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 const char *__attribute_const__ rdma_reject_msg(struct rdma_cm_id *id,
 105                                                 int reason)
 106 {
 107         if (rdma_ib_or_roce(id->device, id->port_num))
 108                 return ibcm_reject_msg(reason);
 109
 110         if (rdma_protocol_iwarp(id->device, id->port_num))
 111                 return iwcm_reject_msg(reason);
 112
 113         WARN_ON_ONCE(1);
 114         return "unrecognized transport";
 115 }
 116 EXPORT_SYMBOL(rdma_reject_msg);
 117
 118 bool rdma_is_consumer_reject(struct rdma_cm_id *id, int reason)
 119 {
 120         if (rdma_ib_or_roce(id->device, id->port_num))
 121                 return reason == IB_CM_REJ_CONSUMER_DEFINED;
 122
 123         if (rdma_protocol_iwarp(id->device, id->port_num))
 124                 return reason == -ECONNREFUSED;
 125
 126         WARN_ON_ONCE(1);
 127         return false;
 128 }
 129 EXPORT_SYMBOL(rdma_is_consumer_reject);
 130
 131 const void *rdma_consumer_reject_data(struct rdma_cm_id *id,
 132                                       struct rdma_cm_event *ev, u8 *data_len)
 133 {
 134         const void *p;
 135
 136         if (rdma_is_consumer_reject(id, ev->status)) {
 137                 *data_len = ev->param.conn.private_data_len;
 138                 p = ev->param.conn.private_data;
 139         } else {
 140                 *data_len = 0;
 141                 p = NULL;
 142         }
 143         return p;
 144 }
 145 EXPORT_SYMBOL(rdma_consumer_reject_data);
 146
 147 static void cma_add_one(struct ib_device *device);
 148 static void cma_remove_one(struct ib_device *device, void *client_data);
 149
 150 static struct ib_client cma_client = {
 151         .name   = "cma",
 152         .add    = cma_add_one,
 153         .remove = cma_remove_one
 154 };
 155
 156 static struct ib_sa_client sa_client;
 157 static struct rdma_addr_client addr_client;
 158 static LIST_HEAD(dev_list);
 159 static LIST_HEAD(listen_any_list);
 160 static DEFINE_MUTEX(lock);
 161 static struct workqueue_struct *cma_wq;
 162 static unsigned int cma_pernet_id;
 163
 164 struct cma_pernet {
 165         struct idr tcp_ps;
 166         struct idr udp_ps;
 167         struct idr ipoib_ps;
 168         struct idr ib_ps;
 169 };
 170
 171 static struct cma_pernet *cma_pernet(struct net *net)
 172 {
 173         return net_generic(net, cma_pernet_id);
 174 }
 175
 176 static struct idr *cma_pernet_idr(struct net *net, enum rdma_port_space ps)
 177 {
 178         struct cma_pernet *pernet = cma_pernet(net);
 179
 180         switch (ps) {
 181         case RDMA_PS_TCP:
 182                 return &pernet->tcp_ps;
 183         case RDMA_PS_UDP:
 184                 return &pernet->udp_ps;
 185         case RDMA_PS_IPOIB:
 186                 return &pernet->ipoib_ps;
 187         case RDMA_PS_IB:
 188                 return &pernet->ib_ps;
 189         default:
 190                 return NULL;
 191         }
 192 }
 193
 194 struct cma_device {
 195         struct list_head        list;
 196         struct ib_device        *device;
 197         struct completion       comp;
 198         atomic_t                refcount;
 199         struct list_head        id_list;
 200         enum ib_gid_type        *default_gid_type;
 201 };
 202
 203 struct rdma_bind_list {
 204         enum rdma_port_space    ps;
 205         struct hlist_head       owners;
 206         unsigned short          port;
 207 };
 208
 209 struct class_port_info_context {
 210         struct ib_class_port_info       *class_port_info;
 211         struct ib_device                *device;
 212         struct completion               done;
 213         struct ib_sa_query              *sa_query;
 214         u8                              port_num;
 215 };
 216
 217 static int cma_ps_alloc(struct net *net, enum rdma_port_space ps,
 218                         struct rdma_bind_list *bind_list, int snum)
 219 {
 220         struct idr *idr = cma_pernet_idr(net, ps);
 221
 222         return idr_alloc(idr, bind_list, snum, snum + 1, GFP_KERNEL);
 223 }
 224
 225 static struct rdma_bind_list *cma_ps_find(struct net *net,
 226                                           enum rdma_port_space ps, int snum)
 227 {
 228         struct idr *idr = cma_pernet_idr(net, ps);
 229
 230         return idr_find(idr, snum);
 231 }
 232
 233 static void cma_ps_remove(struct net *net, enum rdma_port_space ps, int snum)
 234 {
 235         struct idr *idr = cma_pernet_idr(net, ps);
 236
 237         idr_remove(idr, snum);
 238 }
 239
 240 enum {
 241         CMA_OPTION_AFONLY,
 242 };
 243
 244 void cma_ref_dev(struct cma_device *cma_dev)
 245 {
 246         atomic_inc(&cma_dev->refcount);
 247 }
 248
 249 struct cma_device *cma_enum_devices_by_ibdev(cma_device_filter  filter,
 250                                              void               *cookie)
 251 {
 252         struct cma_device *cma_dev;
 253         struct cma_device *found_cma_dev = NULL;
 254
 255         mutex_lock(&lock);
 256
 257         list_for_each_entry(cma_dev, &dev_list, list)
 258                 if (filter(cma_dev->device, cookie)) {
 259                         found_cma_dev = cma_dev;
 260                         break;
 261                 }
 262
 263         if (found_cma_dev)
 264                 cma_ref_dev(found_cma_dev);
 265         mutex_unlock(&lock);
 266         return found_cma_dev;
 267 }
 268
 269 int cma_get_default_gid_type(struct cma_device *cma_dev,
 270                              unsigned int port)
 271 {
 272         if (port < rdma_start_port(cma_dev->device) ||
 273             port > rdma_end_port(cma_dev->device))
 274                 return -EINVAL;
 275
 276         return cma_dev->default_gid_type[port - rdma_start_port(cma_dev->device)];
 277 }
 278
 279 int cma_set_default_gid_type(struct cma_device *cma_dev,
 280                              unsigned int port,
 281                              enum ib_gid_type default_gid_type)
 282 {
 283         unsigned long supported_gids;
 284
 285         if (port < rdma_start_port(cma_dev->device) ||
 286             port > rdma_end_port(cma_dev->device))
 287                 return -EINVAL;
 288
 289         supported_gids = roce_gid_type_mask_support(cma_dev->device, port);
 290
 291         if (!(supported_gids & 1 << default_gid_type))
 292                 return -EINVAL;
 293
 294         cma_dev->default_gid_type[port - rdma_start_port(cma_dev->device)] =
 295                 default_gid_type;
 296
 297         return 0;
 298 }
 299
 300 struct ib_device *cma_get_ib_dev(struct cma_device *cma_dev)
 301 {
 302         return cma_dev->device;
 303 }
 304
 305 /*
 306  * Device removal can occur at anytime, so we need extra handling to
 307  * serialize notifying the user of device removal with other callbacks.
 308  * We do this by disabling removal notification while a callback is in process,
 309  * and reporting it after the callback completes.
 310  */
 311 struct rdma_id_private {
 312         struct rdma_cm_id       id;
 313
 314         struct rdma_bind_list   *bind_list;
 315         struct hlist_node       node;
 316         struct list_head        list; /* listen_any_list or cma_device.list */
 317         struct list_head        listen_list; /* per device listens */
 318         struct cma_device       *cma_dev;
 319         struct list_head        mc_list;
 320
 321         int                     internal_id;
 322         enum rdma_cm_state      state;
 323         spinlock_t              lock;
 324         struct mutex            qp_mutex;
 325
 326         struct completion       comp;
 327         atomic_t                refcount;
 328         struct mutex            handler_mutex;
 329
 330         int                     backlog;
 331         int                     timeout_ms;
 332         struct ib_sa_query      *query;
 333         int                     query_id;
 334         union {
 335                 struct ib_cm_id *ib;
 336                 struct iw_cm_id *iw;
 337         } cm_id;
 338
 339         u32                     seq_num;
 340         u32                     qkey;
 341         u32                     qp_num;
 342         pid_t                   owner;
 343         u32                     options;
 344         u8                      srq;
 345         u8                      tos;
 346         u8                      reuseaddr;
 347         u8                      afonly;
 348         enum ib_gid_type        gid_type;
 349 };
 350
 351 struct cma_multicast {
 352         struct rdma_id_private *id_priv;
 353         union {
 354                 struct ib_sa_multicast *ib;
 355         } multicast;
 356         struct list_head        list;
 357         void                    *context;
 358         struct sockaddr_storage addr;
 359         struct kref             mcref;
 360         bool                    igmp_joined;
 361         u8                      join_state;
 362 };
 363
 364 struct cma_work {
 365         struct work_struct      work;
 366         struct rdma_id_private  *id;
 367         enum rdma_cm_state      old_state;
 368         enum rdma_cm_state      new_state;
 369         struct rdma_cm_event    event;
 370 };
 371
 372 struct cma_ndev_work {
 373         struct work_struct      work;
 374         struct rdma_id_private  *id;
 375         struct rdma_cm_event    event;
 376 };
 377
 378 struct iboe_mcast_work {
 379         struct work_struct       work;
 380         struct rdma_id_private  *id;
 381         struct cma_multicast    *mc;
 382 };
 383
 384 union cma_ip_addr {
 385         struct in6_addr ip6;
 386         struct {
 387                 __be32 pad[3];
 388                 __be32 addr;
 389         } ip4;
 390 };
 391
 392 struct cma_hdr {
 393         u8 cma_version;
 394         u8 ip_version;  /* IP version: 7:4 */
 395         __be16 port;
 396         union cma_ip_addr src_addr;
 397         union cma_ip_addr dst_addr;
 398 };
 399
 400 #define CMA_VERSION 0x00
 401
 402 struct cma_req_info {
 403         struct ib_device *device;
 404         int port;
 405         union ib_gid local_gid;
 406         __be64 service_id;
 407         u16 pkey;
 408         bool has_gid:1;
 409 };
 410
 411 static int cma_comp(struct rdma_id_private *id_priv, enum rdma_cm_state comp)
 412 {
 413         unsigned long flags;
 414         int ret;
 415
 416         spin_lock_irqsave(&id_priv->lock, flags);
 417         ret = (id_priv->state == comp);
 418         spin_unlock_irqrestore(&id_priv->lock, flags);
 419         return ret;
 420 }
 421
 422 static int cma_comp_exch(struct rdma_id_private *id_priv,
 423                          enum rdma_cm_state comp, enum rdma_cm_state exch)
 424 {
 425         unsigned long flags;
 426         int ret;
 427
 428         spin_lock_irqsave(&id_priv->lock, flags);
 429         if ((ret = (id_priv->state == comp)))
 430                 id_priv->state = exch;
 431         spin_unlock_irqrestore(&id_priv->lock, flags);
 432         return ret;
 433 }
 434
 435 static enum rdma_cm_state cma_exch(struct rdma_id_private *id_priv,
 436                                    enum rdma_cm_state exch)
 437 {
 438         unsigned long flags;
 439         enum rdma_cm_state old;
 440
 441         spin_lock_irqsave(&id_priv->lock, flags);
 442         old = id_priv->state;
 443         id_priv->state = exch;
 444         spin_unlock_irqrestore(&id_priv->lock, flags);
 445         return old;
 446 }
 447
 448 static inline u8 cma_get_ip_ver(const struct cma_hdr *hdr)
 449 {
 450         return hdr->ip_version >> 4;
 451 }
 452
 453 static inline void cma_set_ip_ver(struct cma_hdr *hdr, u8 ip_ver)
 454 {
 455         hdr->ip_version = (ip_ver << 4) | (hdr->ip_version & 0xF);
 456 }
 457
 458 static int cma_igmp_send(struct net_device *ndev, union ib_gid *mgid, bool join)
 459 {
 460         struct in_device *in_dev = NULL;
 461
 462         if (ndev) {
 463                 rtnl_lock();
 464                 in_dev = __in_dev_get_rtnl(ndev);
 465                 if (in_dev) {
 466                         if (join)
 467                                 ip_mc_inc_group(in_dev,
 468                                                 *(__be32 *)(mgid->raw + 12));
 469                         else
 470                                 ip_mc_dec_group(in_dev,
 471                                                 *(__be32 *)(mgid->raw + 12));
 472                 }
 473                 rtnl_unlock();
 474         }
 475         return (in_dev) ? 0 : -ENODEV;
 476 }
 477
 478 static void _cma_attach_to_dev(struct rdma_id_private *id_priv,
 479                                struct cma_device *cma_dev)
 480 {
 481         cma_ref_dev(cma_dev);
 482         id_priv->cma_dev = cma_dev;
 483         id_priv->gid_type = 0;
 484         id_priv->id.device = cma_dev->device;
 485         id_priv->id.route.addr.dev_addr.transport =
 486                 rdma_node_get_transport(cma_dev->device->node_type);
 487         list_add_tail(&id_priv->list, &cma_dev->id_list);
 488 }
 489
 490 static void cma_attach_to_dev(struct rdma_id_private *id_priv,
 491                               struct cma_device *cma_dev)
 492 {
 493         _cma_attach_to_dev(id_priv, cma_dev);
 494         id_priv->gid_type =
 495                 cma_dev->default_gid_type[id_priv->id.port_num -
 496                                           rdma_start_port(cma_dev->device)];
 497 }
 498
 499 void cma_deref_dev(struct cma_device *cma_dev)
 500 {
 501         if (atomic_dec_and_test(&cma_dev->refcount))
 502                 complete(&cma_dev->comp);
 503 }
 504
 505 static inline void release_mc(struct kref *kref)
 506 {
 507         struct cma_multicast *mc = container_of(kref, struct cma_multicast, mcref);
 508
 509         kfree(mc->multicast.ib);
 510         kfree(mc);
 511 }
 512
 513 static void cma_release_dev(struct rdma_id_private *id_priv)
 514 {
 515         mutex_lock(&lock);
 516         list_del(&id_priv->list);
 517         cma_deref_dev(id_priv->cma_dev);
 518         id_priv->cma_dev = NULL;
 519         mutex_unlock(&lock);
 520 }
 521
 522 static inline struct sockaddr *cma_src_addr(struct rdma_id_private *id_priv)
 523 {
 524         return (struct sockaddr *) &id_priv->id.route.addr.src_addr;
 525 }
 526
 527 static inline struct sockaddr *cma_dst_addr(struct rdma_id_private *id_priv)
 528 {
 529         return (struct sockaddr *) &id_priv->id.route.addr.dst_addr;
 530 }
 531
 532 static inline unsigned short cma_family(struct rdma_id_private *id_priv)
 533 {
 534         return id_priv->id.route.addr.src_addr.ss_family;
 535 }
 536
 537 static int cma_set_qkey(struct rdma_id_private *id_priv, u32 qkey)
 538 {
 539         struct ib_sa_mcmember_rec rec;
 540         int ret = 0;
 541
 542         if (id_priv->qkey) {
 543                 if (qkey && id_priv->qkey != qkey)
 544                         return -EINVAL;
 545                 return 0;
 546         }
 547
 548         if (qkey) {
 549                 id_priv->qkey = qkey;
 550                 return 0;
 551         }
 552
 553         switch (id_priv->id.ps) {
 554         case RDMA_PS_UDP:
 555         case RDMA_PS_IB:
 556                 id_priv->qkey = RDMA_UDP_QKEY;
 557                 break;
 558         case RDMA_PS_IPOIB:
 559                 ib_addr_get_mgid(&id_priv->id.route.addr.dev_addr, &rec.mgid);
 560                 ret = ib_sa_get_mcmember_rec(id_priv->id.device,
 561                                              id_priv->id.port_num, &rec.mgid,
 562                                              &rec);
 563                 if (!ret)
 564                         id_priv->qkey = be32_to_cpu(rec.qkey);
 565                 break;
 566         default:
 567                 break;
 568         }
 569         return ret;
 570 }
 571
 572 static void cma_translate_ib(struct sockaddr_ib *sib, struct rdma_dev_addr *dev_addr)
 573 {
 574         dev_addr->dev_type = ARPHRD_INFINIBAND;
 575         rdma_addr_set_sgid(dev_addr, (union ib_gid *) &sib->sib_addr);
 576         ib_addr_set_pkey(dev_addr, ntohs(sib->sib_pkey));
 577 }
 578
 579 static int cma_translate_addr(struct sockaddr *addr, struct rdma_dev_addr *dev_addr)
 580 {
 581         int ret;
 582
 583         if (addr->sa_family != AF_IB) {
 584                 ret = rdma_translate_ip(addr, dev_addr, NULL);
 585         } else {
 586                 cma_translate_ib((struct sockaddr_ib *) addr, dev_addr);
 587                 ret = 0;
 588         }
 589
 590         return ret;
 591 }
 592
 593 static inline int cma_validate_port(struct ib_device *device, u8 port,
 594                                     enum ib_gid_type gid_type,
 595                                       union ib_gid *gid, int dev_type,
 596                                       int bound_if_index)
 597 {
 598         int ret = -ENODEV;
 599         struct net_device *ndev = NULL;
 600
 601         if ((dev_type == ARPHRD_INFINIBAND) && !rdma_protocol_ib(device, port))
 602                 return ret;
 603
 604         if ((dev_type != ARPHRD_INFINIBAND) && rdma_protocol_ib(device, port))
 605                 return ret;
 606
 607         if (dev_type == ARPHRD_ETHER && rdma_protocol_roce(device, port)) {
 608                 ndev = dev_get_by_index(&init_net, bound_if_index);
 609                 if (ndev && ndev->flags & IFF_LOOPBACK) {
 610                         pr_info("detected loopback device\n");
 611                         dev_put(ndev);
 612
 613                         if (!device->get_netdev)
 614                                 return -EOPNOTSUPP;
 615
 616                         ndev = device->get_netdev(device, port);
 617                         if (!ndev)
 618                                 return -ENODEV;
 619                 }
 620         } else {
 621                 gid_type = IB_GID_TYPE_IB;
 622         }
 623
 624         ret = ib_find_cached_gid_by_port(device, gid, gid_type, port,
 625                                          ndev, NULL);
 626
 627         if (ndev)
 628                 dev_put(ndev);
 629
 630         return ret;
 631 }
 632
 633 static int cma_acquire_dev(struct rdma_id_private *id_priv,
 634                            struct rdma_id_private *listen_id_priv)
 635 {
 636         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
 637         struct cma_device *cma_dev;
 638         union ib_gid gid, iboe_gid, *gidp;
 639         int ret = -ENODEV;
 640         u8 port;
 641
 642         if (dev_addr->dev_type != ARPHRD_INFINIBAND &&
 643             id_priv->id.ps == RDMA_PS_IPOIB)
 644                 return -EINVAL;
 645
 646         mutex_lock(&lock);
 647         rdma_ip2gid((struct sockaddr *)&id_priv->id.route.addr.src_addr,
 648                     &iboe_gid);
 649
 650         memcpy(&gid, dev_addr->src_dev_addr +
 651                rdma_addr_gid_offset(dev_addr), sizeof gid);
 652
 653         if (listen_id_priv) {
 654                 cma_dev = listen_id_priv->cma_dev;
 655                 port = listen_id_priv->id.port_num;
 656                 gidp = rdma_protocol_roce(cma_dev->device, port) ?
 657                        &iboe_gid : &gid;
 658
 659                 ret = cma_validate_port(cma_dev->device, port,
 660                                         rdma_protocol_ib(cma_dev->device, port) ?
 661                                         IB_GID_TYPE_IB :
 662                                         listen_id_priv->gid_type, gidp,
 663                                         dev_addr->dev_type,
 664                                         dev_addr->bound_dev_if);
 665                 if (!ret) {
 666                         id_priv->id.port_num = port;
 667                         goto out;
 668                 }
 669         }
 670
 671         list_for_each_entry(cma_dev, &dev_list, list) {
 672                 for (port = 1; port <= cma_dev->device->phys_port_cnt; ++port) {
 673                         if (listen_id_priv &&
 674                             listen_id_priv->cma_dev == cma_dev &&
 675                             listen_id_priv->id.port_num == port)
 676                                 continue;
 677
 678                         gidp = rdma_protocol_roce(cma_dev->device, port) ?
 679                                &iboe_gid : &gid;
 680
 681                         ret = cma_validate_port(cma_dev->device, port,
 682                                                 rdma_protocol_ib(cma_dev->device, port) ?
 683                                                 IB_GID_TYPE_IB :
 684                                                 cma_dev->default_gid_type[port - 1],
 685                                                 gidp, dev_addr->dev_type,
 686                                                 dev_addr->bound_dev_if);
 687                         if (!ret) {
 688                                 id_priv->id.port_num = port;
 689                                 goto out;
 690                         }
 691                 }
 692         }
 693
 694 out:
 695         if (!ret)
 696                 cma_attach_to_dev(id_priv, cma_dev);
 697
 698         mutex_unlock(&lock);
 699         return ret;
 700 }
 701
 702 /*
 703  * Select the source IB device and address to reach the destination IB address.
