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 int cma_resolve_iboe_route(struct rdma_id_private *id_priv)
2442 {
2443         struct rdma_route *route = &id_priv->id.route;
2444         struct rdma_addr *addr = &route->addr;
2445         struct cma_work *work;
2446         int ret;
2447         struct net_device *ndev = NULL;
2448
2449
2450         work = kzalloc(sizeof *work, GFP_KERNEL);
2451         if (!work)
2452                 return -ENOMEM;
2453
2454         work->id = id_priv;
2455         INIT_WORK(&work->work, cma_work_handler);
2456
2457         route->path_rec = kzalloc(sizeof *route->path_rec, GFP_KERNEL);
2458         if (!route->path_rec) {
2459                 ret = -ENOMEM;
2460                 goto err1;
2461         }
2462
2463         route->num_paths = 1;
2464
2465         if (addr->dev_addr.bound_dev_if) {
2466                 ndev = dev_get_by_index(&init_net, addr->dev_addr.bound_dev_if);
2467                 if (!ndev) {
2468                         ret = -ENODEV;
2469                         goto err2;
2470                 }
2471
2472                 if (ndev->flags & IFF_LOOPBACK) {
2473                         dev_put(ndev);
2474                         if (!id_priv->id.device->get_netdev) {
2475                                 ret = -EOPNOTSUPP;
2476                                 goto err2;
2477                         }
2478
2479                         ndev = id_priv->id.device->get_netdev(id_priv->id.device,
2480                                                               id_priv->id.port_num);
2481                         if (!ndev) {
2482                                 ret = -ENODEV;
2483                                 goto err2;
2484                         }
2485                 }
2486
2487                 route->path_rec->net = &init_net;
2488                 route->path_rec->ifindex = ndev->ifindex;
2489                 route->path_rec->gid_type = id_priv->gid_type;
2490         }
2491         if (!ndev) {
2492                 ret = -ENODEV;
2493                 goto err2;
2494         }
2495
2496         memcpy(route->path_rec->dmac, addr->dev_addr.dst_dev_addr, ETH_ALEN);
2497
2498         rdma_ip2gid((struct sockaddr *)&id_priv->id.route.addr.src_addr,
2499                     &route->path_rec->sgid);
2500         rdma_ip2gid((struct sockaddr *)&id_priv->id.route.addr.dst_addr,
2501                     &route->path_rec->dgid);
2502
2503         /* Use the hint from IP Stack to select GID Type */
2504         if (route->path_rec->gid_type < ib_network_to_gid_type(addr->dev_addr.network))
2505                 route->path_rec->gid_type = ib_network_to_gid_type(addr->dev_addr.network);
2506         if (((struct sockaddr *)&id_priv->id.route.addr.dst_addr)->sa_family != AF_IB)
2507                 /* TODO: get the hoplimit from the inet/inet6 device */
2508                 route->path_rec->hop_limit = addr->dev_addr.hoplimit;
2509         else
2510                 route->path_rec->hop_limit = 1;
2511         route->path_rec->reversible = 1;
2512         route->path_rec->pkey = cpu_to_be16(0xffff);
2513         route->path_rec->mtu_selector = IB_SA_EQ;
2514         route->path_rec->sl = iboe_tos_to_sl(ndev, id_priv->tos);
2515         route->path_rec->mtu = iboe_get_mtu(ndev->mtu);
2516         route->path_rec->rate_selector = IB_SA_EQ;
2517         route->path_rec->rate = iboe_get_rate(ndev);
2518         dev_put(ndev);
2519         route->path_rec->packet_life_time_selector = IB_SA_EQ;
2520         route->path_rec->packet_life_time = CMA_IBOE_PACKET_LIFETIME;
2521         if (!route->path_rec->mtu) {
2522                 ret = -EINVAL;
2523                 goto err2;
2524         }
2525
2526         work->old_state = RDMA_CM_ROUTE_QUERY;
2527         work->new_state = RDMA_CM_ROUTE_RESOLVED;
2528         work->event.event = RDMA_CM_EVENT_ROUTE_RESOLVED;
2529         work->event.status = 0;
2530
2531         queue_work(cma_wq, &work->work);
2532
2533         return 0;
2534
2535 err2:
2536         kfree(route->path_rec);
2537         route->path_rec = NULL;
2538 err1:
2539         kfree(work);
2540         return ret;
2541 }
2542
2543 int rdma_resolve_route(struct rdma_cm_id *id, int timeout_ms)
2544 {
2545         struct rdma_id_private *id_priv;
2546         int ret;
2547
2548         id_priv = container_of(id, struct rdma_id_private, id);
2549         if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_RESOLVED, RDMA_CM_ROUTE_QUERY))
2550                 return -EINVAL;
2551
2552         atomic_inc(&id_priv->refcount);
2553         if (rdma_cap_ib_sa(id->device, id->port_num))
2554                 ret = cma_resolve_ib_route(id_priv, timeout_ms);
2555         else if (rdma_protocol_roce(id->device, id->port_num))
2556                 ret = cma_resolve_iboe_route(id_priv);
2557         else if (rdma_protocol_iwarp(id->device, id->port_num))
2558                 ret = cma_resolve_iw_route(id_priv, timeout_ms);
2559         else
2560                 ret = -ENOSYS;
2561
2562         if (ret)
2563                 goto err;
2564
2565         return 0;
2566 err:
2567         cma_comp_exch(id_priv, RDMA_CM_ROUTE_QUERY, RDMA_CM_ADDR_RESOLVED);
2568         cma_deref_id(id_priv);
2569         return ret;
2570 }
2571 EXPORT_SYMBOL(rdma_resolve_route);
2572
2573 static void cma_set_loopback(struct sockaddr *addr)
2574 {
2575         switch (addr->sa_family) {
2576         case AF_INET:
2577                 ((struct sockaddr_in *) addr)->sin_addr.s_addr = htonl(INADDR_LOOPBACK);
2578                 break;
2579         case AF_INET6:
2580                 ipv6_addr_set(&((struct sockaddr_in6 *) addr)->sin6_addr,
2581                               0, 0, 0, htonl(1));
2582                 break;
2583         default:
2584                 ib_addr_set(&((struct sockaddr_ib *) addr)->sib_addr,
2585                             0, 0, 0, htonl(1));
2586                 break;
2587         }
2588 }
2589
2590 static int cma_bind_loopback(struct rdma_id_private *id_priv)
2591 {
2592         struct cma_device *cma_dev, *cur_dev;
2593         struct ib_port_attr port_attr;
2594         union ib_gid gid;
2595         u16 pkey;
2596         int ret;
2597         u8 p;
2598
2599         cma_dev = NULL;
2600         mutex_lock(&lock);
2601         list_for_each_entry(cur_dev, &dev_list, list) {
2602                 if (cma_family(id_priv) == AF_IB &&
2603                     !rdma_cap_ib_cm(cur_dev->device, 1))
2604                         continue;
2605
2606                 if (!cma_dev)
2607                         cma_dev = cur_dev;
2608
2609                 for (p = 1; p <= cur_dev->device->phys_port_cnt; ++p) {
2610                         if (!ib_query_port(cur_dev->device, p, &port_attr) &&
2611                             port_attr.state == IB_PORT_ACTIVE) {
2612                                 cma_dev = cur_dev;
2613                                 goto port_found;
2614                         }
2615                 }
2616         }
2617
2618         if (!cma_dev) {
2619                 ret = -ENODEV;
2620                 goto out;
2621         }
2622
2623         p = 1;
2624
2625 port_found:
2626         ret = ib_get_cached_gid(cma_dev->device, p, 0, &gid, NULL);
2627         if (ret)
2628                 goto out;
2629
2630         ret = ib_get_cached_pkey(cma_dev->device, p, 0, &pkey);
2631         if (ret)
2632                 goto out;
2633
2634         id_priv->id.route.addr.dev_addr.dev_type =
2635                 (rdma_protocol_ib(cma_dev->device, p)) ?
