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         enum ib_port_state port_state;
 713         int i;
 714
 715         cma_dev = NULL;
 716         addr = (struct sockaddr_ib *) cma_dst_addr(id_priv);
 717         dgid = (union ib_gid *) &addr->sib_addr;
 718         pkey = ntohs(addr->sib_pkey);
 719
 720         list_for_each_entry(cur_dev, &dev_list, list) {
 721                 for (p = 1; p <= cur_dev->device->phys_port_cnt; ++p) {
 722                         if (!rdma_cap_af_ib(cur_dev->device, p))
 723                                 continue;
 724
 725                         if (ib_find_cached_pkey(cur_dev->device, p, pkey, &index))
 726                                 continue;
 727
 728                         if (ib_get_cached_port_state(cur_dev->device, p, &port_state))
 729                                 continue;
 730                         for (i = 0; !ib_get_cached_gid(cur_dev->device, p, i,
 731                                                        &gid, NULL);
 732                              i++) {
 733                                 if (!memcmp(&gid, dgid, sizeof(gid))) {
 734                                         cma_dev = cur_dev;
 735                                         sgid = gid;
 736                                         id_priv->id.port_num = p;
 737                                         goto found;
 738                                 }
 739
 740                                 if (!cma_dev && (gid.global.subnet_prefix ==
 741                                     dgid->global.subnet_prefix) &&
 742                                     port_state == IB_PORT_ACTIVE) {
 743                                         cma_dev = cur_dev;
 744                                         sgid = gid;
 745                                         id_priv->id.port_num = p;
 746                                 }
 747                         }
 748                 }
 749         }
 750
 751         if (!cma_dev)
 752                 return -ENODEV;
 753
 754 found:
 755         cma_attach_to_dev(id_priv, cma_dev);
 756         addr = (struct sockaddr_ib *) cma_src_addr(id_priv);
 757         memcpy(&addr->sib_addr, &sgid, sizeof sgid);
 758         cma_translate_ib(addr, &id_priv->id.route.addr.dev_addr);
 759         return 0;
 760 }
 761
 762 static void cma_deref_id(struct rdma_id_private *id_priv)
 763 {
 764         if (atomic_dec_and_test(&id_priv->refcount))
 765                 complete(&id_priv->comp);
 766 }
 767
 768 struct rdma_cm_id *rdma_create_id(struct net *net,
 769                                   rdma_cm_event_handler event_handler,
 770                                   void *context, enum rdma_port_space ps,
 771                                   enum ib_qp_type qp_type)
 772 {
 773         struct rdma_id_private *id_priv;
 774
 775         id_priv = kzalloc(sizeof *id_priv, GFP_KERNEL);
 776         if (!id_priv)
 777                 return ERR_PTR(-ENOMEM);
 778
 779         id_priv->owner = task_pid_nr(current);
 780         id_priv->state = RDMA_CM_IDLE;
 781         id_priv->id.context = context;
 782         id_priv->id.event_handler = event_handler;
 783         id_priv->id.ps = ps;
 784         id_priv->id.qp_type = qp_type;
 785         spin_lock_init(&id_priv->lock);
 786         mutex_init(&id_priv->qp_mutex);
 787         init_completion(&id_priv->comp);
 788         atomic_set(&id_priv->refcount, 1);
 789         mutex_init(&id_priv->handler_mutex);
 790         INIT_LIST_HEAD(&id_priv->listen_list);
 791         INIT_LIST_HEAD(&id_priv->mc_list);
 792         get_random_bytes(&id_priv->seq_num, sizeof id_priv->seq_num);
 793         id_priv->id.route.addr.dev_addr.net = get_net(net);
 794
 795         return &id_priv->id;
 796 }
 797 EXPORT_SYMBOL(rdma_create_id);
 798
 799 static int cma_init_ud_qp(struct rdma_id_private *id_priv, struct ib_qp *qp)
 800 {
 801         struct ib_qp_attr qp_attr;