 704  */
 705 static int cma_resolve_ib_dev(struct rdma_id_private *id_priv)
 706 {
 707         struct cma_device *cma_dev, *cur_dev;
 708         struct sockaddr_ib *addr;
 709         union ib_gid gid, sgid, *dgid;
 710         u16 pkey, index;
 711         u8 p;
 712         int i;
 713
 714         cma_dev = NULL;
 715         addr = (struct sockaddr_ib *) cma_dst_addr(id_priv);
 716         dgid = (union ib_gid *) &addr->sib_addr;
 717         pkey = ntohs(addr->sib_pkey);
 718
 719         list_for_each_entry(cur_dev, &dev_list, list) {
 720                 for (p = 1; p <= cur_dev->device->phys_port_cnt; ++p) {
 721                         if (!rdma_cap_af_ib(cur_dev->device, p))
 722                                 continue;
 723
 724                         if (ib_find_cached_pkey(cur_dev->device, p, pkey, &index))
 725                                 continue;
 726
 727                         for (i = 0; !ib_get_cached_gid(cur_dev->device, p, i,
 728                                                        &gid, NULL);
 729                              i++) {
 730                                 if (!memcmp(&gid, dgid, sizeof(gid))) {
 731                                         cma_dev = cur_dev;
 732                                         sgid = gid;
 733                                         id_priv->id.port_num = p;
 734                                         goto found;
 735                                 }
 736
 737                                 if (!cma_dev && (gid.global.subnet_prefix ==
 738                                                  dgid->global.subnet_prefix)) {
 739                                         cma_dev = cur_dev;
 740                                         sgid = gid;
 741                                         id_priv->id.port_num = p;
 742                                 }
 743                         }
 744                 }
 745         }
 746
 747         if (!cma_dev)
 748                 return -ENODEV;
 749
 750 found:
 751         cma_attach_to_dev(id_priv, cma_dev);
 752         addr = (struct sockaddr_ib *) cma_src_addr(id_priv);
 753         memcpy(&addr->sib_addr, &sgid, sizeof sgid);
 754         cma_translate_ib(addr, &id_priv->id.route.addr.dev_addr);
 755         return 0;
 756 }
 757
 758 static void cma_deref_id(struct rdma_id_private *id_priv)
 759 {
 760         if (atomic_dec_and_test(&id_priv->refcount))
 761                 complete(&id_priv->comp);
 762 }
 763
 764 struct rdma_cm_id *rdma_create_id(struct net *net,
 765                                   rdma_cm_event_handler event_handler,
 766                                   void *context, enum rdma_port_space ps,
 767                                   enum ib_qp_type qp_type)
 768 {
 769         struct rdma_id_private *id_priv;
 770
 771         id_priv = kzalloc(sizeof *id_priv, GFP_KERNEL);
 772         if (!id_priv)
 773                 return ERR_PTR(-ENOMEM);
 774
 775         id_priv->owner = task_pid_nr(current);
 776         id_priv->state = RDMA_CM_IDLE;
 777         id_priv->id.context = context;
 778         id_priv->id.event_handler = event_handler;
 779         id_priv->id.ps = ps;
 780         id_priv->id.qp_type = qp_type;
 781         spin_lock_init(&id_priv->lock);
 782         mutex_init(&id_priv->qp_mutex);
 783         init_completion(&id_priv->comp);
 784         atomic_set(&id_priv->refcount, 1);
 785         mutex_init(&id_priv->handler_mutex);
 786         INIT_LIST_HEAD(&id_priv->listen_list);
 787         INIT_LIST_HEAD(&id_priv->mc_list);
 788         get_random_bytes(&id_priv->seq_num, sizeof id_priv->seq_num);
 789         id_priv->id.route.addr.dev_addr.net = get_net(net);
 790
 791         return &id_priv->id;
 792 }
 793 EXPORT_SYMBOL(rdma_create_id);
 794
 795 static int cma_init_ud_qp(struct rdma_id_private *id_priv, struct ib_qp *qp)
 796 {
 797         struct ib_qp_attr qp_attr;
 798         int qp_attr_mask, ret;
 799
 800         qp_attr.qp_state = IB_QPS_INIT;
 801         ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
 802         if (ret)
 803                 return ret;
 804
 805         ret = ib_modify_qp(qp, &qp_attr, qp_attr_mask);
 806         if (ret)
 807                 return ret;
 808
 809         qp_attr.qp_state = IB_QPS_RTR;
 810         ret = ib_modify_qp(qp, &qp_attr, IB_QP_STATE);
 811         if (ret)
 812                 return ret;
 813
 814         qp_attr.qp_state = IB_QPS_RTS;
 815         qp_attr.sq_psn = 0;
 816         ret = ib_modify_qp(qp, &qp_attr, IB_QP_STATE | IB_QP_SQ_PSN);
 817
 818         return ret;
 819 }
 820
 821 static int cma_init_conn_qp(struct rdma_id_private *id_priv, struct ib_qp *qp)
 822 {
 823         struct ib_qp_attr qp_attr;
 824         int qp_attr_mask, ret;
 825
 826         qp_attr.qp_state = IB_QPS_INIT;
 827         ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
 828         if (ret)
 829                 return ret;
 830
 831         return ib_modify_qp(qp, &qp_attr, qp_attr_mask);
 832 }
 833
 834 int rdma_create_qp(struct rdma_cm_id *id, struct ib_pd *pd,
 835                    struct ib_qp_init_attr *qp_init_attr)
 836 {
 837         struct rdma_id_private *id_priv;
 838         struct ib_qp *qp;
 839         int ret;
 840
 841         id_priv = container_of(id, struct rdma_id_private, id);
 842         if (id->device != pd->device)
 843                 return -EINVAL;
 844
 845         qp_init_attr->port_num = id->port_num;
 846         qp = ib_create_qp(pd, qp_init_attr);
 847         if (IS_ERR(qp))
 848                 return PTR_ERR(qp);
 849
 850         if (id->qp_type == IB_QPT_UD)
 851                 ret = cma_init_ud_qp(id_priv, qp);
 852         else
 853                 ret = cma_init_conn_qp(id_priv, qp);
 854         if (ret)
 855                 goto err;
 856
 857         id->qp = qp;
 858         id_priv->qp_num = qp->qp_num;
 859         id_priv->srq = (qp->srq != NULL);
 860         return 0;
 861 err:
 862         ib_destroy_qp(qp);
 863         return ret;
 864 }
 865 EXPORT_SYMBOL(rdma_create_qp);
 866
 867 void rdma_destroy_qp(struct rdma_cm_id *id)
 868 {
 869         struct rdma_id_private *id_priv;
 870
 871         id_priv = container_of(id, struct rdma_id_private, id);
 872         mutex_lock(&id_priv->qp_mutex);
 873         ib_destroy_qp(id_priv->id.qp);
 874         id_priv->id.qp = NULL;
 875         mutex_unlock(&id_priv->qp_mutex);
 876 }
 877 EXPORT_SYMBOL(rdma_destroy_qp);
 878
 879 static int cma_modify_qp_rtr(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         union ib_gid sgid;
 885
 886         mutex_lock(&id_priv->qp_mutex);
 887         if (!id_priv->id.qp) {
 888                 ret = 0;
 889                 goto out;
 890         }
 891
 892         /* Need to update QP attributes from default values. */
 893         qp_attr.qp_state = IB_QPS_INIT;
 894         ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
 895         if (ret)
 896                 goto out;
 897
 898         ret = ib_modify_qp(id_priv->id.qp, &qp_attr, qp_attr_mask);
 899         if (ret)
 900                 goto out;
 901
 902         qp_attr.qp_state = IB_QPS_RTR;
 903         ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
 904         if (ret)
 905                 goto out;
 906
 907         ret = ib_query_gid(id_priv->id.device, id_priv->id.port_num,
 908                            qp_attr.ah_attr.grh.sgid_index, &sgid, NULL);
 909         if (ret)
 910                 goto out;
 911
 912         BUG_ON(id_priv->cma_dev->device != id_priv->id.device);
 913
 914         if (conn_param)
 915                 qp_attr.max_dest_rd_atomic = conn_param->responder_resources;
 916         ret = ib_modify_qp(id_priv->id.qp, &qp_attr, qp_attr_mask);
 917 out:
 918         mutex_unlock(&id_priv->qp_mutex);
 919         return ret;
 920 }
 921
 922 static int cma_modify_qp_rts(struct rdma_id_private *id_priv,
 923                              struct rdma_conn_param *conn_param)
 924 {
 925         struct ib_qp_attr qp_attr;
 926         int qp_attr_mask, ret;
 927
 928         mutex_lock(&id_priv->qp_mutex);
 929         if (!id_priv->id.qp) {
 930                 ret = 0;
 931                 goto out;
 932         }
 933
 934         qp_attr.qp_state = IB_QPS_RTS;
 935         ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
 936         if (ret)
 937                 goto out;
 938
 939         if (conn_param)
 940                 qp_attr.max_rd_atomic = conn_param->initiator_depth;
 941         ret = ib_modify_qp(id_priv->id.qp, &qp_attr, qp_attr_mask);
 942 out:
 943         mutex_unlock(&id_priv->qp_mutex);
 944         return ret;
 945 }
 946
 947 static int cma_modify_qp_err(struct rdma_id_private *id_priv)
 948 {
 949         struct ib_qp_attr qp_attr;
 950         int ret;
 951
 952         mutex_lock(&id_priv->qp_mutex);
 953         if (!id_priv->id.qp) {
 954                 ret = 0;
 955                 goto out;
 956         }
 957
 958         qp_attr.qp_state = IB_QPS_ERR;
 959         ret = ib_modify_qp(id_priv->id.qp, &qp_attr, IB_QP_STATE);
 960 out:
 961         mutex_unlock(&id_priv->qp_mutex);
 962         return ret;
 963 }
 964
 965 static int cma_ib_init_qp_attr(struct rdma_id_private *id_priv,
 966                                struct ib_qp_attr *qp_attr, int *qp_attr_mask)
 967 {
 968         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
 969         int ret;
 970         u16 pkey;
 971
 972         if (rdma_cap_eth_ah(id_priv->id.device, id_priv->id.port_num))
 973                 pkey = 0xffff;
 974         else
 975                 pkey = ib_addr_get_pkey(dev_addr);
 976
 977         ret = ib_find_cached_pkey(id_priv->id.device, id_priv->id.port_num,
 978                                   pkey, &qp_attr->pkey_index);
 979         if (ret)
 980                 return ret;
 981
 982         qp_attr->port_num = id_priv->id.port_num;
 983         *qp_attr_mask = IB_QP_STATE | IB_QP_PKEY_INDEX | IB_QP_PORT;
 984
 985         if (id_priv->id.qp_type == IB_QPT_UD) {
 986                 ret = cma_set_qkey(id_priv, 0);
 987                 if (ret)
 988                         return ret;
 989
 990                 qp_attr->qkey = id_priv->qkey;
 991                 *qp_attr_mask |= IB_QP_QKEY;
 992         } else {
 993                 qp_attr->qp_access_flags = 0;
 994                 *qp_attr_mask |= IB_QP_ACCESS_FLAGS;
 995         }
 996         return 0;
 997 }
 998
 999 int rdma_init_qp_attr(struct rdma_cm_id *id, struct ib_qp_attr *qp_attr,
1000                        int *qp_attr_mask)
1001 {
1002         struct rdma_id_private *id_priv;
1003         int ret = 0;
1004
1005         id_priv = container_of(id, struct rdma_id_private, id);
1006         if (rdma_cap_ib_cm(id->device, id->port_num)) {
1007                 if (!id_priv->cm_id.ib || (id_priv->id.qp_type == IB_QPT_UD))
1008                         ret = cma_ib_init_qp_attr(id_priv, qp_attr, qp_attr_mask);
1009                 else
1010                         ret = ib_cm_init_qp_attr(id_priv->cm_id.ib, qp_attr,
1011                                                  qp_attr_mask);
1012
1013                 if (qp_attr->qp_state == IB_QPS_RTR)
1014                         qp_attr->rq_psn = id_priv->seq_num;
1015         } else if (rdma_cap_iw_cm(id->device, id->port_num)) {
1016                 if (!id_priv->cm_id.iw) {
1017                         qp_attr->qp_access_flags = 0;
1018                         *qp_attr_mask = IB_QP_STATE | IB_QP_ACCESS_FLAGS;
1019                 } else
1020                         ret = iw_cm_init_qp_attr(id_priv->cm_id.iw, qp_attr,
1021                                                  qp_attr_mask);
1022         } else
1023                 ret = -ENOSYS;
1024
1025         return ret;
1026 }
1027 EXPORT_SYMBOL(rdma_init_qp_attr);
1028
1029 static inline int cma_zero_addr(struct sockaddr *addr)
1030 {
1031         switch (addr->sa_family) {
1032         case AF_INET:
1033                 return ipv4_is_zeronet(((struct sockaddr_in *)addr)->sin_addr.s_addr);
1034         case AF_INET6:
1035                 return ipv6_addr_any(&((struct sockaddr_in6 *) addr)->sin6_addr);
1036         case AF_IB:
1037                 return ib_addr_any(&((struct sockaddr_ib *) addr)->sib_addr);
1038         default:
1039                 return 0;
1040         }
1041 }
1042
1043 static inline int cma_loopback_addr(struct sockaddr *addr)
1044 {
1045         switch (addr->sa_family) {
1046         case AF_INET:
1047                 return ipv4_is_loopback(((struct sockaddr_in *) addr)->sin_addr.s_addr);
1048         case AF_INET6:
1049                 return ipv6_addr_loopback(&((struct sockaddr_in6 *) addr)->sin6_addr);
1050         case AF_IB:
1051                 return ib_addr_loopback(&((struct sockaddr_ib *) addr)->sib_addr);
1052         default:
1053                 return 0;
1054         }
1055 }
1056
1057 static inline int cma_any_addr(struct sockaddr *addr)
1058 {
1059         return cma_zero_addr(addr) || cma_loopback_addr(addr);
1060 }
1061
1062 static int cma_addr_cmp(struct sockaddr *src, struct sockaddr *dst)
1063 {
1064         if (src->sa_family != dst->sa_family)
1065                 return -1;
1066
1067         switch (src->sa_family) {
1068         case AF_INET:
1069                 return ((struct sockaddr_in *) src)->sin_addr.s_addr !=
1070                        ((struct sockaddr_in *) dst)->sin_addr.s_addr;
1071         case AF_INET6:
1072                 return ipv6_addr_cmp(&((struct sockaddr_in6 *) src)->sin6_addr,
1073                                      &((struct sockaddr_in6 *) dst)->sin6_addr);
1074         default:
1075                 return ib_addr_cmp(&((struct sockaddr_ib *) src)->sib_addr,
1076                                    &((struct sockaddr_ib *) dst)->sib_addr);
1077         }
1078 }
1079
1080 static __be16 cma_port(struct sockaddr *addr)
1081 {
1082         struct sockaddr_ib *sib;
1083
1084         switch (addr->sa_family) {
1085         case AF_INET:
1086                 return ((struct sockaddr_in *) addr)->sin_port;
1087         case AF_INET6:
1088                 return ((struct sockaddr_in6 *) addr)->sin6_port;
1089         case AF_IB:
1090                 sib = (struct sockaddr_ib *) addr;
1091                 return htons((u16) (be64_to_cpu(sib->sib_sid) &
1092                                     be64_to_cpu(sib->sib_sid_mask)));
1093         default:
1094                 return 0;
1095         }
1096 }
1097
1098 static inline int cma_any_port(struct sockaddr *addr)
1099 {
1100         return !cma_port(addr);
1101 }
1102
1103 static void cma_save_ib_info(struct sockaddr *src_addr,
1104                              struct sockaddr *dst_addr,
1105                              struct rdma_cm_id *listen_id,
1106                              struct ib_sa_path_rec *path)
1107 {
1108         struct sockaddr_ib *listen_ib, *ib;
1109
1110         listen_ib = (struct sockaddr_ib *) &listen_id->route.addr.src_addr;
1111         if (src_addr) {
1112                 ib = (struct sockaddr_ib *)src_addr;
1113                 ib->sib_family = AF_IB;
1114                 if (path) {
1115                         ib->sib_pkey = path->pkey;
1116                         ib->sib_flowinfo = path->flow_label;
1117                         memcpy(&ib->sib_addr, &path->sgid, 16);
1118                         ib->sib_sid = path->service_id;
1119                         ib->sib_scope_id = 0;
1120                 } else {
1121                         ib->sib_pkey = listen_ib->sib_pkey;
1122                         ib->sib_flowinfo = listen_ib->sib_flowinfo;
1123                         ib->sib_addr = listen_ib->sib_addr;
1124                         ib->sib_sid = listen_ib->sib_sid;
1125                         ib->sib_scope_id = listen_ib->sib_scope_id;
1126                 }
1127                 ib->sib_sid_mask = cpu_to_be64(0xffffffffffffffffULL);
1128         }
1129         if (dst_addr) {
1130                 ib = (struct sockaddr_ib *)dst_addr;
1131                 ib->sib_family = AF_IB;
1132                 if (path) {
1133                         ib->sib_pkey = path->pkey;
1134                         ib->sib_flowinfo = path->flow_label;
1135                         memcpy(&ib->sib_addr, &path->dgid, 16);
1136                 }
1137         }
1138 }
1139
1140 static void cma_save_ip4_info(struct sockaddr_in *src_addr,
1141                               struct sockaddr_in *dst_addr,
1142                               struct cma_hdr *hdr,
1143                               __be16 local_port)
1144 {
1145         if (src_addr) {
1146                 *src_addr = (struct sockaddr_in) {
1147                         .sin_family = AF_INET,
1148                         .sin_addr.s_addr = hdr->dst_addr.ip4.addr,
1149                         .sin_port = local_port,
1150                 };
1151         }
1152
1153         if (dst_addr) {
1154                 *dst_addr = (struct sockaddr_in) {
1155                         .sin_family = AF_INET,
1156                         .sin_addr.s_addr = hdr->src_addr.ip4.addr,
1157                         .sin_port = hdr->port,
1158                 };
1159         }
1160 }
1161
1162 static void cma_save_ip6_info(struct sockaddr_in6 *src_addr,
1163                               struct sockaddr_in6 *dst_addr,
1164                               struct cma_hdr *hdr,
1165                               __be16 local_port)
1166 {
1167         if (src_addr) {
1168                 *src_addr = (struct sockaddr_in6) {
1169                         .sin6_family = AF_INET6,
1170                         .sin6_addr = hdr->dst_addr.ip6,
1171                         .sin6_port = local_port,
1172                 };
1173         }
1174
1175         if (dst_addr) {
1176                 *dst_addr = (struct sockaddr_in6) {
1177                         .sin6_family = AF_INET6,
1178                         .sin6_addr = hdr->src_addr.ip6,
1179                         .sin6_port = hdr->port,
1180                 };
1181         }
1182 }
1183
1184 static u16 cma_port_from_service_id(__be64 service_id)
1185 {
1186         return (u16)be64_to_cpu(service_id);
1187 }
1188
1189 static int cma_save_ip_info(struct sockaddr *src_addr,
1190                             struct sockaddr *dst_addr,
1191                             struct ib_cm_event *ib_event,
1192                             __be64 service_id)
1193 {
1194         struct cma_hdr *hdr;
1195         __be16 port;
1196
1197         hdr = ib_event->private_data;
1198         if (hdr->cma_version != CMA_VERSION)
1199                 return -EINVAL;
1200
1201         port = htons(cma_port_from_service_id(service_id));
1202
1203         switch (cma_get_ip_ver(hdr)) {
1204         case 4:
1205                 cma_save_ip4_info((struct sockaddr_in *)src_addr,
1206                                   (struct sockaddr_in *)dst_addr, hdr, port);
1207                 break;
1208         case 6:
1209                 cma_save_ip6_info((struct sockaddr_in6 *)src_addr,
1210                                   (struct sockaddr_in6 *)dst_addr, hdr, port);
1211                 break;
1212         default:
1213                 return -EAFNOSUPPORT;
1214         }
1215
1216         return 0;
1217 }
1218
1219 static int cma_save_net_info(struct sockaddr *src_addr,
1220                              struct sockaddr *dst_addr,
1221                              struct rdma_cm_id *listen_id,
1222                              struct ib_cm_event *ib_event,
1223                              sa_family_t sa_family, __be64 service_id)
1224 {
1225         if (sa_family == AF_IB) {
1226                 if (ib_event->event == IB_CM_REQ_RECEIVED)
1227                         cma_save_ib_info(src_addr, dst_addr, listen_id,
1228                                          ib_event->param.req_rcvd.primary_path);
1229                 else if (ib_event->event == IB_CM_SIDR_REQ_RECEIVED)
1230                         cma_save_ib_info(src_addr, dst_addr, listen_id, NULL);
1231                 return 0;
1232         }
1233
1234         return cma_save_ip_info(src_addr, dst_addr, ib_event, service_id);
1235 }
1236
1237 static int cma_save_req_info(const struct ib_cm_event *ib_event,
1238                              struct cma_req_info *req)
1239 {
1240         const struct ib_cm_req_event_param *req_param =
1241                 &ib_event->param.req_rcvd;
1242         const struct ib_cm_sidr_req_event_param *sidr_param =
1243                 &ib_event->param.sidr_req_rcvd;
1244
1245         switch (ib_event->event) {
1246         case IB_CM_REQ_RECEIVED:
1247                 req->device     = req_param->listen_id->device;
1248                 req->port       = req_param->port;
1249                 memcpy(&req->local_gid, &req_param->primary_path->sgid,
1250                        sizeof(req->local_gid));
1251                 req->has_gid    = true;
1252                 req->service_id = req_param->primary_path->service_id;
1253                 req->pkey       = be16_to_cpu(req_param->primary_path->pkey);
1254                 if (req->pkey != req_param->bth_pkey)
1255                         pr_warn_ratelimited("RDMA CMA: got different BTH P_Key (0x%x) and primary path P_Key (0x%x)\n"
1256                                             "RDMA CMA: in the future this may cause the request to be dropped\n",
1257                                             req_param->bth_pkey, req->pkey);
1258                 break;
1259         case IB_CM_SIDR_REQ_RECEIVED:
1260                 req->device     = sidr_param->listen_id->device;
1261                 req->port       = sidr_param->port;
1262                 req->has_gid    = false;
1263                 req->service_id = sidr_param->service_id;
1264                 req->pkey       = sidr_param->pkey;
1265                 if (req->pkey != sidr_param->bth_pkey)
1266                         pr_warn_ratelimited("RDMA CMA: got different BTH P_Key (0x%x) and SIDR request payload P_Key (0x%x)\n"
1267                                             "RDMA CMA: in the future this may cause the request to be dropped\n",
1268                                             sidr_param->bth_pkey, req->pkey);
1269                 break;
1270         default:
1271                 return -EINVAL;
1272         }
1273
1274         return 0;
1275 }
1276
1277 static bool validate_ipv4_net_dev(struct net_device *net_dev,
1278                                   const struct sockaddr_in *dst_addr,
1279                                   const struct sockaddr_in *src_addr)
1280 {
1281         __be32 daddr = dst_addr->sin_addr.s_addr,
1282                saddr = src_addr->sin_addr.s_addr;
1283         struct fib_result res;
1284         struct flowi4 fl4;
1285         int err;
1286         bool ret;
1287
1288         if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1289             ipv4_is_lbcast(daddr) || ipv4_is_zeronet(saddr) ||
1290             ipv4_is_zeronet(daddr) || ipv4_is_loopback(daddr) ||
1291             ipv4_is_loopback(saddr))
1292                 return false;
1293
1294         memset(&fl4, 0, sizeof(fl4));
1295         fl4.flowi4_iif = net_dev->ifindex;
1296         fl4.daddr = daddr;
1297         fl4.saddr = saddr;
1298
1299         rcu_read_lock();
1300         err = fib_lookup(dev_net(net_dev), &fl4, &res, 0);
1301         ret = err == 0 && FIB_RES_DEV(res) == net_dev;
1302         rcu_read_unlock();
1303
1304         return ret;
1305 }
1306
1307 static bool validate_ipv6_net_dev(struct net_device *net_dev,
1308                                   const struct sockaddr_in6 *dst_addr,
1309                                   const struct sockaddr_in6 *src_addr)
1310 {