2636                 ARPHRD_INFINIBAND : ARPHRD_ETHER;
2637
2638         rdma_addr_set_sgid(&id_priv->id.route.addr.dev_addr, &gid);
2639         ib_addr_set_pkey(&id_priv->id.route.addr.dev_addr, pkey);
2640         id_priv->id.port_num = p;
2641         cma_attach_to_dev(id_priv, cma_dev);
2642         cma_set_loopback(cma_src_addr(id_priv));
2643 out:
2644         mutex_unlock(&lock);
2645         return ret;
2646 }
2647
2648 static void addr_handler(int status, struct sockaddr *src_addr,
2649                          struct rdma_dev_addr *dev_addr, void *context)
2650 {
2651         struct rdma_id_private *id_priv = context;
2652         struct rdma_cm_event event;
2653
2654         memset(&event, 0, sizeof event);
2655         mutex_lock(&id_priv->handler_mutex);
2656         if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_QUERY,
2657                            RDMA_CM_ADDR_RESOLVED))
2658                 goto out;
2659
2660         memcpy(cma_src_addr(id_priv), src_addr, rdma_addr_size(src_addr));
2661         if (!status && !id_priv->cma_dev)
2662                 status = cma_acquire_dev(id_priv, NULL);
2663
2664         if (status) {
2665                 if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_RESOLVED,
2666                                    RDMA_CM_ADDR_BOUND))
2667                         goto out;
2668                 event.event = RDMA_CM_EVENT_ADDR_ERROR;
2669                 event.status = status;
2670         } else
2671                 event.event = RDMA_CM_EVENT_ADDR_RESOLVED;
2672
2673         if (id_priv->id.event_handler(&id_priv->id, &event)) {
2674                 cma_exch(id_priv, RDMA_CM_DESTROYING);
2675                 mutex_unlock(&id_priv->handler_mutex);
2676                 cma_deref_id(id_priv);
2677                 rdma_destroy_id(&id_priv->id);
2678                 return;
2679         }
2680 out:
2681         mutex_unlock(&id_priv->handler_mutex);
2682         cma_deref_id(id_priv);
2683 }
2684
2685 static int cma_resolve_loopback(struct rdma_id_private *id_priv)
2686 {
2687         struct cma_work *work;
2688         union ib_gid gid;
2689         int ret;
2690
2691         work = kzalloc(sizeof *work, GFP_KERNEL);
2692         if (!work)
2693                 return -ENOMEM;
2694
2695         if (!id_priv->cma_dev) {
2696                 ret = cma_bind_loopback(id_priv);
2697                 if (ret)
2698                         goto err;
2699         }
2700
2701         rdma_addr_get_sgid(&id_priv->id.route.addr.dev_addr, &gid);
2702         rdma_addr_set_dgid(&id_priv->id.route.addr.dev_addr, &gid);
2703
2704         work->id = id_priv;
2705         INIT_WORK(&work->work, cma_work_handler);
2706         work->old_state = RDMA_CM_ADDR_QUERY;
2707         work->new_state = RDMA_CM_ADDR_RESOLVED;
2708         work->event.event = RDMA_CM_EVENT_ADDR_RESOLVED;
2709         queue_work(cma_wq, &work->work);
2710         return 0;
2711 err:
2712         kfree(work);
2713         return ret;
2714 }
2715
2716 static int cma_resolve_ib_addr(struct rdma_id_private *id_priv)
2717 {
2718         struct cma_work *work;
2719         int ret;
2720
2721         work = kzalloc(sizeof *work, GFP_KERNEL);
2722         if (!work)
2723                 return -ENOMEM;
2724
2725         if (!id_priv->cma_dev) {
2726                 ret = cma_resolve_ib_dev(id_priv);
2727                 if (ret)
2728                         goto err;
2729         }
2730
2731         rdma_addr_set_dgid(&id_priv->id.route.addr.dev_addr, (union ib_gid *)
2732                 &(((struct sockaddr_ib *) &id_priv->id.route.addr.dst_addr)->sib_addr));
2733
2734         work->id = id_priv;
2735         INIT_WORK(&work->work, cma_work_handler);
2736         work->old_state = RDMA_CM_ADDR_QUERY;
2737         work->new_state = RDMA_CM_ADDR_RESOLVED;
2738         work->event.event = RDMA_CM_EVENT_ADDR_RESOLVED;
2739         queue_work(cma_wq, &work->work);
2740         return 0;
2741 err:
2742         kfree(work);
2743         return ret;
2744 }
2745
2746 static int cma_bind_addr(struct rdma_cm_id *id, struct sockaddr *src_addr,
2747                          struct sockaddr *dst_addr)
2748 {
2749         if (!src_addr || !src_addr->sa_family) {
2750                 src_addr = (struct sockaddr *) &id->route.addr.src_addr;
2751                 src_addr->sa_family = dst_addr->sa_family;
2752                 if (dst_addr->sa_family == AF_INET6) {
2753                         struct sockaddr_in6 *src_addr6 = (struct sockaddr_in6 *) src_addr;
2754                         struct sockaddr_in6 *dst_addr6 = (struct sockaddr_in6 *) dst_addr;
2755                         src_addr6->sin6_scope_id = dst_addr6->sin6_scope_id;
2756                         if (ipv6_addr_type(&dst_addr6->sin6_addr) & IPV6_ADDR_LINKLOCAL)
2757                                 id->route.addr.dev_addr.bound_dev_if = dst_addr6->sin6_scope_id;
2758                 } else if (dst_addr->sa_family == AF_IB) {
2759                         ((struct sockaddr_ib *) src_addr)->sib_pkey =
2760                                 ((struct sockaddr_ib *) dst_addr)->sib_pkey;
2761                 }
2762         }
2763         return rdma_bind_addr(id, src_addr);
2764 }
2765
2766 int rdma_resolve_addr(struct rdma_cm_id *id, struct sockaddr *src_addr,
2767                       struct sockaddr *dst_addr, int timeout_ms)
2768 {
2769         struct rdma_id_private *id_priv;
2770         int ret;
2771
2772         id_priv = container_of(id, struct rdma_id_private, id);
2773         if (id_priv->state == RDMA_CM_IDLE) {
2774                 ret = cma_bind_addr(id, src_addr, dst_addr);
2775                 if (ret)
2776                         return ret;
2777         }
2778
2779         if (cma_family(id_priv) != dst_addr->sa_family)
2780                 return -EINVAL;
2781
2782         if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_BOUND, RDMA_CM_ADDR_QUERY))
2783                 return -EINVAL;
2784
2785         atomic_inc(&id_priv->refcount);
2786         memcpy(cma_dst_addr(id_priv), dst_addr, rdma_addr_size(dst_addr));
2787         if (cma_any_addr(dst_addr)) {
2788                 ret = cma_resolve_loopback(id_priv);
2789         } else {
2790                 if (dst_addr->sa_family == AF_IB) {
2791                         ret = cma_resolve_ib_addr(id_priv);
2792                 } else {
2793                         ret = rdma_resolve_ip(&addr_client, cma_src_addr(id_priv),
2794                                               dst_addr, &id->route.addr.dev_addr,
2795                                               timeout_ms, addr_handler, id_priv);
2796                 }
2797         }
2798         if (ret)
2799                 goto err;
2800
2801         return 0;
2802 err:
2803         cma_comp_exch(id_priv, RDMA_CM_ADDR_QUERY, RDMA_CM_ADDR_BOUND);
2804         cma_deref_id(id_priv);
2805         return ret;
2806 }
2807 EXPORT_SYMBOL(rdma_resolve_addr);
2808
2809 int rdma_set_reuseaddr(struct rdma_cm_id *id, int reuse)
2810 {
2811         struct rdma_id_private *id_priv;
2812         unsigned long flags;
2813         int ret;
2814
2815         id_priv = container_of(id, struct rdma_id_private, id);
2816         spin_lock_irqsave(&id_priv->lock, flags);
2817         if (reuse || id_priv->state == RDMA_CM_IDLE) {
2818                 id_priv->reuseaddr = reuse;
2819                 ret = 0;
2820         } else {
2821                 ret = -EINVAL;
2822         }
2823         spin_unlock_irqrestore(&id_priv->lock, flags);
2824         return ret;
2825 }
2826 EXPORT_SYMBOL(rdma_set_reuseaddr);
2827
2828 int rdma_set_afonly(struct rdma_cm_id *id, int afonly)
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 (id_priv->state == RDMA_CM_IDLE || id_priv->state == RDMA_CM_ADDR_BOUND) {
2837                 id_priv->options |= (1 << CMA_OPTION_AFONLY);
2838                 id_priv->afonly = afonly;
2839                 ret = 0;
2840         } else {
2841                 ret = -EINVAL;
2842         }
2843         spin_unlock_irqrestore(&id_priv->lock, flags);
2844         return ret;
2845 }
2846 EXPORT_SYMBOL(rdma_set_afonly);
2847
2848 static void cma_bind_port(struct rdma_bind_list *bind_list,
2849                           struct rdma_id_private *id_priv)
2850 {
2851         struct sockaddr *addr;
2852         struct sockaddr_ib *sib;
2853         u64 sid, mask;
2854         __be16 port;
2855
2856         addr = cma_src_addr(id_priv);
2857         port = htons(bind_list->port);
2858
2859         switch (addr->sa_family) {
2860         case AF_INET:
2861                 ((struct sockaddr_in *) addr)->sin_port = port;
2862                 break;
2863         case AF_INET6:
2864                 ((struct sockaddr_in6 *) addr)->sin6_port = port;
2865                 break;
2866         case AF_IB:
2867                 sib = (struct sockaddr_ib *) addr;
2868                 sid = be64_to_cpu(sib->sib_sid);
2869                 mask = be64_to_cpu(sib->sib_sid_mask);
2870                 sib->sib_sid = cpu_to_be64((sid & mask) | (u64) ntohs(port));
2871                 sib->sib_sid_mask = cpu_to_be64(~0ULL);
2872                 break;
2873         }
2874         id_priv->bind_list = bind_list;
2875         hlist_add_head(&id_priv->node, &bind_list->owners);
2876 }
2877
2878 static int cma_alloc_port(enum rdma_port_space ps,
2879                           struct rdma_id_private *id_priv, unsigned short snum)
2880 {
2881         struct rdma_bind_list *bind_list;
2882         int ret;
2883
2884         bind_list = kzalloc(sizeof *bind_list, GFP_KERNEL);
2885         if (!bind_list)
2886                 return -ENOMEM;
2887
2888         ret = cma_ps_alloc(id_priv->id.route.addr.dev_addr.net, ps, bind_list,
2889                            snum);
2890         if (ret < 0)
2891                 goto err;
2892
2893         bind_list->ps = ps;
2894         bind_list->port = (unsigned short)ret;
2895         cma_bind_port(bind_list, id_priv);
2896         return 0;
2897 err:
2898         kfree(bind_list);
2899         return ret == -ENOSPC ? -EADDRNOTAVAIL : ret;
2900 }
2901
2902 static int cma_alloc_any_port(enum rdma_port_space ps,
2903                               struct rdma_id_private *id_priv)
2904 {
2905         static unsigned int last_used_port;
2906         int low, high, remaining;
2907         unsigned int rover;
2908         struct net *net = id_priv->id.route.addr.dev_addr.net;
2909
2910         inet_get_local_port_range(net, &low, &high);
2911         remaining = (high - low) + 1;
2912         rover = prandom_u32() % remaining + low;
2913 retry:
2914         if (last_used_port != rover &&
2915             !cma_ps_find(net, ps, (unsigned short)rover)) {
2916                 int ret = cma_alloc_port(ps, id_priv, rover);
2917                 /*
2918                  * Remember previously used port number in order to avoid
2919                  * re-using same port immediately after it is closed.
2920                  */
2921                 if (!ret)
2922                         last_used_port = rover;
2923                 if (ret != -EADDRNOTAVAIL)
2924                         return ret;
2925         }
2926         if (--remaining) {
2927                 rover++;
2928                 if ((rover < low) || (rover > high))
2929                         rover = low;
2930                 goto retry;
2931         }
2932         return -EADDRNOTAVAIL;
2933 }
2934
2935 /*
2936  * Check that the requested port is available.  This is called when trying to
2937  * bind to a specific port, or when trying to listen on a bound port.  In
2938  * the latter case, the provided id_priv may already be on the bind_list, but
2939  * we still need to check that it's okay to start listening.
