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/igmp.h>
  42 #include <linux/idr.h>
  43 #include <linux/inetdevice.h>
  44 #include <linux/slab.h>
  45 #include <linux/module.h>
  46 #include <net/route.h>
  47
  48 #include <net/net_namespace.h>
  49 #include <net/netns/generic.h>
  50 #include <net/tcp.h>
  51 #include <net/ipv6.h>
  52 #include <net/ip_fib.h>
  53 #include <net/ip6_route.h>
  54
  55 #include <rdma/rdma_cm.h>
  56 #include <rdma/rdma_cm_ib.h>
  57 #include <rdma/rdma_netlink.h>
  58 #include <rdma/ib.h>
  59 #include <rdma/ib_cache.h>
  60 #include <rdma/ib_cm.h>
  61 #include <rdma/ib_sa.h>
  62 #include <rdma/iw_cm.h>
  63
  64 #include "core_priv.h"
  65
  66 MODULE_AUTHOR("Sean Hefty");
  67 MODULE_DESCRIPTION("Generic RDMA CM Agent");
  68 MODULE_LICENSE("Dual BSD/GPL");
  69
  70 #define CMA_CM_RESPONSE_TIMEOUT 20
  71 #define CMA_QUERY_CLASSPORT_INFO_TIMEOUT 3000
  72 #define CMA_MAX_CM_RETRIES 15
  73 #define CMA_CM_MRA_SETTING (IB_CM_MRA_FLAG_DELAY | 24)
  74 #define CMA_IBOE_PACKET_LIFETIME 18
  75
  76 static const char * const cma_events[] = {
  77         [RDMA_CM_EVENT_ADDR_RESOLVED]    = "address resolved",
  78         [RDMA_CM_EVENT_ADDR_ERROR]       = "address error",
  79         [RDMA_CM_EVENT_ROUTE_RESOLVED]   = "route resolved ",
  80         [RDMA_CM_EVENT_ROUTE_ERROR]      = "route error",
  81         [RDMA_CM_EVENT_CONNECT_REQUEST]  = "connect request",
  82         [RDMA_CM_EVENT_CONNECT_RESPONSE] = "connect response",
  83         [RDMA_CM_EVENT_CONNECT_ERROR]    = "connect error",
  84         [RDMA_CM_EVENT_UNREACHABLE]      = "unreachable",
  85         [RDMA_CM_EVENT_REJECTED]         = "rejected",
  86         [RDMA_CM_EVENT_ESTABLISHED]      = "established",
  87         [RDMA_CM_EVENT_DISCONNECTED]     = "disconnected",
  88         [RDMA_CM_EVENT_DEVICE_REMOVAL]   = "device removal",
  89         [RDMA_CM_EVENT_MULTICAST_JOIN]   = "multicast join",
  90         [RDMA_CM_EVENT_MULTICAST_ERROR]  = "multicast error",
  91         [RDMA_CM_EVENT_ADDR_CHANGE]      = "address change",
  92         [RDMA_CM_EVENT_TIMEWAIT_EXIT]    = "timewait exit",
  93 };
  94
  95 const char *__attribute_const__ rdma_event_msg(enum rdma_cm_event_type event)
  96 {
  97         size_t index = event;
  98
  99         return (index < ARRAY_SIZE(cma_events) && cma_events[index]) ?
 100                         cma_events[index] : "unrecognized event";
 101 }
 102 EXPORT_SYMBOL(rdma_event_msg);
 103
 104 static void cma_add_one(struct ib_device *device);
 105 static void cma_remove_one(struct ib_device *device, void *client_data);
 106
 107 static struct ib_client cma_client = {
 108         .name   = "cma",
 109         .add    = cma_add_one,
 110         .remove = cma_remove_one
 111 };
 112
 113 static struct ib_sa_client sa_client;
 114 static struct rdma_addr_client addr_client;
 115 static LIST_HEAD(dev_list);
 116 static LIST_HEAD(listen_any_list);
 117 static DEFINE_MUTEX(lock);
 118 static struct workqueue_struct *cma_wq;
 119 static int cma_pernet_id;
 120
 121 struct cma_pernet {
 122         struct idr tcp_ps;
 123         struct idr udp_ps;
 124         struct idr ipoib_ps;
 125         struct idr ib_ps;
 126 };
 127
 128 static struct cma_pernet *cma_pernet(struct net *net)
 129 {
 130         return net_generic(net, cma_pernet_id);
 131 }
 132
 133 static struct idr *cma_pernet_idr(struct net *net, enum rdma_port_space ps)
 134 {
 135         struct cma_pernet *pernet = cma_pernet(net);
 136
 137         switch (ps) {
 138         case RDMA_PS_TCP:
 139                 return &pernet->tcp_ps;
 140         case RDMA_PS_UDP:
 141                 return &pernet->udp_ps;
 142         case RDMA_PS_IPOIB:
 143                 return &pernet->ipoib_ps;
 144         case RDMA_PS_IB:
 145                 return &pernet->ib_ps;
 146         default:
 147                 return NULL;
 148         }
 149 }
 150
 151 struct cma_device {
 152         struct list_head        list;
 153         struct ib_device        *device;
 154         struct completion       comp;
 155         atomic_t                refcount;
 156         struct list_head        id_list;
 157         enum ib_gid_type        *default_gid_type;
 158 };
 159
 160 struct rdma_bind_list {
 161         enum rdma_port_space    ps;
 162         struct hlist_head       owners;
 163         unsigned short          port;
 164 };
 165
 166 struct class_port_info_context {
 167         struct ib_class_port_info       *class_port_info;
 168         struct ib_device                *device;
 169         struct completion               done;
 170         struct ib_sa_query              *sa_query;
 171         u8                              port_num;
 172 };
 173
 174 static int cma_ps_alloc(struct net *net, enum rdma_port_space ps,
 175                         struct rdma_bind_list *bind_list, int snum)
 176 {
 177         struct idr *idr = cma_pernet_idr(net, ps);
 178
 179         return idr_alloc(idr, bind_list, snum, snum + 1, GFP_KERNEL);
 180 }
 181
 182 static struct rdma_bind_list *cma_ps_find(struct net *net,
 183                                           enum rdma_port_space ps, int snum)
 184 {
 185         struct idr *idr = cma_pernet_idr(net, ps);
 186
 187         return idr_find(idr, snum);
 188 }
 189
 190 static void cma_ps_remove(struct net *net, enum rdma_port_space ps, int snum)
 191 {
 192         struct idr *idr = cma_pernet_idr(net, ps);
 193
 194         idr_remove(idr, snum);
 195 }
 196
 197 enum {
 198         CMA_OPTION_AFONLY,
 199 };
 200
 201 void cma_ref_dev(struct cma_device *cma_dev)
 202 {
 203         atomic_inc(&cma_dev->refcount);
 204 }
 205
 206 struct cma_device *cma_enum_devices_by_ibdev(cma_device_filter  filter,
 207                                              void               *cookie)
 208 {
 209         struct cma_device *cma_dev;
 210         struct cma_device *found_cma_dev = NULL;
 211
 212         mutex_lock(&lock);
 213
 214         list_for_each_entry(cma_dev, &dev_list, list)
 215                 if (filter(cma_dev->device, cookie)) {
 216                         found_cma_dev = cma_dev;
 217                         break;
 218                 }
 219
 220         if (found_cma_dev)
 221                 cma_ref_dev(found_cma_dev);
 222         mutex_unlock(&lock);
 223         return found_cma_dev;
 224 }
 225
 226 int cma_get_default_gid_type(struct cma_device *cma_dev,
 227                              unsigned int port)
 228 {
 229         if (port < rdma_start_port(cma_dev->device) ||
 230             port > rdma_end_port(cma_dev->device))
 231                 return -EINVAL;
 232
 233         return cma_dev->default_gid_type[port - rdma_start_port(cma_dev->device)];
 234 }
 235
 236 int cma_set_default_gid_type(struct cma_device *cma_dev,
 237                              unsigned int port,
 238                              enum ib_gid_type default_gid_type)
 239 {
 240         unsigned long supported_gids;
 241
 242         if (port < rdma_start_port(cma_dev->device) ||
 243             port > rdma_end_port(cma_dev->device))
 244                 return -EINVAL;
 245
 246         supported_gids = roce_gid_type_mask_support(cma_dev->device, port);
 247
 248         if (!(supported_gids & 1 << default_gid_type))
 249                 return -EINVAL;
 250
 251         cma_dev->default_gid_type[port - rdma_start_port(cma_dev->device)] =
 252                 default_gid_type;
 253
 254         return 0;
 255 }
 256
 257 struct ib_device *cma_get_ib_dev(struct cma_device *cma_dev)
 258 {
 259         return cma_dev->device;
 260 }
 261
 262 /*
 263  * Device removal can occur at anytime, so we need extra handling to
 264  * serialize notifying the user of device removal with other callbacks.
 265  * We do this by disabling removal notification while a callback is in process,
 266  * and reporting it after the callback completes.
 267  */
 268 struct rdma_id_private {
 269         struct rdma_cm_id       id;
 270
 271         struct rdma_bind_list   *bind_list;
 272         struct hlist_node       node;
 273         struct list_head        list; /* listen_any_list or cma_device.list */
 274         struct list_head        listen_list; /* per device listens */
 275         struct cma_device       *cma_dev;
 276         struct list_head        mc_list;
 277
 278         int                     internal_id;
 279         enum rdma_cm_state      state;
 280         spinlock_t              lock;
 281         struct mutex            qp_mutex;
 282
 283         struct completion       comp;
 284         atomic_t                refcount;
 285         struct mutex            handler_mutex;
 286
 287         int                     backlog;
 288         int                     timeout_ms;
 289         struct ib_sa_query      *query;
 290         int                     query_id;
 291         union {
 292                 struct ib_cm_id *ib;
 293                 struct iw_cm_id *iw;
 294         } cm_id;
 295
 296         u32                     seq_num;
 297         u32                     qkey;
 298         u32                     qp_num;
 299         pid_t                   owner;
 300         u32                     options;
 301         u8                      srq;
 302         u8                      tos;
 303         u8                      reuseaddr;
 304         u8                      afonly;
 305         enum ib_gid_type        gid_type;
 306 };
 307
 308 struct cma_multicast {
 309         struct rdma_id_private *id_priv;
 310         union {
 311                 struct ib_sa_multicast *ib;
 312         } multicast;
 313         struct list_head        list;
 314         void                    *context;
 315         struct sockaddr_storage addr;
 316         struct kref             mcref;
 317         bool                    igmp_joined;
 318         u8                      join_state;
 319 };
 320
 321 struct cma_work {
 322         struct work_struct      work;
 323         struct rdma_id_private  *id;
 324         enum rdma_cm_state      old_state;
 325         enum rdma_cm_state      new_state;
 326         struct rdma_cm_event    event;
 327 };
 328
 329 struct cma_ndev_work {
 330         struct work_struct      work;
 331         struct rdma_id_private  *id;
 332         struct rdma_cm_event    event;
 333 };
 334
 335 struct iboe_mcast_work {
 336         struct work_struct       work;
 337         struct rdma_id_private  *id;
 338         struct cma_multicast    *mc;
 339 };
 340
 341 union cma_ip_addr {
 342         struct in6_addr ip6;
 343         struct {
 344                 __be32 pad[3];
 345                 __be32 addr;
 346         } ip4;
 347 };
 348
 349 struct cma_hdr {
 350         u8 cma_version;
 351         u8 ip_version;  /* IP version: 7:4 */
 352         __be16 port;
 353         union cma_ip_addr src_addr;
 354         union cma_ip_addr dst_addr;
 355 };
 356
 357 #define CMA_VERSION 0x00
 358
 359 struct cma_req_info {
 360         struct ib_device *device;
 361         int port;
 362         union ib_gid local_gid;
 363         __be64 service_id;
 364         u16 pkey;
 365         bool has_gid:1;
 366 };
 367
 368 static int cma_comp(struct rdma_id_private *id_priv, enum rdma_cm_state comp)
 369 {
 370         unsigned long flags;
 371         int ret;
 372
 373         spin_lock_irqsave(&id_priv->lock, flags);
 374         ret = (id_priv->state == comp);
 375         spin_unlock_irqrestore(&id_priv->lock, flags);
 376         return ret;
 377 }
 378
 379 static int cma_comp_exch(struct rdma_id_private *id_priv,
 380                          enum rdma_cm_state comp, enum rdma_cm_state exch)
 381 {
 382         unsigned long flags;
 383         int ret;
 384
 385         spin_lock_irqsave(&id_priv->lock, flags);
 386         if ((ret = (id_priv->state == comp)))
 387                 id_priv->state = exch;
 388         spin_unlock_irqrestore(&id_priv->lock, flags);
 389         return ret;
 390 }
 391
 392 static enum rdma_cm_state cma_exch(struct rdma_id_private *id_priv,
 393                                    enum rdma_cm_state exch)
 394 {
 395         unsigned long flags;
 396         enum rdma_cm_state old;
 397
 398         spin_lock_irqsave(&id_priv->lock, flags);
 399         old = id_priv->state;
 400         id_priv->state = exch;
 401         spin_unlock_irqrestore(&id_priv->lock, flags);
 402         return old;
 403 }
 404
 405 static inline u8 cma_get_ip_ver(const struct cma_hdr *hdr)
 406 {
 407         return hdr->ip_version >> 4;
 408 }
 409
 410 static inline void cma_set_ip_ver(struct cma_hdr *hdr, u8 ip_ver)
 411 {
 412         hdr->ip_version = (ip_ver << 4) | (hdr->ip_version & 0xF);
 413 }
 414
 415 static int cma_igmp_send(struct net_device *ndev, union ib_gid *mgid, bool join)
 416 {
 417         struct in_device *in_dev = NULL;
 418
 419         if (ndev) {
 420                 rtnl_lock();
 421                 in_dev = __in_dev_get_rtnl(ndev);
 422                 if (in_dev) {
 423                         if (join)
 424                                 ip_mc_inc_group(in_dev,
 425                                                 *(__be32 *)(mgid->raw + 12));
 426                         else
 427                                 ip_mc_dec_group(in_dev,
 428                                                 *(__be32 *)(mgid->raw + 12));
 429                 }
 430                 rtnl_unlock();
 431         }
 432         return (in_dev) ? 0 : -ENODEV;
 433 }
 434
 435 static void _cma_attach_to_dev(struct rdma_id_private *id_priv,
 436                                struct cma_device *cma_dev)
 437 {
 438         cma_ref_dev(cma_dev);
 439         id_priv->cma_dev = cma_dev;
 440         id_priv->gid_type = 0;
 441         id_priv->id.device = cma_dev->device;
 442         id_priv->id.route.addr.dev_addr.transport =
 443                 rdma_node_get_transport(cma_dev->device->node_type);
 444         list_add_tail(&id_priv->list, &cma_dev->id_list);
 445 }
 446
 447 static void cma_attach_to_dev(struct rdma_id_private *id_priv,
 448                               struct cma_device *cma_dev)
 449 {
 450         _cma_attach_to_dev(id_priv, cma_dev);
 451         id_priv->gid_type =
 452                 cma_dev->default_gid_type[id_priv->id.port_num -
 453                                           rdma_start_port(cma_dev->device)];
 454 }
 455
 456 void cma_deref_dev(struct cma_device *cma_dev)
 457 {
 458         if (atomic_dec_and_test(&cma_dev->refcount))
 459                 complete(&cma_dev->comp);
 460 }
 461
 462 static inline void release_mc(struct kref *kref)
 463 {
 464         struct cma_multicast *mc = container_of(kref, struct cma_multicast, mcref);
 465
 466         kfree(mc->multicast.ib);
 467         kfree(mc);
 468 }
 469
 470 static void cma_release_dev(struct rdma_id_private *id_priv)
 471 {
 472         mutex_lock(&lock);
 473         list_del(&id_priv->list);
 474         cma_deref_dev(id_priv->cma_dev);
 475         id_priv->cma_dev = NULL;
 476         mutex_unlock(&lock);
 477 }
 478
 479 static inline struct sockaddr *cma_src_addr(struct rdma_id_private *id_priv)
 480 {
 481         return (struct sockaddr *) &id_priv->id.route.addr.src_addr;
 482 }
 483
 484 static inline struct sockaddr *cma_dst_addr(struct rdma_id_private *id_priv)
 485 {
 486         return (struct sockaddr *) &id_priv->id.route.addr.dst_addr;
 487 }
 488
 489 static inline unsigned short cma_family(struct rdma_id_private *id_priv)
 490 {
 491         return id_priv->id.route.addr.src_addr.ss_family;
 492 }
 493
 494 static int cma_set_qkey(struct rdma_id_private *id_priv, u32 qkey)
 495 {
 496         struct ib_sa_mcmember_rec rec;
 497         int ret = 0;
 498
 499         if (id_priv->qkey) {
 500                 if (qkey && id_priv->qkey != qkey)
 501                         return -EINVAL;
 502                 return 0;
 503         }
 504
 505         if (qkey) {
 506                 id_priv->qkey = qkey;
 507                 return 0;
 508         }
 509
 510         switch (id_priv->id.ps) {
 511         case RDMA_PS_UDP:
 512         case RDMA_PS_IB:
 513                 id_priv->qkey = RDMA_UDP_QKEY;
 514                 break;
 515         case RDMA_PS_IPOIB:
 516                 ib_addr_get_mgid(&id_priv->id.route.addr.dev_addr, &rec.mgid);
 517                 ret = ib_sa_get_mcmember_rec(id_priv->id.device,
 518                                              id_priv->id.port_num, &rec.mgid,
 519                                              &rec);
 520                 if (!ret)
 521                         id_priv->qkey = be32_to_cpu(rec.qkey);
 522                 break;
 523         default:
 524                 break;
 525         }
 526         return ret;
 527 }
 528
 529 static void cma_translate_ib(struct sockaddr_ib *sib, struct rdma_dev_addr *dev_addr)
 530 {
 531         dev_addr->dev_type = ARPHRD_INFINIBAND;
 532         rdma_addr_set_sgid(dev_addr, (union ib_gid *) &sib->sib_addr);
 533         ib_addr_set_pkey(dev_addr, ntohs(sib->sib_pkey));
 534 }
 535
 536 static int cma_translate_addr(struct sockaddr *addr, struct rdma_dev_addr *dev_addr)
 537 {
 538         int ret;
 539
 540         if (addr->sa_family != AF_IB) {
 541                 ret = rdma_translate_ip(addr, dev_addr, NULL);
 542         } else {
 543                 cma_translate_ib((struct sockaddr_ib *) addr, dev_addr);
 544                 ret = 0;
 545         }
 546
 547         return ret;
 548 }
 549
 550 static inline int cma_validate_port(struct ib_device *device, u8 port,
 551                                     enum ib_gid_type gid_type,
 552                                       union ib_gid *gid, int dev_type,
 553                                       int bound_if_index)
 554 {
 555         int ret = -ENODEV;
 556         struct net_device *ndev = NULL;
 557
 558         if ((dev_type == ARPHRD_INFINIBAND) && !rdma_protocol_ib(device, port))
 559                 return ret;
 560
 561         if ((dev_type != ARPHRD_INFINIBAND) && rdma_protocol_ib(device, port))
 562                 return ret;
 563
 564         if (dev_type == ARPHRD_ETHER && rdma_protocol_roce(device, port)) {
 565                 ndev = dev_get_by_index(&init_net, bound_if_index);
 566                 if (ndev && ndev->flags & IFF_LOOPBACK) {
 567                         pr_info("detected loopback device\n");
 568                         dev_put(ndev);
 569
 570                         if (!device->get_netdev)
 571                                 return -EOPNOTSUPP;
 572
 573                         ndev = device->get_netdev(device, port);
 574                         if (!ndev)
 575                                 return -ENODEV;
 576                 }
 577         } else {
 578                 gid_type = IB_GID_TYPE_IB;
 579         }
 580
 581         ret = ib_find_cached_gid_by_port(device, gid, gid_type, port,
 582                                          ndev, NULL);
 583
 584         if (ndev)
 585                 dev_put(ndev);
 586
 587         return ret;
 588 }
 589
 590 static int cma_acquire_dev(struct rdma_id_private *id_priv,
 591                            struct rdma_id_private *listen_id_priv)
 592 {
 593         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
 594         struct cma_device *cma_dev;
 595         union ib_gid gid, iboe_gid, *gidp;
 596         int ret = -ENODEV;
 597         u8 port;
 598
 599         if (dev_addr->dev_type != ARPHRD_INFINIBAND &&
 600             id_priv->id.ps == RDMA_PS_IPOIB)
 601                 return -EINVAL;
 602
 603         mutex_lock(&lock);
 604         rdma_ip2gid((struct sockaddr *)&id_priv->id.route.addr.src_addr,
 605                     &iboe_gid);
 606
 607         memcpy(&gid, dev_addr->src_dev_addr +
 608                rdma_addr_gid_offset(dev_addr), sizeof gid);
 609
 610         if (listen_id_priv) {
 611                 cma_dev = listen_id_priv->cma_dev;
 612                 port = listen_id_priv->id.port_num;
 613                 gidp = rdma_protocol_roce(cma_dev->device, port) ?
 614                        &iboe_gid : &gid;
 615
 616                 ret = cma_validate_port(cma_dev->device, port,
 617                                         rdma_protocol_ib(cma_dev->device, port) ?
 618                                         IB_GID_TYPE_IB :
 619                                         listen_id_priv->gid_type, gidp,
 620                                         dev_addr->dev_type,
 621                                         dev_addr->bound_dev_if);
 622                 if (!ret) {
 623                         id_priv->id.port_num = port;
 624                         goto out;
 625                 }
 626         }
 627
 628         list_for_each_entry(cma_dev, &dev_list, list) {
 629                 for (port = 1; port <= cma_dev->device->phys_port_cnt; ++port) {
 630                         if (listen_id_priv &&
 631                             listen_id_priv->cma_dev == cma_dev &&
 632                             listen_id_priv->id.port_num == port)
 633                                 continue;
 634
 635                         gidp = rdma_protocol_roce(cma_dev->device, port) ?
 636                                &iboe_gid : &gid;
 637
 638                         ret = cma_validate_port(cma_dev->device, port,
 639                                                 rdma_protocol_ib(cma_dev->device, port) ?