 802         int qp_attr_mask, ret;
 803
 804         qp_attr.qp_state = IB_QPS_INIT;
 805         ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
 806         if (ret)
 807                 return ret;
 808
 809         ret = ib_modify_qp(qp, &qp_attr, qp_attr_mask);
 810         if (ret)
 811                 return ret;
 812
 813         qp_attr.qp_state = IB_QPS_RTR;
 814         ret = ib_modify_qp(qp, &qp_attr, IB_QP_STATE);
 815         if (ret)
 816                 return ret;
 817
 818         qp_attr.qp_state = IB_QPS_RTS;
 819         qp_attr.sq_psn = 0;
 820         ret = ib_modify_qp(qp, &qp_attr, IB_QP_STATE | IB_QP_SQ_PSN);
 821
 822         return ret;
 823 }
 824
 825 static int cma_init_conn_qp(struct rdma_id_private *id_priv, struct ib_qp *qp)
 826 {
 827         struct ib_qp_attr qp_attr;
 828         int qp_attr_mask, ret;
 829
 830         qp_attr.qp_state = IB_QPS_INIT;
 831         ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
 832         if (ret)
 833                 return ret;
 834
 835         return ib_modify_qp(qp, &qp_attr, qp_attr_mask);
 836 }
 837
 838 int rdma_create_qp(struct rdma_cm_id *id, struct ib_pd *pd,
 839                    struct ib_qp_init_attr *qp_init_attr)
 840 {
 841         struct rdma_id_private *id_priv;
 842         struct ib_qp *qp;
 843         int ret;
 844
 845         id_priv = container_of(id, struct rdma_id_private, id);
 846         if (id->device != pd->device)
 847                 return -EINVAL;
 848
 849         qp_init_attr->port_num = id->port_num;
 850         qp = ib_create_qp(pd, qp_init_attr);
 851         if (IS_ERR(qp))
 852                 return PTR_ERR(qp);
 853
 854         if (id->qp_type == IB_QPT_UD)
 855                 ret = cma_init_ud_qp(id_priv, qp);
 856         else
 857                 ret = cma_init_conn_qp(id_priv, qp);
 858         if (ret)
 859                 goto err;
 860
 861         id->qp = qp;
 862         id_priv->qp_num = qp->qp_num;
 863         id_priv->srq = (qp->srq != NULL);
 864         return 0;
 865 err:
 866         ib_destroy_qp(qp);
 867         return ret;
 868 }
 869 EXPORT_SYMBOL(rdma_create_qp);
 870
 871 void rdma_destroy_qp(struct rdma_cm_id *id)
 872 {
 873         struct rdma_id_private *id_priv;
 874
 875         id_priv = container_of(id, struct rdma_id_private, id);
 876         mutex_lock(&id_priv->qp_mutex);
 877         ib_destroy_qp(id_priv->id.qp);
 878         id_priv->id.qp = NULL;
 879         mutex_unlock(&id_priv->qp_mutex);
 880 }
 881 EXPORT_SYMBOL(rdma_destroy_qp);
 882
 883 static int cma_modify_qp_rtr(struct rdma_id_private *id_priv,
 884                              struct rdma_conn_param *conn_param)
 885 {
 886         struct ib_qp_attr qp_attr;
 887         int qp_attr_mask, ret;
 888         union ib_gid sgid;
 889
 890         mutex_lock(&id_priv->qp_mutex);
 891         if (!id_priv->id.qp) {
 892                 ret = 0;
 893                 goto out;
 894         }
 895
 896         /* Need to update QP attributes from default values. */
 897         qp_attr.qp_state = IB_QPS_INIT;
 898         ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
 899         if (ret)
 900                 goto out;
 901
 902         ret = ib_modify_qp(id_priv->id.qp, &qp_attr, qp_attr_mask);
 903         if (ret)
 904                 goto out;
 905
 906         qp_attr.qp_state = IB_QPS_RTR;
 907         ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
 908         if (ret)
 909                 goto out;
 910
 911         ret = ib_query_gid(id_priv->id.device, id_priv->id.port_num,
 912                            qp_attr.ah_attr.grh.sgid_index, &sgid, NULL);
 913         if (ret)
 914                 goto out;
 915
 916         BUG_ON(id_priv->cma_dev->device != id_priv->id.device);
 917
 918         if (conn_param)
 919                 qp_attr.max_dest_rd_atomic = conn_param->responder_resources;