2940  */
2941 static int cma_check_port(struct rdma_bind_list *bind_list,
2942                           struct rdma_id_private *id_priv, uint8_t reuseaddr)
2943 {
2944         struct rdma_id_private *cur_id;
2945         struct sockaddr *addr, *cur_addr;
2946
2947         addr = cma_src_addr(id_priv);
2948         hlist_for_each_entry(cur_id, &bind_list->owners, node) {
2949                 if (id_priv == cur_id)
2950                         continue;
2951
2952                 if ((cur_id->state != RDMA_CM_LISTEN) && reuseaddr &&
2953                     cur_id->reuseaddr)
2954                         continue;
2955
2956                 cur_addr = cma_src_addr(cur_id);
2957                 if (id_priv->afonly && cur_id->afonly &&
2958                     (addr->sa_family != cur_addr->sa_family))
2959                         continue;
2960
2961                 if (cma_any_addr(addr) || cma_any_addr(cur_addr))
2962                         return -EADDRNOTAVAIL;
2963
2964                 if (!cma_addr_cmp(addr, cur_addr))
2965                         return -EADDRINUSE;
2966         }
2967         return 0;
2968 }
2969
2970 static int cma_use_port(enum rdma_port_space ps,
2971                         struct rdma_id_private *id_priv)
2972 {
2973         struct rdma_bind_list *bind_list;
2974         unsigned short snum;
2975         int ret;
2976
2977         snum = ntohs(cma_port(cma_src_addr(id_priv)));
2978         if (snum < PROT_SOCK && !capable(CAP_NET_BIND_SERVICE))
2979                 return -EACCES;
2980
2981         bind_list = cma_ps_find(id_priv->id.route.addr.dev_addr.net, ps, snum);
2982         if (!bind_list) {
2983                 ret = cma_alloc_port(ps, id_priv, snum);
2984         } else {
2985                 ret = cma_check_port(bind_list, id_priv, id_priv->reuseaddr);
2986                 if (!ret)
2987                         cma_bind_port(bind_list, id_priv);
2988         }
2989         return ret;
2990 }
2991
2992 static int cma_bind_listen(struct rdma_id_private *id_priv)
2993 {
2994         struct rdma_bind_list *bind_list = id_priv->bind_list;
2995         int ret = 0;
2996
2997         mutex_lock(&lock);
2998         if (bind_list->owners.first->next)
2999                 ret = cma_check_port(bind_list, id_priv, 0);
3000         mutex_unlock(&lock);
3001         return ret;
3002 }
3003
3004 static enum rdma_port_space cma_select_inet_ps(
3005                 struct rdma_id_private *id_priv)
3006 {
3007         switch (id_priv->id.ps) {
3008         case RDMA_PS_TCP:
3009         case RDMA_PS_UDP:
3010         case RDMA_PS_IPOIB:
3011         case RDMA_PS_IB:
3012                 return id_priv->id.ps;
3013         default:
3014
3015                 return 0;
3016         }
3017 }
3018
3019 static enum rdma_port_space cma_select_ib_ps(struct rdma_id_private *id_priv)
3020 {
3021         enum rdma_port_space ps = 0;
3022         struct sockaddr_ib *sib;
3023         u64 sid_ps, mask, sid;
3024
3025         sib = (struct sockaddr_ib *) cma_src_addr(id_priv);
3026         mask = be64_to_cpu(sib->sib_sid_mask) & RDMA_IB_IP_PS_MASK;
3027         sid = be64_to_cpu(sib->sib_sid) & mask;
3028
3029         if ((id_priv->id.ps == RDMA_PS_IB) && (sid == (RDMA_IB_IP_PS_IB & mask))) {
3030                 sid_ps = RDMA_IB_IP_PS_IB;
3031                 ps = RDMA_PS_IB;
3032         } else if (((id_priv->id.ps == RDMA_PS_IB) || (id_priv->id.ps == RDMA_PS_TCP)) &&
3033                    (sid == (RDMA_IB_IP_PS_TCP & mask))) {
3034                 sid_ps = RDMA_IB_IP_PS_TCP;
3035                 ps = RDMA_PS_TCP;
3036         } else if (((id_priv->id.ps == RDMA_PS_IB) || (id_priv->id.ps == RDMA_PS_UDP)) &&
3037                    (sid == (RDMA_IB_IP_PS_UDP & mask))) {
3038                 sid_ps = RDMA_IB_IP_PS_UDP;
3039                 ps = RDMA_PS_UDP;
3040         }
3041
3042         if (ps) {
3043                 sib->sib_sid = cpu_to_be64(sid_ps | ntohs(cma_port((struct sockaddr *) sib)));
3044                 sib->sib_sid_mask = cpu_to_be64(RDMA_IB_IP_PS_MASK |
3045                                                 be64_to_cpu(sib->sib_sid_mask));
3046         }
3047         return ps;
3048 }
3049
3050 static int cma_get_port(struct rdma_id_private *id_priv)
3051 {
3052         enum rdma_port_space ps;
3053         int ret;
3054
3055         if (cma_family(id_priv) != AF_IB)
3056                 ps = cma_select_inet_ps(id_priv);
3057         else
3058                 ps = cma_select_ib_ps(id_priv);
3059         if (!ps)
3060                 return -EPROTONOSUPPORT;
3061
3062         mutex_lock(&lock);
3063         if (cma_any_port(cma_src_addr(id_priv)))
3064                 ret = cma_alloc_any_port(ps, id_priv);
3065         else
3066                 ret = cma_use_port(ps, id_priv);
3067         mutex_unlock(&lock);
3068
3069         return ret;
3070 }
3071
3072 static int cma_check_linklocal(struct rdma_dev_addr *dev_addr,
3073                                struct sockaddr *addr)
3074 {
3075 #if IS_ENABLED(CONFIG_IPV6)
3076         struct sockaddr_in6 *sin6;
3077
3078         if (addr->sa_family != AF_INET6)
3079                 return 0;
3080
3081         sin6 = (struct sockaddr_in6 *) addr;
3082
3083         if (!(ipv6_addr_type(&sin6->sin6_addr) & IPV6_ADDR_LINKLOCAL))
3084                 return 0;
3085
3086         if (!sin6->sin6_scope_id)
3087                         return -EINVAL;
3088
3089         dev_addr->bound_dev_if = sin6->sin6_scope_id;
3090 #endif
3091         return 0;
3092 }
3093
3094 int rdma_listen(struct rdma_cm_id *id, int backlog)
3095 {
3096         struct rdma_id_private *id_priv;
3097         int ret;
3098
3099         id_priv = container_of(id, struct rdma_id_private, id);
3100         if (id_priv->state == RDMA_CM_IDLE) {
3101                 id->route.addr.src_addr.ss_family = AF_INET;
3102                 ret = rdma_bind_addr(id, cma_src_addr(id_priv));
3103                 if (ret)
3104                         return ret;
3105         }
3106
3107         if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_BOUND, RDMA_CM_LISTEN))
3108                 return -EINVAL;
3109
3110         if (id_priv->reuseaddr) {
3111                 ret = cma_bind_listen(id_priv);
3112                 if (ret)
3113                         goto err;
3114         }
3115
3116         id_priv->backlog = backlog;
3117         if (id->device) {
3118                 if (rdma_cap_ib_cm(id->device, 1)) {
3119                         ret = cma_ib_listen(id_priv);
3120                         if (ret)
3121                                 goto err;
3122                 } else if (rdma_cap_iw_cm(id->device, 1)) {
3123                         ret = cma_iw_listen(id_priv, backlog);
3124                         if (ret)
3125                                 goto err;
3126                 } else {
3127                         ret = -ENOSYS;
3128                         goto err;
3129                 }
3130         } else
3131                 cma_listen_on_all(id_priv);
3132
3133         return 0;
3134 err:
3135         id_priv->backlog = 0;
3136         cma_comp_exch(id_priv, RDMA_CM_LISTEN, RDMA_CM_ADDR_BOUND);
3137         return ret;
3138 }
3139 EXPORT_SYMBOL(rdma_listen);
3140
3141 int rdma_bind_addr(struct rdma_cm_id *id, struct sockaddr *addr)
3142 {
3143         struct rdma_id_private *id_priv;
3144         int ret;
3145
3146         if (addr->sa_family != AF_INET && addr->sa_family != AF_INET6 &&
3147             addr->sa_family != AF_IB)
3148                 return -EAFNOSUPPORT;
3149
3150         id_priv = container_of(id, struct rdma_id_private, id);
3151         if (!cma_comp_exch(id_priv, RDMA_CM_IDLE, RDMA_CM_ADDR_BOUND))
3152                 return -EINVAL;
3153
3154         ret = cma_check_linklocal(&id->route.addr.dev_addr, addr);
3155         if (ret)
3156                 goto err1;
3157
3158         memcpy(cma_src_addr(id_priv), addr, rdma_addr_size(addr));
3159         if (!cma_any_addr(addr)) {
3160                 ret = cma_translate_addr(addr, &id->route.addr.dev_addr);
3161                 if (ret)
3162                         goto err1;
3163
3164                 ret = cma_acquire_dev(id_priv, NULL);
3165                 if (ret)
3166                         goto err1;
3167         }
3168
3169         if (!(id_priv->options & (1 << CMA_OPTION_AFONLY))) {
3170                 if (addr->sa_family == AF_INET)
3171                         id_priv->afonly = 1;
3172 #if IS_ENABLED(CONFIG_IPV6)
3173                 else if (addr->sa_family == AF_INET6) {
3174                         struct net *net = id_priv->id.route.addr.dev_addr.net;
3175
3176                         id_priv->afonly = net->ipv6.sysctl.bindv6only;
3177                 }
3178 #endif
3179         }
3180         ret = cma_get_port(id_priv);
3181         if (ret)
3182                 goto err2;
3183
3184         return 0;
3185 err2:
3186         if (id_priv->cma_dev)
3187                 cma_release_dev(id_priv);
3188 err1:
3189         cma_comp_exch(id_priv, RDMA_CM_ADDR_BOUND, RDMA_CM_IDLE);
3190         return ret;
3191 }
3192 EXPORT_SYMBOL(rdma_bind_addr);
3193
3194 static int cma_format_hdr(void *hdr, struct rdma_id_private *id_priv)
3195 {
3196         struct cma_hdr *cma_hdr;
3197
3198         cma_hdr = hdr;
3199         cma_hdr->cma_version = CMA_VERSION;
3200         if (cma_family(id_priv) == AF_INET) {
3201                 struct sockaddr_in *src4, *dst4;
3202
3203                 src4 = (struct sockaddr_in *) cma_src_addr(id_priv);
3204                 dst4 = (struct sockaddr_in *) cma_dst_addr(id_priv);
3205
3206                 cma_set_ip_ver(cma_hdr, 4);
3207                 cma_hdr->src_addr.ip4.addr = src4->sin_addr.s_addr;
3208                 cma_hdr->dst_addr.ip4.addr = dst4->sin_addr.s_addr;