 640                                                 IB_GID_TYPE_IB :
 641                                                 cma_dev->default_gid_type[port - 1],
 642                                                 gidp, dev_addr->dev_type,
 643                                                 dev_addr->bound_dev_if);
 644                         if (!ret) {
 645                                 id_priv->id.port_num = port;
 646                                 goto out;
 647                         }
 648                 }
 649         }
 650
 651 out:
 652         if (!ret)
 653                 cma_attach_to_dev(id_priv, cma_dev);
 654
 655         mutex_unlock(&lock);
 656         return ret;
 657 }
 658
 659 /*
 660  * Select the source IB device and address to reach the destination IB address.
 661  */
 662 static int cma_resolve_ib_dev(struct rdma_id_private *id_priv)
 663 {
 664         struct cma_device *cma_dev, *cur_dev;
 665         struct sockaddr_ib *addr;
 666         union ib_gid gid, sgid, *dgid;
 667         u16 pkey, index;
 668         u8 p;
 669         int i;
 670
 671         cma_dev = NULL;
 672         addr = (struct sockaddr_ib *) cma_dst_addr(id_priv);
 673         dgid = (union ib_gid *) &addr->sib_addr;
 674         pkey = ntohs(addr->sib_pkey);
 675
 676         list_for_each_entry(cur_dev, &dev_list, list) {
 677                 for (p = 1; p <= cur_dev->device->phys_port_cnt; ++p) {
 678                         if (!rdma_cap_af_ib(cur_dev->device, p))
 679                                 continue;
 680
 681                         if (ib_find_cached_pkey(cur_dev->device, p, pkey, &index))
 682                                 continue;
 683
 684                         for (i = 0; !ib_get_cached_gid(cur_dev->device, p, i,
 685                                                        &gid, NULL);
 686                              i++) {
 687                                 if (!memcmp(&gid, dgid, sizeof(gid))) {
 688                                         cma_dev = cur_dev;
 689                                         sgid = gid;
 690                                         id_priv->id.port_num = p;
 691                                         goto found;
 692                                 }
 693
 694                                 if (!cma_dev && (gid.global.subnet_prefix ==
 695                                                  dgid->global.subnet_prefix)) {
 696                                         cma_dev = cur_dev;
 697                                         sgid = gid;
 698                                         id_priv->id.port_num = p;
 699                                 }
 700                         }
 701                 }
 702         }
 703
 704         if (!cma_dev)
 705                 return -ENODEV;
 706
 707 found:
 708         cma_attach_to_dev(id_priv, cma_dev);
 709         addr = (struct sockaddr_ib *) cma_src_addr(id_priv);
 710         memcpy(&addr->sib_addr, &sgid, sizeof sgid);
 711         cma_translate_ib(addr, &id_priv->id.route.addr.dev_addr);
 712         return 0;
 713 }
 714
 715 static void cma_deref_id(struct rdma_id_private *id_priv)
 716 {
 717         if (atomic_dec_and_test(&id_priv->refcount))
 718                 complete(&id_priv->comp);
 719 }
 720
 721 struct rdma_cm_id *rdma_create_id(struct net *net,
 722                                   rdma_cm_event_handler event_handler,
 723                                   void *context, enum rdma_port_space ps,
 724                                   enum ib_qp_type qp_type)
 725 {
 726         struct rdma_id_private *id_priv;
 727
 728         id_priv = kzalloc(sizeof *id_priv, GFP_KERNEL);
 729         if (!id_priv)
 730                 return ERR_PTR(-ENOMEM);
 731
 732         id_priv->owner = task_pid_nr(current);
 733         id_priv->state = RDMA_CM_IDLE;
 734         id_priv->id.context = context;
 735         id_priv->id.event_handler = event_handler;
 736         id_priv->id.ps = ps;
 737         id_priv->id.qp_type = qp_type;
 738         spin_lock_init(&id_priv->lock);
 739         mutex_init(&id_priv->qp_mutex);
 740         init_completion(&id_priv->comp);
 741         atomic_set(&id_priv->refcount, 1);
 742         mutex_init(&id_priv->handler_mutex);
 743         INIT_LIST_HEAD(&id_priv->listen_list);
 744         INIT_LIST_HEAD(&id_priv->mc_list);
 745         get_random_bytes(&id_priv->seq_num, sizeof id_priv->seq_num);
 746         id_priv->id.route.addr.dev_addr.net = get_net(net);
 747
 748         return &id_priv->id;
 749 }
 750 EXPORT_SYMBOL(rdma_create_id);
 751
 752 static int cma_init_ud_qp(struct rdma_id_private *id_priv, struct ib_qp *qp)
 753 {
 754         struct ib_qp_attr qp_attr;
 755         int qp_attr_mask, ret;
 756
 757         qp_attr.qp_state = IB_QPS_INIT;
 758         ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
 759         if (ret)
 760                 return ret;
 761
 762         ret = ib_modify_qp(qp, &qp_attr, qp_attr_mask);
 763         if (ret)
 764                 return ret;
 765
 766         qp_attr.qp_state = IB_QPS_RTR;
 767         ret = ib_modify_qp(qp, &qp_attr, IB_QP_STATE);
 768         if (ret)
 769                 return ret;
 770
 771         qp_attr.qp_state = IB_QPS_RTS;
 772         qp_attr.sq_psn = 0;
 773         ret = ib_modify_qp(qp, &qp_attr, IB_QP_STATE | IB_QP_SQ_PSN);
 774
 775         return ret;
 776 }
 777
 778 static int cma_init_conn_qp(struct rdma_id_private *id_priv, struct ib_qp *qp)
 779 {
 780         struct ib_qp_attr qp_attr;
 781         int qp_attr_mask, ret;
 782
 783         qp_attr.qp_state = IB_QPS_INIT;
 784         ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
 785         if (ret)
 786                 return ret;
 787
 788         return ib_modify_qp(qp, &qp_attr, qp_attr_mask);
 789 }
 790
 791 int rdma_create_qp(struct rdma_cm_id *id, struct ib_pd *pd,
 792                    struct ib_qp_init_attr *qp_init_attr)
 793 {
 794         struct rdma_id_private *id_priv;
 795         struct ib_qp *qp;
 796         int ret;
 797
 798         id_priv = container_of(id, struct rdma_id_private, id);
 799         if (id->device != pd->device)
 800                 return -EINVAL;
 801
 802         qp_init_attr->port_num = id->port_num;
 803         qp = ib_create_qp(pd, qp_init_attr);
 804         if (IS_ERR(qp))
 805                 return PTR_ERR(qp);
 806
 807         if (id->qp_type == IB_QPT_UD)
 808                 ret = cma_init_ud_qp(id_priv, qp);
 809         else
 810                 ret = cma_init_conn_qp(id_priv, qp);
 811         if (ret)
 812                 goto err;
 813
 814         id->qp = qp;
 815         id_priv->qp_num = qp->qp_num;
 816         id_priv->srq = (qp->srq != NULL);
 817         return 0;
 818 err:
 819         ib_destroy_qp(qp);
 820         return ret;
 821 }
 822 EXPORT_SYMBOL(rdma_create_qp);
 823
 824 void rdma_destroy_qp(struct rdma_cm_id *id)
 825 {
 826         struct rdma_id_private *id_priv;
 827
 828         id_priv = container_of(id, struct rdma_id_private, id);
 829         mutex_lock(&id_priv->qp_mutex);
 830         ib_destroy_qp(id_priv->id.qp);
 831         id_priv->id.qp = NULL;
 832         mutex_unlock(&id_priv->qp_mutex);
 833 }
 834 EXPORT_SYMBOL(rdma_destroy_qp);
 835
 836 static int cma_modify_qp_rtr(struct rdma_id_private *id_priv,
 837                              struct rdma_conn_param *conn_param)
 838 {
 839         struct ib_qp_attr qp_attr;
 840         int qp_attr_mask, ret;
 841         union ib_gid sgid;
 842
 843         mutex_lock(&id_priv->qp_mutex);
 844         if (!id_priv->id.qp) {
 845                 ret = 0;
 846                 goto out;
 847         }
 848
 849         /* Need to update QP attributes from default values. */
 850         qp_attr.qp_state = IB_QPS_INIT;
 851         ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
 852         if (ret)
 853                 goto out;
 854
 855         ret = ib_modify_qp(id_priv->id.qp, &qp_attr, qp_attr_mask);
 856         if (ret)
 857                 goto out;
 858
 859         qp_attr.qp_state = IB_QPS_RTR;
 860         ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
 861         if (ret)
 862                 goto out;
 863
 864         ret = ib_query_gid(id_priv->id.device, id_priv->id.port_num,
 865                            qp_attr.ah_attr.grh.sgid_index, &sgid, NULL);
 866         if (ret)
 867                 goto out;
 868
 869         BUG_ON(id_priv->cma_dev->device != id_priv->id.device);
 870
 871         if (conn_param)
 872                 qp_attr.max_dest_rd_atomic = conn_param->responder_resources;
 873         ret = ib_modify_qp(id_priv->id.qp, &qp_attr, qp_attr_mask);
 874 out:
 875         mutex_unlock(&id_priv->qp_mutex);
 876         return ret;
 877 }
 878
 879 static int cma_modify_qp_rts(struct rdma_id_private *id_priv,
 880                              struct rdma_conn_param *conn_param)
 881 {
 882         struct ib_qp_attr qp_attr;
 883         int qp_attr_mask, ret;
 884
 885         mutex_lock(&id_priv->qp_mutex);
 886         if (!id_priv->id.qp) {
 887                 ret = 0;
 888                 goto out;
 889         }
 890
 891         qp_attr.qp_state = IB_QPS_RTS;
 892         ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
 893         if (ret)
 894                 goto out;
 895
 896         if (conn_param)
 897                 qp_attr.max_rd_atomic = conn_param->initiator_depth;
 898         ret = ib_modify_qp(id_priv->id.qp, &qp_attr, qp_attr_mask);
 899 out:
 900         mutex_unlock(&id_priv->qp_mutex);
 901         return ret;
 902 }
 903
 904 static int cma_modify_qp_err(struct rdma_id_private *id_priv)
 905 {
 906         struct ib_qp_attr qp_attr;
 907         int ret;
 908
 909         mutex_lock(&id_priv->qp_mutex);
 910         if (!id_priv->id.qp) {
 911                 ret = 0;
 912                 goto out;
 913         }
 914
 915         qp_attr.qp_state = IB_QPS_ERR;
 916         ret = ib_modify_qp(id_priv->id.qp, &qp_attr, IB_QP_STATE);
 917 out:
 918         mutex_unlock(&id_priv->qp_mutex);
 919         return ret;
 920 }
 921
 922 static int cma_ib_init_qp_attr(struct rdma_id_private *id_priv,
 923                                struct ib_qp_attr *qp_attr, int *qp_attr_mask)
 924 {
 925         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
 926         int ret;
 927         u16 pkey;
 928
 929         if (rdma_cap_eth_ah(id_priv->id.device, id_priv->id.port_num))
 930                 pkey = 0xffff;
 931         else
 932                 pkey = ib_addr_get_pkey(dev_addr);
 933
 934         ret = ib_find_cached_pkey(id_priv->id.device, id_priv->id.port_num,
 935                                   pkey, &qp_attr->pkey_index);
 936         if (ret)
 937                 return ret;
 938
 939         qp_attr->port_num = id_priv->id.port_num;
 940         *qp_attr_mask = IB_QP_STATE | IB_QP_PKEY_INDEX | IB_QP_PORT;
 941
 942         if (id_priv->id.qp_type == IB_QPT_UD) {
 943                 ret = cma_set_qkey(id_priv, 0);
 944                 if (ret)
 945                         return ret;
 946
 947                 qp_attr->qkey = id_priv->qkey;
 948                 *qp_attr_mask |= IB_QP_QKEY;
 949         } else {
 950                 qp_attr->qp_access_flags = 0;
 951                 *qp_attr_mask |= IB_QP_ACCESS_FLAGS;
 952         }
 953         return 0;
 954 }
 955
 956 int rdma_init_qp_attr(struct rdma_cm_id *id, struct ib_qp_attr *qp_attr,
 957                        int *qp_attr_mask)
 958 {
 959         struct rdma_id_private *id_priv;
 960         int ret = 0;
 961
 962         id_priv = container_of(id, struct rdma_id_private, id);
 963         if (rdma_cap_ib_cm(id->device, id->port_num)) {
 964                 if (!id_priv->cm_id.ib || (id_priv->id.qp_type == IB_QPT_UD))
 965                         ret = cma_ib_init_qp_attr(id_priv, qp_attr, qp_attr_mask);
 966                 else
 967                         ret = ib_cm_init_qp_attr(id_priv->cm_id.ib, qp_attr,
 968                                                  qp_attr_mask);
 969
 970                 if (qp_attr->qp_state == IB_QPS_RTR)
 971                         qp_attr->rq_psn = id_priv->seq_num;
 972         } else if (rdma_cap_iw_cm(id->device, id->port_num)) {
 973                 if (!id_priv->cm_id.iw) {
 974                         qp_attr->qp_access_flags = 0;
 975                         *qp_attr_mask = IB_QP_STATE | IB_QP_ACCESS_FLAGS;
 976                 } else
 977                         ret = iw_cm_init_qp_attr(id_priv->cm_id.iw, qp_attr,
 978                                                  qp_attr_mask);
 979         } else
 980                 ret = -ENOSYS;
 981
 982         return ret;
 983 }
 984 EXPORT_SYMBOL(rdma_init_qp_attr);
 985
 986 static inline int cma_zero_addr(struct sockaddr *addr)
 987 {
 988         switch (addr->sa_family) {
 989         case AF_INET:
 990                 return ipv4_is_zeronet(((struct sockaddr_in *)addr)->sin_addr.s_addr);
 991         case AF_INET6:
 992                 return ipv6_addr_any(&((struct sockaddr_in6 *) addr)->sin6_addr);
 993         case AF_IB:
 994                 return ib_addr_any(&((struct sockaddr_ib *) addr)->sib_addr);
 995         default:
 996                 return 0;
 997         }
 998 }
 999
1000 static inline int cma_loopback_addr(struct sockaddr *addr)
1001 {
1002         switch (addr->sa_family) {
1003         case AF_INET:
1004                 return ipv4_is_loopback(((struct sockaddr_in *) addr)->sin_addr.s_addr);
1005         case AF_INET6:
1006                 return ipv6_addr_loopback(&((struct sockaddr_in6 *) addr)->sin6_addr);
1007         case AF_IB:
1008                 return ib_addr_loopback(&((struct sockaddr_ib *) addr)->sib_addr);
1009         default:
1010                 return 0;
1011         }
1012 }
1013
1014 static inline int cma_any_addr(struct sockaddr *addr)
1015 {
1016         return cma_zero_addr(addr) || cma_loopback_addr(addr);
1017 }
1018
1019 static int cma_addr_cmp(struct sockaddr *src, struct sockaddr *dst)
1020 {
1021         if (src->sa_family != dst->sa_family)
1022                 return -1;
1023
1024         switch (src->sa_family) {
1025         case AF_INET:
1026                 return ((struct sockaddr_in *) src)->sin_addr.s_addr !=
1027                        ((struct sockaddr_in *) dst)->sin_addr.s_addr;
1028         case AF_INET6:
1029                 return ipv6_addr_cmp(&((struct sockaddr_in6 *) src)->sin6_addr,
1030                                      &((struct sockaddr_in6 *) dst)->sin6_addr);
1031         default:
1032                 return ib_addr_cmp(&((struct sockaddr_ib *) src)->sib_addr,
1033                                    &((struct sockaddr_ib *) dst)->sib_addr);
1034         }
1035 }
1036
1037 static __be16 cma_port(struct sockaddr *addr)
1038 {
1039         struct sockaddr_ib *sib;
1040
1041         switch (addr->sa_family) {
1042         case AF_INET:
1043                 return ((struct sockaddr_in *) addr)->sin_port;
1044         case AF_INET6:
1045                 return ((struct sockaddr_in6 *) addr)->sin6_port;
1046         case AF_IB:
1047                 sib = (struct sockaddr_ib *) addr;
1048                 return htons((u16) (be64_to_cpu(sib->sib_sid) &
1049                                     be64_to_cpu(sib->sib_sid_mask)));
1050         default:
1051                 return 0;
1052         }
1053 }
1054
1055 static inline int cma_any_port(struct sockaddr *addr)
1056 {
1057         return !cma_port(addr);
1058 }
1059
1060 static void cma_save_ib_info(struct sockaddr *src_addr,
1061                              struct sockaddr *dst_addr,
1062                              struct rdma_cm_id *listen_id,
1063                              struct ib_sa_path_rec *path)
1064 {
1065         struct sockaddr_ib *listen_ib, *ib;
1066
1067         listen_ib = (struct sockaddr_ib *) &listen_id->route.addr.src_addr;
1068         if (src_addr) {
1069                 ib = (struct sockaddr_ib *)src_addr;
1070                 ib->sib_family = AF_IB;
1071                 if (path) {
1072                         ib->sib_pkey = path->pkey;
1073                         ib->sib_flowinfo = path->flow_label;
1074                         memcpy(&ib->sib_addr, &path->sgid, 16);
1075                         ib->sib_sid = path->service_id;
1076                         ib->sib_scope_id = 0;
1077                 } else {
1078                         ib->sib_pkey = listen_ib->sib_pkey;
1079                         ib->sib_flowinfo = listen_ib->sib_flowinfo;
1080                         ib->sib_addr = listen_ib->sib_addr;
1081                         ib->sib_sid = listen_ib->sib_sid;
1082                         ib->sib_scope_id = listen_ib->sib_scope_id;
1083                 }
1084                 ib->sib_sid_mask = cpu_to_be64(0xffffffffffffffffULL);
1085         }
1086         if (dst_addr) {
1087                 ib = (struct sockaddr_ib *)dst_addr;
1088                 ib->sib_family = AF_IB;
1089                 if (path) {
1090                         ib->sib_pkey = path->pkey;
1091                         ib->sib_flowinfo = path->flow_label;
1092                         memcpy(&ib->sib_addr, &path->dgid, 16);
1093                 }
1094         }
1095 }
1096
1097 static void cma_save_ip4_info(struct sockaddr_in *src_addr,
1098                               struct sockaddr_in *dst_addr,
1099                               struct cma_hdr *hdr,
1100                               __be16 local_port)
1101 {
1102         if (src_addr) {
1103                 *src_addr = (struct sockaddr_in) {
1104                         .sin_family = AF_INET,
1105                         .sin_addr.s_addr = hdr->dst_addr.ip4.addr,
1106                         .sin_port = local_port,
1107                 };
1108         }
1109
1110         if (dst_addr) {
1111                 *dst_addr = (struct sockaddr_in) {
1112                         .sin_family = AF_INET,
1113                         .sin_addr.s_addr = hdr->src_addr.ip4.addr,
1114                         .sin_port = hdr->port,
1115                 };
1116         }
1117 }
1118
1119 static void cma_save_ip6_info(struct sockaddr_in6 *src_addr,
1120                               struct sockaddr_in6 *dst_addr,
1121                               struct cma_hdr *hdr,
1122                               __be16 local_port)
1123 {
1124         if (src_addr) {
1125                 *src_addr = (struct sockaddr_in6) {
1126                         .sin6_family = AF_INET6,
1127                         .sin6_addr = hdr->dst_addr.ip6,
1128                         .sin6_port = local_port,
1129                 };
1130         }
1131
1132         if (dst_addr) {
1133                 *dst_addr = (struct sockaddr_in6) {
1134                         .sin6_family = AF_INET6,
1135                         .sin6_addr = hdr->src_addr.ip6,
1136                         .sin6_port = hdr->port,
1137                 };
1138         }
1139 }
1140
1141 static u16 cma_port_from_service_id(__be64 service_id)
1142 {
1143         return (u16)be64_to_cpu(service_id);
1144 }
1145
1146 static int cma_save_ip_info(struct sockaddr *src_addr,
1147                             struct sockaddr *dst_addr,
