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)
 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_id_private *id_priv,
1269                               const struct net_device *net_dev)
1270 {
1271         const struct rdma_addr *addr = &id_priv->id.route.addr;
1272
1273         if (!net_dev)
1274                 /* This request is an AF_IB request or a RoCE request */
1275                 return addr->src_addr.ss_family == AF_IB ||
1276                        cma_protocol_roce(&id_priv->id);
1277
1278         return !addr->dev_addr.bound_dev_if ||
1279                (net_eq(dev_net(net_dev), addr->dev_addr.net) &&
1280                 addr->dev_addr.bound_dev_if == net_dev->ifindex);
1281 }
1282
1283 static struct rdma_id_private *cma_find_listener(
1284                 const struct rdma_bind_list *bind_list,
1285                 const struct ib_cm_id *cm_id,
1286                 const struct ib_cm_event *ib_event,
1287                 const struct cma_req_info *req,
1288                 const struct net_device *net_dev)
1289 {
1290         struct rdma_id_private *id_priv, *id_priv_dev;
1291
1292         if (!bind_list)
1293                 return ERR_PTR(-EINVAL);
1294
1295         hlist_for_each_entry(id_priv, &bind_list->owners, node) {
1296                 if (cma_match_private_data(id_priv, ib_event->private_data)) {
1297                         if (id_priv->id.device == cm_id->device &&
1298                             cma_match_net_dev(id_priv, net_dev))
1299                                 return id_priv;
1300                         list_for_each_entry(id_priv_dev,
1301                                             &id_priv->listen_list,
1302                                             listen_list) {
1303                                 if (id_priv_dev->id.device == cm_id->device &&
1304                                     cma_match_net_dev(id_priv_dev, net_dev))
1305                                         return id_priv_dev;
1306                         }
1307                 }
1308         }
1309
1310         return ERR_PTR(-EINVAL);
1311 }
1312
1313 static struct rdma_id_private *cma_id_from_event(struct ib_cm_id *cm_id,
1314                                                  struct ib_cm_event *ib_event,
1315                                                  struct net_device **net_dev)
1316 {
1317         struct cma_req_info req;
1318         struct rdma_bind_list *bind_list;
1319         struct rdma_id_private *id_priv;
1320         int err;
1321
1322         err = cma_save_req_info(ib_event, &req);
1323         if (err)
1324                 return ERR_PTR(err);
1325
1326         *net_dev = cma_get_net_dev(ib_event, &req);
1327         if (IS_ERR(*net_dev)) {
1328                 if (PTR_ERR(*net_dev) == -EAFNOSUPPORT) {
1329                         /* Assuming the protocol is AF_IB */
1330                         *net_dev = NULL;
1331                 } else if (cma_protocol_roce_dev_port(req.device, req.port)) {
1332                         /* TODO find the net dev matching the request parameters
1333                          * through the RoCE GID table */
1334                         *net_dev = NULL;
1335                 } else {
1336                         return ERR_CAST(*net_dev);
1337                 }
1338         }
1339
1340         bind_list = cma_ps_find(*net_dev ? dev_net(*net_dev) : &init_net,
1341                                 rdma_ps_from_service_id(req.service_id),
1342                                 cma_port_from_service_id(req.service_id));
1343         id_priv = cma_find_listener(bind_list, cm_id, ib_event, &req, *net_dev);
1344         if (IS_ERR(id_priv) && *net_dev) {
1345                 dev_put(*net_dev);
1346                 *net_dev = NULL;
1347         }
1348
1349         return id_priv;
1350 }
1351
1352 static inline int cma_user_data_offset(struct rdma_id_private *id_priv)
1353 {
1354         return cma_family(id_priv) == AF_IB ? 0 : sizeof(struct cma_hdr);
1355 }
1356
1357 static void cma_cancel_route(struct rdma_id_private *id_priv)
1358 {
1359         if (rdma_cap_ib_sa(id_priv->id.device, id_priv->id.port_num)) {
1360                 if (id_priv->query)
1361                         ib_sa_cancel_query(id_priv->query_id, id_priv->query);
1362         }
1363 }
1364
1365 static void cma_cancel_listens(struct rdma_id_private *id_priv)
1366 {
1367         struct rdma_id_private *dev_id_priv;
1368
1369         /*
1370          * Remove from listen_any_list to prevent added devices from spawning
1371          * additional listen requests.
1372          */
1373         mutex_lock(&lock);
1374         list_del(&id_priv->list);
1375
1376         while (!list_empty(&id_priv->listen_list)) {
1377                 dev_id_priv = list_entry(id_priv->listen_list.next,
1378                                          struct rdma_id_private, listen_list);
1379                 /* sync with device removal to avoid duplicate destruction */
1380                 list_del_init(&dev_id_priv->list);
1381                 list_del(&dev_id_priv->listen_list);
1382                 mutex_unlock(&lock);
1383
1384                 rdma_destroy_id(&dev_id_priv->id);
1385                 mutex_lock(&lock);
1386         }
1387         mutex_unlock(&lock);
1388 }
1389
1390 static void cma_cancel_operation(struct rdma_id_private *id_priv,
1391                                  enum rdma_cm_state state)
1392 {
1393         switch (state) {
1394         case RDMA_CM_ADDR_QUERY:
1395                 rdma_addr_cancel(&id_priv->id.route.addr.dev_addr);
1396                 break;
1397         case RDMA_CM_ROUTE_QUERY:
1398                 cma_cancel_route(id_priv);
1399                 break;
1400         case RDMA_CM_LISTEN:
1401                 if (cma_any_addr(cma_src_addr(id_priv)) && !id_priv->cma_dev)
1402                         cma_cancel_listens(id_priv);
1403                 break;
1404         default:
1405                 break;
1406         }
1407 }
1408
1409 static void cma_release_port(struct rdma_id_private *id_priv)
1410 {
1411         struct rdma_bind_list *bind_list = id_priv->bind_list;
1412         struct net *net = id_priv->id.route.addr.dev_addr.net;
1413
1414         if (!bind_list)
1415                 return;
1416
1417         mutex_lock(&lock);
1418         hlist_del(&id_priv->node);
1419         if (hlist_empty(&bind_list->owners)) {
1420                 cma_ps_remove(net, bind_list->ps, bind_list->port);
1421                 kfree(bind_list);
1422         }
1423         mutex_unlock(&lock);
1424 }
1425
1426 static void cma_leave_mc_groups(struct rdma_id_private *id_priv)
1427 {
1428         struct cma_multicast *mc;
1429
1430         while (!list_empty(&id_priv->mc_list)) {
1431                 mc = container_of(id_priv->mc_list.next,
1432                                   struct cma_multicast, list);
1433                 list_del(&mc->list);
1434                 if (rdma_cap_ib_mcast(id_priv->cma_dev->device,
1435                                       id_priv->id.port_num)) {
1436                         ib_sa_free_multicast(mc->multicast.ib);
1437                         kfree(mc);
1438                 } else
1439                         kref_put(&mc->mcref, release_mc);
1440         }
1441 }
1442
1443 void rdma_destroy_id(struct rdma_cm_id *id)
1444 {
1445         struct rdma_id_private *id_priv;
1446         enum rdma_cm_state state;
1447
1448         id_priv = container_of(id, struct rdma_id_private, id);
1449         state = cma_exch(id_priv, RDMA_CM_DESTROYING);
1450         cma_cancel_operation(id_priv, state);
1451
1452         /*
1453          * Wait for any active callback to finish.  New callbacks will find
1454          * the id_priv state set to destroying and abort.
1455          */
1456         mutex_lock(&id_priv->handler_mutex);
1457         mutex_unlock(&id_priv->handler_mutex);
1458
1459         if (id_priv->cma_dev) {
1460                 if (rdma_cap_ib_cm(id_priv->id.device, 1)) {
1461                         if (id_priv->cm_id.ib)
1462                                 ib_destroy_cm_id(id_priv->cm_id.ib);
1463                 } else if (rdma_cap_iw_cm(id_priv->id.device, 1)) {
1464                         if (id_priv->cm_id.iw)
1465                                 iw_destroy_cm_id(id_priv->cm_id.iw);
1466                 }
1467                 cma_leave_mc_groups(id_priv);
1468                 cma_release_dev(id_priv);
1469         }
1470
1471         cma_release_port(id_priv);
1472         cma_deref_id(id_priv);
1473         wait_for_completion(&id_priv->comp);
1474
1475         if (id_priv->internal_id)
1476                 cma_deref_id(id_priv->id.context);
1477
1478         kfree(id_priv->id.route.path_rec);
1479         put_net(id_priv->id.route.addr.dev_addr.net);
1480         kfree(id_priv);
1481 }
1482 EXPORT_SYMBOL(rdma_destroy_id);
1483
1484 static int cma_rep_recv(struct rdma_id_private *id_priv)
1485 {
1486         int ret;
1487
1488         ret = cma_modify_qp_rtr(id_priv, NULL);
1489         if (ret)
1490                 goto reject;
1491
1492         ret = cma_modify_qp_rts(id_priv, NULL);
1493         if (ret)
1494                 goto reject;
1495
1496         ret = ib_send_cm_rtu(id_priv->cm_id.ib, NULL, 0);
1497         if (ret)
1498                 goto reject;
1499
1500         return 0;
1501 reject:
1502         cma_modify_qp_err(id_priv);
1503         ib_send_cm_rej(id_priv->cm_id.ib, IB_CM_REJ_CONSUMER_DEFINED,
1504                        NULL, 0, NULL, 0);
1505         return ret;
1506 }
1507
1508 static void cma_set_rep_event_data(struct rdma_cm_event *event,
1509                                    struct ib_cm_rep_event_param *rep_data,
1510                                    void *private_data)
1511 {
1512         event->param.conn.private_data = private_data;
1513         event->param.conn.private_data_len = IB_CM_REP_PRIVATE_DATA_SIZE;
1514         event->param.conn.responder_resources = rep_data->responder_resources;
1515         event->param.conn.initiator_depth = rep_data->initiator_depth;
1516         event->param.conn.flow_control = rep_data->flow_control;
1517         event->param.conn.rnr_retry_count = rep_data->rnr_retry_count;
1518         event->param.conn.srq = rep_data->srq;
1519         event->param.conn.qp_num = rep_data->remote_qpn;
1520 }
1521
1522 static int cma_ib_handler(struct ib_cm_id *cm_id, struct ib_cm_event *ib_event)
1523 {
1524         struct rdma_id_private *id_priv = cm_id->context;
1525         struct rdma_cm_event event;
1526         int ret = 0;
1527
1528         if ((ib_event->event != IB_CM_TIMEWAIT_EXIT &&
1529                 cma_disable_callback(id_priv, RDMA_CM_CONNECT)) ||
1530             (ib_event->event == IB_CM_TIMEWAIT_EXIT &&
1531                 cma_disable_callback(id_priv, RDMA_CM_DISCONNECT)))
1532                 return 0;
1533
1534         memset(&event, 0, sizeof event);
1535         switch (ib_event->event) {
1536         case IB_CM_REQ_ERROR:
1537         case IB_CM_REP_ERROR:
1538                 event.event = RDMA_CM_EVENT_UNREACHABLE;
1539                 event.status = -ETIMEDOUT;
1540                 break;
1541         case IB_CM_REP_RECEIVED:
1542                 if (id_priv->id.qp) {
1543                         event.status = cma_rep_recv(id_priv);
1544                         event.event = event.status ? RDMA_CM_EVENT_CONNECT_ERROR :
1545                                                      RDMA_CM_EVENT_ESTABLISHED;
1546                 } else {
1547                         event.event = RDMA_CM_EVENT_CONNECT_RESPONSE;
1548                 }
1549                 cma_set_rep_event_data(&event, &ib_event->param.rep_rcvd,
1550                                        ib_event->private_data);
1551                 break;
1552         case IB_CM_RTU_RECEIVED:
1553         case IB_CM_USER_ESTABLISHED:
1554                 event.event = RDMA_CM_EVENT_ESTABLISHED;
1555                 break;
1556         case IB_CM_DREQ_ERROR:
1557                 event.status = -ETIMEDOUT; /* fall through */
1558         case IB_CM_DREQ_RECEIVED:
1559         case IB_CM_DREP_RECEIVED:
1560                 if (!cma_comp_exch(id_priv, RDMA_CM_CONNECT,
1561                                    RDMA_CM_DISCONNECT))
1562                         goto out;
1563                 event.event = RDMA_CM_EVENT_DISCONNECTED;
1564                 break;
1565         case IB_CM_TIMEWAIT_EXIT:
1566                 event.event = RDMA_CM_EVENT_TIMEWAIT_EXIT;
1567                 break;
1568         case IB_CM_MRA_RECEIVED:
1569                 /* ignore event */
1570                 goto out;
1571         case IB_CM_REJ_RECEIVED:
1572                 cma_modify_qp_err(id_priv);
1573                 event.status = ib_event->param.rej_rcvd.reason;
1574                 event.event = RDMA_CM_EVENT_REJECTED;
1575                 event.param.conn.private_data = ib_event->private_data;
1576                 event.param.conn.private_data_len = IB_CM_REJ_PRIVATE_DATA_SIZE;
1577                 break;
1578         default:
1579                 printk(KERN_ERR "RDMA CMA: unexpected IB CM event: %d\n",
1580                        ib_event->event);
1581                 goto out;
1582         }
1583
1584         ret = id_priv->id.event_handler(&id_priv->id, &event);
1585         if (ret) {
1586                 /* Destroy the CM ID by returning a non-zero value. */
1587                 id_priv->cm_id.ib = NULL;
1588                 cma_exch(id_priv, RDMA_CM_DESTROYING);
1589                 mutex_unlock(&id_priv->handler_mutex);
1590                 rdma_destroy_id(&id_priv->id);
1591                 return ret;
1592         }
1593 out:
1594         mutex_unlock(&id_priv->handler_mutex);
1595         return ret;
1596 }
1597
1598 static struct rdma_id_private *cma_new_conn_id(struct rdma_cm_id *listen_id,
1599                                                struct ib_cm_event *ib_event,
1600                                                struct net_device *net_dev)
1601 {
1602         struct rdma_id_private *id_priv;
1603         struct rdma_cm_id *id;
1604         struct rdma_route *rt;
1605         const sa_family_t ss_family = listen_id->route.addr.src_addr.ss_family;
1606         const __be64 service_id =
1607                       ib_event->param.req_rcvd.primary_path->service_id;
1608         int ret;
1609
1610         id = rdma_create_id(listen_id->route.addr.dev_addr.net,
1611                             listen_id->event_handler, listen_id->context,
1612                             listen_id->ps, ib_event->param.req_rcvd.qp_type);
1613         if (IS_ERR(id))
1614                 return NULL;
1615
1616         id_priv = container_of(id, struct rdma_id_private, id);