 920         ret = ib_modify_qp(id_priv->id.qp, &qp_attr, qp_attr_mask);
 921 out:
 922         mutex_unlock(&id_priv->qp_mutex);
 923         return ret;
 924 }
 925
 926 static int cma_modify_qp_rts(struct rdma_id_private *id_priv,
 927                              struct rdma_conn_param *conn_param)
 928 {
 929         struct ib_qp_attr qp_attr;
 930         int qp_attr_mask, ret;
 931
 932         mutex_lock(&id_priv->qp_mutex);
 933         if (!id_priv->id.qp) {
 934                 ret = 0;
 935                 goto out;
 936         }
 937
 938         qp_attr.qp_state = IB_QPS_RTS;
 939         ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
 940         if (ret)
 941                 goto out;
 942
 943         if (conn_param)
 944                 qp_attr.max_rd_atomic = conn_param->initiator_depth;
 945         ret = ib_modify_qp(id_priv->id.qp, &qp_attr, qp_attr_mask);
 946 out:
 947         mutex_unlock(&id_priv->qp_mutex);
 948         return ret;
 949 }
 950
 951 static int cma_modify_qp_err(struct rdma_id_private *id_priv)
 952 {
 953         struct ib_qp_attr qp_attr;
 954         int ret;
 955
 956         mutex_lock(&id_priv->qp_mutex);
 957         if (!id_priv->id.qp) {
 958                 ret = 0;
 959                 goto out;
 960         }
 961
 962         qp_attr.qp_state = IB_QPS_ERR;
 963         ret = ib_modify_qp(id_priv->id.qp, &qp_attr, IB_QP_STATE);
 964 out:
 965         mutex_unlock(&id_priv->qp_mutex);
 966         return ret;
 967 }
 968
 969 static int cma_ib_init_qp_attr(struct rdma_id_private *id_priv,
 970                                struct ib_qp_attr *qp_attr, int *qp_attr_mask)
 971 {
 972         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
 973         int ret;
 974         u16 pkey;
 975
 976         if (rdma_cap_eth_ah(id_priv->id.device, id_priv->id.port_num))
 977                 pkey = 0xffff;
 978         else
 979                 pkey = ib_addr_get_pkey(dev_addr);
 980
 981         ret = ib_find_cached_pkey(id_priv->id.device, id_priv->id.port_num,
 982                                   pkey, &qp_attr->pkey_index);
 983         if (ret)
 984                 return ret;
 985
 986         qp_attr->port_num = id_priv->id.port_num;
 987         *qp_attr_mask = IB_QP_STATE | IB_QP_PKEY_INDEX | IB_QP_PORT;
 988
 989         if (id_priv->id.qp_type == IB_QPT_UD) {
 990                 ret = cma_set_qkey(id_priv, 0);
 991                 if (ret)
 992                         return ret;
 993
 994                 qp_attr->qkey = id_priv->qkey;
 995                 *qp_attr_mask |= IB_QP_QKEY;
 996         } else {
 997                 qp_attr->qp_access_flags = 0;
 998                 *qp_attr_mask |= IB_QP_ACCESS_FLAGS;
 999         }
1000         return 0;
1001 }
1002
1003 int rdma_init_qp_attr(struct rdma_cm_id *id, struct ib_qp_attr *qp_attr,
1004                        int *qp_attr_mask)
1005 {
1006         struct rdma_id_private *id_priv;
1007         int ret = 0;
1008
1009         id_priv = container_of(id, struct rdma_id_private, id);
1010         if (rdma_cap_ib_cm(id->device, id->port_num)) {
1011                 if (!id_priv->cm_id.ib || (id_priv->id.qp_type == IB_QPT_UD))
1012                         ret = cma_ib_init_qp_attr(id_priv, qp_attr, qp_attr_mask);
1013                 else
1014                         ret = ib_cm_init_qp_attr(id_priv->cm_id.ib, qp_attr,
1015                                                  qp_attr_mask);
1016
1017                 if (qp_attr->qp_state == IB_QPS_RTR)
1018                         qp_attr->rq_psn = id_priv->seq_num;
1019         } else if (rdma_cap_iw_cm(id->device, id->port_num)) {
1020                 if (!id_priv->cm_id.iw) {
1021                         qp_attr->qp_access_flags = 0;
1022                         *qp_attr_mask = IB_QP_STATE | IB_QP_ACCESS_FLAGS;
1023                 } else
1024                         ret = iw_cm_init_qp_attr(id_priv->cm_id.iw, qp_attr,