1148                             struct ib_cm_event *ib_event,
1149                             __be64 service_id)
1150 {
1151         struct cma_hdr *hdr;
1152         __be16 port;
1153
1154         hdr = ib_event->private_data;
1155         if (hdr->cma_version != CMA_VERSION)
1156                 return -EINVAL;
1157
1158         port = htons(cma_port_from_service_id(service_id));
1159
1160         switch (cma_get_ip_ver(hdr)) {
1161         case 4:
1162                 cma_save_ip4_info((struct sockaddr_in *)src_addr,
1163                                   (struct sockaddr_in *)dst_addr, hdr, port);
1164                 break;
1165         case 6:
1166                 cma_save_ip6_info((struct sockaddr_in6 *)src_addr,
1167                                   (struct sockaddr_in6 *)dst_addr, hdr, port);
1168                 break;
1169         default:
1170                 return -EAFNOSUPPORT;
1171         }
1172
1173         return 0;
1174 }
1175
1176 static int cma_save_net_info(struct sockaddr *src_addr,
1177                              struct sockaddr *dst_addr,
1178                              struct rdma_cm_id *listen_id,
1179                              struct ib_cm_event *ib_event,
1180                              sa_family_t sa_family, __be64 service_id)
1181 {
1182         if (sa_family == AF_IB) {
1183                 if (ib_event->event == IB_CM_REQ_RECEIVED)
1184                         cma_save_ib_info(src_addr, dst_addr, listen_id,
1185                                          ib_event->param.req_rcvd.primary_path);
1186                 else if (ib_event->event == IB_CM_SIDR_REQ_RECEIVED)
1187                         cma_save_ib_info(src_addr, dst_addr, listen_id, NULL);
1188                 return 0;
1189         }
1190
1191         return cma_save_ip_info(src_addr, dst_addr, ib_event, service_id);
1192 }
1193
1194 static int cma_save_req_info(const struct ib_cm_event *ib_event,
1195                              struct cma_req_info *req)
1196 {
1197         const struct ib_cm_req_event_param *req_param =
1198                 &ib_event->param.req_rcvd;
1199         const struct ib_cm_sidr_req_event_param *sidr_param =
1200                 &ib_event->param.sidr_req_rcvd;
1201
1202         switch (ib_event->event) {
1203         case IB_CM_REQ_RECEIVED:
1204                 req->device     = req_param->listen_id->device;
1205                 req->port       = req_param->port;
1206                 memcpy(&req->local_gid, &req_param->primary_path->sgid,
1207                        sizeof(req->local_gid));
1208                 req->has_gid    = true;
1209                 req->service_id = req_param->primary_path->service_id;
1210                 req->pkey       = be16_to_cpu(req_param->primary_path->pkey);
1211                 if (req->pkey != req_param->bth_pkey)
1212                         pr_warn_ratelimited("RDMA CMA: got different BTH P_Key (0x%x) and primary path P_Key (0x%x)\n"
1213                                             "RDMA CMA: in the future this may cause the request to be dropped\n",
1214                                             req_param->bth_pkey, req->pkey);
1215                 break;
1216         case IB_CM_SIDR_REQ_RECEIVED:
1217                 req->device     = sidr_param->listen_id->device;
1218                 req->port       = sidr_param->port;
1219                 req->has_gid    = false;
1220                 req->service_id = sidr_param->service_id;
1221                 req->pkey       = sidr_param->pkey;
1222                 if (req->pkey != sidr_param->bth_pkey)
1223                         pr_warn_ratelimited("RDMA CMA: got different BTH P_Key (0x%x) and SIDR request payload P_Key (0x%x)\n"
1224                                             "RDMA CMA: in the future this may cause the request to be dropped\n",
1225                                             sidr_param->bth_pkey, req->pkey);
1226                 break;
1227         default:
1228                 return -EINVAL;
1229         }
1230
1231         return 0;
1232 }
1233
1234 static bool validate_ipv4_net_dev(struct net_device *net_dev,
1235                                   const struct sockaddr_in *dst_addr,
1236                                   const struct sockaddr_in *src_addr)
1237 {
1238         __be32 daddr = dst_addr->sin_addr.s_addr,
1239                saddr = src_addr->sin_addr.s_addr;
1240         struct fib_result res;
1241         struct flowi4 fl4;
1242         int err;
1243         bool ret;
1244
1245         if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1246             ipv4_is_lbcast(daddr) || ipv4_is_zeronet(saddr) ||
1247             ipv4_is_zeronet(daddr) || ipv4_is_loopback(daddr) ||
1248             ipv4_is_loopback(saddr))
1249                 return false;
1250
1251         memset(&fl4, 0, sizeof(fl4));
1252         fl4.flowi4_iif = net_dev->ifindex;
1253         fl4.daddr = daddr;
1254         fl4.saddr = saddr;
1255
1256         rcu_read_lock();
1257         err = fib_lookup(dev_net(net_dev), &fl4, &res, 0);
1258         ret = err == 0 && FIB_RES_DEV(res) == net_dev;
1259         rcu_read_unlock();
1260
1261         return ret;
1262 }
1263
1264 static bool validate_ipv6_net_dev(struct net_device *net_dev,
1265                                   const struct sockaddr_in6 *dst_addr,
1266                                   const struct sockaddr_in6 *src_addr)
1267 {
1268 #if IS_ENABLED(CONFIG_IPV6)
1269         const int strict = ipv6_addr_type(&dst_addr->sin6_addr) &
1270                            IPV6_ADDR_LINKLOCAL;
1271         struct rt6_info *rt = rt6_lookup(dev_net(net_dev), &dst_addr->sin6_addr,
1272                                          &src_addr->sin6_addr, net_dev->ifindex,
1273                                          strict);
1274         bool ret;
1275
1276         if (!rt)
1277                 return false;
1278
1279         ret = rt->rt6i_idev->dev == net_dev;
1280         ip6_rt_put(rt);
1281
1282         return ret;
1283 #else
1284         return false;
1285 #endif
1286 }
1287
1288 static bool validate_net_dev(struct net_device *net_dev,
1289                              const struct sockaddr *daddr,
1290                              const struct sockaddr *saddr)
1291 {
1292         const struct sockaddr_in *daddr4 = (const struct sockaddr_in *)daddr;
1293         const struct sockaddr_in *saddr4 = (const struct sockaddr_in *)saddr;
1294         const struct sockaddr_in6 *daddr6 = (const struct sockaddr_in6 *)daddr;
1295         const struct sockaddr_in6 *saddr6 = (const struct sockaddr_in6 *)saddr;
1296
1297         switch (daddr->sa_family) {
1298         case AF_INET:
1299                 return saddr->sa_family == AF_INET &&
1300                        validate_ipv4_net_dev(net_dev, daddr4, saddr4);
1301
1302         case AF_INET6:
1303                 return saddr->sa_family == AF_INET6 &&
1304                        validate_ipv6_net_dev(net_dev, daddr6, saddr6);
1305
1306         default:
1307                 return false;
1308         }
1309 }
1310
1311 static struct net_device *cma_get_net_dev(struct ib_cm_event *ib_event,
1312                                           const struct cma_req_info *req)
1313 {
1314         struct sockaddr_storage listen_addr_storage, src_addr_storage;
1315         struct sockaddr *listen_addr = (struct sockaddr *)&listen_addr_storage,
1316                         *src_addr = (struct sockaddr *)&src_addr_storage;
1317         struct net_device *net_dev;
1318         const union ib_gid *gid = req->has_gid ? &req->local_gid : NULL;
1319         int err;
1320
1321         err = cma_save_ip_info(listen_addr, src_addr, ib_event,
1322                                req->service_id);
1323         if (err)
1324                 return ERR_PTR(err);
1325
1326         net_dev = ib_get_net_dev_by_params(req->device, req->port, req->pkey,
1327                                            gid, listen_addr);
1328         if (!net_dev)
1329                 return ERR_PTR(-ENODEV);
1330
1331         if (!validate_net_dev(net_dev, listen_addr, src_addr)) {
1332                 dev_put(net_dev);
1333                 return ERR_PTR(-EHOSTUNREACH);
1334         }
1335
1336         return net_dev;
1337 }
1338
1339 static enum rdma_port_space rdma_ps_from_service_id(__be64 service_id)
1340 {
1341         return (be64_to_cpu(service_id) >> 16) & 0xffff;
1342 }
1343
1344 static bool cma_match_private_data(struct rdma_id_private *id_priv,
1345                                    const struct cma_hdr *hdr)
1346 {
1347         struct sockaddr *addr = cma_src_addr(id_priv);
1348         __be32 ip4_addr;
1349         struct in6_addr ip6_addr;
1350
1351         if (cma_any_addr(addr) && !id_priv->afonly)
1352                 return true;
1353
1354         switch (addr->sa_family) {
1355         case AF_INET:
1356                 ip4_addr = ((struct sockaddr_in *)addr)->sin_addr.s_addr;
1357                 if (cma_get_ip_ver(hdr) != 4)
1358                         return false;
1359                 if (!cma_any_addr(addr) &&
1360                     hdr->dst_addr.ip4.addr != ip4_addr)
1361                         return false;
1362                 break;
1363         case AF_INET6:
1364                 ip6_addr = ((struct sockaddr_in6 *)addr)->sin6_addr;
1365                 if (cma_get_ip_ver(hdr) != 6)
1366                         return false;
1367                 if (!cma_any_addr(addr) &&
1368                     memcmp(&hdr->dst_addr.ip6, &ip6_addr, sizeof(ip6_addr)))
1369                         return false;
1370                 break;
1371         case AF_IB:
1372                 return true;
1373         default:
1374                 return false;
1375         }
1376
1377         return true;
1378 }
1379
1380 static bool cma_protocol_roce_dev_port(struct ib_device *device, int port_num)
1381 {
1382         enum rdma_link_layer ll = rdma_port_get_link_layer(device, port_num);
1383         enum rdma_transport_type transport =
1384                 rdma_node_get_transport(device->node_type);
1385
1386         return ll == IB_LINK_LAYER_ETHERNET && transport == RDMA_TRANSPORT_IB;
1387 }
1388
1389 static bool cma_protocol_roce(const struct rdma_cm_id *id)
1390 {
1391         struct ib_device *device = id->device;
1392         const int port_num = id->port_num ?: rdma_start_port(device);
1393
1394         return cma_protocol_roce_dev_port(device, port_num);
1395 }
1396
1397 static bool cma_match_net_dev(const struct rdma_cm_id *id,
1398                               const struct net_device *net_dev,
1399                               u8 port_num)
1400 {
1401         const struct rdma_addr *addr = &id->route.addr;
1402
1403         if (!net_dev)
1404                 /* This request is an AF_IB request or a RoCE request */
1405                 return (!id->port_num || id->port_num == port_num) &&
1406                        (addr->src_addr.ss_family == AF_IB ||
1407                         cma_protocol_roce_dev_port(id->device, port_num));
1408
1409         return !addr->dev_addr.bound_dev_if ||
1410                (net_eq(dev_net(net_dev), addr->dev_addr.net) &&
1411                 addr->dev_addr.bound_dev_if == net_dev->ifindex);
1412 }
1413
1414 static struct rdma_id_private *cma_find_listener(
1415                 const struct rdma_bind_list *bind_list,
1416                 const struct ib_cm_id *cm_id,
1417                 const struct ib_cm_event *ib_event,
1418                 const struct cma_req_info *req,
1419                 const struct net_device *net_dev)
1420 {
1421         struct rdma_id_private *id_priv, *id_priv_dev;
1422
1423         if (!bind_list)
1424                 return ERR_PTR(-EINVAL);
1425
1426         hlist_for_each_entry(id_priv, &bind_list->owners, node) {
1427                 if (cma_match_private_data(id_priv, ib_event->private_data)) {
1428                         if (id_priv->id.device == cm_id->device &&
1429                             cma_match_net_dev(&id_priv->id, net_dev, req->port))
1430                                 return id_priv;
1431                         list_for_each_entry(id_priv_dev,
1432                                             &id_priv->listen_list,
1433                                             listen_list) {
1434                                 if (id_priv_dev->id.device == cm_id->device &&
1435                                     cma_match_net_dev(&id_priv_dev->id, net_dev, req->port))
1436                                         return id_priv_dev;
1437                         }
1438                 }
1439         }
1440
1441         return ERR_PTR(-EINVAL);
1442 }
1443
1444 static struct rdma_id_private *cma_id_from_event(struct ib_cm_id *cm_id,
1445                                                  struct ib_cm_event *ib_event,
1446                                                  struct net_device **net_dev)
1447 {
1448         struct cma_req_info req;
1449         struct rdma_bind_list *bind_list;
1450         struct rdma_id_private *id_priv;
1451         int err;
1452
1453         err = cma_save_req_info(ib_event, &req);
1454         if (err)
1455                 return ERR_PTR(err);
1456
1457         *net_dev = cma_get_net_dev(ib_event, &req);
1458         if (IS_ERR(*net_dev)) {
1459                 if (PTR_ERR(*net_dev) == -EAFNOSUPPORT) {
1460                         /* Assuming the protocol is AF_IB */
1461                         *net_dev = NULL;
1462                 } else if (cma_protocol_roce_dev_port(req.device, req.port)) {
1463                         /* TODO find the net dev matching the request parameters
1464                          * through the RoCE GID table */
1465                         *net_dev = NULL;
1466                 } else {
1467                         return ERR_CAST(*net_dev);
1468                 }
1469         }
1470
1471         bind_list = cma_ps_find(*net_dev ? dev_net(*net_dev) : &init_net,
1472                                 rdma_ps_from_service_id(req.service_id),
1473                                 cma_port_from_service_id(req.service_id));
1474         id_priv = cma_find_listener(bind_list, cm_id, ib_event, &req, *net_dev);
1475         if (IS_ERR(id_priv) && *net_dev) {
1476                 dev_put(*net_dev);
1477                 *net_dev = NULL;
1478         }
1479
1480         return id_priv;
1481 }
1482
1483 static inline int cma_user_data_offset(struct rdma_id_private *id_priv)
1484 {
1485         return cma_family(id_priv) == AF_IB ? 0 : sizeof(struct cma_hdr);
1486 }
1487
1488 static void cma_cancel_route(struct rdma_id_private *id_priv)
1489 {
1490         if (rdma_cap_ib_sa(id_priv->id.device, id_priv->id.port_num)) {
1491                 if (id_priv->query)
1492                         ib_sa_cancel_query(id_priv->query_id, id_priv->query);
1493         }
1494 }
1495
1496 static void cma_cancel_listens(struct rdma_id_private *id_priv)
1497 {
1498         struct rdma_id_private *dev_id_priv;
1499
1500         /*
1501          * Remove from listen_any_list to prevent added devices from spawning
1502          * additional listen requests.
1503          */
1504         mutex_lock(&lock);
1505         list_del(&id_priv->list);
1506
1507         while (!list_empty(&id_priv->listen_list)) {
1508                 dev_id_priv = list_entry(id_priv->listen_list.next,
1509                                          struct rdma_id_private, listen_list);
1510                 /* sync with device removal to avoid duplicate destruction */
1511                 list_del_init(&dev_id_priv->list);
1512                 list_del(&dev_id_priv->listen_list);
1513                 mutex_unlock(&lock);
1514
1515                 rdma_destroy_id(&dev_id_priv->id);
1516                 mutex_lock(&lock);
1517         }
1518         mutex_unlock(&lock);
1519 }
1520
1521 static void cma_cancel_operation(struct rdma_id_private *id_priv,
1522                                  enum rdma_cm_state state)
1523 {
1524         switch (state) {
1525         case RDMA_CM_ADDR_QUERY:
1526                 rdma_addr_cancel(&id_priv->id.route.addr.dev_addr);
1527                 break;
1528         case RDMA_CM_ROUTE_QUERY:
1529                 cma_cancel_route(id_priv);
1530                 break;
1531         case RDMA_CM_LISTEN:
1532                 if (cma_any_addr(cma_src_addr(id_priv)) && !id_priv->cma_dev)
1533                         cma_cancel_listens(id_priv);
1534                 break;
1535         default:
1536                 break;
1537         }
1538 }
1539
1540 static void cma_release_port(struct rdma_id_private *id_priv)
1541 {
1542         struct rdma_bind_list *bind_list = id_priv->bind_list;
1543         struct net *net = id_priv->id.route.addr.dev_addr.net;
1544
1545         if (!bind_list)
1546                 return;
1547
1548         mutex_lock(&lock);
1549         hlist_del(&id_priv->node);
1550         if (hlist_empty(&bind_list->owners)) {
1551                 cma_ps_remove(net, bind_list->ps, bind_list->port);
1552                 kfree(bind_list);
1553         }
1554         mutex_unlock(&lock);
1555 }
1556
1557 static void cma_leave_mc_groups(struct rdma_id_private *id_priv)
1558 {
1559         struct cma_multicast *mc;
1560
1561         while (!list_empty(&id_priv->mc_list)) {
1562                 mc = container_of(id_priv->mc_list.next,
1563                                   struct cma_multicast, list);
1564                 list_del(&mc->list);
1565                 if (rdma_cap_ib_mcast(id_priv->cma_dev->device,
1566                                       id_priv->id.port_num)) {
1567                         ib_sa_free_multicast(mc->multicast.ib);
1568                         kfree(mc);
1569                 } else {
1570                         if (mc->igmp_joined) {
1571                                 struct rdma_dev_addr *dev_addr =
1572                                         &id_priv->id.route.addr.dev_addr;
1573                                 struct net_device *ndev = NULL;
1574
1575                                 if (dev_addr->bound_dev_if)
1576                                         ndev = dev_get_by_index(&init_net,
1577                                                                 dev_addr->bound_dev_if);
1578                                 if (ndev) {
1579                                         cma_igmp_send(ndev,
1580                                                       &mc->multicast.ib->rec.mgid,
1581                                                       false);
1582                                         dev_put(ndev);
1583                                 }
1584                         }
1585                         kref_put(&mc->mcref, release_mc);
1586                 }
1587         }
1588 }
1589
1590 void rdma_destroy_id(struct rdma_cm_id *id)
1591 {
1592         struct rdma_id_private *id_priv;
1593         enum rdma_cm_state state;
1594
1595         id_priv = container_of(id, struct rdma_id_private, id);
1596         state = cma_exch(id_priv, RDMA_CM_DESTROYING);
1597         cma_cancel_operation(id_priv, state);
1598
1599         /*
1600          * Wait for any active callback to finish.  New callbacks will find
1601          * the id_priv state set to destroying and abort.
