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 *src_addr,
1098                               struct sockaddr *dst_addr,
1099                               struct cma_hdr *hdr,
1100                               __be16 local_port)
1101 {
1102         struct sockaddr_in *ip4;
1103
1104         if (src_addr) {
1105                 ip4 = (struct sockaddr_in *)src_addr;
1106                 ip4->sin_family = AF_INET;
1107                 ip4->sin_addr.s_addr = hdr->dst_addr.ip4.addr;
1108                 ip4->sin_port = local_port;
1109         }
1110
1111         if (dst_addr) {
1112                 ip4 = (struct sockaddr_in *)dst_addr;
1113                 ip4->sin_family = AF_INET;
1114                 ip4->sin_addr.s_addr = hdr->src_addr.ip4.addr;
1115                 ip4->sin_port = hdr->port;
1116         }
1117 }
1118
1119 static void cma_save_ip6_info(struct sockaddr *src_addr,
1120                               struct sockaddr *dst_addr,
1121                               struct cma_hdr *hdr,
1122                               __be16 local_port)
1123 {
1124         struct sockaddr_in6 *ip6;
1125
1126         if (src_addr) {
1127                 ip6 = (struct sockaddr_in6 *)src_addr;
1128                 ip6->sin6_family = AF_INET6;
1129                 ip6->sin6_addr = hdr->dst_addr.ip6;
1130                 ip6->sin6_port = local_port;
1131         }
1132
1133         if (dst_addr) {
1134                 ip6 = (struct sockaddr_in6 *)dst_addr;
1135                 ip6->sin6_family = AF_INET6;
1136                 ip6->sin6_addr = hdr->src_addr.ip6;
1137                 ip6->sin6_port = hdr->port;
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(src_addr, dst_addr, hdr, port);
1163                 break;
1164         case 6:
1165                 cma_save_ip6_info(src_addr, dst_addr, hdr, port);
1166                 break;
1167         default:
1168                 return -EAFNOSUPPORT;
1169         }
1170
1171         return 0;
1172 }
1173
1174 static int cma_save_net_info(struct sockaddr *src_addr,
1175                              struct sockaddr *dst_addr,
1176                              struct rdma_cm_id *listen_id,
1177                              struct ib_cm_event *ib_event,
1178                              sa_family_t sa_family, __be64 service_id)
1179 {
1180         if (sa_family == AF_IB) {
1181                 if (ib_event->event == IB_CM_REQ_RECEIVED)
1182                         cma_save_ib_info(src_addr, dst_addr, listen_id,
1183                                          ib_event->param.req_rcvd.primary_path);
1184                 else if (ib_event->event == IB_CM_SIDR_REQ_RECEIVED)
1185                         cma_save_ib_info(src_addr, dst_addr, listen_id, NULL);
1186                 return 0;
1187         }
1188
1189         return cma_save_ip_info(src_addr, dst_addr, ib_event, service_id);
1190 }
1191
1192 static int cma_save_req_info(const struct ib_cm_event *ib_event,
1193                              struct cma_req_info *req)
1194 {
1195         const struct ib_cm_req_event_param *req_param =
1196                 &ib_event->param.req_rcvd;
1197         const struct ib_cm_sidr_req_event_param *sidr_param =
1198                 &ib_event->param.sidr_req_rcvd;
1199
1200         switch (ib_event->event) {
1201         case IB_CM_REQ_RECEIVED:
1202                 req->device     = req_param->listen_id->device;
1203                 req->port       = req_param->port;
1204                 memcpy(&req->local_gid, &req_param->primary_path->sgid,
1205                        sizeof(req->local_gid));
1206                 req->has_gid    = true;
1207                 req->service_id = req_param->primary_path->service_id;
1208                 req->pkey       = be16_to_cpu(req_param->primary_path->pkey);
1209                 if (req->pkey != req_param->bth_pkey)
1210                         pr_warn_ratelimited("RDMA CMA: got different BTH P_Key (0x%x) and primary path P_Key (0x%x)\n"
1211                                             "RDMA CMA: in the future this may cause the request to be dropped\n",
1212                                             req_param->bth_pkey, req->pkey);
1213                 break;
1214         case IB_CM_SIDR_REQ_RECEIVED:
1215                 req->device     = sidr_param->listen_id->device;
1216                 req->port       = sidr_param->port;
1217                 req->has_gid    = false;
1218                 req->service_id = sidr_param->service_id;
1219                 req->pkey       = sidr_param->pkey;
1220                 if (req->pkey != sidr_param->bth_pkey)
1221                         pr_warn_ratelimited("RDMA CMA: got different BTH P_Key (0x%x) and SIDR request payload P_Key (0x%x)\n"
1222                                             "RDMA CMA: in the future this may cause the request to be dropped\n",
1223                                             sidr_param->bth_pkey, req->pkey);
1224                 break;
1225         default:
1226                 return -EINVAL;
1227         }
1228
1229         return 0;
1230 }
1231
1232 static bool validate_ipv4_net_dev(struct net_device *net_dev,
1233                                   const struct sockaddr_in *dst_addr,
1234                                   const struct sockaddr_in *src_addr)
1235 {
1236         __be32 daddr = dst_addr->sin_addr.s_addr,
1237                saddr = src_addr->sin_addr.s_addr;
1238         struct fib_result res;
1239         struct flowi4 fl4;
1240         int err;
1241         bool ret;
1242
1243         if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1244             ipv4_is_lbcast(daddr) || ipv4_is_zeronet(saddr) ||
1245             ipv4_is_zeronet(daddr) || ipv4_is_loopback(daddr) ||
1246             ipv4_is_loopback(saddr))
1247                 return false;
1248
1249         memset(&fl4, 0, sizeof(fl4));
1250         fl4.flowi4_iif = net_dev->ifindex;
1251         fl4.daddr = daddr;
1252         fl4.saddr = saddr;
1253
1254         rcu_read_lock();
1255         err = fib_lookup(dev_net(net_dev), &fl4, &res, 0);
1256         ret = err == 0 && FIB_RES_DEV(res) == net_dev;
1257         rcu_read_unlock();
1258
1259         return ret;
1260 }
1261
1262 static bool validate_ipv6_net_dev(struct net_device *net_dev,
1263                                   const struct sockaddr_in6 *dst_addr,
1264                                   const struct sockaddr_in6 *src_addr)
1265 {
1266 #if IS_ENABLED(CONFIG_IPV6)
1267         const int strict = ipv6_addr_type(&dst_addr->sin6_addr) &
1268                            IPV6_ADDR_LINKLOCAL;
1269         struct rt6_info *rt = rt6_lookup(dev_net(net_dev), &dst_addr->sin6_addr,
1270                                          &src_addr->sin6_addr, net_dev->ifindex,
1271                                          strict);
1272         bool ret;
1273
1274         if (!rt)
1275                 return false;
1276
1277         ret = rt->rt6i_idev->dev == net_dev;
1278         ip6_rt_put(rt);
1279
1280         return ret;
1281 #else
1282         return false;
1283 #endif
1284 }
1285
1286 static bool validate_net_dev(struct net_device *net_dev,
1287                              const struct sockaddr *daddr,
1288                              const struct sockaddr *saddr)
1289 {
1290         const struct sockaddr_in *daddr4 = (const struct sockaddr_in *)daddr;
1291         const struct sockaddr_in *saddr4 = (const struct sockaddr_in *)saddr;
1292         const struct sockaddr_in6 *daddr6 = (const struct sockaddr_in6 *)daddr;
1293         const struct sockaddr_in6 *saddr6 = (const struct sockaddr_in6 *)saddr;
1294
1295         switch (daddr->sa_family) {
1296         case AF_INET:
1297                 return saddr->sa_family == AF_INET &&
1298                        validate_ipv4_net_dev(net_dev, daddr4, saddr4);
1299
1300         case AF_INET6:
1301                 return saddr->sa_family == AF_INET6 &&
1302                        validate_ipv6_net_dev(net_dev, daddr6, saddr6);
1303
1304         default:
1305                 return false;
1306         }
1307 }
1308
1309 static struct net_device *cma_get_net_dev(struct ib_cm_event *ib_event,
1310                                           const struct cma_req_info *req)
1311 {
1312         struct sockaddr_storage listen_addr_storage, src_addr_storage;
1313         struct sockaddr *listen_addr = (struct sockaddr *)&listen_addr_storage,
1314                         *src_addr = (struct sockaddr *)&src_addr_storage;
1315         struct net_device *net_dev;
1316         const union ib_gid *gid = req->has_gid ? &req->local_gid : NULL;
1317         int err;
1318
1319         err = cma_save_ip_info(listen_addr, src_addr, ib_event,
1320                                req->service_id);
1321         if (err)
1322                 return ERR_PTR(err);
1323
1324         net_dev = ib_get_net_dev_by_params(req->device, req->port, req->pkey,
1325                                            gid, listen_addr);
1326         if (!net_dev)
1327                 return ERR_PTR(-ENODEV);
1328
1329         if (!validate_net_dev(net_dev, listen_addr, src_addr)) {
1330                 dev_put(net_dev);
1331                 return ERR_PTR(-EHOSTUNREACH);
1332         }
1333
1334         return net_dev;
1335 }
1336
1337 static enum rdma_port_space rdma_ps_from_service_id(__be64 service_id)
1338 {
1339         return (be64_to_cpu(service_id) >> 16) & 0xffff;
1340 }
1341
1342 static bool cma_match_private_data(struct rdma_id_private *id_priv,
1343                                    const struct cma_hdr *hdr)
1344 {
1345         struct sockaddr *addr = cma_src_addr(id_priv);
1346         __be32 ip4_addr;
1347         struct in6_addr ip6_addr;
1348
1349         if (cma_any_addr(addr) && !id_priv->afonly)
1350                 return true;
1351
1352         switch (addr->sa_family) {
1353         case AF_INET:
1354                 ip4_addr = ((struct sockaddr_in *)addr)->sin_addr.s_addr;
1355                 if (cma_get_ip_ver(hdr) != 4)
1356                         return false;
1357                 if (!cma_any_addr(addr) &&
1358                     hdr->dst_addr.ip4.addr != ip4_addr)
1359                         return false;
1360                 break;
1361         case AF_INET6:
1362                 ip6_addr = ((struct sockaddr_in6 *)addr)->sin6_addr;
1363                 if (cma_get_ip_ver(hdr) != 6)
1364                         return false;
1365                 if (!cma_any_addr(addr) &&
1366                     memcmp(&hdr->dst_addr.ip6, &ip6_addr, sizeof(ip6_addr)))
1367                         return false;
1368                 break;
1369         case AF_IB:
1370                 return true;
1371         default:
1372                 return false;
1373         }
1374
1375         return true;
1376 }
1377
1378 static bool cma_protocol_roce_dev_port(struct ib_device *device, int port_num)
1379 {
1380         enum rdma_link_layer ll = rdma_port_get_link_layer(device, port_num);
1381         enum rdma_transport_type transport =
1382                 rdma_node_get_transport(device->node_type);
1383
1384         return ll == IB_LINK_LAYER_ETHERNET && transport == RDMA_TRANSPORT_IB;
1385 }
1386
1387 static bool cma_protocol_roce(const struct rdma_cm_id *id)
1388 {
1389         struct ib_device *device = id->device;
1390         const int port_num = id->port_num ?: rdma_start_port(device);
1391
1392         return cma_protocol_roce_dev_port(device, port_num);
1393 }
1394
1395 static bool cma_match_net_dev(const struct rdma_cm_id *id,
1396                               const struct net_device *net_dev,
1397                               u8 port_num)
1398 {
1399         const struct rdma_addr *addr = &id->route.addr;
1400
1401         if (!net_dev)
1402                 /* This request is an AF_IB request or a RoCE request */
1403                 return (!id->port_num || id->port_num == port_num) &&
1404                        (addr->src_addr.ss_family == AF_IB ||
1405                         cma_protocol_roce_dev_port(id->device, port_num));
1406
1407         return !addr->dev_addr.bound_dev_if ||
1408                (net_eq(dev_net(net_dev), addr->dev_addr.net) &&
1409                 addr->dev_addr.bound_dev_if == net_dev->ifindex);
1410 }
1411
1412 static struct rdma_id_private *cma_find_listener(
1413                 const struct rdma_bind_list *bind_list,
1414                 const struct ib_cm_id *cm_id,
1415                 const struct ib_cm_event *ib_event,
1416                 const struct cma_req_info *req,
1417                 const struct net_device *net_dev)
1418 {
1419         struct rdma_id_private *id_priv, *id_priv_dev;
1420
1421         if (!bind_list)
1422                 return ERR_PTR(-EINVAL);
1423
1424         hlist_for_each_entry(id_priv, &bind_list->owners, node) {
1425                 if (cma_match_private_data(id_priv, ib_event->private_data)) {
1426                         if (id_priv->id.device == cm_id->device &&
1427                             cma_match_net_dev(&id_priv->id, net_dev, req->port))
1428                                 return id_priv;
1429                         list_for_each_entry(id_priv_dev,
1430                                             &id_priv->listen_list,
1431                                             listen_list) {
1432                                 if (id_priv_dev->id.device == cm_id->device &&
1433                                     cma_match_net_dev(&id_priv_dev->id, net_dev, req->port))
1434                                         return id_priv_dev;
1435                         }
1436                 }
1437         }
1438
1439         return ERR_PTR(-EINVAL);
1440 }
1441
1442 static struct rdma_id_private *cma_id_from_event(struct ib_cm_id *cm_id,
1443                                                  struct ib_cm_event *ib_event,
1444                                                  struct net_device **net_dev)
1445 {
1446         struct cma_req_info req;
1447         struct rdma_bind_list *bind_list;
1448         struct rdma_id_private *id_priv;
1449         int err;
1450
1451         err = cma_save_req_info(ib_event, &req);
1452         if (err)
1453                 return ERR_PTR(err);
1454
1455         *net_dev = cma_get_net_dev(ib_event, &req);
1456         if (IS_ERR(*net_dev)) {
1457                 if (PTR_ERR(*net_dev) == -EAFNOSUPPORT) {
1458                         /* Assuming the protocol is AF_IB */
1459                         *net_dev = NULL;
1460                 } else if (cma_protocol_roce_dev_port(req.device, req.port)) {
1461                         /* TODO find the net dev matching the request parameters
1462                          * through the RoCE GID table */
1463                         *net_dev = NULL;
1464                 } else {
1465                         return ERR_CAST(*net_dev);
1466                 }
1467         }
1468
1469         bind_list = cma_ps_find(*net_dev ? dev_net(*net_dev) : &init_net,
1470                                 rdma_ps_from_service_id(req.service_id),
1471                                 cma_port_from_service_id(req.service_id));
1472         id_priv = cma_find_listener(bind_list, cm_id, ib_event, &req, *net_dev);
1473         if (IS_ERR(id_priv) && *net_dev) {
1474                 dev_put(*net_dev);
1475                 *net_dev = NULL;
1476         }
1477
1478         return id_priv;
1479 }
1480
1481 static inline int cma_user_data_offset(struct rdma_id_private *id_priv)
1482 {
1483         return cma_family(id_priv) == AF_IB ? 0 : sizeof(struct cma_hdr);
1484 }
1485
1486 static void cma_cancel_route(struct rdma_id_private *id_priv)
1487 {
1488         if (rdma_cap_ib_sa(id_priv->id.device, id_priv->id.port_num)) {
1489                 if (id_priv->query)
1490                         ib_sa_cancel_query(id_priv->query_id, id_priv->query);
1491         }
1492 }
1493
1494 static void cma_cancel_listens(struct rdma_id_private *id_priv)
1495 {
1496         struct rdma_id_private *dev_id_priv;
1497
1498         /*
1499          * Remove from listen_any_list to prevent added devices from spawning
1500          * additional listen requests.