1311 #if IS_ENABLED(CONFIG_IPV6)
1312         const int strict = ipv6_addr_type(&dst_addr->sin6_addr) &
1313                            IPV6_ADDR_LINKLOCAL;
1314         struct rt6_info *rt = rt6_lookup(dev_net(net_dev), &dst_addr->sin6_addr,
1315                                          &src_addr->sin6_addr, net_dev->ifindex,
1316                                          strict);
1317         bool ret;
1318
1319         if (!rt)
1320                 return false;
1321
1322         ret = rt->rt6i_idev->dev == net_dev;
1323         ip6_rt_put(rt);
1324
1325         return ret;
1326 #else
1327         return false;
1328 #endif
1329 }
1330
1331 static bool validate_net_dev(struct net_device *net_dev,
1332                              const struct sockaddr *daddr,
1333                              const struct sockaddr *saddr)
1334 {
1335         const struct sockaddr_in *daddr4 = (const struct sockaddr_in *)daddr;
1336         const struct sockaddr_in *saddr4 = (const struct sockaddr_in *)saddr;
1337         const struct sockaddr_in6 *daddr6 = (const struct sockaddr_in6 *)daddr;
1338         const struct sockaddr_in6 *saddr6 = (const struct sockaddr_in6 *)saddr;
1339
1340         switch (daddr->sa_family) {
1341         case AF_INET:
1342                 return saddr->sa_family == AF_INET &&
1343                        validate_ipv4_net_dev(net_dev, daddr4, saddr4);
1344
1345         case AF_INET6:
1346                 return saddr->sa_family == AF_INET6 &&
1347                        validate_ipv6_net_dev(net_dev, daddr6, saddr6);
1348
1349         default:
1350                 return false;
1351         }
1352 }
1353
1354 static struct net_device *cma_get_net_dev(struct ib_cm_event *ib_event,
1355                                           const struct cma_req_info *req)
1356 {
1357         struct sockaddr_storage listen_addr_storage, src_addr_storage;
1358         struct sockaddr *listen_addr = (struct sockaddr *)&listen_addr_storage,
1359                         *src_addr = (struct sockaddr *)&src_addr_storage;
1360         struct net_device *net_dev;
1361         const union ib_gid *gid = req->has_gid ? &req->local_gid : NULL;
1362         int err;
1363
1364         err = cma_save_ip_info(listen_addr, src_addr, ib_event,
1365                                req->service_id);
1366         if (err)
1367                 return ERR_PTR(err);
1368
1369         net_dev = ib_get_net_dev_by_params(req->device, req->port, req->pkey,
1370                                            gid, listen_addr);
1371         if (!net_dev)
1372                 return ERR_PTR(-ENODEV);
1373
1374         if (!validate_net_dev(net_dev, listen_addr, src_addr)) {
1375                 dev_put(net_dev);
1376                 return ERR_PTR(-EHOSTUNREACH);
1377         }
1378
1379         return net_dev;
1380 }
1381
1382 static enum rdma_port_space rdma_ps_from_service_id(__be64 service_id)
1383 {
1384         return (be64_to_cpu(service_id) >> 16) & 0xffff;
1385 }
1386
1387 static bool cma_match_private_data(struct rdma_id_private *id_priv,
1388                                    const struct cma_hdr *hdr)
1389 {
1390         struct sockaddr *addr = cma_src_addr(id_priv);
1391         __be32 ip4_addr;
1392         struct in6_addr ip6_addr;
1393
1394         if (cma_any_addr(addr) && !id_priv->afonly)
1395                 return true;
1396
1397         switch (addr->sa_family) {
1398         case AF_INET:
1399                 ip4_addr = ((struct sockaddr_in *)addr)->sin_addr.s_addr;
1400                 if (cma_get_ip_ver(hdr) != 4)
1401                         return false;
1402                 if (!cma_any_addr(addr) &&
1403                     hdr->dst_addr.ip4.addr != ip4_addr)
1404                         return false;
1405                 break;
1406         case AF_INET6:
1407                 ip6_addr = ((struct sockaddr_in6 *)addr)->sin6_addr;
1408                 if (cma_get_ip_ver(hdr) != 6)
1409                         return false;
1410                 if (!cma_any_addr(addr) &&
1411                     memcmp(&hdr->dst_addr.ip6, &ip6_addr, sizeof(ip6_addr)))
1412                         return false;
1413                 break;
1414         case AF_IB:
1415                 return true;
1416         default:
1417                 return false;
1418         }
1419
1420         return true;
1421 }
1422
1423 static bool cma_protocol_roce_dev_port(struct ib_device *device, int port_num)
1424 {
1425         enum rdma_link_layer ll = rdma_port_get_link_layer(device, port_num);
1426         enum rdma_transport_type transport =
1427                 rdma_node_get_transport(device->node_type);
1428
1429         return ll == IB_LINK_LAYER_ETHERNET && transport == RDMA_TRANSPORT_IB;
1430 }
1431
1432 static bool cma_protocol_roce(const struct rdma_cm_id *id)
1433 {
1434         struct ib_device *device = id->device;
1435         const int port_num = id->port_num ?: rdma_start_port(device);
1436
1437         return cma_protocol_roce_dev_port(device, port_num);
1438 }
1439
1440 static bool cma_match_net_dev(const struct rdma_cm_id *id,
1441                               const struct net_device *net_dev,
1442                               u8 port_num)
1443 {
1444         const struct rdma_addr *addr = &id->route.addr;
1445
1446         if (!net_dev)
1447                 /* This request is an AF_IB request or a RoCE request */
1448                 return (!id->port_num || id->port_num == port_num) &&
1449                        (addr->src_addr.ss_family == AF_IB ||
1450                         cma_protocol_roce_dev_port(id->device, port_num));
1451
1452         return !addr->dev_addr.bound_dev_if ||
1453                (net_eq(dev_net(net_dev), addr->dev_addr.net) &&
1454                 addr->dev_addr.bound_dev_if == net_dev->ifindex);
1455 }
1456
1457 static struct rdma_id_private *cma_find_listener(
1458                 const struct rdma_bind_list *bind_list,
1459                 const struct ib_cm_id *cm_id,
1460                 const struct ib_cm_event *ib_event,
1461                 const struct cma_req_info *req,
1462                 const struct net_device *net_dev)
1463 {
1464         struct rdma_id_private *id_priv, *id_priv_dev;
1465
1466         if (!bind_list)
1467                 return ERR_PTR(-EINVAL);
1468
1469         hlist_for_each_entry(id_priv, &bind_list->owners, node) {
1470                 if (cma_match_private_data(id_priv, ib_event->private_data)) {
1471                         if (id_priv->id.device == cm_id->device &&
1472                             cma_match_net_dev(&id_priv->id, net_dev, req->port))
1473                                 return id_priv;
1474                         list_for_each_entry(id_priv_dev,
1475                                             &id_priv->listen_list,
1476                                             listen_list) {
1477                                 if (id_priv_dev->id.device == cm_id->device &&
1478                                     cma_match_net_dev(&id_priv_dev->id, net_dev, req->port))
1479                                         return id_priv_dev;
1480                         }
1481                 }
1482         }
1483
1484         return ERR_PTR(-EINVAL);
1485 }
1486
1487 static struct rdma_id_private *cma_id_from_event(struct ib_cm_id *cm_id,
1488                                                  struct ib_cm_event *ib_event,
1489                                                  struct net_device **net_dev)
1490 {
1491         struct cma_req_info req;
1492         struct rdma_bind_list *bind_list;
1493         struct rdma_id_private *id_priv;
1494         int err;
1495
1496         err = cma_save_req_info(ib_event, &req);
1497         if (err)
1498                 return ERR_PTR(err);
1499
1500         *net_dev = cma_get_net_dev(ib_event, &req);
1501         if (IS_ERR(*net_dev)) {
1502                 if (PTR_ERR(*net_dev) == -EAFNOSUPPORT) {
1503                         /* Assuming the protocol is AF_IB */
1504                         *net_dev = NULL;
1505                 } else if (cma_protocol_roce_dev_port(req.device, req.port)) {
1506                         /* TODO find the net dev matching the request parameters
1507                          * through the RoCE GID table */
1508                         *net_dev = NULL;
1509                 } else {
1510                         return ERR_CAST(*net_dev);
1511                 }
1512         }
1513
1514         bind_list = cma_ps_find(*net_dev ? dev_net(*net_dev) : &init_net,
1515                                 rdma_ps_from_service_id(req.service_id),
1516                                 cma_port_from_service_id(req.service_id));
1517         id_priv = cma_find_listener(bind_list, cm_id, ib_event, &req, *net_dev);
1518         if (IS_ERR(id_priv) && *net_dev) {
1519                 dev_put(*net_dev);
1520                 *net_dev = NULL;
1521         }
1522
1523         return id_priv;
1524 }
1525
1526 static inline int cma_user_data_offset(struct rdma_id_private *id_priv)
1527 {
1528         return cma_family(id_priv) == AF_IB ? 0 : sizeof(struct cma_hdr);
1529 }
1530
1531 static void cma_cancel_route(struct rdma_id_private *id_priv)
1532 {
1533         if (rdma_cap_ib_sa(id_priv->id.device, id_priv->id.port_num)) {
1534                 if (id_priv->query)
1535                         ib_sa_cancel_query(id_priv->query_id, id_priv->query);
1536         }
1537 }
1538
1539 static void cma_cancel_listens(struct rdma_id_private *id_priv)
1540 {
1541         struct rdma_id_private *dev_id_priv;
1542
1543         /*
1544          * Remove from listen_any_list to prevent added devices from spawning
1545          * additional listen requests.
1546          */
1547         mutex_lock(&lock);
1548         list_del(&id_priv->list);
1549
1550         while (!list_empty(&id_priv->listen_list)) {
1551                 dev_id_priv = list_entry(id_priv->listen_list.next,
1552                                          struct rdma_id_private, listen_list);
1553                 /* sync with device removal to avoid duplicate destruction */
1554                 list_del_init(&dev_id_priv->list);
1555                 list_del(&dev_id_priv->listen_list);
1556                 mutex_unlock(&lock);
1557
1558                 rdma_destroy_id(&dev_id_priv->id);
1559                 mutex_lock(&lock);
1560         }
1561         mutex_unlock(&lock);
1562 }
1563
1564 static void cma_cancel_operation(struct rdma_id_private *id_priv,
1565                                  enum rdma_cm_state state)
1566 {
1567         switch (state) {
1568         case RDMA_CM_ADDR_QUERY:
1569                 rdma_addr_cancel(&id_priv->id.route.addr.dev_addr);
1570                 break;
1571         case RDMA_CM_ROUTE_QUERY:
1572                 cma_cancel_route(id_priv);
1573                 break;
1574         case RDMA_CM_LISTEN:
1575                 if (cma_any_addr(cma_src_addr(id_priv)) && !id_priv->cma_dev)
1576                         cma_cancel_listens(id_priv);
1577                 break;
1578         default:
1579                 break;
1580         }
1581 }
1582
1583 static void cma_release_port(struct rdma_id_private *id_priv)
1584 {
1585         struct rdma_bind_list *bind_list = id_priv->bind_list;
1586         struct net *net = id_priv->id.route.addr.dev_addr.net;
1587
1588         if (!bind_list)
1589                 return;
1590
1591         mutex_lock(&lock);
1592         hlist_del(&id_priv->node);
1593         if (hlist_empty(&bind_list->owners)) {
1594                 cma_ps_remove(net, bind_list->ps, bind_list->port);
1595                 kfree(bind_list);
1596         }
1597         mutex_unlock(&lock);
1598 }
1599
1600 static void cma_leave_mc_groups(struct rdma_id_private *id_priv)
1601 {
1602         struct cma_multicast *mc;
1603
1604         while (!list_empty(&id_priv->mc_list)) {
1605                 mc = container_of(id_priv->mc_list.next,
1606                                   struct cma_multicast, list);
1607                 list_del(&mc->list);
1608                 if (rdma_cap_ib_mcast(id_priv->cma_dev->device,
1609                                       id_priv->id.port_num)) {
1610                         ib_sa_free_multicast(mc->multicast.ib);
1611                         kfree(mc);
1612                 } else {
1613                         if (mc->igmp_joined) {
1614                                 struct rdma_dev_addr *dev_addr =
1615                                         &id_priv->id.route.addr.dev_addr;
1616                                 struct net_device *ndev = NULL;
1617
1618                                 if (dev_addr->bound_dev_if)
1619                                         ndev = dev_get_by_index(&init_net,
1620                                                                 dev_addr->bound_dev_if);
1621                                 if (ndev) {
1622                                         cma_igmp_send(ndev,
1623                                                       &mc->multicast.ib->rec.mgid,
1624                                                       false);
1625                                         dev_put(ndev);
1626                                 }
1627                         }
1628                         kref_put(&mc->mcref, release_mc);
1629                 }
1630         }
1631 }
1632
1633 void rdma_destroy_id(struct rdma_cm_id *id)
1634 {
1635         struct rdma_id_private *id_priv;
1636         enum rdma_cm_state state;
1637
1638         id_priv = container_of(id, struct rdma_id_private, id);
1639         state = cma_exch(id_priv, RDMA_CM_DESTROYING);
1640         cma_cancel_operation(id_priv, state);
1641
1642         /*
1643          * Wait for any active callback to finish.  New callbacks will find
1644          * the id_priv state set to destroying and abort.
1645          */
1646         mutex_lock(&id_priv->handler_mutex);
1647         mutex_unlock(&id_priv->handler_mutex);
1648
1649         if (id_priv->cma_dev) {
1650                 if (rdma_cap_ib_cm(id_priv->id.device, 1)) {
1651                         if (id_priv->cm_id.ib)
1652                                 ib_destroy_cm_id(id_priv->cm_id.ib);
1653                 } else if (rdma_cap_iw_cm(id_priv->id.device, 1)) {
1654                         if (id_priv->cm_id.iw)
1655                                 iw_destroy_cm_id(id_priv->cm_id.iw);
1656                 }
1657                 cma_leave_mc_groups(id_priv);
1658                 cma_release_dev(id_priv);
1659         }
1660
1661         cma_release_port(id_priv);
1662         cma_deref_id(id_priv);
1663         wait_for_completion(&id_priv->comp);
1664
1665         if (id_priv->internal_id)
1666                 cma_deref_id(id_priv->id.context);
1667
1668         kfree(id_priv->id.route.path_rec);
1669         put_net(id_priv->id.route.addr.dev_addr.net);
1670         kfree(id_priv);
1671 }
1672 EXPORT_SYMBOL(rdma_destroy_id);
1673
1674 static int cma_rep_recv(struct rdma_id_private *id_priv)
1675 {
1676         int ret;
1677
1678         ret = cma_modify_qp_rtr(id_priv, NULL);
1679         if (ret)
1680                 goto reject;
1681
1682         ret = cma_modify_qp_rts(id_priv, NULL);
1683         if (ret)
1684                 goto reject;
1685
1686         ret = ib_send_cm_rtu(id_priv->cm_id.ib, NULL, 0);
1687         if (ret)
1688                 goto reject;
1689
1690         return 0;
1691 reject:
1692         cma_modify_qp_err(id_priv);
1693         ib_send_cm_rej(id_priv->cm_id.ib, IB_CM_REJ_CONSUMER_DEFINED,
1694                        NULL, 0, NULL, 0);
1695         return ret;
1696 }
1697
1698 static void cma_set_rep_event_data(struct rdma_cm_event *event,
1699                                    struct ib_cm_rep_event_param *rep_data,
1700                                    void *private_data)
1701 {
1702         event->param.conn.private_data = private_data;
1703         event->param.conn.private_data_len = IB_CM_REP_PRIVATE_DATA_SIZE;
1704         event->param.conn.responder_resources = rep_data->responder_resources;
1705         event->param.conn.initiator_depth = rep_data->initiator_depth;
1706         event->param.conn.flow_control = rep_data->flow_control;
1707         event->param.conn.rnr_retry_count = rep_data->rnr_retry_count;
1708         event->param.conn.srq = rep_data->srq;
1709         event->param.conn.qp_num = rep_data->remote_qpn;
1710 }
1711
1712 static int cma_ib_handler(struct ib_cm_id *cm_id, struct ib_cm_event *ib_event)
1713 {
1714         struct rdma_id_private *id_priv = cm_id->context;
1715         struct rdma_cm_event event;
1716         int ret = 0;
1717
1718         mutex_lock(&id_priv->handler_mutex);
1719         if ((ib_event->event != IB_CM_TIMEWAIT_EXIT &&
1720              id_priv->state != RDMA_CM_CONNECT) ||
1721             (ib_event->event == IB_CM_TIMEWAIT_EXIT &&
1722              id_priv->state != RDMA_CM_DISCONNECT))
1723                 goto out;
1724
1725         memset(&event, 0, sizeof event);
1726         switch (ib_event->event) {
1727         case IB_CM_REQ_ERROR:
1728         case IB_CM_REP_ERROR:
1729                 event.event = RDMA_CM_EVENT_UNREACHABLE;
1730                 event.status = -ETIMEDOUT;
1731                 break;
1732         case IB_CM_REP_RECEIVED:
1733                 if (id_priv->id.qp) {
1734                         event.status = cma_rep_recv(id_priv);
1735                         event.event = event.status ? RDMA_CM_EVENT_CONNECT_ERROR :
1736                                                      RDMA_CM_EVENT_ESTABLISHED;
1737                 } else {
1738                         event.event = RDMA_CM_EVENT_CONNECT_RESPONSE;
1739                 }
1740                 cma_set_rep_event_data(&event, &ib_event->param.rep_rcvd,
1741                                        ib_event->private_data);
1742                 break;
1743         case IB_CM_RTU_RECEIVED:
1744         case IB_CM_USER_ESTABLISHED:
1745                 event.event = RDMA_CM_EVENT_ESTABLISHED;
1746                 break;
1747         case IB_CM_DREQ_ERROR:
1748                 event.status = -ETIMEDOUT; /* fall through */
1749         case IB_CM_DREQ_RECEIVED:
1750         case IB_CM_DREP_RECEIVED:
1751                 if (!cma_comp_exch(id_priv, RDMA_CM_CONNECT,
1752                                    RDMA_CM_DISCONNECT))
1753                         goto out;
1754                 event.event = RDMA_CM_EVENT_DISCONNECTED;
1755                 break;
1756         case IB_CM_TIMEWAIT_EXIT:
1757                 event.event = RDMA_CM_EVENT_TIMEWAIT_EXIT;
1758                 break;
1759         case IB_CM_MRA_RECEIVED:
1760                 /* ignore event */
1761                 goto out;
1762         case IB_CM_REJ_RECEIVED:
1763                 cma_modify_qp_err(id_priv);
1764                 event.status = ib_event->param.rej_rcvd.reason;
1765                 event.event = RDMA_CM_EVENT_REJECTED;
1766                 event.param.conn.private_data = ib_event->private_data;
1767                 event.param.conn.private_data_len = IB_CM_REJ_PRIVATE_DATA_SIZE;
1768                 break;
1769         default:
1770                 pr_err("RDMA CMA: unexpected IB CM event: %d\n",
1771                        ib_event->event);
1772                 goto out;
1773         }
1774
1775         ret = id_priv->id.event_handler(&id_priv->id, &event);
1776         if (ret) {
1777                 /* Destroy the CM ID by returning a non-zero value. */
1778                 id_priv->cm_id.ib = NULL;
1779                 cma_exch(id_priv, RDMA_CM_DESTROYING);
1780                 mutex_unlock(&id_priv->handler_mutex);
1781                 rdma_destroy_id(&id_priv->id);
1782                 return ret;
1783         }
1784 out:
1785         mutex_unlock(&id_priv->handler_mutex);
1786         return ret;
1787 }
1788
1789 static struct rdma_id_private *cma_new_conn_id(struct rdma_cm_id *listen_id,
1790                                                struct ib_cm_event *ib_event,
1791                                                struct net_device *net_dev)
1792 {
1793         struct rdma_id_private *id_priv;
1794         struct rdma_cm_id *id;
1795         struct rdma_route *rt;
1796         const sa_family_t ss_family = listen_id->route.addr.src_addr.ss_family;
1797         const __be64 service_id =
1798                       ib_event->param.req_rcvd.primary_path->service_id;
1799         int ret;
1800
1801         id = rdma_create_id(listen_id->route.addr.dev_addr.net,
1802                             listen_id->event_handler, listen_id->context,
1803                             listen_id->ps, ib_event->param.req_rcvd.qp_type);
1804         if (IS_ERR(id))
1805                 return NULL;
1806
1807         id_priv = container_of(id, struct rdma_id_private, id);
1808         if (cma_save_net_info((struct sockaddr *)&id->route.addr.src_addr,
1809                               (struct sockaddr *)&id->route.addr.dst_addr,
1810                               listen_id, ib_event, ss_family, service_id))
1811                 goto err;
1812
1813         rt = &id->route;
1814         rt->num_paths = ib_event->param.req_rcvd.alternate_path ? 2 : 1;
1815         rt->path_rec = kmalloc(sizeof *rt->path_rec * rt->num_paths,
1816                                GFP_KERNEL);
1817         if (!rt->path_rec)
1818                 goto err;
1819
1820         rt->path_rec[0] = *ib_event->param.req_rcvd.primary_path;
1821         if (rt->num_paths == 2)
1822                 rt->path_rec[1] = *ib_event->param.req_rcvd.alternate_path;
1823
1824         if (net_dev) {
1825                 ret = rdma_copy_addr(&rt->addr.dev_addr, net_dev, NULL);
1826                 if (ret)
1827                         goto err;
1828         } else {
1829                 if (!cma_protocol_roce(listen_id) &&
1830                     cma_any_addr(cma_src_addr(id_priv))) {
1831                         rt->addr.dev_addr.dev_type = ARPHRD_INFINIBAND;
1832                         rdma_addr_set_sgid(&rt->addr.dev_addr, &rt->path_rec[0].sgid);
1833                         ib_addr_set_pkey(&rt->addr.dev_addr, be16_to_cpu(rt->path_rec[0].pkey));
1834                 } else if (!cma_any_addr(cma_src_addr(id_priv))) {
1835                         ret = cma_translate_addr(cma_src_addr(id_priv), &rt->addr.dev_addr);
1836                         if (ret)
1837                                 goto err;
1838                 }
1839         }
1840         rdma_addr_set_dgid(&rt->addr.dev_addr, &rt->path_rec[0].dgid);
1841
1842         id_priv->state = RDMA_CM_CONNECT;
1843         return id_priv;
1844
1845 err:
1846         rdma_destroy_id(id);
1847         return NULL;
1848 }
1849
1850 static struct rdma_id_private *cma_new_udp_id(struct rdma_cm_id *listen_id,
1851                                               struct ib_cm_event *ib_event,
1852                                               struct net_device *net_dev)
1853 {
1854         struct rdma_id_private *id_priv;
1855         struct rdma_cm_id *id;
1856         const sa_family_t ss_family = listen_id->route.addr.src_addr.ss_family;
1857         struct net *net = listen_id->route.addr.dev_addr.net;
1858         int ret;
1859
1860         id = rdma_create_id(net, listen_id->event_handler, listen_id->context,
1861                             listen_id->ps, IB_QPT_UD);
1862         if (IS_ERR(id))
1863                 return NULL;
1864
1865         id_priv = container_of(id, struct rdma_id_private, id);
1866         if (cma_save_net_info((struct sockaddr *)&id->route.addr.src_addr,
1867                               (struct sockaddr *)&id->route.addr.dst_addr,
1868                               listen_id, ib_event, ss_family,