1617         if (cma_save_net_info((struct sockaddr *)&id->route.addr.src_addr,
1618                               (struct sockaddr *)&id->route.addr.dst_addr,
1619                               listen_id, ib_event, ss_family, service_id))
1620                 goto err;
1621
1622         rt = &id->route;
1623         rt->num_paths = ib_event->param.req_rcvd.alternate_path ? 2 : 1;
1624         rt->path_rec = kmalloc(sizeof *rt->path_rec * rt->num_paths,
1625                                GFP_KERNEL);
1626         if (!rt->path_rec)
1627                 goto err;
1628
1629         rt->path_rec[0] = *ib_event->param.req_rcvd.primary_path;
1630         if (rt->num_paths == 2)
1631                 rt->path_rec[1] = *ib_event->param.req_rcvd.alternate_path;
1632
1633         if (net_dev) {
1634                 ret = rdma_copy_addr(&rt->addr.dev_addr, net_dev, NULL);
1635                 if (ret)
1636                         goto err;
1637         } else {
1638                 if (!cma_protocol_roce(listen_id) &&
1639                     cma_any_addr(cma_src_addr(id_priv))) {
1640                         rt->addr.dev_addr.dev_type = ARPHRD_INFINIBAND;
1641                         rdma_addr_set_sgid(&rt->addr.dev_addr, &rt->path_rec[0].sgid);
1642                         ib_addr_set_pkey(&rt->addr.dev_addr, be16_to_cpu(rt->path_rec[0].pkey));
1643                 } else if (!cma_any_addr(cma_src_addr(id_priv))) {
1644                         ret = cma_translate_addr(cma_src_addr(id_priv), &rt->addr.dev_addr);
1645                         if (ret)
1646                                 goto err;
1647                 }
1648         }
1649         rdma_addr_set_dgid(&rt->addr.dev_addr, &rt->path_rec[0].dgid);
1650
1651         id_priv->state = RDMA_CM_CONNECT;
1652         return id_priv;
1653
1654 err:
1655         rdma_destroy_id(id);
1656         return NULL;
1657 }
1658
1659 static struct rdma_id_private *cma_new_udp_id(struct rdma_cm_id *listen_id,
1660                                               struct ib_cm_event *ib_event,
1661                                               struct net_device *net_dev)
1662 {
1663         struct rdma_id_private *id_priv;
1664         struct rdma_cm_id *id;
1665         const sa_family_t ss_family = listen_id->route.addr.src_addr.ss_family;
1666         struct net *net = listen_id->route.addr.dev_addr.net;
1667         int ret;
1668
1669         id = rdma_create_id(net, listen_id->event_handler, listen_id->context,
1670                             listen_id->ps, IB_QPT_UD);
1671         if (IS_ERR(id))
1672                 return NULL;
1673
1674         id_priv = container_of(id, struct rdma_id_private, id);
1675         if (cma_save_net_info((struct sockaddr *)&id->route.addr.src_addr,
1676                               (struct sockaddr *)&id->route.addr.dst_addr,
1677                               listen_id, ib_event, ss_family,
1678                               ib_event->param.sidr_req_rcvd.service_id))
1679                 goto err;
1680
1681         if (net_dev) {
1682                 ret = rdma_copy_addr(&id->route.addr.dev_addr, net_dev, NULL);
1683                 if (ret)
1684                         goto err;
1685         } else {
1686                 if (!cma_any_addr(cma_src_addr(id_priv))) {
1687                         ret = cma_translate_addr(cma_src_addr(id_priv),
1688                                                  &id->route.addr.dev_addr);
1689                         if (ret)
1690                                 goto err;
1691                 }
1692         }
1693
1694         id_priv->state = RDMA_CM_CONNECT;
1695         return id_priv;
1696 err:
1697         rdma_destroy_id(id);
1698         return NULL;
1699 }
1700
1701 static void cma_set_req_event_data(struct rdma_cm_event *event,
1702                                    struct ib_cm_req_event_param *req_data,
1703                                    void *private_data, int offset)
1704 {
1705         event->param.conn.private_data = private_data + offset;
1706         event->param.conn.private_data_len = IB_CM_REQ_PRIVATE_DATA_SIZE - offset;
1707         event->param.conn.responder_resources = req_data->responder_resources;
1708         event->param.conn.initiator_depth = req_data->initiator_depth;
1709         event->param.conn.flow_control = req_data->flow_control;
1710         event->param.conn.retry_count = req_data->retry_count;
1711         event->param.conn.rnr_retry_count = req_data->rnr_retry_count;
1712         event->param.conn.srq = req_data->srq;
1713         event->param.conn.qp_num = req_data->remote_qpn;
1714 }
1715
1716 static int cma_check_req_qp_type(struct rdma_cm_id *id, struct ib_cm_event *ib_event)
1717 {
1718         return (((ib_event->event == IB_CM_REQ_RECEIVED) &&
1719                  (ib_event->param.req_rcvd.qp_type == id->qp_type)) ||
1720                 ((ib_event->event == IB_CM_SIDR_REQ_RECEIVED) &&
1721                  (id->qp_type == IB_QPT_UD)) ||
1722                 (!id->qp_type));
1723 }
1724
1725 static int cma_req_handler(struct ib_cm_id *cm_id, struct ib_cm_event *ib_event)
1726 {
1727         struct rdma_id_private *listen_id, *conn_id;
1728         struct rdma_cm_event event;
1729         struct net_device *net_dev;
1730         int offset, ret;
1731
1732         listen_id = cma_id_from_event(cm_id, ib_event, &net_dev);
1733         if (IS_ERR(listen_id))
1734                 return PTR_ERR(listen_id);
1735
1736         if (!cma_check_req_qp_type(&listen_id->id, ib_event)) {
1737                 ret = -EINVAL;
1738                 goto net_dev_put;
1739         }
1740
1741         if (cma_disable_callback(listen_id, RDMA_CM_LISTEN)) {
1742                 ret = -ECONNABORTED;
1743                 goto net_dev_put;
1744         }
1745
1746         memset(&event, 0, sizeof event);
1747         offset = cma_user_data_offset(listen_id);
1748         event.event = RDMA_CM_EVENT_CONNECT_REQUEST;
1749         if (ib_event->event == IB_CM_SIDR_REQ_RECEIVED) {
1750                 conn_id = cma_new_udp_id(&listen_id->id, ib_event, net_dev);
1751                 event.param.ud.private_data = ib_event->private_data + offset;
1752                 event.param.ud.private_data_len =
1753                                 IB_CM_SIDR_REQ_PRIVATE_DATA_SIZE - offset;
1754         } else {
1755                 conn_id = cma_new_conn_id(&listen_id->id, ib_event, net_dev);
1756                 cma_set_req_event_data(&event, &ib_event->param.req_rcvd,
1757                                        ib_event->private_data, offset);
1758         }
1759         if (!conn_id) {
1760                 ret = -ENOMEM;
1761                 goto err1;
1762         }
1763
1764         mutex_lock_nested(&conn_id->handler_mutex, SINGLE_DEPTH_NESTING);
1765         ret = cma_acquire_dev(conn_id, listen_id);
1766         if (ret)
1767                 goto err2;
1768
1769         conn_id->cm_id.ib = cm_id;
1770         cm_id->context = conn_id;
1771         cm_id->cm_handler = cma_ib_handler;
1772
1773         /*
1774          * Protect against the user destroying conn_id from another thread
1775          * until we're done accessing it.
1776          */
1777         atomic_inc(&conn_id->refcount);
1778         ret = conn_id->id.event_handler(&conn_id->id, &event);
1779         if (ret)
1780                 goto err3;
1781         /*
1782          * Acquire mutex to prevent user executing rdma_destroy_id()
1783          * while we're accessing the cm_id.
1784          */
1785         mutex_lock(&lock);
1786         if (cma_comp(conn_id, RDMA_CM_CONNECT) &&
1787             (conn_id->id.qp_type != IB_QPT_UD))
1788                 ib_send_cm_mra(cm_id, CMA_CM_MRA_SETTING, NULL, 0);
1789         mutex_unlock(&lock);
1790         mutex_unlock(&conn_id->handler_mutex);
1791         mutex_unlock(&listen_id->handler_mutex);
1792         cma_deref_id(conn_id);
1793         if (net_dev)
1794                 dev_put(net_dev);
1795         return 0;
1796
1797 err3:
1798         cma_deref_id(conn_id);
1799         /* Destroy the CM ID by returning a non-zero value. */
1800         conn_id->cm_id.ib = NULL;
1801 err2:
1802         cma_exch(conn_id, RDMA_CM_DESTROYING);
1803         mutex_unlock(&conn_id->handler_mutex);
1804 err1:
1805         mutex_unlock(&listen_id->handler_mutex);
1806         if (conn_id)
1807                 rdma_destroy_id(&conn_id->id);
1808
1809 net_dev_put:
1810         if (net_dev)
1811                 dev_put(net_dev);
1812
1813         return ret;
1814 }
1815
1816 __be64 rdma_get_service_id(struct rdma_cm_id *id, struct sockaddr *addr)
1817 {
1818         if (addr->sa_family == AF_IB)
1819                 return ((struct sockaddr_ib *) addr)->sib_sid;
1820
1821         return cpu_to_be64(((u64)id->ps << 16) + be16_to_cpu(cma_port(addr)));
1822 }
1823 EXPORT_SYMBOL(rdma_get_service_id);
1824
1825 static int cma_iw_handler(struct iw_cm_id *iw_id, struct iw_cm_event *iw_event)
1826 {
1827         struct rdma_id_private *id_priv = iw_id->context;
1828         struct rdma_cm_event event;
1829         int ret = 0;
1830         struct sockaddr *laddr = (struct sockaddr *)&iw_event->local_addr;
1831         struct sockaddr *raddr = (struct sockaddr *)&iw_event->remote_addr;
1832
1833         if (cma_disable_callback(id_priv, RDMA_CM_CONNECT))
1834                 return 0;
1835
1836         memset(&event, 0, sizeof event);
1837         switch (iw_event->event) {
1838         case IW_CM_EVENT_CLOSE:
1839                 event.event = RDMA_CM_EVENT_DISCONNECTED;
1840                 break;
1841         case IW_CM_EVENT_CONNECT_REPLY:
1842                 memcpy(cma_src_addr(id_priv), laddr,
1843                        rdma_addr_size(laddr));
1844                 memcpy(cma_dst_addr(id_priv), raddr,
1845                        rdma_addr_size(raddr));
1846                 switch (iw_event->status) {
1847                 case 0:
1848                         event.event = RDMA_CM_EVENT_ESTABLISHED;
1849                         event.param.conn.initiator_depth = iw_event->ird;
1850                         event.param.conn.responder_resources = iw_event->ord;
1851                         break;
1852                 case -ECONNRESET:
1853                 case -ECONNREFUSED:
1854                         event.event = RDMA_CM_EVENT_REJECTED;
1855                         break;
1856                 case -ETIMEDOUT:
1857                         event.event = RDMA_CM_EVENT_UNREACHABLE;
1858                         break;
1859                 default:
1860                         event.event = RDMA_CM_EVENT_CONNECT_ERROR;
1861                         break;
1862                 }
1863                 break;
1864         case IW_CM_EVENT_ESTABLISHED:
1865                 event.event = RDMA_CM_EVENT_ESTABLISHED;
1866                 event.param.conn.initiator_depth = iw_event->ird;
1867                 event.param.conn.responder_resources = iw_event->ord;
1868                 break;
1869         default:
1870                 BUG_ON(1);
1871         }
1872
1873         event.status = iw_event->status;
1874         event.param.conn.private_data = iw_event->private_data;
1875         event.param.conn.private_data_len = iw_event->private_data_len;
1876         ret = id_priv->id.event_handler(&id_priv->id, &event);
1877         if (ret) {
1878                 /* Destroy the CM ID by returning a non-zero value. */
1879                 id_priv->cm_id.iw = NULL;
1880                 cma_exch(id_priv, RDMA_CM_DESTROYING);
1881                 mutex_unlock(&id_priv->handler_mutex);
1882                 rdma_destroy_id(&id_priv->id);
1883                 return ret;
1884         }
1885
1886         mutex_unlock(&id_priv->handler_mutex);
1887         return ret;
1888 }
1889
1890 static int iw_conn_req_handler(struct iw_cm_id *cm_id,
1891                                struct iw_cm_event *iw_event)
1892 {
1893         struct rdma_cm_id *new_cm_id;
1894         struct rdma_id_private *listen_id, *conn_id;
1895         struct rdma_cm_event event;
1896         int ret;
1897         struct ib_device_attr attr;
1898         struct sockaddr *laddr = (struct sockaddr *)&iw_event->local_addr;
1899         struct sockaddr *raddr = (struct sockaddr *)&iw_event->remote_addr;
1900
1901         listen_id = cm_id->context;
1902         if (cma_disable_callback(listen_id, RDMA_CM_LISTEN))
1903                 return -ECONNABORTED;
1904
1905         /* Create a new RDMA id for the new IW CM ID */
1906         new_cm_id = rdma_create_id(listen_id->id.route.addr.dev_addr.net,
1907                                    listen_id->id.event_handler,
1908                                    listen_id->id.context,
1909                                    RDMA_PS_TCP, IB_QPT_RC);
1910         if (IS_ERR(new_cm_id)) {
1911                 ret = -ENOMEM;
1912                 goto out;
1913         }
1914         conn_id = container_of(new_cm_id, struct rdma_id_private, id);
1915         mutex_lock_nested(&conn_id->handler_mutex, SINGLE_DEPTH_NESTING);
1916         conn_id->state = RDMA_CM_CONNECT;
1917
1918         ret = rdma_translate_ip(laddr, &conn_id->id.route.addr.dev_addr, NULL);
1919         if (ret) {
1920                 mutex_unlock(&conn_id->handler_mutex);
1921                 rdma_destroy_id(new_cm_id);
1922                 goto out;
1923         }
1924
1925         ret = cma_acquire_dev(conn_id, listen_id);
1926         if (ret) {
1927                 mutex_unlock(&conn_id->handler_mutex);
1928                 rdma_destroy_id(new_cm_id);
1929                 goto out;
1930         }
1931
1932         conn_id->cm_id.iw = cm_id;
1933         cm_id->context = conn_id;
1934         cm_id->cm_handler = cma_iw_handler;
1935
1936         memcpy(cma_src_addr(conn_id), laddr, rdma_addr_size(laddr));
1937         memcpy(cma_dst_addr(conn_id), raddr, rdma_addr_size(raddr));
1938
1939         ret = ib_query_device(conn_id->id.device, &attr);
1940         if (ret) {
1941                 mutex_unlock(&conn_id->handler_mutex);
1942                 rdma_destroy_id(new_cm_id);
1943                 goto out;
1944         }
1945
1946         memset(&event, 0, sizeof event);
1947         event.event = RDMA_CM_EVENT_CONNECT_REQUEST;
1948         event.param.conn.private_data = iw_event->private_data;
1949         event.param.conn.private_data_len = iw_event->private_data_len;
1950         event.param.conn.initiator_depth = iw_event->ird;
1951         event.param.conn.responder_resources = iw_event->ord;
1952
1953         /*
1954          * Protect against the user destroying conn_id from another thread
1955          * until we're done accessing it.