1501          */
1502         mutex_lock(&lock);
1503         list_del(&id_priv->list);
1504
1505         while (!list_empty(&id_priv->listen_list)) {
1506                 dev_id_priv = list_entry(id_priv->listen_list.next,
1507                                          struct rdma_id_private, listen_list);
1508                 /* sync with device removal to avoid duplicate destruction */
1509                 list_del_init(&dev_id_priv->list);
1510                 list_del(&dev_id_priv->listen_list);
1511                 mutex_unlock(&lock);
1512
1513                 rdma_destroy_id(&dev_id_priv->id);
1514                 mutex_lock(&lock);
1515         }
1516         mutex_unlock(&lock);
1517 }
1518
1519 static void cma_cancel_operation(struct rdma_id_private *id_priv,
1520                                  enum rdma_cm_state state)
1521 {
1522         switch (state) {
1523         case RDMA_CM_ADDR_QUERY:
1524                 rdma_addr_cancel(&id_priv->id.route.addr.dev_addr);
1525                 break;
1526         case RDMA_CM_ROUTE_QUERY:
1527                 cma_cancel_route(id_priv);
1528                 break;
1529         case RDMA_CM_LISTEN:
1530                 if (cma_any_addr(cma_src_addr(id_priv)) && !id_priv->cma_dev)
1531                         cma_cancel_listens(id_priv);
1532                 break;
1533         default:
1534                 break;
1535         }
1536 }
1537
1538 static void cma_release_port(struct rdma_id_private *id_priv)
1539 {
1540         struct rdma_bind_list *bind_list = id_priv->bind_list;
1541         struct net *net = id_priv->id.route.addr.dev_addr.net;
1542
1543         if (!bind_list)
1544                 return;
1545
1546         mutex_lock(&lock);
1547         hlist_del(&id_priv->node);
1548         if (hlist_empty(&bind_list->owners)) {
1549                 cma_ps_remove(net, bind_list->ps, bind_list->port);
1550                 kfree(bind_list);
1551         }
1552         mutex_unlock(&lock);
1553 }
1554
1555 static void cma_leave_mc_groups(struct rdma_id_private *id_priv)
1556 {
1557         struct cma_multicast *mc;
1558
1559         while (!list_empty(&id_priv->mc_list)) {
1560                 mc = container_of(id_priv->mc_list.next,
1561                                   struct cma_multicast, list);
1562                 list_del(&mc->list);
1563                 if (rdma_cap_ib_mcast(id_priv->cma_dev->device,
1564                                       id_priv->id.port_num)) {
1565                         ib_sa_free_multicast(mc->multicast.ib);
1566                         kfree(mc);
1567                 } else {
1568                         if (mc->igmp_joined) {
1569                                 struct rdma_dev_addr *dev_addr =
1570                                         &id_priv->id.route.addr.dev_addr;
1571                                 struct net_device *ndev = NULL;
1572
1573                                 if (dev_addr->bound_dev_if)
1574                                         ndev = dev_get_by_index(&init_net,
1575                                                                 dev_addr->bound_dev_if);
1576                                 if (ndev) {
1577                                         cma_igmp_send(ndev,
1578                                                       &mc->multicast.ib->rec.mgid,
1579                                                       false);
1580                                         dev_put(ndev);
1581                                 }
1582                         }
1583                         kref_put(&mc->mcref, release_mc);
1584                 }
1585         }
1586 }
1587
1588 void rdma_destroy_id(struct rdma_cm_id *id)
1589 {
1590         struct rdma_id_private *id_priv;
1591         enum rdma_cm_state state;
1592
1593         id_priv = container_of(id, struct rdma_id_private, id);
1594         state = cma_exch(id_priv, RDMA_CM_DESTROYING);
1595         cma_cancel_operation(id_priv, state);
1596
1597         /*
1598          * Wait for any active callback to finish.  New callbacks will find
1599          * the id_priv state set to destroying and abort.
1600          */
1601         mutex_lock(&id_priv->handler_mutex);
1602         mutex_unlock(&id_priv->handler_mutex);
1603
1604         if (id_priv->cma_dev) {
1605                 if (rdma_cap_ib_cm(id_priv->id.device, 1)) {
1606                         if (id_priv->cm_id.ib)
1607                                 ib_destroy_cm_id(id_priv->cm_id.ib);
1608                 } else if (rdma_cap_iw_cm(id_priv->id.device, 1)) {
1609                         if (id_priv->cm_id.iw)
1610                                 iw_destroy_cm_id(id_priv->cm_id.iw);
1611                 }
1612                 cma_leave_mc_groups(id_priv);
1613                 cma_release_dev(id_priv);
1614         }
1615
1616         cma_release_port(id_priv);
1617         cma_deref_id(id_priv);
1618         wait_for_completion(&id_priv->comp);
1619
1620         if (id_priv->internal_id)
1621                 cma_deref_id(id_priv->id.context);
1622
1623         kfree(id_priv->id.route.path_rec);
1624         put_net(id_priv->id.route.addr.dev_addr.net);
1625         kfree(id_priv);
1626 }
1627 EXPORT_SYMBOL(rdma_destroy_id);
1628
1629 static int cma_rep_recv(struct rdma_id_private *id_priv)
1630 {
1631         int ret;
1632
1633         ret = cma_modify_qp_rtr(id_priv, NULL);
1634         if (ret)
1635                 goto reject;
1636
1637         ret = cma_modify_qp_rts(id_priv, NULL);
1638         if (ret)
1639                 goto reject;
1640
1641         ret = ib_send_cm_rtu(id_priv->cm_id.ib, NULL, 0);
1642         if (ret)
1643                 goto reject;
1644
1645         return 0;
1646 reject:
1647         cma_modify_qp_err(id_priv);
1648         ib_send_cm_rej(id_priv->cm_id.ib, IB_CM_REJ_CONSUMER_DEFINED,
1649                        NULL, 0, NULL, 0);
1650         return ret;
1651 }
1652
1653 static void cma_set_rep_event_data(struct rdma_cm_event *event,
1654                                    struct ib_cm_rep_event_param *rep_data,
1655                                    void *private_data)
1656 {
1657         event->param.conn.private_data = private_data;
1658         event->param.conn.private_data_len = IB_CM_REP_PRIVATE_DATA_SIZE;
1659         event->param.conn.responder_resources = rep_data->responder_resources;
1660         event->param.conn.initiator_depth = rep_data->initiator_depth;
1661         event->param.conn.flow_control = rep_data->flow_control;
1662         event->param.conn.rnr_retry_count = rep_data->rnr_retry_count;
1663         event->param.conn.srq = rep_data->srq;
1664         event->param.conn.qp_num = rep_data->remote_qpn;
1665 }
1666
1667 static int cma_ib_handler(struct ib_cm_id *cm_id, struct ib_cm_event *ib_event)
1668 {
1669         struct rdma_id_private *id_priv = cm_id->context;
1670         struct rdma_cm_event event;
1671         int ret = 0;
1672
1673         mutex_lock(&id_priv->handler_mutex);
1674         if ((ib_event->event != IB_CM_TIMEWAIT_EXIT &&
1675              id_priv->state != RDMA_CM_CONNECT) ||
1676             (ib_event->event == IB_CM_TIMEWAIT_EXIT &&
1677              id_priv->state != RDMA_CM_DISCONNECT))
1678                 goto out;
1679
1680         memset(&event, 0, sizeof event);
1681         switch (ib_event->event) {
1682         case IB_CM_REQ_ERROR:
1683         case IB_CM_REP_ERROR:
1684                 event.event = RDMA_CM_EVENT_UNREACHABLE;
1685                 event.status = -ETIMEDOUT;
1686                 break;
1687         case IB_CM_REP_RECEIVED:
1688                 if (id_priv->id.qp) {
1689                         event.status = cma_rep_recv(id_priv);
1690                         event.event = event.status ? RDMA_CM_EVENT_CONNECT_ERROR :
1691                                                      RDMA_CM_EVENT_ESTABLISHED;
1692                 } else {
1693                         event.event = RDMA_CM_EVENT_CONNECT_RESPONSE;
1694                 }
1695                 cma_set_rep_event_data(&event, &ib_event->param.rep_rcvd,
1696                                        ib_event->private_data);
1697                 break;
1698         case IB_CM_RTU_RECEIVED:
1699         case IB_CM_USER_ESTABLISHED:
1700                 event.event = RDMA_CM_EVENT_ESTABLISHED;
1701                 break;
1702         case IB_CM_DREQ_ERROR:
1703                 event.status = -ETIMEDOUT; /* fall through */
1704         case IB_CM_DREQ_RECEIVED:
1705         case IB_CM_DREP_RECEIVED:
1706                 if (!cma_comp_exch(id_priv, RDMA_CM_CONNECT,
1707                                    RDMA_CM_DISCONNECT))
1708                         goto out;
1709                 event.event = RDMA_CM_EVENT_DISCONNECTED;
1710                 break;
1711         case IB_CM_TIMEWAIT_EXIT:
1712                 event.event = RDMA_CM_EVENT_TIMEWAIT_EXIT;
1713                 break;
1714         case IB_CM_MRA_RECEIVED:
1715                 /* ignore event */
1716                 goto out;
1717         case IB_CM_REJ_RECEIVED:
1718                 cma_modify_qp_err(id_priv);
1719                 event.status = ib_event->param.rej_rcvd.reason;
1720                 event.event = RDMA_CM_EVENT_REJECTED;
1721                 event.param.conn.private_data = ib_event->private_data;
1722                 event.param.conn.private_data_len = IB_CM_REJ_PRIVATE_DATA_SIZE;
1723                 break;
1724         default:
1725                 pr_err("RDMA CMA: unexpected IB CM event: %d\n",
1726                        ib_event->event);
1727                 goto out;
1728         }
1729
1730         ret = id_priv->id.event_handler(&id_priv->id, &event);
1731         if (ret) {
1732                 /* Destroy the CM ID by returning a non-zero value. */
1733                 id_priv->cm_id.ib = NULL;
1734                 cma_exch(id_priv, RDMA_CM_DESTROYING);
1735                 mutex_unlock(&id_priv->handler_mutex);
1736                 rdma_destroy_id(&id_priv->id);
1737                 return ret;
1738         }
1739 out:
1740         mutex_unlock(&id_priv->handler_mutex);
1741         return ret;
1742 }
1743
1744 static struct rdma_id_private *cma_new_conn_id(struct rdma_cm_id *listen_id,
1745                                                struct ib_cm_event *ib_event,
1746                                                struct net_device *net_dev)
1747 {
1748         struct rdma_id_private *id_priv;
1749         struct rdma_cm_id *id;
1750         struct rdma_route *rt;
1751         const sa_family_t ss_family = listen_id->route.addr.src_addr.ss_family;
1752         const __be64 service_id =
1753                       ib_event->param.req_rcvd.primary_path->service_id;
1754         int ret;
1755
1756         id = rdma_create_id(listen_id->route.addr.dev_addr.net,
1757                             listen_id->event_handler, listen_id->context,
1758                             listen_id->ps, ib_event->param.req_rcvd.qp_type);
1759         if (IS_ERR(id))
1760                 return NULL;
1761
1762         id_priv = container_of(id, struct rdma_id_private, id);
1763         if (cma_save_net_info((struct sockaddr *)&id->route.addr.src_addr,
1764                               (struct sockaddr *)&id->route.addr.dst_addr,
1765                               listen_id, ib_event, ss_family, service_id))
1766                 goto err;
1767
1768         rt = &id->route;
1769         rt->num_paths = ib_event->param.req_rcvd.alternate_path ? 2 : 1;
1770         rt->path_rec = kmalloc(sizeof *rt->path_rec * rt->num_paths,
1771                                GFP_KERNEL);
1772         if (!rt->path_rec)
1773                 goto err;
1774
1775         rt->path_rec[0] = *ib_event->param.req_rcvd.primary_path;
1776         if (rt->num_paths == 2)
1777                 rt->path_rec[1] = *ib_event->param.req_rcvd.alternate_path;
1778
1779         if (net_dev) {
1780                 ret = rdma_copy_addr(&rt->addr.dev_addr, net_dev, NULL);
1781                 if (ret)
1782                         goto err;
1783         } else {
1784                 if (!cma_protocol_roce(listen_id) &&
1785                     cma_any_addr(cma_src_addr(id_priv))) {
1786                         rt->addr.dev_addr.dev_type = ARPHRD_INFINIBAND;
1787                         rdma_addr_set_sgid(&rt->addr.dev_addr, &rt->path_rec[0].sgid);
1788                         ib_addr_set_pkey(&rt->addr.dev_addr, be16_to_cpu(rt->path_rec[0].pkey));
1789                 } else if (!cma_any_addr(cma_src_addr(id_priv))) {
1790                         ret = cma_translate_addr(cma_src_addr(id_priv), &rt->addr.dev_addr);
1791                         if (ret)
1792                                 goto err;
1793                 }
1794         }
1795         rdma_addr_set_dgid(&rt->addr.dev_addr, &rt->path_rec[0].dgid);
1796
1797         id_priv->state = RDMA_CM_CONNECT;
1798         return id_priv;
1799
1800 err:
1801         rdma_destroy_id(id);
1802         return NULL;
1803 }
1804
1805 static struct rdma_id_private *cma_new_udp_id(struct rdma_cm_id *listen_id,
1806                                               struct ib_cm_event *ib_event,
1807                                               struct net_device *net_dev)
1808 {
1809         struct rdma_id_private *id_priv;
1810         struct rdma_cm_id *id;
1811         const sa_family_t ss_family = listen_id->route.addr.src_addr.ss_family;
1812         struct net *net = listen_id->route.addr.dev_addr.net;
1813         int ret;
1814
1815         id = rdma_create_id(net, listen_id->event_handler, listen_id->context,
1816                             listen_id->ps, IB_QPT_UD);
1817         if (IS_ERR(id))
1818                 return NULL;
1819
1820         id_priv = container_of(id, struct rdma_id_private, id);
1821         if (cma_save_net_info((struct sockaddr *)&id->route.addr.src_addr,
1822                               (struct sockaddr *)&id->route.addr.dst_addr,
1823                               listen_id, ib_event, ss_family,
1824                               ib_event->param.sidr_req_rcvd.service_id))
1825                 goto err;
1826
1827         if (net_dev) {
1828                 ret = rdma_copy_addr(&id->route.addr.dev_addr, net_dev, NULL);
1829                 if (ret)
1830                         goto err;
1831         } else {
1832                 if (!cma_any_addr(cma_src_addr(id_priv))) {
1833                         ret = cma_translate_addr(cma_src_addr(id_priv),
1834                                                  &id->route.addr.dev_addr);
1835                         if (ret)
1836                                 goto err;
1837                 }
1838         }
1839
1840         id_priv->state = RDMA_CM_CONNECT;
1841         return id_priv;
1842 err:
1843         rdma_destroy_id(id);
1844         return NULL;
1845 }
1846
1847 static void cma_set_req_event_data(struct rdma_cm_event *event,
1848                                    struct ib_cm_req_event_param *req_data,
1849                                    void *private_data, int offset)
1850 {
1851         event->param.conn.private_data = private_data + offset;
1852         event->param.conn.private_data_len = IB_CM_REQ_PRIVATE_DATA_SIZE - offset;
1853         event->param.conn.responder_resources = req_data->responder_resources;
1854         event->param.conn.initiator_depth = req_data->initiator_depth;
1855         event->param.conn.flow_control = req_data->flow_control;
1856         event->param.conn.retry_count = req_data->retry_count;
1857         event->param.conn.rnr_retry_count = req_data->rnr_retry_count;
1858         event->param.conn.srq = req_data->srq;
1859         event->param.conn.qp_num = req_data->remote_qpn;
1860 }
1861
1862 static int cma_check_req_qp_type(struct rdma_cm_id *id, struct ib_cm_event *ib_event)
1863 {
1864         return (((ib_event->event == IB_CM_REQ_RECEIVED) &&
1865                  (ib_event->param.req_rcvd.qp_type == id->qp_type)) ||
1866                 ((ib_event->event == IB_CM_SIDR_REQ_RECEIVED) &&
1867                  (id->qp_type == IB_QPT_UD)) ||
1868                 (!id->qp_type));
1869 }
1870
1871 static int cma_req_handler(struct ib_cm_id *cm_id, struct ib_cm_event *ib_event)
1872 {
1873         struct rdma_id_private *listen_id, *conn_id = NULL;
1874         struct rdma_cm_event event;
1875         struct net_device *net_dev;
1876         int offset, ret;
1877
1878         listen_id = cma_id_from_event(cm_id, ib_event, &net_dev);
1879         if (IS_ERR(listen_id))
1880                 return PTR_ERR(listen_id);
1881
1882         if (!cma_check_req_qp_type(&listen_id->id, ib_event)) {
1883                 ret = -EINVAL;
1884                 goto net_dev_put;
1885         }
1886
1887         mutex_lock(&listen_id->handler_mutex);
1888         if (listen_id->state != RDMA_CM_LISTEN) {
1889                 ret = -ECONNABORTED;
1890                 goto err1;
1891         }
1892
1893         memset(&event, 0, sizeof event);
1894         offset = cma_user_data_offset(listen_id);
1895         event.event = RDMA_CM_EVENT_CONNECT_REQUEST;
1896         if (ib_event->event == IB_CM_SIDR_REQ_RECEIVED) {
1897                 conn_id = cma_new_udp_id(&listen_id->id, ib_event, net_dev);
1898                 event.param.ud.private_data = ib_event->private_data + offset;
1899                 event.param.ud.private_data_len =
1900                                 IB_CM_SIDR_REQ_PRIVATE_DATA_SIZE - offset;
1901         } else {
1902                 conn_id = cma_new_conn_id(&listen_id->id, ib_event, net_dev);
1903                 cma_set_req_event_data(&event, &ib_event->param.req_rcvd,
1904                                        ib_event->private_data, offset);
1905         }
1906         if (!conn_id) {
1907                 ret = -ENOMEM;
1908                 goto err1;
1909         }
1910
1911         mutex_lock_nested(&conn_id->handler_mutex, SINGLE_DEPTH_NESTING);
1912         ret = cma_acquire_dev(conn_id, listen_id);
1913         if (ret)
1914                 goto err2;
1915
1916         conn_id->cm_id.ib = cm_id;
1917         cm_id->context = conn_id;
1918         cm_id->cm_handler = cma_ib_handler;
1919
1920         /*
1921          * Protect against the user destroying conn_id from another thread
1922          * until we're done accessing it.
1923          */
1924         atomic_inc(&conn_id->refcount);
1925         ret = conn_id->id.event_handler(&conn_id->id, &event);
1926         if (ret)
1927                 goto err3;
1928         /*
1929          * Acquire mutex to prevent user executing rdma_destroy_id()
1930          * while we're accessing the cm_id.