1869                               ib_event->param.sidr_req_rcvd.service_id))
1870                 goto err;
1871
1872         if (net_dev) {
1873                 ret = rdma_copy_addr(&id->route.addr.dev_addr, net_dev, NULL);
1874                 if (ret)
1875                         goto err;
1876         } else {
1877                 if (!cma_any_addr(cma_src_addr(id_priv))) {
1878                         ret = cma_translate_addr(cma_src_addr(id_priv),
1879                                                  &id->route.addr.dev_addr);
1880                         if (ret)
1881                                 goto err;
1882                 }
1883         }
1884
1885         id_priv->state = RDMA_CM_CONNECT;
1886         return id_priv;
1887 err:
1888         rdma_destroy_id(id);
1889         return NULL;
1890 }
1891
1892 static void cma_set_req_event_data(struct rdma_cm_event *event,
1893                                    struct ib_cm_req_event_param *req_data,
1894                                    void *private_data, int offset)
1895 {
1896         event->param.conn.private_data = private_data + offset;
1897         event->param.conn.private_data_len = IB_CM_REQ_PRIVATE_DATA_SIZE - offset;
1898         event->param.conn.responder_resources = req_data->responder_resources;
1899         event->param.conn.initiator_depth = req_data->initiator_depth;
1900         event->param.conn.flow_control = req_data->flow_control;
1901         event->param.conn.retry_count = req_data->retry_count;
1902         event->param.conn.rnr_retry_count = req_data->rnr_retry_count;
1903         event->param.conn.srq = req_data->srq;
1904         event->param.conn.qp_num = req_data->remote_qpn;
1905 }
1906
1907 static int cma_check_req_qp_type(struct rdma_cm_id *id, struct ib_cm_event *ib_event)
1908 {
1909         return (((ib_event->event == IB_CM_REQ_RECEIVED) &&
1910                  (ib_event->param.req_rcvd.qp_type == id->qp_type)) ||
1911                 ((ib_event->event == IB_CM_SIDR_REQ_RECEIVED) &&
1912                  (id->qp_type == IB_QPT_UD)) ||
1913                 (!id->qp_type));
1914 }
1915
1916 static int cma_req_handler(struct ib_cm_id *cm_id, struct ib_cm_event *ib_event)
1917 {
1918         struct rdma_id_private *listen_id, *conn_id = NULL;
1919         struct rdma_cm_event event;
1920         struct net_device *net_dev;
1921         int offset, ret;
1922
1923         listen_id = cma_id_from_event(cm_id, ib_event, &net_dev);
1924         if (IS_ERR(listen_id))
1925                 return PTR_ERR(listen_id);
1926
1927         if (!cma_check_req_qp_type(&listen_id->id, ib_event)) {
1928                 ret = -EINVAL;
1929                 goto net_dev_put;
1930         }
1931
1932         mutex_lock(&listen_id->handler_mutex);
1933         if (listen_id->state != RDMA_CM_LISTEN) {
1934                 ret = -ECONNABORTED;
1935                 goto err1;
1936         }
1937
1938         memset(&event, 0, sizeof event);
1939         offset = cma_user_data_offset(listen_id);
1940         event.event = RDMA_CM_EVENT_CONNECT_REQUEST;
1941         if (ib_event->event == IB_CM_SIDR_REQ_RECEIVED) {
1942                 conn_id = cma_new_udp_id(&listen_id->id, ib_event, net_dev);
1943                 event.param.ud.private_data = ib_event->private_data + offset;
1944                 event.param.ud.private_data_len =
1945                                 IB_CM_SIDR_REQ_PRIVATE_DATA_SIZE - offset;
1946         } else {
1947                 conn_id = cma_new_conn_id(&listen_id->id, ib_event, net_dev);
1948                 cma_set_req_event_data(&event, &ib_event->param.req_rcvd,
1949                                        ib_event->private_data, offset);
1950         }
1951         if (!conn_id) {
1952                 ret = -ENOMEM;
1953                 goto err1;
1954         }
1955
1956         mutex_lock_nested(&conn_id->handler_mutex, SINGLE_DEPTH_NESTING);
1957         ret = cma_acquire_dev(conn_id, listen_id);
1958         if (ret)
1959                 goto err2;
1960
1961         conn_id->cm_id.ib = cm_id;
1962         cm_id->context = conn_id;
1963         cm_id->cm_handler = cma_ib_handler;
1964
1965         /*
1966          * Protect against the user destroying conn_id from another thread
1967          * until we're done accessing it.
1968          */
1969         atomic_inc(&conn_id->refcount);
1970         ret = conn_id->id.event_handler(&conn_id->id, &event);
1971         if (ret)
1972                 goto err3;
1973         /*
1974          * Acquire mutex to prevent user executing rdma_destroy_id()
1975          * while we're accessing the cm_id.
1976          */
1977         mutex_lock(&lock);
1978         if (cma_comp(conn_id, RDMA_CM_CONNECT) &&
1979             (conn_id->id.qp_type != IB_QPT_UD))
1980                 ib_send_cm_mra(cm_id, CMA_CM_MRA_SETTING, NULL, 0);
1981         mutex_unlock(&lock);
1982         mutex_unlock(&conn_id->handler_mutex);
1983         mutex_unlock(&listen_id->handler_mutex);
1984         cma_deref_id(conn_id);
1985         if (net_dev)
1986                 dev_put(net_dev);
1987         return 0;
1988
1989 err3:
1990         cma_deref_id(conn_id);
1991         /* Destroy the CM ID by returning a non-zero value. */
1992         conn_id->cm_id.ib = NULL;
1993 err2:
1994         cma_exch(conn_id, RDMA_CM_DESTROYING);
1995         mutex_unlock(&conn_id->handler_mutex);
1996 err1:
1997         mutex_unlock(&listen_id->handler_mutex);
1998         if (conn_id)
1999                 rdma_destroy_id(&conn_id->id);
2000
2001 net_dev_put:
2002         if (net_dev)
2003                 dev_put(net_dev);
2004
2005         return ret;
2006 }
2007
2008 __be64 rdma_get_service_id(struct rdma_cm_id *id, struct sockaddr *addr)
2009 {
2010         if (addr->sa_family == AF_IB)
2011                 return ((struct sockaddr_ib *) addr)->sib_sid;
2012
2013         return cpu_to_be64(((u64)id->ps << 16) + be16_to_cpu(cma_port(addr)));
2014 }
2015 EXPORT_SYMBOL(rdma_get_service_id);
2016
2017 static int cma_iw_handler(struct iw_cm_id *iw_id, struct iw_cm_event *iw_event)
2018 {
2019         struct rdma_id_private *id_priv = iw_id->context;
2020         struct rdma_cm_event event;
2021         int ret = 0;
2022         struct sockaddr *laddr = (struct sockaddr *)&iw_event->local_addr;
2023         struct sockaddr *raddr = (struct sockaddr *)&iw_event->remote_addr;
2024
2025         mutex_lock(&id_priv->handler_mutex);
2026         if (id_priv->state != RDMA_CM_CONNECT)
2027                 goto out;
2028
2029         memset(&event, 0, sizeof event);
2030         switch (iw_event->event) {
2031         case IW_CM_EVENT_CLOSE:
2032                 event.event = RDMA_CM_EVENT_DISCONNECTED;
2033                 break;
2034         case IW_CM_EVENT_CONNECT_REPLY:
2035                 memcpy(cma_src_addr(id_priv), laddr,
2036                        rdma_addr_size(laddr));
2037                 memcpy(cma_dst_addr(id_priv), raddr,
2038                        rdma_addr_size(raddr));
2039                 switch (iw_event->status) {
2040                 case 0:
2041                         event.event = RDMA_CM_EVENT_ESTABLISHED;
2042                         event.param.conn.initiator_depth = iw_event->ird;
2043                         event.param.conn.responder_resources = iw_event->ord;
2044                         break;
2045                 case -ECONNRESET:
2046                 case -ECONNREFUSED:
2047                         event.event = RDMA_CM_EVENT_REJECTED;
2048                         break;
2049                 case -ETIMEDOUT:
2050                         event.event = RDMA_CM_EVENT_UNREACHABLE;
2051                         break;
2052                 default:
2053                         event.event = RDMA_CM_EVENT_CONNECT_ERROR;
2054                         break;
2055                 }
2056                 break;
2057         case IW_CM_EVENT_ESTABLISHED:
2058                 event.event = RDMA_CM_EVENT_ESTABLISHED;
2059                 event.param.conn.initiator_depth = iw_event->ird;
2060                 event.param.conn.responder_resources = iw_event->ord;
2061                 break;
2062         default:
2063                 BUG_ON(1);
2064         }
2065
2066         event.status = iw_event->status;
2067         event.param.conn.private_data = iw_event->private_data;
2068         event.param.conn.private_data_len = iw_event->private_data_len;
2069         ret = id_priv->id.event_handler(&id_priv->id, &event);
2070         if (ret) {
2071                 /* Destroy the CM ID by returning a non-zero value. */
2072                 id_priv->cm_id.iw = NULL;
2073                 cma_exch(id_priv, RDMA_CM_DESTROYING);
2074                 mutex_unlock(&id_priv->handler_mutex);
2075                 rdma_destroy_id(&id_priv->id);
2076                 return ret;
2077         }
2078
2079 out:
2080         mutex_unlock(&id_priv->handler_mutex);
2081         return ret;
2082 }
2083
2084 static int iw_conn_req_handler(struct iw_cm_id *cm_id,
2085                                struct iw_cm_event *iw_event)
2086 {
2087         struct rdma_cm_id *new_cm_id;
2088         struct rdma_id_private *listen_id, *conn_id;
2089         struct rdma_cm_event event;
2090         int ret = -ECONNABORTED;
2091         struct sockaddr *laddr = (struct sockaddr *)&iw_event->local_addr;
2092         struct sockaddr *raddr = (struct sockaddr *)&iw_event->remote_addr;
2093
2094         listen_id = cm_id->context;
2095
2096         mutex_lock(&listen_id->handler_mutex);
2097         if (listen_id->state != RDMA_CM_LISTEN)
2098                 goto out;
2099
2100         /* Create a new RDMA id for the new IW CM ID */
2101         new_cm_id = rdma_create_id(listen_id->id.route.addr.dev_addr.net,
2102                                    listen_id->id.event_handler,
2103                                    listen_id->id.context,
2104                                    RDMA_PS_TCP, IB_QPT_RC);
2105         if (IS_ERR(new_cm_id)) {
2106                 ret = -ENOMEM;
2107                 goto out;
2108         }
2109         conn_id = container_of(new_cm_id, struct rdma_id_private, id);
2110         mutex_lock_nested(&conn_id->handler_mutex, SINGLE_DEPTH_NESTING);
2111         conn_id->state = RDMA_CM_CONNECT;
2112
2113         ret = rdma_translate_ip(laddr, &conn_id->id.route.addr.dev_addr, NULL);
2114         if (ret) {
2115                 mutex_unlock(&conn_id->handler_mutex);
2116                 rdma_destroy_id(new_cm_id);
2117                 goto out;
2118         }
2119
2120         ret = cma_acquire_dev(conn_id, listen_id);
2121         if (ret) {
2122                 mutex_unlock(&conn_id->handler_mutex);
2123                 rdma_destroy_id(new_cm_id);
2124                 goto out;
2125         }
2126
2127         conn_id->cm_id.iw = cm_id;
2128         cm_id->context = conn_id;
2129         cm_id->cm_handler = cma_iw_handler;
2130
2131         memcpy(cma_src_addr(conn_id), laddr, rdma_addr_size(laddr));
2132         memcpy(cma_dst_addr(conn_id), raddr, rdma_addr_size(raddr));
2133
2134         memset(&event, 0, sizeof event);
2135         event.event = RDMA_CM_EVENT_CONNECT_REQUEST;
2136         event.param.conn.private_data = iw_event->private_data;
2137         event.param.conn.private_data_len = iw_event->private_data_len;
2138         event.param.conn.initiator_depth = iw_event->ird;
2139         event.param.conn.responder_resources = iw_event->ord;
2140
2141         /*
2142          * Protect against the user destroying conn_id from another thread
2143          * until we're done accessing it.
2144          */
2145         atomic_inc(&conn_id->refcount);
2146         ret = conn_id->id.event_handler(&conn_id->id, &event);
2147         if (ret) {
2148                 /* User wants to destroy the CM ID */
2149                 conn_id->cm_id.iw = NULL;
2150                 cma_exch(conn_id, RDMA_CM_DESTROYING);
2151                 mutex_unlock(&conn_id->handler_mutex);
2152                 cma_deref_id(conn_id);
2153                 rdma_destroy_id(&conn_id->id);
2154                 goto out;
2155         }
2156
2157         mutex_unlock(&conn_id->handler_mutex);
2158         cma_deref_id(conn_id);
2159
2160 out:
2161         mutex_unlock(&listen_id->handler_mutex);
2162         return ret;
2163 }
2164
2165 static int cma_ib_listen(struct rdma_id_private *id_priv)
2166 {
2167         struct sockaddr *addr;
2168         struct ib_cm_id *id;
2169         __be64 svc_id;
2170
2171         addr = cma_src_addr(id_priv);
2172         svc_id = rdma_get_service_id(&id_priv->id, addr);
2173         id = ib_cm_insert_listen(id_priv->id.device, cma_req_handler, svc_id);
2174         if (IS_ERR(id))
2175                 return PTR_ERR(id);
2176         id_priv->cm_id.ib = id;
2177
2178         return 0;
2179 }
2180
2181 static int cma_iw_listen(struct rdma_id_private *id_priv, int backlog)
2182 {
2183         int ret;
2184         struct iw_cm_id *id;
2185
2186         id = iw_create_cm_id(id_priv->id.device,
2187                              iw_conn_req_handler,
2188                              id_priv);
2189         if (IS_ERR(id))
2190                 return PTR_ERR(id);
2191
2192         id->tos = id_priv->tos;
2193         id_priv->cm_id.iw = id;
2194
2195         memcpy(&id_priv->cm_id.iw->local_addr, cma_src_addr(id_priv),
2196                rdma_addr_size(cma_src_addr(id_priv)));
2197
2198         ret = iw_cm_listen(id_priv->cm_id.iw, backlog);
2199
2200         if (ret) {
2201                 iw_destroy_cm_id(id_priv->cm_id.iw);
2202                 id_priv->cm_id.iw = NULL;
2203         }
2204
2205         return ret;
2206 }
2207
2208 static int cma_listen_handler(struct rdma_cm_id *id,
2209                               struct rdma_cm_event *event)
2210 {
2211         struct rdma_id_private *id_priv = id->context;
2212
2213         id->context = id_priv->id.context;
2214         id->event_handler = id_priv->id.event_handler;
2215         return id_priv->id.event_handler(id, event);
2216 }
2217
2218 static void cma_listen_on_dev(struct rdma_id_private *id_priv,
2219                               struct cma_device *cma_dev)
2220 {
2221         struct rdma_id_private *dev_id_priv;
2222         struct rdma_cm_id *id;
2223         struct net *net = id_priv->id.route.addr.dev_addr.net;
2224         int ret;
2225
2226         if (cma_family(id_priv) == AF_IB && !rdma_cap_ib_cm(cma_dev->device, 1))
2227                 return;
2228
2229         id = rdma_create_id(net, cma_listen_handler, id_priv, id_priv->id.ps,
2230                             id_priv->id.qp_type);
2231         if (IS_ERR(id))
2232                 return;
2233
2234         dev_id_priv = container_of(id, struct rdma_id_private, id);
2235
2236         dev_id_priv->state = RDMA_CM_ADDR_BOUND;
2237         memcpy(cma_src_addr(dev_id_priv), cma_src_addr(id_priv),
2238                rdma_addr_size(cma_src_addr(id_priv)));
2239
2240         _cma_attach_to_dev(dev_id_priv, cma_dev);
2241         list_add_tail(&dev_id_priv->listen_list, &id_priv->listen_list);
2242         atomic_inc(&id_priv->refcount);
2243         dev_id_priv->internal_id = 1;
2244         dev_id_priv->afonly = id_priv->afonly;
2245
2246         ret = rdma_listen(id, id_priv->backlog);
2247         if (ret)
2248                 pr_warn("RDMA CMA: cma_listen_on_dev, error %d, listening on device %s\n",
2249                         ret, cma_dev->device->name);
2250 }
2251
2252 static void cma_listen_on_all(struct rdma_id_private *id_priv)
2253 {
2254         struct cma_device *cma_dev;
2255
2256         mutex_lock(&lock);
2257         list_add_tail(&id_priv->list, &listen_any_list);
2258         list_for_each_entry(cma_dev, &dev_list, list)
2259                 cma_listen_on_dev(id_priv, cma_dev);
2260         mutex_unlock(&lock);
2261 }
2262
2263 void rdma_set_service_type(struct rdma_cm_id *id, int tos)
2264 {
2265         struct rdma_id_private *id_priv;
2266
2267         id_priv = container_of(id, struct rdma_id_private, id);
2268         id_priv->tos = (u8) tos;
2269 }
2270 EXPORT_SYMBOL(rdma_set_service_type);
2271
2272 static void cma_query_handler(int status, struct ib_sa_path_rec *path_rec,
2273                               void *context)
2274 {
2275         struct cma_work *work = context;
2276         struct rdma_route *route;
2277
2278         route = &work->id->id.route;
2279
2280         if (!status) {
2281                 route->num_paths = 1;
2282                 *route->path_rec = *path_rec;
2283         } else {
2284                 work->old_state = RDMA_CM_ROUTE_QUERY;
2285                 work->new_state = RDMA_CM_ADDR_RESOLVED;
2286                 work->event.event = RDMA_CM_EVENT_ROUTE_ERROR;
2287                 work->event.status = status;
2288         }
2289
2290         queue_work(cma_wq, &work->work);
2291 }
2292
2293 static int cma_query_ib_route(struct rdma_id_private *id_priv, int timeout_ms,
2294                               struct cma_work *work)
2295 {
2296         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
2297         struct ib_sa_path_rec path_rec;
2298         ib_sa_comp_mask comp_mask;
2299         struct sockaddr_in6 *sin6;
2300         struct sockaddr_ib *sib;
2301
2302         memset(&path_rec, 0, sizeof path_rec);
2303         rdma_addr_get_sgid(dev_addr, &path_rec.sgid);
2304         rdma_addr_get_dgid(dev_addr, &path_rec.dgid);
2305         path_rec.pkey = cpu_to_be16(ib_addr_get_pkey(dev_addr));
2306         path_rec.numb_path = 1;
2307         path_rec.reversible = 1;
2308         path_rec.service_id = rdma_get_service_id(&id_priv->id, cma_dst_addr(id_priv));
2309
2310         comp_mask = IB_SA_PATH_REC_DGID | IB_SA_PATH_REC_SGID |
2311                     IB_SA_PATH_REC_PKEY | IB_SA_PATH_REC_NUMB_PATH |
2312                     IB_SA_PATH_REC_REVERSIBLE | IB_SA_PATH_REC_SERVICE_ID;
2313
2314         switch (cma_family(id_priv)) {
2315         case AF_INET:
2316                 path_rec.qos_class = cpu_to_be16((u16) id_priv->tos);
2317                 comp_mask |= IB_SA_PATH_REC_QOS_CLASS;
2318                 break;
2319         case AF_INET6:
2320                 sin6 = (struct sockaddr_in6 *) cma_src_addr(id_priv);
2321                 path_rec.traffic_class = (u8) (be32_to_cpu(sin6->sin6_flowinfo) >> 20);
2322                 comp_mask |= IB_SA_PATH_REC_TRAFFIC_CLASS;
2323                 break;
2324         case AF_IB:
2325                 sib = (struct sockaddr_ib *) cma_src_addr(id_priv);
2326                 path_rec.traffic_class = (u8) (be32_to_cpu(sib->sib_flowinfo) >> 20);
2327                 comp_mask |= IB_SA_PATH_REC_TRAFFIC_CLASS;
2328                 break;
2329         }
2330
2331         id_priv->query_id = ib_sa_path_rec_get(&sa_client, id_priv->id.device,
2332                                                id_priv->id.port_num, &path_rec,
2333                                                comp_mask, timeout_ms,
2334                                                GFP_KERNEL, cma_query_handler,
2335                                                work, &id_priv->query);
2336
2337         return (id_priv->query_id < 0) ? id_priv->query_id : 0;
2338 }
2339
2340 static void cma_work_handler(struct work_struct *_work)
2341 {
2342         struct cma_work *work = container_of(_work, struct cma_work, work);
2343         struct rdma_id_private *id_priv = work->id;
2344         int destroy = 0;
2345
2346         mutex_lock(&id_priv->handler_mutex);
2347         if (!cma_comp_exch(id_priv, work->old_state, work->new_state))
2348                 goto out;
2349
2350         if (id_priv->id.event_handler(&id_priv->id, &work->event)) {
2351                 cma_exch(id_priv, RDMA_CM_DESTROYING);
2352                 destroy = 1;
2353         }
2354 out:
2355         mutex_unlock(&id_priv->handler_mutex);
2356         cma_deref_id(id_priv);
2357         if (destroy)
2358                 rdma_destroy_id(&id_priv->id);
2359         kfree(work);
2360 }
2361
2362 static void cma_ndev_work_handler(struct work_struct *_work)
2363 {
2364         struct cma_ndev_work *work = container_of(_work, struct cma_ndev_work, work);
2365         struct rdma_id_private *id_priv = work->id;
2366         int destroy = 0;
2367
2368         mutex_lock(&id_priv->handler_mutex);
2369         if (id_priv->state == RDMA_CM_DESTROYING ||
2370             id_priv->state == RDMA_CM_DEVICE_REMOVAL)
2371                 goto out;
2372
2373         if (id_priv->id.event_handler(&id_priv->id, &work->event)) {
2374                 cma_exch(id_priv, RDMA_CM_DESTROYING);
2375                 destroy = 1;
2376         }
2377
2378 out:
2379         mutex_unlock(&id_priv->handler_mutex);
2380         cma_deref_id(id_priv);
2381         if (destroy)
2382                 rdma_destroy_id(&id_priv->id);
2383         kfree(work);
2384 }
2385
2386 static int cma_resolve_ib_route(struct rdma_id_private *id_priv, int timeout_ms)
2387 {
2388         struct rdma_route *route = &id_priv->id.route;
2389         struct cma_work *work;
2390         int ret;
2391
2392         work = kzalloc(sizeof *work, GFP_KERNEL);
2393         if (!work)
2394                 return -ENOMEM;
2395
2396         work->id = id_priv;
2397         INIT_WORK(&work->work, cma_work_handler);
2398         work->old_state = RDMA_CM_ROUTE_QUERY;
2399         work->new_state = RDMA_CM_ROUTE_RESOLVED;
2400         work->event.event = RDMA_CM_EVENT_ROUTE_RESOLVED;
2401
2402         route->path_rec = kmalloc(sizeof *route->path_rec, GFP_KERNEL);
2403         if (!route->path_rec) {
2404                 ret = -ENOMEM;
2405                 goto err1;
2406         }
2407
2408         ret = cma_query_ib_route(id_priv, timeout_ms, work);
2409         if (ret)
2410                 goto err2;
2411
2412         return 0;
2413 err2:
2414         kfree(route->path_rec);
2415         route->path_rec = NULL;
2416 err1:
2417         kfree(work);
2418         return ret;
2419 }
2420
2421 int rdma_set_ib_paths(struct rdma_cm_id *id,
2422                       struct ib_sa_path_rec *path_rec, int num_paths)
2423 {
2424         struct rdma_id_private *id_priv;
2425         int ret;
2426
2427         id_priv = container_of(id, struct rdma_id_private, id);
2428         if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_RESOLVED,
2429                            RDMA_CM_ROUTE_RESOLVED))
2430                 return -EINVAL;
2431
2432         id->route.path_rec = kmemdup(path_rec, sizeof *path_rec * num_paths,
2433                                      GFP_KERNEL);
2434         if (!id->route.path_rec) {
2435                 ret = -ENOMEM;
2436                 goto err;
2437         }
2438
2439         id->route.num_paths = num_paths;
2440         return 0;
2441 err:
2442         cma_comp_exch(id_priv, RDMA_CM_ROUTE_RESOLVED, RDMA_CM_ADDR_RESOLVED);
2443         return ret;
2444 }
2445 EXPORT_SYMBOL(rdma_set_ib_paths);
2446
2447 static int cma_resolve_iw_route(struct rdma_id_private *id_priv, int timeout_ms)
2448 {
2449         struct cma_work *work;
2450
2451         work = kzalloc(sizeof *work, GFP_KERNEL);
2452         if (!work)
2453                 return -ENOMEM;
2454
2455         work->id = id_priv;
2456         INIT_WORK(&work->work, cma_work_handler);
2457         work->old_state = RDMA_CM_ROUTE_QUERY;
2458         work->new_state = RDMA_CM_ROUTE_RESOLVED;
2459         work->event.event = RDMA_CM_EVENT_ROUTE_RESOLVED;
2460         queue_work(cma_wq, &work->work);
2461         return 0;
2462 }
2463
2464 static int iboe_tos_to_sl(struct net_device *ndev, int tos)
2465 {
2466         int prio;
2467         struct net_device *dev;
2468
2469         prio = rt_tos2priority(tos);
2470         dev = ndev->priv_flags & IFF_802_1Q_VLAN ?