1602          */
1603         mutex_lock(&id_priv->handler_mutex);
1604         mutex_unlock(&id_priv->handler_mutex);
1605
1606         if (id_priv->cma_dev) {
1607                 if (rdma_cap_ib_cm(id_priv->id.device, 1)) {
1608                         if (id_priv->cm_id.ib)
1609                                 ib_destroy_cm_id(id_priv->cm_id.ib);
1610                 } else if (rdma_cap_iw_cm(id_priv->id.device, 1)) {
1611                         if (id_priv->cm_id.iw)
1612                                 iw_destroy_cm_id(id_priv->cm_id.iw);
1613                 }
1614                 cma_leave_mc_groups(id_priv);
1615                 cma_release_dev(id_priv);
1616         }
1617
1618         cma_release_port(id_priv);
1619         cma_deref_id(id_priv);
1620         wait_for_completion(&id_priv->comp);
1621
1622         if (id_priv->internal_id)
1623                 cma_deref_id(id_priv->id.context);
1624
1625         kfree(id_priv->id.route.path_rec);
1626         put_net(id_priv->id.route.addr.dev_addr.net);
1627         kfree(id_priv);
1628 }
1629 EXPORT_SYMBOL(rdma_destroy_id);
1630
1631 static int cma_rep_recv(struct rdma_id_private *id_priv)
1632 {
1633         int ret;
1634
1635         ret = cma_modify_qp_rtr(id_priv, NULL);
1636         if (ret)
1637                 goto reject;
1638
1639         ret = cma_modify_qp_rts(id_priv, NULL);
1640         if (ret)
1641                 goto reject;
1642
1643         ret = ib_send_cm_rtu(id_priv->cm_id.ib, NULL, 0);
1644         if (ret)
1645                 goto reject;
1646
1647         return 0;
1648 reject:
1649         cma_modify_qp_err(id_priv);
1650         ib_send_cm_rej(id_priv->cm_id.ib, IB_CM_REJ_CONSUMER_DEFINED,
1651                        NULL, 0, NULL, 0);
1652         return ret;
1653 }
1654
1655 static void cma_set_rep_event_data(struct rdma_cm_event *event,
1656                                    struct ib_cm_rep_event_param *rep_data,
1657                                    void *private_data)
1658 {
1659         event->param.conn.private_data = private_data;
1660         event->param.conn.private_data_len = IB_CM_REP_PRIVATE_DATA_SIZE;
1661         event->param.conn.responder_resources = rep_data->responder_resources;
1662         event->param.conn.initiator_depth = rep_data->initiator_depth;
1663         event->param.conn.flow_control = rep_data->flow_control;
1664         event->param.conn.rnr_retry_count = rep_data->rnr_retry_count;
1665         event->param.conn.srq = rep_data->srq;
1666         event->param.conn.qp_num = rep_data->remote_qpn;
1667 }
1668
1669 static int cma_ib_handler(struct ib_cm_id *cm_id, struct ib_cm_event *ib_event)
1670 {
1671         struct rdma_id_private *id_priv = cm_id->context;
1672         struct rdma_cm_event event;
1673         int ret = 0;
1674
1675         mutex_lock(&id_priv->handler_mutex);
1676         if ((ib_event->event != IB_CM_TIMEWAIT_EXIT &&
1677              id_priv->state != RDMA_CM_CONNECT) ||
1678             (ib_event->event == IB_CM_TIMEWAIT_EXIT &&
1679              id_priv->state != RDMA_CM_DISCONNECT))
1680                 goto out;
1681
1682         memset(&event, 0, sizeof event);
1683         switch (ib_event->event) {
1684         case IB_CM_REQ_ERROR:
1685         case IB_CM_REP_ERROR:
1686                 event.event = RDMA_CM_EVENT_UNREACHABLE;
1687                 event.status = -ETIMEDOUT;
1688                 break;
1689         case IB_CM_REP_RECEIVED:
1690                 if (id_priv->id.qp) {
1691                         event.status = cma_rep_recv(id_priv);
1692                         event.event = event.status ? RDMA_CM_EVENT_CONNECT_ERROR :
1693                                                      RDMA_CM_EVENT_ESTABLISHED;
1694                 } else {
1695                         event.event = RDMA_CM_EVENT_CONNECT_RESPONSE;
1696                 }
1697                 cma_set_rep_event_data(&event, &ib_event->param.rep_rcvd,
1698                                        ib_event->private_data);
1699                 break;
1700         case IB_CM_RTU_RECEIVED:
1701         case IB_CM_USER_ESTABLISHED:
1702                 event.event = RDMA_CM_EVENT_ESTABLISHED;
1703                 break;
1704         case IB_CM_DREQ_ERROR:
1705                 event.status = -ETIMEDOUT; /* fall through */
1706         case IB_CM_DREQ_RECEIVED:
1707         case IB_CM_DREP_RECEIVED:
1708                 if (!cma_comp_exch(id_priv, RDMA_CM_CONNECT,
1709                                    RDMA_CM_DISCONNECT))
1710                         goto out;
1711                 event.event = RDMA_CM_EVENT_DISCONNECTED;
1712                 break;
1713         case IB_CM_TIMEWAIT_EXIT:
1714                 event.event = RDMA_CM_EVENT_TIMEWAIT_EXIT;
1715                 break;
1716         case IB_CM_MRA_RECEIVED:
1717                 /* ignore event */
1718                 goto out;
1719         case IB_CM_REJ_RECEIVED:
1720                 cma_modify_qp_err(id_priv);
1721                 event.status = ib_event->param.rej_rcvd.reason;
1722                 event.event = RDMA_CM_EVENT_REJECTED;
1723                 event.param.conn.private_data = ib_event->private_data;
1724                 event.param.conn.private_data_len = IB_CM_REJ_PRIVATE_DATA_SIZE;
1725                 break;
1726         default:
1727                 pr_err("RDMA CMA: unexpected IB CM event: %d\n",
1728                        ib_event->event);
1729                 goto out;
1730         }
1731
1732         ret = id_priv->id.event_handler(&id_priv->id, &event);
1733         if (ret) {
1734                 /* Destroy the CM ID by returning a non-zero value. */
1735                 id_priv->cm_id.ib = NULL;
1736                 cma_exch(id_priv, RDMA_CM_DESTROYING);
1737                 mutex_unlock(&id_priv->handler_mutex);
1738                 rdma_destroy_id(&id_priv->id);
1739                 return ret;
1740         }
1741 out:
1742         mutex_unlock(&id_priv->handler_mutex);
1743         return ret;
1744 }
1745
1746 static struct rdma_id_private *cma_new_conn_id(struct rdma_cm_id *listen_id,
1747                                                struct ib_cm_event *ib_event,
1748                                                struct net_device *net_dev)
1749 {
1750         struct rdma_id_private *id_priv;
1751         struct rdma_cm_id *id;
1752         struct rdma_route *rt;
1753         const sa_family_t ss_family = listen_id->route.addr.src_addr.ss_family;
1754         const __be64 service_id =
1755                       ib_event->param.req_rcvd.primary_path->service_id;
1756         int ret;
1757
1758         id = rdma_create_id(listen_id->route.addr.dev_addr.net,
1759                             listen_id->event_handler, listen_id->context,
1760                             listen_id->ps, ib_event->param.req_rcvd.qp_type);
1761         if (IS_ERR(id))
1762                 return NULL;
1763
1764         id_priv = container_of(id, struct rdma_id_private, id);
1765         if (cma_save_net_info((struct sockaddr *)&id->route.addr.src_addr,
1766                               (struct sockaddr *)&id->route.addr.dst_addr,
1767                               listen_id, ib_event, ss_family, service_id))
1768                 goto err;
1769
1770         rt = &id->route;
1771         rt->num_paths = ib_event->param.req_rcvd.alternate_path ? 2 : 1;
1772         rt->path_rec = kmalloc(sizeof *rt->path_rec * rt->num_paths,
1773                                GFP_KERNEL);
1774         if (!rt->path_rec)
1775                 goto err;
1776
1777         rt->path_rec[0] = *ib_event->param.req_rcvd.primary_path;
1778         if (rt->num_paths == 2)
1779                 rt->path_rec[1] = *ib_event->param.req_rcvd.alternate_path;
1780
1781         if (net_dev) {
1782                 ret = rdma_copy_addr(&rt->addr.dev_addr, net_dev, NULL);
1783                 if (ret)
1784                         goto err;
1785         } else {
1786                 if (!cma_protocol_roce(listen_id) &&
1787                     cma_any_addr(cma_src_addr(id_priv))) {
1788                         rt->addr.dev_addr.dev_type = ARPHRD_INFINIBAND;
1789                         rdma_addr_set_sgid(&rt->addr.dev_addr, &rt->path_rec[0].sgid);
1790                         ib_addr_set_pkey(&rt->addr.dev_addr, be16_to_cpu(rt->path_rec[0].pkey));
1791                 } else if (!cma_any_addr(cma_src_addr(id_priv))) {
1792                         ret = cma_translate_addr(cma_src_addr(id_priv), &rt->addr.dev_addr);
1793                         if (ret)
1794                                 goto err;
1795                 }
1796         }
1797         rdma_addr_set_dgid(&rt->addr.dev_addr, &rt->path_rec[0].dgid);
1798
1799         id_priv->state = RDMA_CM_CONNECT;
1800         return id_priv;
1801
1802 err:
1803         rdma_destroy_id(id);
1804         return NULL;
1805 }
1806
1807 static struct rdma_id_private *cma_new_udp_id(struct rdma_cm_id *listen_id,
1808                                               struct ib_cm_event *ib_event,
1809                                               struct net_device *net_dev)
1810 {
1811         struct rdma_id_private *id_priv;
1812         struct rdma_cm_id *id;
1813         const sa_family_t ss_family = listen_id->route.addr.src_addr.ss_family;
1814         struct net *net = listen_id->route.addr.dev_addr.net;
1815         int ret;
1816
1817         id = rdma_create_id(net, listen_id->event_handler, listen_id->context,
1818                             listen_id->ps, IB_QPT_UD);
1819         if (IS_ERR(id))
1820                 return NULL;
1821
1822         id_priv = container_of(id, struct rdma_id_private, id);
1823         if (cma_save_net_info((struct sockaddr *)&id->route.addr.src_addr,
1824                               (struct sockaddr *)&id->route.addr.dst_addr,
1825                               listen_id, ib_event, ss_family,
1826                               ib_event->param.sidr_req_rcvd.service_id))
1827                 goto err;
1828
1829         if (net_dev) {
1830                 ret = rdma_copy_addr(&id->route.addr.dev_addr, net_dev, NULL);
1831                 if (ret)
1832                         goto err;
1833         } else {
1834                 if (!cma_any_addr(cma_src_addr(id_priv))) {
1835                         ret = cma_translate_addr(cma_src_addr(id_priv),
1836                                                  &id->route.addr.dev_addr);
1837                         if (ret)
1838                                 goto err;
1839                 }
1840         }
1841
1842         id_priv->state = RDMA_CM_CONNECT;
1843         return id_priv;
1844 err:
1845         rdma_destroy_id(id);
1846         return NULL;
1847 }
1848
1849 static void cma_set_req_event_data(struct rdma_cm_event *event,
1850                                    struct ib_cm_req_event_param *req_data,
1851                                    void *private_data, int offset)
1852 {
1853         event->param.conn.private_data = private_data + offset;
1854         event->param.conn.private_data_len = IB_CM_REQ_PRIVATE_DATA_SIZE - offset;
1855         event->param.conn.responder_resources = req_data->responder_resources;
1856         event->param.conn.initiator_depth = req_data->initiator_depth;
1857         event->param.conn.flow_control = req_data->flow_control;
1858         event->param.conn.retry_count = req_data->retry_count;
1859         event->param.conn.rnr_retry_count = req_data->rnr_retry_count;
1860         event->param.conn.srq = req_data->srq;
1861         event->param.conn.qp_num = req_data->remote_qpn;
1862 }
1863
1864 static int cma_check_req_qp_type(struct rdma_cm_id *id, struct ib_cm_event *ib_event)
1865 {
1866         return (((ib_event->event == IB_CM_REQ_RECEIVED) &&
1867                  (ib_event->param.req_rcvd.qp_type == id->qp_type)) ||
1868                 ((ib_event->event == IB_CM_SIDR_REQ_RECEIVED) &&
1869                  (id->qp_type == IB_QPT_UD)) ||
1870                 (!id->qp_type));
1871 }
1872
1873 static int cma_req_handler(struct ib_cm_id *cm_id, struct ib_cm_event *ib_event)
1874 {
1875         struct rdma_id_private *listen_id, *conn_id = NULL;
1876         struct rdma_cm_event event;
1877         struct net_device *net_dev;
1878         int offset, ret;
1879
1880         listen_id = cma_id_from_event(cm_id, ib_event, &net_dev);
1881         if (IS_ERR(listen_id))
1882                 return PTR_ERR(listen_id);
1883
1884         if (!cma_check_req_qp_type(&listen_id->id, ib_event)) {
1885                 ret = -EINVAL;
1886                 goto net_dev_put;
1887         }
1888
1889         mutex_lock(&listen_id->handler_mutex);
1890         if (listen_id->state != RDMA_CM_LISTEN) {
1891                 ret = -ECONNABORTED;
1892                 goto err1;
1893         }
1894
1895         memset(&event, 0, sizeof event);
1896         offset = cma_user_data_offset(listen_id);
1897         event.event = RDMA_CM_EVENT_CONNECT_REQUEST;
1898         if (ib_event->event == IB_CM_SIDR_REQ_RECEIVED) {
1899                 conn_id = cma_new_udp_id(&listen_id->id, ib_event, net_dev);
1900                 event.param.ud.private_data = ib_event->private_data + offset;
1901                 event.param.ud.private_data_len =
1902                                 IB_CM_SIDR_REQ_PRIVATE_DATA_SIZE - offset;
1903         } else {
1904                 conn_id = cma_new_conn_id(&listen_id->id, ib_event, net_dev);
1905                 cma_set_req_event_data(&event, &ib_event->param.req_rcvd,
1906                                        ib_event->private_data, offset);
1907         }
1908         if (!conn_id) {
1909                 ret = -ENOMEM;
1910                 goto err1;
1911         }
1912
1913         mutex_lock_nested(&conn_id->handler_mutex, SINGLE_DEPTH_NESTING);
1914         ret = cma_acquire_dev(conn_id, listen_id);
1915         if (ret)
1916                 goto err2;
1917
1918         conn_id->cm_id.ib = cm_id;
1919         cm_id->context = conn_id;
1920         cm_id->cm_handler = cma_ib_handler;
1921
1922         /*
1923          * Protect against the user destroying conn_id from another thread
1924          * until we're done accessing it.
1925          */
1926         atomic_inc(&conn_id->refcount);
1927         ret = conn_id->id.event_handler(&conn_id->id, &event);
1928         if (ret)
1929                 goto err3;
1930         /*
1931          * Acquire mutex to prevent user executing rdma_destroy_id()
1932          * while we're accessing the cm_id.
1933          */
1934         mutex_lock(&lock);
1935         if (cma_comp(conn_id, RDMA_CM_CONNECT) &&
1936             (conn_id->id.qp_type != IB_QPT_UD))
1937                 ib_send_cm_mra(cm_id, CMA_CM_MRA_SETTING, NULL, 0);
1938         mutex_unlock(&lock);
1939         mutex_unlock(&conn_id->handler_mutex);
1940         mutex_unlock(&listen_id->handler_mutex);
1941         cma_deref_id(conn_id);
1942         if (net_dev)
1943                 dev_put(net_dev);
1944         return 0;
1945
1946 err3:
1947         cma_deref_id(conn_id);
1948         /* Destroy the CM ID by returning a non-zero value. */
1949         conn_id->cm_id.ib = NULL;
1950 err2:
1951         cma_exch(conn_id, RDMA_CM_DESTROYING);
1952         mutex_unlock(&conn_id->handler_mutex);
1953 err1:
1954         mutex_unlock(&listen_id->handler_mutex);
1955         if (conn_id)
1956                 rdma_destroy_id(&conn_id->id);
1957
1958 net_dev_put:
1959         if (net_dev)
1960                 dev_put(net_dev);
1961
1962         return ret;
1963 }
1964
1965 __be64 rdma_get_service_id(struct rdma_cm_id *id, struct sockaddr *addr)
1966 {
1967         if (addr->sa_family == AF_IB)
1968                 return ((struct sockaddr_ib *) addr)->sib_sid;
1969
1970         return cpu_to_be64(((u64)id->ps << 16) + be16_to_cpu(cma_port(addr)));
1971 }
1972 EXPORT_SYMBOL(rdma_get_service_id);
1973
1974 static int cma_iw_handler(struct iw_cm_id *iw_id, struct iw_cm_event *iw_event)
1975 {
1976         struct rdma_id_private *id_priv = iw_id->context;
1977         struct rdma_cm_event event;
1978         int ret = 0;
1979         struct sockaddr *laddr = (struct sockaddr *)&iw_event->local_addr;
1980         struct sockaddr *raddr = (struct sockaddr *)&iw_event->remote_addr;
1981
1982         mutex_lock(&id_priv->handler_mutex);
1983         if (id_priv->state != RDMA_CM_CONNECT)
1984                 goto out;
1985
1986         memset(&event, 0, sizeof event);
1987         switch (iw_event->event) {
1988         case IW_CM_EVENT_CLOSE:
1989                 event.event = RDMA_CM_EVENT_DISCONNECTED;
1990                 break;
1991         case IW_CM_EVENT_CONNECT_REPLY:
1992                 memcpy(cma_src_addr(id_priv), laddr,
1993                        rdma_addr_size(laddr));
1994                 memcpy(cma_dst_addr(id_priv), raddr,
1995                        rdma_addr_size(raddr));
1996                 switch (iw_event->status) {
1997                 case 0:
1998                         event.event = RDMA_CM_EVENT_ESTABLISHED;
1999                         event.param.conn.initiator_depth = iw_event->ird;
2000                         event.param.conn.responder_resources = iw_event->ord;
2001                         break;
2002                 case -ECONNRESET:
2003                 case -ECONNREFUSED:
2004                         event.event = RDMA_CM_EVENT_REJECTED;
2005                         break;
2006                 case -ETIMEDOUT:
2007                         event.event = RDMA_CM_EVENT_UNREACHABLE;
2008                         break;
2009                 default:
2010                         event.event = RDMA_CM_EVENT_CONNECT_ERROR;
2011                         break;
2012                 }
2013                 break;
2014         case IW_CM_EVENT_ESTABLISHED:
2015                 event.event = RDMA_CM_EVENT_ESTABLISHED;
2016                 event.param.conn.initiator_depth = iw_event->ird;
2017                 event.param.conn.responder_resources = iw_event->ord;
2018                 break;
2019         default:
2020                 BUG_ON(1);
2021         }
2022
2023         event.status = iw_event->status;
2024         event.param.conn.private_data = iw_event->private_data;
2025         event.param.conn.private_data_len = iw_event->private_data_len;
2026         ret = id_priv->id.event_handler(&id_priv->id, &event);
2027         if (ret) {
2028                 /* Destroy the CM ID by returning a non-zero value. */
2029                 id_priv->cm_id.iw = NULL;
2030                 cma_exch(id_priv, RDMA_CM_DESTROYING);
2031                 mutex_unlock(&id_priv->handler_mutex);
2032                 rdma_destroy_id(&id_priv->id);
2033                 return ret;
2034         }
2035
2036 out:
2037         mutex_unlock(&id_priv->handler_mutex);
2038         return ret;
2039 }
2040
2041 static int iw_conn_req_handler(struct iw_cm_id *cm_id,
2042                                struct iw_cm_event *iw_event)
2043 {
2044         struct rdma_cm_id *new_cm_id;
2045         struct rdma_id_private *listen_id, *conn_id;
2046         struct rdma_cm_event event;
2047         int ret = -ECONNABORTED;
2048         struct sockaddr *laddr = (struct sockaddr *)&iw_event->local_addr;
2049         struct sockaddr *raddr = (struct sockaddr *)&iw_event->remote_addr;
2050
2051         listen_id = cm_id->context;
2052
2053         mutex_lock(&listen_id->handler_mutex);
2054         if (listen_id->state != RDMA_CM_LISTEN)
2055                 goto out;
2056
2057         /* Create a new RDMA id for the new IW CM ID */
2058         new_cm_id = rdma_create_id(listen_id->id.route.addr.dev_addr.net,
2059                                    listen_id->id.event_handler,
2060                                    listen_id->id.context,
2061                                    RDMA_PS_TCP, IB_QPT_RC);
2062         if (IS_ERR(new_cm_id)) {
2063                 ret = -ENOMEM;
2064                 goto out;
2065         }
2066         conn_id = container_of(new_cm_id, struct rdma_id_private, id);
2067         mutex_lock_nested(&conn_id->handler_mutex, SINGLE_DEPTH_NESTING);
2068         conn_id->state = RDMA_CM_CONNECT;
2069
2070         ret = rdma_translate_ip(laddr, &conn_id->id.route.addr.dev_addr, NULL);
2071         if (ret) {
2072                 mutex_unlock(&conn_id->handler_mutex);
2073                 rdma_destroy_id(new_cm_id);
2074                 goto out;
2075         }
2076
2077         ret = cma_acquire_dev(conn_id, listen_id);
2078         if (ret) {
2079                 mutex_unlock(&conn_id->handler_mutex);
2080                 rdma_destroy_id(new_cm_id);
2081                 goto out;
2082         }
2083
2084         conn_id->cm_id.iw = cm_id;
2085         cm_id->context = conn_id;
2086         cm_id->cm_handler = cma_iw_handler;
2087
2088         memcpy(cma_src_addr(conn_id), laddr, rdma_addr_size(laddr));
2089         memcpy(cma_dst_addr(conn_id), raddr, rdma_addr_size(raddr));
2090
2091         memset(&event, 0, sizeof event);
2092         event.event = RDMA_CM_EVENT_CONNECT_REQUEST;
2093         event.param.conn.private_data = iw_event->private_data;
2094         event.param.conn.private_data_len = iw_event->private_data_len;
2095         event.param.conn.initiator_depth = iw_event->ird;
2096         event.param.conn.responder_resources = iw_event->ord;
2097
2098         /*
2099          * Protect against the user destroying conn_id from another thread
2100          * until we're done accessing it.
2101          */
2102         atomic_inc(&conn_id->refcount);
2103         ret = conn_id->id.event_handler(&conn_id->id, &event);
2104         if (ret) {
2105                 /* User wants to destroy the CM ID */
2106                 conn_id->cm_id.iw = NULL;
2107                 cma_exch(conn_id, RDMA_CM_DESTROYING);
2108                 mutex_unlock(&conn_id->handler_mutex);
2109                 cma_deref_id(conn_id);
2110                 rdma_destroy_id(&conn_id->id);
2111                 goto out;
2112         }
2113
2114         mutex_unlock(&conn_id->handler_mutex);
2115         cma_deref_id(conn_id);
2116
2117 out:
2118         mutex_unlock(&listen_id->handler_mutex);
2119         return ret;
2120 }
2121
2122 static int cma_ib_listen(struct rdma_id_private *id_priv)
2123 {
2124         struct sockaddr *addr;
2125         struct ib_cm_id *id;
2126         __be64 svc_id;
2127
2128         addr = cma_src_addr(id_priv);
2129         svc_id = rdma_get_service_id(&id_priv->id, addr);
2130         id = ib_cm_insert_listen(id_priv->id.device, cma_req_handler, svc_id);
2131         if (IS_ERR(id))
2132                 return PTR_ERR(id);
2133         id_priv->cm_id.ib = id;
2134
2135         return 0;
2136 }
2137
2138 static int cma_iw_listen(struct rdma_id_private *id_priv, int backlog)
2139 {
2140         int ret;
2141         struct iw_cm_id *id;
2142
2143         id = iw_create_cm_id(id_priv->id.device,
2144                              iw_conn_req_handler,
2145                              id_priv);
2146         if (IS_ERR(id))
2147                 return PTR_ERR(id);
2148
2149         id->tos = id_priv->tos;
2150         id_priv->cm_id.iw = id;
2151
2152         memcpy(&id_priv->cm_id.iw->local_addr, cma_src_addr(id_priv),
2153                rdma_addr_size(cma_src_addr(id_priv)));
2154
2155         ret = iw_cm_listen(id_priv->cm_id.iw, backlog);
2156
2157         if (ret) {
2158                 iw_destroy_cm_id(id_priv->cm_id.iw);
2159                 id_priv->cm_id.iw = NULL;
2160         }
2161
2162         return ret;
2163 }
2164
2165 static int cma_listen_handler(struct rdma_cm_id *id,
2166                               struct rdma_cm_event *event)
2167 {
2168         struct rdma_id_private *id_priv = id->context;
2169
2170         id->context = id_priv->id.context;
2171         id->event_handler = id_priv->id.event_handler;
2172         return id_priv->id.event_handler(id, event);
2173 }
2174
2175 static void cma_listen_on_dev(struct rdma_id_private *id_priv,
2176                               struct cma_device *cma_dev)
2177 {
2178         struct rdma_id_private *dev_id_priv;
2179         struct rdma_cm_id *id;
2180         struct net *net = id_priv->id.route.addr.dev_addr.net;
2181         int ret;
2182
2183         if (cma_family(id_priv) == AF_IB && !rdma_cap_ib_cm(cma_dev->device, 1))
2184                 return;
2185
2186         id = rdma_create_id(net, cma_listen_handler, id_priv, id_priv->id.ps,
2187                             id_priv->id.qp_type);
2188         if (IS_ERR(id))
2189                 return;
2190
2191         dev_id_priv = container_of(id, struct rdma_id_private, id);
2192
2193         dev_id_priv->state = RDMA_CM_ADDR_BOUND;
2194         memcpy(cma_src_addr(dev_id_priv), cma_src_addr(id_priv),
2195                rdma_addr_size(cma_src_addr(id_priv)));
2196
2197         _cma_attach_to_dev(dev_id_priv, cma_dev);
2198         list_add_tail(&dev_id_priv->listen_list, &id_priv->listen_list);
2199         atomic_inc(&id_priv->refcount);
2200         dev_id_priv->internal_id = 1;
2201         dev_id_priv->afonly = id_priv->afonly;
2202
2203         ret = rdma_listen(id, id_priv->backlog);
2204         if (ret)
2205                 pr_warn("RDMA CMA: cma_listen_on_dev, error %d, listening on device %s\n",
2206                         ret, cma_dev->device->name);
2207 }
2208
2209 static void cma_listen_on_all(struct rdma_id_private *id_priv)
2210 {
2211         struct cma_device *cma_dev;
2212
2213         mutex_lock(&lock);
2214         list_add_tail(&id_priv->list, &listen_any_list);
2215         list_for_each_entry(cma_dev, &dev_list, list)
2216                 cma_listen_on_dev(id_priv, cma_dev);
2217         mutex_unlock(&lock);
2218 }
2219
2220 void rdma_set_service_type(struct rdma_cm_id *id, int tos)
2221 {
2222         struct rdma_id_private *id_priv;
2223
2224         id_priv = container_of(id, struct rdma_id_private, id);
2225         id_priv->tos = (u8) tos;
2226 }
2227 EXPORT_SYMBOL(rdma_set_service_type);
2228
2229 static void cma_query_handler(int status, struct ib_sa_path_rec *path_rec,
2230                               void *context)
2231 {
2232         struct cma_work *work = context;
2233         struct rdma_route *route;
2234
2235         route = &work->id->id.route;
2236
2237         if (!status) {
2238                 route->num_paths = 1;
2239                 *route->path_rec = *path_rec;
2240         } else {
2241                 work->old_state = RDMA_CM_ROUTE_QUERY;
2242                 work->new_state = RDMA_CM_ADDR_RESOLVED;
2243                 work->event.event = RDMA_CM_EVENT_ROUTE_ERROR;
2244                 work->event.status = status;
2245         }
2246
2247         queue_work(cma_wq, &work->work);
2248 }
2249
2250 static int cma_query_ib_route(struct rdma_id_private *id_priv, int timeout_ms,
2251                               struct cma_work *work)
2252 {
2253         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
2254         struct ib_sa_path_rec path_rec;
2255         ib_sa_comp_mask comp_mask;
2256         struct sockaddr_in6 *sin6;
2257         struct sockaddr_ib *sib;
2258
2259         memset(&path_rec, 0, sizeof path_rec);
2260         rdma_addr_get_sgid(dev_addr, &path_rec.sgid);
2261         rdma_addr_get_dgid(dev_addr, &path_rec.dgid);
2262         path_rec.pkey = cpu_to_be16(ib_addr_get_pkey(dev_addr));
2263         path_rec.numb_path = 1;
2264         path_rec.reversible = 1;
2265         path_rec.service_id = rdma_get_service_id(&id_priv->id, cma_dst_addr(id_priv));
2266
2267         comp_mask = IB_SA_PATH_REC_DGID | IB_SA_PATH_REC_SGID |
2268                     IB_SA_PATH_REC_PKEY | IB_SA_PATH_REC_NUMB_PATH |
2269                     IB_SA_PATH_REC_REVERSIBLE | IB_SA_PATH_REC_SERVICE_ID;
2270
2271         switch (cma_family(id_priv)) {
2272         case AF_INET:
2273                 path_rec.qos_class = cpu_to_be16((u16) id_priv->tos);
2274                 comp_mask |= IB_SA_PATH_REC_QOS_CLASS;
2275                 break;
2276         case AF_INET6:
2277                 sin6 = (struct sockaddr_in6 *) cma_src_addr(id_priv);
2278                 path_rec.traffic_class = (u8) (be32_to_cpu(sin6->sin6_flowinfo) >> 20);
2279                 comp_mask |= IB_SA_PATH_REC_TRAFFIC_CLASS;
2280                 break;
2281         case AF_IB:
2282                 sib = (struct sockaddr_ib *) cma_src_addr(id_priv);
2283                 path_rec.traffic_class = (u8) (be32_to_cpu(sib->sib_flowinfo) >> 20);
2284                 comp_mask |= IB_SA_PATH_REC_TRAFFIC_CLASS;
2285                 break;
2286         }
2287
2288         id_priv->query_id = ib_sa_path_rec_get(&sa_client, id_priv->id.device,
2289                                                id_priv->id.port_num, &path_rec,
2290                                                comp_mask, timeout_ms,
2291                                                GFP_KERNEL, cma_query_handler,
2292                                                work, &id_priv->query);
2293
2294         return (id_priv->query_id < 0) ? id_priv->query_id : 0;
2295 }
2296
2297 static void cma_work_handler(struct work_struct *_work)
2298 {
2299         struct cma_work *work = container_of(_work, struct cma_work, work);
2300         struct rdma_id_private *id_priv = work->id;
2301         int destroy = 0;
2302
2303         mutex_lock(&id_priv->handler_mutex);
2304         if (!cma_comp_exch(id_priv, work->old_state, work->new_state))
2305                 goto out;
2306
2307         if (id_priv->id.event_handler(&id_priv->id, &work->event)) {
2308                 cma_exch(id_priv, RDMA_CM_DESTROYING);
2309                 destroy = 1;
2310         }
2311 out:
2312         mutex_unlock(&id_priv->handler_mutex);
2313         cma_deref_id(id_priv);
2314         if (destroy)
2315                 rdma_destroy_id(&id_priv->id);
2316         kfree(work);
2317 }
2318
2319 static void cma_ndev_work_handler(struct work_struct *_work)
2320 {
2321         struct cma_ndev_work *work = container_of(_work, struct cma_ndev_work, work);
2322         struct rdma_id_private *id_priv = work->id;
2323         int destroy = 0;
2324
2325         mutex_lock(&id_priv->handler_mutex);
2326         if (id_priv->state == RDMA_CM_DESTROYING ||
2327             id_priv->state == RDMA_CM_DEVICE_REMOVAL)
2328                 goto out;
2329
2330         if (id_priv->id.event_handler(&id_priv->id, &work->event)) {
2331                 cma_exch(id_priv, RDMA_CM_DESTROYING);
2332                 destroy = 1;
2333         }
2334
2335 out:
2336         mutex_unlock(&id_priv->handler_mutex);
2337         cma_deref_id(id_priv);
2338         if (destroy)
2339                 rdma_destroy_id(&id_priv->id);
2340         kfree(work);
2341 }
2342
2343 static int cma_resolve_ib_route(struct rdma_id_private *id_priv, int timeout_ms)
2344 {
2345         struct rdma_route *route = &id_priv->id.route;
2346         struct cma_work *work;
2347         int ret;
2348
2349         work = kzalloc(sizeof *work, GFP_KERNEL);
2350         if (!work)
2351                 return -ENOMEM;
2352
2353         work->id = id_priv;
2354         INIT_WORK(&work->work, cma_work_handler);
2355         work->old_state = RDMA_CM_ROUTE_QUERY;
2356         work->new_state = RDMA_CM_ROUTE_RESOLVED;
2357         work->event.event = RDMA_CM_EVENT_ROUTE_RESOLVED;
2358
2359         route->path_rec = kmalloc(sizeof *route->path_rec, GFP_KERNEL);
2360         if (!route->path_rec) {
2361                 ret = -ENOMEM;
2362                 goto err1;
2363         }
2364
2365         ret = cma_query_ib_route(id_priv, timeout_ms, work);
2366         if (ret)
2367                 goto err2;
2368
2369         return 0;
2370 err2:
2371         kfree(route->path_rec);
2372         route->path_rec = NULL;
2373 err1:
2374         kfree(work);
2375         return ret;
2376 }
2377
2378 int rdma_set_ib_paths(struct rdma_cm_id *id,
2379                       struct ib_sa_path_rec *path_rec, int num_paths)
2380 {
2381         struct rdma_id_private *id_priv;
2382         int ret;
2383
2384         id_priv = container_of(id, struct rdma_id_private, id);
2385         if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_RESOLVED,
2386                            RDMA_CM_ROUTE_RESOLVED))
2387                 return -EINVAL;
2388
2389         id->route.path_rec = kmemdup(path_rec, sizeof *path_rec * num_paths,
2390                                      GFP_KERNEL);
2391         if (!id->route.path_rec) {
2392                 ret = -ENOMEM;
2393                 goto err;
2394         }
2395
2396         id->route.num_paths = num_paths;
2397         return 0;
2398 err:
2399         cma_comp_exch(id_priv, RDMA_CM_ROUTE_RESOLVED, RDMA_CM_ADDR_RESOLVED);
2400         return ret;
2401 }
2402 EXPORT_SYMBOL(rdma_set_ib_paths);
2403
2404 static int cma_resolve_iw_route(struct rdma_id_private *id_priv, int timeout_ms)
2405 {
2406         struct cma_work *work;
2407
2408         work = kzalloc(sizeof *work, GFP_KERNEL);
2409         if (!work)
2410                 return -ENOMEM;
2411
2412         work->id = id_priv;
2413         INIT_WORK(&work->work, cma_work_handler);
2414         work->old_state = RDMA_CM_ROUTE_QUERY;
2415         work->new_state = RDMA_CM_ROUTE_RESOLVED;
2416         work->event.event = RDMA_CM_EVENT_ROUTE_RESOLVED;
2417         queue_work(cma_wq, &work->work);
2418         return 0;
2419 }
2420
2421 static int iboe_tos_to_sl(struct net_device *ndev, int tos)
2422 {
2423         int prio;
2424         struct net_device *dev;
2425
2426         prio = rt_tos2priority(tos);
2427         dev = ndev->priv_flags & IFF_802_1Q_VLAN ?