1931          */
1932         mutex_lock(&lock);
1933         if (cma_comp(conn_id, RDMA_CM_CONNECT) &&
1934             (conn_id->id.qp_type != IB_QPT_UD))
1935                 ib_send_cm_mra(cm_id, CMA_CM_MRA_SETTING, NULL, 0);
1936         mutex_unlock(&lock);
1937         mutex_unlock(&conn_id->handler_mutex);
1938         mutex_unlock(&listen_id->handler_mutex);
1939         cma_deref_id(conn_id);
1940         if (net_dev)
1941                 dev_put(net_dev);
1942         return 0;
1943
1944 err3:
1945         cma_deref_id(conn_id);
1946         /* Destroy the CM ID by returning a non-zero value. */
1947         conn_id->cm_id.ib = NULL;
1948 err2:
1949         cma_exch(conn_id, RDMA_CM_DESTROYING);
1950         mutex_unlock(&conn_id->handler_mutex);
1951 err1:
1952         mutex_unlock(&listen_id->handler_mutex);
1953         if (conn_id)
1954                 rdma_destroy_id(&conn_id->id);
1955
1956 net_dev_put:
1957         if (net_dev)
1958                 dev_put(net_dev);
1959
1960         return ret;
1961 }
1962
1963 __be64 rdma_get_service_id(struct rdma_cm_id *id, struct sockaddr *addr)
1964 {
1965         if (addr->sa_family == AF_IB)
1966                 return ((struct sockaddr_ib *) addr)->sib_sid;
1967
1968         return cpu_to_be64(((u64)id->ps << 16) + be16_to_cpu(cma_port(addr)));
1969 }
1970 EXPORT_SYMBOL(rdma_get_service_id);
1971
1972 static int cma_iw_handler(struct iw_cm_id *iw_id, struct iw_cm_event *iw_event)
1973 {
1974         struct rdma_id_private *id_priv = iw_id->context;
1975         struct rdma_cm_event event;
1976         int ret = 0;
1977         struct sockaddr *laddr = (struct sockaddr *)&iw_event->local_addr;
1978         struct sockaddr *raddr = (struct sockaddr *)&iw_event->remote_addr;
1979
1980         mutex_lock(&id_priv->handler_mutex);
1981         if (id_priv->state != RDMA_CM_CONNECT)
1982                 goto out;
1983
1984         memset(&event, 0, sizeof event);
1985         switch (iw_event->event) {
1986         case IW_CM_EVENT_CLOSE:
1987                 event.event = RDMA_CM_EVENT_DISCONNECTED;
1988                 break;
1989         case IW_CM_EVENT_CONNECT_REPLY:
1990                 memcpy(cma_src_addr(id_priv), laddr,
1991                        rdma_addr_size(laddr));
1992                 memcpy(cma_dst_addr(id_priv), raddr,
1993                        rdma_addr_size(raddr));
1994                 switch (iw_event->status) {
1995                 case 0:
1996                         event.event = RDMA_CM_EVENT_ESTABLISHED;
1997                         event.param.conn.initiator_depth = iw_event->ird;
1998                         event.param.conn.responder_resources = iw_event->ord;
1999                         break;
2000                 case -ECONNRESET:
2001                 case -ECONNREFUSED:
2002                         event.event = RDMA_CM_EVENT_REJECTED;
2003                         break;
2004                 case -ETIMEDOUT:
2005                         event.event = RDMA_CM_EVENT_UNREACHABLE;
2006                         break;
2007                 default:
2008                         event.event = RDMA_CM_EVENT_CONNECT_ERROR;
2009                         break;
2010                 }
2011                 break;
2012         case IW_CM_EVENT_ESTABLISHED:
2013                 event.event = RDMA_CM_EVENT_ESTABLISHED;
2014                 event.param.conn.initiator_depth = iw_event->ird;
2015                 event.param.conn.responder_resources = iw_event->ord;
2016                 break;
2017         default:
2018                 BUG_ON(1);
2019         }
2020
2021         event.status = iw_event->status;
2022         event.param.conn.private_data = iw_event->private_data;
2023         event.param.conn.private_data_len = iw_event->private_data_len;
2024         ret = id_priv->id.event_handler(&id_priv->id, &event);
2025         if (ret) {
2026                 /* Destroy the CM ID by returning a non-zero value. */
2027                 id_priv->cm_id.iw = NULL;
2028                 cma_exch(id_priv, RDMA_CM_DESTROYING);
2029                 mutex_unlock(&id_priv->handler_mutex);
2030                 rdma_destroy_id(&id_priv->id);
2031                 return ret;
2032         }
2033
2034 out:
2035         mutex_unlock(&id_priv->handler_mutex);
2036         return ret;
2037 }
2038
2039 static int iw_conn_req_handler(struct iw_cm_id *cm_id,
2040                                struct iw_cm_event *iw_event)
2041 {
2042         struct rdma_cm_id *new_cm_id;
2043         struct rdma_id_private *listen_id, *conn_id;
2044         struct rdma_cm_event event;
2045         int ret = -ECONNABORTED;
2046         struct sockaddr *laddr = (struct sockaddr *)&iw_event->local_addr;
2047         struct sockaddr *raddr = (struct sockaddr *)&iw_event->remote_addr;
2048
2049         listen_id = cm_id->context;
2050
2051         mutex_lock(&listen_id->handler_mutex);
2052         if (listen_id->state != RDMA_CM_LISTEN)
2053                 goto out;
2054
2055         /* Create a new RDMA id for the new IW CM ID */
2056         new_cm_id = rdma_create_id(listen_id->id.route.addr.dev_addr.net,
2057                                    listen_id->id.event_handler,
2058                                    listen_id->id.context,
2059                                    RDMA_PS_TCP, IB_QPT_RC);
2060         if (IS_ERR(new_cm_id)) {
2061                 ret = -ENOMEM;
2062                 goto out;
2063         }
2064         conn_id = container_of(new_cm_id, struct rdma_id_private, id);
2065         mutex_lock_nested(&conn_id->handler_mutex, SINGLE_DEPTH_NESTING);
2066         conn_id->state = RDMA_CM_CONNECT;
2067
2068         ret = rdma_translate_ip(laddr, &conn_id->id.route.addr.dev_addr, NULL);
2069         if (ret) {
2070                 mutex_unlock(&conn_id->handler_mutex);
2071                 rdma_destroy_id(new_cm_id);
2072                 goto out;
2073         }
2074
2075         ret = cma_acquire_dev(conn_id, listen_id);
2076         if (ret) {
2077                 mutex_unlock(&conn_id->handler_mutex);
2078                 rdma_destroy_id(new_cm_id);
2079                 goto out;
2080         }
2081
2082         conn_id->cm_id.iw = cm_id;
2083         cm_id->context = conn_id;
2084         cm_id->cm_handler = cma_iw_handler;
2085
2086         memcpy(cma_src_addr(conn_id), laddr, rdma_addr_size(laddr));
2087         memcpy(cma_dst_addr(conn_id), raddr, rdma_addr_size(raddr));
2088
2089         memset(&event, 0, sizeof event);
2090         event.event = RDMA_CM_EVENT_CONNECT_REQUEST;
2091         event.param.conn.private_data = iw_event->private_data;
2092         event.param.conn.private_data_len = iw_event->private_data_len;
2093         event.param.conn.initiator_depth = iw_event->ird;
2094         event.param.conn.responder_resources = iw_event->ord;
2095
2096         /*
2097          * Protect against the user destroying conn_id from another thread
2098          * until we're done accessing it.
2099          */
2100         atomic_inc(&conn_id->refcount);
2101         ret = conn_id->id.event_handler(&conn_id->id, &event);
2102         if (ret) {
2103                 /* User wants to destroy the CM ID */
2104                 conn_id->cm_id.iw = NULL;
2105                 cma_exch(conn_id, RDMA_CM_DESTROYING);
2106                 mutex_unlock(&conn_id->handler_mutex);
2107                 cma_deref_id(conn_id);
2108                 rdma_destroy_id(&conn_id->id);
2109                 goto out;
2110         }
2111
2112         mutex_unlock(&conn_id->handler_mutex);
2113         cma_deref_id(conn_id);
2114
2115 out:
2116         mutex_unlock(&listen_id->handler_mutex);
2117         return ret;
2118 }
2119
2120 static int cma_ib_listen(struct rdma_id_private *id_priv)
2121 {
2122         struct sockaddr *addr;
2123         struct ib_cm_id *id;
2124         __be64 svc_id;
2125
2126         addr = cma_src_addr(id_priv);
2127         svc_id = rdma_get_service_id(&id_priv->id, addr);
2128         id = ib_cm_insert_listen(id_priv->id.device, cma_req_handler, svc_id);
2129         if (IS_ERR(id))
2130                 return PTR_ERR(id);
2131         id_priv->cm_id.ib = id;
2132
2133         return 0;
2134 }
2135
2136 static int cma_iw_listen(struct rdma_id_private *id_priv, int backlog)
2137 {
2138         int ret;
2139         struct iw_cm_id *id;
2140
2141         id = iw_create_cm_id(id_priv->id.device,
2142                              iw_conn_req_handler,
2143                              id_priv);
2144         if (IS_ERR(id))
2145                 return PTR_ERR(id);
2146
2147         id->tos = id_priv->tos;
2148         id_priv->cm_id.iw = id;
2149
2150         memcpy(&id_priv->cm_id.iw->local_addr, cma_src_addr(id_priv),
2151                rdma_addr_size(cma_src_addr(id_priv)));
2152
2153         ret = iw_cm_listen(id_priv->cm_id.iw, backlog);
2154
2155         if (ret) {
2156                 iw_destroy_cm_id(id_priv->cm_id.iw);
2157                 id_priv->cm_id.iw = NULL;
2158         }
2159
2160         return ret;
2161 }
2162
2163 static int cma_listen_handler(struct rdma_cm_id *id,
2164                               struct rdma_cm_event *event)
2165 {
2166         struct rdma_id_private *id_priv = id->context;
2167
2168         id->context = id_priv->id.context;
2169         id->event_handler = id_priv->id.event_handler;
2170         return id_priv->id.event_handler(id, event);
2171 }
2172
2173 static void cma_listen_on_dev(struct rdma_id_private *id_priv,
2174                               struct cma_device *cma_dev)
2175 {
2176         struct rdma_id_private *dev_id_priv;
2177         struct rdma_cm_id *id;
2178         struct net *net = id_priv->id.route.addr.dev_addr.net;
2179         int ret;
2180
2181         if (cma_family(id_priv) == AF_IB && !rdma_cap_ib_cm(cma_dev->device, 1))
2182                 return;
2183
2184         id = rdma_create_id(net, cma_listen_handler, id_priv, id_priv->id.ps,
2185                             id_priv->id.qp_type);
2186         if (IS_ERR(id))
2187                 return;
2188
2189         dev_id_priv = container_of(id, struct rdma_id_private, id);
2190
2191         dev_id_priv->state = RDMA_CM_ADDR_BOUND;
2192         memcpy(cma_src_addr(dev_id_priv), cma_src_addr(id_priv),
2193                rdma_addr_size(cma_src_addr(id_priv)));
2194
2195         _cma_attach_to_dev(dev_id_priv, cma_dev);
2196         list_add_tail(&dev_id_priv->listen_list, &id_priv->listen_list);
2197         atomic_inc(&id_priv->refcount);
2198         dev_id_priv->internal_id = 1;
2199         dev_id_priv->afonly = id_priv->afonly;
2200
2201         ret = rdma_listen(id, id_priv->backlog);
2202         if (ret)
2203                 pr_warn("RDMA CMA: cma_listen_on_dev, error %d, listening on device %s\n",
2204                         ret, cma_dev->device->name);
2205 }
2206
2207 static void cma_listen_on_all(struct rdma_id_private *id_priv)
2208 {
2209         struct cma_device *cma_dev;
2210
2211         mutex_lock(&lock);
2212         list_add_tail(&id_priv->list, &listen_any_list);
2213         list_for_each_entry(cma_dev, &dev_list, list)
2214                 cma_listen_on_dev(id_priv, cma_dev);
2215         mutex_unlock(&lock);
2216 }
2217
2218 void rdma_set_service_type(struct rdma_cm_id *id, int tos)
2219 {
2220         struct rdma_id_private *id_priv;
2221
2222         id_priv = container_of(id, struct rdma_id_private, id);
2223         id_priv->tos = (u8) tos;
2224 }
2225 EXPORT_SYMBOL(rdma_set_service_type);
2226
2227 static void cma_query_handler(int status, struct ib_sa_path_rec *path_rec,
2228                               void *context)
2229 {
2230         struct cma_work *work = context;
2231         struct rdma_route *route;
2232
2233         route = &work->id->id.route;
2234
2235         if (!status) {
2236                 route->num_paths = 1;
2237                 *route->path_rec = *path_rec;
2238         } else {
2239                 work->old_state = RDMA_CM_ROUTE_QUERY;
2240                 work->new_state = RDMA_CM_ADDR_RESOLVED;
2241                 work->event.event = RDMA_CM_EVENT_ROUTE_ERROR;
2242                 work->event.status = status;
2243         }
2244
2245         queue_work(cma_wq, &work->work);
2246 }
2247
2248 static int cma_query_ib_route(struct rdma_id_private *id_priv, int timeout_ms,
2249                               struct cma_work *work)
2250 {
2251         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
2252         struct ib_sa_path_rec path_rec;
2253         ib_sa_comp_mask comp_mask;
2254         struct sockaddr_in6 *sin6;
2255         struct sockaddr_ib *sib;
2256
2257         memset(&path_rec, 0, sizeof path_rec);
2258         rdma_addr_get_sgid(dev_addr, &path_rec.sgid);
2259         rdma_addr_get_dgid(dev_addr, &path_rec.dgid);
2260         path_rec.pkey = cpu_to_be16(ib_addr_get_pkey(dev_addr));
2261         path_rec.numb_path = 1;
2262         path_rec.reversible = 1;
2263         path_rec.service_id = rdma_get_service_id(&id_priv->id, cma_dst_addr(id_priv));
2264
2265         comp_mask = IB_SA_PATH_REC_DGID | IB_SA_PATH_REC_SGID |
2266                     IB_SA_PATH_REC_PKEY | IB_SA_PATH_REC_NUMB_PATH |
2267                     IB_SA_PATH_REC_REVERSIBLE | IB_SA_PATH_REC_SERVICE_ID;
2268
2269         switch (cma_family(id_priv)) {
2270         case AF_INET:
2271                 path_rec.qos_class = cpu_to_be16((u16) id_priv->tos);
2272                 comp_mask |= IB_SA_PATH_REC_QOS_CLASS;
2273                 break;
2274         case AF_INET6:
2275                 sin6 = (struct sockaddr_in6 *) cma_src_addr(id_priv);
2276                 path_rec.traffic_class = (u8) (be32_to_cpu(sin6->sin6_flowinfo) >> 20);
2277                 comp_mask |= IB_SA_PATH_REC_TRAFFIC_CLASS;
2278                 break;
2279         case AF_IB:
2280                 sib = (struct sockaddr_ib *) cma_src_addr(id_priv);
2281                 path_rec.traffic_class = (u8) (be32_to_cpu(sib->sib_flowinfo) >> 20);
2282                 comp_mask |= IB_SA_PATH_REC_TRAFFIC_CLASS;
2283                 break;
2284         }
2285
2286         id_priv->query_id = ib_sa_path_rec_get(&sa_client, id_priv->id.device,
2287                                                id_priv->id.port_num, &path_rec,
2288                                                comp_mask, timeout_ms,
2289                                                GFP_KERNEL, cma_query_handler,
2290                                                work, &id_priv->query);
2291
2292         return (id_priv->query_id < 0) ? id_priv->query_id : 0;
2293 }
2294
2295 static void cma_work_handler(struct work_struct *_work)
2296 {
2297         struct cma_work *work = container_of(_work, struct cma_work, work);
2298         struct rdma_id_private *id_priv = work->id;
2299         int destroy = 0;
2300
2301         mutex_lock(&id_priv->handler_mutex);
2302         if (!cma_comp_exch(id_priv, work->old_state, work->new_state))
2303                 goto out;
2304
2305         if (id_priv->id.event_handler(&id_priv->id, &work->event)) {
2306                 cma_exch(id_priv, RDMA_CM_DESTROYING);
2307                 destroy = 1;
2308         }
2309 out:
2310         mutex_unlock(&id_priv->handler_mutex);
2311         cma_deref_id(id_priv);
2312         if (destroy)
2313                 rdma_destroy_id(&id_priv->id);
2314         kfree(work);
2315 }
2316
2317 static void cma_ndev_work_handler(struct work_struct *_work)
2318 {
2319         struct cma_ndev_work *work = container_of(_work, struct cma_ndev_work, work);
2320         struct rdma_id_private *id_priv = work->id;
2321         int destroy = 0;
2322
2323         mutex_lock(&id_priv->handler_mutex);
2324         if (id_priv->state == RDMA_CM_DESTROYING ||
2325             id_priv->state == RDMA_CM_DEVICE_REMOVAL)
2326                 goto out;
2327
2328         if (id_priv->id.event_handler(&id_priv->id, &work->event)) {
2329                 cma_exch(id_priv, RDMA_CM_DESTROYING);
2330                 destroy = 1;
2331         }
2332
2333 out:
2334         mutex_unlock(&id_priv->handler_mutex);
2335         cma_deref_id(id_priv);
2336         if (destroy)
2337                 rdma_destroy_id(&id_priv->id);
2338         kfree(work);
2339 }
2340
2341 static int cma_resolve_ib_route(struct rdma_id_private *id_priv, int timeout_ms)
2342 {
2343         struct rdma_route *route = &id_priv->id.route;
2344         struct cma_work *work;
2345         int ret;
2346
2347         work = kzalloc(sizeof *work, GFP_KERNEL);
2348         if (!work)
2349                 return -ENOMEM;
2350
2351         work->id = id_priv;
2352         INIT_WORK(&work->work, cma_work_handler);
2353         work->old_state = RDMA_CM_ROUTE_QUERY;
2354         work->new_state = RDMA_CM_ROUTE_RESOLVED;
2355         work->event.event = RDMA_CM_EVENT_ROUTE_RESOLVED;
2356
2357         route->path_rec = kmalloc(sizeof *route->path_rec, GFP_KERNEL);
2358         if (!route->path_rec) {
2359                 ret = -ENOMEM;
2360                 goto err1;
2361         }
2362
2363         ret = cma_query_ib_route(id_priv, timeout_ms, work);
2364         if (ret)
2365                 goto err2;
2366
2367         return 0;
2368 err2:
2369         kfree(route->path_rec);
2370         route->path_rec = NULL;
2371 err1:
2372         kfree(work);
2373         return ret;
2374 }
2375
2376 int rdma_set_ib_paths(struct rdma_cm_id *id,
2377                       struct ib_sa_path_rec *path_rec, int num_paths)
2378 {
2379         struct rdma_id_private *id_priv;
2380         int ret;
2381
2382         id_priv = container_of(id, struct rdma_id_private, id);
2383         if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_RESOLVED,
2384                            RDMA_CM_ROUTE_RESOLVED))
2385                 return -EINVAL;
2386
2387         id->route.path_rec = kmemdup(path_rec, sizeof *path_rec * num_paths,
2388                                      GFP_KERNEL);
2389         if (!id->route.path_rec) {
2390                 ret = -ENOMEM;
2391                 goto err;
2392         }
2393
2394         id->route.num_paths = num_paths;
2395         return 0;
2396 err:
2397         cma_comp_exch(id_priv, RDMA_CM_ROUTE_RESOLVED, RDMA_CM_ADDR_RESOLVED);
2398         return ret;
2399 }
2400 EXPORT_SYMBOL(rdma_set_ib_paths);
2401
2402 static int cma_resolve_iw_route(struct rdma_id_private *id_priv, int timeout_ms)
2403 {
2404         struct cma_work *work;
2405
2406         work = kzalloc(sizeof *work, GFP_KERNEL);
2407         if (!work)
2408                 return -ENOMEM;
2409
2410         work->id = id_priv;
2411         INIT_WORK(&work->work, cma_work_handler);
2412         work->old_state = RDMA_CM_ROUTE_QUERY;
2413         work->new_state = RDMA_CM_ROUTE_RESOLVED;
2414         work->event.event = RDMA_CM_EVENT_ROUTE_RESOLVED;
2415         queue_work(cma_wq, &work->work);
2416         return 0;
2417 }
2418
2419 static int iboe_tos_to_sl(struct net_device *ndev, int tos)
2420 {
2421         int prio;
2422         struct net_device *dev;
2423
2424         prio = rt_tos2priority(tos);
2425         dev = ndev->priv_flags & IFF_802_1Q_VLAN ?