2471                 vlan_dev_real_dev(ndev) : ndev;
2472
2473         if (dev->num_tc)
2474                 return netdev_get_prio_tc_map(dev, prio);
2475
2476 #if IS_ENABLED(CONFIG_VLAN_8021Q)
2477         if (ndev->priv_flags & IFF_802_1Q_VLAN)
2478                 return (vlan_dev_get_egress_qos_mask(ndev, prio) &
2479                         VLAN_PRIO_MASK) >> VLAN_PRIO_SHIFT;
2480 #endif
2481         return 0;
2482 }
2483
2484 static enum ib_gid_type cma_route_gid_type(enum rdma_network_type network_type,
2485                                            unsigned long supported_gids,
2486                                            enum ib_gid_type default_gid)
2487 {
2488         if ((network_type == RDMA_NETWORK_IPV4 ||
2489              network_type == RDMA_NETWORK_IPV6) &&
2490             test_bit(IB_GID_TYPE_ROCE_UDP_ENCAP, &supported_gids))
2491                 return IB_GID_TYPE_ROCE_UDP_ENCAP;
2492
2493         return default_gid;
2494 }
2495
2496 static int cma_resolve_iboe_route(struct rdma_id_private *id_priv)
2497 {
2498         struct rdma_route *route = &id_priv->id.route;
2499         struct rdma_addr *addr = &route->addr;
2500         struct cma_work *work;
2501         int ret;
2502         struct net_device *ndev = NULL;
2503
2504
2505         work = kzalloc(sizeof *work, GFP_KERNEL);
2506         if (!work)
2507                 return -ENOMEM;
2508
2509         work->id = id_priv;
2510         INIT_WORK(&work->work, cma_work_handler);
2511
2512         route->path_rec = kzalloc(sizeof *route->path_rec, GFP_KERNEL);
2513         if (!route->path_rec) {
2514                 ret = -ENOMEM;
2515                 goto err1;
2516         }
2517
2518         route->num_paths = 1;
2519
2520         if (addr->dev_addr.bound_dev_if) {
2521                 unsigned long supported_gids;
2522
2523                 ndev = dev_get_by_index(&init_net, addr->dev_addr.bound_dev_if);
2524                 if (!ndev) {
2525                         ret = -ENODEV;
2526                         goto err2;
2527                 }
2528
2529                 if (ndev->flags & IFF_LOOPBACK) {
2530                         dev_put(ndev);
2531                         if (!id_priv->id.device->get_netdev) {
2532                                 ret = -EOPNOTSUPP;
2533                                 goto err2;
2534                         }
2535
2536                         ndev = id_priv->id.device->get_netdev(id_priv->id.device,
2537                                                               id_priv->id.port_num);
2538                         if (!ndev) {
2539                                 ret = -ENODEV;
2540                                 goto err2;
2541                         }
2542                 }
2543
2544                 route->path_rec->net = &init_net;
2545                 route->path_rec->ifindex = ndev->ifindex;
2546                 supported_gids = roce_gid_type_mask_support(id_priv->id.device,
2547                                                             id_priv->id.port_num);
2548                 route->path_rec->gid_type =
2549                         cma_route_gid_type(addr->dev_addr.network,
2550                                            supported_gids,
2551                                            id_priv->gid_type);
2552         }
2553         if (!ndev) {
2554                 ret = -ENODEV;
2555                 goto err2;
2556         }
2557
2558         memcpy(route->path_rec->dmac, addr->dev_addr.dst_dev_addr, ETH_ALEN);
2559
2560         rdma_ip2gid((struct sockaddr *)&id_priv->id.route.addr.src_addr,
2561                     &route->path_rec->sgid);
2562         rdma_ip2gid((struct sockaddr *)&id_priv->id.route.addr.dst_addr,
2563                     &route->path_rec->dgid);
2564
2565         /* Use the hint from IP Stack to select GID Type */
2566         if (route->path_rec->gid_type < ib_network_to_gid_type(addr->dev_addr.network))
2567                 route->path_rec->gid_type = ib_network_to_gid_type(addr->dev_addr.network);
2568         if (((struct sockaddr *)&id_priv->id.route.addr.dst_addr)->sa_family != AF_IB)
2569                 /* TODO: get the hoplimit from the inet/inet6 device */
2570                 route->path_rec->hop_limit = addr->dev_addr.hoplimit;
2571         else
2572                 route->path_rec->hop_limit = 1;
2573         route->path_rec->reversible = 1;
2574         route->path_rec->pkey = cpu_to_be16(0xffff);
2575         route->path_rec->mtu_selector = IB_SA_EQ;
2576         route->path_rec->sl = iboe_tos_to_sl(ndev, id_priv->tos);
2577         route->path_rec->mtu = iboe_get_mtu(ndev->mtu);
2578         route->path_rec->rate_selector = IB_SA_EQ;
2579         route->path_rec->rate = iboe_get_rate(ndev);
2580         dev_put(ndev);
2581         route->path_rec->packet_life_time_selector = IB_SA_EQ;
2582         route->path_rec->packet_life_time = CMA_IBOE_PACKET_LIFETIME;
2583         if (!route->path_rec->mtu) {
2584                 ret = -EINVAL;
2585                 goto err2;
2586         }
2587
2588         work->old_state = RDMA_CM_ROUTE_QUERY;
2589         work->new_state = RDMA_CM_ROUTE_RESOLVED;
2590         work->event.event = RDMA_CM_EVENT_ROUTE_RESOLVED;
2591         work->event.status = 0;
2592
2593         queue_work(cma_wq, &work->work);
2594
2595         return 0;
2596
2597 err2:
2598         kfree(route->path_rec);
2599         route->path_rec = NULL;
2600 err1:
2601         kfree(work);
2602         return ret;
2603 }
2604
2605 int rdma_resolve_route(struct rdma_cm_id *id, int timeout_ms)
2606 {
2607         struct rdma_id_private *id_priv;
2608         int ret;
2609
2610         id_priv = container_of(id, struct rdma_id_private, id);
2611         if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_RESOLVED, RDMA_CM_ROUTE_QUERY))
2612                 return -EINVAL;
2613
2614         atomic_inc(&id_priv->refcount);
2615         if (rdma_cap_ib_sa(id->device, id->port_num))
2616                 ret = cma_resolve_ib_route(id_priv, timeout_ms);
2617         else if (rdma_protocol_roce(id->device, id->port_num))
2618                 ret = cma_resolve_iboe_route(id_priv);
2619         else if (rdma_protocol_iwarp(id->device, id->port_num))
2620                 ret = cma_resolve_iw_route(id_priv, timeout_ms);
2621         else
2622                 ret = -ENOSYS;
2623
2624         if (ret)
2625                 goto err;
2626
2627         return 0;
2628 err:
2629         cma_comp_exch(id_priv, RDMA_CM_ROUTE_QUERY, RDMA_CM_ADDR_RESOLVED);
2630         cma_deref_id(id_priv);
2631         return ret;
2632 }
2633 EXPORT_SYMBOL(rdma_resolve_route);
2634
2635 static void cma_set_loopback(struct sockaddr *addr)
2636 {
2637         switch (addr->sa_family) {
2638         case AF_INET:
2639                 ((struct sockaddr_in *) addr)->sin_addr.s_addr = htonl(INADDR_LOOPBACK);
2640                 break;
2641         case AF_INET6:
2642                 ipv6_addr_set(&((struct sockaddr_in6 *) addr)->sin6_addr,
2643                               0, 0, 0, htonl(1));
2644                 break;
2645         default:
2646                 ib_addr_set(&((struct sockaddr_ib *) addr)->sib_addr,
2647                             0, 0, 0, htonl(1));
2648                 break;
2649         }
2650 }
2651
2652 static int cma_bind_loopback(struct rdma_id_private *id_priv)
2653 {
2654         struct cma_device *cma_dev, *cur_dev;
2655         struct ib_port_attr port_attr;
2656         union ib_gid gid;
2657         u16 pkey;
2658         int ret;
2659         u8 p;
2660
2661         cma_dev = NULL;
2662         mutex_lock(&lock);
2663         list_for_each_entry(cur_dev, &dev_list, list) {
2664                 if (cma_family(id_priv) == AF_IB &&
2665                     !rdma_cap_ib_cm(cur_dev->device, 1))
2666                         continue;
2667
2668                 if (!cma_dev)
2669                         cma_dev = cur_dev;
2670
2671                 for (p = 1; p <= cur_dev->device->phys_port_cnt; ++p) {
2672                         if (!ib_query_port(cur_dev->device, p, &port_attr) &&
2673                             port_attr.state == IB_PORT_ACTIVE) {
2674                                 cma_dev = cur_dev;
2675                                 goto port_found;
2676                         }
2677                 }
2678         }
2679
2680         if (!cma_dev) {
2681                 ret = -ENODEV;
2682                 goto out;
2683         }
2684
2685         p = 1;
2686
2687 port_found:
2688         ret = ib_get_cached_gid(cma_dev->device, p, 0, &gid, NULL);
2689         if (ret)
2690                 goto out;
2691
2692         ret = ib_get_cached_pkey(cma_dev->device, p, 0, &pkey);
2693         if (ret)
2694                 goto out;
2695
2696         id_priv->id.route.addr.dev_addr.dev_type =
2697                 (rdma_protocol_ib(cma_dev->device, p)) ?
2698                 ARPHRD_INFINIBAND : ARPHRD_ETHER;
2699
2700         rdma_addr_set_sgid(&id_priv->id.route.addr.dev_addr, &gid);
2701         ib_addr_set_pkey(&id_priv->id.route.addr.dev_addr, pkey);
2702         id_priv->id.port_num = p;
2703         cma_attach_to_dev(id_priv, cma_dev);
2704         cma_set_loopback(cma_src_addr(id_priv));
2705 out:
2706         mutex_unlock(&lock);
2707         return ret;
2708 }
2709
2710 static void addr_handler(int status, struct sockaddr *src_addr,
2711                          struct rdma_dev_addr *dev_addr, void *context)
2712 {
2713         struct rdma_id_private *id_priv = context;
2714         struct rdma_cm_event event;
2715
2716         memset(&event, 0, sizeof event);
2717         mutex_lock(&id_priv->handler_mutex);
2718         if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_QUERY,
2719                            RDMA_CM_ADDR_RESOLVED))
2720                 goto out;
2721
2722         memcpy(cma_src_addr(id_priv), src_addr, rdma_addr_size(src_addr));
2723         if (!status && !id_priv->cma_dev)
2724                 status = cma_acquire_dev(id_priv, NULL);
2725
2726         if (status) {
2727                 if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_RESOLVED,
2728                                    RDMA_CM_ADDR_BOUND))
2729                         goto out;
2730                 event.event = RDMA_CM_EVENT_ADDR_ERROR;
2731                 event.status = status;
2732         } else
2733                 event.event = RDMA_CM_EVENT_ADDR_RESOLVED;
2734
2735         if (id_priv->id.event_handler(&id_priv->id, &event)) {
2736                 cma_exch(id_priv, RDMA_CM_DESTROYING);
2737                 mutex_unlock(&id_priv->handler_mutex);
2738                 cma_deref_id(id_priv);
2739                 rdma_destroy_id(&id_priv->id);
2740                 return;
2741         }
2742 out:
2743         mutex_unlock(&id_priv->handler_mutex);
2744         cma_deref_id(id_priv);
2745 }
2746
2747 static int cma_resolve_loopback(struct rdma_id_private *id_priv)
2748 {
2749         struct cma_work *work;
2750         union ib_gid gid;
2751         int ret;
2752
2753         work = kzalloc(sizeof *work, GFP_KERNEL);
2754         if (!work)
2755                 return -ENOMEM;
2756
2757         if (!id_priv->cma_dev) {
2758                 ret = cma_bind_loopback(id_priv);
2759                 if (ret)
2760                         goto err;
2761         }
2762
2763         rdma_addr_get_sgid(&id_priv->id.route.addr.dev_addr, &gid);
2764         rdma_addr_set_dgid(&id_priv->id.route.addr.dev_addr, &gid);
2765
2766         work->id = id_priv;
2767         INIT_WORK(&work->work, cma_work_handler);
2768         work->old_state = RDMA_CM_ADDR_QUERY;
2769         work->new_state = RDMA_CM_ADDR_RESOLVED;
2770         work->event.event = RDMA_CM_EVENT_ADDR_RESOLVED;
2771         queue_work(cma_wq, &work->work);
2772         return 0;
2773 err:
2774         kfree(work);
2775         return ret;
2776 }
2777
2778 static int cma_resolve_ib_addr(struct rdma_id_private *id_priv)
2779 {
2780         struct cma_work *work;
2781         int ret;
2782
2783         work = kzalloc(sizeof *work, GFP_KERNEL);
2784         if (!work)
2785                 return -ENOMEM;
2786
2787         if (!id_priv->cma_dev) {
2788                 ret = cma_resolve_ib_dev(id_priv);
2789                 if (ret)
2790                         goto err;
2791         }
2792
2793         rdma_addr_set_dgid(&id_priv->id.route.addr.dev_addr, (union ib_gid *)
2794                 &(((struct sockaddr_ib *) &id_priv->id.route.addr.dst_addr)->sib_addr));
2795
2796         work->id = id_priv;
2797         INIT_WORK(&work->work, cma_work_handler);
2798         work->old_state = RDMA_CM_ADDR_QUERY;
2799         work->new_state = RDMA_CM_ADDR_RESOLVED;
2800         work->event.event = RDMA_CM_EVENT_ADDR_RESOLVED;
2801         queue_work(cma_wq, &work->work);
2802         return 0;
2803 err:
2804         kfree(work);
2805         return ret;
2806 }
2807
2808 static int cma_bind_addr(struct rdma_cm_id *id, struct sockaddr *src_addr,
2809                          struct sockaddr *dst_addr)
2810 {
2811         if (!src_addr || !src_addr->sa_family) {
2812                 src_addr = (struct sockaddr *) &id->route.addr.src_addr;
2813                 src_addr->sa_family = dst_addr->sa_family;
2814                 if (dst_addr->sa_family == AF_INET6) {
2815                         struct sockaddr_in6 *src_addr6 = (struct sockaddr_in6 *) src_addr;
2816                         struct sockaddr_in6 *dst_addr6 = (struct sockaddr_in6 *) dst_addr;
2817                         src_addr6->sin6_scope_id = dst_addr6->sin6_scope_id;
2818                         if (ipv6_addr_type(&dst_addr6->sin6_addr) & IPV6_ADDR_LINKLOCAL)
2819                                 id->route.addr.dev_addr.bound_dev_if = dst_addr6->sin6_scope_id;
2820                 } else if (dst_addr->sa_family == AF_IB) {
2821                         ((struct sockaddr_ib *) src_addr)->sib_pkey =
2822                                 ((struct sockaddr_ib *) dst_addr)->sib_pkey;
2823                 }
2824         }
2825         return rdma_bind_addr(id, src_addr);
2826 }
2827
2828 int rdma_resolve_addr(struct rdma_cm_id *id, struct sockaddr *src_addr,
2829                       struct sockaddr *dst_addr, int timeout_ms)
2830 {
2831         struct rdma_id_private *id_priv;
2832         int ret;
2833
2834         id_priv = container_of(id, struct rdma_id_private, id);
2835         if (id_priv->state == RDMA_CM_IDLE) {
2836                 ret = cma_bind_addr(id, src_addr, dst_addr);
2837                 if (ret)
2838                         return ret;
2839         }
2840
2841         if (cma_family(id_priv) != dst_addr->sa_family)
2842                 return -EINVAL;
2843
2844         if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_BOUND, RDMA_CM_ADDR_QUERY))
2845                 return -EINVAL;
2846
2847         atomic_inc(&id_priv->refcount);
2848         memcpy(cma_dst_addr(id_priv), dst_addr, rdma_addr_size(dst_addr));
2849         if (cma_any_addr(dst_addr)) {
2850                 ret = cma_resolve_loopback(id_priv);
2851         } else {
2852                 if (dst_addr->sa_family == AF_IB) {
2853                         ret = cma_resolve_ib_addr(id_priv);
2854                 } else {
2855                         ret = rdma_resolve_ip(&addr_client, cma_src_addr(id_priv),
2856                                               dst_addr, &id->route.addr.dev_addr,
2857                                               timeout_ms, addr_handler, id_priv);
2858                 }
2859         }
2860         if (ret)
2861                 goto err;
2862
2863         return 0;
2864 err:
2865         cma_comp_exch(id_priv, RDMA_CM_ADDR_QUERY, RDMA_CM_ADDR_BOUND);
2866         cma_deref_id(id_priv);
2867         return ret;
2868 }
2869 EXPORT_SYMBOL(rdma_resolve_addr);
2870
2871 int rdma_set_reuseaddr(struct rdma_cm_id *id, int reuse)
2872 {
2873         struct rdma_id_private *id_priv;
2874         unsigned long flags;
2875         int ret;
2876
2877         id_priv = container_of(id, struct rdma_id_private, id);
2878         spin_lock_irqsave(&id_priv->lock, flags);
2879         if (reuse || id_priv->state == RDMA_CM_IDLE) {
2880                 id_priv->reuseaddr = reuse;
2881                 ret = 0;
2882         } else {
2883                 ret = -EINVAL;
2884         }
2885         spin_unlock_irqrestore(&id_priv->lock, flags);
2886         return ret;
2887 }
2888 EXPORT_SYMBOL(rdma_set_reuseaddr);
2889
2890 int rdma_set_afonly(struct rdma_cm_id *id, int afonly)
2891 {
2892         struct rdma_id_private *id_priv;
2893         unsigned long flags;
2894         int ret;
2895
2896         id_priv = container_of(id, struct rdma_id_private, id);
2897         spin_lock_irqsave(&id_priv->lock, flags);
2898         if (id_priv->state == RDMA_CM_IDLE || id_priv->state == RDMA_CM_ADDR_BOUND) {
2899                 id_priv->options |= (1 << CMA_OPTION_AFONLY);
2900                 id_priv->afonly = afonly;
2901                 ret = 0;
2902         } else {
2903                 ret = -EINVAL;
2904         }
2905         spin_unlock_irqrestore(&id_priv->lock, flags);
2906         return ret;
2907 }
2908 EXPORT_SYMBOL(rdma_set_afonly);
2909
2910 static void cma_bind_port(struct rdma_bind_list *bind_list,
2911                           struct rdma_id_private *id_priv)
2912 {
2913         struct sockaddr *addr;
2914         struct sockaddr_ib *sib;
2915         u64 sid, mask;
2916         __be16 port;
2917
2918         addr = cma_src_addr(id_priv);
2919         port = htons(bind_list->port);
2920
2921         switch (addr->sa_family) {
2922         case AF_INET:
2923                 ((struct sockaddr_in *) addr)->sin_port = port;
2924                 break;
2925         case AF_INET6:
2926                 ((struct sockaddr_in6 *) addr)->sin6_port = port;
2927                 break;
2928         case AF_IB:
2929                 sib = (struct sockaddr_ib *) addr;
2930                 sid = be64_to_cpu(sib->sib_sid);
2931                 mask = be64_to_cpu(sib->sib_sid_mask);
2932                 sib->sib_sid = cpu_to_be64((sid & mask) | (u64) ntohs(port));
2933                 sib->sib_sid_mask = cpu_to_be64(~0ULL);
2934                 break;
2935         }
2936         id_priv->bind_list = bind_list;
2937         hlist_add_head(&id_priv->node, &bind_list->owners);
2938 }
2939
2940 static int cma_alloc_port(enum rdma_port_space ps,
2941                           struct rdma_id_private *id_priv, unsigned short snum)
2942 {
2943         struct rdma_bind_list *bind_list;
2944         int ret;
2945
2946         bind_list = kzalloc(sizeof *bind_list, GFP_KERNEL);
2947         if (!bind_list)
2948                 return -ENOMEM;
2949
2950         ret = cma_ps_alloc(id_priv->id.route.addr.dev_addr.net, ps, bind_list,
2951                            snum);
2952         if (ret < 0)
2953                 goto err;
2954
2955         bind_list->ps = ps;
2956         bind_list->port = (unsigned short)ret;
2957         cma_bind_port(bind_list, id_priv);
2958         return 0;
2959 err:
2960         kfree(bind_list);
2961         return ret == -ENOSPC ? -EADDRNOTAVAIL : ret;
2962 }
2963
2964 static int cma_alloc_any_port(enum rdma_port_space ps,
2965                               struct rdma_id_private *id_priv)
2966 {
2967         static unsigned int last_used_port;
2968         int low, high, remaining;
2969         unsigned int rover;
2970         struct net *net = id_priv->id.route.addr.dev_addr.net;
2971
2972         inet_get_local_port_range(net, &low, &high);
2973         remaining = (high - low) + 1;
2974         rover = prandom_u32() % remaining + low;
2975 retry:
2976         if (last_used_port != rover &&
2977             !cma_ps_find(net, ps, (unsigned short)rover)) {
2978                 int ret = cma_alloc_port(ps, id_priv, rover);
2979                 /*
2980                  * Remember previously used port number in order to avoid
2981                  * re-using same port immediately after it is closed.