2428                 vlan_dev_real_dev(ndev) : ndev;
2429
2430         if (dev->num_tc)
2431                 return netdev_get_prio_tc_map(dev, prio);
2432
2433 #if IS_ENABLED(CONFIG_VLAN_8021Q)
2434         if (ndev->priv_flags & IFF_802_1Q_VLAN)
2435                 return (vlan_dev_get_egress_qos_mask(ndev, prio) &
2436                         VLAN_PRIO_MASK) >> VLAN_PRIO_SHIFT;
2437 #endif
2438         return 0;
2439 }
2440
2441 static enum ib_gid_type cma_route_gid_type(enum rdma_network_type network_type,
2442                                            unsigned long supported_gids,
2443                                            enum ib_gid_type default_gid)
2444 {
2445         if ((network_type == RDMA_NETWORK_IPV4 ||
2446              network_type == RDMA_NETWORK_IPV6) &&
2447             test_bit(IB_GID_TYPE_ROCE_UDP_ENCAP, &supported_gids))
2448                 return IB_GID_TYPE_ROCE_UDP_ENCAP;
2449
2450         return default_gid;
2451 }
2452
2453 static int cma_resolve_iboe_route(struct rdma_id_private *id_priv)
2454 {
2455         struct rdma_route *route = &id_priv->id.route;
2456         struct rdma_addr *addr = &route->addr;
2457         struct cma_work *work;
2458         int ret;
2459         struct net_device *ndev = NULL;
2460
2461
2462         work = kzalloc(sizeof *work, GFP_KERNEL);
2463         if (!work)
2464                 return -ENOMEM;
2465
2466         work->id = id_priv;
2467         INIT_WORK(&work->work, cma_work_handler);
2468
2469         route->path_rec = kzalloc(sizeof *route->path_rec, GFP_KERNEL);
2470         if (!route->path_rec) {
2471                 ret = -ENOMEM;
2472                 goto err1;
2473         }
2474
2475         route->num_paths = 1;
2476
2477         if (addr->dev_addr.bound_dev_if) {
2478                 unsigned long supported_gids;
2479
2480                 ndev = dev_get_by_index(&init_net, addr->dev_addr.bound_dev_if);
2481                 if (!ndev) {
2482                         ret = -ENODEV;
2483                         goto err2;
2484                 }
2485
2486                 if (ndev->flags & IFF_LOOPBACK) {
2487                         dev_put(ndev);
2488                         if (!id_priv->id.device->get_netdev) {
2489                                 ret = -EOPNOTSUPP;
2490                                 goto err2;
2491                         }
2492
2493                         ndev = id_priv->id.device->get_netdev(id_priv->id.device,
2494                                                               id_priv->id.port_num);
2495                         if (!ndev) {
2496                                 ret = -ENODEV;
2497                                 goto err2;
2498                         }
2499                 }
2500
2501                 route->path_rec->net = &init_net;
2502                 route->path_rec->ifindex = ndev->ifindex;
2503                 supported_gids = roce_gid_type_mask_support(id_priv->id.device,
2504                                                             id_priv->id.port_num);
2505                 route->path_rec->gid_type =
2506                         cma_route_gid_type(addr->dev_addr.network,
2507                                            supported_gids,
2508                                            id_priv->gid_type);
2509         }
2510         if (!ndev) {
2511                 ret = -ENODEV;
2512                 goto err2;
2513         }
2514
2515         memcpy(route->path_rec->dmac, addr->dev_addr.dst_dev_addr, ETH_ALEN);
2516
2517         rdma_ip2gid((struct sockaddr *)&id_priv->id.route.addr.src_addr,
2518                     &route->path_rec->sgid);
2519         rdma_ip2gid((struct sockaddr *)&id_priv->id.route.addr.dst_addr,
2520                     &route->path_rec->dgid);
2521
2522         /* Use the hint from IP Stack to select GID Type */
2523         if (route->path_rec->gid_type < ib_network_to_gid_type(addr->dev_addr.network))
2524                 route->path_rec->gid_type = ib_network_to_gid_type(addr->dev_addr.network);
2525         if (((struct sockaddr *)&id_priv->id.route.addr.dst_addr)->sa_family != AF_IB)
2526                 /* TODO: get the hoplimit from the inet/inet6 device */
2527                 route->path_rec->hop_limit = addr->dev_addr.hoplimit;
2528         else
2529                 route->path_rec->hop_limit = 1;
2530         route->path_rec->reversible = 1;
2531         route->path_rec->pkey = cpu_to_be16(0xffff);
2532         route->path_rec->mtu_selector = IB_SA_EQ;
2533         route->path_rec->sl = iboe_tos_to_sl(ndev, id_priv->tos);
2534         route->path_rec->mtu = iboe_get_mtu(ndev->mtu);
2535         route->path_rec->rate_selector = IB_SA_EQ;
2536         route->path_rec->rate = iboe_get_rate(ndev);
2537         dev_put(ndev);
2538         route->path_rec->packet_life_time_selector = IB_SA_EQ;
2539         route->path_rec->packet_life_time = CMA_IBOE_PACKET_LIFETIME;
2540         if (!route->path_rec->mtu) {
2541                 ret = -EINVAL;
2542                 goto err2;
2543         }
2544
2545         work->old_state = RDMA_CM_ROUTE_QUERY;
2546         work->new_state = RDMA_CM_ROUTE_RESOLVED;
2547         work->event.event = RDMA_CM_EVENT_ROUTE_RESOLVED;
2548         work->event.status = 0;
2549
2550         queue_work(cma_wq, &work->work);
2551
2552         return 0;
2553
2554 err2:
2555         kfree(route->path_rec);
2556         route->path_rec = NULL;
2557 err1:
2558         kfree(work);
2559         return ret;
2560 }
2561
2562 int rdma_resolve_route(struct rdma_cm_id *id, int timeout_ms)
2563 {
2564         struct rdma_id_private *id_priv;
2565         int ret;
2566
2567         id_priv = container_of(id, struct rdma_id_private, id);
2568         if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_RESOLVED, RDMA_CM_ROUTE_QUERY))
2569                 return -EINVAL;
2570
2571         atomic_inc(&id_priv->refcount);
2572         if (rdma_cap_ib_sa(id->device, id->port_num))
2573                 ret = cma_resolve_ib_route(id_priv, timeout_ms);
2574         else if (rdma_protocol_roce(id->device, id->port_num))
2575                 ret = cma_resolve_iboe_route(id_priv);
2576         else if (rdma_protocol_iwarp(id->device, id->port_num))
2577                 ret = cma_resolve_iw_route(id_priv, timeout_ms);
2578         else
2579                 ret = -ENOSYS;
2580
2581         if (ret)
2582                 goto err;
2583
2584         return 0;
2585 err:
2586         cma_comp_exch(id_priv, RDMA_CM_ROUTE_QUERY, RDMA_CM_ADDR_RESOLVED);
2587         cma_deref_id(id_priv);
2588         return ret;
2589 }
2590 EXPORT_SYMBOL(rdma_resolve_route);
2591
2592 static void cma_set_loopback(struct sockaddr *addr)
2593 {
2594         switch (addr->sa_family) {
2595         case AF_INET:
2596                 ((struct sockaddr_in *) addr)->sin_addr.s_addr = htonl(INADDR_LOOPBACK);
2597                 break;
2598         case AF_INET6:
2599                 ipv6_addr_set(&((struct sockaddr_in6 *) addr)->sin6_addr,
2600                               0, 0, 0, htonl(1));
2601                 break;
2602         default:
2603                 ib_addr_set(&((struct sockaddr_ib *) addr)->sib_addr,
2604                             0, 0, 0, htonl(1));
2605                 break;
2606         }
2607 }
2608
2609 static int cma_bind_loopback(struct rdma_id_private *id_priv)
2610 {
2611         struct cma_device *cma_dev, *cur_dev;
2612         struct ib_port_attr port_attr;
2613         union ib_gid gid;
2614         u16 pkey;
2615         int ret;
2616         u8 p;
2617
2618         cma_dev = NULL;
2619         mutex_lock(&lock);
2620         list_for_each_entry(cur_dev, &dev_list, list) {
2621                 if (cma_family(id_priv) == AF_IB &&
2622                     !rdma_cap_ib_cm(cur_dev->device, 1))
2623                         continue;
2624
2625                 if (!cma_dev)
2626                         cma_dev = cur_dev;
2627
2628                 for (p = 1; p <= cur_dev->device->phys_port_cnt; ++p) {
2629                         if (!ib_query_port(cur_dev->device, p, &port_attr) &&
2630                             port_attr.state == IB_PORT_ACTIVE) {
2631                                 cma_dev = cur_dev;
2632                                 goto port_found;
2633                         }
2634                 }
2635         }
2636
2637         if (!cma_dev) {
2638                 ret = -ENODEV;
2639                 goto out;
2640         }
2641
2642         p = 1;
2643
2644 port_found:
2645         ret = ib_get_cached_gid(cma_dev->device, p, 0, &gid, NULL);
2646         if (ret)
2647                 goto out;
2648
2649         ret = ib_get_cached_pkey(cma_dev->device, p, 0, &pkey);
2650         if (ret)
2651                 goto out;
2652
2653         id_priv->id.route.addr.dev_addr.dev_type =
2654                 (rdma_protocol_ib(cma_dev->device, p)) ?
2655                 ARPHRD_INFINIBAND : ARPHRD_ETHER;
2656
2657         rdma_addr_set_sgid(&id_priv->id.route.addr.dev_addr, &gid);
2658         ib_addr_set_pkey(&id_priv->id.route.addr.dev_addr, pkey);
2659         id_priv->id.port_num = p;
2660         cma_attach_to_dev(id_priv, cma_dev);
2661         cma_set_loopback(cma_src_addr(id_priv));
2662 out:
2663         mutex_unlock(&lock);
2664         return ret;
2665 }
2666
2667 static void addr_handler(int status, struct sockaddr *src_addr,
2668                          struct rdma_dev_addr *dev_addr, void *context)
2669 {
2670         struct rdma_id_private *id_priv = context;
2671         struct rdma_cm_event event;
2672
2673         memset(&event, 0, sizeof event);
2674         mutex_lock(&id_priv->handler_mutex);
2675         if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_QUERY,
2676                            RDMA_CM_ADDR_RESOLVED))
2677                 goto out;
2678
2679         memcpy(cma_src_addr(id_priv), src_addr, rdma_addr_size(src_addr));
2680         if (!status && !id_priv->cma_dev)
2681                 status = cma_acquire_dev(id_priv, NULL);
2682
2683         if (status) {
2684                 if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_RESOLVED,
2685                                    RDMA_CM_ADDR_BOUND))
2686                         goto out;
2687                 event.event = RDMA_CM_EVENT_ADDR_ERROR;
2688                 event.status = status;
2689         } else
2690                 event.event = RDMA_CM_EVENT_ADDR_RESOLVED;
2691
2692         if (id_priv->id.event_handler(&id_priv->id, &event)) {
2693                 cma_exch(id_priv, RDMA_CM_DESTROYING);
2694                 mutex_unlock(&id_priv->handler_mutex);
2695                 cma_deref_id(id_priv);
2696                 rdma_destroy_id(&id_priv->id);
2697                 return;
2698         }
2699 out:
2700         mutex_unlock(&id_priv->handler_mutex);
2701         cma_deref_id(id_priv);
2702 }
2703
2704 static int cma_resolve_loopback(struct rdma_id_private *id_priv)
2705 {
2706         struct cma_work *work;
2707         union ib_gid gid;
2708         int ret;
2709
2710         work = kzalloc(sizeof *work, GFP_KERNEL);
2711         if (!work)
2712                 return -ENOMEM;
2713
2714         if (!id_priv->cma_dev) {
2715                 ret = cma_bind_loopback(id_priv);
2716                 if (ret)
2717                         goto err;
2718         }
2719
2720         rdma_addr_get_sgid(&id_priv->id.route.addr.dev_addr, &gid);
2721         rdma_addr_set_dgid(&id_priv->id.route.addr.dev_addr, &gid);
2722
2723         work->id = id_priv;
2724         INIT_WORK(&work->work, cma_work_handler);
2725         work->old_state = RDMA_CM_ADDR_QUERY;
2726         work->new_state = RDMA_CM_ADDR_RESOLVED;
2727         work->event.event = RDMA_CM_EVENT_ADDR_RESOLVED;
2728         queue_work(cma_wq, &work->work);
2729         return 0;
2730 err:
2731         kfree(work);
2732         return ret;
2733 }
2734
2735 static int cma_resolve_ib_addr(struct rdma_id_private *id_priv)
2736 {
2737         struct cma_work *work;
2738         int ret;
2739
2740         work = kzalloc(sizeof *work, GFP_KERNEL);
2741         if (!work)
2742                 return -ENOMEM;
2743
2744         if (!id_priv->cma_dev) {
2745                 ret = cma_resolve_ib_dev(id_priv);
2746                 if (ret)
2747                         goto err;
2748         }
2749
2750         rdma_addr_set_dgid(&id_priv->id.route.addr.dev_addr, (union ib_gid *)
2751                 &(((struct sockaddr_ib *) &id_priv->id.route.addr.dst_addr)->sib_addr));
2752
2753         work->id = id_priv;
2754         INIT_WORK(&work->work, cma_work_handler);
2755         work->old_state = RDMA_CM_ADDR_QUERY;
2756         work->new_state = RDMA_CM_ADDR_RESOLVED;
2757         work->event.event = RDMA_CM_EVENT_ADDR_RESOLVED;
2758         queue_work(cma_wq, &work->work);
2759         return 0;
2760 err:
2761         kfree(work);
2762         return ret;
2763 }
2764
2765 static int cma_bind_addr(struct rdma_cm_id *id, struct sockaddr *src_addr,
2766                          struct sockaddr *dst_addr)
2767 {
2768         if (!src_addr || !src_addr->sa_family) {
2769                 src_addr = (struct sockaddr *) &id->route.addr.src_addr;
2770                 src_addr->sa_family = dst_addr->sa_family;
2771                 if (dst_addr->sa_family == AF_INET6) {
2772                         struct sockaddr_in6 *src_addr6 = (struct sockaddr_in6 *) src_addr;
2773                         struct sockaddr_in6 *dst_addr6 = (struct sockaddr_in6 *) dst_addr;
2774                         src_addr6->sin6_scope_id = dst_addr6->sin6_scope_id;
2775                         if (ipv6_addr_type(&dst_addr6->sin6_addr) & IPV6_ADDR_LINKLOCAL)
2776                                 id->route.addr.dev_addr.bound_dev_if = dst_addr6->sin6_scope_id;
2777                 } else if (dst_addr->sa_family == AF_IB) {
2778                         ((struct sockaddr_ib *) src_addr)->sib_pkey =
2779                                 ((struct sockaddr_ib *) dst_addr)->sib_pkey;
2780                 }
2781         }
2782         return rdma_bind_addr(id, src_addr);
2783 }
2784
2785 int rdma_resolve_addr(struct rdma_cm_id *id, struct sockaddr *src_addr,
2786                       struct sockaddr *dst_addr, int timeout_ms)
2787 {
2788         struct rdma_id_private *id_priv;
2789         int ret;
2790
2791         id_priv = container_of(id, struct rdma_id_private, id);
2792         if (id_priv->state == RDMA_CM_IDLE) {
2793                 ret = cma_bind_addr(id, src_addr, dst_addr);
2794                 if (ret)
2795                         return ret;
2796         }
2797
2798         if (cma_family(id_priv) != dst_addr->sa_family)
2799                 return -EINVAL;
2800
2801         if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_BOUND, RDMA_CM_ADDR_QUERY))
2802                 return -EINVAL;
2803
2804         atomic_inc(&id_priv->refcount);
2805         memcpy(cma_dst_addr(id_priv), dst_addr, rdma_addr_size(dst_addr));
2806         if (cma_any_addr(dst_addr)) {
2807                 ret = cma_resolve_loopback(id_priv);
2808         } else {
2809                 if (dst_addr->sa_family == AF_IB) {
2810                         ret = cma_resolve_ib_addr(id_priv);
2811                 } else {
2812                         ret = rdma_resolve_ip(&addr_client, cma_src_addr(id_priv),
2813                                               dst_addr, &id->route.addr.dev_addr,
2814                                               timeout_ms, addr_handler, id_priv);
2815                 }
2816         }
2817         if (ret)
2818                 goto err;
2819
2820         return 0;
2821 err:
2822         cma_comp_exch(id_priv, RDMA_CM_ADDR_QUERY, RDMA_CM_ADDR_BOUND);
2823         cma_deref_id(id_priv);
2824         return ret;
2825 }
2826 EXPORT_SYMBOL(rdma_resolve_addr);
2827
2828 int rdma_set_reuseaddr(struct rdma_cm_id *id, int reuse)
2829 {
2830         struct rdma_id_private *id_priv;
2831         unsigned long flags;
2832         int ret;
2833
2834         id_priv = container_of(id, struct rdma_id_private, id);
2835         spin_lock_irqsave(&id_priv->lock, flags);
2836         if (reuse || id_priv->state == RDMA_CM_IDLE) {
2837                 id_priv->reuseaddr = reuse;
2838                 ret = 0;
2839         } else {
2840                 ret = -EINVAL;
2841         }
2842         spin_unlock_irqrestore(&id_priv->lock, flags);
2843         return ret;
2844 }
2845 EXPORT_SYMBOL(rdma_set_reuseaddr);
2846
2847 int rdma_set_afonly(struct rdma_cm_id *id, int afonly)
2848 {
2849         struct rdma_id_private *id_priv;
2850         unsigned long flags;
2851         int ret;
2852
2853         id_priv = container_of(id, struct rdma_id_private, id);
2854         spin_lock_irqsave(&id_priv->lock, flags);
2855         if (id_priv->state == RDMA_CM_IDLE || id_priv->state == RDMA_CM_ADDR_BOUND) {
2856                 id_priv->options |= (1 << CMA_OPTION_AFONLY);
2857                 id_priv->afonly = afonly;
2858                 ret = 0;
2859         } else {
2860                 ret = -EINVAL;
2861         }
2862         spin_unlock_irqrestore(&id_priv->lock, flags);
2863         return ret;
2864 }
2865 EXPORT_SYMBOL(rdma_set_afonly);
2866
2867 static void cma_bind_port(struct rdma_bind_list *bind_list,
2868                           struct rdma_id_private *id_priv)
2869 {
2870         struct sockaddr *addr;
2871         struct sockaddr_ib *sib;
2872         u64 sid, mask;
2873         __be16 port;
2874
2875         addr = cma_src_addr(id_priv);
2876         port = htons(bind_list->port);
2877
2878         switch (addr->sa_family) {
2879         case AF_INET:
2880                 ((struct sockaddr_in *) addr)->sin_port = port;
2881                 break;
2882         case AF_INET6:
2883                 ((struct sockaddr_in6 *) addr)->sin6_port = port;
2884                 break;
2885         case AF_IB:
2886                 sib = (struct sockaddr_ib *) addr;
2887                 sid = be64_to_cpu(sib->sib_sid);
2888                 mask = be64_to_cpu(sib->sib_sid_mask);
2889                 sib->sib_sid = cpu_to_be64((sid & mask) | (u64) ntohs(port));
2890                 sib->sib_sid_mask = cpu_to_be64(~0ULL);
2891                 break;
2892         }
2893         id_priv->bind_list = bind_list;
2894         hlist_add_head(&id_priv->node, &bind_list->owners);
2895 }
2896
2897 static int cma_alloc_port(enum rdma_port_space ps,
2898                           struct rdma_id_private *id_priv, unsigned short snum)
2899 {
2900         struct rdma_bind_list *bind_list;
2901         int ret;
2902
2903         bind_list = kzalloc(sizeof *bind_list, GFP_KERNEL);
2904         if (!bind_list)
2905                 return -ENOMEM;
2906
2907         ret = cma_ps_alloc(id_priv->id.route.addr.dev_addr.net, ps, bind_list,
2908                            snum);
2909         if (ret < 0)
2910                 goto err;
2911
2912         bind_list->ps = ps;
2913         bind_list->port = (unsigned short)ret;
2914         cma_bind_port(bind_list, id_priv);
2915         return 0;
2916 err:
2917         kfree(bind_list);
2918         return ret == -ENOSPC ? -EADDRNOTAVAIL : ret;
2919 }
2920
2921 static int cma_alloc_any_port(enum rdma_port_space ps,
2922                               struct rdma_id_private *id_priv)
2923 {
2924         static unsigned int last_used_port;
2925         int low, high, remaining;
2926         unsigned int rover;
2927         struct net *net = id_priv->id.route.addr.dev_addr.net;
2928
2929         inet_get_local_port_range(net, &low, &high);
2930         remaining = (high - low) + 1;
2931         rover = prandom_u32() % remaining + low;
2932 retry:
2933         if (last_used_port != rover &&
2934             !cma_ps_find(net, ps, (unsigned short)rover)) {
2935                 int ret = cma_alloc_port(ps, id_priv, rover);
2936                 /*
2937                  * Remember previously used port number in order to avoid
2938                  * re-using same port immediately after it is closed.