2426                 vlan_dev_real_dev(ndev) : ndev;
2427
2428         if (dev->num_tc)
2429                 return netdev_get_prio_tc_map(dev, prio);
2430
2431 #if IS_ENABLED(CONFIG_VLAN_8021Q)
2432         if (ndev->priv_flags & IFF_802_1Q_VLAN)
2433                 return (vlan_dev_get_egress_qos_mask(ndev, prio) &
2434                         VLAN_PRIO_MASK) >> VLAN_PRIO_SHIFT;
2435 #endif
2436         return 0;
2437 }
2438
2439 static int cma_resolve_iboe_route(struct rdma_id_private *id_priv)
2440 {
2441         struct rdma_route *route = &id_priv->id.route;
2442         struct rdma_addr *addr = &route->addr;
2443         struct cma_work *work;
2444         int ret;
2445         struct net_device *ndev = NULL;
2446
2447
2448         work = kzalloc(sizeof *work, GFP_KERNEL);
2449         if (!work)
2450                 return -ENOMEM;
2451
2452         work->id = id_priv;
2453         INIT_WORK(&work->work, cma_work_handler);
2454
2455         route->path_rec = kzalloc(sizeof *route->path_rec, GFP_KERNEL);
2456         if (!route->path_rec) {
2457                 ret = -ENOMEM;
2458                 goto err1;
2459         }
2460
2461         route->num_paths = 1;
2462
2463         if (addr->dev_addr.bound_dev_if) {
2464                 ndev = dev_get_by_index(&init_net, addr->dev_addr.bound_dev_if);
2465                 if (!ndev) {
2466                         ret = -ENODEV;
2467                         goto err2;
2468                 }
2469
2470                 if (ndev->flags & IFF_LOOPBACK) {
2471                         dev_put(ndev);
2472                         if (!id_priv->id.device->get_netdev) {
2473                                 ret = -EOPNOTSUPP;
2474                                 goto err2;
2475                         }
2476
2477                         ndev = id_priv->id.device->get_netdev(id_priv->id.device,
2478                                                               id_priv->id.port_num);
2479                         if (!ndev) {
2480                                 ret = -ENODEV;
2481                                 goto err2;
2482                         }
2483                 }
2484
2485                 route->path_rec->net = &init_net;
2486                 route->path_rec->ifindex = ndev->ifindex;
2487                 route->path_rec->gid_type = id_priv->gid_type;
2488         }
2489         if (!ndev) {
2490                 ret = -ENODEV;
2491                 goto err2;
2492         }
2493
2494         memcpy(route->path_rec->dmac, addr->dev_addr.dst_dev_addr, ETH_ALEN);
2495
2496         rdma_ip2gid((struct sockaddr *)&id_priv->id.route.addr.src_addr,
2497                     &route->path_rec->sgid);
2498         rdma_ip2gid((struct sockaddr *)&id_priv->id.route.addr.dst_addr,
2499                     &route->path_rec->dgid);
2500
2501         /* Use the hint from IP Stack to select GID Type */
2502         if (route->path_rec->gid_type < ib_network_to_gid_type(addr->dev_addr.network))
2503                 route->path_rec->gid_type = ib_network_to_gid_type(addr->dev_addr.network);
2504         if (((struct sockaddr *)&id_priv->id.route.addr.dst_addr)->sa_family != AF_IB)
2505                 /* TODO: get the hoplimit from the inet/inet6 device */
2506                 route->path_rec->hop_limit = addr->dev_addr.hoplimit;
2507         else
2508                 route->path_rec->hop_limit = 1;
2509         route->path_rec->reversible = 1;
2510         route->path_rec->pkey = cpu_to_be16(0xffff);
2511         route->path_rec->mtu_selector = IB_SA_EQ;
2512         route->path_rec->sl = iboe_tos_to_sl(ndev, id_priv->tos);
2513         route->path_rec->mtu = iboe_get_mtu(ndev->mtu);
2514         route->path_rec->rate_selector = IB_SA_EQ;
2515         route->path_rec->rate = iboe_get_rate(ndev);
2516         dev_put(ndev);
2517         route->path_rec->packet_life_time_selector = IB_SA_EQ;
2518         route->path_rec->packet_life_time = CMA_IBOE_PACKET_LIFETIME;
2519         if (!route->path_rec->mtu) {
2520                 ret = -EINVAL;
2521                 goto err2;
2522         }
2523
2524         work->old_state = RDMA_CM_ROUTE_QUERY;
2525         work->new_state = RDMA_CM_ROUTE_RESOLVED;
2526         work->event.event = RDMA_CM_EVENT_ROUTE_RESOLVED;
2527         work->event.status = 0;
2528
2529         queue_work(cma_wq, &work->work);
2530
2531         return 0;
2532
2533 err2:
2534         kfree(route->path_rec);
2535         route->path_rec = NULL;
2536 err1:
2537         kfree(work);
2538         return ret;
2539 }
2540
2541 int rdma_resolve_route(struct rdma_cm_id *id, int timeout_ms)
2542 {
2543         struct rdma_id_private *id_priv;
2544         int ret;
2545
2546         id_priv = container_of(id, struct rdma_id_private, id);
2547         if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_RESOLVED, RDMA_CM_ROUTE_QUERY))
2548                 return -EINVAL;
2549
2550         atomic_inc(&id_priv->refcount);
2551         if (rdma_cap_ib_sa(id->device, id->port_num))
2552                 ret = cma_resolve_ib_route(id_priv, timeout_ms);
2553         else if (rdma_protocol_roce(id->device, id->port_num))
2554                 ret = cma_resolve_iboe_route(id_priv);
2555         else if (rdma_protocol_iwarp(id->device, id->port_num))
2556                 ret = cma_resolve_iw_route(id_priv, timeout_ms);
2557         else
2558                 ret = -ENOSYS;
2559
2560         if (ret)
2561                 goto err;
2562
2563         return 0;
2564 err:
2565         cma_comp_exch(id_priv, RDMA_CM_ROUTE_QUERY, RDMA_CM_ADDR_RESOLVED);
2566         cma_deref_id(id_priv);
2567         return ret;
2568 }
2569 EXPORT_SYMBOL(rdma_resolve_route);
2570
2571 static void cma_set_loopback(struct sockaddr *addr)
2572 {
2573         switch (addr->sa_family) {
2574         case AF_INET:
2575                 ((struct sockaddr_in *) addr)->sin_addr.s_addr = htonl(INADDR_LOOPBACK);
2576                 break;
2577         case AF_INET6:
2578                 ipv6_addr_set(&((struct sockaddr_in6 *) addr)->sin6_addr,
2579                               0, 0, 0, htonl(1));
2580                 break;
2581         default:
2582                 ib_addr_set(&((struct sockaddr_ib *) addr)->sib_addr,
2583                             0, 0, 0, htonl(1));
2584                 break;
2585         }
2586 }
2587
2588 static int cma_bind_loopback(struct rdma_id_private *id_priv)
2589 {
2590         struct cma_device *cma_dev, *cur_dev;
2591         struct ib_port_attr port_attr;
2592         union ib_gid gid;
2593         u16 pkey;
2594         int ret;
2595         u8 p;
2596
2597         cma_dev = NULL;
2598         mutex_lock(&lock);
2599         list_for_each_entry(cur_dev, &dev_list, list) {
2600                 if (cma_family(id_priv) == AF_IB &&
2601                     !rdma_cap_ib_cm(cur_dev->device, 1))
2602                         continue;
2603
2604                 if (!cma_dev)
2605                         cma_dev = cur_dev;
2606
2607                 for (p = 1; p <= cur_dev->device->phys_port_cnt; ++p) {
2608                         if (!ib_query_port(cur_dev->device, p, &port_attr) &&
2609                             port_attr.state == IB_PORT_ACTIVE) {
2610                                 cma_dev = cur_dev;
2611                                 goto port_found;
2612                         }
2613                 }
2614         }
2615
2616         if (!cma_dev) {
2617                 ret = -ENODEV;
2618                 goto out;
2619         }
2620
2621         p = 1;
2622
2623 port_found:
2624         ret = ib_get_cached_gid(cma_dev->device, p, 0, &gid, NULL);
2625         if (ret)
2626                 goto out;
2627
2628         ret = ib_get_cached_pkey(cma_dev->device, p, 0, &pkey);
2629         if (ret)
2630                 goto out;
2631
2632         id_priv->id.route.addr.dev_addr.dev_type =
2633                 (rdma_protocol_ib(cma_dev->device, p)) ?