1956          */
1957         atomic_inc(&conn_id->refcount);
1958         ret = conn_id->id.event_handler(&conn_id->id, &event);
1959         if (ret) {
1960                 /* User wants to destroy the CM ID */
1961                 conn_id->cm_id.iw = NULL;
1962                 cma_exch(conn_id, RDMA_CM_DESTROYING);
1963                 mutex_unlock(&conn_id->handler_mutex);
1964                 cma_deref_id(conn_id);
1965                 rdma_destroy_id(&conn_id->id);
1966                 goto out;
1967         }
1968
1969         mutex_unlock(&conn_id->handler_mutex);
1970         cma_deref_id(conn_id);
1971
1972 out:
1973         mutex_unlock(&listen_id->handler_mutex);
1974         return ret;
1975 }
1976
1977 static int cma_ib_listen(struct rdma_id_private *id_priv)
1978 {
1979         struct sockaddr *addr;
1980         struct ib_cm_id *id;
1981         __be64 svc_id;
1982
1983         addr = cma_src_addr(id_priv);
1984         svc_id = rdma_get_service_id(&id_priv->id, addr);
1985         id = ib_cm_insert_listen(id_priv->id.device, cma_req_handler, svc_id);
1986         if (IS_ERR(id))
1987                 return PTR_ERR(id);
1988         id_priv->cm_id.ib = id;
1989
1990         return 0;
1991 }
1992
1993 static int cma_iw_listen(struct rdma_id_private *id_priv, int backlog)
1994 {
1995         int ret;
1996         struct iw_cm_id *id;
1997
1998         id = iw_create_cm_id(id_priv->id.device,
1999                              iw_conn_req_handler,
2000                              id_priv);
2001         if (IS_ERR(id))
2002                 return PTR_ERR(id);
2003
2004         id->tos = id_priv->tos;
2005         id_priv->cm_id.iw = id;
2006
2007         memcpy(&id_priv->cm_id.iw->local_addr, cma_src_addr(id_priv),
2008                rdma_addr_size(cma_src_addr(id_priv)));
2009
2010         ret = iw_cm_listen(id_priv->cm_id.iw, backlog);
2011
2012         if (ret) {
2013                 iw_destroy_cm_id(id_priv->cm_id.iw);
2014                 id_priv->cm_id.iw = NULL;
2015         }
2016
2017         return ret;
2018 }
2019
2020 static int cma_listen_handler(struct rdma_cm_id *id,
2021                               struct rdma_cm_event *event)
2022 {
2023         struct rdma_id_private *id_priv = id->context;
2024
2025         id->context = id_priv->id.context;
2026         id->event_handler = id_priv->id.event_handler;
2027         return id_priv->id.event_handler(id, event);
2028 }
2029
2030 static void cma_listen_on_dev(struct rdma_id_private *id_priv,
2031                               struct cma_device *cma_dev)
2032 {
2033         struct rdma_id_private *dev_id_priv;
2034         struct rdma_cm_id *id;
2035         struct net *net = id_priv->id.route.addr.dev_addr.net;
2036         int ret;
2037
2038         if (cma_family(id_priv) == AF_IB && !rdma_cap_ib_cm(cma_dev->device, 1))
2039                 return;
2040
2041         id = rdma_create_id(net, cma_listen_handler, id_priv, id_priv->id.ps,
2042                             id_priv->id.qp_type);
2043         if (IS_ERR(id))
2044                 return;
2045
2046         dev_id_priv = container_of(id, struct rdma_id_private, id);
2047
2048         dev_id_priv->state = RDMA_CM_ADDR_BOUND;
2049         memcpy(cma_src_addr(dev_id_priv), cma_src_addr(id_priv),
2050                rdma_addr_size(cma_src_addr(id_priv)));
2051
2052         cma_attach_to_dev(dev_id_priv, cma_dev);
2053         list_add_tail(&dev_id_priv->listen_list, &id_priv->listen_list);
2054         atomic_inc(&id_priv->refcount);
2055         dev_id_priv->internal_id = 1;
2056         dev_id_priv->afonly = id_priv->afonly;
2057
2058         ret = rdma_listen(id, id_priv->backlog);
2059         if (ret)
2060                 printk(KERN_WARNING "RDMA CMA: cma_listen_on_dev, error %d, "
2061                        "listening on device %s\n", ret, cma_dev->device->name);
2062 }
2063
2064 static void cma_listen_on_all(struct rdma_id_private *id_priv)
2065 {
2066         struct cma_device *cma_dev;
2067
2068         mutex_lock(&lock);
2069         list_add_tail(&id_priv->list, &listen_any_list);
2070         list_for_each_entry(cma_dev, &dev_list, list)
2071                 cma_listen_on_dev(id_priv, cma_dev);
2072         mutex_unlock(&lock);
2073 }
2074
2075 void rdma_set_service_type(struct rdma_cm_id *id, int tos)
2076 {
2077         struct rdma_id_private *id_priv;
2078
2079         id_priv = container_of(id, struct rdma_id_private, id);
2080         id_priv->tos = (u8) tos;
2081 }
2082 EXPORT_SYMBOL(rdma_set_service_type);
2083
2084 static void cma_query_handler(int status, struct ib_sa_path_rec *path_rec,
2085                               void *context)
2086 {
2087         struct cma_work *work = context;
2088         struct rdma_route *route;
2089
2090         route = &work->id->id.route;
2091
2092         if (!status) {
2093                 route->num_paths = 1;
2094                 *route->path_rec = *path_rec;
2095         } else {
2096                 work->old_state = RDMA_CM_ROUTE_QUERY;
2097                 work->new_state = RDMA_CM_ADDR_RESOLVED;
2098                 work->event.event = RDMA_CM_EVENT_ROUTE_ERROR;
2099                 work->event.status = status;
2100         }
2101
2102         queue_work(cma_wq, &work->work);
2103 }
2104
2105 static int cma_query_ib_route(struct rdma_id_private *id_priv, int timeout_ms,
2106                               struct cma_work *work)
2107 {
2108         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
2109         struct ib_sa_path_rec path_rec;
2110         ib_sa_comp_mask comp_mask;
2111         struct sockaddr_in6 *sin6;
2112         struct sockaddr_ib *sib;
2113
2114         memset(&path_rec, 0, sizeof path_rec);
2115         rdma_addr_get_sgid(dev_addr, &path_rec.sgid);
2116         rdma_addr_get_dgid(dev_addr, &path_rec.dgid);
2117         path_rec.pkey = cpu_to_be16(ib_addr_get_pkey(dev_addr));
2118         path_rec.numb_path = 1;
2119         path_rec.reversible = 1;
2120         path_rec.service_id = rdma_get_service_id(&id_priv->id, cma_dst_addr(id_priv));
2121
2122         comp_mask = IB_SA_PATH_REC_DGID | IB_SA_PATH_REC_SGID |
2123                     IB_SA_PATH_REC_PKEY | IB_SA_PATH_REC_NUMB_PATH |
2124                     IB_SA_PATH_REC_REVERSIBLE | IB_SA_PATH_REC_SERVICE_ID;
2125
2126         switch (cma_family(id_priv)) {
2127         case AF_INET:
2128                 path_rec.qos_class = cpu_to_be16((u16) id_priv->tos);
2129                 comp_mask |= IB_SA_PATH_REC_QOS_CLASS;
2130                 break;
2131         case AF_INET6:
2132                 sin6 = (struct sockaddr_in6 *) cma_src_addr(id_priv);
2133                 path_rec.traffic_class = (u8) (be32_to_cpu(sin6->sin6_flowinfo) >> 20);
2134                 comp_mask |= IB_SA_PATH_REC_TRAFFIC_CLASS;
2135                 break;
2136         case AF_IB:
2137                 sib = (struct sockaddr_ib *) cma_src_addr(id_priv);
2138                 path_rec.traffic_class = (u8) (be32_to_cpu(sib->sib_flowinfo) >> 20);
2139                 comp_mask |= IB_SA_PATH_REC_TRAFFIC_CLASS;
2140                 break;
2141         }
2142
2143         id_priv->query_id = ib_sa_path_rec_get(&sa_client, id_priv->id.device,
2144                                                id_priv->id.port_num, &path_rec,
2145                                                comp_mask, timeout_ms,
2146                                                GFP_KERNEL, cma_query_handler,
2147                                                work, &id_priv->query);
2148
2149         return (id_priv->query_id < 0) ? id_priv->query_id : 0;
2150 }
2151
2152 static void cma_work_handler(struct work_struct *_work)
2153 {
2154         struct cma_work *work = container_of(_work, struct cma_work, work);
2155         struct rdma_id_private *id_priv = work->id;
2156         int destroy = 0;
2157
2158         mutex_lock(&id_priv->handler_mutex);
2159         if (!cma_comp_exch(id_priv, work->old_state, work->new_state))
2160                 goto out;
2161
2162         if (id_priv->id.event_handler(&id_priv->id, &work->event)) {
2163                 cma_exch(id_priv, RDMA_CM_DESTROYING);
2164                 destroy = 1;
2165         }
2166 out:
2167         mutex_unlock(&id_priv->handler_mutex);
2168         cma_deref_id(id_priv);
2169         if (destroy)
2170                 rdma_destroy_id(&id_priv->id);
2171         kfree(work);
2172 }
2173
2174 static void cma_ndev_work_handler(struct work_struct *_work)
2175 {
2176         struct cma_ndev_work *work = container_of(_work, struct cma_ndev_work, work);
2177         struct rdma_id_private *id_priv = work->id;
2178         int destroy = 0;
2179
2180         mutex_lock(&id_priv->handler_mutex);
2181         if (id_priv->state == RDMA_CM_DESTROYING ||
2182             id_priv->state == RDMA_CM_DEVICE_REMOVAL)
2183                 goto out;
2184
2185         if (id_priv->id.event_handler(&id_priv->id, &work->event)) {
2186                 cma_exch(id_priv, RDMA_CM_DESTROYING);
2187                 destroy = 1;
2188         }
2189
2190 out:
2191         mutex_unlock(&id_priv->handler_mutex);
2192         cma_deref_id(id_priv);
2193         if (destroy)
2194                 rdma_destroy_id(&id_priv->id);
2195         kfree(work);
2196 }
2197
2198 static int cma_resolve_ib_route(struct rdma_id_private *id_priv, int timeout_ms)
2199 {
2200         struct rdma_route *route = &id_priv->id.route;
2201         struct cma_work *work;
2202         int ret;
2203
2204         work = kzalloc(sizeof *work, GFP_KERNEL);
2205         if (!work)
2206                 return -ENOMEM;
2207
2208         work->id = id_priv;
2209         INIT_WORK(&work->work, cma_work_handler);
2210         work->old_state = RDMA_CM_ROUTE_QUERY;
2211         work->new_state = RDMA_CM_ROUTE_RESOLVED;
2212         work->event.event = RDMA_CM_EVENT_ROUTE_RESOLVED;
2213
2214         route->path_rec = kmalloc(sizeof *route->path_rec, GFP_KERNEL);
2215         if (!route->path_rec) {
2216                 ret = -ENOMEM;
2217                 goto err1;
2218         }
2219
2220         ret = cma_query_ib_route(id_priv, timeout_ms, work);
2221         if (ret)
2222                 goto err2;
2223
2224         return 0;
2225 err2:
2226         kfree(route->path_rec);
2227         route->path_rec = NULL;
2228 err1:
2229         kfree(work);
2230         return ret;
2231 }
2232
2233 int rdma_set_ib_paths(struct rdma_cm_id *id,
2234                       struct ib_sa_path_rec *path_rec, int num_paths)
2235 {
2236         struct rdma_id_private *id_priv;
2237         int ret;
2238
2239         id_priv = container_of(id, struct rdma_id_private, id);
2240         if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_RESOLVED,
2241                            RDMA_CM_ROUTE_RESOLVED))
2242                 return -EINVAL;
2243
2244         id->route.path_rec = kmemdup(path_rec, sizeof *path_rec * num_paths,
2245                                      GFP_KERNEL);
2246         if (!id->route.path_rec) {
2247                 ret = -ENOMEM;
2248                 goto err;
2249         }
2250
2251         id->route.num_paths = num_paths;
2252         return 0;
2253 err:
2254         cma_comp_exch(id_priv, RDMA_CM_ROUTE_RESOLVED, RDMA_CM_ADDR_RESOLVED);
2255         return ret;
2256 }
2257 EXPORT_SYMBOL(rdma_set_ib_paths);
2258
2259 static int cma_resolve_iw_route(struct rdma_id_private *id_priv, int timeout_ms)
2260 {
2261         struct cma_work *work;
2262
2263         work = kzalloc(sizeof *work, GFP_KERNEL);
2264         if (!work)
2265                 return -ENOMEM;
2266
2267         work->id = id_priv;
2268         INIT_WORK(&work->work, cma_work_handler);
2269         work->old_state = RDMA_CM_ROUTE_QUERY;
2270         work->new_state = RDMA_CM_ROUTE_RESOLVED;
2271         work->event.event = RDMA_CM_EVENT_ROUTE_RESOLVED;
2272         queue_work(cma_wq, &work->work);
2273         return 0;
2274 }
2275
2276 static int iboe_tos_to_sl(struct net_device *ndev, int tos)
2277 {
2278         int prio;
2279         struct net_device *dev;
2280
2281         prio = rt_tos2priority(tos);
2282         dev = ndev->priv_flags & IFF_802_1Q_VLAN ?