2982                  */
2983                 if (!ret)
2984                         last_used_port = rover;
2985                 if (ret != -EADDRNOTAVAIL)
2986                         return ret;
2987         }
2988         if (--remaining) {
2989                 rover++;
2990                 if ((rover < low) || (rover > high))
2991                         rover = low;
2992                 goto retry;
2993         }
2994         return -EADDRNOTAVAIL;
2995 }
2996
2997 /*
2998  * Check that the requested port is available.  This is called when trying to
2999  * bind to a specific port, or when trying to listen on a bound port.  In
3000  * the latter case, the provided id_priv may already be on the bind_list, but
3001  * we still need to check that it's okay to start listening.
3002  */
3003 static int cma_check_port(struct rdma_bind_list *bind_list,
3004                           struct rdma_id_private *id_priv, uint8_t reuseaddr)
3005 {
3006         struct rdma_id_private *cur_id;
3007         struct sockaddr *addr, *cur_addr;
3008
3009         addr = cma_src_addr(id_priv);
3010         hlist_for_each_entry(cur_id, &bind_list->owners, node) {
3011                 if (id_priv == cur_id)
3012                         continue;
3013
3014                 if ((cur_id->state != RDMA_CM_LISTEN) && reuseaddr &&
3015                     cur_id->reuseaddr)
3016                         continue;
3017
3018                 cur_addr = cma_src_addr(cur_id);
3019                 if (id_priv->afonly && cur_id->afonly &&
3020                     (addr->sa_family != cur_addr->sa_family))
3021                         continue;
3022
3023                 if (cma_any_addr(addr) || cma_any_addr(cur_addr))
3024                         return -EADDRNOTAVAIL;
3025
3026                 if (!cma_addr_cmp(addr, cur_addr))
3027                         return -EADDRINUSE;
3028         }
3029         return 0;
3030 }
3031
3032 static int cma_use_port(enum rdma_port_space ps,
3033                         struct rdma_id_private *id_priv)
3034 {
3035         struct rdma_bind_list *bind_list;
3036         unsigned short snum;
3037         int ret;
3038
3039         snum = ntohs(cma_port(cma_src_addr(id_priv)));
3040         if (snum < PROT_SOCK && !capable(CAP_NET_BIND_SERVICE))
3041                 return -EACCES;
3042
3043         bind_list = cma_ps_find(id_priv->id.route.addr.dev_addr.net, ps, snum);
3044         if (!bind_list) {
3045                 ret = cma_alloc_port(ps, id_priv, snum);
3046         } else {
3047                 ret = cma_check_port(bind_list, id_priv, id_priv->reuseaddr);
3048                 if (!ret)
3049                         cma_bind_port(bind_list, id_priv);
3050         }
3051         return ret;
3052 }
3053
3054 static int cma_bind_listen(struct rdma_id_private *id_priv)
3055 {
3056         struct rdma_bind_list *bind_list = id_priv->bind_list;
3057         int ret = 0;
3058
3059         mutex_lock(&lock);
3060         if (bind_list->owners.first->next)
3061                 ret = cma_check_port(bind_list, id_priv, 0);
3062         mutex_unlock(&lock);
3063         return ret;
3064 }
3065
3066 static enum rdma_port_space cma_select_inet_ps(
3067                 struct rdma_id_private *id_priv)
3068 {
3069         switch (id_priv->id.ps) {
3070         case RDMA_PS_TCP:
3071         case RDMA_PS_UDP:
3072         case RDMA_PS_IPOIB:
3073         case RDMA_PS_IB:
3074                 return id_priv->id.ps;
3075         default:
3076
3077                 return 0;
3078         }
3079 }
3080
3081 static enum rdma_port_space cma_select_ib_ps(struct rdma_id_private *id_priv)
3082 {
3083         enum rdma_port_space ps = 0;
3084         struct sockaddr_ib *sib;
3085         u64 sid_ps, mask, sid;
3086
3087         sib = (struct sockaddr_ib *) cma_src_addr(id_priv);
3088         mask = be64_to_cpu(sib->sib_sid_mask) & RDMA_IB_IP_PS_MASK;
3089         sid = be64_to_cpu(sib->sib_sid) & mask;
3090
3091         if ((id_priv->id.ps == RDMA_PS_IB) && (sid == (RDMA_IB_IP_PS_IB & mask))) {
3092                 sid_ps = RDMA_IB_IP_PS_IB;
3093                 ps = RDMA_PS_IB;
3094         } else if (((id_priv->id.ps == RDMA_PS_IB) || (id_priv->id.ps == RDMA_PS_TCP)) &&
3095                    (sid == (RDMA_IB_IP_PS_TCP & mask))) {
3096                 sid_ps = RDMA_IB_IP_PS_TCP;
3097                 ps = RDMA_PS_TCP;
3098         } else if (((id_priv->id.ps == RDMA_PS_IB) || (id_priv->id.ps == RDMA_PS_UDP)) &&
3099                    (sid == (RDMA_IB_IP_PS_UDP & mask))) {
3100                 sid_ps = RDMA_IB_IP_PS_UDP;
3101                 ps = RDMA_PS_UDP;
3102         }
3103
3104         if (ps) {
3105                 sib->sib_sid = cpu_to_be64(sid_ps | ntohs(cma_port((struct sockaddr *) sib)));
3106                 sib->sib_sid_mask = cpu_to_be64(RDMA_IB_IP_PS_MASK |
3107                                                 be64_to_cpu(sib->sib_sid_mask));
3108         }
3109         return ps;
3110 }
3111
3112 static int cma_get_port(struct rdma_id_private *id_priv)
3113 {
3114         enum rdma_port_space ps;
3115         int ret;
3116
3117         if (cma_family(id_priv) != AF_IB)
3118                 ps = cma_select_inet_ps(id_priv);
3119         else
3120                 ps = cma_select_ib_ps(id_priv);
3121         if (!ps)
3122                 return -EPROTONOSUPPORT;
3123
3124         mutex_lock(&lock);
3125         if (cma_any_port(cma_src_addr(id_priv)))
3126                 ret = cma_alloc_any_port(ps, id_priv);
3127         else
3128                 ret = cma_use_port(ps, id_priv);
3129         mutex_unlock(&lock);
3130
3131         return ret;
3132 }
3133
3134 static int cma_check_linklocal(struct rdma_dev_addr *dev_addr,
3135                                struct sockaddr *addr)
3136 {
3137 #if IS_ENABLED(CONFIG_IPV6)
3138         struct sockaddr_in6 *sin6;
3139
3140         if (addr->sa_family != AF_INET6)
3141                 return 0;
3142
3143         sin6 = (struct sockaddr_in6 *) addr;
3144
3145         if (!(ipv6_addr_type(&sin6->sin6_addr) & IPV6_ADDR_LINKLOCAL))
3146                 return 0;
3147
3148         if (!sin6->sin6_scope_id)
3149                         return -EINVAL;
3150
3151         dev_addr->bound_dev_if = sin6->sin6_scope_id;
3152 #endif
3153         return 0;
3154 }
3155
3156 int rdma_listen(struct rdma_cm_id *id, int backlog)
3157 {
3158         struct rdma_id_private *id_priv;
3159         int ret;
3160
3161         id_priv = container_of(id, struct rdma_id_private, id);
3162         if (id_priv->state == RDMA_CM_IDLE) {
3163                 id->route.addr.src_addr.ss_family = AF_INET;
3164                 ret = rdma_bind_addr(id, cma_src_addr(id_priv));
3165                 if (ret)
3166                         return ret;
3167         }
3168
3169         if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_BOUND, RDMA_CM_LISTEN))
3170                 return -EINVAL;
3171
3172         if (id_priv->reuseaddr) {
3173                 ret = cma_bind_listen(id_priv);
3174                 if (ret)
3175                         goto err;
3176         }
3177
3178         id_priv->backlog = backlog;
3179         if (id->device) {
3180                 if (rdma_cap_ib_cm(id->device, 1)) {
3181                         ret = cma_ib_listen(id_priv);
3182                         if (ret)
3183                                 goto err;
3184                 } else if (rdma_cap_iw_cm(id->device, 1)) {
3185                         ret = cma_iw_listen(id_priv, backlog);
3186                         if (ret)
3187                                 goto err;
3188                 } else {
3189                         ret = -ENOSYS;
3190                         goto err;
3191                 }
3192         } else
3193                 cma_listen_on_all(id_priv);
3194
3195         return 0;
3196 err:
3197         id_priv->backlog = 0;
3198         cma_comp_exch(id_priv, RDMA_CM_LISTEN, RDMA_CM_ADDR_BOUND);
3199         return ret;
3200 }
3201 EXPORT_SYMBOL(rdma_listen);
3202
3203 int rdma_bind_addr(struct rdma_cm_id *id, struct sockaddr *addr)
3204 {
3205         struct rdma_id_private *id_priv;
3206         int ret;
3207
3208         if (addr->sa_family != AF_INET && addr->sa_family != AF_INET6 &&
3209             addr->sa_family != AF_IB)
3210                 return -EAFNOSUPPORT;
3211
3212         id_priv = container_of(id, struct rdma_id_private, id);
3213         if (!cma_comp_exch(id_priv, RDMA_CM_IDLE, RDMA_CM_ADDR_BOUND))
3214                 return -EINVAL;
3215
3216         ret = cma_check_linklocal(&id->route.addr.dev_addr, addr);
3217         if (ret)
3218                 goto err1;
3219
3220         memcpy(cma_src_addr(id_priv), addr, rdma_addr_size(addr));
3221         if (!cma_any_addr(addr)) {
3222                 ret = cma_translate_addr(addr, &id->route.addr.dev_addr);
3223                 if (ret)
3224                         goto err1;
3225
3226                 ret = cma_acquire_dev(id_priv, NULL);
3227                 if (ret)
3228                         goto err1;
3229         }
3230
3231         if (!(id_priv->options & (1 << CMA_OPTION_AFONLY))) {
3232                 if (addr->sa_family == AF_INET)
3233                         id_priv->afonly = 1;
3234 #if IS_ENABLED(CONFIG_IPV6)
3235                 else if (addr->sa_family == AF_INET6) {
3236                         struct net *net = id_priv->id.route.addr.dev_addr.net;
3237
3238                         id_priv->afonly = net->ipv6.sysctl.bindv6only;
3239                 }
3240 #endif
3241         }
3242         ret = cma_get_port(id_priv);
3243         if (ret)
3244                 goto err2;
3245
3246         return 0;
3247 err2:
3248         if (id_priv->cma_dev)
3249                 cma_release_dev(id_priv);
3250 err1:
3251         cma_comp_exch(id_priv, RDMA_CM_ADDR_BOUND, RDMA_CM_IDLE);
3252         return ret;
3253 }
3254 EXPORT_SYMBOL(rdma_bind_addr);
3255
3256 static int cma_format_hdr(void *hdr, struct rdma_id_private *id_priv)
3257 {
3258         struct cma_hdr *cma_hdr;
3259
3260         cma_hdr = hdr;
3261         cma_hdr->cma_version = CMA_VERSION;
3262         if (cma_family(id_priv) == AF_INET) {
3263                 struct sockaddr_in *src4, *dst4;
3264
3265                 src4 = (struct sockaddr_in *) cma_src_addr(id_priv);
3266                 dst4 = (struct sockaddr_in *) cma_dst_addr(id_priv);
3267
3268                 cma_set_ip_ver(cma_hdr, 4);
3269                 cma_hdr->src_addr.ip4.addr = src4->sin_addr.s_addr;
3270                 cma_hdr->dst_addr.ip4.addr = dst4->sin_addr.s_addr;
3271                 cma_hdr->port = src4->sin_port;
3272         } else if (cma_family(id_priv) == AF_INET6) {
3273                 struct sockaddr_in6 *src6, *dst6;
3274
3275                 src6 = (struct sockaddr_in6 *) cma_src_addr(id_priv);
3276                 dst6 = (struct sockaddr_in6 *) cma_dst_addr(id_priv);
3277
3278                 cma_set_ip_ver(cma_hdr, 6);
3279                 cma_hdr->src_addr.ip6 = src6->sin6_addr;
3280                 cma_hdr->dst_addr.ip6 = dst6->sin6_addr;
3281                 cma_hdr->port = src6->sin6_port;
3282         }
3283         return 0;
3284 }
3285
3286 static int cma_sidr_rep_handler(struct ib_cm_id *cm_id,
3287                                 struct ib_cm_event *ib_event)
3288 {
3289         struct rdma_id_private *id_priv = cm_id->context;
3290         struct rdma_cm_event event;
3291         struct ib_cm_sidr_rep_event_param *rep = &ib_event->param.sidr_rep_rcvd;
3292         int ret = 0;
3293
3294         mutex_lock(&id_priv->handler_mutex);
3295         if (id_priv->state != RDMA_CM_CONNECT)
3296                 goto out;
3297
3298         memset(&event, 0, sizeof event);
3299         switch (ib_event->event) {
3300         case IB_CM_SIDR_REQ_ERROR:
3301                 event.event = RDMA_CM_EVENT_UNREACHABLE;
3302                 event.status = -ETIMEDOUT;
3303                 break;
3304         case IB_CM_SIDR_REP_RECEIVED:
3305                 event.param.ud.private_data = ib_event->private_data;
3306                 event.param.ud.private_data_len = IB_CM_SIDR_REP_PRIVATE_DATA_SIZE;
3307                 if (rep->status != IB_SIDR_SUCCESS) {
3308                         event.event = RDMA_CM_EVENT_UNREACHABLE;
3309                         event.status = ib_event->param.sidr_rep_rcvd.status;
3310                         break;
3311                 }
3312                 ret = cma_set_qkey(id_priv, rep->qkey);
3313                 if (ret) {
3314                         event.event = RDMA_CM_EVENT_ADDR_ERROR;
3315                         event.status = ret;
3316                         break;
3317                 }
3318                 ib_init_ah_from_path(id_priv->id.device, id_priv->id.port_num,
3319                                      id_priv->id.route.path_rec,
3320                                      &event.param.ud.ah_attr);
3321                 event.param.ud.qp_num = rep->qpn;
3322                 event.param.ud.qkey = rep->qkey;
3323                 event.event = RDMA_CM_EVENT_ESTABLISHED;
3324                 event.status = 0;
3325                 break;
3326         default:
3327                 pr_err("RDMA CMA: unexpected IB CM event: %d\n",
3328                        ib_event->event);
3329                 goto out;
3330         }
3331
3332         ret = id_priv->id.event_handler(&id_priv->id, &event);
3333         if (ret) {
3334                 /* Destroy the CM ID by returning a non-zero value. */
3335                 id_priv->cm_id.ib = NULL;
3336                 cma_exch(id_priv, RDMA_CM_DESTROYING);
3337                 mutex_unlock(&id_priv->handler_mutex);
3338                 rdma_destroy_id(&id_priv->id);
3339                 return ret;
3340         }
3341 out:
3342         mutex_unlock(&id_priv->handler_mutex);
3343         return ret;
3344 }
3345
3346 static int cma_resolve_ib_udp(struct rdma_id_private *id_priv,
3347                               struct rdma_conn_param *conn_param)
3348 {
3349         struct ib_cm_sidr_req_param req;
3350         struct ib_cm_id *id;
3351         void *private_data;
3352         int offset, ret;
3353
3354         memset(&req, 0, sizeof req);
3355         offset = cma_user_data_offset(id_priv);
3356         req.private_data_len = offset + conn_param->private_data_len;
3357         if (req.private_data_len < conn_param->private_data_len)
3358                 return -EINVAL;
3359
3360         if (req.private_data_len) {
3361                 private_data = kzalloc(req.private_data_len, GFP_ATOMIC);
3362                 if (!private_data)
3363                         return -ENOMEM;
3364         } else {
3365                 private_data = NULL;
3366         }
3367
3368         if (conn_param->private_data && conn_param->private_data_len)
3369                 memcpy(private_data + offset, conn_param->private_data,
3370                        conn_param->private_data_len);
3371
3372         if (private_data) {
3373                 ret = cma_format_hdr(private_data, id_priv);
3374                 if (ret)
3375                         goto out;
3376                 req.private_data = private_data;
3377         }
3378
3379         id = ib_create_cm_id(id_priv->id.device, cma_sidr_rep_handler,
3380                              id_priv);
3381         if (IS_ERR(id)) {
3382                 ret = PTR_ERR(id);
3383                 goto out;
3384         }
3385         id_priv->cm_id.ib = id;
3386
3387         req.path = id_priv->id.route.path_rec;
3388         req.service_id = rdma_get_service_id(&id_priv->id, cma_dst_addr(id_priv));
3389         req.timeout_ms = 1 << (CMA_CM_RESPONSE_TIMEOUT - 8);
3390         req.max_cm_retries = CMA_MAX_CM_RETRIES;
3391
3392         ret = ib_send_cm_sidr_req(id_priv->cm_id.ib, &req);
3393         if (ret) {
3394                 ib_destroy_cm_id(id_priv->cm_id.ib);
3395                 id_priv->cm_id.ib = NULL;
3396         }
3397 out:
3398         kfree(private_data);
3399         return ret;
3400 }
3401
3402 static int cma_connect_ib(struct rdma_id_private *id_priv,
3403                           struct rdma_conn_param *conn_param)
3404 {
3405         struct ib_cm_req_param req;
3406         struct rdma_route *route;
3407         void *private_data;
3408         struct ib_cm_id *id;
3409         int offset, ret;
3410
3411         memset(&req, 0, sizeof req);
3412         offset = cma_user_data_offset(id_priv);
3413         req.private_data_len = offset + conn_param->private_data_len;
3414         if (req.private_data_len < conn_param->private_data_len)
3415                 return -EINVAL;
3416
3417         if (req.private_data_len) {
3418                 private_data = kzalloc(req.private_data_len, GFP_ATOMIC);
3419                 if (!private_data)
3420                         return -ENOMEM;
3421         } else {
3422                 private_data = NULL;
3423         }
3424
3425         if (conn_param->private_data && conn_param->private_data_len)
3426                 memcpy(private_data + offset, conn_param->private_data,
3427                        conn_param->private_data_len);
3428
3429         id = ib_create_cm_id(id_priv->id.device, cma_ib_handler, id_priv);
3430         if (IS_ERR(id)) {
3431                 ret = PTR_ERR(id);
3432                 goto out;
3433         }
3434         id_priv->cm_id.ib = id;
3435
3436         route = &id_priv->id.route;
3437         if (private_data) {
3438                 ret = cma_format_hdr(private_data, id_priv);
3439                 if (ret)
3440                         goto out;
3441                 req.private_data = private_data;
3442         }
3443
3444         req.primary_path = &route->path_rec[0];
3445         if (route->num_paths == 2)
3446                 req.alternate_path = &route->path_rec[1];
3447
3448         req.service_id = rdma_get_service_id(&id_priv->id, cma_dst_addr(id_priv));
3449         req.qp_num = id_priv->qp_num;
3450         req.qp_type = id_priv->id.qp_type;
3451         req.starting_psn = id_priv->seq_num;
3452         req.responder_resources = conn_param->responder_resources;
3453         req.initiator_depth = conn_param->initiator_depth;
3454         req.flow_control = conn_param->flow_control;
3455         req.retry_count = min_t(u8, 7, conn_param->retry_count);
3456         req.rnr_retry_count = min_t(u8, 7, conn_param->rnr_retry_count);
3457         req.remote_cm_response_timeout = CMA_CM_RESPONSE_TIMEOUT;
3458         req.local_cm_response_timeout = CMA_CM_RESPONSE_TIMEOUT;
3459         req.max_cm_retries = CMA_MAX_CM_RETRIES;
3460         req.srq = id_priv->srq ? 1 : 0;
3461
3462         ret = ib_send_cm_req(id_priv->cm_id.ib, &req);
3463 out:
3464         if (ret && !IS_ERR(id)) {
3465                 ib_destroy_cm_id(id);
3466                 id_priv->cm_id.ib = NULL;
3467         }
3468
3469         kfree(private_data);
3470         return ret;
3471 }
3472
3473 static int cma_connect_iw(struct rdma_id_private *id_priv,
3474                           struct rdma_conn_param *conn_param)
3475 {
3476         struct iw_cm_id *cm_id;
3477         int ret;
3478         struct iw_cm_conn_param iw_param;
3479
3480         cm_id = iw_create_cm_id(id_priv->id.device, cma_iw_handler, id_priv);
3481         if (IS_ERR(cm_id))
3482                 return PTR_ERR(cm_id);
3483
3484         cm_id->tos = id_priv->tos;
3485         id_priv->cm_id.iw = cm_id;
3486
3487         memcpy(&cm_id->local_addr, cma_src_addr(id_priv),
3488                rdma_addr_size(cma_src_addr(id_priv)));
3489         memcpy(&cm_id->remote_addr, cma_dst_addr(id_priv),
3490                rdma_addr_size(cma_dst_addr(id_priv)));
3491
3492         ret = cma_modify_qp_rtr(id_priv, conn_param);
3493         if (ret)
3494                 goto out;
3495
3496         if (conn_param) {
3497                 iw_param.ord = conn_param->initiator_depth;