2634                 ARPHRD_INFINIBAND : ARPHRD_ETHER;
2635
2636         rdma_addr_set_sgid(&id_priv->id.route.addr.dev_addr, &gid);
2637         ib_addr_set_pkey(&id_priv->id.route.addr.dev_addr, pkey);
2638         id_priv->id.port_num = p;
2639         cma_attach_to_dev(id_priv, cma_dev);
2640         cma_set_loopback(cma_src_addr(id_priv));
2641 out:
2642         mutex_unlock(&lock);
2643         return ret;
2644 }
2645
2646 static void addr_handler(int status, struct sockaddr *src_addr,
2647                          struct rdma_dev_addr *dev_addr, void *context)
2648 {
2649         struct rdma_id_private *id_priv = context;
2650         struct rdma_cm_event event;
2651
2652         memset(&event, 0, sizeof event);
2653         mutex_lock(&id_priv->handler_mutex);
2654         if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_QUERY,
2655                            RDMA_CM_ADDR_RESOLVED))
2656                 goto out;
2657
2658         memcpy(cma_src_addr(id_priv), src_addr, rdma_addr_size(src_addr));
2659         if (!status && !id_priv->cma_dev)
2660                 status = cma_acquire_dev(id_priv, NULL);
2661
2662         if (status) {
2663                 if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_RESOLVED,
2664                                    RDMA_CM_ADDR_BOUND))
2665                         goto out;
2666                 event.event = RDMA_CM_EVENT_ADDR_ERROR;
2667                 event.status = status;
2668         } else
2669                 event.event = RDMA_CM_EVENT_ADDR_RESOLVED;
2670
2671         if (id_priv->id.event_handler(&id_priv->id, &event)) {
2672                 cma_exch(id_priv, RDMA_CM_DESTROYING);
2673                 mutex_unlock(&id_priv->handler_mutex);
2674                 cma_deref_id(id_priv);
2675                 rdma_destroy_id(&id_priv->id);
2676                 return;
2677         }
2678 out:
2679         mutex_unlock(&id_priv->handler_mutex);
2680         cma_deref_id(id_priv);
2681 }
2682
2683 static int cma_resolve_loopback(struct rdma_id_private *id_priv)
2684 {
2685         struct cma_work *work;
2686         union ib_gid gid;
2687         int ret;
2688
2689         work = kzalloc(sizeof *work, GFP_KERNEL);
2690         if (!work)
2691                 return -ENOMEM;
2692
2693         if (!id_priv->cma_dev) {
2694                 ret = cma_bind_loopback(id_priv);
2695                 if (ret)
2696                         goto err;
2697         }
2698
2699         rdma_addr_get_sgid(&id_priv->id.route.addr.dev_addr, &gid);
2700         rdma_addr_set_dgid(&id_priv->id.route.addr.dev_addr, &gid);
2701
2702         work->id = id_priv;
2703         INIT_WORK(&work->work, cma_work_handler);
2704         work->old_state = RDMA_CM_ADDR_QUERY;
2705         work->new_state = RDMA_CM_ADDR_RESOLVED;
2706         work->event.event = RDMA_CM_EVENT_ADDR_RESOLVED;
2707         queue_work(cma_wq, &work->work);
2708         return 0;
2709 err:
2710         kfree(work);
2711         return ret;
2712 }
2713
2714 static int cma_resolve_ib_addr(struct rdma_id_private *id_priv)
2715 {
2716         struct cma_work *work;
2717         int ret;
2718
2719         work = kzalloc(sizeof *work, GFP_KERNEL);
2720         if (!work)
2721                 return -ENOMEM;
2722
2723         if (!id_priv->cma_dev) {
2724                 ret = cma_resolve_ib_dev(id_priv);
2725                 if (ret)
2726                         goto err;
2727         }
2728
2729         rdma_addr_set_dgid(&id_priv->id.route.addr.dev_addr, (union ib_gid *)
2730                 &(((struct sockaddr_ib *) &id_priv->id.route.addr.dst_addr)->sib_addr));
2731
2732         work->id = id_priv;
2733         INIT_WORK(&work->work, cma_work_handler);
2734         work->old_state = RDMA_CM_ADDR_QUERY;
2735         work->new_state = RDMA_CM_ADDR_RESOLVED;
2736         work->event.event = RDMA_CM_EVENT_ADDR_RESOLVED;
2737         queue_work(cma_wq, &work->work);
2738         return 0;
2739 err:
2740         kfree(work);
2741         return ret;
2742 }
2743
2744 static int cma_bind_addr(struct rdma_cm_id *id, struct sockaddr *src_addr,
2745                          struct sockaddr *dst_addr)
2746 {
2747         if (!src_addr || !src_addr->sa_family) {
2748                 src_addr = (struct sockaddr *) &id->route.addr.src_addr;
2749                 src_addr->sa_family = dst_addr->sa_family;
2750                 if (dst_addr->sa_family == AF_INET6) {
2751                         struct sockaddr_in6 *src_addr6 = (struct sockaddr_in6 *) src_addr;
2752                         struct sockaddr_in6 *dst_addr6 = (struct sockaddr_in6 *) dst_addr;
2753                         src_addr6->sin6_scope_id = dst_addr6->sin6_scope_id;
2754                         if (ipv6_addr_type(&dst_addr6->sin6_addr) & IPV6_ADDR_LINKLOCAL)
2755                                 id->route.addr.dev_addr.bound_dev_if = dst_addr6->sin6_scope_id;
2756                 } else if (dst_addr->sa_family == AF_IB) {
2757                         ((struct sockaddr_ib *) src_addr)->sib_pkey =
2758                                 ((struct sockaddr_ib *) dst_addr)->sib_pkey;
2759                 }
2760         }
2761         return rdma_bind_addr(id, src_addr);
2762 }
2763
2764 int rdma_resolve_addr(struct rdma_cm_id *id, struct sockaddr *src_addr,
2765                       struct sockaddr *dst_addr, int timeout_ms)
2766 {
2767         struct rdma_id_private *id_priv;
2768         int ret;
2769
2770         id_priv = container_of(id, struct rdma_id_private, id);
2771         if (id_priv->state == RDMA_CM_IDLE) {
2772                 ret = cma_bind_addr(id, src_addr, dst_addr);
2773                 if (ret)
2774                         return ret;
2775         }
2776
2777         if (cma_family(id_priv) != dst_addr->sa_family)
2778                 return -EINVAL;
2779
2780         if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_BOUND, RDMA_CM_ADDR_QUERY))
2781                 return -EINVAL;
2782
2783         atomic_inc(&id_priv->refcount);
2784         memcpy(cma_dst_addr(id_priv), dst_addr, rdma_addr_size(dst_addr));
2785         if (cma_any_addr(dst_addr)) {
2786                 ret = cma_resolve_loopback(id_priv);
2787         } else {
2788                 if (dst_addr->sa_family == AF_IB) {
2789                         ret = cma_resolve_ib_addr(id_priv);
2790                 } else {
2791                         ret = rdma_resolve_ip(&addr_client, cma_src_addr(id_priv),
2792                                               dst_addr, &id->route.addr.dev_addr,
2793                                               timeout_ms, addr_handler, id_priv);
2794                 }
2795         }
2796         if (ret)
2797                 goto err;
2798
2799         return 0;
2800 err:
2801         cma_comp_exch(id_priv, RDMA_CM_ADDR_QUERY, RDMA_CM_ADDR_BOUND);
2802         cma_deref_id(id_priv);
2803         return ret;
2804 }
2805 EXPORT_SYMBOL(rdma_resolve_addr);
2806
2807 int rdma_set_reuseaddr(struct rdma_cm_id *id, int reuse)
2808 {
2809         struct rdma_id_private *id_priv;
2810         unsigned long flags;
2811         int ret;
2812
2813         id_priv = container_of(id, struct rdma_id_private, id);
2814         spin_lock_irqsave(&id_priv->lock, flags);
2815         if (reuse || id_priv->state == RDMA_CM_IDLE) {
2816                 id_priv->reuseaddr = reuse;
2817                 ret = 0;
2818         } else {
2819                 ret = -EINVAL;
2820         }
2821         spin_unlock_irqrestore(&id_priv->lock, flags);
2822         return ret;
2823 }
2824 EXPORT_SYMBOL(rdma_set_reuseaddr);
2825
2826 int rdma_set_afonly(struct rdma_cm_id *id, int afonly)
2827 {
2828         struct rdma_id_private *id_priv;
2829         unsigned long flags;
2830         int ret;
2831
2832         id_priv = container_of(id, struct rdma_id_private, id);
2833         spin_lock_irqsave(&id_priv->lock, flags);
2834         if (id_priv->state == RDMA_CM_IDLE || id_priv->state == RDMA_CM_ADDR_BOUND) {
2835                 id_priv->options |= (1 << CMA_OPTION_AFONLY);
2836                 id_priv->afonly = afonly;
2837                 ret = 0;
2838         } else {
2839                 ret = -EINVAL;
2840         }
2841         spin_unlock_irqrestore(&id_priv->lock, flags);
2842         return ret;
2843 }
2844 EXPORT_SYMBOL(rdma_set_afonly);
2845
2846 static void cma_bind_port(struct rdma_bind_list *bind_list,
2847                           struct rdma_id_private *id_priv)
2848 {
2849         struct sockaddr *addr;
2850         struct sockaddr_ib *sib;
2851         u64 sid, mask;
2852         __be16 port;
2853
2854         addr = cma_src_addr(id_priv);
2855         port = htons(bind_list->port);
2856
2857         switch (addr->sa_family) {
2858         case AF_INET:
2859                 ((struct sockaddr_in *) addr)->sin_port = port;
2860                 break;
2861         case AF_INET6:
2862                 ((struct sockaddr_in6 *) addr)->sin6_port = port;
2863                 break;
2864         case AF_IB:
2865                 sib = (struct sockaddr_ib *) addr;
2866                 sid = be64_to_cpu(sib->sib_sid);
2867                 mask = be64_to_cpu(sib->sib_sid_mask);
2868                 sib->sib_sid = cpu_to_be64((sid & mask) | (u64) ntohs(port));
2869                 sib->sib_sid_mask = cpu_to_be64(~0ULL);
2870                 break;
2871         }
2872         id_priv->bind_list = bind_list;
2873         hlist_add_head(&id_priv->node, &bind_list->owners);
2874 }
2875
2876 static int cma_alloc_port(enum rdma_port_space ps,
2877                           struct rdma_id_private *id_priv, unsigned short snum)
2878 {
2879         struct rdma_bind_list *bind_list;
2880         int ret;
2881
2882         bind_list = kzalloc(sizeof *bind_list, GFP_KERNEL);
2883         if (!bind_list)
2884                 return -ENOMEM;
2885
2886         ret = cma_ps_alloc(id_priv->id.route.addr.dev_addr.net, ps, bind_list,
2887                            snum);
2888         if (ret < 0)
2889                 goto err;
2890
2891         bind_list->ps = ps;
2892         bind_list->port = (unsigned short)ret;
2893         cma_bind_port(bind_list, id_priv);
2894         return 0;
2895 err:
2896         kfree(bind_list);
2897         return ret == -ENOSPC ? -EADDRNOTAVAIL : ret;
2898 }
2899
2900 static int cma_alloc_any_port(enum rdma_port_space ps,
2901                               struct rdma_id_private *id_priv)
2902 {
2903         static unsigned int last_used_port;
2904         int low, high, remaining;
2905         unsigned int rover;
2906         struct net *net = id_priv->id.route.addr.dev_addr.net;
2907
2908         inet_get_local_port_range(net, &low, &high);
2909         remaining = (high - low) + 1;
2910         rover = prandom_u32() % remaining + low;
2911 retry:
2912         if (last_used_port != rover &&
2913             !cma_ps_find(net, ps, (unsigned short)rover)) {
2914                 int ret = cma_alloc_port(ps, id_priv, rover);
2915                 /*
2916                  * Remember previously used port number in order to avoid
2917                  * re-using same port immediately after it is closed.
2918                  */
2919                 if (!ret)
2920                         last_used_port = rover;
2921                 if (ret != -EADDRNOTAVAIL)
2922                         return ret;
2923         }
2924         if (--remaining) {
2925                 rover++;
2926                 if ((rover < low) || (rover > high))
2927                         rover = low;
2928                 goto retry;
2929         }
2930         return -EADDRNOTAVAIL;
2931 }
2932
2933 /*
2934  * Check that the requested port is available.  This is called when trying to
2935  * bind to a specific port, or when trying to listen on a bound port.  In
2936  * the latter case, the provided id_priv may already be on the bind_list, but
2937  * we still need to check that it's okay to start listening.
2938  */
2939 static int cma_check_port(struct rdma_bind_list *bind_list,
2940                           struct rdma_id_private *id_priv, uint8_t reuseaddr)
2941 {
2942         struct rdma_id_private *cur_id;
2943         struct sockaddr *addr, *cur_addr;
2944
2945         addr = cma_src_addr(id_priv);
2946         hlist_for_each_entry(cur_id, &bind_list->owners, node) {
2947                 if (id_priv == cur_id)
2948                         continue;
2949
2950                 if ((cur_id->state != RDMA_CM_LISTEN) && reuseaddr &&
2951                     cur_id->reuseaddr)
2952                         continue;
2953
2954                 cur_addr = cma_src_addr(cur_id);
2955                 if (id_priv->afonly && cur_id->afonly &&
2956                     (addr->sa_family != cur_addr->sa_family))
2957                         continue;
2958
2959                 if (cma_any_addr(addr) || cma_any_addr(cur_addr))
2960                         return -EADDRNOTAVAIL;
2961
2962                 if (!cma_addr_cmp(addr, cur_addr))
2963                         return -EADDRINUSE;
2964         }
2965         return 0;
2966 }
2967
2968 static int cma_use_port(enum rdma_port_space ps,
2969                         struct rdma_id_private *id_priv)
2970 {
2971         struct rdma_bind_list *bind_list;
2972         unsigned short snum;
2973         int ret;
2974
2975         snum = ntohs(cma_port(cma_src_addr(id_priv)));
2976         if (snum < PROT_SOCK && !capable(CAP_NET_BIND_SERVICE))
2977                 return -EACCES;
2978
2979         bind_list = cma_ps_find(id_priv->id.route.addr.dev_addr.net, ps, snum);
2980         if (!bind_list) {
2981                 ret = cma_alloc_port(ps, id_priv, snum);
2982         } else {
2983                 ret = cma_check_port(bind_list, id_priv, id_priv->reuseaddr);
2984                 if (!ret)
2985                         cma_bind_port(bind_list, id_priv);
2986         }
2987         return ret;
2988 }
2989
2990 static int cma_bind_listen(struct rdma_id_private *id_priv)
2991 {
2992         struct rdma_bind_list *bind_list = id_priv->bind_list;
2993         int ret = 0;
2994
2995         mutex_lock(&lock);
2996         if (bind_list->owners.first->next)
2997                 ret = cma_check_port(bind_list, id_priv, 0);
2998         mutex_unlock(&lock);
2999         return ret;
3000 }
3001
3002 static enum rdma_port_space cma_select_inet_ps(
3003                 struct rdma_id_private *id_priv)
3004 {
3005         switch (id_priv->id.ps) {
3006         case RDMA_PS_TCP:
3007         case RDMA_PS_UDP:
3008         case RDMA_PS_IPOIB:
3009         case RDMA_PS_IB:
3010                 return id_priv->id.ps;
3011         default:
3012
3013                 return 0;
3014         }
3015 }
3016
3017 static enum rdma_port_space cma_select_ib_ps(struct rdma_id_private *id_priv)
3018 {
3019         enum rdma_port_space ps = 0;
3020         struct sockaddr_ib *sib;
3021         u64 sid_ps, mask, sid;
3022
3023         sib = (struct sockaddr_ib *) cma_src_addr(id_priv);
3024         mask = be64_to_cpu(sib->sib_sid_mask) & RDMA_IB_IP_PS_MASK;
3025         sid = be64_to_cpu(sib->sib_sid) & mask;
3026
3027         if ((id_priv->id.ps == RDMA_PS_IB) && (sid == (RDMA_IB_IP_PS_IB & mask))) {
3028                 sid_ps = RDMA_IB_IP_PS_IB;
3029                 ps = RDMA_PS_IB;
3030         } else if (((id_priv->id.ps == RDMA_PS_IB) || (id_priv->id.ps == RDMA_PS_TCP)) &&
3031                    (sid == (RDMA_IB_IP_PS_TCP & mask))) {
3032                 sid_ps = RDMA_IB_IP_PS_TCP;
3033                 ps = RDMA_PS_TCP;
3034         } else if (((id_priv->id.ps == RDMA_PS_IB) || (id_priv->id.ps == RDMA_PS_UDP)) &&
3035                    (sid == (RDMA_IB_IP_PS_UDP & mask))) {
3036                 sid_ps = RDMA_IB_IP_PS_UDP;
3037                 ps = RDMA_PS_UDP;
3038         }
3039
3040         if (ps) {
3041                 sib->sib_sid = cpu_to_be64(sid_ps | ntohs(cma_port((struct sockaddr *) sib)));
3042                 sib->sib_sid_mask = cpu_to_be64(RDMA_IB_IP_PS_MASK |
3043                                                 be64_to_cpu(sib->sib_sid_mask));
3044         }
3045         return ps;
3046 }
3047
3048 static int cma_get_port(struct rdma_id_private *id_priv)
3049 {
3050         enum rdma_port_space ps;
3051         int ret;
3052
3053         if (cma_family(id_priv) != AF_IB)
3054                 ps = cma_select_inet_ps(id_priv);
3055         else
3056                 ps = cma_select_ib_ps(id_priv);
3057         if (!ps)
3058                 return -EPROTONOSUPPORT;
3059
3060         mutex_lock(&lock);
3061         if (cma_any_port(cma_src_addr(id_priv)))
3062                 ret = cma_alloc_any_port(ps, id_priv);
3063         else
3064                 ret = cma_use_port(ps, id_priv);
3065         mutex_unlock(&lock);
3066
3067         return ret;
3068 }
3069
3070 static int cma_check_linklocal(struct rdma_dev_addr *dev_addr,
3071                                struct sockaddr *addr)
3072 {
3073 #if IS_ENABLED(CONFIG_IPV6)
3074         struct sockaddr_in6 *sin6;
3075
3076         if (addr->sa_family != AF_INET6)
3077                 return 0;
3078
3079         sin6 = (struct sockaddr_in6 *) addr;
3080
3081         if (!(ipv6_addr_type(&sin6->sin6_addr) & IPV6_ADDR_LINKLOCAL))
3082                 return 0;
3083
3084         if (!sin6->sin6_scope_id)
3085                         return -EINVAL;
3086
3087         dev_addr->bound_dev_if = sin6->sin6_scope_id;
3088 #endif
3089         return 0;
3090 }
3091
3092 int rdma_listen(struct rdma_cm_id *id, int backlog)
3093 {
3094         struct rdma_id_private *id_priv;
3095         int ret;
3096
3097         id_priv = container_of(id, struct rdma_id_private, id);
3098         if (id_priv->state == RDMA_CM_IDLE) {
