id_dev(struct rdma_id_private *id_priv)
3884 {
3885         struct rdma_cm_event event;
3886         enum rdma_cm_state state;
3887         int ret = 0;
3888
3889         /* Record that we want to remove the device */
3890         state = cma_exch(id_priv, RDMA_CM_DEVICE_REMOVAL);
3891         if (state == RDMA_CM_DESTROYING)
3892                 return 0;
3893
3894         cma_cancel_operation(id_priv, state);
3895         mutex_lock(&id_priv->handler_mutex);
3896
3897         /* Check for destruction from another callback. */
3898         if (!cma_comp(id_priv, RDMA_CM_DEVICE_REMOVAL))
3899                 goto out;
3900
3901         memset(&event, 0, sizeof event);
3902         event.event = RDMA_CM_EVENT_DEVICE_REMOVAL;
3903         ret = id_priv->id.event_handler(&id_priv->id, &event);
3904 out:
3905         mutex_unlock(&id_priv->handler_mutex);
3906         return ret;
3907 }
3908
3909 static void cma_process_remove(struct cma_device *cma_dev)
3910 {
3911         struct rdma_id_private *id_priv;
3912         int ret;
3913
3914         mutex_lock(&lock);
3915         while (!list_empty(&cma_dev->id_list)) {
3916                 id_priv = list_entry(cma_dev->id_list.next,
3917                                      struct rdma_id_private, list);
3918
3919                 list_del(&id_priv->listen_list);
3920                 list_del_init(&id_priv->list);
3921                 atomic_inc(&id_priv->refcount);
3922                 mutex_unlock(&lock);
3923
3924                 ret = id_priv->internal_id ? 1 : cma_remove_id_dev(id_priv);
3925                 cma_deref_id(id_priv);
3926                 if (ret)
3927                         rdma_destroy_id(&id_priv->id);
3928
3929                 mutex_lock(&lock);
3930         }
3931         mutex_unlock(&lock);
3932
3933         cma_deref_dev(cma_dev);
3934         wait_for_completion(&cma_dev->comp);
3935 }
3936
3937 static void cma_remove_one(struct ib_device *device, void *client_data)
3938 {
3939         struct cma_device *cma_dev = client_data;
3940
3941         if (!cma_dev)
3942                 return;
3943
3944         mutex_lock(&lock);
3945         list_del(&cma_dev->list);
3946         mutex_unlock(&lock);
3947
3948         cma_process_remove(cma_dev);
3949         kfree(cma_dev);
3950 }
3951
3952 static int cma_get_id_stats(struct sk_buff *skb, struct netlink_callback *cb)
3953 {
3954         struct nlmsghdr *nlh;
3955         struct rdma_cm_id_stats *id_stats;
3956         struct rdma_id_private *id_priv;
3957         struct rdma_cm_id *id = NULL;
3958         struct cma_device *cma_dev;
3959         int i_dev = 0, i_id = 0;
3960
3961         /*
3962          * We export all of the IDs as a sequence of messages.  Each
3963          * ID gets its own netlink message.