2283                 vlan_dev_real_dev(ndev) : ndev;
2284
2285         if (dev->num_tc)
2286                 return netdev_get_prio_tc_map(dev, prio);
2287
2288 #if IS_ENABLED(CONFIG_VLAN_8021Q)
2289         if (ndev->priv_flags & IFF_802_1Q_VLAN)
2290                 return (vlan_dev_get_egress_qos_mask(ndev, prio) &
2291                         VLAN_PRIO_MASK) >> VLAN_PRIO_SHIFT;
2292 #endif
2293         return 0;
2294 }
2295
2296 static int cma_resolve_iboe_route(struct rdma_id_private *id_priv)
2297 {
2298         struct rdma_route *route = &id_priv->id.route;
2299         struct rdma_addr *addr = &route->addr;
2300         struct cma_work *work;
2301         int ret;
2302         struct net_device *ndev = NULL;
2303
2304
2305         work = kzalloc(sizeof *work, GFP_KERNEL);
2306         if (!work)
2307                 return -ENOMEM;
2308
2309         work->id = id_priv;
2310         INIT_WORK(&work->work, cma_work_handler);
2311
2312         route->path_rec = kzalloc(sizeof *route->path_rec, GFP_KERNEL);
2313         if (!route->path_rec) {
2314                 ret = -ENOMEM;
2315                 goto err1;
2316         }
2317
2318         route->num_paths = 1;
2319
2320         if (addr->dev_addr.bound_dev_if) {
2321                 ndev = dev_get_by_index(&init_net, addr->dev_addr.bound_dev_if);
2322                 route->path_rec->net = &init_net;
2323                 route->path_rec->ifindex = addr->dev_addr.bound_dev_if;
2324         }
2325         if (!ndev) {
2326                 ret = -ENODEV;
2327                 goto err2;
2328         }
2329
2330         memcpy(route->path_rec->dmac, addr->dev_addr.dst_dev_addr, ETH_ALEN);
2331
2332         rdma_ip2gid((struct sockaddr *)&id_priv->id.route.addr.src_addr,
2333                     &route->path_rec->sgid);
2334         rdma_ip2gid((struct sockaddr *)&id_priv->id.route.addr.dst_addr,
2335                     &route->path_rec->dgid);
2336
2337         route->path_rec->hop_limit = 1;
2338         route->path_rec->reversible = 1;
2339         route->path_rec->pkey = cpu_to_be16(0xffff);
2340         route->path_rec->mtu_selector = IB_SA_EQ;
2341         route->path_rec->sl = iboe_tos_to_sl(ndev, id_priv->tos);
2342         route->path_rec->mtu = iboe_get_mtu(ndev->mtu);
2343         route->path_rec->rate_selector = IB_SA_EQ;
2344         route->path_rec->rate = iboe_get_rate(ndev);
2345         dev_put(ndev);
2346         route->path_rec->packet_life_time_selector = IB_SA_EQ;
2347         route->path_rec->packet_life_time = CMA_IBOE_PACKET_LIFETIME;
2348         if (!route->path_rec->mtu) {
2349                 ret = -EINVAL;
2350                 goto err2;
2351         }
2352
2353         work->old_state = RDMA_CM_ROUTE_QUERY;
2354         work->new_state = RDMA_CM_ROUTE_RESOLVED;
2355         work->event.event = RDMA_CM_EVENT_ROUTE_RESOLVED;
2356         work->event.status = 0;
2357
2358         queue_work(cma_wq, &work->work);
2359
2360         return 0;
2361
2362 err2:
2363         kfree(route->path_rec);
2364         route->path_rec = NULL;
2365 err1:
2366         kfree(work);
2367         return ret;
2368 }
2369
2370 int rdma_resolve_route(struct rdma_cm_id *id, int timeout_ms)
2371 {
2372         struct rdma_id_private *id_priv;
2373         int ret;
2374
2375         id_priv = container_of(id, struct rdma_id_private, id);
2376         if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_RESOLVED, RDMA_CM_ROUTE_QUERY))
2377                 return -EINVAL;
2378
2379         atomic_inc(&id_priv->refcount);
2380         if (rdma_cap_ib_sa(id->device, id->port_num))
2381                 ret = cma_resolve_ib_route(id_priv, timeout_ms);
2382         else if (rdma_protocol_roce(id->device, id->port_num))
2383                 ret = cma_resolve_iboe_route(id_priv);
2384         else if (rdma_protocol_iwarp(id->device, id->port_num))
2385                 ret = cma_resolve_iw_route(id_priv, timeout_ms);
2386         else
2387                 ret = -ENOSYS;
2388
2389         if (ret)
2390                 goto err;
2391
2392         return 0;
2393 err:
2394         cma_comp_exch(id_priv, RDMA_CM_ROUTE_QUERY, RDMA_CM_ADDR_RESOLVED);
2395         cma_deref_id(id_priv);
2396         return ret;
2397 }
2398 EXPORT_SYMBOL(rdma_resolve_route);
2399
2400 static void cma_set_loopback(struct sockaddr *addr)
2401 {
2402         switch (addr->sa_family) {
2403         case AF_INET:
2404                 ((struct sockaddr_in *) addr)->sin_addr.s_addr = htonl(INADDR_LOOPBACK);
2405                 break;
2406         case AF_INET6:
2407                 ipv6_addr_set(&((struct sockaddr_in6 *) addr)->sin6_addr,
2408                               0, 0, 0, htonl(1));
2409                 break;
2410         default:
2411                 ib_addr_set(&((struct sockaddr_ib *) addr)->sib_addr,
2412                             0, 0, 0, htonl(1));
2413                 break;
2414         }
2415 }
2416
2417 static int cma_bind_loopback(struct rdma_id_private *id_priv)
2418 {
2419         struct cma_device *cma_dev, *cur_dev;
2420         struct ib_port_attr port_attr;
2421         union ib_gid gid;
2422         u16 pkey;
2423         int ret;
2424         u8 p;
2425
2426         cma_dev = NULL;
2427         mutex_lock(&lock);
2428         list_for_each_entry(cur_dev, &dev_list, list) {
2429                 if (cma_family(id_priv) == AF_IB &&
2430                     !rdma_cap_ib_cm(cur_dev->device, 1))
2431                         continue;
2432
2433                 if (!cma_dev)
2434                         cma_dev = cur_dev;
2435
2436                 for (p = 1; p <= cur_dev->device->phys_port_cnt; ++p) {
2437                         if (!ib_query_port(cur_dev->device, p, &port_attr) &&
2438                             port_attr.state == IB_PORT_ACTIVE) {
2439                                 cma_dev = cur_dev;
2440                                 goto port_found;
2441                         }
2442                 }
2443         }
2444
2445         if (!cma_dev) {
2446                 ret = -ENODEV;
2447                 goto out;
2448         }
2449
2450         p = 1;
2451
2452 port_found:
2453         ret = ib_get_cached_gid(cma_dev->device, p, 0, &gid, NULL);
2454         if (ret)
2455                 goto out;
2456
2457         ret = ib_get_cached_pkey(cma_dev->device, p, 0, &pkey);
2458         if (ret)
2459                 goto out;
2460
2461         id_priv->id.route.addr.dev_addr.dev_type =
2462                 (rdma_protocol_ib(cma_dev->device, p)) ?
2463                 ARPHRD_INFINIBAND : ARPHRD_ETHER;
2464
2465         rdma_addr_set_sgid(&id_priv->id.route.addr.dev_addr, &gid);
2466         ib_addr_set_pkey(&id_priv->id.route.addr.dev_addr, pkey);
2467         id_priv->id.port_num = p;
2468         cma_attach_to_dev(id_priv, cma_dev);
2469         cma_set_loopback(cma_src_addr(id_priv));
2470 out:
2471         mutex_unlock(&lock);
2472         return ret;
2473 }
2474
2475 static void addr_handler(int status, struct sockaddr *src_addr,
2476                          struct rdma_dev_addr *dev_addr, void *context)
2477 {
2478         struct rdma_id_private *id_priv = context;
2479         struct rdma_cm_event event;
2480
2481         memset(&event, 0, sizeof event);
2482         mutex_lock(&id_priv->handler_mutex);
2483         if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_QUERY,
2484                            RDMA_CM_ADDR_RESOLVED))
2485                 goto out;
2486
2487         memcpy(cma_src_addr(id_priv), src_addr, rdma_addr_size(src_addr));
2488         if (!status && !id_priv->cma_dev)
2489                 status = cma_acquire_dev(id_priv, NULL);
2490
2491         if (status) {
2492                 if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_RESOLVED,
2493                                    RDMA_CM_ADDR_BOUND))
2494                         goto out;
2495                 event.event = RDMA_CM_EVENT_ADDR_ERROR;
2496                 event.status = status;
2497         } else
2498                 event.event = RDMA_CM_EVENT_ADDR_RESOLVED;
2499
2500         if (id_priv->id.event_handler(&id_priv->id, &event)) {
2501                 cma_exch(id_priv, RDMA_CM_DESTROYING);
2502                 mutex_unlock(&id_priv->handler_mutex);
2503                 cma_deref_id(id_priv);
2504                 rdma_destroy_id(&id_priv->id);
2505                 return;
2506         }
2507 out:
2508         mutex_unlock(&id_priv->handler_mutex);
2509         cma_deref_id(id_priv);
2510 }
2511
2512 static int cma_resolve_loopback(struct rdma_id_private *id_priv)
2513 {
2514         struct cma_work *work;
2515         union ib_gid gid;
2516         int ret;
2517
2518         work = kzalloc(sizeof *work, GFP_KERNEL);
2519         if (!work)
2520                 return -ENOMEM;
2521
2522         if (!id_priv->cma_dev) {
2523                 ret = cma_bind_loopback(id_priv);
2524                 if (ret)
2525                         goto err;
2526         }
2527
2528         rdma_addr_get_sgid(&id_priv->id.route.addr.dev_addr, &gid);
2529         rdma_addr_set_dgid(&id_priv->id.route.addr.dev_addr, &gid);
2530
2531         work->id = id_priv;
2532         INIT_WORK(&work->work, cma_work_handler);
2533         work->old_state = RDMA_CM_ADDR_QUERY;
2534         work->new_state = RDMA_CM_ADDR_RESOLVED;
2535         work->event.event = RDMA_CM_EVENT_ADDR_RESOLVED;
2536         queue_work(cma_wq, &work->work);
2537         return 0;
2538 err:
2539         kfree(work);
2540         return ret;
2541 }
2542
2543 static int cma_resolve_ib_addr(struct rdma_id_private *id_priv)
2544 {
2545         struct cma_work *work;
2546         int ret;
2547
2548         work = kzalloc(sizeof *work, GFP_KERNEL);
2549         if (!work)
2550                 return -ENOMEM;
2551
2552         if (!id_priv->cma_dev) {
2553                 ret = cma_resolve_ib_dev(id_priv);
2554                 if (ret)
2555                         goto err;
2556         }
2557
2558         rdma_addr_set_dgid(&id_priv->id.route.addr.dev_addr, (union ib_gid *)
2559                 &(((struct sockaddr_ib *) &id_priv->id.route.addr.dst_addr)->sib_addr));
2560
2561         work->id = id_priv;
2562         INIT_WORK(&work->work, cma_work_handler);
2563         work->old_state = RDMA_CM_ADDR_QUERY;
2564         work->new_state = RDMA_CM_ADDR_RESOLVED;
2565         work->event.event = RDMA_CM_EVENT_ADDR_RESOLVED;
2566         queue_work(cma_wq, &work->work);
2567         return 0;
2568 err:
2569         kfree(work);
2570         return ret;
2571 }
2572
2573 static int cma_bind_addr(struct rdma_cm_id *id, struct sockaddr *src_addr,
2574                          struct sockaddr *dst_addr)
2575 {
2576         if (!src_addr || !src_addr->sa_family) {
2577                 src_addr = (struct sockaddr *) &id->route.addr.src_addr;
2578                 src_addr->sa_family = dst_addr->sa_family;
2579                 if (dst_addr->sa_family == AF_INET6) {
2580                         struct sockaddr_in6 *src_addr6 = (struct sockaddr_in6 *) src_addr;
2581                         struct sockaddr_in6 *dst_addr6 = (struct sockaddr_in6 *) dst_addr;
2582                         src_addr6->sin6_scope_id = dst_addr6->sin6_scope_id;
2583                         if (ipv6_addr_type(&dst_addr6->sin6_addr) & IPV6_ADDR_LINKLOCAL)
2584                                 id->route.addr.dev_addr.bound_dev_if = dst_addr6->sin6_scope_id;
2585                 } else if (dst_addr->sa_family == AF_IB) {
2586                         ((struct sockaddr_ib *) src_addr)->sib_pkey =
2587                                 ((struct sockaddr_ib *) dst_addr)->sib_pkey;
2588                 }
2589         }
2590         return rdma_bind_addr(id, src_addr);
2591 }
2592
2593 int rdma_resolve_addr(struct rdma_cm_id *id, struct sockaddr *src_addr,
2594                       struct sockaddr *dst_addr, int timeout_ms)
2595 {
2596         struct rdma_id_private *id_priv;
2597         int ret;
2598
2599         id_priv = container_of(id, struct rdma_id_private, id);
2600         if (id_priv->state == RDMA_CM_IDLE) {
2601                 ret = cma_bind_addr(id, src_addr, dst_addr);
2602                 if (ret)
2603                         return ret;
2604         }
2605
2606         if (cma_family(id_priv) != dst_addr->sa_family)
2607                 return -EINVAL;
2608
2609         if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_BOUND, RDMA_CM_ADDR_QUERY))
2610                 return -EINVAL;
2611
2612         atomic_inc(&id_priv->refcount);
2613         memcpy(cma_dst_addr(id_priv), dst_addr, rdma_addr_size(dst_addr));
2614         if (cma_any_addr(dst_addr)) {
2615                 ret = cma_resolve_loopback(id_priv);
2616         } else {
2617                 if (dst_addr->sa_family == AF_IB) {
2618                         ret = cma_resolve_ib_addr(id_priv);
2619                 } else {
2620                         ret = rdma_resolve_ip(&addr_client, cma_src_addr(id_priv),
2621                                               dst_addr, &id->route.addr.dev_addr,
2622                                               timeout_ms, addr_handler, id_priv);
2623                 }
2624         }
2625         if (ret)
2626                 goto err;
2627
2628         return 0;
2629 err:
2630         cma_comp_exch(id_priv, RDMA_CM_ADDR_QUERY, RDMA_CM_ADDR_BOUND);
2631         cma_deref_id(id_priv);
2632         return ret;
2633 }
2634 EXPORT_SYMBOL(rdma_resolve_addr);
2635
2636 int rdma_set_reuseaddr(struct rdma_cm_id *id, int reuse)
2637 {
2638         struct rdma_id_private *id_priv;
2639         unsigned long flags;
2640         int ret;
2641
2642         id_priv = container_of(id, struct rdma_id_private, id);
2643         spin_lock_irqsave(&id_priv->lock, flags);
2644         if (reuse || id_priv->state == RDMA_CM_IDLE) {
2645                 id_priv->reuseaddr = reuse;
2646                 ret = 0;
2647         } else {
2648                 ret = -EINVAL;
2649         }
2650         spin_unlock_irqrestore(&id_priv->lock, flags);
2651         return ret;
2652 }
2653 EXPORT_SYMBOL(rdma_set_reuseaddr);
2654
2655 int rdma_set_afonly(struct rdma_cm_id *id, int afonly)
2656 {
2657         struct rdma_id_private *id_priv;
2658         unsigned long flags;
2659         int ret;
2660
2661         id_priv = container_of(id, struct rdma_id_private, id);
2662         spin_lock_irqsave(&id_priv->lock, flags);
2663         if (id_priv->state == RDMA_CM_IDLE || id_priv->state == RDMA_CM_ADDR_BOUND) {
2664                 id_priv->options |= (1 << CMA_OPTION_AFONLY);
2665                 id_priv->afonly = afonly;
2666                 ret = 0;
2667         } else {
2668                 ret = -EINVAL;
2669         }
2670         spin_unlock_irqrestore(&id_priv->lock, flags);
2671         return ret;
2672 }
2673 EXPORT_SYMBOL(rdma_set_afonly);
2674
2675 static void cma_bind_port(struct rdma_bind_list *bind_list,
2676                           struct rdma_id_private *id_priv)
2677 {
2678         struct sockaddr *addr;
2679         struct sockaddr_ib *sib;
2680         u64 sid, mask;
2681         __be16 port;
2682
2683         addr = cma_src_addr(id_priv);
2684         port = htons(bind_list->port);
2685
2686         switch (addr->sa_family) {
2687         case AF_INET:
2688                 ((struct sockaddr_in *) addr)->sin_port = port;
2689                 break;
2690         case AF_INET6:
2691                 ((struct sockaddr_in6 *) addr)->sin6_port = port;
2692                 break;
2693         case AF_IB:
2694                 sib = (struct sockaddr_ib *) addr;
2695                 sid = be64_to_cpu(sib->sib_sid);
2696                 mask = be64_to_cpu(sib->sib_sid_mask);
2697                 sib->sib_sid = cpu_to_be64((sid & mask) | (u64) ntohs(port));
2698                 sib->sib_sid_mask = cpu_to_be64(~0ULL);
2699                 break;
2700         }
2701         id_priv->bind_list = bind_list;
2702         hlist_add_head(&id_priv->node, &bind_list->owners);
2703 }
2704
2705 static int cma_alloc_port(enum rdma_port_space ps,
2706                           struct rdma_id_private *id_priv, unsigned short snum)
2707 {
2708         struct rdma_bind_list *bind_list;
2709         int ret;
2710
2711         bind_list = kzalloc(sizeof *bind_list, GFP_KERNEL);
2712         if (!bind_list)
2713                 return -ENOMEM;
2714
2715         ret = cma_ps_alloc(id_priv->id.route.addr.dev_addr.net, ps, bind_list,
2716                            snum);
2717         if (ret < 0)
2718                 goto err;
2719
2720         bind_list->ps = ps;
2721         bind_list->port = (unsigned short)ret;
2722         cma_bind_port(bind_list, id_priv);
2723         return 0;
2724 err:
2725         kfree(bind_list);
2726         return ret == -ENOSPC ? -EADDRNOTAVAIL : ret;
2727 }
2728
2729 static int cma_alloc_any_port(enum rdma_port_space ps,
2730                               struct rdma_id_private *id_priv)
2731 {
2732         static unsigned int last_used_port;
2733         int low, high, remaining;
2734         unsigned int rover;
2735         struct net *net = id_priv->id.route.addr.dev_addr.net;
2736
2737         inet_get_local_port_range(net, &low, &high);
2738         remaining = (high - low) + 1;
2739         rover = prandom_u32() % remaining + low;
2740 retry:
2741         if (last_used_port != rover &&
2742             !cma_ps_find(net, ps, (unsigned short)rover)) {
2743                 int ret = cma_alloc_port(ps, id_priv, rover);
2744                 /*
2745                  * Remember previously used port number in order to avoid
2746                  * re-using same port immediately after it is closed.