3099                 id->route.addr.src_addr.ss_family = AF_INET;
3100                 ret = rdma_bind_addr(id, cma_src_addr(id_priv));
3101                 if (ret)
3102                         return ret;
3103         }
3104
3105         if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_BOUND, RDMA_CM_LISTEN))
3106                 return -EINVAL;
3107
3108         if (id_priv->reuseaddr) {
3109                 ret = cma_bind_listen(id_priv);
3110                 if (ret)
3111                         goto err;
3112         }
3113
3114         id_priv->backlog = backlog;
3115         if (id->device) {
3116                 if (rdma_cap_ib_cm(id->device, 1)) {
3117                         ret = cma_ib_listen(id_priv);
3118                         if (ret)
3119                                 goto err;
3120                 } else if (rdma_cap_iw_cm(id->device, 1)) {
3121                         ret = cma_iw_listen(id_priv, backlog);
3122                         if (ret)
3123                                 goto err;
3124                 } else {
3125                         ret = -ENOSYS;
3126                         goto err;
3127                 }
3128         } else
3129                 cma_listen_on_all(id_priv);
3130
3131         return 0;
3132 err:
3133         id_priv->backlog = 0;
3134         cma_comp_exch(id_priv, RDMA_CM_LISTEN, RDMA_CM_ADDR_BOUND);
3135         return ret;
3136 }
3137 EXPORT_SYMBOL(rdma_listen);
3138
3139 int rdma_bind_addr(struct rdma_cm_id *id, struct sockaddr *addr)
3140 {
3141         struct rdma_id_private *id_priv;
3142         int ret;
3143
3144         if (addr->sa_family != AF_INET && addr->sa_family != AF_INET6 &&
3145             addr->sa_family != AF_IB)
3146                 return -EAFNOSUPPORT;
3147
3148         id_priv = container_of(id, struct rdma_id_private, id);
3149         if (!cma_comp_exch(id_priv, RDMA_CM_IDLE, RDMA_CM_ADDR_BOUND))
3150                 return -EINVAL;
3151
3152         ret = cma_check_linklocal(&id->route.addr.dev_addr, addr);
3153         if (ret)
3154                 goto err1;
3155
3156         memcpy(cma_src_addr(id_priv), addr, rdma_addr_size(addr));
3157         if (!cma_any_addr(addr)) {
3158                 ret = cma_translate_addr(addr, &id->route.addr.dev_addr);
3159                 if (ret)
3160                         goto err1;
3161
3162                 ret = cma_acquire_dev(id_priv, NULL);
3163                 if (ret)
3164                         goto err1;
3165         }
3166
3167         if (!(id_priv->options & (1 << CMA_OPTION_AFONLY))) {
3168                 if (addr->sa_family == AF_INET)
3169                         id_priv->afonly = 1;
3170 #if IS_ENABLED(CONFIG_IPV6)
3171                 else if (addr->sa_family == AF_INET6) {
3172                         struct net *net = id_priv->id.route.addr.dev_addr.net;
3173
3174                         id_priv->afonly = net->ipv6.sysctl.bindv6only;
3175                 }
3176 #endif
3177         }
3178         ret = cma_get_port(id_priv);
3179         if (ret)
3180                 goto err2;
3181
3182         return 0;
3183 err2:
3184         if (id_priv->cma_dev)
3185                 cma_release_dev(id_priv);
3186 err1:
3187         cma_comp_exch(id_priv, RDMA_CM_ADDR_BOUND, RDMA_CM_IDLE);
3188         return ret;
3189 }
3190 EXPORT_SYMBOL(rdma_bind_addr);
3191
3192 static int cma_format_hdr(void *hdr, struct rdma_id_private *id_priv)
3193 {
3194         struct cma_hdr *cma_hdr;
3195
3196         cma_hdr = hdr;
3197         cma_hdr->cma_version = CMA_VERSION;
3198         if (cma_family(id_priv) == AF_INET) {
3199                 struct sockaddr_in *src4, *dst4;
3200
3201                 src4 = (struct sockaddr_in *) cma_src_addr(id_priv);
3202                 dst4 = (struct sockaddr_in *) cma_dst_addr(id_priv);
3203
3204                 cma_set_ip_ver(cma_hdr, 4);
3205                 cma_hdr->src_addr.ip4.addr = src4->sin_addr.s_addr;
3206                 cma_hdr->dst_addr.ip4.addr = dst4->sin_addr.s_addr;
3207                 cma_hdr->port = src4->sin_port;
3208         } else if (cma_family(id_priv) == AF_INET6) {
3209                 struct sockaddr_in6 *src6, *dst6;
3210
3211                 src6 = (struct sockaddr_in6 *) cma_src_addr(id_priv);
3212                 dst6 = (struct sockaddr_in6 *) cma_dst_addr(id_priv);
3213
3214                 cma_set_ip_ver(cma_hdr, 6);
3215                 cma_hdr->src_addr.ip6 = src6->sin6_addr;
3216                 cma_hdr->dst_addr.ip6 = dst6->sin6_addr;
3217                 cma_hdr->port = src6->sin6_port;
3218         }
3219         return 0;
3220 }
3221
3222 static int cma_sidr_rep_handler(struct ib_cm_id *cm_id,
3223                                 struct ib_cm_event *ib_event)
3224 {
3225         struct rdma_id_private *id_priv = cm_id->context;
3226         struct rdma_cm_event event;
3227         struct ib_cm_sidr_rep_event_param *rep = &ib_event->param.sidr_rep_rcvd;
3228         int ret = 0;
3229
3230         mutex_lock(&id_priv->handler_mutex);
3231         if (id_priv->state != RDMA_CM_CONNECT)
3232                 goto out;
3233
3234         memset(&event, 0, sizeof event);
3235         switch (ib_event->event) {
3236         case IB_CM_SIDR_REQ_ERROR:
3237                 event.event = RDMA_CM_EVENT_UNREACHABLE;
3238                 event.status = -ETIMEDOUT;
3239                 break;
3240         case IB_CM_SIDR_REP_RECEIVED:
3241                 event.param.ud.private_data = ib_event->private_data;
3242                 event.param.ud.private_data_len = IB_CM_SIDR_REP_PRIVATE_DATA_SIZE;
3243                 if (rep->status != IB_SIDR_SUCCESS) {
3244                         event.event = RDMA_CM_EVENT_UNREACHABLE;
3245                         event.status = ib_event->param.sidr_rep_rcvd.status;
3246                         break;
3247                 }
3248                 ret = cma_set_qkey(id_priv, rep->qkey);
3249                 if (ret) {
3250                         event.event = RDMA_CM_EVENT_ADDR_ERROR;
3251                         event.status = ret;
3252                         break;
3253                 }
3254                 ib_init_ah_from_path(id_priv->id.device, id_priv->id.port_num,
3255                                      id_priv->id.route.path_rec,
3256                                      &event.param.ud.ah_attr);
3257                 event.param.ud.qp_num = rep->qpn;
3258                 event.param.ud.qkey = rep->qkey;
3259                 event.event = RDMA_CM_EVENT_ESTABLISHED;
3260                 event.status = 0;
3261                 break;
3262         default:
3263                 pr_err("RDMA CMA: unexpected IB CM event: %d\n",
3264                        ib_event->event);
3265                 goto out;
3266         }
3267
3268         ret = id_priv->id.event_handler(&id_priv->id, &event);
3269         if (ret) {
3270                 /* Destroy the CM ID by returning a non-zero value. */
3271                 id_priv->cm_id.ib = NULL;
3272                 cma_exch(id_priv, RDMA_CM_DESTROYING);
3273                 mutex_unlock(&id_priv->handler_mutex);
3274                 rdma_destroy_id(&id_priv->id);
3275                 return ret;
3276         }
3277 out:
3278         mutex_unlock(&id_priv->handler_mutex);
3279         return ret;
3280 }
3281
3282 static int cma_resolve_ib_udp(struct rdma_id_private *id_priv,
3283                               struct rdma_conn_param *conn_param)
3284 {
3285         struct ib_cm_sidr_req_param req;
3286         struct ib_cm_id *id;
3287         void *private_data;
3288         int offset, ret;
3289
3290         memset(&req, 0, sizeof req);
3291         offset = cma_user_data_offset(id_priv);
3292         req.private_data_len = offset + conn_param->private_data_len;
3293         if (req.private_data_len < conn_param->private_data_len)
3294                 return -EINVAL;
3295
3296         if (req.private_data_len) {
3297                 private_data = kzalloc(req.private_data_len, GFP_ATOMIC);
3298                 if (!private_data)
3299                         return -ENOMEM;
3300         } else {
3301                 private_data = NULL;
3302         }
3303
3304         if (conn_param->private_data && conn_param->private_data_len)
3305                 memcpy(private_data + offset, conn_param->private_data,
3306                        conn_param->private_data_len);
3307
3308         if (private_data) {
3309                 ret = cma_format_hdr(private_data, id_priv);
3310                 if (ret)
3311                         goto out;
3312                 req.private_data = private_data;
3313         }
3314
3315         id = ib_create_cm_id(id_priv->id.device, cma_sidr_rep_handler,
3316                              id_priv);
3317         if (IS_ERR(id)) {
3318                 ret = PTR_ERR(id);
3319                 goto out;
3320         }
3321         id_priv->cm_id.ib = id;
3322
3323         req.path = id_priv->id.route.path_rec;
3324         req.service_id = rdma_get_service_id(&id_priv->id, cma_dst_addr(id_priv));
3325         req.timeout_ms = 1 << (CMA_CM_RESPONSE_TIMEOUT - 8);
3326         req.max_cm_retries = CMA_MAX_CM_RETRIES;
3327
3328         ret = ib_send_cm_sidr_req(id_priv->cm_id.ib, &req);
3329         if (ret) {
3330                 ib_destroy_cm_id(id_priv->cm_id.ib);
3331                 id_priv->cm_id.ib = NULL;
3332         }
3333 out:
3334         kfree(private_data);
3335         return ret;
3336 }
3337
3338 static int cma_connect_ib(struct rdma_id_private *id_priv,
3339                           struct rdma_conn_param *conn_param)
3340 {
3341         struct ib_cm_req_param req;
3342         struct rdma_route *route;
3343         void *private_data;
3344         struct ib_cm_id *id;
3345         int offset, ret;
3346
3347         memset(&req, 0, sizeof req);
3348         offset = cma_user_data_offset(id_priv);
3349         req.private_data_len = offset + conn_param->private_data_len;
3350         if (req.private_data_len < conn_param->private_data_len)
3351                 return -EINVAL;
3352
3353         if (req.private_data_len) {
3354                 private_data = kzalloc(req.private_data_len, GFP_ATOMIC);
3355                 if (!private_data)
3356                         return -ENOMEM;
3357         } else {
3358                 private_data = NULL;
3359         }
3360
3361         if (conn_param->private_data && conn_param->private_data_len)
3362                 memcpy(private_data + offset, conn_param->private_data,
3363                        conn_param->private_data_len);
3364
3365         id = ib_create_cm_id(id_priv->id.device, cma_ib_handler, id_priv);
3366         if (IS_ERR(id)) {
3367                 ret = PTR_ERR(id);
3368                 goto out;
3369         }
3370         id_priv->cm_id.ib = id;
3371
3372         route = &id_priv->id.route;
3373         if (private_data) {
3374                 ret = cma_format_hdr(private_data, id_priv);
3375                 if (ret)
3376                         goto out;
3377                 req.private_data = private_data;
3378         }
3379
3380         req.primary_path = &route->path_rec[0];
3381         if (route->num_paths == 2)
3382                 req.alternate_path = &route->path_rec[1];
3383
3384         req.service_id = rdma_get_service_id(&id_priv->id, cma_dst_addr(id_priv));
3385         req.qp_num = id_priv->qp_num;
3386         req.qp_type = id_priv->id.qp_type;
3387         req.starting_psn = id_priv->seq_num;
3388         req.responder_resources = conn_param->responder_resources;
3389         req.initiator_depth = conn_param->initiator_depth;
3390         req.flow_control = conn_param->flow_control;
3391         req.retry_count = min_t(u8, 7, conn_param->retry_count);
3392         req.rnr_retry_count = min_t(u8, 7, conn_param->rnr_retry_count);
3393         req.remote_cm_response_timeout = CMA_CM_RESPONSE_TIMEOUT;
3394         req.local_cm_response_timeout = CMA_CM_RESPONSE_TIMEOUT;
3395         req.max_cm_retries = CMA_MAX_CM_RETRIES;
3396         req.srq = id_priv->srq ? 1 : 0;
3397
3398         ret = ib_send_cm_req(id_priv->cm_id.ib, &req);
3399 out:
3400         if (ret && !IS_ERR(id)) {
3401                 ib_destroy_cm_id(id);
3402                 id_priv->cm_id.ib = NULL;
3403         }
3404
3405         kfree(private_data);
3406         return ret;
3407 }
3408
3409 static int cma_connect_iw(struct rdma_id_private *id_priv,
3410                           struct rdma_conn_param *conn_param)
3411 {
3412         struct iw_cm_id *cm_id;
3413         int ret;
3414         struct iw_cm_conn_param iw_param;
3415
3416         cm_id = iw_create_cm_id(id_priv->id.device, cma_iw_handler, id_priv);
3417         if (IS_ERR(cm_id))
3418                 return PTR_ERR(cm_id);
3419
3420         cm_id->tos = id_priv->tos;
3421         id_priv->cm_id.iw = cm_id;
3422
3423         memcpy(&cm_id->local_addr, cma_src_addr(id_priv),
3424                rdma_addr_size(cma_src_addr(id_priv)));
3425         memcpy(&cm_id->remote_addr, cma_dst_addr(id_priv),
3426                rdma_addr_size(cma_dst_addr(id_priv)));
3427
3428         ret = cma_modify_qp_rtr(id_priv, conn_param);
3429         if (ret)
3430                 goto out;
3431
3432         if (conn_param) {
3433                 iw_param.ord = conn_param->initiator_depth;
3434                 iw_param.ird = conn_param->responder_resources;
3435                 iw_param.private_data = conn_param->private_data;
3436                 iw_param.private_data_len = conn_param->private_data_len;
3437                 iw_param.qpn = id_priv->id.qp ? id_priv->qp_num : conn_param->qp_num;
3438         } else {
3439                 memset(&iw_param, 0, sizeof iw_param);
3440                 iw_param.qpn = id_priv->qp_num;
3441         }
3442         ret = iw_cm_connect(cm_id, &iw_param);
3443 out:
3444         if (ret) {
3445                 iw_destroy_cm_id(cm_id);
3446                 id_priv->cm_id.iw = NULL;
3447         }
3448         return ret;
3449 }
3450
3451 int rdma_connect(struct rdma_cm_id *id, struct rdma_conn_param *conn_param)
3452 {
3453         struct rdma_id_private *id_priv;
3454         int ret;
3455
3456         id_priv = container_of(id, struct rdma_id_private, id);
3457         if (!cma_comp_exch(id_priv, RDMA_CM_ROUTE_RESOLVED, RDMA_CM_CONNECT))
3458                 return -EINVAL;
3459
3460         if (!id->qp) {
3461                 id_priv->qp_num = conn_param->qp_num;
3462                 id_priv->srq = conn_param->srq;
3463         }
3464
3465         if (rdma_cap_ib_cm(id->device, id->port_num)) {
3466                 if (id->qp_type == IB_QPT_UD)
3467                         ret = cma_resolve_ib_udp(id_priv, conn_param);
3468                 else
3469                         ret = cma_connect_ib(id_priv, conn_param);
3470         } else if (rdma_cap_iw_cm(id->device, id->port_num))
3471                 ret = cma_connect_iw(id_priv, conn_param);
3472         else
3473                 ret = -ENOSYS;
3474         if (ret)
3475                 goto err;
3476
3477         return 0;
3478 err:
3479         cma_comp_exch(id_priv, RDMA_CM_CONNECT, RDMA_CM_ROUTE_RESOLVED);
3480         return ret;
3481 }
3482 EXPORT_SYMBOL(rdma_connect);
3483
3484 static int cma_accept_ib(struct rdma_id_private *id_priv,
3485                          struct rdma_conn_param *conn_param)
3486 {
3487         struct ib_cm_rep_param rep;
3488         int ret;
3489
3490         ret = cma_modify_qp_rtr(id_priv, conn_param);
3491         if (ret)
3492                 goto out;
3493
3494         ret = cma_modify_qp_rts(id_priv, conn_param);
3495         if (ret)
3496                 goto out;
3497
3498         memset(&rep, 0, sizeof rep);
3499         rep.qp_num = id_priv->qp_num;
3500         rep.starting_psn = id_priv->seq_num;
3501         rep.private_data = conn_param->private_data;
3502         rep.private_data_len = conn_param->private_data_len;
3503         rep.responder_resources = conn_param->responder_resources;
3504         rep.initiator_depth = conn_param->initiator_depth;
3505         rep.failover_accepted = 0;
3506         rep.flow_control = conn_param->flow_control;
3507         rep.rnr_retry_count = min_t(u8, 7, conn_param->rnr_retry_count);
3508         rep.srq = id_priv->srq ? 1 : 0;
3509
3510         ret = ib_send_cm_rep(id_priv->cm_id.ib, &rep);
3511 out:
3512         return ret;
3513 }
3514
3515 static int cma_accept_iw(struct rdma_id_private *id_priv,
3516                   struct rdma_conn_param *conn_param)
3517 {
3518         struct iw_cm_conn_param iw_param;
3519         int ret;
3520
3521         ret = cma_modify_qp_rtr(id_priv, conn_param);
3522         if (ret)
3523                 return ret;
3524
3525         iw_param.ord = conn_param->initiator_depth;
3526         iw_param.ird = conn_param->responder_resources;
3527         iw_param.private_data = conn_param->private_data;
3528         iw_param.private_data_len = conn_param->private_data_len;
3529         if (id_priv->id.qp) {
3530                 iw_param.qpn = id_priv->qp_num;
3531         } else
3532                 iw_param.qpn = conn_param->qp_num;
3533
3534         return iw_cm_accept(id_priv->cm_id.iw, &iw_param);
3535 }
3536
3537 static int cma_send_sidr_rep(struct rdma_id_private *id_priv,
3538                              enum ib_cm_sidr_status status, u32 qkey,
3539                              const void *private_data, int private_data_len)
3540 {
3541         struct ib_cm_sidr_rep_param rep;
3542         int ret;
3543
3544         memset(&rep, 0, sizeof rep);
3545         rep.status = status;
3546         if (status == IB_SIDR_SUCCESS) {
3547                 ret = cma_set_qkey(id_priv, qkey);