3964          */
3965         mutex_lock(&lock);
3966
3967         list_for_each_entry(cma_dev, &dev_list, list) {
3968                 if (i_dev < cb->args[0]) {
3969                         i_dev++;
3970                         continue;
3971                 }
3972
3973                 i_id = 0;
3974                 list_for_each_entry(id_priv, &cma_dev->id_list, list) {
3975                         if (i_id < cb->args[1]) {
3976                                 i_id++;
3977                                 continue;
3978                         }
3979
3980                         id_stats = ibnl_put_msg(skb, &nlh, cb->nlh->nlmsg_seq,
3981                                                 sizeof *id_stats, RDMA_NL_RDMA_CM,
3982                                                 RDMA_NL_RDMA_CM_ID_STATS,
3983                                                 NLM_F_MULTI);
3984                         if (!id_stats)
3985                                 goto out;
3986
3987                         memset(id_stats, 0, sizeof *id_stats);
3988                         id = &id_priv->id;
3989                         id_stats->node_type = id->route.addr.dev_addr.dev_type;
3990                         id_stats->port_num = id->port_num;
3991                         id_stats->bound_dev_if =
3992                                 id->route.addr.dev_addr.bound_dev_if;
3993
3994                         if (ibnl_put_attr(skb, nlh,
3995                                           rdma_addr_size(cma_src_addr(id_priv)),
3996                                           cma_src_addr(id_priv),
3997                                           RDMA_NL_RDMA_CM_ATTR_SRC_ADDR))
3998                                 goto out;
3999                         if (ibnl_put_attr(skb, nlh,
4000                                           rdma_addr_size(cma_src_addr(id_priv)),
4001                                           cma_dst_addr(id_priv),
4002                                           RDMA_NL_RDMA_CM_ATTR_DST_ADDR))
4003                                 goto out;
4004
4005                         id_stats->pid           = id_priv->owner;
4006                         id_stats->port_space    = id->ps;
4007                         id_stats->cm_state      = id_priv->state;
4008                         id_stats->qp_num        = id_priv->qp_num;
4009                         id_stats->qp_type       = id->qp_type;
4010
4011                         i_id++;
4012                 }
4013
4014                 cb->args[1] = 0;
4015                 i_dev++;
4016         }
4017
4018 out:
4019         mutex_unlock(&lock);
4020         cb->args[0] = i_dev;
4021         cb->args[1] = i_id;
4022
4023         return skb->len;
4024 }
4025
4026 static const struct ibnl_client_cbs cma_cb_table[] = {
4027         [RDMA_NL_RDMA_CM_ID_STATS] = { .dump = cma_get_id_stats,
4028                                        .module = THIS_MODULE },
4029 };
4030
4031 static int cma_init_net(struct net *net)
4032 {
4033         struct cma_pernet *pernet = cma_pernet(net);
4034
4035         idr_init(&pernet->tcp_ps);
4036         idr_init(&pernet->udp_ps);
4037         idr_init(&pernet->ipoib_ps);
4038         idr_init(&pernet->ib_ps);
4039
4040         return 0;
4041 }
4042
4043 static void cma_exit_net(struct net *net)
4044 {
4045         struct cma_pernet *pernet = cma_pernet(net);
4046
4047         idr_destroy(&pernet->tcp_ps);
4048         idr_destroy(&pernet->udp_ps);
4049         idr_destroy(&pernet->ipoib_ps);
4050         idr_destroy(&pernet->ib_ps);
4051 }
4052
4053 static struct pernet_operations cma_pernet_operations = {
4054         .init = cma_init_net,
4055         .exit = cma_exit_net,
4056         .id = &cma_pernet_id,
4057         .size = sizeof(struct cma_pernet),
4058 };
4059
4060 static int __init cma_init(void)
4061 {
4062         int ret;
4063
4064         cma_wq = create_singlethread_workqueue("rdma_cm");
4065         if (!cma_wq)
4066                 return -ENOMEM;
4067
4068         ret = register_pernet_subsys(&cma_pernet_operations);
4069         if (ret)
4070                 goto err_wq;
4071
4072         ib_sa_register_client(&sa_client);
4073         rdma_addr_register_client(&addr_client);
4074         register_netdevice_notifier(&cma_nb);
4075
4076         ret = ib_register_client(&cma_client);
4077         if (ret)
4078                 goto err;
4079
4080         if (ibnl_add_client(RDMA_NL_RDMA_CM, RDMA_NL_RDMA_CM_NUM_OPS, cma_cb_table))
4081                 printk(KERN_WARNING "RDMA CMA: failed to add netlink callback\n");
4082
4083         return 0;
4084
4085 err:
4086         unregister_netdevice_notifier(&cma_nb);
4087         rdma_addr_unregister_client(&addr_client);
4088         ib_sa_unregister_client(&sa_client);
4089 err_wq:
4090         destroy_workqueue(cma_wq);
4091         return ret;
4092 }
4093
4094 static void __exit cma_cleanup(void)
4095 {
4096         ibnl_remove_client(RDMA_NL_RDMA_CM);
4097         ib_unregister_client(&cma_client);
4098         unregister_netdevice_notifier(&cma_nb);
4099         rdma_addr_unregister_client(&addr_client);
4100         ib_sa_unregister_client(&sa_client);
4101         unregister_pernet_subsys(&cma_pernet_operations);
4102         destroy_workqueue(cma_wq);
4103 }
4104
4105 module_init(cma_init);
4106 module_exit(cma_cleanup);