2939                  */
2940                 if (!ret)
2941                         last_used_port = rover;
2942                 if (ret != -EADDRNOTAVAIL)
2943                         return ret;
2944         }
2945         if (--remaining) {
2946                 rover++;
2947                 if ((rover < low) || (rover > high))
2948                         rover = low;
2949                 goto retry;
2950         }
2951         return -EADDRNOTAVAIL;
2952 }
2953
2954 /*
2955  * Check that the requested port is available.  This is called when trying to
2956  * bind to a specific port, or when trying to listen on a bound port.  In
2957  * the latter case, the provided id_priv may already be on the bind_list, but
2958  * we still need to check that it's okay to start listening.
2959  */
2960 static int cma_check_port(struct rdma_bind_list *bind_list,
2961                           struct rdma_id_private *id_priv, uint8_t reuseaddr)
2962 {
2963         struct rdma_id_private *cur_id;
2964         struct sockaddr *addr, *cur_addr;
2965
2966         addr = cma_src_addr(id_priv);
2967         hlist_for_each_entry(cur_id, &bind_list->owners, node) {
2968                 if (id_priv == cur_id)
2969                         continue;
2970
2971                 if ((cur_id->state != RDMA_CM_LISTEN) && reuseaddr &&
2972                     cur_id->reuseaddr)
2973                         continue;
2974
2975                 cur_addr = cma_src_addr(cur_id);
2976                 if (id_priv->afonly && cur_id->afonly &&
2977                     (addr->sa_family != cur_addr->sa_family))
2978                         continue;
2979
2980                 if (cma_any_addr(addr) || cma_any_addr(cur_addr))
2981                         return -EADDRNOTAVAIL;
2982
2983                 if (!cma_addr_cmp(addr, cur_addr))
2984                         return -EADDRINUSE;
2985         }
2986         return 0;
2987 }
2988
2989 static int cma_use_port(enum rdma_port_space ps,
2990                         struct rdma_id_private *id_priv)
2991 {
2992         struct rdma_bind_list *bind_list;
2993         unsigned short snum;
2994         int ret;
2995
2996         snum = ntohs(cma_port(cma_src_addr(id_priv)));
2997         if (snum < PROT_SOCK && !capable(CAP_NET_BIND_SERVICE))
2998                 return -EACCES;
2999
3000         bind_list = cma_ps_find(id_priv->id.route.addr.dev_addr.net, ps, snum);
3001         if (!bind_list) {
3002                 ret = cma_alloc_port(ps, id_priv, snum);
3003         } else {
3004                 ret = cma_check_port(bind_list, id_priv, id_priv->reuseaddr);
3005                 if (!ret)
3006                         cma_bind_port(bind_list, id_priv);
3007         }
3008         return ret;
3009 }
3010
3011 static int cma_bind_listen(struct rdma_id_private *id_priv)
3012 {
3013         struct rdma_bind_list *bind_list = id_priv->bind_list;
3014         int ret = 0;
3015
3016         mutex_lock(&lock);
3017         if (bind_list->owners.first->next)
3018                 ret = cma_check_port(bind_list, id_priv, 0);
3019         mutex_unlock(&lock);
3020         return ret;
3021 }
3022
3023 static enum rdma_port_space cma_select_inet_ps(
3024                 struct rdma_id_private *id_priv)
3025 {
3026         switch (id_priv->id.ps) {
3027         case RDMA_PS_TCP:
3028         case RDMA_PS_UDP:
3029         case RDMA_PS_IPOIB:
3030         case RDMA_PS_IB:
3031                 return id_priv->id.ps;
3032         default:
3033
3034                 return 0;
3035         }
3036 }
3037
3038 static enum rdma_port_space cma_select_ib_ps(struct rdma_id_private *id_priv)
3039 {
3040         enum rdma_port_space ps = 0;
3041         struct sockaddr_ib *sib;
3042         u64 sid_ps, mask, sid;
3043
3044         sib = (struct sockaddr_ib *) cma_src_addr(id_priv);
3045         mask = be64_to_cpu(sib->sib_sid_mask) & RDMA_IB_IP_PS_MASK;
3046         sid = be64_to_cpu(sib->sib_sid) & mask;
3047
3048         if ((id_priv->id.ps == RDMA_PS_IB) && (sid == (RDMA_IB_IP_PS_IB & mask))) {
3049                 sid_ps = RDMA_IB_IP_PS_IB;
3050                 ps = RDMA_PS_IB;
3051         } else if (((id_priv->id.ps == RDMA_PS_IB) || (id_priv->id.ps == RDMA_PS_TCP)) &&
3052                    (sid == (RDMA_IB_IP_PS_TCP & mask))) {
3053                 sid_ps = RDMA_IB_IP_PS_TCP;
3054                 ps = RDMA_PS_TCP;
3055         } else if (((id_priv->id.ps == RDMA_PS_IB) || (id_priv->id.ps == RDMA_PS_UDP)) &&
3056                    (sid == (RDMA_IB_IP_PS_UDP & mask))) {
3057                 sid_ps = RDMA_IB_IP_PS_UDP;
3058                 ps = RDMA_PS_UDP;
3059         }
3060
3061         if (ps) {
3062                 sib->sib_sid = cpu_to_be64(sid_ps | ntohs(cma_port((struct sockaddr *) sib)));
3063                 sib->sib_sid_mask = cpu_to_be64(RDMA_IB_IP_PS_MASK |
3064                                                 be64_to_cpu(sib->sib_sid_mask));
3065         }
3066         return ps;
3067 }
3068
3069 static int cma_get_port(struct rdma_id_private *id_priv)
3070 {
3071         enum rdma_port_space ps;
3072         int ret;
3073
3074         if (cma_family(id_priv) != AF_IB)
3075                 ps = cma_select_inet_ps(id_priv);
3076         else
3077                 ps = cma_select_ib_ps(id_priv);
3078         if (!ps)
3079                 return -EPROTONOSUPPORT;
3080
3081         mutex_lock(&lock);
3082         if (cma_any_port(cma_src_addr(id_priv)))
3083                 ret = cma_alloc_any_port(ps, id_priv);
3084         else
3085                 ret = cma_use_port(ps, id_priv);
3086         mutex_unlock(&lock);
3087
3088         return ret;
3089 }
3090
3091 static int cma_check_linklocal(struct rdma_dev_addr *dev_addr,
3092                                struct sockaddr *addr)
3093 {
3094 #if IS_ENABLED(CONFIG_IPV6)
3095         struct sockaddr_in6 *sin6;
3096
3097         if (addr->sa_family != AF_INET6)
3098                 return 0;
3099
3100         sin6 = (struct sockaddr_in6 *) addr;
3101
3102         if (!(ipv6_addr_type(&sin6->sin6_addr) & IPV6_ADDR_LINKLOCAL))
3103                 return 0;
3104
3105         if (!sin6->sin6_scope_id)
3106                         return -EINVAL;
3107
3108         dev_addr->bound_dev_if = sin6->sin6_scope_id;
3109 #endif
3110         return 0;
3111 }
3112
3113 int rdma_listen(struct rdma_cm_id *id, int backlog)
3114 {
3115         struct rdma_id_private *id_priv;
3116         int ret;
3117
3118         id_priv = container_of(id, struct rdma_id_private, id);
3119         if (id_priv->state == RDMA_CM_IDLE) {
3120                 id->route.addr.src_addr.ss_family = AF_INET;
3121                 ret = rdma_bind_addr(id, cma_src_addr(id_priv));
3122                 if (ret)
3123                         return ret;
3124         }
3125
3126         if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_BOUND, RDMA_CM_LISTEN))
3127                 return -EINVAL;
3128
3129         if (id_priv->reuseaddr) {
3130                 ret = cma_bind_listen(id_priv);
3131                 if (ret)
3132                         goto err;
3133         }
3134
3135         id_priv->backlog = backlog;
3136         if (id->device) {
3137                 if (rdma_cap_ib_cm(id->device, 1)) {
3138                         ret = cma_ib_listen(id_priv);
3139                         if (ret)
3140                                 goto err;
3141                 } else if (rdma_cap_iw_cm(id->device, 1)) {
3142                         ret = cma_iw_listen(id_priv, backlog);
3143                         if (ret)
3144                                 goto err;
3145                 } else {
3146                         ret = -ENOSYS;
3147                         goto err;
3148                 }
3149         } else
3150                 cma_listen_on_all(id_priv);
3151
3152         return 0;
3153 err:
3154         id_priv->backlog = 0;
3155         cma_comp_exch(id_priv, RDMA_CM_LISTEN, RDMA_CM_ADDR_BOUND);
3156         return ret;
3157 }
3158 EXPORT_SYMBOL(rdma_listen);
3159
3160 int rdma_bind_addr(struct rdma_cm_id *id, struct sockaddr *addr)
3161 {
3162         struct rdma_id_private *id_priv;
3163         int ret;
3164
3165         if (addr->sa_family != AF_INET && addr->sa_family != AF_INET6 &&
3166             addr->sa_family != AF_IB)
3167                 return -EAFNOSUPPORT;
3168
3169         id_priv = container_of(id, struct rdma_id_private, id);
3170         if (!cma_comp_exch(id_priv, RDMA_CM_IDLE, RDMA_CM_ADDR_BOUND))
3171                 return -EINVAL;
3172
3173         ret = cma_check_linklocal(&id->route.addr.dev_addr, addr);
3174         if (ret)
3175                 goto err1;
3176
3177         memcpy(cma_src_addr(id_priv), addr, rdma_addr_size(addr));
3178         if (!cma_any_addr(addr)) {
3179                 ret = cma_translate_addr(addr, &id->route.addr.dev_addr);
3180                 if (ret)
3181                         goto err1;
3182
3183                 ret = cma_acquire_dev(id_priv, NULL);
3184                 if (ret)
3185                         goto err1;
3186         }
3187
3188         if (!(id_priv->options & (1 << CMA_OPTION_AFONLY))) {
3189                 if (addr->sa_family == AF_INET)
3190                         id_priv->afonly = 1;
3191 #if IS_ENABLED(CONFIG_IPV6)
3192                 else if (addr->sa_family == AF_INET6) {
3193                         struct net *net = id_priv->id.route.addr.dev_addr.net;
3194
3195                         id_priv->afonly = net->ipv6.sysctl.bindv6only;
3196                 }
3197 #endif
3198         }
3199         ret = cma_get_port(id_priv);
3200         if (ret)
3201                 goto err2;
3202
3203         return 0;
3204 err2:
3205         if (id_priv->cma_dev)
3206                 cma_release_dev(id_priv);
3207 err1:
3208         cma_comp_exch(id_priv, RDMA_CM_ADDR_BOUND, RDMA_CM_IDLE);
3209         return ret;
3210 }
3211 EXPORT_SYMBOL(rdma_bind_addr);
3212
3213 static int cma_format_hdr(void *hdr, struct rdma_id_private *id_priv)
3214 {
3215         struct cma_hdr *cma_hdr;
3216
3217         cma_hdr = hdr;
3218         cma_hdr->cma_version = CMA_VERSION;
3219         if (cma_family(id_priv) == AF_INET) {
3220                 struct sockaddr_in *src4, *dst4;
3221
3222                 src4 = (struct sockaddr_in *) cma_src_addr(id_priv);
3223                 dst4 = (struct sockaddr_in *) cma_dst_addr(id_priv);
3224
3225                 cma_set_ip_ver(cma_hdr, 4);
3226                 cma_hdr->src_addr.ip4.addr = src4->sin_addr.s_addr;
3227                 cma_hdr->dst_addr.ip4.addr = dst4->sin_addr.s_addr;
3228                 cma_hdr->port = src4->sin_port;
3229         } else if (cma_family(id_priv) == AF_INET6) {
3230                 struct sockaddr_in6 *src6, *dst6;
3231
3232                 src6 = (struct sockaddr_in6 *) cma_src_addr(id_priv);
3233                 dst6 = (struct sockaddr_in6 *) cma_dst_addr(id_priv);
3234
3235                 cma_set_ip_ver(cma_hdr, 6);
3236                 cma_hdr->src_addr.ip6 = src6->sin6_addr;
3237                 cma_hdr->dst_addr.ip6 = dst6->sin6_addr;
3238                 cma_hdr->port = src6->sin6_port;
3239         }
3240         return 0;
3241 }
3242
3243 static int cma_sidr_rep_handler(struct ib_cm_id *cm_id,
3244                                 struct ib_cm_event *ib_event)
3245 {
3246         struct rdma_id_private *id_priv = cm_id->context;
3247         struct rdma_cm_event event;
3248         struct ib_cm_sidr_rep_event_param *rep = &ib_event->param.sidr_rep_rcvd;
3249         int ret = 0;
3250
3251         mutex_lock(&id_priv->handler_mutex);
3252         if (id_priv->state != RDMA_CM_CONNECT)
3253                 goto out;
3254
3255         memset(&event, 0, sizeof event);
3256         switch (ib_event->event) {
3257         case IB_CM_SIDR_REQ_ERROR:
3258                 event.event = RDMA_CM_EVENT_UNREACHABLE;
3259                 event.status = -ETIMEDOUT;
3260                 break;
3261         case IB_CM_SIDR_REP_RECEIVED:
3262                 event.param.ud.private_data = ib_event->private_data;
3263                 event.param.ud.private_data_len = IB_CM_SIDR_REP_PRIVATE_DATA_SIZE;
3264                 if (rep->status != IB_SIDR_SUCCESS) {
3265                         event.event = RDMA_CM_EVENT_UNREACHABLE;
3266                         event.status = ib_event->param.sidr_rep_rcvd.status;
3267                         break;
3268                 }
3269                 ret = cma_set_qkey(id_priv, rep->qkey);
3270                 if (ret) {
3271                         event.event = RDMA_CM_EVENT_ADDR_ERROR;
3272                         event.status = ret;
3273                         break;
3274                 }
3275                 ib_init_ah_from_path(id_priv->id.device, id_priv->id.port_num,
3276                                      id_priv->id.route.path_rec,
3277                                      &event.param.ud.ah_attr);
3278                 event.param.ud.qp_num = rep->qpn;
3279                 event.param.ud.qkey = rep->qkey;
3280                 event.event = RDMA_CM_EVENT_ESTABLISHED;
3281                 event.status = 0;
3282                 break;
3283         default:
3284                 pr_err("RDMA CMA: unexpected IB CM event: %d\n",
3285                        ib_event->event);
3286                 goto out;
3287         }
3288
3289         ret = id_priv->id.event_handler(&id_priv->id, &event);
3290         if (ret) {
3291                 /* Destroy the CM ID by returning a non-zero value. */
3292                 id_priv->cm_id.ib = NULL;
3293                 cma_exch(id_priv, RDMA_CM_DESTROYING);
3294                 mutex_unlock(&id_priv->handler_mutex);
3295                 rdma_destroy_id(&id_priv->id);
3296                 return ret;
3297         }
3298 out:
3299         mutex_unlock(&id_priv->handler_mutex);
3300         return ret;
3301 }
3302
3303 static int cma_resolve_ib_udp(struct rdma_id_private *id_priv,
3304                               struct rdma_conn_param *conn_param)
3305 {
3306         struct ib_cm_sidr_req_param req;
3307         struct ib_cm_id *id;
3308         void *private_data;
3309         int offset, ret;
3310
3311         memset(&req, 0, sizeof req);
3312         offset = cma_user_data_offset(id_priv);
3313         req.private_data_len = offset + conn_param->private_data_len;
3314         if (req.private_data_len < conn_param->private_data_len)
3315                 return -EINVAL;
3316
3317         if (req.private_data_len) {
3318                 private_data = kzalloc(req.private_data_len, GFP_ATOMIC);
3319                 if (!private_data)
3320                         return -ENOMEM;
3321         } else {
3322                 private_data = NULL;
3323         }
3324
3325         if (conn_param->private_data && conn_param->private_data_len)
3326                 memcpy(private_data + offset, conn_param->private_data,
3327                        conn_param->private_data_len);
3328
3329         if (private_data) {
3330                 ret = cma_format_hdr(private_data, id_priv);
3331                 if (ret)
3332                         goto out;
3333                 req.private_data = private_data;
3334         }
3335
3336         id = ib_create_cm_id(id_priv->id.device, cma_sidr_rep_handler,
3337                              id_priv);
3338         if (IS_ERR(id)) {
3339                 ret = PTR_ERR(id);
3340                 goto out;
3341         }
3342         id_priv->cm_id.ib = id;
3343
3344         req.path = id_priv->id.route.path_rec;
3345         req.service_id = rdma_get_service_id(&id_priv->id, cma_dst_addr(id_priv));
3346         req.timeout_ms = 1 << (CMA_CM_RESPONSE_TIMEOUT - 8);
3347         req.max_cm_retries = CMA_MAX_CM_RETRIES;
3348
3349         ret = ib_send_cm_sidr_req(id_priv->cm_id.ib, &req);
3350         if (ret) {
3351                 ib_destroy_cm_id(id_priv->cm_id.ib);
3352                 id_priv->cm_id.ib = NULL;
3353         }
3354 out:
3355         kfree(private_data);
3356         return ret;
3357 }
3358
3359 static int cma_connect_ib(struct rdma_id_private *id_priv,
3360                           struct rdma_conn_param *conn_param)
3361 {
3362         struct ib_cm_req_param req;
3363         struct rdma_route *route;
3364         void *private_data;
3365         struct ib_cm_id *id;
3366         int offset, ret;
3367
3368         memset(&req, 0, sizeof req);
3369         offset = cma_user_data_offset(id_priv);
3370         req.private_data_len = offset + conn_param->private_data_len;
3371         if (req.private_data_len < conn_param->private_data_len)
3372                 return -EINVAL;
3373
3374         if (req.private_data_len) {
3375                 private_data = kzalloc(req.private_data_len, GFP_ATOMIC);
3376                 if (!private_data)
3377                         return -ENOMEM;
3378         } else {
3379                 private_data = NULL;
3380         }
3381
3382         if (conn_param->private_data && conn_param->private_data_len)
3383                 memcpy(private_data + offset, conn_param->private_data,
3384                        conn_param->private_data_len);
3385
3386         id = ib_create_cm_id(id_priv->id.device, cma_ib_handler, id_priv);
3387         if (IS_ERR(id)) {
3388                 ret = PTR_ERR(id);
3389                 goto out;
3390         }
3391         id_priv->cm_id.ib = id;
3392
3393         route = &id_priv->id.route;
3394         if (private_data) {
3395                 ret = cma_format_hdr(private_data, id_priv);
3396                 if (ret)
3397                         goto out;
3398                 req.private_data = private_data;
3399         }
3400
3401         req.primary_path = &route->path_rec[0];
3402         if (route->num_paths == 2)
3403                 req.alternate_path = &route->path_rec[1];
3404
3405         req.service_id = rdma_get_service_id(&id_priv->id, cma_dst_addr(id_priv));
3406         req.qp_num = id_priv->qp_num;
3407         req.qp_type = id_priv->id.qp_type;
3408         req.starting_psn = id_priv->seq_num;
3409         req.responder_resources = conn_param->responder_resources;
3410         req.initiator_depth = conn_param->initiator_depth;
3411         req.flow_control = conn_param->flow_control;
3412         req.retry_count = min_t(u8, 7, conn_param->retry_count);
3413         req.rnr_retry_count = min_t(u8, 7, conn_param->rnr_retry_count);
3414         req.remote_cm_response_timeout = CMA_CM_RESPONSE_TIMEOUT;
3415         req.local_cm_response_timeout = CMA_CM_RESPONSE_TIMEOUT;
3416         req.max_cm_retries = CMA_MAX_CM_RETRIES;
3417         req.srq = id_priv->srq ? 1 : 0;
3418
3419         ret = ib_send_cm_req(id_priv->cm_id.ib, &req);
3420 out:
3421         if (ret && !IS_ERR(id)) {
3422                 ib_destroy_cm_id(id);
3423                 id_priv->cm_id.ib = NULL;
3424         }
3425
3426         kfree(private_data);
3427         return ret;
3428 }
3429
3430 static int cma_connect_iw(struct rdma_id_private *id_priv,
3431                           struct rdma_conn_param *conn_param)
3432 {
3433         struct iw_cm_id *cm_id;
3434         int ret;
3435         struct iw_cm_conn_param iw_param;
3436
3437         cm_id = iw_create_cm_id(id_priv->id.device, cma_iw_handler, id_priv);
3438         if (IS_ERR(cm_id))
3439                 return PTR_ERR(cm_id);
3440
3441         cm_id->tos = id_priv->tos;
3442         id_priv->cm_id.iw = cm_id;
3443
3444         memcpy(&cm_id->local_addr, cma_src_addr(id_priv),
3445                rdma_addr_size(cma_src_addr(id_priv)));
3446         memcpy(&cm_id->remote_addr, cma_dst_addr(id_priv),
3447                rdma_addr_size(cma_dst_addr(id_priv)));
3448
3449         ret = cma_modify_qp_rtr(id_priv, conn_param);
3450         if (ret)
3451                 goto out;
3452
3453         if (conn_param) {
3454                 iw_param.ord = conn_param->initiator_depth;