3548                 if (ret)
3549                         return ret;
3550                 rep.qp_num = id_priv->qp_num;
3551                 rep.qkey = id_priv->qkey;
3552         }
3553         rep.private_data = private_data;
3554         rep.private_data_len = private_data_len;
3555
3556         return ib_send_cm_sidr_rep(id_priv->cm_id.ib, &rep);
3557 }
3558
3559 int rdma_accept(struct rdma_cm_id *id, struct rdma_conn_param *conn_param)
3560 {
3561         struct rdma_id_private *id_priv;
3562         int ret;
3563
3564         id_priv = container_of(id, struct rdma_id_private, id);
3565
3566         id_priv->owner = task_pid_nr(current);
3567
3568         if (!cma_comp(id_priv, RDMA_CM_CONNECT))
3569                 return -EINVAL;
3570
3571         if (!id->qp && conn_param) {
3572                 id_priv->qp_num = conn_param->qp_num;
3573                 id_priv->srq = conn_param->srq;
3574         }
3575
3576         if (rdma_cap_ib_cm(id->device, id->port_num)) {
3577                 if (id->qp_type == IB_QPT_UD) {
3578                         if (conn_param)
3579                                 ret = cma_send_sidr_rep(id_priv, IB_SIDR_SUCCESS,
3580                                                         conn_param->qkey,
3581                                                         conn_param->private_data,
3582                                                         conn_param->private_data_len);
3583                         else
3584                                 ret = cma_send_sidr_rep(id_priv, IB_SIDR_SUCCESS,
3585                                                         0, NULL, 0);
3586                 } else {
3587                         if (conn_param)
3588                                 ret = cma_accept_ib(id_priv, conn_param);
3589                         else
3590                                 ret = cma_rep_recv(id_priv);
3591                 }
3592         } else if (rdma_cap_iw_cm(id->device, id->port_num))
3593                 ret = cma_accept_iw(id_priv, conn_param);
3594         else
3595                 ret = -ENOSYS;
3596
3597         if (ret)
3598                 goto reject;
3599
3600         return 0;
3601 reject:
3602         cma_modify_qp_err(id_priv);
3603         rdma_reject(id, NULL, 0);
3604         return ret;
3605 }
3606 EXPORT_SYMBOL(rdma_accept);
3607
3608 int rdma_notify(struct rdma_cm_id *id, enum ib_event_type event)
3609 {
3610         struct rdma_id_private *id_priv;
3611         int ret;
3612
3613         id_priv = container_of(id, struct rdma_id_private, id);
3614         if (!id_priv->cm_id.ib)
3615                 return -EINVAL;
3616
3617         switch (id->device->node_type) {
3618         case RDMA_NODE_IB_CA:
3619                 ret = ib_cm_notify(id_priv->cm_id.ib, event);
3620                 break;
3621         default:
3622                 ret = 0;
3623                 break;
3624         }
3625         return ret;
3626 }
3627 EXPORT_SYMBOL(rdma_notify);
3628
3629 int rdma_reject(struct rdma_cm_id *id, const void *private_data,
3630                 u8 private_data_len)
3631 {
3632         struct rdma_id_private *id_priv;
3633         int ret;
3634
3635         id_priv = container_of(id, struct rdma_id_private, id);
3636         if (!id_priv->cm_id.ib)
3637                 return -EINVAL;
3638
3639         if (rdma_cap_ib_cm(id->device, id->port_num)) {
3640                 if (id->qp_type == IB_QPT_UD)
3641                         ret = cma_send_sidr_rep(id_priv, IB_SIDR_REJECT, 0,
3642                                                 private_data, private_data_len);
3643                 else
3644                         ret = ib_send_cm_rej(id_priv->cm_id.ib,
3645                                              IB_CM_REJ_CONSUMER_DEFINED, NULL,
3646                                              0, private_data, private_data_len);
3647         } else if (rdma_cap_iw_cm(id->device, id->port_num)) {
3648                 ret = iw_cm_reject(id_priv->cm_id.iw,
3649                                    private_data, private_data_len);
3650         } else
3651                 ret = -ENOSYS;
3652
3653         return ret;
3654 }
3655 EXPORT_SYMBOL(rdma_reject);
3656
3657 int rdma_disconnect(struct rdma_cm_id *id)
3658 {
3659         struct rdma_id_private *id_priv;
3660         int ret;
3661
3662         id_priv = container_of(id, struct rdma_id_private, id);
3663         if (!id_priv->cm_id.ib)
3664                 return -EINVAL;
3665
3666         if (rdma_cap_ib_cm(id->device, id->port_num)) {
3667                 ret = cma_modify_qp_err(id_priv);
3668                 if (ret)
3669                         goto out;
3670                 /* Initiate or respond to a disconnect. */
3671                 if (ib_send_cm_dreq(id_priv->cm_id.ib, NULL, 0))
3672                         ib_send_cm_drep(id_priv->cm_id.ib, NULL, 0);
3673         } else if (rdma_cap_iw_cm(id->device, id->port_num)) {
3674                 ret = iw_cm_disconnect(id_priv->cm_id.iw, 0);
3675         } else
3676                 ret = -EINVAL;
3677
3678 out:
3679         return ret;
3680 }
3681 EXPORT_SYMBOL(rdma_disconnect);
3682
3683 static int cma_ib_mc_handler(int status, struct ib_sa_multicast *multicast)
3684 {
3685         struct rdma_id_private *id_priv;
3686         struct cma_multicast *mc = multicast->context;
3687         struct rdma_cm_event event;
3688         int ret = 0;
3689
3690         id_priv = mc->id_priv;
3691         mutex_lock(&id_priv->handler_mutex);
3692         if (id_priv->state != RDMA_CM_ADDR_BOUND &&
3693             id_priv->state != RDMA_CM_ADDR_RESOLVED)
3694                 goto out;
3695
3696         if (!status)
3697                 status = cma_set_qkey(id_priv, be32_to_cpu(multicast->rec.qkey));
3698         mutex_lock(&id_priv->qp_mutex);
3699         if (!status && id_priv->id.qp)
3700                 status = ib_attach_mcast(id_priv->id.qp, &multicast->rec.mgid,
3701                                          be16_to_cpu(multicast->rec.mlid));
3702         mutex_unlock(&id_priv->qp_mutex);
3703
3704         memset(&event, 0, sizeof event);
3705         event.status = status;
3706         event.param.ud.private_data = mc->context;
3707         if (!status) {
3708                 struct rdma_dev_addr *dev_addr =
3709                         &id_priv->id.route.addr.dev_addr;
3710                 struct net_device *ndev =
3711                         dev_get_by_index(&init_net, dev_addr->bound_dev_if);
3712                 enum ib_gid_type gid_type =
3713                         id_priv->cma_dev->default_gid_type[id_priv->id.port_num -
3714                         rdma_start_port(id_priv->cma_dev->device)];
3715
3716                 event.event = RDMA_CM_EVENT_MULTICAST_JOIN;
3717                 ib_init_ah_from_mcmember(id_priv->id.device,
3718                                          id_priv->id.port_num, &multicast->rec,
3719                                          ndev, gid_type,
3720                                          &event.param.ud.ah_attr);
3721                 event.param.ud.qp_num = 0xFFFFFF;
3722                 event.param.ud.qkey = be32_to_cpu(multicast->rec.qkey);
3723                 if (ndev)
3724                         dev_put(ndev);
3725         } else
3726                 event.event = RDMA_CM_EVENT_MULTICAST_ERROR;
3727
3728         ret = id_priv->id.event_handler(&id_priv->id, &event);
3729         if (ret) {
3730                 cma_exch(id_priv, RDMA_CM_DESTROYING);
3731                 mutex_unlock(&id_priv->handler_mutex);
3732                 rdma_destroy_id(&id_priv->id);
3733                 return 0;
3734         }
3735
3736 out:
3737         mutex_unlock(&id_priv->handler_mutex);
3738         return 0;
3739 }
3740
3741 static void cma_set_mgid(struct rdma_id_private *id_priv,
3742                          struct sockaddr *addr, union ib_gid *mgid)
3743 {
3744         unsigned char mc_map[MAX_ADDR_LEN];
3745         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
3746         struct sockaddr_in *sin = (struct sockaddr_in *) addr;
3747         struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *) addr;
3748
3749         if (cma_any_addr(addr)) {
3750                 memset(mgid, 0, sizeof *mgid);
3751         } else if ((addr->sa_family == AF_INET6) &&
3752                    ((be32_to_cpu(sin6->sin6_addr.s6_addr32[0]) & 0xFFF0FFFF) ==
3753                                                                  0xFF10A01B)) {
3754                 /* IPv6 address is an SA assigned MGID. */
3755                 memcpy(mgid, &sin6->sin6_addr, sizeof *mgid);
3756         } else if (addr->sa_family == AF_IB) {
3757                 memcpy(mgid, &((struct sockaddr_ib *) addr)->sib_addr, sizeof *mgid);
3758         } else if ((addr->sa_family == AF_INET6)) {
3759                 ipv6_ib_mc_map(&sin6->sin6_addr, dev_addr->broadcast, mc_map);
3760                 if (id_priv->id.ps == RDMA_PS_UDP)
3761                         mc_map[7] = 0x01;       /* Use RDMA CM signature */
3762                 *mgid = *(union ib_gid *) (mc_map + 4);
3763         } else {
3764                 ip_ib_mc_map(sin->sin_addr.s_addr, dev_addr->broadcast, mc_map);
3765                 if (id_priv->id.ps == RDMA_PS_UDP)
3766                         mc_map[7] = 0x01;       /* Use RDMA CM signature */
3767                 *mgid = *(union ib_gid *) (mc_map + 4);
3768         }
3769 }
3770
3771 static void cma_query_sa_classport_info_cb(int status,
3772                                            struct ib_class_port_info *rec,
3773                                            void *context)
3774 {
3775         struct class_port_info_context *cb_ctx = context;
3776
3777         WARN_ON(!context);
3778
3779         if (status || !rec) {
3780                 pr_debug("RDMA CM: %s port %u failed query ClassPortInfo status: %d\n",
3781                          cb_ctx->device->name, cb_ctx->port_num, status);
3782                 goto out;
3783         }
3784
3785         memcpy(cb_ctx->class_port_info, rec, sizeof(struct ib_class_port_info));
3786
3787 out:
3788         complete(&cb_ctx->done);
3789 }
3790
3791 static int cma_query_sa_classport_info(struct ib_device *device, u8 port_num,
3792                                        struct ib_class_port_info *class_port_info)
3793 {
3794         struct class_port_info_context *cb_ctx;
3795         int ret;
3796
3797         cb_ctx = kmalloc(sizeof(*cb_ctx), GFP_KERNEL);
3798         if (!cb_ctx)
3799                 return -ENOMEM;
3800
3801         cb_ctx->device = device;
3802         cb_ctx->class_port_info = class_port_info;
3803         cb_ctx->port_num = port_num;
3804         init_completion(&cb_ctx->done);
3805
3806         ret = ib_sa_classport_info_rec_query(&sa_client, device, port_num,
3807                                              CMA_QUERY_CLASSPORT_INFO_TIMEOUT,
3808                                              GFP_KERNEL, cma_query_sa_classport_info_cb,
3809                                              cb_ctx, &cb_ctx->sa_query);
3810         if (ret < 0) {
3811                 pr_err("RDMA CM: %s port %u failed to send ClassPortInfo query, ret: %d\n",
3812                        device->name, port_num, ret);
3813                 goto out;
3814         }
3815
3816         wait_for_completion(&cb_ctx->done);
3817
3818 out:
3819         kfree(cb_ctx);
3820         return ret;
3821 }
3822
3823 static int cma_join_ib_multicast(struct rdma_id_private *id_priv,
3824                                  struct cma_multicast *mc)
3825 {
3826         struct ib_sa_mcmember_rec rec;
3827         struct ib_class_port_info class_port_info;
3828         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
3829         ib_sa_comp_mask comp_mask;
3830         int ret;
3831
3832         ib_addr_get_mgid(dev_addr, &rec.mgid);
3833         ret = ib_sa_get_mcmember_rec(id_priv->id.device, id_priv->id.port_num,
3834                                      &rec.mgid, &rec);
3835         if (ret)
3836                 return ret;
3837
3838         ret = cma_set_qkey(id_priv, 0);
3839         if (ret)
3840                 return ret;
3841
3842         cma_set_mgid(id_priv, (struct sockaddr *) &mc->addr, &rec.mgid);
3843         rec.qkey = cpu_to_be32(id_priv->qkey);
3844         rdma_addr_get_sgid(dev_addr, &rec.port_gid);
3845         rec.pkey = cpu_to_be16(ib_addr_get_pkey(dev_addr));
3846         rec.join_state = mc->join_state;
3847
3848         if (rec.join_state == BIT(SENDONLY_FULLMEMBER_JOIN)) {
3849                 ret = cma_query_sa_classport_info(id_priv->id.device,
3850                                                   id_priv->id.port_num,
3851                                                   &class_port_info);
3852
3853                 if (ret)
3854                         return ret;
3855
3856                 if (!(ib_get_cpi_capmask2(&class_port_info) &
3857                       IB_SA_CAP_MASK2_SENDONLY_FULL_MEM_SUPPORT)) {
3858                         pr_warn("RDMA CM: %s port %u Unable to multicast join\n"
3859                                 "RDMA CM: SM doesn't support Send Only Full Member option\n",
3860                                 id_priv->id.device->name, id_priv->id.port_num);
3861                         return -EOPNOTSUPP;
3862                 }
3863         }
3864
3865         comp_mask = IB_SA_MCMEMBER_REC_MGID | IB_SA_MCMEMBER_REC_PORT_GID |
3866                     IB_SA_MCMEMBER_REC_PKEY | IB_SA_MCMEMBER_REC_JOIN_STATE |
3867                     IB_SA_MCMEMBER_REC_QKEY | IB_SA_MCMEMBER_REC_SL |
3868                     IB_SA_MCMEMBER_REC_FLOW_LABEL |
3869                     IB_SA_MCMEMBER_REC_TRAFFIC_CLASS;
3870
3871         if (id_priv->id.ps == RDMA_PS_IPOIB)
3872                 comp_mask |= IB_SA_MCMEMBER_REC_RATE |
3873                              IB_SA_MCMEMBER_REC_RATE_SELECTOR |
3874                              IB_SA_MCMEMBER_REC_MTU_SELECTOR |
3875                              IB_SA_MCMEMBER_REC_MTU |
3876                              IB_SA_MCMEMBER_REC_HOP_LIMIT;
3877
3878         mc->multicast.ib = ib_sa_join_multicast(&sa_client, id_priv->id.device,
3879                                                 id_priv->id.port_num, &rec,
3880                                                 comp_mask, GFP_KERNEL,
3881                                                 cma_ib_mc_handler, mc);
3882         return PTR_ERR_OR_ZERO(mc->multicast.ib);
3883 }
3884
3885 static void iboe_mcast_work_handler(struct work_struct *work)
3886 {
3887         struct iboe_mcast_work *mw = container_of(work, struct iboe_mcast_work, work);
3888         struct cma_multicast *mc = mw->mc;
3889         struct ib_sa_multicast *m = mc->multicast.ib;
3890
3891         mc->multicast.ib->context = mc;
3892         cma_ib_mc_handler(0, m);
3893         kref_put(&mc->mcref, release_mc);
3894         kfree(mw);
3895 }
3896
3897 static void cma_iboe_set_mgid(struct sockaddr *addr, union ib_gid *mgid)
3898 {
3899         struct sockaddr_in *sin = (struct sockaddr_in *)addr;
3900         struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)addr;
3901
3902         if (cma_any_addr(addr)) {
3903                 memset(mgid, 0, sizeof *mgid);
3904         } else if (addr->sa_family == AF_INET6) {
3905                 memcpy(mgid, &sin6->sin6_addr, sizeof *mgid);
3906         } else {
3907                 mgid->raw[0] = 0xff;
3908                 mgid->raw[1] = 0x0e;
3909                 mgid->raw[2] = 0;
3910                 mgid->raw[3] = 0;
3911                 mgid->raw[4] = 0;
3912                 mgid->raw[5] = 0;
3913                 mgid->raw[6] = 0;
3914                 mgid->raw[7] = 0;
3915                 mgid->raw[8] = 0;
3916                 mgid->raw[9] = 0;
3917                 mgid->raw[10] = 0xff;
3918                 mgid->raw[11] = 0xff;
3919                 *(__be32 *)(&mgid->raw[12]) = sin->sin_addr.s_addr;
3920         }
3921 }
3922
3923 static int cma_iboe_join_multicast(struct rdma_id_private *id_priv,
3924                                    struct cma_multicast *mc)
3925 {
3926         struct iboe_mcast_work *work;
3927         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
3928         int err = 0;
3929         struct sockaddr *addr = (struct sockaddr *)&mc->addr;
3930         struct net_device *ndev = NULL;
3931         enum ib_gid_type gid_type;
3932         bool send_only;
3933
3934         send_only = mc->join_state == BIT(SENDONLY_FULLMEMBER_JOIN);
3935
3936         if (cma_zero_addr((struct sockaddr *)&mc->addr))
3937                 return -EINVAL;
3938
3939         work = kzalloc(sizeof *work, GFP_KERNEL);
3940         if (!work)
3941                 return -ENOMEM;
3942
3943         mc->multicast.ib = kzalloc(sizeof(struct ib_sa_multicast), GFP_KERNEL);
3944         if (!mc->multicast.ib) {
3945                 err = -ENOMEM;
3946                 goto out1;
3947         }
3948
3949         cma_iboe_set_mgid(addr, &mc->multicast.ib->rec.mgid);
3950
3951         mc->multicast.ib->rec.pkey = cpu_to_be16(0xffff);
3952         if (id_priv->id.ps == RDMA_PS_UDP)
3953                 mc->multicast.ib->rec.qkey = cpu_to_be32(RDMA_UDP_QKEY);
3954
3955         if (dev_addr->bound_dev_if)
3956                 ndev = dev_get_by_index(&init_net, dev_addr->bound_dev_if);
3957         if (!ndev) {
3958                 err = -ENODEV;
3959                 goto out2;
3960         }
3961         mc->multicast.ib->rec.rate = iboe_get_rate(ndev);
3962         mc->multicast.ib->rec.hop_limit = 1;
3963         mc->multicast.ib->rec.mtu = iboe_get_mtu(ndev->mtu);
3964
3965         gid_type = id_priv->cma_dev->default_gid_type[id_priv->id.port_num -