3498                 iw_param.ird = conn_param->responder_resources;
3499                 iw_param.private_data = conn_param->private_data;
3500                 iw_param.private_data_len = conn_param->private_data_len;
3501                 iw_param.qpn = id_priv->id.qp ? id_priv->qp_num : conn_param->qp_num;
3502         } else {
3503                 memset(&iw_param, 0, sizeof iw_param);
3504                 iw_param.qpn = id_priv->qp_num;
3505         }
3506         ret = iw_cm_connect(cm_id, &iw_param);
3507 out:
3508         if (ret) {
3509                 iw_destroy_cm_id(cm_id);
3510                 id_priv->cm_id.iw = NULL;
3511         }
3512         return ret;
3513 }
3514
3515 int rdma_connect(struct rdma_cm_id *id, struct rdma_conn_param *conn_param)
3516 {
3517         struct rdma_id_private *id_priv;
3518         int ret;
3519
3520         id_priv = container_of(id, struct rdma_id_private, id);
3521         if (!cma_comp_exch(id_priv, RDMA_CM_ROUTE_RESOLVED, RDMA_CM_CONNECT))
3522                 return -EINVAL;
3523
3524         if (!id->qp) {
3525                 id_priv->qp_num = conn_param->qp_num;
3526                 id_priv->srq = conn_param->srq;
3527         }
3528
3529         if (rdma_cap_ib_cm(id->device, id->port_num)) {
3530                 if (id->qp_type == IB_QPT_UD)
3531                         ret = cma_resolve_ib_udp(id_priv, conn_param);
3532                 else
3533                         ret = cma_connect_ib(id_priv, conn_param);
3534         } else if (rdma_cap_iw_cm(id->device, id->port_num))
3535                 ret = cma_connect_iw(id_priv, conn_param);
3536         else
3537                 ret = -ENOSYS;
3538         if (ret)
3539                 goto err;
3540
3541         return 0;
3542 err:
3543         cma_comp_exch(id_priv, RDMA_CM_CONNECT, RDMA_CM_ROUTE_RESOLVED);
3544         return ret;
3545 }
3546 EXPORT_SYMBOL(rdma_connect);
3547
3548 static int cma_accept_ib(struct rdma_id_private *id_priv,
3549                          struct rdma_conn_param *conn_param)
3550 {
3551         struct ib_cm_rep_param rep;
3552         int ret;
3553
3554         ret = cma_modify_qp_rtr(id_priv, conn_param);
3555         if (ret)
3556                 goto out;
3557
3558         ret = cma_modify_qp_rts(id_priv, conn_param);
3559         if (ret)
3560                 goto out;
3561
3562         memset(&rep, 0, sizeof rep);
3563         rep.qp_num = id_priv->qp_num;
3564         rep.starting_psn = id_priv->seq_num;
3565         rep.private_data = conn_param->private_data;
3566         rep.private_data_len = conn_param->private_data_len;
3567         rep.responder_resources = conn_param->responder_resources;
3568         rep.initiator_depth = conn_param->initiator_depth;
3569         rep.failover_accepted = 0;
3570         rep.flow_control = conn_param->flow_control;
3571         rep.rnr_retry_count = min_t(u8, 7, conn_param->rnr_retry_count);
3572         rep.srq = id_priv->srq ? 1 : 0;
3573
3574         ret = ib_send_cm_rep(id_priv->cm_id.ib, &rep);
3575 out:
3576         return ret;
3577 }
3578
3579 static int cma_accept_iw(struct rdma_id_private *id_priv,
3580                   struct rdma_conn_param *conn_param)
3581 {
3582         struct iw_cm_conn_param iw_param;
3583         int ret;
3584
3585         ret = cma_modify_qp_rtr(id_priv, conn_param);
3586         if (ret)
3587                 return ret;
3588
3589         iw_param.ord = conn_param->initiator_depth;
3590         iw_param.ird = conn_param->responder_resources;
3591         iw_param.private_data = conn_param->private_data;
3592         iw_param.private_data_len = conn_param->private_data_len;
3593         if (id_priv->id.qp) {
3594                 iw_param.qpn = id_priv->qp_num;
3595         } else
3596                 iw_param.qpn = conn_param->qp_num;
3597
3598         return iw_cm_accept(id_priv->cm_id.iw, &iw_param);
3599 }
3600
3601 static int cma_send_sidr_rep(struct rdma_id_private *id_priv,
3602                              enum ib_cm_sidr_status status, u32 qkey,
3603                              const void *private_data, int private_data_len)
3604 {
3605         struct ib_cm_sidr_rep_param rep;
3606         int ret;
3607
3608         memset(&rep, 0, sizeof rep);
3609         rep.status = status;
3610         if (status == IB_SIDR_SUCCESS) {
3611                 ret = cma_set_qkey(id_priv, qkey);
3612                 if (ret)
3613                         return ret;
3614                 rep.qp_num = id_priv->qp_num;
3615                 rep.qkey = id_priv->qkey;
3616         }
3617         rep.private_data = private_data;
3618         rep.private_data_len = private_data_len;
3619
3620         return ib_send_cm_sidr_rep(id_priv->cm_id.ib, &rep);
3621 }
3622
3623 int rdma_accept(struct rdma_cm_id *id, struct rdma_conn_param *conn_param)
3624 {
3625         struct rdma_id_private *id_priv;
3626         int ret;
3627
3628         id_priv = container_of(id, struct rdma_id_private, id);
3629
3630         id_priv->owner = task_pid_nr(current);
3631
3632         if (!cma_comp(id_priv, RDMA_CM_CONNECT))
3633                 return -EINVAL;
3634
3635         if (!id->qp && conn_param) {
3636                 id_priv->qp_num = conn_param->qp_num;
3637                 id_priv->srq = conn_param->srq;
3638         }
3639
3640         if (rdma_cap_ib_cm(id->device, id->port_num)) {
3641                 if (id->qp_type == IB_QPT_UD) {
3642                         if (conn_param)
3643                                 ret = cma_send_sidr_rep(id_priv, IB_SIDR_SUCCESS,
3644                                                         conn_param->qkey,
3645                                                         conn_param->private_data,
3646                                                         conn_param->private_data_len);
3647                         else
3648                                 ret = cma_send_sidr_rep(id_priv, IB_SIDR_SUCCESS,
3649                                                         0, NULL, 0);
3650                 } else {
3651                         if (conn_param)
3652                                 ret = cma_accept_ib(id_priv, conn_param);
3653                         else
3654                                 ret = cma_rep_recv(id_priv);
3655                 }
3656         } else if (rdma_cap_iw_cm(id->device, id->port_num))
3657                 ret = cma_accept_iw(id_priv, conn_param);
3658         else
3659                 ret = -ENOSYS;
3660
3661         if (ret)
3662                 goto reject;
3663
3664         return 0;
3665 reject:
3666         cma_modify_qp_err(id_priv);
3667         rdma_reject(id, NULL, 0);
3668         return ret;
3669 }
3670 EXPORT_SYMBOL(rdma_accept);
3671
3672 int rdma_notify(struct rdma_cm_id *id, enum ib_event_type event)
3673 {
3674         struct rdma_id_private *id_priv;
3675         int ret;
3676
3677         id_priv = container_of(id, struct rdma_id_private, id);
3678         if (!id_priv->cm_id.ib)
3679                 return -EINVAL;
3680
3681         switch (id->device->node_type) {
3682         case RDMA_NODE_IB_CA:
3683                 ret = ib_cm_notify(id_priv->cm_id.ib, event);
3684                 break;
3685         default:
3686                 ret = 0;
3687                 break;
3688         }
3689         return ret;
3690 }
3691 EXPORT_SYMBOL(rdma_notify);
3692
3693 int rdma_reject(struct rdma_cm_id *id, const void *private_data,
3694                 u8 private_data_len)
3695 {
3696         struct rdma_id_private *id_priv;
3697         int ret;
3698
3699         id_priv = container_of(id, struct rdma_id_private, id);
3700         if (!id_priv->cm_id.ib)
3701                 return -EINVAL;
3702
3703         if (rdma_cap_ib_cm(id->device, id->port_num)) {
3704                 if (id->qp_type == IB_QPT_UD)
3705                         ret = cma_send_sidr_rep(id_priv, IB_SIDR_REJECT, 0,
3706                                                 private_data, private_data_len);
3707                 else
3708                         ret = ib_send_cm_rej(id_priv->cm_id.ib,
3709                                              IB_CM_REJ_CONSUMER_DEFINED, NULL,
3710                                              0, private_data, private_data_len);
3711         } else if (rdma_cap_iw_cm(id->device, id->port_num)) {
3712                 ret = iw_cm_reject(id_priv->cm_id.iw,
3713                                    private_data, private_data_len);
3714         } else
3715                 ret = -ENOSYS;
3716
3717         return ret;
3718 }
3719 EXPORT_SYMBOL(rdma_reject);
3720
3721 int rdma_disconnect(struct rdma_cm_id *id)
3722 {
3723         struct rdma_id_private *id_priv;
3724         int ret;
3725
3726         id_priv = container_of(id, struct rdma_id_private, id);
3727         if (!id_priv->cm_id.ib)
3728                 return -EINVAL;
3729
3730         if (rdma_cap_ib_cm(id->device, id->port_num)) {
3731                 ret = cma_modify_qp_err(id_priv);
3732                 if (ret)
3733                         goto out;
3734                 /* Initiate or respond to a disconnect. */
3735                 if (ib_send_cm_dreq(id_priv->cm_id.ib, NULL, 0))
3736                         ib_send_cm_drep(id_priv->cm_id.ib, NULL, 0);
3737         } else if (rdma_cap_iw_cm(id->device, id->port_num)) {
3738                 ret = iw_cm_disconnect(id_priv->cm_id.iw, 0);
3739         } else
3740                 ret = -EINVAL;
3741
3742 out:
3743         return ret;
3744 }
3745 EXPORT_SYMBOL(rdma_disconnect);
3746
3747 static int cma_ib_mc_handler(int status, struct ib_sa_multicast *multicast)
3748 {
3749         struct rdma_id_private *id_priv;
3750         struct cma_multicast *mc = multicast->context;
3751         struct rdma_cm_event event;
3752         int ret = 0;
3753
3754         id_priv = mc->id_priv;
3755         mutex_lock(&id_priv->handler_mutex);
3756         if (id_priv->state != RDMA_CM_ADDR_BOUND &&
3757             id_priv->state != RDMA_CM_ADDR_RESOLVED)
3758                 goto out;
3759
3760         if (!status)
3761                 status = cma_set_qkey(id_priv, be32_to_cpu(multicast->rec.qkey));
3762         mutex_lock(&id_priv->qp_mutex);
3763         if (!status && id_priv->id.qp)
3764                 status = ib_attach_mcast(id_priv->id.qp, &multicast->rec.mgid,
3765                                          be16_to_cpu(multicast->rec.mlid));
3766         mutex_unlock(&id_priv->qp_mutex);
3767
3768         memset(&event, 0, sizeof event);
3769         event.status = status;
3770         event.param.ud.private_data = mc->context;
3771         if (!status) {
3772                 struct rdma_dev_addr *dev_addr =
3773                         &id_priv->id.route.addr.dev_addr;
3774                 struct net_device *ndev =
3775                         dev_get_by_index(&init_net, dev_addr->bound_dev_if);
3776                 enum ib_gid_type gid_type =
3777                         id_priv->cma_dev->default_gid_type[id_priv->id.port_num -
3778                         rdma_start_port(id_priv->cma_dev->device)];
3779
3780                 event.event = RDMA_CM_EVENT_MULTICAST_JOIN;
3781                 ib_init_ah_from_mcmember(id_priv->id.device,
3782                                          id_priv->id.port_num, &multicast->rec,
3783                                          ndev, gid_type,
3784                                          &event.param.ud.ah_attr);
3785                 event.param.ud.qp_num = 0xFFFFFF;
3786                 event.param.ud.qkey = be32_to_cpu(multicast->rec.qkey);
3787                 if (ndev)
3788                         dev_put(ndev);
3789         } else
3790                 event.event = RDMA_CM_EVENT_MULTICAST_ERROR;
3791
3792         ret = id_priv->id.event_handler(&id_priv->id, &event);
3793         if (ret) {
3794                 cma_exch(id_priv, RDMA_CM_DESTROYING);
3795                 mutex_unlock(&id_priv->handler_mutex);
3796                 rdma_destroy_id(&id_priv->id);
3797                 return 0;
3798         }
3799
3800 out:
3801         mutex_unlock(&id_priv->handler_mutex);
3802         return 0;
3803 }
3804
3805 static void cma_set_mgid(struct rdma_id_private *id_priv,
3806                          struct sockaddr *addr, union ib_gid *mgid)
3807 {
3808         unsigned char mc_map[MAX_ADDR_LEN];
3809         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
3810         struct sockaddr_in *sin = (struct sockaddr_in *) addr;
3811         struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *) addr;
3812
3813         if (cma_any_addr(addr)) {
3814                 memset(mgid, 0, sizeof *mgid);
3815         } else if ((addr->sa_family == AF_INET6) &&
3816                    ((be32_to_cpu(sin6->sin6_addr.s6_addr32[0]) & 0xFFF0FFFF) ==
3817                                                                  0xFF10A01B)) {
3818                 /* IPv6 address is an SA assigned MGID. */
3819                 memcpy(mgid, &sin6->sin6_addr, sizeof *mgid);
3820         } else if (addr->sa_family == AF_IB) {
3821                 memcpy(mgid, &((struct sockaddr_ib *) addr)->sib_addr, sizeof *mgid);
3822         } else if ((addr->sa_family == AF_INET6)) {
3823                 ipv6_ib_mc_map(&sin6->sin6_addr, dev_addr->broadcast, mc_map);
3824                 if (id_priv->id.ps == RDMA_PS_UDP)
3825                         mc_map[7] = 0x01;       /* Use RDMA CM signature */
3826                 *mgid = *(union ib_gid *) (mc_map + 4);
3827         } else {
3828                 ip_ib_mc_map(sin->sin_addr.s_addr, dev_addr->broadcast, mc_map);
3829                 if (id_priv->id.ps == RDMA_PS_UDP)
3830                         mc_map[7] = 0x01;       /* Use RDMA CM signature */
3831                 *mgid = *(union ib_gid *) (mc_map + 4);
3832         }
3833 }
3834
3835 static void cma_query_sa_classport_info_cb(int status,
3836                                            struct ib_class_port_info *rec,
3837                                            void *context)
3838 {
3839         struct class_port_info_context *cb_ctx = context;
3840
3841         WARN_ON(!context);
3842
3843         if (status || !rec) {
3844                 pr_debug("RDMA CM: %s port %u failed query ClassPortInfo status: %d\n",
3845                          cb_ctx->device->name, cb_ctx->port_num, status);
3846                 goto out;
3847         }
3848
3849         memcpy(cb_ctx->class_port_info, rec, sizeof(struct ib_class_port_info));
3850
3851 out:
3852         complete(&cb_ctx->done);
3853 }
3854
3855 static int cma_query_sa_classport_info(struct ib_device *device, u8 port_num,
3856                                        struct ib_class_port_info *class_port_info)
3857 {
3858         struct class_port_info_context *cb_ctx;
3859         int ret;
3860
3861         cb_ctx = kmalloc(sizeof(*cb_ctx), GFP_KERNEL);
3862         if (!cb_ctx)
3863                 return -ENOMEM;
3864
3865         cb_ctx->device = device;
3866         cb_ctx->class_port_info = class_port_info;
3867         cb_ctx->port_num = port_num;
3868         init_completion(&cb_ctx->done);
3869
3870         ret = ib_sa_classport_info_rec_query(&sa_client, device, port_num,
3871                                              CMA_QUERY_CLASSPORT_INFO_TIMEOUT,
3872                                              GFP_KERNEL, cma_query_sa_classport_info_cb,
3873                                              cb_ctx, &cb_ctx->sa_query);
3874         if (ret < 0) {
3875                 pr_err("RDMA CM: %s port %u failed to send ClassPortInfo query, ret: %d\n",
3876                        device->name, port_num, ret);
3877                 goto out;
3878         }
3879
3880         wait_for_completion(&cb_ctx->done);
3881
3882 out:
3883         kfree(cb_ctx);
3884         return ret;
3885 }
3886
3887 static int cma_join_ib_multicast(struct rdma_id_private *id_priv,
3888                                  struct cma_multicast *mc)
3889 {
3890         struct ib_sa_mcmember_rec rec;
3891         struct ib_class_port_info class_port_info;
3892         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
3893         ib_sa_comp_mask comp_mask;
3894         int ret;
3895
3896         ib_addr_get_mgid(dev_addr, &rec.mgid);
3897         ret = ib_sa_get_mcmember_rec(id_priv->id.device, id_priv->id.port_num,
3898                                      &rec.mgid, &rec);
3899         if (ret)
3900                 return ret;
3901
3902         ret = cma_set_qkey(id_priv, 0);
3903         if (ret)
3904                 return ret;
3905
3906         cma_set_mgid(id_priv, (struct sockaddr *) &mc->addr, &rec.mgid);
3907         rec.qkey = cpu_to_be32(id_priv->qkey);
3908         rdma_addr_get_sgid(dev_addr, &rec.port_gid);
3909         rec.pkey = cpu_to_be16(ib_addr_get_pkey(dev_addr));
3910         rec.join_state = mc->join_state;
3911
3912         if (rec.join_state == BIT(SENDONLY_FULLMEMBER_JOIN)) {
3913                 ret = cma_query_sa_classport_info(id_priv->id.device,
3914                                                   id_priv->id.port_num,
3915                                                   &class_port_info);
3916
3917                 if (ret)
3918                         return ret;
3919
3920                 if (!(ib_get_cpi_capmask2(&class_port_info) &
3921                       IB_SA_CAP_MASK2_SENDONLY_FULL_MEM_SUPPORT)) {
3922                         pr_warn("RDMA CM: %s port %u Unable to multicast join\n"
3923                                 "RDMA CM: SM doesn't support Send Only Full Member option\n",
3924                                 id_priv->id.device->name, id_priv->id.port_num);
3925                         return -EOPNOTSUPP;
3926                 }
3927         }
3928
3929         comp_mask = IB_SA_MCMEMBER_REC_MGID | IB_SA_MCMEMBER_REC_PORT_GID |
3930                     IB_SA_MCMEMBER_REC_PKEY | IB_SA_MCMEMBER_REC_JOIN_STATE |
3931                     IB_SA_MCMEMBER_REC_QKEY | IB_SA_MCMEMBER_REC_SL |
3932                     IB_SA_MCMEMBER_REC_FLOW_LABEL |
3933                     IB_SA_MCMEMBER_REC_TRAFFIC_CLASS;
3934
3935         if (id_priv->id.ps == RDMA_PS_IPOIB)
3936                 comp_mask |= IB_SA_MCMEMBER_REC_RATE |
3937                              IB_SA_MCMEMBER_REC_RATE_SELECTOR |
3938                              IB_SA_MCMEMBER_REC_MTU_SELECTOR |
3939                              IB_SA_MCMEMBER_REC_MTU |
3940                              IB_SA_MCMEMBER_REC_HOP_LIMIT;
3941
3942         mc->multicast.ib = ib_sa_join_multicast(&sa_client, id_priv->id.device,
3943                                                 id_priv->id.port_num, &rec,
3944                                                 comp_mask, GFP_KERNEL,
3945                                                 cma_ib_mc_handler, mc);
3946         return PTR_ERR_OR_ZERO(mc->multicast.ib);
3947 }
3948
3949 static void iboe_mcast_work_handler(struct work_struct *work)
3950 {
3951         struct iboe_mcast_work *mw = container_of(work, struct iboe_mcast_work, work);
3952         struct cma_multicast *mc = mw->mc;
3953         struct ib_sa_multicast *m = mc->multicast.ib;
3954
3955         mc->multicast.ib->context = mc;
3956         cma_ib_mc_handler(0, m);
3957         kref_put(&mc->mcref, release_mc);
3958         kfree(mw);
3959 }
3960
3961 static void cma_iboe_set_mgid(struct sockaddr *addr, union ib_gid *mgid)
3962 {
3963         struct sockaddr_in *sin = (struct sockaddr_in *)addr;
3964         struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)addr;
3965
3966         if (cma_any_addr(addr)) {
3967                 memset(mgid, 0, sizeof *mgid);
3968         } else if (addr->sa_family == AF_INET6) {
3969                 memcpy(mgid, &sin6->sin6_addr, sizeof *mgid);
3970         } else {
3971                 mgid->raw[0] = 0xff;
3972                 mgid->raw[1] = 0x0e;
3973                 mgid->raw[2] = 0;
3974                 mgid->raw[3] = 0;