3209                 cma_hdr->port = src4->sin_port;
3210         } else if (cma_family(id_priv) == AF_INET6) {
3211                 struct sockaddr_in6 *src6, *dst6;
3212
3213                 src6 = (struct sockaddr_in6 *) cma_src_addr(id_priv);
3214                 dst6 = (struct sockaddr_in6 *) cma_dst_addr(id_priv);
3215
3216                 cma_set_ip_ver(cma_hdr, 6);
3217                 cma_hdr->src_addr.ip6 = src6->sin6_addr;
3218                 cma_hdr->dst_addr.ip6 = dst6->sin6_addr;
3219                 cma_hdr->port = src6->sin6_port;
3220         }
3221         return 0;
3222 }
3223
3224 static int cma_sidr_rep_handler(struct ib_cm_id *cm_id,
3225                                 struct ib_cm_event *ib_event)
3226 {
3227         struct rdma_id_private *id_priv = cm_id->context;
3228         struct rdma_cm_event event;
3229         struct ib_cm_sidr_rep_event_param *rep = &ib_event->param.sidr_rep_rcvd;
3230         int ret = 0;
3231
3232         mutex_lock(&id_priv->handler_mutex);
3233         if (id_priv->state != RDMA_CM_CONNECT)
3234                 goto out;
3235
3236         memset(&event, 0, sizeof event);
3237         switch (ib_event->event) {
3238         case IB_CM_SIDR_REQ_ERROR:
3239                 event.event = RDMA_CM_EVENT_UNREACHABLE;
3240                 event.status = -ETIMEDOUT;
3241                 break;
3242         case IB_CM_SIDR_REP_RECEIVED:
3243                 event.param.ud.private_data = ib_event->private_data;
3244                 event.param.ud.private_data_len = IB_CM_SIDR_REP_PRIVATE_DATA_SIZE;
3245                 if (rep->status != IB_SIDR_SUCCESS) {
3246                         event.event = RDMA_CM_EVENT_UNREACHABLE;
3247                         event.status = ib_event->param.sidr_rep_rcvd.status;
3248                         break;
3249                 }
3250                 ret = cma_set_qkey(id_priv, rep->qkey);
3251                 if (ret) {
3252                         event.event = RDMA_CM_EVENT_ADDR_ERROR;
3253                         event.status = ret;
3254                         break;
3255                 }
3256                 ib_init_ah_from_path(id_priv->id.device, id_priv->id.port_num,
3257                                      id_priv->id.route.path_rec,
3258                                      &event.param.ud.ah_attr);
3259                 event.param.ud.qp_num = rep->qpn;
3260                 event.param.ud.qkey = rep->qkey;
3261                 event.event = RDMA_CM_EVENT_ESTABLISHED;
3262                 event.status = 0;
3263                 break;
3264         default:
3265                 pr_err("RDMA CMA: unexpected IB CM event: %d\n",
3266                        ib_event->event);
3267                 goto out;
3268         }
3269
3270         ret = id_priv->id.event_handler(&id_priv->id, &event);
3271         if (ret) {
3272                 /* Destroy the CM ID by returning a non-zero value. */
3273                 id_priv->cm_id.ib = NULL;
3274                 cma_exch(id_priv, RDMA_CM_DESTROYING);
3275                 mutex_unlock(&id_priv->handler_mutex);
3276                 rdma_destroy_id(&id_priv->id);
3277                 return ret;
3278         }
3279 out:
3280         mutex_unlock(&id_priv->handler_mutex);
3281         return ret;
3282 }
3283
3284 static int cma_resolve_ib_udp(struct rdma_id_private *id_priv,
3285                               struct rdma_conn_param *conn_param)
3286 {
3287         struct ib_cm_sidr_req_param req;
3288         struct ib_cm_id *id;
3289         void *private_data;
3290         int offset, ret;
3291
3292         memset(&req, 0, sizeof req);
3293         offset = cma_user_data_offset(id_priv);
3294         req.private_data_len = offset + conn_param->private_data_len;
3295         if (req.private_data_len < conn_param->private_data_len)
3296                 return -EINVAL;
3297
3298         if (req.private_data_len) {
3299                 private_data = kzalloc(req.private_data_len, GFP_ATOMIC);
3300                 if (!private_data)
3301                         return -ENOMEM;
3302         } else {
3303                 private_data = NULL;
3304         }
3305
3306         if (conn_param->private_data && conn_param->private_data_len)
3307                 memcpy(private_data + offset, conn_param->private_data,
3308                        conn_param->private_data_len);
3309
3310         if (private_data) {
3311                 ret = cma_format_hdr(private_data, id_priv);
3312                 if (ret)
3313                         goto out;
3314                 req.private_data = private_data;
3315         }
3316
3317         id = ib_create_cm_id(id_priv->id.device, cma_sidr_rep_handler,
3318                              id_priv);
3319         if (IS_ERR(id)) {
3320                 ret = PTR_ERR(id);
3321                 goto out;
3322         }
3323         id_priv->cm_id.ib = id;
3324
3325         req.path = id_priv->id.route.path_rec;
3326         req.service_id = rdma_get_service_id(&id_priv->id, cma_dst_addr(id_priv));
3327         req.timeout_ms = 1 << (CMA_CM_RESPONSE_TIMEOUT - 8);
3328         req.max_cm_retries = CMA_MAX_CM_RETRIES;
3329
3330         ret = ib_send_cm_sidr_req(id_priv->cm_id.ib, &req);
3331         if (ret) {
3332                 ib_destroy_cm_id(id_priv->cm_id.ib);
3333                 id_priv->cm_id.ib = NULL;
3334         }
3335 out:
3336         kfree(private_data);
3337         return ret;
3338 }
3339
3340 static int cma_connect_ib(struct rdma_id_private *id_priv,
3341                           struct rdma_conn_param *conn_param)
3342 {
3343         struct ib_cm_req_param req;
3344         struct rdma_route *route;
3345         void *private_data;
3346         struct ib_cm_id *id;
3347         int offset, ret;
3348
3349         memset(&req, 0, sizeof req);
3350         offset = cma_user_data_offset(id_priv);
3351         req.private_data_len = offset + conn_param->private_data_len;
3352         if (req.private_data_len < conn_param->private_data_len)
3353                 return -EINVAL;
3354
3355         if (req.private_data_len) {
3356                 private_data = kzalloc(req.private_data_len, GFP_ATOMIC);
3357                 if (!private_data)
3358                         return -ENOMEM;
3359         } else {
3360                 private_data = NULL;
3361         }
3362
3363         if (conn_param->private_data && conn_param->private_data_len)
3364                 memcpy(private_data + offset, conn_param->private_data,
3365                        conn_param->private_data_len);
3366
3367         id = ib_create_cm_id(id_priv->id.device, cma_ib_handler, id_priv);
3368         if (IS_ERR(id)) {
3369                 ret = PTR_ERR(id);
3370                 goto out;
3371         }
3372         id_priv->cm_id.ib = id;
3373
3374         route = &id_priv->id.route;
3375         if (private_data) {
3376                 ret = cma_format_hdr(private_data, id_priv);
3377                 if (ret)
3378                         goto out;
3379                 req.private_data = private_data;
3380         }
3381
3382         req.primary_path = &route->path_rec[0];
3383         if (route->num_paths == 2)
3384                 req.alternate_path = &route->path_rec[1];
3385
3386         req.service_id = rdma_get_service_id(&id_priv->id, cma_dst_addr(id_priv));
3387         req.qp_num = id_priv->qp_num;
3388         req.qp_type = id_priv->id.qp_type;
3389         req.starting_psn = id_priv->seq_num;
3390         req.responder_resources = conn_param->responder_resources;
3391         req.initiator_depth = conn_param->initiator_depth;
3392         req.flow_control = conn_param->flow_control;
3393         req.retry_count = min_t(u8, 7, conn_param->retry_count);
3394         req.rnr_retry_count = min_t(u8, 7, conn_param->rnr_retry_count);
3395         req.remote_cm_response_timeout = CMA_CM_RESPONSE_TIMEOUT;
3396         req.local_cm_response_timeout = CMA_CM_RESPONSE_TIMEOUT;
3397         req.max_cm_retries = CMA_MAX_CM_RETRIES;
3398         req.srq = id_priv->srq ? 1 : 0;
3399
3400         ret = ib_send_cm_req(id_priv->cm_id.ib, &req);
3401 out:
3402         if (ret && !IS_ERR(id)) {
3403                 ib_destroy_cm_id(id);
3404                 id_priv->cm_id.ib = NULL;
3405         }
3406
3407         kfree(private_data);
3408         return ret;
3409 }
3410
3411 static int cma_connect_iw(struct rdma_id_private *id_priv,
3412                           struct rdma_conn_param *conn_param)
3413 {
3414         struct iw_cm_id *cm_id;
3415         int ret;
3416         struct iw_cm_conn_param iw_param;
3417
3418         cm_id = iw_create_cm_id(id_priv->id.device, cma_iw_handler, id_priv);
3419         if (IS_ERR(cm_id))
3420                 return PTR_ERR(cm_id);
3421
3422         cm_id->tos = id_priv->tos;
3423         id_priv->cm_id.iw = cm_id;
3424
3425         memcpy(&cm_id->local_addr, cma_src_addr(id_priv),
3426                rdma_addr_size(cma_src_addr(id_priv)));
3427         memcpy(&cm_id->remote_addr, cma_dst_addr(id_priv),
3428                rdma_addr_size(cma_dst_addr(id_priv)));
3429
3430         ret = cma_modify_qp_rtr(id_priv, conn_param);
3431         if (ret)
3432                 goto out;
3433
3434         if (conn_param) {
3435                 iw_param.ord = conn_param->initiator_depth;
3436                 iw_param.ird = conn_param->responder_resources;
3437                 iw_param.private_data = conn_param->private_data;
3438                 iw_param.private_data_len = conn_param->private_data_len;
3439                 iw_param.qpn = id_priv->id.qp ? id_priv->qp_num : conn_param->qp_num;
3440         } else {
3441                 memset(&iw_param, 0, sizeof iw_param);
3442                 iw_param.qpn = id_priv->qp_num;
3443         }
3444         ret = iw_cm_connect(cm_id, &iw_param);
3445 out:
3446         if (ret) {
3447                 iw_destroy_cm_id(cm_id);
3448                 id_priv->cm_id.iw = NULL;
3449         }
3450         return ret;
3451 }
3452
3453 int rdma_connect(struct rdma_cm_id *id, struct rdma_conn_param *conn_param)
3454 {
3455         struct rdma_id_private *id_priv;
3456         int ret;
3457
3458         id_priv = container_of(id, struct rdma_id_private, id);
3459         if (!cma_comp_exch(id_priv, RDMA_CM_ROUTE_RESOLVED, RDMA_CM_CONNECT))
3460                 return -EINVAL;
3461
3462         if (!id->qp) {
3463                 id_priv->qp_num = conn_param->qp_num;
3464                 id_priv->srq = conn_param->srq;
3465         }
3466
3467         if (rdma_cap_ib_cm(id->device, id->port_num)) {
3468                 if (id->qp_type == IB_QPT_UD)
3469                         ret = cma_resolve_ib_udp(id_priv, conn_param);
3470                 else
3471                         ret = cma_connect_ib(id_priv, conn_param);
3472         } else if (rdma_cap_iw_cm(id->device, id->port_num))
3473                 ret = cma_connect_iw(id_priv, conn_param);
3474         else
3475                 ret = -ENOSYS;
3476         if (ret)
3477                 goto err;
3478
3479         return 0;
3480 err:
3481         cma_comp_exch(id_priv, RDMA_CM_CONNECT, RDMA_CM_ROUTE_RESOLVED);
3482         return ret;
3483 }
3484 EXPORT_SYMBOL(rdma_connect);
3485
3486 static int cma_accept_ib(struct rdma_id_private *id_priv,
3487                          struct rdma_conn_param *conn_param)
3488 {
3489         struct ib_cm_rep_param rep;
3490         int ret;
3491
3492         ret = cma_modify_qp_rtr(id_priv, conn_param);
3493         if (ret)
3494                 goto out;
3495
3496         ret = cma_modify_qp_rts(id_priv, conn_param);
3497         if (ret)
3498                 goto out;
3499
3500         memset(&rep, 0, sizeof rep);
3501         rep.qp_num = id_priv->qp_num;
3502         rep.starting_psn = id_priv->seq_num;
3503         rep.private_data = conn_param->private_data;
3504         rep.private_data_len = conn_param->private_data_len;
3505         rep.responder_resources = conn_param->responder_resources;
3506         rep.initiator_depth = conn_param->initiator_depth;
3507         rep.failover_accepted = 0;
3508         rep.flow_control = conn_param->flow_control;
3509         rep.rnr_retry_count = min_t(u8, 7, conn_param->rnr_retry_count);
3510         rep.srq = id_priv->srq ? 1 : 0;
3511
3512         ret = ib_send_cm_rep(id_priv->cm_id.ib, &rep);
3513 out:
3514         return ret;
3515 }
3516
3517 static int cma_accept_iw(struct rdma_id_private *id_priv,
3518                   struct rdma_conn_param *conn_param)
3519 {
3520         struct iw_cm_conn_param iw_param;
3521         int ret;
3522
3523         ret = cma_modify_qp_rtr(id_priv, conn_param);
3524         if (ret)
3525                 return ret;
3526
3527         iw_param.ord = conn_param->initiator_depth;
3528         iw_param.ird = conn_param->responder_resources;
3529         iw_param.private_data = conn_param->private_data;
3530         iw_param.private_data_len = conn_param->private_data_len;
3531         if (id_priv->id.qp) {
3532                 iw_param.qpn = id_priv->qp_num;
3533         } else
3534                 iw_param.qpn = conn_param->qp_num;
3535
3536         return iw_cm_accept(id_priv->cm_id.iw, &iw_param);
3537 }
3538
3539 static int cma_send_sidr_rep(struct rdma_id_private *id_priv,
3540                              enum ib_cm_sidr_status status, u32 qkey,
3541                              const void *private_data, int private_data_len)
3542 {
3543         struct ib_cm_sidr_rep_param rep;
3544         int ret;
3545
3546         memset(&rep, 0, sizeof rep);
3547         rep.status = status;
3548         if (status == IB_SIDR_SUCCESS) {
3549                 ret = cma_set_qkey(id_priv, qkey);
3550                 if (ret)
3551                         return ret;
3552                 rep.qp_num = id_priv->qp_num;
3553                 rep.qkey = id_priv->qkey;
3554         }
3555         rep.private_data = private_data;
3556         rep.private_data_len = private_data_len;
3557
3558         return ib_send_cm_sidr_rep(id_priv->cm_id.ib, &rep);
3559 }
3560
3561 int rdma_accept(struct rdma_cm_id *id, struct rdma_conn_param *conn_param)
3562 {
3563         struct rdma_id_private *id_priv;
3564         int ret;
3565
3566         id_priv = container_of(id, struct rdma_id_private, id);
3567
3568         id_priv->owner = task_pid_nr(current);
3569
3570         if (!cma_comp(id_priv, RDMA_CM_CONNECT))
3571                 return -EINVAL;
3572
3573         if (!id->qp && conn_param) {
3574                 id_priv->qp_num = conn_param->qp_num;
3575                 id_priv->srq = conn_param->srq;
3576         }
3577
3578         if (rdma_cap_ib_cm(id->device, id->port_num)) {
3579                 if (id->qp_type == IB_QPT_UD) {
3580                         if (conn_param)
3581                                 ret = cma_send_sidr_rep(id_priv, IB_SIDR_SUCCESS,
3582                                                         conn_param->qkey,
3583                                                         conn_param->private_data,
3584                                                         conn_param->private_data_len);
3585                         else
3586                                 ret = cma_send_sidr_rep(id_priv, IB_SIDR_SUCCESS,
3587                                                         0, NULL, 0);
3588                 } else {
3589                         if (conn_param)
3590                                 ret = cma_accept_ib(id_priv, conn_param);
3591                         else
3592                                 ret = cma_rep_recv(id_priv);
3593                 }
3594         } else if (rdma_cap_iw_cm(id->device, id->port_num))
3595                 ret = cma_accept_iw(id_priv, conn_param);
3596         else
3597                 ret = -ENOSYS;
3598
3599         if (ret)
3600                 goto reject;
3601
3602         return 0;
3603 reject:
3604         cma_modify_qp_err(id_priv);
3605         rdma_reject(id, NULL, 0);
3606         return ret;
3607 }
3608 EXPORT_SYMBOL(rdma_accept);
3609
3610 int rdma_notify(struct rdma_cm_id *id, enum ib_event_type event)
3611 {
3612         struct rdma_id_private *id_priv;
3613         int ret;
3614
3615         id_priv = container_of(id, struct rdma_id_private, id);
3616         if (!id_priv->cm_id.ib)
3617                 return -EINVAL;
3618
3619         switch (id->device->node_type) {
3620         case RDMA_NODE_IB_CA:
3621                 ret = ib_cm_notify(id_priv->cm_id.ib, event);
3622                 break;
3623         default:
3624                 ret = 0;
3625                 break;
3626         }
3627         return ret;
3628 }
3629 EXPORT_SYMBOL(rdma_notify);
3630
3631 int rdma_reject(struct rdma_cm_id *id, const void *private_data,
3632                 u8 private_data_len)
3633 {
3634         struct rdma_id_private *id_priv;
3635         int ret;
3636
3637         id_priv = container_of(id, struct rdma_id_private, id);
3638         if (!id_priv->cm_id.ib)
3639                 return -EINVAL;
3640
3641         if (rdma_cap_ib_cm(id->device, id->port_num)) {
3642                 if (id->qp_type == IB_QPT_UD)
3643                         ret = cma_send_sidr_rep(id_priv, IB_SIDR_REJECT, 0,
3644                                                 private_data, private_data_len);
3645                 else
3646                         ret = ib_send_cm_rej(id_priv->cm_id.ib,
3647                                              IB_CM_REJ_CONSUMER_DEFINED, NULL,
3648                                              0, private_data, private_data_len);
3649         } else if (rdma_cap_iw_cm(id->device, id->port_num)) {
3650                 ret = iw_cm_reject(id_priv->cm_id.iw,
3651                                    private_data, private_data_len);
3652         } else
3653                 ret = -ENOSYS;
3654
3655         return ret;
3656 }
3657 EXPORT_SYMBOL(rdma_reject);
3658
3659 int rdma_disconnect(struct rdma_cm_id *id)
3660 {
3661         struct rdma_id_private *id_priv;
3662         int ret;
3663
3664         id_priv = container_of(id, struct rdma_id_private, id);
3665         if (!id_priv->cm_id.ib)
3666                 return -EINVAL;
3667
3668         if (rdma_cap_ib_cm(id->device, id->port_num)) {
3669                 ret = cma_modify_qp_err(id_priv);
3670                 if (ret)
3671                         goto out;
3672                 /* Initiate or respond to a disconnect. */
3673                 if (ib_send_cm_dreq(id_priv->cm_id.ib, NULL, 0))
3674                         ib_send_cm_drep(id_priv->cm_id.ib, NULL, 0);
3675         } else if (rdma_cap_iw_cm(id->device, id->port_num)) {
3676                 ret = iw_cm_disconnect(id_priv->cm_id.iw, 0);
3677         } else
3678                 ret = -EINVAL;
3679
3680 out:
3681         return ret;
3682 }
3683 EXPORT_SYMBOL(rdma_disconnect);
3684
3685 static int cma_ib_mc_handler(int status, struct ib_sa_multicast *multicast)
3686 {
3687         struct rdma_id_private *id_priv;
3688         struct cma_multicast *mc = multicast->context;
3689         struct rdma_cm_event event;
3690         int ret = 0;
3691
3692         id_priv = mc->id_priv;
3693         mutex_lock(&id_priv->handler_mutex);
3694         if (id_priv->state != RDMA_CM_ADDR_BOUND &&
3695             id_priv->state != RDMA_CM_ADDR_RESOLVED)
3696                 goto out;
3697
3698         if (!status)
3699                 status = cma_set_qkey(id_priv, be32_to_cpu(multicast->rec.qkey));
3700         mutex_lock(&id_priv->qp_mutex);
3701         if (!status && id_priv->id.qp)
3702                 status = ib_attach_mcast(id_priv->id.qp, &multicast->rec.mgid,