3966                    rdma_start_port(id_priv->cma_dev->device)];
3967         if (addr->sa_family == AF_INET) {
3968                 if (gid_type == IB_GID_TYPE_ROCE_UDP_ENCAP) {
3969                         mc->multicast.ib->rec.hop_limit = IPV6_DEFAULT_HOPLIMIT;
3970                         if (!send_only) {
3971                                 err = cma_igmp_send(ndev, &mc->multicast.ib->rec.mgid,
3972                                                     true);
3973                                 if (!err)
3974                                         mc->igmp_joined = true;
3975                         }
3976                 }
3977         } else {
3978                 if (gid_type == IB_GID_TYPE_ROCE_UDP_ENCAP)
3979                         err = -ENOTSUPP;
3980         }
3981         dev_put(ndev);
3982         if (err || !mc->multicast.ib->rec.mtu) {
3983                 if (!err)
3984                         err = -EINVAL;
3985                 goto out2;
3986         }
3987         rdma_ip2gid((struct sockaddr *)&id_priv->id.route.addr.src_addr,
3988                     &mc->multicast.ib->rec.port_gid);
3989         work->id = id_priv;
3990         work->mc = mc;
3991         INIT_WORK(&work->work, iboe_mcast_work_handler);
3992         kref_get(&mc->mcref);
3993         queue_work(cma_wq, &work->work);
3994
3995         return 0;
3996
3997 out2:
3998         kfree(mc->multicast.ib);
3999 out1:
4000         kfree(work);
4001         return err;
4002 }
4003
4004 int rdma_join_multicast(struct rdma_cm_id *id, struct sockaddr *addr,
4005                         u8 join_state, void *context)
4006 {
4007         struct rdma_id_private *id_priv;
4008         struct cma_multicast *mc;
4009         int ret;
4010
4011         id_priv = container_of(id, struct rdma_id_private, id);
4012         if (!cma_comp(id_priv, RDMA_CM_ADDR_BOUND) &&
4013             !cma_comp(id_priv, RDMA_CM_ADDR_RESOLVED))
4014                 return -EINVAL;
4015
4016         mc = kmalloc(sizeof *mc, GFP_KERNEL);
4017         if (!mc)
4018                 return -ENOMEM;
4019
4020         memcpy(&mc->addr, addr, rdma_addr_size(addr));
4021         mc->context = context;
4022         mc->id_priv = id_priv;
4023         mc->igmp_joined = false;
4024         mc->join_state = join_state;
4025         spin_lock(&id_priv->lock);
4026         list_add(&mc->list, &id_priv->mc_list);
4027         spin_unlock(&id_priv->lock);
4028
4029         if (rdma_protocol_roce(id->device, id->port_num)) {
4030                 kref_init(&mc->mcref);
4031                 ret = cma_iboe_join_multicast(id_priv, mc);
4032         } else if (rdma_cap_ib_mcast(id->device, id->port_num))
4033                 ret = cma_join_ib_multicast(id_priv, mc);
4034         else
4035                 ret = -ENOSYS;
4036
4037         if (ret) {
4038                 spin_lock_irq(&id_priv->lock);
4039                 list_del(&mc->list);
4040                 spin_unlock_irq(&id_priv->lock);
4041                 kfree(mc);
4042         }
4043         return ret;
4044 }
4045 EXPORT_SYMBOL(rdma_join_multicast);
4046
4047 void rdma_leave_multicast(struct rdma_cm_id *id, struct sockaddr *addr)
4048 {
4049         struct rdma_id_private *id_priv;
4050         struct cma_multicast *mc;
4051
4052         id_priv = container_of(id, struct rdma_id_private, id);
4053         spin_lock_irq(&id_priv->lock);
4054         list_for_each_entry(mc, &id_priv->mc_list, list) {
4055                 if (!memcmp(&mc->addr, addr, rdma_addr_size(addr))) {
4056                         list_del(&mc->list);
4057                         spin_unlock_irq(&id_priv->lock);
4058
4059                         if (id->qp)
4060                                 ib_detach_mcast(id->qp,
4061                                                 &mc->multicast.ib->rec.mgid,
4062                                                 be16_to_cpu(mc->multicast.ib->rec.mlid));
4063
4064                         BUG_ON(id_priv->cma_dev->device != id->device);
4065
4066                         if (rdma_cap_ib_mcast(id->device, id->port_num)) {
4067                                 ib_sa_free_multicast(mc->multicast.ib);
4068                                 kfree(mc);
4069                         } else if (rdma_protocol_roce(id->device, id->port_num)) {
4070                                 if (mc->igmp_joined) {
4071                                         struct rdma_dev_addr *dev_addr =
4072                                                 &id->route.addr.dev_addr;
4073                                         struct net_device *ndev = NULL;
4074
4075                                         if (dev_addr->bound_dev_if)
4076                                                 ndev = dev_get_by_index(&init_net,
4077                                                                         dev_addr->bound_dev_if);
4078                                         if (ndev) {
4079                                                 cma_igmp_send(ndev,
4080                                                               &mc->multicast.ib->rec.mgid,
4081                                                               false);
4082                                                 dev_put(ndev);
4083                                         }
4084                                         mc->igmp_joined = false;
4085                                 }
4086                                 kref_put(&mc->mcref, release_mc);
4087                         }
4088                         return;
4089                 }
4090         }
4091         spin_unlock_irq(&id_priv->lock);
4092 }
4093 EXPORT_SYMBOL(rdma_leave_multicast);
4094
4095 static int cma_netdev_change(struct net_device *ndev, struct rdma_id_private *id_priv)
4096 {
4097         struct rdma_dev_addr *dev_addr;
4098         struct cma_ndev_work *work;
4099
4100         dev_addr = &id_priv->id.route.addr.dev_addr;
4101
4102         if ((dev_addr->bound_dev_if == ndev->ifindex) &&
4103             (net_eq(dev_net(ndev), dev_addr->net)) &&
4104             memcmp(dev_addr->src_dev_addr, ndev->dev_addr, ndev->addr_len)) {
4105                 pr_info("RDMA CM addr change for ndev %s used by id %p\n",
4106                         ndev->name, &id_priv->id);
4107                 work = kzalloc(sizeof *work, GFP_KERNEL);
4108                 if (!work)
4109                         return -ENOMEM;
4110
4111                 INIT_WORK(&work->work, cma_ndev_work_handler);
4112                 work->id = id_priv;
4113                 work->event.event = RDMA_CM_EVENT_ADDR_CHANGE;
4114                 atomic_inc(&id_priv->refcount);
4115                 queue_work(cma_wq, &work->work);
4116         }
4117
4118         return 0;
4119 }
4120
4121 static int cma_netdev_callback(struct notifier_block *self, unsigned long event,
4122                                void *ptr)
4123 {
4124         struct net_device *ndev = netdev_notifier_info_to_dev(ptr);
4125         struct cma_device *cma_dev;
4126         struct rdma_id_private *id_priv;
4127         int ret = NOTIFY_DONE;
4128
4129         if (event != NETDEV_BONDING_FAILOVER)
4130                 return NOTIFY_DONE;
4131
4132         if (!(ndev->flags & IFF_MASTER) || !(ndev->priv_flags & IFF_BONDING))
4133                 return NOTIFY_DONE;
4134
4135         mutex_lock(&lock);
4136         list_for_each_entry(cma_dev, &dev_list, list)
4137                 list_for_each_entry(id_priv, &cma_dev->id_list, list) {
4138                         ret = cma_netdev_change(ndev, id_priv);
4139                         if (ret)
4140                                 goto out;
4141                 }
4142
4143 out:
4144         mutex_unlock(&lock);
4145         return ret;
4146 }
4147
4148 static struct notifier_block cma_nb = {
4149         .notifier_call = cma_netdev_callback
4150 };
4151
4152 static void cma_add_one(struct ib_device *device)
4153 {
4154         struct cma_device *cma_dev;
4155         struct rdma_id_private *id_priv;
4156         unsigned int i;
4157         unsigned long supported_gids = 0;
4158
4159         cma_dev = kmalloc(sizeof *cma_dev, GFP_KERNEL);
4160         if (!cma_dev)
4161                 return;
4162
4163         cma_dev->device = device;
4164         cma_dev->default_gid_type = kcalloc(device->phys_port_cnt,
4165                                             sizeof(*cma_dev->default_gid_type),
4166                                             GFP_KERNEL);
4167         if (!cma_dev->default_gid_type) {
4168                 kfree(cma_dev);
4169                 return;
4170         }
4171         for (i = rdma_start_port(device); i <= rdma_end_port(device); i++) {
4172                 supported_gids = roce_gid_type_mask_support(device, i);
4173                 WARN_ON(!supported_gids);
4174                 cma_dev->default_gid_type[i - rdma_start_port(device)] =
4175                         find_first_bit(&supported_gids, BITS_PER_LONG);
4176         }
4177
4178         init_completion(&cma_dev->comp);
4179         atomic_set(&cma_dev->refcount, 1);
4180         INIT_LIST_HEAD(&cma_dev->id_list);
4181         ib_set_client_data(device, &cma_client, cma_dev);
4182
4183         mutex_lock(&lock);
4184         list_add_tail(&cma_dev->list, &dev_list);
4185         list_for_each_entry(id_priv, &listen_any_list, list)
4186                 cma_listen_on_dev(id_priv, cma_dev);
4187         mutex_unlock(&lock);
4188 }
4189
4190 static int cma_remove_id_dev(struct rdma_id_private *id_priv)
4191 {
4192         struct rdma_cm_event event;
4193         enum rdma_cm_state state;
4194         int ret = 0;
4195
4196         /* Record that we want to remove the device */
4197         state = cma_exch(id_priv, RDMA_CM_DEVICE_REMOVAL);
4198         if (state == RDMA_CM_DESTROYING)
4199                 return 0;
4200
4201         cma_cancel_operation(id_priv, state);
4202         mutex_lock(&id_priv->handler_mutex);
4203
4204         /* Check for destruction from another callback. */
4205         if (!cma_comp(id_priv, RDMA_CM_DEVICE_REMOVAL))
4206                 goto out;
4207
4208         memset(&event, 0, sizeof event);
4209         event.event = RDMA_CM_EVENT_DEVICE_REMOVAL;
4210         ret = id_priv->id.event_handler(&id_priv->id, &event);
4211 out:
4212         mutex_unlock(&id_priv->handler_mutex);
4213         return ret;
4214 }
4215
4216 static void cma_process_remove(struct cma_device *cma_dev)
4217 {
4218         struct rdma_id_private *id_priv;
4219         int ret;
4220
4221         mutex_lock(&lock);
4222         while (!list_empty(&cma_dev->id_list)) {
4223                 id_priv = list_entry(cma_dev->id_list.next,
4224                                      struct rdma_id_private, list);
4225
4226                 list_del(&id_priv->listen_list);
4227                 list_del_init(&id_priv->list);
4228                 atomic_inc(&id_priv->refcount);
4229                 mutex_unlock(&lock);
4230
4231                 ret = id_priv->internal_id ? 1 : cma_remove_id_dev(id_priv);
4232                 cma_deref_id(id_priv);
4233                 if (ret)
4234                         rdma_destroy_id(&id_priv->id);
4235
4236                 mutex_lock(&lock);
4237         }
4238         mutex_unlock(&lock);
4239
4240         cma_deref_dev(cma_dev);
4241         wait_for_completion(&cma_dev->comp);
4242 }
4243
4244 static void cma_remove_one(struct ib_device *device, void *client_data)
4245 {
4246         struct cma_device *cma_dev = client_data;
4247
4248         if (!cma_dev)
4249                 return;
4250
4251         mutex_lock(&lock);
4252         list_del(&cma_dev->list);
4253         mutex_unlock(&lock);
4254
4255         cma_process_remove(cma_dev);
4256         kfree(cma_dev->default_gid_type);
4257         kfree(cma_dev);
4258 }
4259
4260 static int cma_get_id_stats(struct sk_buff *skb, struct netlink_callback *cb)
4261 {
4262         struct nlmsghdr *nlh;
4263         struct rdma_cm_id_stats *id_stats;
4264         struct rdma_id_private *id_priv;
4265         struct rdma_cm_id *id = NULL;
4266         struct cma_device *cma_dev;
4267         int i_dev = 0, i_id = 0;
4268
4269         /*
4270          * We export all of the IDs as a sequence of messages.  Each
4271          * ID gets its own netlink message.
4272          */
4273         mutex_lock(&lock);
4274
4275         list_for_each_entry(cma_dev, &dev_list, list) {
4276                 if (i_dev < cb->args[0]) {
4277                         i_dev++;
4278                         continue;
4279                 }
4280
4281                 i_id = 0;
4282                 list_for_each_entry(id_priv, &cma_dev->id_list, list) {
4283                         if (i_id < cb->args[1]) {
4284                                 i_id++;
4285                                 continue;
4286                         }
4287
4288                         id_stats = ibnl_put_msg(skb, &nlh, cb->nlh->nlmsg_seq,
4289                                                 sizeof *id_stats, RDMA_NL_RDMA_CM,
4290                                                 RDMA_NL_RDMA_CM_ID_STATS,
4291                                                 NLM_F_MULTI);
4292                         if (!id_stats)
4293                                 goto out;
4294
4295                         memset(id_stats, 0, sizeof *id_stats);
4296                         id = &id_priv->id;
4297                         id_stats->node_type = id->route.addr.dev_addr.dev_type;
4298                         id_stats->port_num = id->port_num;
4299                         id_stats->bound_dev_if =
4300                                 id->route.addr.dev_addr.bound_dev_if;
4301
4302                         if (ibnl_put_attr(skb, nlh,
4303                                           rdma_addr_size(cma_src_addr(id_priv)),
4304                                           cma_src_addr(id_priv),
4305                                           RDMA_NL_RDMA_CM_ATTR_SRC_ADDR))
4306                                 goto out;
4307                         if (ibnl_put_attr(skb, nlh,
4308                                           rdma_addr_size(cma_src_addr(id_priv)),
4309                                           cma_dst_addr(id_priv),
4310                                           RDMA_NL_RDMA_CM_ATTR_DST_ADDR))
4311                                 goto out;
4312
4313                         id_stats->pid           = id_priv->owner;
4314                         id_stats->port_space    = id->ps;
4315                         id_stats->cm_state      = id_priv->state;
4316                         id_stats->qp_num        = id_priv->qp_num;
4317                         id_stats->qp_type       = id->qp_type;
4318
4319                         i_id++;
4320                 }
4321
4322                 cb->args[1] = 0;
4323                 i_dev++;
4324         }
4325
4326 out:
4327         mutex_unlock(&lock);
4328         cb->args[0] = i_dev;
4329         cb->args[1] = i_id;
4330
4331         return skb->len;
4332 }
4333
4334 static const struct ibnl_client_cbs cma_cb_table[] = {
4335         [RDMA_NL_RDMA_CM_ID_STATS] = { .dump = cma_get_id_stats,
4336                                        .module = THIS_MODULE },
4337 };
4338
4339 static int cma_init_net(struct net *net)
4340 {
4341         struct cma_pernet *pernet = cma_pernet(net);
4342
4343         idr_init(&pernet->tcp_ps);
4344         idr_init(&pernet->udp_ps);
4345         idr_init(&pernet->ipoib_ps);
4346         idr_init(&pernet->ib_ps);
4347
4348         return 0;
4349 }
4350
4351 static void cma_exit_net(struct net *net)
4352 {
4353         struct cma_pernet *pernet = cma_pernet(net);
4354
4355         idr_destroy(&pernet->tcp_ps);
4356         idr_destroy(&pernet->udp_ps);
4357         idr_destroy(&pernet->ipoib_ps);
4358         idr_destroy(&pernet->ib_ps);
4359 }
4360
4361 static struct pernet_operations cma_pernet_operations = {
4362         .init = cma_init_net,
4363         .exit = cma_exit_net,
4364         .id = &cma_pernet_id,
4365         .size = sizeof(struct cma_pernet),
4366 };
4367
4368 static int __init cma_init(void)
4369 {
4370         int ret;
4371
4372         cma_wq = create_singlethread_workqueue("rdma_cm");
4373         if (!cma_wq)
4374                 return -ENOMEM;
4375
4376         ret = register_pernet_subsys(&cma_pernet_operations);
4377         if (ret)
4378                 goto err_wq;
4379
4380         ib_sa_register_client(&sa_client);
4381         rdma_addr_register_client(&addr_client);
4382         register_netdevice_notifier(&cma_nb);
4383
4384         ret = ib_register_client(&cma_client);
4385         if (ret)
4386                 goto err;
4387
4388         if (ibnl_add_client(RDMA_NL_RDMA_CM, ARRAY_SIZE(cma_cb_table),
4389                             cma_cb_table))
4390                 pr_warn("RDMA CMA: failed to add netlink callback\n");
4391         cma_configfs_init();
4392
4393         return 0;
4394
4395 err:
4396         unregister_netdevice_notifier(&cma_nb);
4397         rdma_addr_unregister_client(&addr_client);
4398         ib_sa_unregister_client(&sa_client);
4399 err_wq:
4400         destroy_workqueue(cma_wq);
4401         return ret;
4402 }
4403
4404 static void __exit cma_cleanup(void)
4405 {
4406         cma_configfs_exit();
4407         ibnl_remove_client(RDMA_NL_RDMA_CM);
4408         ib_unregister_client(&cma_client);
4409         unregister_netdevice_notifier(&cma_nb);
4410         rdma_addr_unregister_client(&addr_client);
4411         ib_sa_unregister_client(&sa_client);
4412         unregister_pernet_subsys(&cma_pernet_operations);
4413         destroy_workqueue(cma_wq);
4414 }
4415
4416 module_init(cma_init);
4417 module_exit(cma_cleanup);