3455                 iw_param.ird = conn_param->responder_resources;
3456                 iw_param.private_data = conn_param->private_data;
3457                 iw_param.private_data_len = conn_param->private_data_len;
3458                 iw_param.qpn = id_priv->id.qp ? id_priv->qp_num : conn_param->qp_num;
3459         } else {
3460                 memset(&iw_param, 0, sizeof iw_param);
3461                 iw_param.qpn = id_priv->qp_num;
3462         }
3463         ret = iw_cm_connect(cm_id, &iw_param);
3464 out:
3465         if (ret) {
3466                 iw_destroy_cm_id(cm_id);
3467                 id_priv->cm_id.iw = NULL;
3468         }
3469         return ret;
3470 }
3471
3472 int rdma_connect(struct rdma_cm_id *id, struct rdma_conn_param *conn_param)
3473 {
3474         struct rdma_id_private *id_priv;
3475         int ret;
3476
3477         id_priv = container_of(id, struct rdma_id_private, id);
3478         if (!cma_comp_exch(id_priv, RDMA_CM_ROUTE_RESOLVED, RDMA_CM_CONNECT))
3479                 return -EINVAL;
3480
3481         if (!id->qp) {
3482                 id_priv->qp_num = conn_param->qp_num;
3483                 id_priv->srq = conn_param->srq;
3484         }
3485
3486         if (rdma_cap_ib_cm(id->device, id->port_num)) {
3487                 if (id->qp_type == IB_QPT_UD)
3488                         ret = cma_resolve_ib_udp(id_priv, conn_param);
3489                 else
3490                         ret = cma_connect_ib(id_priv, conn_param);
3491         } else if (rdma_cap_iw_cm(id->device, id->port_num))
3492                 ret = cma_connect_iw(id_priv, conn_param);
3493         else
3494                 ret = -ENOSYS;
3495         if (ret)
3496                 goto err;
3497
3498         return 0;
3499 err:
3500         cma_comp_exch(id_priv, RDMA_CM_CONNECT, RDMA_CM_ROUTE_RESOLVED);
3501         return ret;
3502 }
3503 EXPORT_SYMBOL(rdma_connect);
3504
3505 static int cma_accept_ib(struct rdma_id_private *id_priv,
3506                          struct rdma_conn_param *conn_param)
3507 {
3508         struct ib_cm_rep_param rep;
3509         int ret;
3510
3511         ret = cma_modify_qp_rtr(id_priv, conn_param);
3512         if (ret)
3513                 goto out;
3514
3515         ret = cma_modify_qp_rts(id_priv, conn_param);
3516         if (ret)
3517                 goto out;
3518
3519         memset(&rep, 0, sizeof rep);
3520         rep.qp_num = id_priv->qp_num;
3521         rep.starting_psn = id_priv->seq_num;
3522         rep.private_data = conn_param->private_data;
3523         rep.private_data_len = conn_param->private_data_len;
3524         rep.responder_resources = conn_param->responder_resources;
3525         rep.initiator_depth = conn_param->initiator_depth;
3526         rep.failover_accepted = 0;
3527         rep.flow_control = conn_param->flow_control;
3528         rep.rnr_retry_count = min_t(u8, 7, conn_param->rnr_retry_count);
3529         rep.srq = id_priv->srq ? 1 : 0;
3530
3531         ret = ib_send_cm_rep(id_priv->cm_id.ib, &rep);
3532 out:
3533         return ret;
3534 }
3535
3536 static int cma_accept_iw(struct rdma_id_private *id_priv,
3537                   struct rdma_conn_param *conn_param)
3538 {
3539         struct iw_cm_conn_param iw_param;
3540         int ret;
3541
3542         ret = cma_modify_qp_rtr(id_priv, conn_param);
3543         if (ret)
3544                 return ret;
3545
3546         iw_param.ord = conn_param->initiator_depth;
3547         iw_param.ird = conn_param->responder_resources;
3548         iw_param.private_data = conn_param->private_data;
3549         iw_param.private_data_len = conn_param->private_data_len;
3550         if (id_priv->id.qp) {
3551                 iw_param.qpn = id_priv->qp_num;
3552         } else
3553                 iw_param.qpn = conn_param->qp_num;
3554
3555         return iw_cm_accept(id_priv->cm_id.iw, &iw_param);
3556 }
3557
3558 static int cma_send_sidr_rep(struct rdma_id_private *id_priv,
3559                              enum ib_cm_sidr_status status, u32 qkey,
3560                              const void *private_data, int private_data_len)
3561 {
3562         struct ib_cm_sidr_rep_param rep;
3563         int ret;
3564
3565         memset(&rep, 0, sizeof rep);
3566         rep.status = status;
3567         if (status == IB_SIDR_SUCCESS) {
3568                 ret = cma_set_qkey(id_priv, qkey);
3569                 if (ret)
3570                         return ret;
3571                 rep.qp_num = id_priv->qp_num;
3572                 rep.qkey = id_priv->qkey;
3573         }
3574         rep.private_data = private_data;
3575         rep.private_data_len = private_data_len;
3576
3577         return ib_send_cm_sidr_rep(id_priv->cm_id.ib, &rep);
3578 }
3579
3580 int rdma_accept(struct rdma_cm_id *id, struct rdma_conn_param *conn_param)
3581 {
3582         struct rdma_id_private *id_priv;
3583         int ret;
3584
3585         id_priv = container_of(id, struct rdma_id_private, id);
3586
3587         id_priv->owner = task_pid_nr(current);
3588
3589         if (!cma_comp(id_priv, RDMA_CM_CONNECT))
3590                 return -EINVAL;
3591
3592         if (!id->qp && conn_param) {
3593                 id_priv->qp_num = conn_param->qp_num;
3594                 id_priv->srq = conn_param->srq;
3595         }
3596
3597         if (rdma_cap_ib_cm(id->device, id->port_num)) {
3598                 if (id->qp_type == IB_QPT_UD) {
3599                         if (conn_param)
3600                                 ret = cma_send_sidr_rep(id_priv, IB_SIDR_SUCCESS,
3601                                                         conn_param->qkey,
3602                                                         conn_param->private_data,
3603                                                         conn_param->private_data_len);
3604                         else
3605                                 ret = cma_send_sidr_rep(id_priv, IB_SIDR_SUCCESS,
3606                                                         0, NULL, 0);
3607                 } else {
3608                         if (conn_param)
3609                                 ret = cma_accept_ib(id_priv, conn_param);
3610                         else
3611                                 ret = cma_rep_recv(id_priv);
3612                 }
3613         } else if (rdma_cap_iw_cm(id->device, id->port_num))
3614                 ret = cma_accept_iw(id_priv, conn_param);
3615         else
3616                 ret = -ENOSYS;
3617
3618         if (ret)
3619                 goto reject;
3620
3621         return 0;
3622 reject:
3623         cma_modify_qp_err(id_priv);
3624         rdma_reject(id, NULL, 0);
3625         return ret;
3626 }
3627 EXPORT_SYMBOL(rdma_accept);
3628
3629 int rdma_notify(struct rdma_cm_id *id, enum ib_event_type event)
3630 {
3631         struct rdma_id_private *id_priv;
3632         int ret;
3633
3634         id_priv = container_of(id, struct rdma_id_private, id);
3635         if (!id_priv->cm_id.ib)
3636                 return -EINVAL;
3637
3638         switch (id->device->node_type) {
3639         case RDMA_NODE_IB_CA:
3640                 ret = ib_cm_notify(id_priv->cm_id.ib, event);
3641                 break;
3642         default:
3643                 ret = 0;
3644                 break;
3645         }
3646         return ret;
3647 }
3648 EXPORT_SYMBOL(rdma_notify);
3649
3650 int rdma_reject(struct rdma_cm_id *id, const void *private_data,
3651                 u8 private_data_len)
3652 {
3653         struct rdma_id_private *id_priv;
3654         int ret;
3655
3656         id_priv = container_of(id, struct rdma_id_private, id);
3657         if (!id_priv->cm_id.ib)
3658                 return -EINVAL;
3659
3660         if (rdma_cap_ib_cm(id->device, id->port_num)) {
3661                 if (id->qp_type == IB_QPT_UD)
3662                         ret = cma_send_sidr_rep(id_priv, IB_SIDR_REJECT, 0,
3663                                                 private_data, private_data_len);
3664                 else
3665                         ret = ib_send_cm_rej(id_priv->cm_id.ib,
3666                                              IB_CM_REJ_CONSUMER_DEFINED, NULL,
3667                                              0, private_data, private_data_len);
3668         } else if (rdma_cap_iw_cm(id->device, id->port_num)) {
3669                 ret = iw_cm_reject(id_priv->cm_id.iw,
3670                                    private_data, private_data_len);
3671         } else
3672                 ret = -ENOSYS;
3673
3674         return ret;
3675 }
3676 EXPORT_SYMBOL(rdma_reject);
3677
3678 int rdma_disconnect(struct rdma_cm_id *id)
3679 {
3680         struct rdma_id_private *id_priv;
3681         int ret;
3682
3683         id_priv = container_of(id, struct rdma_id_private, id);
3684         if (!id_priv->cm_id.ib)
3685                 return -EINVAL;
3686
3687         if (rdma_cap_ib_cm(id->device, id->port_num)) {
3688                 ret = cma_modify_qp_err(id_priv);
3689                 if (ret)
3690                         goto out;
3691                 /* Initiate or respond to a disconnect. */
3692                 if (ib_send_cm_dreq(id_priv->cm_id.ib, NULL, 0))
3693                         ib_send_cm_drep(id_priv->cm_id.ib, NULL, 0);
3694         } else if (rdma_cap_iw_cm(id->device, id->port_num)) {
3695                 ret = iw_cm_disconnect(id_priv->cm_id.iw, 0);
3696         } else
3697                 ret = -EINVAL;
3698
3699 out:
3700         return ret;
3701 }
3702 EXPORT_SYMBOL(rdma_disconnect);
3703
3704 static int cma_ib_mc_handler(int status, struct ib_sa_multicast *multicast)
3705 {
3706         struct rdma_id_private *id_priv;
3707         struct cma_multicast *mc = multicast->context;
3708         struct rdma_cm_event event;
3709         int ret = 0;
3710
3711         id_priv = mc->id_priv;
3712         mutex_lock(&id_priv->handler_mutex);
3713         if (id_priv->state != RDMA_CM_ADDR_BOUND &&
3714             id_priv->state != RDMA_CM_ADDR_RESOLVED)
3715                 goto out;
3716
3717         if (!status)
3718                 status = cma_set_qkey(id_priv, be32_to_cpu(multicast->rec.qkey));
3719         mutex_lock(&id_priv->qp_mutex);
3720         if (!status && id_priv->id.qp)
3721                 status = ib_attach_mcast(id_priv->id.qp, &multicast->rec.mgid,
3722                                          be16_to_cpu(multicast->rec.mlid));
3723         mutex_unlock(&id_priv->qp_mutex);
3724
3725         memset(&event, 0, sizeof event);
3726         event.status = status;
3727         event.param.ud.private_data = mc->context;
3728         if (!status) {
3729                 struct rdma_dev_addr *dev_addr =
3730                         &id_priv->id.route.addr.dev_addr;
3731                 struct net_device *ndev =
3732                         dev_get_by_index(&init_net, dev_addr->bound_dev_if);
3733                 enum ib_gid_type gid_type =
3734                         id_priv->cma_dev->default_gid_type[id_priv->id.port_num -
3735                         rdma_start_port(id_priv->cma_dev->device)];
3736
3737                 event.event = RDMA_CM_EVENT_MULTICAST_JOIN;
3738                 ib_init_ah_from_mcmember(id_priv->id.device,
3739                                          id_priv->id.port_num, &multicast->rec,
3740                                          ndev, gid_type,
3741                                          &event.param.ud.ah_attr);
3742                 event.param.ud.qp_num = 0xFFFFFF;
3743                 event.param.ud.qkey = be32_to_cpu(multicast->rec.qkey);
3744                 if (ndev)
3745                         dev_put(ndev);
3746         } else
3747                 event.event = RDMA_CM_EVENT_MULTICAST_ERROR;
3748
3749         ret = id_priv->id.event_handler(&id_priv->id, &event);
3750         if (ret) {
3751                 cma_exch(id_priv, RDMA_CM_DESTROYING);
3752                 mutex_unlock(&id_priv->handler_mutex);
3753                 rdma_destroy_id(&id_priv->id);
3754                 return 0;
3755         }
3756
3757 out:
3758         mutex_unlock(&id_priv->handler_mutex);
3759         return 0;
3760 }
3761
3762 static void cma_set_mgid(struct rdma_id_private *id_priv,
3763                          struct sockaddr *addr, union ib_gid *mgid)
3764 {
3765         unsigned char mc_map[MAX_ADDR_LEN];
3766         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
3767         struct sockaddr_in *sin = (struct sockaddr_in *) addr;
3768         struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *) addr;
3769
3770         if (cma_any_addr(addr)) {
3771                 memset(mgid, 0, sizeof *mgid);
3772         } else if ((addr->sa_family == AF_INET6) &&
3773                    ((be32_to_cpu(sin6->sin6_addr.s6_addr32[0]) & 0xFFF0FFFF) ==
3774                                                                  0xFF10A01B)) {
3775                 /* IPv6 address is an SA assigned MGID. */
3776                 memcpy(mgid, &sin6->sin6_addr, sizeof *mgid);
3777         } else if (addr->sa_family == AF_IB) {
3778                 memcpy(mgid, &((struct sockaddr_ib *) addr)->sib_addr, sizeof *mgid);
3779         } else if ((addr->sa_family == AF_INET6)) {
3780                 ipv6_ib_mc_map(&sin6->sin6_addr, dev_addr->broadcast, mc_map);
3781                 if (id_priv->id.ps == RDMA_PS_UDP)
3782                         mc_map[7] = 0x01;       /* Use RDMA CM signature */
3783                 *mgid = *(union ib_gid *) (mc_map + 4);
3784         } else {
3785                 ip_ib_mc_map(sin->sin_addr.s_addr, dev_addr->broadcast, mc_map);
3786                 if (id_priv->id.ps == RDMA_PS_UDP)
3787                         mc_map[7] = 0x01;       /* Use RDMA CM signature */
3788                 *mgid = *(union ib_gid *) (mc_map + 4);
3789         }
3790 }
3791
3792 static void cma_query_sa_classport_info_cb(int status,
3793                                            struct ib_class_port_info *rec,
3794                                            void *context)
3795 {
3796         struct class_port_info_context *cb_ctx = context;
3797
3798         WARN_ON(!context);
3799
3800         if (status || !rec) {
3801                 pr_debug("RDMA CM: %s port %u failed query ClassPortInfo status: %d\n",
3802                          cb_ctx->device->name, cb_ctx->port_num, status);
3803                 goto out;
3804         }
3805
3806         memcpy(cb_ctx->class_port_info, rec, sizeof(struct ib_class_port_info));
3807
3808 out:
3809         complete(&cb_ctx->done);
3810 }
3811
3812 static int cma_query_sa_classport_info(struct ib_device *device, u8 port_num,
3813                                        struct ib_class_port_info *class_port_info)
3814 {
3815         struct class_port_info_context *cb_ctx;
3816         int ret;
3817
3818         cb_ctx = kmalloc(sizeof(*cb_ctx), GFP_KERNEL);
3819         if (!cb_ctx)
3820                 return -ENOMEM;
3821
3822         cb_ctx->device = device;
3823         cb_ctx->class_port_info = class_port_info;
3824         cb_ctx->port_num = port_num;
3825         init_completion(&cb_ctx->done);
3826
3827         ret = ib_sa_classport_info_rec_query(&sa_client, device, port_num,
3828                                              CMA_QUERY_CLASSPORT_INFO_TIMEOUT,
3829                                              GFP_KERNEL, cma_query_sa_classport_info_cb,
3830                                              cb_ctx, &cb_ctx->sa_query);
3831         if (ret < 0) {
3832                 pr_err("RDMA CM: %s port %u failed to send ClassPortInfo query, ret: %d\n",
3833                        device->name, port_num, ret);
3834                 goto out;
3835         }
3836
3837         wait_for_completion(&cb_ctx->done);
3838
3839 out:
3840         kfree(cb_ctx);
3841         return ret;
3842 }
3843
3844 static int cma_join_ib_multicast(struct rdma_id_private *id_priv,
3845                                  struct cma_multicast *mc)
3846 {
3847         struct ib_sa_mcmember_rec rec;
3848         struct ib_class_port_info class_port_info;
3849         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
3850         ib_sa_comp_mask comp_mask;
3851         int ret;
3852
3853         ib_addr_get_mgid(dev_addr, &rec.mgid);
3854         ret = ib_sa_get_mcmember_rec(id_priv->id.device, id_priv->id.port_num,
3855                                      &rec.mgid, &rec);
3856         if (ret)
3857                 return ret;
3858
3859         ret = cma_set_qkey(id_priv, 0);
3860         if (ret)
3861                 return ret;
3862
3863         cma_set_mgid(id_priv, (struct sockaddr *) &mc->addr, &rec.mgid);
3864         rec.qkey = cpu_to_be32(id_priv->qkey);
3865         rdma_addr_get_sgid(dev_addr, &rec.port_gid);
3866         rec.pkey = cpu_to_be16(ib_addr_get_pkey(dev_addr));
3867         rec.join_state = mc->join_state;
3868
3869         if (rec.join_state == BIT(SENDONLY_FULLMEMBER_JOIN)) {
3870                 ret = cma_query_sa_classport_info(id_priv->id.device,
3871                                                   id_priv->id.port_num,
3872                                                   &class_port_info);
3873
3874                 if (ret)
3875                         return ret;
3876
3877                 if (!(ib_get_cpi_capmask2(&class_port_info) &
3878                       IB_SA_CAP_MASK2_SENDONLY_FULL_MEM_SUPPORT)) {
3879                         pr_warn("RDMA CM: %s port %u Unable to multicast join\n"
3880                                 "RDMA CM: SM doesn't support Send Only Full Member option\n",
3881                                 id_priv->id.device->name, id_priv->id.port_num);
3882                         return -EOPNOTSUPP;
3883                 }
3884         }
3885
3886         comp_mask = IB_SA_MCMEMBER_REC_MGID | IB_SA_MCMEMBER_REC_PORT_GID |
3887                     IB_SA_MCMEMBER_REC_PKEY | IB_SA_MCMEMBER_REC_JOIN_STATE |
3888                     IB_SA_MCMEMBER_REC_QKEY | IB_SA_MCMEMBER_REC_SL |
3889                     IB_SA_MCMEMBER_REC_FLOW_LABEL |
3890                     IB_SA_MCMEMBER_REC_TRAFFIC_CLASS;
3891
3892         if (id_priv->id.ps == RDMA_PS_IPOIB)
3893                 comp_mask |= IB_SA_MCMEMBER_REC_RATE |
3894                              IB_SA_MCMEMBER_REC_RATE_SELECTOR |
3895                              IB_SA_MCMEMBER_REC_MTU_SELECTOR |
3896                              IB_SA_MCMEMBER_REC_MTU |
3897                              IB_SA_MCMEMBER_REC_HOP_LIMIT;
3898
3899         mc->multicast.ib = ib_sa_join_multicast(&sa_client, id_priv->id.device,
3900                                                 id_priv->id.port_num, &rec,
3901                                                 comp_mask, GFP_KERNEL,
3902                                                 cma_ib_mc_handler, mc);
3903         return PTR_ERR_OR_ZERO(mc->multicast.ib);
3904 }
3905
3906 static void iboe_mcast_work_handler(struct work_struct *work)
3907 {
3908         struct iboe_mcast_work *mw = container_of(work, struct iboe_mcast_work, work);
3909         struct cma_multicast *mc = mw->mc;
3910         struct ib_sa_multicast *m = mc->multicast.ib;
3911
3912         mc->multicast.ib->context = mc;
3913         cma_ib_mc_handler(0, m);
3914         kref_put(&mc->mcref, release_mc);
3915         kfree(mw);
3916 }
3917
3918 static void cma_iboe_set_mgid(struct sockaddr *addr, union ib_gid *mgid)
3919 {
3920         struct sockaddr_in *sin = (struct sockaddr_in *)addr;
3921         struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)addr;
3922
3923         if (cma_any_addr(addr)) {
3924                 memset(mgid, 0, sizeof *mgid);
3925         } else if (addr->sa_family == AF_INET6) {
3926                 memcpy(mgid, &sin6->sin6_addr, sizeof *mgid);
3927         } else {
3928                 mgid->raw[0] = 0xff;
3929                 mgid->raw[1] = 0x0e;
3930                 mgid->raw[2] = 0;
3931                 mgid->raw[3] = 0;