3703                                          be16_to_cpu(multicast->rec.mlid));
3704         mutex_unlock(&id_priv->qp_mutex);
3705
3706         memset(&event, 0, sizeof event);
3707         event.status = status;
3708         event.param.ud.private_data = mc->context;
3709         if (!status) {
3710                 struct rdma_dev_addr *dev_addr =
3711                         &id_priv->id.route.addr.dev_addr;
3712                 struct net_device *ndev =
3713                         dev_get_by_index(&init_net, dev_addr->bound_dev_if);
3714                 enum ib_gid_type gid_type =
3715                         id_priv->cma_dev->default_gid_type[id_priv->id.port_num -
3716                         rdma_start_port(id_priv->cma_dev->device)];
3717
3718                 event.event = RDMA_CM_EVENT_MULTICAST_JOIN;
3719                 ib_init_ah_from_mcmember(id_priv->id.device,
3720                                          id_priv->id.port_num, &multicast->rec,
3721                                          ndev, gid_type,
3722                                          &event.param.ud.ah_attr);
3723                 event.param.ud.qp_num = 0xFFFFFF;
3724                 event.param.ud.qkey = be32_to_cpu(multicast->rec.qkey);
3725                 if (ndev)
3726                         dev_put(ndev);
3727         } else
3728                 event.event = RDMA_CM_EVENT_MULTICAST_ERROR;
3729
3730         ret = id_priv->id.event_handler(&id_priv->id, &event);
3731         if (ret) {
3732                 cma_exch(id_priv, RDMA_CM_DESTROYING);
3733                 mutex_unlock(&id_priv->handler_mutex);
3734                 rdma_destroy_id(&id_priv->id);
3735                 return 0;
3736         }
3737
3738 out:
3739         mutex_unlock(&id_priv->handler_mutex);
3740         return 0;
3741 }
3742
3743 static void cma_set_mgid(struct rdma_id_private *id_priv,
3744                          struct sockaddr *addr, union ib_gid *mgid)
3745 {
3746         unsigned char mc_map[MAX_ADDR_LEN];
3747         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
3748         struct sockaddr_in *sin = (struct sockaddr_in *) addr;
3749         struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *) addr;
3750
3751         if (cma_any_addr(addr)) {
3752                 memset(mgid, 0, sizeof *mgid);
3753         } else if ((addr->sa_family == AF_INET6) &&
3754                    ((be32_to_cpu(sin6->sin6_addr.s6_addr32[0]) & 0xFFF0FFFF) ==
3755                                                                  0xFF10A01B)) {
3756                 /* IPv6 address is an SA assigned MGID. */
3757                 memcpy(mgid, &sin6->sin6_addr, sizeof *mgid);
3758         } else if (addr->sa_family == AF_IB) {
3759                 memcpy(mgid, &((struct sockaddr_ib *) addr)->sib_addr, sizeof *mgid);
3760         } else if ((addr->sa_family == AF_INET6)) {
3761                 ipv6_ib_mc_map(&sin6->sin6_addr, dev_addr->broadcast, mc_map);
3762                 if (id_priv->id.ps == RDMA_PS_UDP)
3763                         mc_map[7] = 0x01;       /* Use RDMA CM signature */
3764                 *mgid = *(union ib_gid *) (mc_map + 4);
3765         } else {
3766                 ip_ib_mc_map(sin->sin_addr.s_addr, dev_addr->broadcast, mc_map);
3767                 if (id_priv->id.ps == RDMA_PS_UDP)
3768                         mc_map[7] = 0x01;       /* Use RDMA CM signature */
3769                 *mgid = *(union ib_gid *) (mc_map + 4);
3770         }
3771 }
3772
3773 static void cma_query_sa_classport_info_cb(int status,
3774                                            struct ib_class_port_info *rec,
3775                                            void *context)
3776 {
3777         struct class_port_info_context *cb_ctx = context;
3778
3779         WARN_ON(!context);
3780
3781         if (status || !rec) {
3782                 pr_debug("RDMA CM: %s port %u failed query ClassPortInfo status: %d\n",
3783                          cb_ctx->device->name, cb_ctx->port_num, status);
3784                 goto out;
3785         }
3786
3787         memcpy(cb_ctx->class_port_info, rec, sizeof(struct ib_class_port_info));
3788
3789 out:
3790         complete(&cb_ctx->done);
3791 }
3792
3793 static int cma_query_sa_classport_info(struct ib_device *device, u8 port_num,
3794                                        struct ib_class_port_info *class_port_info)
3795 {
3796         struct class_port_info_context *cb_ctx;
3797         int ret;
3798
3799         cb_ctx = kmalloc(sizeof(*cb_ctx), GFP_KERNEL);
3800         if (!cb_ctx)
3801                 return -ENOMEM;
3802
3803         cb_ctx->device = device;
3804         cb_ctx->class_port_info = class_port_info;
3805         cb_ctx->port_num = port_num;
3806         init_completion(&cb_ctx->done);
3807
3808         ret = ib_sa_classport_info_rec_query(&sa_client, device, port_num,
3809                                              CMA_QUERY_CLASSPORT_INFO_TIMEOUT,
3810                                              GFP_KERNEL, cma_query_sa_classport_info_cb,
3811                                              cb_ctx, &cb_ctx->sa_query);
3812         if (ret < 0) {
3813                 pr_err("RDMA CM: %s port %u failed to send ClassPortInfo query, ret: %d\n",
3814                        device->name, port_num, ret);
3815                 goto out;
3816         }
3817
3818         wait_for_completion(&cb_ctx->done);
3819
3820 out:
3821         kfree(cb_ctx);
3822         return ret;
3823 }
3824
3825 static int cma_join_ib_multicast(struct rdma_id_private *id_priv,
3826                                  struct cma_multicast *mc)
3827 {
3828         struct ib_sa_mcmember_rec rec;
3829         struct ib_class_port_info class_port_info;
3830         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
3831         ib_sa_comp_mask comp_mask;
3832         int ret;
3833
3834         ib_addr_get_mgid(dev_addr, &rec.mgid);
3835         ret = ib_sa_get_mcmember_rec(id_priv->id.device, id_priv->id.port_num,
3836                                      &rec.mgid, &rec);
3837         if (ret)
3838                 return ret;
3839
3840         ret = cma_set_qkey(id_priv, 0);
3841         if (ret)
3842                 return ret;
3843
3844         cma_set_mgid(id_priv, (struct sockaddr *) &mc->addr, &rec.mgid);
3845         rec.qkey = cpu_to_be32(id_priv->qkey);
3846         rdma_addr_get_sgid(dev_addr, &rec.port_gid);
3847         rec.pkey = cpu_to_be16(ib_addr_get_pkey(dev_addr));
3848         rec.join_state = mc->join_state;
3849
3850         if (rec.join_state == BIT(SENDONLY_FULLMEMBER_JOIN)) {
3851                 ret = cma_query_sa_classport_info(id_priv->id.device,
3852                                                   id_priv->id.port_num,
3853                                                   &class_port_info);
3854
3855                 if (ret)
3856                         return ret;
3857
3858                 if (!(ib_get_cpi_capmask2(&class_port_info) &
3859                       IB_SA_CAP_MASK2_SENDONLY_FULL_MEM_SUPPORT)) {
3860                         pr_warn("RDMA CM: %s port %u Unable to multicast join\n"
3861                                 "RDMA CM: SM doesn't support Send Only Full Member option\n",
3862                                 id_priv->id.device->name, id_priv->id.port_num);
3863                         return -EOPNOTSUPP;
3864                 }
3865         }
3866
3867         comp_mask = IB_SA_MCMEMBER_REC_MGID | IB_SA_MCMEMBER_REC_PORT_GID |
3868                     IB_SA_MCMEMBER_REC_PKEY | IB_SA_MCMEMBER_REC_JOIN_STATE |
3869                     IB_SA_MCMEMBER_REC_QKEY | IB_SA_MCMEMBER_REC_SL |
3870                     IB_SA_MCMEMBER_REC_FLOW_LABEL |
3871                     IB_SA_MCMEMBER_REC_TRAFFIC_CLASS;
3872
3873         if (id_priv->id.ps == RDMA_PS_IPOIB)
3874                 comp_mask |= IB_SA_MCMEMBER_REC_RATE |
3875                              IB_SA_MCMEMBER_REC_RATE_SELECTOR |
3876                              IB_SA_MCMEMBER_REC_MTU_SELECTOR |
3877                              IB_SA_MCMEMBER_REC_MTU |
3878                              IB_SA_MCMEMBER_REC_HOP_LIMIT;
3879
3880         mc->multicast.ib = ib_sa_join_multicast(&sa_client, id_priv->id.device,
3881                                                 id_priv->id.port_num, &rec,
3882                                                 comp_mask, GFP_KERNEL,
3883                                                 cma_ib_mc_handler, mc);
3884         return PTR_ERR_OR_ZERO(mc->multicast.ib);
3885 }
3886
3887 static void iboe_mcast_work_handler(struct work_struct *work)
3888 {
3889         struct iboe_mcast_work *mw = container_of(work, struct iboe_mcast_work, work);
3890         struct cma_multicast *mc = mw->mc;
3891         struct ib_sa_multicast *m = mc->multicast.ib;
3892
3893         mc->multicast.ib->context = mc;
3894         cma_ib_mc_handler(0, m);
3895         kref_put(&mc->mcref, release_mc);
3896         kfree(mw);
3897 }
3898
3899 static void cma_iboe_set_mgid(struct sockaddr *addr, union ib_gid *mgid)
3900 {
3901         struct sockaddr_in *sin = (struct sockaddr_in *)addr;
3902         struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)addr;
3903
3904         if (cma_any_addr(addr)) {
3905                 memset(mgid, 0, sizeof *mgid);
3906         } else if (addr->sa_family == AF_INET6) {
3907                 memcpy(mgid, &sin6->sin6_addr, sizeof *mgid);
3908         } else {
3909                 mgid->raw[0] = 0xff;
3910                 mgid->raw[1] = 0x0e;
3911                 mgid->raw[2] = 0;
3912                 mgid->raw[3] = 0;
3913                 mgid->raw[4] = 0;
3914                 mgid->raw[5] = 0;
3915                 mgid->raw[6] = 0;
3916                 mgid->raw[7] = 0;
3917                 mgid->raw[8] = 0;
3918                 mgid->raw[9] = 0;