2747                  */
2748                 if (!ret)
2749                         last_used_port = rover;
2750                 if (ret != -EADDRNOTAVAIL)
2751                         return ret;
2752         }
2753         if (--remaining) {
2754                 rover++;
2755                 if ((rover < low) || (rover > high))
2756                         rover = low;
2757                 goto retry;
2758         }
2759         return -EADDRNOTAVAIL;
2760 }
2761
2762 /*
2763  * Check that the requested port is available.  This is called when trying to
2764  * bind to a specific port, or when trying to listen on a bound port.  In
2765  * the latter case, the provided id_priv may already be on the bind_list, but
2766  * we still need to check that it's okay to start listening.
2767  */
2768 static int cma_check_port(struct rdma_bind_list *bind_list,
2769                           struct rdma_id_private *id_priv, uint8_t reuseaddr)
2770 {
2771         struct rdma_id_private *cur_id;
2772         struct sockaddr *addr, *cur_addr;
2773
2774         addr = cma_src_addr(id_priv);
2775         hlist_for_each_entry(cur_id, &bind_list->owners, node) {
2776                 if (id_priv == cur_id)
2777                         continue;
2778
2779                 if ((cur_id->state != RDMA_CM_LISTEN) && reuseaddr &&
2780                     cur_id->reuseaddr)
2781                         continue;
2782
2783                 cur_addr = cma_src_addr(cur_id);
2784                 if (id_priv->afonly && cur_id->afonly &&
2785                     (addr->sa_family != cur_addr->sa_family))
2786                         continue;
2787
2788                 if (cma_any_addr(addr) || cma_any_addr(cur_addr))
2789                         return -EADDRNOTAVAIL;
2790
2791                 if (!cma_addr_cmp(addr, cur_addr))
2792                         return -EADDRINUSE;
2793         }
2794         return 0;
2795 }
2796
2797 static int cma_use_port(enum rdma_port_space ps,
2798                         struct rdma_id_private *id_priv)
2799 {
2800         struct rdma_bind_list *bind_list;
2801         unsigned short snum;
2802         int ret;
2803
2804         snum = ntohs(cma_port(cma_src_addr(id_priv)));
2805         if (snum < PROT_SOCK && !capable(CAP_NET_BIND_SERVICE))
2806                 return -EACCES;
2807
2808         bind_list = cma_ps_find(id_priv->id.route.addr.dev_addr.net, ps, snum);
2809         if (!bind_list) {
2810                 ret = cma_alloc_port(ps, id_priv, snum);
2811         } else {
2812                 ret = cma_check_port(bind_list, id_priv, id_priv->reuseaddr);
2813                 if (!ret)
2814                         cma_bind_port(bind_list, id_priv);
2815         }
2816         return ret;
2817 }
2818
2819 static int cma_bind_listen(struct rdma_id_private *id_priv)
2820 {
2821         struct rdma_bind_list *bind_list = id_priv->bind_list;
2822         int ret = 0;
2823
2824         mutex_lock(&lock);
2825         if (bind_list->owners.first->next)
2826                 ret = cma_check_port(bind_list, id_priv, 0);
2827         mutex_unlock(&lock);
2828         return ret;
2829 }
2830
2831 static enum rdma_port_space cma_select_inet_ps(
2832                 struct rdma_id_private *id_priv)
2833 {
2834         switch (id_priv->id.ps) {
2835         case RDMA_PS_TCP:
2836         case RDMA_PS_UDP:
2837         case RDMA_PS_IPOIB:
2838         case RDMA_PS_IB:
2839                 return id_priv->id.ps;
2840         default:
2841
2842                 return 0;
2843         }
2844 }
2845
2846 static enum rdma_port_space cma_select_ib_ps(struct rdma_id_private *id_priv)
2847 {
2848         enum rdma_port_space ps = 0;
2849         struct sockaddr_ib *sib;
2850         u64 sid_ps, mask, sid;
2851
2852         sib = (struct sockaddr_ib *) cma_src_addr(id_priv);
2853         mask = be64_to_cpu(sib->sib_sid_mask) & RDMA_IB_IP_PS_MASK;
2854         sid = be64_to_cpu(sib->sib_sid) & mask;
2855
2856         if ((id_priv->id.ps == RDMA_PS_IB) && (sid == (RDMA_IB_IP_PS_IB & mask))) {
2857                 sid_ps = RDMA_IB_IP_PS_IB;
2858                 ps = RDMA_PS_IB;
2859         } else if (((id_priv->id.ps == RDMA_PS_IB) || (id_priv->id.ps == RDMA_PS_TCP)) &&
2860                    (sid == (RDMA_IB_IP_PS_TCP & mask))) {
2861                 sid_ps = RDMA_IB_IP_PS_TCP;
2862                 ps = RDMA_PS_TCP;
2863         } else if (((id_priv->id.ps == RDMA_PS_IB) || (id_priv->id.ps == RDMA_PS_UDP)) &&
2864                    (sid == (RDMA_IB_IP_PS_UDP & mask))) {
2865                 sid_ps = RDMA_IB_IP_PS_UDP;
2866                 ps = RDMA_PS_UDP;
2867         }
2868
2869         if (ps) {
2870                 sib->sib_sid = cpu_to_be64(sid_ps | ntohs(cma_port((struct sockaddr *) sib)));
2871                 sib->sib_sid_mask = cpu_to_be64(RDMA_IB_IP_PS_MASK |
2872                                                 be64_to_cpu(sib->sib_sid_mask));
2873         }
2874         return ps;
2875 }
2876
2877 static int cma_get_port(struct rdma_id_private *id_priv)
2878 {
2879         enum rdma_port_space ps;
2880         int ret;
2881
2882         if (cma_family(id_priv) != AF_IB)
2883                 ps = cma_select_inet_ps(id_priv);
2884         else
2885                 ps = cma_select_ib_ps(id_priv);
2886         if (!ps)
2887                 return -EPROTONOSUPPORT;
2888
2889         mutex_lock(&lock);
2890         if (cma_any_port(cma_src_addr(id_priv)))
2891                 ret = cma_alloc_any_port(ps, id_priv);
2892         else
2893                 ret = cma_use_port(ps, id_priv);
2894         mutex_unlock(&lock);
2895
2896         return ret;
2897 }
2898
2899 static int cma_check_linklocal(struct rdma_dev_addr *dev_addr,
2900                                struct sockaddr *addr)
2901 {
2902 #if IS_ENABLED(CONFIG_IPV6)
2903         struct sockaddr_in6 *sin6;
2904
2905         if (addr->sa_family != AF_INET6)
2906                 return 0;
2907
2908         sin6 = (struct sockaddr_in6 *) addr;
2909
2910         if (!(ipv6_addr_type(&sin6->sin6_addr) & IPV6_ADDR_LINKLOCAL))
2911                 return 0;
2912
2913         if (!sin6->sin6_scope_id)
2914                         return -EINVAL;
2915
2916         dev_addr->bound_dev_if = sin6->sin6_scope_id;
2917 #endif
2918         return 0;
2919 }
2920
2921 int rdma_listen(struct rdma_cm_id *id, int backlog)
2922 {
2923         struct rdma_id_private *id_priv;
2924         int ret;
2925
2926         id_priv = container_of(id, struct rdma_id_private, id);
2927         if (id_priv->state == RDMA_CM_IDLE) {
2928                 id->route.addr.src_addr.ss_family = AF_INET;
2929                 ret = rdma_bind_addr(id, cma_src_addr(id_priv));
2930                 if (ret)
2931                         return ret;
2932         }
2933
2934         if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_BOUND, RDMA_CM_LISTEN))
2935                 return -EINVAL;
2936
2937         if (id_priv->reuseaddr) {
2938                 ret = cma_bind_listen(id_priv);
2939                 if (ret)
2940                         goto err;
2941         }
2942
2943         id_priv->backlog = backlog;
2944         if (id->device) {
2945                 if (rdma_cap_ib_cm(id->device, 1)) {
2946                         ret = cma_ib_listen(id_priv);
2947                         if (ret)
2948                                 goto err;
2949                 } else if (rdma_cap_iw_cm(id->device, 1)) {
2950                         ret = cma_iw_listen(id_priv, backlog);
2951                         if (ret)
2952                                 goto err;
2953                 } else {
2954                         ret = -ENOSYS;
2955                         goto err;
2956                 }
2957         } else
2958                 cma_listen_on_all(id_priv);
2959
2960         return 0;
2961 err:
2962         id_priv->backlog = 0;
2963         cma_comp_exch(id_priv, RDMA_CM_LISTEN, RDMA_CM_ADDR_BOUND);
2964         return ret;
2965 }
2966 EXPORT_SYMBOL(rdma_listen);
2967
2968 int rdma_bind_addr(struct rdma_cm_id *id, struct sockaddr *addr)
2969 {
2970         struct rdma_id_private *id_priv;
2971         int ret;
2972
2973         if (addr->sa_family != AF_INET && addr->sa_family != AF_INET6 &&
2974             addr->sa_family != AF_IB)
2975                 return -EAFNOSUPPORT;
2976
2977         id_priv = container_of(id, struct rdma_id_private, id);
2978         if (!cma_comp_exch(id_priv, RDMA_CM_IDLE, RDMA_CM_ADDR_BOUND))
2979                 return -EINVAL;
2980
2981         ret = cma_check_linklocal(&id->route.addr.dev_addr, addr);
2982         if (ret)
2983                 goto err1;
2984
2985         memcpy(cma_src_addr(id_priv), addr, rdma_addr_size(addr));
2986         if (!cma_any_addr(addr)) {
2987                 ret = cma_translate_addr(addr, &id->route.addr.dev_addr);
2988                 if (ret)
2989                         goto err1;
2990
2991                 ret = cma_acquire_dev(id_priv, NULL);
2992                 if (ret)
2993                         goto err1;
2994         }
2995
2996         if (!(id_priv->options & (1 << CMA_OPTION_AFONLY))) {
2997                 if (addr->sa_family == AF_INET)
2998                         id_priv->afonly = 1;
2999 #if IS_ENABLED(CONFIG_IPV6)
3000                 else if (addr->sa_family == AF_INET6) {
3001                         struct net *net = id_priv->id.route.addr.dev_addr.net;
3002
3003                         id_priv->afonly = net->ipv6.sysctl.bindv6only;
3004                 }
3005 #endif
3006         }
3007         ret = cma_get_port(id_priv);
3008         if (ret)
3009                 goto err2;
3010
3011         return 0;
3012 err2:
3013         if (id_priv->cma_dev)
3014                 cma_release_dev(id_priv);
3015 err1:
3016         cma_comp_exch(id_priv, RDMA_CM_ADDR_BOUND, RDMA_CM_IDLE);
3017         return ret;
3018 }
3019 EXPORT_SYMBOL(rdma_bind_addr);
3020
3021 static int cma_format_hdr(void *hdr, struct rdma_id_private *id_priv)
3022 {
3023         struct cma_hdr *cma_hdr;
3024
3025         cma_hdr = hdr;
3026         cma_hdr->cma_version = CMA_VERSION;
3027         if (cma_family(id_priv) == AF_INET) {
3028                 struct sockaddr_in *src4, *dst4;
3029
3030                 src4 = (struct sockaddr_in *) cma_src_addr(id_priv);
3031                 dst4 = (struct sockaddr_in *) cma_dst_addr(id_priv);
3032
3033                 cma_set_ip_ver(cma_hdr, 4);
3034                 cma_hdr->src_addr.ip4.addr = src4->sin_addr.s_addr;
3035                 cma_hdr->dst_addr.ip4.addr = dst4->sin_addr.s_addr;
3036                 cma_hdr->port = src4->sin_port;
3037         } else if (cma_family(id_priv) == AF_INET6) {
3038                 struct sockaddr_in6 *src6, *dst6;
3039
3040                 src6 = (struct sockaddr_in6 *) cma_src_addr(id_priv);
3041                 dst6 = (struct sockaddr_in6 *) cma_dst_addr(id_priv);
3042
3043                 cma_set_ip_ver(cma_hdr, 6);
3044                 cma_hdr->src_addr.ip6 = src6->sin6_addr;
3045                 cma_hdr->dst_addr.ip6 = dst6->sin6_addr;
3046                 cma_hdr->port = src6->sin6_port;
3047         }
3048         return 0;
3049 }
3050
3051 static int cma_sidr_rep_handler(struct ib_cm_id *cm_id,
3052                                 struct ib_cm_event *ib_event)
3053 {
3054         struct rdma_id_private *id_priv = cm_id->context;
3055         struct rdma_cm_event event;
3056         struct ib_cm_sidr_rep_event_param *rep = &ib_event->param.sidr_rep_rcvd;
3057         int ret = 0;
3058
3059         if (cma_disable_callback(id_priv, RDMA_CM_CONNECT))
3060                 return 0;
3061
3062         memset(&event, 0, sizeof event);
3063         switch (ib_event->event) {
3064         case IB_CM_SIDR_REQ_ERROR:
3065                 event.event = RDMA_CM_EVENT_UNREACHABLE;
3066                 event.status = -ETIMEDOUT;
3067                 break;
3068         case IB_CM_SIDR_REP_RECEIVED:
3069                 event.param.ud.private_data = ib_event->private_data;
3070                 event.param.ud.private_data_len = IB_CM_SIDR_REP_PRIVATE_DATA_SIZE;
3071                 if (rep->status != IB_SIDR_SUCCESS) {
3072                         event.event = RDMA_CM_EVENT_UNREACHABLE;
3073                         event.status = ib_event->param.sidr_rep_rcvd.status;
3074                         break;
3075                 }
3076                 ret = cma_set_qkey(id_priv, rep->qkey);
3077                 if (ret) {
3078                         event.event = RDMA_CM_EVENT_ADDR_ERROR;
3079                         event.status = ret;
3080                         break;
3081                 }
3082                 ib_init_ah_from_path(id_priv->id.device, id_priv->id.port_num,
3083                                      id_priv->id.route.path_rec,
3084                                      &event.param.ud.ah_attr);
3085                 event.param.ud.qp_num = rep->qpn;
3086                 event.param.ud.qkey = rep->qkey;
3087                 event.event = RDMA_CM_EVENT_ESTABLISHED;
3088                 event.status = 0;
3089                 break;
3090         default:
3091                 printk(KERN_ERR "RDMA CMA: unexpected IB CM event: %d\n",