3932                 mgid->raw[4] = 0;
3933                 mgid->raw[5] = 0;
3934                 mgid->raw[6] = 0;
3935                 mgid->raw[7] = 0;
3936                 mgid->raw[8] = 0;
3937                 mgid->raw[9] = 0;
3938                 mgid->raw[10] = 0xff;
3939                 mgid->raw[11] = 0xff;
3940                 *(__be32 *)(&mgid->raw[12]) = sin->sin_addr.s_addr;
3941         }
3942 }
3943
3944 static int cma_iboe_join_multicast(struct rdma_id_private *id_priv,
3945                                    struct cma_multicast *mc)
3946 {
3947         struct iboe_mcast_work *work;
3948         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
3949         int err = 0;
3950         struct sockaddr *addr = (struct sockaddr *)&mc->addr;
3951         struct net_device *ndev = NULL;
3952         enum ib_gid_type gid_type;
3953         bool send_only;
3954
3955         send_only = mc->join_state == BIT(SENDONLY_FULLMEMBER_JOIN);
3956
3957         if (cma_zero_addr((struct sockaddr *)&mc->addr))
3958                 return -EINVAL;
3959
3960         work = kzalloc(sizeof *work, GFP_KERNEL);
3961         if (!work)
3962                 return -ENOMEM;
3963
3964         mc->multicast.ib = kzalloc(sizeof(struct ib_sa_multicast), GFP_KERNEL);
3965         if (!mc->multicast.ib) {
3966                 err = -ENOMEM;
3967                 goto out1;
3968         }
3969
3970         cma_iboe_set_mgid(addr, &mc->multicast.ib->rec.mgid);
3971
3972         mc->multicast.ib->rec.pkey = cpu_to_be16(0xffff);
3973         if (id_priv->id.ps == RDMA_PS_UDP)
3974                 mc->multicast.ib->rec.qkey = cpu_to_be32(RDMA_UDP_QKEY);
3975
3976         if (dev_addr->bound_dev_if)
3977                 ndev = dev_get_by_index(&init_net, dev_addr->bound_dev_if);
3978         if (!ndev) {
3979                 err = -ENODEV;
3980                 goto out2;
3981         }
3982         mc->multicast.ib->rec.rate = iboe_get_rate(ndev);
3983         mc->multicast.ib->rec.hop_limit = 1;
3984         mc->multicast.ib->rec.mtu = iboe_get_mtu(ndev->mtu);
3985
3986         gid_type = id_priv->cma_dev->default_gid_type[id_priv->id.port_num -
3987                    rdma_start_port(id_priv->cma_dev->device)];
3988         if (addr->sa_family == AF_INET) {
3989                 if (gid_type == IB_GID_TYPE_ROCE_UDP_ENCAP) {
3990                         mc->multicast.ib->rec.hop_limit = IPV6_DEFAULT_HOPLIMIT;
3991                         if (!send_only) {
3992                                 err = cma_igmp_send(ndev, &mc->multicast.ib->rec.mgid,
3993                                                     true);
3994                                 if (!err)
3995                                         mc->igmp_joined = true;
3996                         }
3997                 }
3998         } else {
3999                 if (gid_type == IB_GID_TYPE_ROCE_UDP_ENCAP)
4000                         err = -ENOTSUPP;
4001         }
4002         dev_put(ndev);
4003         if (err || !mc->multicast.ib->rec.mtu) {
4004                 if (!err)
4005                         err = -EINVAL;
4006                 goto out2;
4007         }
4008         rdma_ip2gid((struct sockaddr *)&id_priv->id.route.addr.src_addr,
4009                     &mc->multicast.ib->rec.port_gid);
4010         work->id = id_priv;
4011         work->mc = mc;
4012         INIT_WORK(&work->work, iboe_mcast_work_handler);
4013         kref_get(&mc->mcref);
4014         queue_work(cma_wq, &work->work);
4015
4016         return 0;
4017
4018 out2:
4019         kfree(mc->multicast.ib);
4020 out1:
4021         kfree(work);
4022         return err;
4023 }
4024
4025 int rdma_join_multicast(struct rdma_cm_id *id, struct sockaddr *addr,
4026                         u8 join_state, void *context)
4027 {
4028         struct rdma_id_private *id_priv;
4029         struct cma_multicast *mc;
4030         int ret;
4031
4032         id_priv = container_of(id, struct rdma_id_private, id);
4033         if (!cma_comp(id_priv, RDMA_CM_ADDR_BOUND) &&
4034             !cma_comp(id_priv, RDMA_CM_ADDR_RESOLVED))
4035                 return -EINVAL;
4036
4037         mc = kmalloc(sizeof *mc, GFP_KERNEL);
4038         if (!mc)
4039                 return -ENOMEM;
4040
4041         memcpy(&mc->addr, addr, rdma_addr_size(addr));
4042         mc->context = context;
4043         mc->id_priv = id_priv;
4044         mc->igmp_joined = false;
4045         mc->join_state = join_state;
4046         spin_lock(&id_priv->lock);
4047         list_add(&mc->list, &id_priv->mc_list);
4048         spin_unlock(&id_priv->lock);
4049
4050         if (rdma_protocol_roce(id->device, id->port_num)) {
4051                 kref_init(&mc->mcref);
4052                 ret = cma_iboe_join_multicast(id_priv, mc);
4053         } else if (rdma_cap_ib_mcast(id->device, id->port_num))
4054                 ret = cma_join_ib_multicast(id_priv, mc);
4055         else
4056                 ret = -ENOSYS;
4057
4058         if (ret) {
4059                 spin_lock_irq(&id_priv->lock);
4060                 list_del(&mc->list);
4061                 spin_unlock_irq(&id_priv->lock);
4062                 kfree(mc);
4063         }
4064         return ret;
4065 }
4066 EXPORT_SYMBOL(rdma_join_multicast);
4067
4068 void rdma_leave_multicast(struct rdma_cm_id *id, struct sockaddr *addr)
4069 {
4070         struct rdma_id_private *id_priv;
4071         struct cma_multicast *mc;
4072
4073         id_priv = container_of(id, struct rdma_id_private, id);
4074         spin_lock_irq(&id_priv->lock);
4075         list_for_each_entry(mc, &id_priv->mc_list, list) {
4076                 if (!memcmp(&mc->addr, addr, rdma_addr_size(addr))) {
4077                         list_del(&mc->list);
4078                         spin_unlock_irq(&id_priv->lock);
4079
4080                         if (id->qp)
4081                                 ib_detach_mcast(id->qp,
4082                                                 &mc->multicast.ib->rec.mgid,
4083                                                 be16_to_cpu(mc->multicast.ib->rec.mlid));
4084
4085                         BUG_ON(id_priv->cma_dev->device != id->device);
4086
4087                         if (rdma_cap_ib_mcast(id->device, id->port_num)) {
4088                                 ib_sa_free_multicast(mc->multicast.ib);
4089                                 kfree(mc);
4090                         } else if (rdma_protocol_roce(id->device, id->port_num)) {
4091                                 if (mc->igmp_joined) {
4092                                         struct rdma_dev_addr *dev_addr =
4093                                                 &id->route.addr.dev_addr;
4094                                         struct net_device *ndev = NULL;
4095
4096                                         if (dev_addr->bound_dev_if)
4097                                                 ndev = dev_get_by_index(&init_net,
4098                                                                         dev_addr->bound_dev_if);
4099                                         if (ndev) {
4100                                                 cma_igmp_send(ndev,
4101                                                               &mc->multicast.ib->rec.mgid,
4102                                                               false);
4103                                                 dev_put(ndev);
4104                                         }
4105                                         mc->igmp_joined = false;
4106                                 }
4107                                 kref_put(&mc->mcref, release_mc);
4108                         }
4109                         return;
4110                 }
4111         }
4112         spin_unlock_irq(&id_priv->lock);
4113 }
4114 EXPORT_SYMBOL(rdma_leave_multicast);
4115
4116 static int cma_netdev_change(struct net_device *ndev, struct rdma_id_private *id_priv)
4117 {
4118         struct rdma_dev_addr *dev_addr;
4119         struct cma_ndev_work *work;
4120
4121         dev_addr = &id_priv->id.route.addr.dev_addr;
4122
4123         if ((dev_addr->bound_dev_if == ndev->ifindex) &&
4124             (net_eq(dev_net(ndev), dev_addr->net)) &&
4125             memcmp(dev_addr->src_dev_addr, ndev->dev_addr, ndev->addr_len)) {
4126                 pr_info("RDMA CM addr change for ndev %s used by id %p\n",
4127                         ndev->name, &id_priv->id);
4128                 work = kzalloc(sizeof *work, GFP_KERNEL);
4129                 if (!work)
4130                         return -ENOMEM;
4131
4132                 INIT_WORK(&work->work, cma_ndev_work_handler);
4133                 work->id = id_priv;
4134                 work->event.event = RDMA_CM_EVENT_ADDR_CHANGE;
4135                 atomic_inc(&id_priv->refcount);
4136                 queue_work(cma_wq, &work->work);
4137         }
4138
4139         return 0;
4140 }
4141
4142 static int cma_netdev_callback(struct notifier_block *self, unsigned long event,
4143                                void *ptr)
4144 {
4145         struct net_device *ndev = netdev_notifier_info_to_dev(ptr);
4146         struct cma_device *cma_dev;
4147         struct rdma_id_private *id_priv;
4148         int ret = NOTIFY_DONE;
4149
4150         if (event != NETDEV_BONDING_FAILOVER)
4151                 return NOTIFY_DONE;
4152
4153         if (!(ndev->flags & IFF_MASTER) || !(ndev->priv_flags & IFF_BONDING))
4154                 return NOTIFY_DONE;
4155
4156         mutex_lock(&lock);
4157         list_for_each_entry(cma_dev, &dev_list, list)
4158                 list_for_each_entry(id_priv, &cma_dev->id_list, list) {
4159                         ret = cma_netdev_change(ndev, id_priv);
4160                         if (ret)
4161                                 goto out;
4162                 }
4163
4164 out:
4165         mutex_unlock(&lock);
4166         return ret;
4167 }
4168
4169 static struct notifier_block cma_nb = {
4170         .notifier_call = cma_netdev_callback
4171 };
4172
4173 static void cma_add_one(struct ib_device *device)
4174 {
4175         struct cma_device *cma_dev;
4176         struct rdma_id_private *id_priv;
4177         unsigned int i;
4178         unsigned long supported_gids = 0;
4179
4180         cma_dev = kmalloc(sizeof *cma_dev, GFP_KERNEL);
4181         if (!cma_dev)
4182                 return;
4183
4184         cma_dev->device = device;
4185         cma_dev->default_gid_type = kcalloc(device->phys_port_cnt,
4186                                             sizeof(*cma_dev->default_gid_type),
4187                                             GFP_KERNEL);
4188         if (!cma_dev->default_gid_type) {
4189                 kfree(cma_dev);
4190                 return;
4191         }
4192         for (i = rdma_start_port(device); i <= rdma_end_port(device); i++) {
4193                 supported_gids = roce_gid_type_mask_support(device, i);
4194                 WARN_ON(!supported_gids);
4195                 cma_dev->default_gid_type[i - rdma_start_port(device)] =
4196                         find_first_bit(&supported_gids, BITS_PER_LONG);
4197         }
4198
4199         init_completion(&cma_dev->comp);
4200         atomic_set(&cma_dev->refcount, 1);
4201         INIT_LIST_HEAD(&cma_dev->id_list);
4202         ib_set_client_data(device, &cma_client, cma_dev);
4203
4204         mutex_lock(&lock);
4205         list_add_tail(&cma_dev->list, &dev_list);
4206         list_for_each_entry(id_priv, &listen_any_list, list)
4207                 cma_listen_on_dev(id_priv, cma_dev);
4208         mutex_unlock(&lock);
4209 }
4210
4211 static int cma_remove_id_dev(struct rdma_id_private *id_priv)
4212 {
4213         struct rdma_cm_event event;
4214         enum rdma_cm_state state;
4215         int ret = 0;
4216
4217         /* Record that we want to remove the device */
4218         state = cma_exch(id_priv, RDMA_CM_DEVICE_REMOVAL);
4219         if (state == RDMA_CM_DESTROYING)
4220                 return 0;
4221
4222         cma_cancel_operation(id_priv, state);
4223         mutex_lock(&id_priv->handler_mutex);
4224
4225         /* Check for destruction from another callback. */
4226         if (!cma_comp(id_priv, RDMA_CM_DEVICE_REMOVAL))
4227                 goto out;
4228
4229         memset(&event, 0, sizeof event);
4230         event.event = RDMA_CM_EVENT_DEVICE_REMOVAL;
4231         ret = id_priv->id.event_handler(&id_priv->id, &event);
4232 out:
4233         mutex_unlock(&id_priv->handler_mutex);
4234         return ret;
4235 }
4236
4237 static void cma_process_remove(struct cma_device *cma_dev)
4238 {
4239         struct rdma_id_private *id_priv;
4240         int ret;
4241
4242         mutex_lock(&lock);
4243         while (!list_empty(&cma_dev->id_list)) {
4244                 id_priv = list_entry(cma_dev->id_list.next,
4245                                      struct rdma_id_private, list);
4246
4247                 list_del(&id_priv->listen_list);
4248                 list_del_init(&id_priv->list);
4249                 atomic_inc(&id_priv->refcount);
4250                 mutex_unlock(&lock);
4251
4252                 ret = id_priv->internal_id ? 1 : cma_remove_id_dev(id_priv);
4253                 cma_deref_id(id_priv);
4254                 if (ret)
4255                         rdma_destroy_id(&id_priv->id);
4256
4257                 mutex_lock(&lock);
4258         }
4259         mutex_unlock(&lock);
4260
4261         cma_deref_dev(cma_dev);
4262         wait_for_completion(&cma_dev->comp);
4263 }
4264
4265 static void cma_remove_one(struct ib_device *device, void *client_data)
4266 {
4267         struct cma_device *cma_dev = client_data;
4268
4269         if (!cma_dev)
4270                 return;
4271
4272         mutex_lock(&lock);
4273         list_del(&cma_dev->list);
4274         mutex_unlock(&lock);
4275
4276         cma_process_remove(cma_dev);
4277         kfree(cma_dev->default_gid_type);
4278         kfree(cma_dev);
4279 }
4280
4281 static int cma_get_id_stats(struct sk_buff *skb, struct netlink_callback *cb)
4282 {
4283         struct nlmsghdr *nlh;
4284         struct rdma_cm_id_stats *id_stats;
4285         struct rdma_id_private *id_priv;
4286         struct rdma_cm_id *id = NULL;
4287         struct cma_device *cma_dev;
4288         int i_dev = 0, i_id = 0;
4289
4290         /*
4291          * We export all of the IDs as a sequence of messages.  Each
4292          * ID gets its own netlink message.
4293          */
4294         mutex_lock(&lock);
4295
4296         list_for_each_entry(cma_dev, &dev_list, list) {
4297                 if (i_dev < cb->args[0]) {
4298                         i_dev++;
4299                         continue;
4300                 }
4301
4302                 i_id = 0;
4303                 list_for_each_entry(id_priv, &cma_dev->id_list, list) {
4304                         if (i_id < cb->args[1]) {
4305                                 i_id++;
4306                                 continue;
4307                         }
4308
4309                         id_stats = ibnl_put_msg(skb, &nlh, cb->nlh->nlmsg_seq,
4310                                                 sizeof *id_stats, RDMA_NL_RDMA_CM,
4311                                                 RDMA_NL_RDMA_CM_ID_STATS,
4312                                                 NLM_F_MULTI);
4313                         if (!id_stats)
4314                                 goto out;
4315
4316                         memset(id_stats, 0, sizeof *id_stats);
4317                         id = &id_priv->id;
4318                         id_stats->node_type = id->route.addr.dev_addr.dev_type;
4319                         id_stats->port_num = id->port_num;
4320                         id_stats->bound_dev_if =
4321                                 id->route.addr.dev_addr.bound_dev_if;
4322
4323                         if (ibnl_put_attr(skb, nlh,
4324                                           rdma_addr_size(cma_src_addr(id_priv)),
4325                                           cma_src_addr(id_priv),
4326                                           RDMA_NL_RDMA_CM_ATTR_SRC_ADDR))
4327                                 goto out;
4328                         if (ibnl_put_attr(skb, nlh,
4329                                           rdma_addr_size(cma_src_addr(id_priv)),
4330                                           cma_dst_addr(id_priv),
4331                                           RDMA_NL_RDMA_CM_ATTR_DST_ADDR))
4332                                 goto out;
4333
4334                         id_stats->pid           = id_priv->owner;
4335                         id_stats->port_space    = id->ps;
4336                         id_stats->cm_state      = id_priv->state;
4337                         id_stats->qp_num        = id_priv->qp_num;
4338                         id_stats->qp_type       = id->qp_type;
4339
4340                         i_id++;
4341                 }
4342
4343                 cb->args[1] = 0;
4344                 i_dev++;
4345         }
4346
4347 out:
4348         mutex_unlock(&lock);
4349         cb->args[0] = i_dev;
4350         cb->args[1] = i_id;
4351
4352         return skb->len;
4353 }
4354
4355 static const struct ibnl_client_cbs cma_cb_table[] = {
4356         [RDMA_NL_RDMA_CM_ID_STATS] = { .dump = cma_get_id_stats,
4357                                        .module = THIS_MODULE },
4358 };
4359
4360 static int cma_init_net(struct net *net)
4361 {
4362         struct cma_pernet *pernet = cma_pernet(net);
4363
4364         idr_init(&pernet->tcp_ps);
4365         idr_init(&pernet->udp_ps);
4366         idr_init(&pernet->ipoib_ps);
4367         idr_init(&pernet->ib_ps);
4368
4369         return 0;
4370 }
4371
4372 static void cma_exit_net(struct net *net)
4373 {
4374         struct cma_pernet *pernet = cma_pernet(net);
4375
4376         idr_destroy(&pernet->tcp_ps);
4377         idr_destroy(&pernet->udp_ps);
4378         idr_destroy(&pernet->ipoib_ps);
4379         idr_destroy(&pernet->ib_ps);
4380 }
4381
4382 static struct pernet_operations cma_pernet_operations = {
4383         .init = cma_init_net,
4384         .exit = cma_exit_net,
4385         .id = &cma_pernet_id,
4386         .size = sizeof(struct cma_pernet),
4387 };
4388
4389 static int __init cma_init(void)
4390 {
4391         int ret;
4392
4393         cma_wq = alloc_ordered_workqueue("rdma_cm", WQ_MEM_RECLAIM);
4394         if (!cma_wq)
4395                 return -ENOMEM;
4396
4397         ret = register_pernet_subsys(&cma_pernet_operations);
4398         if (ret)
4399                 goto err_wq;
4400
4401         ib_sa_register_client(&sa_client);
4402         rdma_addr_register_client(&addr_client);
4403         register_netdevice_notifier(&cma_nb);
4404
4405         ret = ib_register_client(&cma_client);
4406         if (ret)
4407                 goto err;
4408
4409         if (ibnl_add_client(RDMA_NL_RDMA_CM, ARRAY_SIZE(cma_cb_table),
4410                             cma_cb_table))
4411                 pr_warn("RDMA CMA: failed to add netlink callback\n");
4412         cma_configfs_init();
4413
4414         return 0;
4415
4416 err:
4417         unregister_netdevice_notifier(&cma_nb);
4418         rdma_addr_unregister_client(&addr_client);
4419         ib_sa_unregister_client(&sa_client);
4420 err_wq:
4421         destroy_workqueue(cma_wq);
4422         return ret;
4423 }
4424
4425 static void __exit cma_cleanup(void)
4426 {
4427         cma_configfs_exit();
4428         ibnl_remove_client(RDMA_NL_RDMA_CM);
4429         ib_unregister_client(&cma_client);
4430         unregister_netdevice_notifier(&cma_nb);
4431         rdma_addr_unregister_client(&addr_client);
4432         ib_sa_unregister_client(&sa_client);
4433         unregister_pernet_subsys(&cma_pernet_operations);
4434         destroy_workqueue(cma_wq);
4435 }
4436
4437 module_init(cma_init);
4438 module_exit(cma_cleanup);