3919                 mgid->raw[10] = 0xff;
3920                 mgid->raw[11] = 0xff;
3921                 *(__be32 *)(&mgid->raw[12]) = sin->sin_addr.s_addr;
3922         }
3923 }
3924
3925 static int cma_iboe_join_multicast(struct rdma_id_private *id_priv,
3926                                    struct cma_multicast *mc)
3927 {
3928         struct iboe_mcast_work *work;
3929         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
3930         int err = 0;
3931         struct sockaddr *addr = (struct sockaddr *)&mc->addr;
3932         struct net_device *ndev = NULL;
3933         enum ib_gid_type gid_type;
3934         bool send_only;
3935
3936         send_only = mc->join_state == BIT(SENDONLY_FULLMEMBER_JOIN);
3937
3938         if (cma_zero_addr((struct sockaddr *)&mc->addr))
3939                 return -EINVAL;
3940
3941         work = kzalloc(sizeof *work, GFP_KERNEL);
3942         if (!work)
3943                 return -ENOMEM;
3944
3945         mc->multicast.ib = kzalloc(sizeof(struct ib_sa_multicast), GFP_KERNEL);
3946         if (!mc->multicast.ib) {
3947                 err = -ENOMEM;
3948                 goto out1;
3949         }
3950
3951         cma_iboe_set_mgid(addr, &mc->multicast.ib->rec.mgid);
3952
3953         mc->multicast.ib->rec.pkey = cpu_to_be16(0xffff);
3954         if (id_priv->id.ps == RDMA_PS_UDP)
3955                 mc->multicast.ib->rec.qkey = cpu_to_be32(RDMA_UDP_QKEY);
3956
3957         if (dev_addr->bound_dev_if)
3958                 ndev = dev_get_by_index(&init_net, dev_addr->bound_dev_if);
3959         if (!ndev) {
3960                 err = -ENODEV;
3961                 goto out2;
3962         }
3963         mc->multicast.ib->rec.rate = iboe_get_rate(ndev);
3964         mc->multicast.ib->rec.hop_limit = 1;
3965         mc->multicast.ib->rec.mtu = iboe_get_mtu(ndev->mtu);
3966
3967         gid_type = id_priv->cma_dev->default_gid_type[id_priv->id.port_num -
3968                    rdma_start_port(id_priv->cma_dev->device)];
3969         if (addr->sa_family == AF_INET) {
3970                 if (gid_type == IB_GID_TYPE_ROCE_UDP_ENCAP) {
3971                         mc->multicast.ib->rec.hop_limit = IPV6_DEFAULT_HOPLIMIT;
3972                         if (!send_only) {
3973                                 err = cma_igmp_send(ndev, &mc->multicast.ib->rec.mgid,
3974                                                     true);
3975                                 if (!err)
3976                                         mc->igmp_joined = true;
3977                         }
3978                 }
3979         } else {
3980                 if (gid_type == IB_GID_TYPE_ROCE_UDP_ENCAP)
3981                         err = -ENOTSUPP;
3982         }
3983         dev_put(ndev);
3984         if (err || !mc->multicast.ib->rec.mtu) {
3985                 if (!err)
3986                         err = -EINVAL;
3987                 goto out2;
3988         }
3989         rdma_ip2gid((struct sockaddr *)&id_priv->id.route.addr.src_addr,
3990                     &mc->multicast.ib->rec.port_gid);
3991         work->id = id_priv;
3992         work->mc = mc;
3993         INIT_WORK(&work->work, iboe_mcast_work_handler);
3994         kref_get(&mc->mcref);
3995         queue_work(cma_wq, &work->work);
3996
3997         return 0;
3998
3999 out2:
4000         kfree(mc->multicast.ib);
4001 out1:
4002         kfree(work);
4003         return err;
4004 }
4005
4006 int rdma_join_multicast(struct rdma_cm_id *id, struct sockaddr *addr,
4007                         u8 join_state, void *context)
4008 {
4009         struct rdma_id_private *id_priv;
4010         struct cma_multicast *mc;
4011         int ret;
4012
4013         id_priv = container_of(id, struct rdma_id_private, id);
4014         if (!cma_comp(id_priv, RDMA_CM_ADDR_BOUND) &&
4015             !cma_comp(id_priv, RDMA_CM_ADDR_RESOLVED))
4016                 return -EINVAL;
4017
4018         mc = kmalloc(sizeof *mc, GFP_KERNEL);
4019         if (!mc)
4020                 return -ENOMEM;
4021
4022         memcpy(&mc->addr, addr, rdma_addr_size(addr));
4023         mc->context = context;
4024         mc->id_priv = id_priv;
4025         mc->igmp_joined = false;
4026         mc->join_state = join_state;
4027         spin_lock(&id_priv->lock);
4028         list_add(&mc->list, &id_priv->mc_list);
4029         spin_unlock(&id_priv->lock);
4030
4031         if (rdma_protocol_roce(id->device, id->port_num)) {
4032                 kref_init(&mc->mcref);
4033                 ret = cma_iboe_join_multicast(id_priv, mc);
4034         } else if (rdma_cap_ib_mcast(id->device, id->port_num))
4035                 ret = cma_join_ib_multicast(id_priv, mc);
4036         else
4037                 ret = -ENOSYS;
4038
4039         if (ret) {
4040                 spin_lock_irq(&id_priv->lock);
4041                 list_del(&mc->list);
4042                 spin_unlock_irq(&id_priv->lock);
4043                 kfree(mc);
4044         }
4045         return ret;
4046 }
4047 EXPORT_SYMBOL(rdma_join_multicast);
4048
4049 void rdma_leave_multicast(struct rdma_cm_id *id, struct sockaddr *addr)
4050 {
4051         struct rdma_id_private *id_priv;
4052         struct cma_multicast *mc;
4053
4054         id_priv = container_of(id, struct rdma_id_private, id);
4055         spin_lock_irq(&id_priv->lock);
4056         list_for_each_entry(mc, &id_priv->mc_list, list) {
4057                 if (!memcmp(&mc->addr, addr, rdma_addr_size(addr))) {
4058                         list_del(&mc->list);
4059                         spin_unlock_irq(&id_priv->lock);
4060
4061                         if (id->qp)
4062                                 ib_detach_mcast(id->qp,
4063                                                 &mc->multicast.ib->rec.mgid,
4064                                                 be16_to_cpu(mc->multicast.ib->rec.mlid));
4065
4066                         BUG_ON(id_priv->cma_dev->device != id->device);
4067
4068                         if (rdma_cap_ib_mcast(id->device, id->port_num)) {
4069                                 ib_sa_free_multicast(mc->multicast.ib);
4070                                 kfree(mc);
4071                         } else if (rdma_protocol_roce(id->device, id->port_num)) {
4072                                 if (mc->igmp_joined) {
4073                                         struct rdma_dev_addr *dev_addr =
4074                                                 &id->route.addr.dev_addr;
4075                                         struct net_device *ndev = NULL;
4076
4077                                         if (dev_addr->bound_dev_if)
4078                                                 ndev = dev_get_by_index(&init_net,
4079                                                                         dev_addr->bound_dev_if);
4080                                         if (ndev) {
4081                                                 cma_igmp_send(ndev,
4082                                                               &mc->multicast.ib->rec.mgid,
4083                                                               false);
4084                                                 dev_put(ndev);
4085                                         }
4086                                         mc->igmp_joined = false;
4087                                 }
4088                                 kref_put(&mc->mcref, release_mc);
4089                         }
4090                         return;
4091                 }
4092         }
4093         spin_unlock_irq(&id_priv->lock);
4094 }
4095 EXPORT_SYMBOL(rdma_leave_multicast);
4096
4097 static int cma_netdev_change(struct net_device *ndev, struct rdma_id_private *id_priv)
4098 {
4099         struct rdma_dev_addr *dev_addr;
4100         struct cma_ndev_work *work;
4101
4102         dev_addr = &id_priv->id.route.addr.dev_addr;
4103
4104         if ((dev_addr->bound_dev_if == ndev->ifindex) &&
4105             (net_eq(dev_net(ndev), dev_addr->net)) &&
4106             memcmp(dev_addr->src_dev_addr, ndev->dev_addr, ndev->addr_len)) {
4107                 pr_info("RDMA CM addr change for ndev %s used by id %p\n",
4108                         ndev->name, &id_priv->id);
4109                 work = kzalloc(sizeof *work, GFP_KERNEL);
4110                 if (!work)
4111                         return -ENOMEM;
4112
4113                 INIT_WORK(&work->work, cma_ndev_work_handler);
4114                 work->id = id_priv;
4115                 work->event.event = RDMA_CM_EVENT_ADDR_CHANGE;
4116                 atomic_inc(&id_priv->refcount);
4117                 queue_work(cma_wq, &work->work);
4118         }
4119
4120         return 0;
4121 }
4122
4123 static int cma_netdev_callback(struct notifier_block *self, unsigned long event,
4124                                void *ptr)
4125 {
4126         struct net_device *ndev = netdev_notifier_info_to_dev(ptr);
4127         struct cma_device *cma_dev;
4128         struct rdma_id_private *id_priv;
4129         int ret = NOTIFY_DONE;
4130
4131         if (event != NETDEV_BONDING_FAILOVER)
4132                 return NOTIFY_DONE;
4133
4134         if (!(ndev->flags & IFF_MASTER) || !(ndev->priv_flags & IFF_BONDING))
4135                 return NOTIFY_DONE;
4136
4137         mutex_lock(&lock);
4138         list_for_each_entry(cma_dev, &dev_list, list)
4139                 list_for_each_entry(id_priv, &cma_dev->id_list, list) {
4140                         ret = cma_netdev_change(ndev, id_priv);
4141                         if (ret)
4142                                 goto out;
4143                 }
4144
4145 out:
4146         mutex_unlock(&lock);
4147         return ret;
4148 }
4149
4150 static struct notifier_block cma_nb = {
4151         .notifier_call = cma_netdev_callback
4152 };
4153
4154 static void cma_add_one(struct ib_device *device)