3092                        ib_event->event);
3093                 goto out;
3094         }
3095
3096         ret = id_priv->id.event_handler(&id_priv->id, &event);
3097         if (ret) {
3098                 /* Destroy the CM ID by returning a non-zero value. */
3099                 id_priv->cm_id.ib = NULL;
3100                 cma_exch(id_priv, RDMA_CM_DESTROYING);
3101                 mutex_unlock(&id_priv->handler_mutex);
3102                 rdma_destroy_id(&id_priv->id);
3103                 return ret;
3104         }
3105 out:
3106         mutex_unlock(&id_priv->handler_mutex);
3107         return ret;
3108 }
3109
3110 static int cma_resolve_ib_udp(struct rdma_id_private *id_priv,
3111                               struct rdma_conn_param *conn_param)
3112 {
3113         struct ib_cm_sidr_req_param req;
3114         struct ib_cm_id *id;
3115         void *private_data;
3116         int offset, ret;
3117
3118         memset(&req, 0, sizeof req);
3119         offset = cma_user_data_offset(id_priv);
3120         req.private_data_len = offset + conn_param->private_data_len;
3121         if (req.private_data_len < conn_param->private_data_len)
3122                 return -EINVAL;
3123
3124         if (req.private_data_len) {
3125                 private_data = kzalloc(req.private_data_len, GFP_ATOMIC);
3126                 if (!private_data)
3127                         return -ENOMEM;
3128         } else {
3129                 private_data = NULL;
3130         }
3131
3132         if (conn_param->private_data && conn_param->private_data_len)
3133                 memcpy(private_data + offset, conn_param->private_data,
3134                        conn_param->private_data_len);
3135
3136         if (private_data) {
3137                 ret = cma_format_hdr(private_data, id_priv);
3138                 if (ret)
3139                         goto out;
3140                 req.private_data = private_data;
3141         }
3142
3143         id = ib_create_cm_id(id_priv->id.device, cma_sidr_rep_handler,
3144                              id_priv);
3145         if (IS_ERR(id)) {
3146                 ret = PTR_ERR(id);
3147                 goto out;
3148         }
3149         id_priv->cm_id.ib = id;
3150
3151         req.path = id_priv->id.route.path_rec;
3152         req.service_id = rdma_get_service_id(&id_priv->id, cma_dst_addr(id_priv));
3153         req.timeout_ms = 1 << (CMA_CM_RESPONSE_TIMEOUT - 8);
3154         req.max_cm_retries = CMA_MAX_CM_RETRIES;
3155
3156         ret = ib_send_cm_sidr_req(id_priv->cm_id.ib, &req);
3157         if (ret) {
3158                 ib_destroy_cm_id(id_priv->cm_id.ib);
3159                 id_priv->cm_id.ib = NULL;
3160         }
3161 out:
3162         kfree(private_data);
3163         return ret;
3164 }
3165
3166 static int cma_connect_ib(struct rdma_id_private *id_priv,
3167                           struct rdma_conn_param *conn_param)
3168 {
3169         struct ib_cm_req_param req;
3170         struct rdma_route *route;
3171         void *private_data;
3172         struct ib_cm_id *id;
3173         int offset, ret;
3174
3175         memset(&req, 0, sizeof req);
3176         offset = cma_user_data_offset(id_priv);
3177         req.private_data_len = offset + conn_param->private_data_len;
3178         if (req.private_data_len < conn_param->private_data_len)
3179                 return -EINVAL;
3180
3181         if (req.private_data_len) {
3182                 private_data = kzalloc(req.private_data_len, GFP_ATOMIC);
3183                 if (!private_data)
3184                         return -ENOMEM;
3185         } else {
3186                 private_data = NULL;
3187         }
3188
3189         if (conn_param->private_data && conn_param->private_data_len)
3190                 memcpy(private_data + offset, conn_param->private_data,
3191                        conn_param->private_data_len);
3192
3193         id = ib_create_cm_id(id_priv->id.device, cma_ib_handler, id_priv);
3194         if (IS_ERR(id)) {
3195                 ret = PTR_ERR(id);
3196                 goto out;
3197         }
3198         id_priv->cm_id.ib = id;
3199
3200         route = &id_priv->id.route;
3201         if (private_data) {
3202                 ret = cma_format_hdr(private_data, id_priv);
3203                 if (ret)
3204                         goto out;
3205                 req.private_data = private_data;
3206         }
3207
3208         req.primary_path = &route->path_rec[0];
3209         if (route->num_paths == 2)
3210                 req.alternate_path = &route->path_rec[1];
3211
3212         req.service_id = rdma_get_service_id(&id_priv->id, cma_dst_addr(id_priv));
3213         req.qp_num = id_priv->qp_num;
3214         req.qp_type = id_priv->id.qp_type;
3215         req.starting_psn = id_priv->seq_num;
3216         req.responder_resources = conn_param->responder_resources;
3217         req.initiator_depth = conn_param->initiator_depth;
3218         req.flow_control = conn_param->flow_control;
3219         req.retry_count = min_t(u8, 7, conn_param->retry_count);
3220         req.rnr_retry_count = min_t(u8, 7, conn_param->rnr_retry_count);
3221         req.remote_cm_response_timeout = CMA_CM_RESPONSE_TIMEOUT;
3222         req.local_cm_response_timeout = CMA_CM_RESPONSE_TIMEOUT;
3223         req.max_cm_retries = CMA_MAX_CM_RETRIES;
3224         req.srq = id_priv->srq ? 1 : 0;
3225
3226         ret = ib_send_cm_req(id_priv->cm_id.ib, &req);
3227 out:
3228         if (ret && !IS_ERR(id)) {
3229                 ib_destroy_cm_id(id);
3230                 id_priv->cm_id.ib = NULL;
3231         }
3232
3233         kfree(private_data);
3234         return ret;
3235 }
3236
3237 static int cma_connect_iw(struct rdma_id_private *id_priv,
3238                           struct rdma_conn_param *conn_param)
3239 {
3240         struct iw_cm_id *cm_id;
3241         int ret;
3242         struct iw_cm_conn_param iw_param;
3243
3244         cm_id = iw_create_cm_id(id_priv->id.device, cma_iw_handler, id_priv);
3245         if (IS_ERR(cm_id))
3246                 return PTR_ERR(cm_id);
3247
3248         cm_id->tos = id_priv->tos;
3249         id_priv->cm_id.iw = cm_id;
3250
3251         memcpy(&cm_id->local_addr, cma_src_addr(id_priv),
3252                rdma_addr_size(cma_src_addr(id_priv)));
3253         memcpy(&cm_id->remote_addr, cma_dst_addr(id_priv),
3254                rdma_addr_size(cma_dst_addr(id_priv)));
3255
3256         ret = cma_modify_qp_rtr(id_priv, conn_param);
3257         if (ret)
3258                 goto out;
3259
3260         if (conn_param) {
3261                 iw_param.ord = conn_param->initiator_depth;
3262                 iw_param.ird = conn_param->responder_resources;
3263                 iw_param.private_data = conn_param->private_data;
3264                 iw_param.private_data_len = conn_param->private_data_len;
3265                 iw_param.qpn = id_priv->id.qp ? id_priv->qp_num : conn_param->qp_num;
3266         } else {
3267                 memset(&iw_param, 0, sizeof iw_param);
3268                 iw_param.qpn = id_priv->qp_num;
3269         }
3270         ret = iw_cm_connect(cm_id, &iw_param);
3271 out:
3272         if (ret) {
3273                 iw_destroy_cm_id(cm_id);
3274                 id_priv->cm_id.iw = NULL;
3275         }
3276         return ret;
3277 }
3278
3279 int rdma_connect(struct rdma_cm_id *id, struct rdma_conn_param *conn_param)
3280 {
3281         struct rdma_id_private *id_priv;
3282         int ret;
3283
3284         id_priv = container_of(id, struct rdma_id_private, id);
3285         if (!cma_comp_exch(id_priv, RDMA_CM_ROUTE_RESOLVED, RDMA_CM_CONNECT))
3286                 return -EINVAL;
3287
3288         if (!id->qp) {
3289                 id_priv->qp_num = conn_param->qp_num;
3290                 id_priv->srq = conn_param->srq;
3291         }
3292
3293         if (rdma_cap_ib_cm(id->device, id->port_num)) {
3294                 if (id->qp_type == IB_QPT_UD)
3295                         ret = cma_resolve_ib_udp(id_priv, conn_param);
3296                 else
3297                         ret = cma_connect_ib(id_priv, conn_param);
3298         } else if (rdma_cap_iw_cm(id->device, id->port_num))
3299                 ret = cma_connect_iw(id_priv, conn_param);
3300         else
3301                 ret = -ENOSYS;
3302         if (ret)
3303                 goto err;
3304
3305         return 0;
3306 err:
3307         cma_comp_exch(id_priv, RDMA_CM_CONNECT, RDMA_CM_ROUTE_RESOLVED);
3308         return ret;
3309 }
3310 EXPORT_SYMBOL(rdma_connect);
3311
3312 static int cma_accept_ib(struct rdma_id_private *id_priv,
3313                          struct rdma_conn_param *conn_param)
3314 {
3315         struct ib_cm_rep_param rep;
3316         int ret;
3317
3318         ret = cma_modify_qp_rtr(id_priv, conn_param);
3319         if (ret)
3320                 goto out;
3321
3322         ret = cma_modify_qp_rts(id_priv, conn_param);
3323         if (ret)
3324                 goto out;
3325
3326         memset(&rep, 0, sizeof rep);
3327         rep.qp_num = id_priv->qp_num;
3328         rep.starting_psn = id_priv->seq_num;
3329         rep.private_data = conn_param->private_data;
3330         rep.private_data_len = conn_param->private_data_len;
3331         rep.responder_resources = conn_param->responder_resources;
3332         rep.initiator_depth = conn_param->initiator_depth;
3333         rep.failover_accepted = 0;
3334         rep.flow_control = conn_param->flow_control;
3335         rep.rnr_retry_count = min_t(u8, 7, conn_param->rnr_retry_count);
3336         rep.srq = id_priv->srq ? 1 : 0;
3337
3338         ret = ib_send_cm_rep(id_priv->cm_id.ib, &rep);
3339 out:
3340         return ret;
3341 }
3342
3343 static int cma_accept_iw(struct rdma_id_private *id_priv,
3344                   struct rdma_conn_param *conn_param)
3345 {
3346         struct iw_cm_conn_param iw_param;
3347         int ret;
3348
3349         ret = cma_modify_qp_rtr(id_priv, conn_param);
3350         if (ret)
3351                 return ret;
3352
3353         iw_param.ord = conn_param->initiator_depth;
3354         iw_param.ird = conn_param->responder_resources;
3355         iw_param.private_data = conn_param->private_data;
3356         iw_param.private_data_len = conn_param->private_data_len;
3357         if (id_priv->id.qp) {
3358                 iw_param.qpn = id_priv->qp_num;
3359         } else
3360                 iw_param.qpn = conn_param->qp_num;
3361
3362         return iw_cm_accept(id_priv->cm_id.iw, &iw_param);
3363 }
3364
3365 static int cma_send_sidr_rep(struct rdma_id_private *id_priv,
3366                              enum ib_cm_sidr_status status, u32 qkey,
3367                              const void *private_data, int private_data_len)
3368 {
3369         struct ib_cm_sidr_rep_param rep;
3370         int ret;
3371
3372         memset(&rep, 0, sizeof rep);
3373         rep.status = status;
3374         if (status == IB_SIDR_SUCCESS) {
3375                 ret = cma_set_qkey(id_priv, qkey);
3376                 if (ret)
3377                         return ret;
3378                 rep.qp_num = id_priv->qp_num;
3379                 rep.qkey = id_priv->qkey;
3380         }
3381         rep.private_data = private_data;
3382         rep.private_data_len = private_data_len;
3383
3384         return ib_send_cm_sidr_rep(id_priv->cm_id.ib, &rep);
3385 }
3386
3387 int rdma_accept(struct rdma_cm_id *id, struct rdma_conn_param *conn_param)
3388 {
3389         struct rdma_id_private *id_priv;
3390         int ret;
3391
3392         id_priv = container_of(id, struct rdma_id_private, id);
3393
3394         id_priv->owner = task_pid_nr(current);
3395
3396         if (!cma_comp(id_priv, RDMA_CM_CONNECT))
3397                 return -EINVAL;
3398
3399         if (!id->qp && conn_param) {
3400                 id_priv->qp_num = conn_param->qp_num;
3401                 id_priv->srq = conn_param->srq;
3402         }
3403
3404         if (rdma_cap_ib_cm(id->device, id->port_num)) {
3405                 if (id->qp_type == IB_QPT_UD) {
3406                         if (conn_param)
3407                                 ret = cma_send_sidr_rep(id_priv, IB_SIDR_SUCCESS,
3408                                                         conn_param->qkey,
3409                                                         conn_param->private_data,
3410                                                         conn_param->private_data_len);
3411                         else
3412                                 ret = cma_send_sidr_rep(id_priv, IB_SIDR_SUCCESS,
3413                                                         0, NULL, 0);
3414                 } else {
3415                         if (conn_param)
3416                                 ret = cma_accept_ib(id_priv, conn_param);
3417                         else
3418                                 ret = cma_rep_recv(id_priv);
3419                 }
3420         } else if (rdma_cap_iw_cm(id->device, id->port_num))
3421                 ret = cma_accept_iw(id_priv, conn_param);
3422         else
3423                 ret = -ENOSYS;
3424
3425         if (ret)
3426                 goto reject;
3427
3428         return 0;
3429 reject:
3430         cma_modify_qp_err(id_priv);
3431         rdma_reject(id, NULL, 0);
3432         return ret;
3433 }
3434 EXPORT_SYMBOL(rdma_accept);
3435
3436 int rdma_notify(struct rdma_cm_id *id, enum ib_event_type event)
3437 {
3438         struct rdma_id_private *id_priv;
3439         int ret;
3440
3441         id_priv = container_of(id, struct rdma_id_private, id);
3442         if (!id_priv->cm_id.ib)
3443                 return -EINVAL;
3444
3445         switch (id->device->node_type) {
3446         case RDMA_NODE_IB_CA:
3447                 ret = ib_cm_notify(id_priv->cm_id.ib, event);
3448                 break;
3449         default:
3450                 ret = 0;
3451                 break;
3452         }
3453         return ret;
3454 }
3455 EXPORT_SYMBOL(rdma_notify);
3456
3457 int rdma_reject(struct rdma_cm_id *id, const void *private_data,
3458                 u8 private_data_len)
3459 {
3460         struct rdma_id_private *id_priv;
3461         int ret;
3462
3463         id_priv = container_of(id, struct rdma_id_private, id);
3464         if (!id_priv->cm_id.ib)
3465                 return -EINVAL;
3466
3467         if (rdma_cap_ib_cm(id->device, id->port_num)) {
3468                 if (id->qp_type == IB_QPT_UD)
3469                         ret = cma_send_sidr_rep(id_priv, IB_SIDR_REJECT, 0,
3470                                                 private_data, private_data_len);
3471                 else
3472                         ret = ib_send_cm_rej(id_priv->cm_id.ib,
3473                                              IB_CM_REJ_CONSUMER_DEFINED, NULL,
3474                                              0, private_data, private_data_len);
3475         } else if (rdma_cap_iw_cm(id->device, id->port_num)) {
3476                 ret = iw_cm_reject(id_priv->cm_id.iw,
3477                                    private_data, private_data_len);
3478         } else
3479                 ret = -ENOSYS;
3480
3481         return ret;
3482 }
3483 EXPORT_SYMBOL(rdma_reject);
3484
3485 int rdma_disconnect(struct rdma_cm_id *id)
3486 {
3487         struct rdma_id_private *id_priv;
3488         int ret;
3489
3490         id_priv = container_of(id, struct rdma_id_private, id);
3491         if (!id_priv->cm_id.ib)
3492                 return -EINVAL;
3493
3494         if (rdma_cap_ib_cm(id->device, id->port_num)) {
3495                 ret = cma_modify_qp_err(id_priv);
3496                 if (ret)
3497                         goto out;
3498                 /* Initiate or respond to a disconnect. */
3499                 if (ib_send_cm_dreq(id_priv->cm_id.ib, NULL, 0))
3500                         ib_send_cm_drep(id_priv->cm_id.ib, NULL, 0);
3501         } else if (rdma_cap_iw_cm(id->device, id->port_num)) {
3502                 ret = iw_cm_disconnect(id_priv->cm_id.iw, 0);
3503         } else
3504                 ret = -EINVAL;
3505
3506 out:
3507         return ret;
3508 }
3509 EXPORT_SYMBOL(rdma_disconnect);
3510
3511 static int cma_ib_mc_handler(int status, struct ib_sa_multicast *multicast)
3512 {
3513         struct rdma_id_private *id_priv;
3514         struct cma_multicast *mc = multicast->context;
3515         struct rdma_cm_event event;
3516         int ret;
3517
3518         id_priv = mc->id_priv;
3519         if (cma_disable_callback(id_priv, RDMA_CM_ADDR_BOUND) &&
3520             cma_disable_callback(id_priv, RDMA_CM_ADDR_RESOLVED))
3521                 return 0;
3522
3523         if (!status)
3524                 status = cma_set_qkey(id_priv, be32_to_cpu(multicast->rec.qkey));
3525         mutex_lock(&id_priv->qp_mutex);
3526         if (!status && id_priv->id.qp)
3527                 status = ib_attach_mcast(id_priv->id.qp, &multicast->rec.mgid,
3528                                          be16_to_cpu(multicast->rec.mlid));
3529         mutex_unlock(&id_priv->qp_mutex);
3530
3531         memset(&event, 0, sizeof event);
3532         event.status = status;
3533         event.param.ud.private_data = mc->context;
3534         if (!status) {
3535                 event.event = RDMA_CM_EVENT_MULTICAST_JOIN;
3536                 ib_init_ah_from_mcmember(id_priv->id.device,
3537                                          id_priv->id.port_num, &multicast->rec,
3538                                          &event.param.ud.ah_attr);
3539                 event.param.ud.qp_num = 0xFFFFFF;
3540                 event.param.ud.qkey = be32_to_cpu(multicast->rec.qkey);
3541         } else
3542                 event.event = RDMA_CM_EVENT_MULTICAST_ERROR;
3543
3544         ret = id_priv->id.event_handler(&id_priv->id, &event);
3545         if (ret) {
3546                 cma_exch(id_priv, RDMA_CM_DESTROYING);
3547                 mutex_unlock(&id_priv->handler_mutex);
3548                 rdma_destroy_id(&id_priv->id);
3549                 return 0;
3550         }
3551
3552         mutex_unlock(&id_priv->handler_mutex);
3553         return 0;
3554 }
3555
3556 static void cma_set_mgid(struct rdma_id_private *id_priv,
3557                          struct sockaddr *addr, union ib_gid *mgid)
3558 {
3559         unsigned char mc_map[MAX_ADDR_LEN];
3560         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
3561         struct sockaddr_in *sin = (struct sockaddr_in *) addr;
3562         struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *) addr;
3563
3564         if (cma_any_addr(addr)) {
3565                 memset(mgid, 0, sizeof *mgid);
3566         } else if ((addr->sa_family == AF_INET6) &&
3567                    ((be32_to_cpu(sin6->sin6_addr.s6_addr32[0]) & 0xFFF0FFFF) ==
3568                                                                  0xFF10A01B)) {
3569                 /* IPv6 address is an SA assigned MGID. */
3570                 memcpy(mgid, &sin6->sin6_addr, sizeof *mgid);
3571         } else if (addr->sa_family == AF_IB) {
3572                 memcpy(mgid, &((struct sockaddr_ib *) addr)->sib_addr, sizeof *mgid);
3573         } else if ((addr->sa_family == AF_INET6)) {
3574                 ipv6_ib_mc_map(&sin6->sin6_addr, dev_addr->broadcast, mc_map);
3575                 if (id_priv->id.ps == RDMA_PS_UDP)
3576                         mc_map[7] = 0x01;       /* Use RDMA CM signature */
3577                 *mgid = *(union ib_gid *) (mc_map + 4);
3578         } else {
3579                 ip_ib_mc_map(sin->sin_addr.s_addr, dev_addr->broadcast, mc_map);
3580                 if (id_priv->id.ps == RDMA_PS_UDP)
3581                         mc_map[7] = 0x01;       /* Use RDMA CM signature */
3582                 *mgid = *(union ib_gid *) (mc_map + 4);
3583         }
3584 }
3585
3586 static int cma_join_ib_multicast(struct rdma_id_private *id_priv,
3587                                  struct cma_multicast *mc)
3588 {
3589         struct ib_sa_mcmember_rec rec;
3590         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
3591         ib_sa_comp_mask comp_mask;
3592         int ret;
3593
3594         ib_addr_get_mgid(dev_addr, &rec.mgid);
3595         ret = ib_sa_get_mcmember_rec(id_priv->id.device, id_priv->id.port_num,
3596                                      &rec.mgid, &rec);
3597         if (ret)
3598                 return ret;
3599
3600         ret = cma_set_qkey(id_priv, 0);
3601         if (ret)
3602                 return ret;
3603
3604         cma_set_mgid(id_priv, (struct sockaddr *) &mc->addr, &rec.mgid);
3605         rec.qkey = cpu_to_be32(id_priv->qkey);
3606         rdma_addr_get_sgid(dev_addr, &rec.port_gid);
3607         rec.pkey = cpu_to_be16(ib_addr_get_pkey(dev_addr));
3608         rec.join_state = 1;
3609
3610         comp_mask = IB_SA_MCMEMBER_REC_MGID | IB_SA_MCMEMBER_REC_PORT_GID |
3611                     IB_SA_MCMEMBER_REC_PKEY | IB_SA_MCMEMBER_REC_JOIN_STATE |
3612                     IB_SA_MCMEMBER_REC_QKEY | IB_SA_MCMEMBER_REC_SL |
3613                     IB_SA_MCMEMBER_REC_FLOW_LABEL |
3614                     IB_SA_MCMEMBER_REC_TRAFFIC_CLASS;
3615
3616         if (id_priv->id.ps == RDMA_PS_IPOIB)
3617                 comp_mask |= IB_SA_MCMEMBER_REC_RATE |
3618                              IB_SA_MCMEMBER_REC_RATE_SELECTOR |
3619                              IB_SA_MCMEMBER_REC_MTU_SELECTOR |
3620                              IB_SA_MCMEMBER_REC_MTU |
3621                              IB_SA_MCMEMBER_REC_HOP_LIMIT;
3622
3623         mc->multicast.ib = ib_sa_join_multicast(&sa_client, id_priv->id.device,
3624                                                 id_priv->id.port_num, &rec,
3625                                                 comp_mask, GFP_KERNEL,
3626                                                 cma_ib_mc_handler, mc);
3627         return PTR_ERR_OR_ZERO(mc->multicast.ib);
3628 }
3629
3630 static void iboe_mcast_work_handler(struct work_struct *work)
3631 {
3632         struct iboe_mcast_work *mw = container_of(work, struct iboe_mcast_work, work);
3633         struct cma_multicast *mc = mw->mc;
3634         struct ib_sa_multicast *m = mc->multicast.ib;
3635
3636         mc->multicast.ib->context = mc;
3637         cma_ib_mc_handler(0, m);
3638         kref_put(&mc->mcref, release_mc);
3639         kfree(mw);
3640 }
3641
3642 static void cma_iboe_set_mgid(struct sockaddr *addr, union ib_gid *mgid)
3643 {
3644         struct sockaddr_in *sin = (struct sockaddr_in *)addr;
3645         struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)addr;
3646
3647         if (cma_any_addr(addr)) {
3648                 memset(mgid, 0, sizeof *mgid);
3649         } else if (addr->sa_family == AF_INET6) {
3650                 memcpy(mgid, &sin6->sin6_addr, sizeof *mgid);
3651         } else {
3652                 mgid->raw[0] = 0xff;
3653                 mgid->raw[1] = 0x0e;
3654                 mgid->raw[2] = 0;
3655                 mgid->raw[3] = 0;
3656                 mgid->raw[4] = 0;
3657                 mgid->raw[5] = 0;
3658                 mgid->raw[6] = 0;
3659                 mgid->raw[7] = 0;
3660                 mgid->raw[8] = 0;
3661                 mgid->raw[9] = 0;
3662                 mgid->raw[10] = 0xff;
3663                 mgid->raw[11] = 0xff;
3664                 *(__be32 *)(&mgid->raw[12]) = sin->sin_addr.s_addr;
3665         }
3666 }
3667
3668 static int cma_iboe_join_multicast(struct rdma_id_private *id_priv,
3669                                    struct cma_multicast *mc)
3670 {
3671         struct iboe_mcast_work *work;
3672         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
3673         int err;
3674         struct sockaddr *addr = (struct sockaddr *)&mc->addr;
3675         struct net_device *ndev = NULL;
3676
3677         if (cma_zero_addr((struct sockaddr *)&mc->addr))
3678                 return -EINVAL;
3679
3680         work = kzalloc(sizeof *work, GFP_KERNEL);
3681         if (!work)
3682                 return -ENOMEM;
3683
3684         mc->multicast.ib = kzalloc(sizeof(struct ib_sa_multicast), GFP_KERNEL);
3685         if (!mc->multicast.ib) {
3686                 err = -ENOMEM;
3687                 goto out1;
3688         }
3689
3690         cma_iboe_set_mgid(addr, &mc->multicast.ib->rec.mgid);
3691
3692         mc->multicast.ib->rec.pkey = cpu_to_be16(0xffff);
3693         if (id_priv->id.ps == RDMA_PS_UDP)
3694                 mc->multicast.ib->rec.qkey = cpu_to_be32(RDMA_UDP_QKEY);
3695
3696         if (dev_addr->bound_dev_if)
3697                 ndev = dev_get_by_index(&init_net, dev_addr->bound_dev_if);
3698         if (!ndev) {
3699                 err = -ENODEV;
3700                 goto out2;
3701         }
3702         mc->multicast.ib->rec.rate = iboe_get_rate(ndev);
3703         mc->multicast.ib->rec.hop_limit = 1;
3704         mc->multicast.ib->rec.mtu = iboe_get_mtu(ndev->mtu);
3705         dev_put(ndev);
3706         if (!mc->multicast.ib->rec.mtu) {
3707                 err = -EINVAL;
3708                 goto out2;
3709         }
3710         rdma_ip2gid((struct sockaddr *)&id_priv->id.route.addr.src_addr,
3711                     &mc->multicast.ib->rec.port_gid);
3712         work->id = id_priv;
3713         work->mc = mc;
3714         INIT_WORK(&work->work, iboe_mcast_work_handler);
3715         kref_get(&mc->mcref);
3716         queue_work(cma_wq, &work->work);
3717
3718         return 0;
3719
3720 out2:
3721         kfree(mc->multicast.ib);
3722 out1:
3723         kfree(work);
3724         return err;
3725 }
3726
3727 int rdma_join_multicast(struct rdma_cm_id *id, struct sockaddr *addr,
3728                         void *context)
3729 {
3730         struct rdma_id_private *id_priv;
3731         struct cma_multicast *mc;
3732         int ret;
3733
3734         id_priv = container_of(id, struct rdma_id_private, id);
3735         if (!cma_comp(id_priv, RDMA_CM_ADDR_BOUND) &&
3736             !cma_comp(id_priv, RDMA_CM_ADDR_RESOLVED))
3737                 return -EINVAL;
3738
3739         mc = kmalloc(sizeof *mc, GFP_KERNEL);
3740         if (!mc)
3741                 return -ENOMEM;
3742
3743         memcpy(&mc->addr, addr, rdma_addr_size(addr));
3744         mc->context = context;
3745         mc->id_priv = id_priv;
3746
3747         spin_lock(&id_priv->lock);
3748         list_add(&mc->list, &id_priv->mc_list);
3749         spin_unlock(&id_priv->lock);
3750