3975                 mgid->raw[4] = 0;
3976                 mgid->raw[5] = 0;
3977                 mgid->raw[6] = 0;
3978                 mgid->raw[7] = 0;
3979                 mgid->raw[8] = 0;
3980                 mgid->raw[9] = 0;
3981                 mgid->raw[10] = 0xff;
3982                 mgid->raw[11] = 0xff;
3983                 *(__be32 *)(&mgid->raw[12]) = sin->sin_addr.s_addr;
3984         }
3985 }
3986
3987 static int cma_iboe_join_multicast(struct rdma_id_private *id_priv,
3988                                    struct cma_multicast *mc)
3989 {
3990         struct iboe_mcast_work *work;
3991         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
3992         int err = 0;
3993         struct sockaddr *addr = (struct sockaddr *)&mc->addr;
3994         struct net_device *ndev = NULL;
3995         enum ib_gid_type gid_type;
3996         bool send_only;
3997
3998         send_only = mc->join_state == BIT(SENDONLY_FULLMEMBER_JOIN);
3999
4000         if (cma_zero_addr((struct sockaddr *)&mc->addr))
4001                 return -EINVAL;
4002
4003         work = kzalloc(sizeof *work, GFP_KERNEL);
4004         if (!work)
4005                 return -ENOMEM;
4006
4007         mc->multicast.ib = kzalloc(sizeof(struct ib_sa_multicast), GFP_KERNEL);
4008         if (!mc->multicast.ib) {
4009                 err = -ENOMEM;
4010                 goto out1;
4011         }
4012
4013         cma_iboe_set_mgid(addr, &mc->multicast.ib->rec.mgid);
4014
4015         mc->multicast.ib->rec.pkey = cpu_to_be16(0xffff);
4016         if (id_priv->id.ps == RDMA_PS_UDP)
4017                 mc->multicast.ib->rec.qkey = cpu_to_be32(RDMA_UDP_QKEY);
4018
4019         if (dev_addr->bound_dev_if)
4020                 ndev = dev_get_by_index(&init_net, dev_addr->bound_dev_if);
4021         if (!ndev) {
4022                 err = -ENODEV;
4023                 goto out2;
4024         }
4025         mc->multicast.ib->rec.rate = iboe_get_rate(ndev);
4026         mc->multicast.ib->rec.hop_limit = 1;
4027         mc->multicast.ib->rec.mtu = iboe_get_mtu(ndev->mtu);
4028
4029         gid_type = id_priv->cma_dev->default_gid_type[id_priv->id.port_num -
4030                    rdma_start_port(id_priv->cma_dev->device)];
4031         if (addr->sa_family == AF_INET) {
4032                 if (gid_type == IB_GID_TYPE_ROCE_UDP_ENCAP) {
4033                         mc->multicast.ib->rec.hop_limit = IPV6_DEFAULT_HOPLIMIT;
4034                         if (!send_only) {
4035                                 err = cma_igmp_send(ndev, &mc->multicast.ib->rec.mgid,
4036                                                     true);
4037                                 if (!err)
4038                                         mc->igmp_joined = true;
4039                         }
4040                 }
4041         } else {
4042                 if (gid_type == IB_GID_TYPE_ROCE_UDP_ENCAP)
4043                         err = -ENOTSUPP;
4044         }
4045         dev_put(ndev);
4046         if (err || !mc->multicast.ib->rec.mtu) {
4047                 if (!err)
4048                         err = -EINVAL;
4049                 goto out2;
4050         }
4051         rdma_ip2gid((struct sockaddr *)&id_priv->id.route.addr.src_addr,
4052                     &mc->multicast.ib->rec.port_gid);
4053         work->id = id_priv;
4054         work->mc = mc;
4055         INIT_WORK(&work->work, iboe_mcast_work_handler);
4056         kref_get(&mc->mcref);
4057         queue_work(cma_wq, &work->work);
4058
4059         return 0;
4060
4061 out2:
4062         kfree(mc->multicast.ib);
4063 out1:
4064         kfree(work);
4065         return err;
4066 }
4067
4068 int rdma_join_multicast(struct rdma_cm_id *id, struct sockaddr *addr,
4069                         u8 join_state, void *context)
4070 {
4071         struct rdma_id_private *id_priv;
4072         struct cma_multicast *mc;
4073         int ret;
4074
4075         id_priv = container_of(id, struct rdma_id_private, id);
4076         if (!cma_comp(id_priv, RDMA_CM_ADDR_BOUND) &&
4077             !cma_comp(id_priv, RDMA_CM_ADDR_RESOLVED))
4078                 return -EINVAL;
4079
4080         mc = kmalloc(sizeof *mc, GFP_KERNEL);
4081         if (!mc)
4082                 return -ENOMEM;
4083
4084         memcpy(&mc->addr, addr, rdma_addr_size(addr));
4085         mc->context = context;
4086         mc->id_priv = id_priv;
4087         mc->igmp_joined = false;
4088         mc->join_state = join_state;
4089         spin_lock(&id_priv->lock);
4090         list_add(&mc->list, &id_priv->mc_list);
4091         spin_unlock(&id_priv->lock);
4092
4093         if (rdma_protocol_roce(id->device, id->port_num)) {
4094                 kref_init(&mc->mcref);
4095                 ret = cma_iboe_join_multicast(id_priv, mc);
4096         } else if (rdma_cap_ib_mcast(id->device, id->port_num))
4097                 ret = cma_join_ib_multicast(id_priv, mc);
4098         else
4099                 ret = -ENOSYS;
4100
4101         if (ret) {
4102                 spin_lock_irq(&id_priv->lock);
4103                 list_del(&mc->list);
4104                 spin_unlock_irq(&id_priv->lock);
4105                 kfree(mc);
4106         }
4107         return ret;
4108 }
4109 EXPORT_SYMBOL(rdma_join_multicast);
4110
4111 void rdma_leave_multicast(struct rdma_cm_id *id, struct sockaddr *addr)
4112 {
4113         struct rdma_id_private *id_priv;
4114         struct cma_multicast *mc;
4115
4116         id_priv = container_of(id, struct rdma_id_private, id);
4117         spin_lock_irq(&id_priv->lock);
4118         list_for_each_entry(mc, &id_priv->mc_list, list) {
4119                 if (!memcmp(&mc->addr, addr, rdma_addr_size(addr))) {
4120                         list_del(&mc->list);
4121                         spin_unlock_irq(&id_priv->lock);
4122
4123                         if (id->qp)
4124                                 ib_detach_mcast(id->qp,
4125                                                 &mc->multicast.ib->rec.mgid,
4126                                                 be16_to_cpu(mc->multicast.ib->rec.mlid));
4127
4128                         BUG_ON(id_priv->cma_dev->device != id->device);
4129
4130                         if (rdma_cap_ib_mcast(id->device, id->port_num)) {
4131                                 ib_sa_free_multicast(mc->multicast.ib);
4132                                 kfree(mc);
4133                         } else if (rdma_protocol_roce(id->device, id->port_num)) {
4134                                 if (mc->igmp_joined) {
4135                                         struct rdma_dev_addr *dev_addr =
4136                                                 &id->route.addr.dev_addr;
4137                                         struct net_device *ndev = NULL;
4138
4139                                         if (dev_addr->bound_dev_if)
4140                                                 ndev = dev_get_by_index(&init_net,
4141                                                                         dev_addr->bound_dev_if);
4142                                         if (ndev) {
4143                                                 cma_igmp_send(ndev,
4144                                                               &mc->multicast.ib->rec.mgid,
4145                                                               false);
4146                                                 dev_put(ndev);
4147                                         }
4148                                         mc->igmp_joined = false;
4149                                 }
4150                                 kref_put(&mc->mcref, release_mc);
4151                         }
4152                         return;
4153                 }
4154         }
4155         spin_unlock_irq(&id_priv->lock);
4156 }
4157 EXPORT_SYMBOL(rdma_leave_multicast);
4158
4159 static int cma_netdev_change(struct net_device *ndev, struct rdma_id_private *id_priv)
4160 {
4161         struct rdma_dev_addr *dev_addr;
4162         struct cma_ndev_work *work;
4163
4164         dev_addr = &id_priv->id.route.addr.dev_addr;
4165
4166         if ((dev_addr->bound_dev_if == ndev->ifindex) &&
4167             (net_eq(dev_net(ndev), dev_addr->net)) &&
4168             memcmp(dev_addr->src_dev_addr, ndev->dev_addr, ndev->addr_len)) {
4169                 pr_info("RDMA CM addr change for ndev %s used by id %p\n",
4170                         ndev->name, &id_priv->id);
4171                 work = kzalloc(sizeof *work, GFP_KERNEL);
4172                 if (!work)
4173                         return -ENOMEM;
4174
4175                 INIT_WORK(&work->work, cma_ndev_work_handler);
4176                 work->id = id_priv;
4177                 work->event.event = RDMA_CM_EVENT_ADDR_CHANGE;
4178                 atomic_inc(&id_priv->refcount);
4179                 queue_work(cma_wq, &work->work);
4180         }
4181
4182         return 0;
4183 }
4184
4185 static int cma_netdev_callback(struct notifier_block *self, unsigned long event,
4186                                void *ptr)
4187 {
4188         struct net_device *ndev = netdev_notifier_info_to_dev(ptr);
4189         struct cma_device *cma_dev;
4190         struct rdma_id_private *id_priv;
4191         int ret = NOTIFY_DONE;
4192
4193         if (event != NETDEV_BONDING_FAILOVER)
4194                 return NOTIFY_DONE;
4195
4196         if (!(ndev->flags & IFF_MASTER) || !(ndev->priv_flags & IFF_BONDING))
4197                 return NOTIFY_DONE;
4198
4199         mutex_lock(&lock);
4200         list_for_each_entry(cma_dev, &dev_list, list)
4201                 list_for_each_entry(id_priv, &cma_dev->id_list, list) {
4202                         ret = cma_netdev_change(ndev, id_priv);
4203                         if (ret)
4204                                 goto out;
4205                 }
4206
4207 out:
4208         mutex_unlock(&lock);
4209         return ret;
4210 }
4211
4212 static struct notifier_block cma_nb = {
4213         .notifier_call = cma_netdev_callback
4214 };
4215
4216 static void cma_add_one(struct ib_device *device)
4217 {
4218         struct cma_device *cma_dev;
4219         struct rdma_id_private *id_priv;
4220         unsigned int i;
4221         unsigned long supported_gids = 0;
4222
4223         cma_dev = kmalloc(sizeof *cma_dev, GFP_KERNEL);
4224         if (!cma_dev)
4225                 return;
4226
4227         cma_dev->device = device;
4228         cma_dev->default_gid_type = kcalloc(device->phys_port_cnt,
4229                                             sizeof(*cma_dev->default_gid_type),
4230                                             GFP_KERNEL);
4231         if (!cma_dev->default_gid_type) {
4232                 kfree(cma_dev);
4233                 return;
4234         }
4235         for (i = rdma_start_port(device); i <= rdma_end_port(device); i++) {
4236                 supported_gids = roce_gid_type_mask_support(device, i);
4237                 WARN_ON(!supported_gids);
4238                 cma_dev->default_gid_type[i - rdma_start_port(device)] =
4239                         find_first_bit(&supported_gids, BITS_PER_LONG);
4240         }
4241
4242         init_completion(&cma_dev->comp);
4243         atomic_set(&cma_dev->refcount, 1);
4244         INIT_LIST_HEAD(&cma_dev->id_list);
4245         ib_set_client_data(device, &cma_client, cma_dev);
4246
4247         mutex_lock(&lock);
4248         list_add_tail(&cma_dev->list, &dev_list);
4249         list_for_each_entry(id_priv, &listen_any_list, list)
4250                 cma_listen_on_dev(id_priv, cma_dev);
4251         mutex_unlock(&lock);
4252 }
4253
4254 static int cma_remove_id_dev(struct rdma_id_private *id_priv)
4255 {
4256         struct rdma_cm_event event;
4257         enum rdma_cm_state state;
4258         int ret = 0;
4259
4260         /* Record that we want to remove the device */
4261         state = cma_exch(id_priv, RDMA_CM_DEVICE_REMOVAL);
4262         if (state == RDMA_CM_DESTROYING)
4263                 return 0;
4264
4265         cma_cancel_operation(id_priv, state);
4266         mutex_lock(&id_priv->handler_mutex);
4267
4268         /* Check for destruction from another callback. */
4269         if (!cma_comp(id_priv, RDMA_CM_DEVICE_REMOVAL))
4270                 goto out;
4271
4272         memset(&event, 0, sizeof event);
4273         event.event = RDMA_CM_EVENT_DEVICE_REMOVAL;
4274         ret = id_priv->id.event_handler(&id_priv->id, &event);
4275 out:
4276         mutex_unlock(&id_priv->handler_mutex);
4277         return ret;
4278 }
4279
4280 static void cma_process_remove(struct cma_device *cma_dev)
4281 {
4282         struct rdma_id_private *id_priv;
4283         int ret;
4284
4285         mutex_lock(&lock);
4286         while (!list_empty(&cma_dev->id_list)) {
4287                 id_priv = list_entry(cma_dev->id_list.next,
4288                                      struct rdma_id_private, list);
4289
4290                 list_del(&id_priv->listen_list);
4291                 list_del_init(&id_priv->list);
4292                 atomic_inc(&id_priv->refcount);
4293                 mutex_unlock(&lock);
4294
4295                 ret = id_priv->internal_id ? 1 : cma_remove_id_dev(id_priv);
4296                 cma_deref_id(id_priv);
4297                 if (ret)
4298                         rdma_destroy_id(&id_priv->id);
4299
4300                 mutex_lock(&lock);
4301         }
4302         mutex_unlock(&lock);
4303
4304         cma_deref_dev(cma_dev);
4305         wait_for_completion(&cma_dev->comp);
4306 }
4307
4308 static void cma_remove_one(struct ib_device *device, void *client_data)
4309 {
4310         struct cma_device *cma_dev = client_data;
4311
4312         if (!cma_dev)
4313                 return;
4314
4315         mutex_lock(&lock);
4316         list_del(&cma_dev->list);
4317         mutex_unlock(&lock);
4318
4319         cma_process_remove(cma_dev);
4320         kfree(cma_dev->default_gid_type);
4321         kfree(cma_dev);
4322 }
4323
4324 static int cma_get_id_stats(struct sk_buff *skb, struct netlink_callback *cb)
4325 {
4326         struct nlmsghdr *nlh;
4327         struct rdma_cm_id_stats *id_stats;
4328         struct rdma_id_private *id_priv;
4329         struct rdma_cm_id *id = NULL;
4330         struct cma_device *cma_dev;
4331         int i_dev = 0, i_id = 0;
4332
4333         /*
4334          * We export all of the IDs as a sequence of messages.  Each
4335          * ID gets its own netlink message.
4336          */
4337         mutex_lock(&lock);
4338
4339         list_for_each_entry(cma_dev, &dev_list, list) {
4340                 if (i_dev < cb->args[0]) {
4341                         i_dev++;
4342                         continue;
4343                 }
4344
4345                 i_id = 0;
4346                 list_for_each_entry(id_priv, &cma_dev->id_list, list) {
4347                         if (i_id < cb->args[1]) {
4348                                 i_id++;
4349                                 continue;
4350                         }
4351
4352                         id_stats = ibnl_put_msg(skb, &nlh, cb->nlh->nlmsg_seq,
4353                                                 sizeof *id_stats, RDMA_NL_RDMA_CM,
4354                                                 RDMA_NL_RDMA_CM_ID_STATS,
4355                                                 NLM_F_MULTI);
4356                         if (!id_stats)
4357                                 goto out;
4358
4359                         memset(id_stats, 0, sizeof *id_stats);
4360                         id = &id_priv->id;
4361                         id_stats->node_type = id->route.addr.dev_addr.dev_type;
4362                         id_stats->port_num = id->port_num;
4363                         id_stats->bound_dev_if =
4364                                 id->route.addr.dev_addr.bound_dev_if;
4365
4366                         if (ibnl_put_attr(skb, nlh,
4367                                           rdma_addr_size(cma_src_addr(id_priv)),
4368                                           cma_src_addr(id_priv),
4369                                           RDMA_NL_RDMA_CM_ATTR_SRC_ADDR))
4370                                 goto out;
4371                         if (ibnl_put_attr(skb, nlh,
4372                                           rdma_addr_size(cma_src_addr(id_priv)),
4373                                           cma_dst_addr(id_priv),
4374                                           RDMA_NL_RDMA_CM_ATTR_DST_ADDR))
4375                                 goto out;
4376
4377                         id_stats->pid           = id_priv->owner;
4378                         id_stats->port_space    = id->ps;
4379                         id_stats->cm_state      = id_priv->state;
4380                         id_stats->qp_num        = id_priv->qp_num;
4381                         id_stats->qp_type       = id->qp_type;
4382
4383                         i_id++;
4384                 }
4385
4386                 cb->args[1] = 0;
4387                 i_dev++;
4388         }
4389
4390 out:
4391         mutex_unlock(&lock);
4392         cb->args[0] = i_dev;
4393         cb->args[1] = i_id;
4394
4395         return skb->len;
4396 }
4397
4398 static const struct ibnl_client_cbs cma_cb_table[] = {
4399         [RDMA_NL_RDMA_CM_ID_STATS] = { .dump = cma_get_id_stats,
4400                                        .module = THIS_MODULE },
4401 };
4402
4403 static int cma_init_net(struct net *net)
4404 {
4405         struct cma_pernet *pernet = cma_pernet(net);
4406
4407         idr_init(&pernet->tcp_ps);
4408         idr_init(&pernet->udp_ps);
4409         idr_init(&pernet->ipoib_ps);
4410         idr_init(&pernet->ib_ps);
4411
4412         return 0;
4413 }
4414
4415 static void cma_exit_net(struct net *net)
4416 {
4417         struct cma_pernet *pernet = cma_pernet(net);
4418
4419         idr_destroy(&pernet->tcp_ps);
4420         idr_destroy(&pernet->udp_ps);
4421         idr_destroy(&pernet->ipoib_ps);
4422         idr_destroy(&pernet->ib_ps);
4423 }
4424
4425 static struct pernet_operations cma_pernet_operations = {
4426         .init = cma_init_net,
4427         .exit = cma_exit_net,
4428         .id = &cma_pernet_id,
4429         .size = sizeof(struct cma_pernet),
4430 };
4431
4432 static int __init cma_init(void)
4433 {
4434         int ret;
4435
4436         cma_wq = alloc_ordered_workqueue("rdma_cm", WQ_MEM_RECLAIM);
4437         if (!cma_wq)
4438                 return -ENOMEM;
4439
4440         ret = register_pernet_subsys(&cma_pernet_operations);
4441         if (ret)
4442                 goto err_wq;
4443
4444         ib_sa_register_client(&sa_client);
4445         rdma_addr_register_client(&addr_client);
4446         register_netdevice_notifier(&cma_nb);
4447
4448         ret = ib_register_client(&cma_client);
4449         if (ret)
4450                 goto err;
4451
4452         if (ibnl_add_client(RDMA_NL_RDMA_CM, ARRAY_SIZE(cma_cb_table),
4453                             cma_cb_table))
4454                 pr_warn("RDMA CMA: failed to add netlink callback\n");
4455         cma_configfs_init();
4456
4457         return 0;
4458
4459 err:
4460         unregister_netdevice_notifier(&cma_nb);
4461         rdma_addr_unregister_client(&addr_client);
4462         ib_sa_unregister_client(&sa_client);
4463 err_wq:
4464         destroy_workqueue(cma_wq);
4465         return ret;
4466 }
4467
4468 static void __exit cma_cleanup(void)
4469 {
4470         cma_configfs_exit();
4471         ibnl_remove_client(RDMA_NL_RDMA_CM);
4472         ib_unregister_client(&cma_client);
4473         unregister_netdevice_notifier(&cma_nb);
4474         rdma_addr_unregister_client(&addr_client);
4475         ib_sa_unregister_client(&sa_client);
4476         unregister_pernet_subsys(&cma_pernet_operations);
4477         destroy_workqueue(cma_wq);
4478 }
4479
4480 module_init(cma_init);
4481 module_exit(cma_cleanup);