4155 {
4156         struct cma_device *cma_dev;
4157         struct rdma_id_private *id_priv;
4158         unsigned int i;
4159         unsigned long supported_gids = 0;
4160
4161         cma_dev = kmalloc(sizeof *cma_dev, GFP_KERNEL);
4162         if (!cma_dev)
4163                 return;
4164
4165         cma_dev->device = device;
4166         cma_dev->default_gid_type = kcalloc(device->phys_port_cnt,
4167                                             sizeof(*cma_dev->default_gid_type),
4168                                             GFP_KERNEL);
4169         if (!cma_dev->default_gid_type) {
4170                 kfree(cma_dev);
4171                 return;
4172         }
4173         for (i = rdma_start_port(device); i <= rdma_end_port(device); i++) {
4174                 supported_gids = roce_gid_type_mask_support(device, i);
4175                 WARN_ON(!supported_gids);
4176                 cma_dev->default_gid_type[i - rdma_start_port(device)] =
4177                         find_first_bit(&supported_gids, BITS_PER_LONG);
4178         }
4179
4180         init_completion(&cma_dev->comp);
4181         atomic_set(&cma_dev->refcount, 1);
4182         INIT_LIST_HEAD(&cma_dev->id_list);
4183         ib_set_client_data(device, &cma_client, cma_dev);
4184
4185         mutex_lock(&lock);
4186         list_add_tail(&cma_dev->list, &dev_list);
4187         list_for_each_entry(id_priv, &listen_any_list, list)
4188                 cma_listen_on_dev(id_priv, cma_dev);
4189         mutex_unlock(&lock);
4190 }
4191
4192 static int cma_remove_id_dev(struct rdma_id_private *id_priv)
4193 {
4194         struct rdma_cm_event event;
4195         enum rdma_cm_state state;
4196         int ret = 0;
4197
4198         /* Record that we want to remove the device */
4199         state = cma_exch(id_priv, RDMA_CM_DEVICE_REMOVAL);
4200         if (state == RDMA_CM_DESTROYING)
4201                 return 0;
4202
4203         cma_cancel_operation(id_priv, state);
4204         mutex_lock(&id_priv->handler_mutex);
4205
4206         /* Check for destruction from another callback. */
4207         if (!cma_comp(id_priv, RDMA_CM_DEVICE_REMOVAL))
4208                 goto out;
4209
4210         memset(&event, 0, sizeof event);
4211         event.event = RDMA_CM_EVENT_DEVICE_REMOVAL;
4212         ret = id_priv->id.event_handler(&id_priv->id, &event);
4213 out:
4214         mutex_unlock(&id_priv->handler_mutex);
4215         return ret;
4216 }
4217
4218 static void cma_process_remove(struct cma_device *cma_dev)
4219 {
4220         struct rdma_id_private *id_priv;
4221         int ret;
4222
4223         mutex_lock(&lock);
4224         while (!list_empty(&cma_dev->id_list)) {
4225                 id_priv = list_entry(cma_dev->id_list.next,
4226                                      struct rdma_id_private, list);
4227
4228                 list_del(&id_priv->listen_list);
4229                 list_del_init(&id_priv->list);
4230                 atomic_inc(&id_priv->refcount);
4231                 mutex_unlock(&lock);
4232
4233                 ret = id_priv->internal_id ? 1 : cma_remove_id_dev(id_priv);
4234                 cma_deref_id(id_priv);
4235                 if (ret)
4236                         rdma_destroy_id(&id_priv->id);
4237
4238                 mutex_lock(&lock);
4239         }
4240         mutex_unlock(&lock);
4241
4242         cma_deref_dev(cma_dev);
4243         wait_for_completion(&cma_dev->comp);
4244 }
4245
4246 static void cma_remove_one(struct ib_device *device, void *client_data)
4247 {
4248         struct cma_device *cma_dev = client_data;
4249
4250         if (!cma_dev)
4251                 return;
4252
4253         mutex_lock(&lock);
4254         list_del(&cma_dev->list);
4255         mutex_unlock(&lock);
4256
4257         cma_process_remove(cma_dev);
4258         kfree(cma_dev->default_gid_type);
4259         kfree(cma_dev);
4260 }
4261
4262 static int cma_get_id_stats(struct sk_buff *skb, struct netlink_callback *cb)
4263 {
4264         struct nlmsghdr *nlh;
4265         struct rdma_cm_id_stats *id_stats;
4266         struct rdma_id_private *id_priv;
4267         struct rdma_cm_id *id = NULL;
4268         struct cma_device *cma_dev;
4269         int i_dev = 0, i_id = 0;
4270
4271         /*
4272          * We export all of the IDs as a sequence of messages.  Each
4273          * ID gets its own netlink message.
4274          */
4275         mutex_lock(&lock);
4276
4277         list_for_each_entry(cma_dev, &dev_list, list) {
4278                 if (i_dev < cb->args[0]) {
4279                         i_dev++;
4280                         continue;
4281                 }
4282
4283                 i_id = 0;
4284                 list_for_each_entry(id_priv, &cma_dev->id_list, list) {
4285                         if (i_id < cb->args[1]) {
4286                                 i_id++;
4287                                 continue;
4288                         }
4289
4290                         id_stats = ibnl_put_msg(skb, &nlh, cb->nlh->nlmsg_seq,
4291                                                 sizeof *id_stats, RDMA_NL_RDMA_CM,
4292                                                 RDMA_NL_RDMA_CM_ID_STATS,
4293                                                 NLM_F_MULTI);
4294                         if (!id_stats)
4295                                 goto out;
4296
4297                         memset(id_stats, 0, sizeof *id_stats);
4298                         id = &id_priv->id;
4299                         id_stats->node_type = id->route.addr.dev_addr.dev_type;
4300                         id_stats->port_num = id->port_num;
4301                         id_stats->bound_dev_if =
4302                                 id->route.addr.dev_addr.bound_dev_if;
4303
4304                         if (ibnl_put_attr(skb, nlh,
4305                                           rdma_addr_size(cma_src_addr(id_priv)),
4306                                           cma_src_addr(id_priv),
4307                                           RDMA_NL_RDMA_CM_ATTR_SRC_ADDR))
4308                                 goto out;
4309                         if (ibnl_put_attr(skb, nlh,
4310                                           rdma_addr_size(cma_src_addr(id_priv)),
4311                                           cma_dst_addr(id_priv),
4312                                           RDMA_NL_RDMA_CM_ATTR_DST_ADDR))
4313                                 goto out;
4314
4315                         id_stats->pid           = id_priv->owner;
4316                         id_stats->port_space    = id->ps;
4317                         id_stats->cm_state      = id_priv->state;
4318                         id_stats->qp_num        = id_priv->qp_num;
4319                         id_stats->qp_type       = id->qp_type;
4320
4321                         i_id++;
4322                 }
4323
4324                 cb->args[1] = 0;
4325                 i_dev++;
4326         }
4327
4328 out:
4329         mutex_unlock(&lock);
4330         cb->args[0] = i_dev;
4331         cb->args[1] = i_id;
4332
4333         return skb->len;
4334 }
4335
4336 static const struct ibnl_client_cbs cma_cb_table[] = {
4337         [RDMA_NL_RDMA_CM_ID_STATS] = { .dump = cma_get_id_stats,
4338                                        .module = THIS_MODULE },
4339 };
4340
4341 static int cma_init_net(struct net *net)
4342 {
4343         struct cma_pernet *pernet = cma_pernet(net);
4344
4345         idr_init(&pernet->tcp_ps);
4346         idr_init(&pernet->udp_ps);
4347         idr_init(&pernet->ipoib_ps);
4348         idr_init(&pernet->ib_ps);
4349
4350         return 0;
4351 }
4352
4353 static void cma_exit_net(struct net *net)
4354 {
4355         struct cma_pernet *pernet = cma_pernet(net);
4356
4357         idr_destroy(&pernet->tcp_ps);
4358         idr_destroy(&pernet->udp_ps);
4359         idr_destroy(&pernet->ipoib_ps);
4360         idr_destroy(&pernet->ib_ps);
4361 }
4362
4363 static struct pernet_operations cma_pernet_operations = {
4364         .init = cma_init_net,
4365         .exit = cma_exit_net,
4366         .id = &cma_pernet_id,
4367         .size = sizeof(struct cma_pernet),
4368 };
4369
4370 static int __init cma_init(void)
4371 {
4372         int ret;
4373
4374         cma_wq = alloc_ordered_workqueue("rdma_cm", WQ_MEM_RECLAIM);
4375         if (!cma_wq)
4376                 return -ENOMEM;
4377
4378         ret = register_pernet_subsys(&cma_pernet_operations);
4379         if (ret)
4380                 goto err_wq;
4381
4382         ib_sa_register_client(&sa_client);
4383         rdma_addr_register_client(&addr_client);
4384         register_netdevice_notifier(&cma_nb);
4385
4386         ret = ib_register_client(&cma_client);
4387         if (ret)
4388                 goto err;
4389
4390         if (ibnl_add_client(RDMA_NL_RDMA_CM, ARRAY_SIZE(cma_cb_table),
4391                             cma_cb_table))
4392                 pr_warn("RDMA CMA: failed to add netlink callback\n");
4393         cma_configfs_init();
4394
4395         return 0;
4396
4397 err:
4398         unregister_netdevice_notifier(&cma_nb);
4399         rdma_addr_unregister_client(&addr_client);
4400         ib_sa_unregister_client(&sa_client);
4401 err_wq:
4402         destroy_workqueue(cma_wq);
4403         return ret;
4404 }
4405
4406 static void __exit cma_cleanup(void)
4407 {
4408         cma_configfs_exit();
4409         ibnl_remove_client(RDMA_NL_RDMA_CM);
4410         ib_unregister_client(&cma_client);
4411         unregister_netdevice_notifier(&cma_nb);
4412         rdma_addr_unregister_client(&addr_client);
4413         ib_sa_unregister_client(&sa_client);
4414         unregister_pernet_subsys(&cma_pernet_operations);
4415         destroy_workqueue(cma_wq);
4416 }
4417
4418 module_init(cma_init);
4419 module_exit(cma_cleanup);