3751         if (rdma_protocol_roce(id->device, id->port_num)) {
3752                 kref_init(&mc->mcref);
3753                 ret = cma_iboe_join_multicast(id_priv, mc);
3754         } else if (rdma_cap_ib_mcast(id->device, id->port_num))
3755                 ret = cma_join_ib_multicast(id_priv, mc);
3756         else
3757                 ret = -ENOSYS;
3758
3759         if (ret) {
3760                 spin_lock_irq(&id_priv->lock);
3761                 list_del(&mc->list);
3762                 spin_unlock_irq(&id_priv->lock);
3763                 kfree(mc);
3764         }
3765         return ret;
3766 }
3767 EXPORT_SYMBOL(rdma_join_multicast);
3768
3769 void rdma_leave_multicast(struct rdma_cm_id *id, struct sockaddr *addr)
3770 {
3771         struct rdma_id_private *id_priv;
3772         struct cma_multicast *mc;
3773
3774         id_priv = container_of(id, struct rdma_id_private, id);
3775         spin_lock_irq(&id_priv->lock);
3776         list_for_each_entry(mc, &id_priv->mc_list, list) {
3777                 if (!memcmp(&mc->addr, addr, rdma_addr_size(addr))) {
3778                         list_del(&mc->list);
3779                         spin_unlock_irq(&id_priv->lock);
3780
3781                         if (id->qp)
3782                                 ib_detach_mcast(id->qp,
3783                                                 &mc->multicast.ib->rec.mgid,
3784                                                 be16_to_cpu(mc->multicast.ib->rec.mlid));
3785
3786                         BUG_ON(id_priv->cma_dev->device != id->device);
3787
3788                         if (rdma_cap_ib_mcast(id->device, id->port_num)) {
3789                                 ib_sa_free_multicast(mc->multicast.ib);
3790                                 kfree(mc);
3791                         } else if (rdma_protocol_roce(id->device, id->port_num))
3792                                 kref_put(&mc->mcref, release_mc);
3793
3794                         return;
3795                 }
3796         }
3797         spin_unlock_irq(&id_priv->lock);
3798 }
3799 EXPORT_SYMBOL(rdma_leave_multicast);
3800
3801 static int cma_netdev_change(struct net_device *ndev, struct rdma_id_private *id_priv)
3802 {
3803         struct rdma_dev_addr *dev_addr;
3804         struct cma_ndev_work *work;
3805
3806         dev_addr = &id_priv->id.route.addr.dev_addr;
3807
3808         if ((dev_addr->bound_dev_if == ndev->ifindex) &&
3809             (net_eq(dev_net(ndev), dev_addr->net)) &&
3810             memcmp(dev_addr->src_dev_addr, ndev->dev_addr, ndev->addr_len)) {
3811                 printk(KERN_INFO "RDMA CM addr change for ndev %s used by id %p\n",
3812                        ndev->name, &id_priv->id);
3813                 work = kzalloc(sizeof *work, GFP_KERNEL);
3814                 if (!work)
3815                         return -ENOMEM;
3816
3817                 INIT_WORK(&work->work, cma_ndev_work_handler);
3818                 work->id = id_priv;
3819                 work->event.event = RDMA_CM_EVENT_ADDR_CHANGE;
3820                 atomic_inc(&id_priv->refcount);
3821                 queue_work(cma_wq, &work->work);
3822         }
3823
3824         return 0;
3825 }
3826
3827 static int cma_netdev_callback(struct notifier_block *self, unsigned long event,
3828                                void *ptr)
3829 {
3830         struct net_device *ndev = netdev_notifier_info_to_dev(ptr);
3831         struct cma_device *cma_dev;
3832         struct rdma_id_private *id_priv;
3833         int ret = NOTIFY_DONE;
3834
3835         if (event != NETDEV_BONDING_FAILOVER)
3836                 return NOTIFY_DONE;
3837
3838         if (!(ndev->flags & IFF_MASTER) || !(ndev->priv_flags & IFF_BONDING))
3839                 return NOTIFY_DONE;
3840
3841         mutex_lock(&lock);
3842         list_for_each_entry(cma_dev, &dev_list, list)
3843                 list_for_each_entry(id_priv, &cma_dev->id_list, list) {
3844                         ret = cma_netdev_change(ndev, id_priv);
3845                         if (ret)
3846                                 goto out;
3847                 }
3848
3849 out:
3850         mutex_unlock(&lock);
3851         return ret;
3852 }
3853
3854 static struct notifier_block cma_nb = {
3855         .notifier_call = cma_netdev_callback
3856 };
3857
3858 static void cma_add_one(struct ib_device *device)
3859 {
3860         struct cma_device *cma_dev;
3861         struct rdma_id_private *id_priv;
3862
3863         cma_dev = kmalloc(sizeof *cma_dev, GFP_KERNEL);
3864         if (!cma_dev)
3865                 return;
3866
3867         cma_dev->device = device;
3868
3869         init_completion(&cma_dev->comp);
3870         atomic_set(&cma_dev->refcount, 1);
3871         INIT_LIST_HEAD(&cma_dev->id_list);
3872         ib_set_client_data(device, &cma_client, cma_dev);
3873
3874         mutex_lock(&lock);
3875         list_add_tail(&cma_dev->list, &dev_list);
3876         list_for_each_entry(id_priv, &listen_any_list, list)
3877                 cma_listen_on_dev(id_priv, cma_dev);
3878         mutex_unlock(&lock);
3879 }
3880
3881 static int cma_remove_id_dev(struct rdma_id_private *id_priv)
3882 {
3883         struct rdma_cm_event event;
3884         enum rdma_cm_state state;
3885         int ret = 0;
3886
3887         /* Record that we want to remove the device */
3888         state = cma_exch(id_priv, RDMA_CM_DEVICE_REMOVAL);
3889         if (state == RDMA_CM_DESTROYING)
3890                 return 0;
3891
3892         cma_cancel_operation(id_priv, state);
3893         mutex_lock(&id_priv->handler_mutex);
3894
3895         /* Check for destruction from another callback. */
3896         if (!cma_comp(id_priv, RDMA_CM_DEVICE_REMOVAL))
3897                 goto out;
3898
3899         memset(&event, 0, sizeof event);
3900         event.event = RDMA_CM_EVENT_DEVICE_REMOVAL;
3901         ret = id_priv->id.event_handler(&id_priv->id, &event);
3902 out:
3903         mutex_unlock(&id_priv->handler_mutex);
3904         return ret;
3905 }
3906
3907 static void cma_process_remove(struct cma_device *cma_dev)
3908 {
3909         struct rdma_id_private *id_priv;
3910         int ret;
3911
3912         mutex_lock(&lock);
3913         while (!list_empty(&cma_dev->id_list)) {
3914                 id_priv = list_entry(cma_dev->id_list.next,
3915                                      struct rdma_id_private, list);
3916
3917                 list_del(&id_priv->listen_list);
3918                 list_del_init(&id_priv->list);
3919                 atomic_inc(&id_priv->refcount);
3920                 mutex_unlock(&lock);
3921
3922                 ret = id_priv->internal_id ? 1 : cma_remove_id_dev(id_priv);
3923                 cma_deref_id(id_priv);
3924                 if (ret)
3925                         rdma_destroy_id(&id_priv->id);
3926
3927                 mutex_lock(&lock);
3928         }
3929         mutex_unlock(&lock);
3930
3931         cma_deref_dev(cma_dev);
3932         wait_for_completion(&cma_dev->comp);
3933 }
3934
3935 static void cma_remove_one(struct ib_device *device, void *client_data)
3936 {
3937         struct cma_device *cma_dev = client_data;
3938
3939         if (!cma_dev)
3940                 return;
3941
3942         mutex_lock(&lock);
3943         list_del(&cma_dev->list);
3944         mutex_unlock(&lock);
3945
3946         cma_process_remove(cma_dev);
3947         kfree(cma_dev);
3948 }
3949
3950 static int cma_get_id_stats(struct sk_buff *skb, struct netlink_callback *cb)
3951 {
3952         struct nlmsghdr *nlh;
3953         struct rdma_cm_id_stats *id_stats;
3954         struct rdma_id_private *id_priv;
3955         struct rdma_cm_id *id = NULL;
3956         struct cma_device *cma_dev;
3957         int i_dev = 0, i_id = 0;
3958
3959         /*
3960          * We export all of the IDs as a sequence of messages.  Each
3961          * ID gets its own netlink message.
3962          */
3963         mutex_lock(&lock);
3964
3965         list_for_each_entry(cma_dev, &dev_list, list) {
3966                 if (i_dev < cb->args[0]) {
3967                         i_dev++;
3968                         continue;
3969                 }
3970
3971                 i_id = 0;
3972                 list_for_each_entry(id_priv, &cma_dev->id_list, list) {
3973                         if (i_id < cb->args[1]) {
3974                                 i_id++;
3975                                 continue;
3976                         }
3977
3978                         id_stats = ibnl_put_msg(skb, &nlh, cb->nlh->nlmsg_seq,
3979                                                 sizeof *id_stats, RDMA_NL_RDMA_CM,
3980                                                 RDMA_NL_RDMA_CM_ID_STATS,
3981                                                 NLM_F_MULTI);
3982                         if (!id_stats)
3983                                 goto out;
3984
3985                         memset(id_stats, 0, sizeof *id_stats);
3986                         id = &id_priv->id;
3987                         id_stats->node_type = id->route.addr.dev_addr.dev_type;
3988                         id_stats->port_num = id->port_num;
3989                         id_stats->bound_dev_if =
3990                                 id->route.addr.dev_addr.bound_dev_if;
3991
3992                         if (ibnl_put_attr(skb, nlh,
3993                                           rdma_addr_size(cma_src_addr(id_priv)),
3994                                           cma_src_addr(id_priv),
3995                                           RDMA_NL_RDMA_CM_ATTR_SRC_ADDR))
3996                                 goto out;
3997                         if (ibnl_put_attr(skb, nlh,
3998                                           rdma_addr_size(cma_src_addr(id_priv)),
3999                                           cma_dst_addr(id_priv),
4000                                           RDMA_NL_RDMA_CM_ATTR_DST_ADDR))
4001                                 goto out;
4002
4003                         id_stats->pid           = id_priv->owner;
4004                         id_stats->port_space    = id->ps;
4005                         id_stats->cm_state      = id_priv->state;
4006                         id_stats->qp_num        = id_priv->qp_num;
4007                         id_stats->qp_type       = id->qp_type;
4008
4009                         i_id++;
4010                 }
4011
4012                 cb->args[1] = 0;
4013                 i_dev++;
4014         }
4015
4016 out:
4017         mutex_unlock(&lock);
4018         cb->args[0] = i_dev;
4019         cb->args[1] = i_id;
4020
4021         return skb->len;
4022 }
4023
4024 static const struct ibnl_client_cbs cma_cb_table[] = {
4025         [RDMA_NL_RDMA_CM_ID_STATS] = { .dump = cma_get_id_stats,
4026                                        .module = THIS_MODULE },
4027 };
4028
4029 static int cma_init_net(struct net *net)
4030 {
4031         struct cma_pernet *pernet = cma_pernet(net);
4032
4033         idr_init(&pernet->tcp_ps);
4034         idr_init(&pernet->udp_ps);
4035         idr_init(&pernet->ipoib_ps);
4036         idr_init(&pernet->ib_ps);
4037
4038         return 0;
4039 }
4040
4041 static void cma_exit_net(struct net *net)
4042 {
4043         struct cma_pernet *pernet = cma_pernet(net);
4044
4045         idr_destroy(&pernet->tcp_ps);
4046         idr_destroy(&pernet->udp_ps);
4047         idr_destroy(&pernet->ipoib_ps);
4048         idr_destroy(&pernet->ib_ps);
4049 }
4050
4051 static struct pernet_operations cma_pernet_operations = {
4052         .init = cma_init_net,
4053         .exit = cma_exit_net,
4054         .id = &cma_pernet_id,
4055         .size = sizeof(struct cma_pernet),
4056 };
4057
4058 static int __init cma_init(void)
4059 {
4060         int ret;
4061
4062         cma_wq = create_singlethread_workqueue("rdma_cm");
4063         if (!cma_wq)
4064                 return -ENOMEM;
4065
4066         ret = register_pernet_subsys(&cma_pernet_operations);
4067         if (ret)
4068                 goto err_wq;
4069
4070         ib_sa_register_client(&sa_client);
4071         rdma_addr_register_client(&addr_client);
4072         register_netdevice_notifier(&cma_nb);
4073
4074         ret = ib_register_client(&cma_client);
4075         if (ret)
4076                 goto err;
4077
4078         if (ibnl_add_client(RDMA_NL_RDMA_CM, RDMA_NL_RDMA_CM_NUM_OPS, cma_cb_table))
4079                 printk(KERN_WARNING "RDMA CMA: failed to add netlink callback\n");
4080
4081         return 0;
4082
4083 err:
4084         unregister_netdevice_notifier(&cma_nb);
4085         rdma_addr_unregister_client(&addr_client);
4086         ib_sa_unregister_client(&sa_client);
4087 err_wq:
4088         destroy_workqueue(cma_wq);
4089         return ret;
4090 }
4091
4092 static void __exit cma_cleanup(void)
4093 {
4094         ibnl_remove_client(RDMA_NL_RDMA_CM);
4095         ib_unregister_client(&cma_client);
4096         unregister_netdevice_notifier(&cma_nb);
4097         rdma_addr_unregister_client(&addr_client);
4098         ib_sa_unregister_client(&sa_client);
4099         unregister_pernet_subsys(&cma_pernet_operations);
4100         destroy_workqueue(cma_wq);
4101 }
4102
4103 module_init(cma_init);
4104 module_exit(cma_cleanup);