1025                                                  qp_attr_mask);
1026         } else
1027                 ret = -ENOSYS;
1028
1029         return ret;
1030 }
1031 EXPORT_SYMBOL(rdma_init_qp_attr);
1032
1033 static inline int cma_zero_addr(struct sockaddr *addr)
1034 {
1035         switch (addr->sa_family) {
1036         case AF_INET:
1037                 return ipv4_is_zeronet(((struct sockaddr_in *)addr)->sin_addr.s_addr);
1038         case AF_INET6:
1039                 return ipv6_addr_any(&((struct sockaddr_in6 *) addr)->sin6_addr);
1040         case AF_IB:
1041                 return ib_addr_any(&((struct sockaddr_ib *) addr)->sib_addr);
1042         default:
1043                 return 0;
1044         }
1045 }
1046
1047 static inline int cma_loopback_addr(struct sockaddr *addr)
1048 {
1049         switch (addr->sa_family) {
1050         case AF_INET:
1051                 return ipv4_is_loopback(((struct sockaddr_in *) addr)->sin_addr.s_addr);
1052         case AF_INET6:
1053                 return ipv6_addr_loopback(&((struct sockaddr_in6 *) addr)->sin6_addr);
1054         case AF_IB:
1055                 return ib_addr_loopback(&((struct sockaddr_ib *) addr)->sib_addr);
1056         default:
1057                 return 0;
1058         }
1059 }
1060
1061 static inline int cma_any_addr(struct sockaddr *addr)
1062 {
1063         return cma_zero_addr(addr) || cma_loopback_addr(addr);
1064 }
1065
1066 static int cma_addr_cmp(struct sockaddr *src, struct sockaddr *dst)
1067 {
1068         if (src->sa_family != dst->sa_family)
1069                 return -1;
1070
1071         switch (src->sa_family) {
1072         case AF_INET:
1073                 return ((struct sockaddr_in *) src)->sin_addr.s_addr !=
1074                        ((struct sockaddr_in *) dst)->sin_addr.s_addr;
1075         case AF_INET6:
1076                 return ipv6_addr_cmp(&((struct sockaddr_in6 *) src)->sin6_addr,
1077                                      &((struct sockaddr_in6 *) dst)->sin6_addr);
1078         default:
1079                 return ib_addr_cmp(&((struct sockaddr_ib *) src)->sib_addr,
1080                                    &((struct sockaddr_ib *) dst)->sib_addr);
1081         }
1082 }
1083
1084 static __be16 cma_port(struct sockaddr *addr)
1085 {
1086         struct sockaddr_ib *sib;
1087
1088         switch (addr->sa_family) {
1089         case AF_INET:
1090                 return ((struct sockaddr_in *) addr)->sin_port;
1091         case AF_INET6:
1092                 return ((struct sockaddr_in6 *) addr)->sin6_port;
1093         case AF_IB:
1094                 sib = (struct sockaddr_ib *) addr;
1095                 return htons((u16) (be64_to_cpu(sib->sib_sid) &
1096                                     be64_to_cpu(sib->sib_sid_mask)));
1097         default:
1098                 return 0;
1099         }
1100 }
1101
1102 static inline int cma_any_port(struct sockaddr *addr)
1103 {
1104         return !cma_port(addr);
1105 }
1106
1107 static void cma_save_ib_info(struct sockaddr *src_addr,
1108                              struct sockaddr *dst_addr,
1109                              struct rdma_cm_id *listen_id,
1110                              struct ib_sa_path_rec *path)
1111 {
1112         struct sockaddr_ib *listen_ib, *ib;
1113
1114         listen_ib = (struct sockaddr_ib *) &listen_id->route.addr.src_addr;
1115         if (src_addr) {
1116                 ib = (struct sockaddr_ib *)src_addr;
1117                 ib->sib_family = AF_IB;
1118                 if (path) {
1119                         ib->sib_pkey = path->pkey;
1120                         ib->sib_flowinfo = path->flow_label;
1121                         memcpy(&ib->sib_addr, &path->sgid, 16);
1122                         ib->sib_sid = path->service_id;
1123                         ib->sib_scope_id = 0;
1124                 } else {
1125                         ib->sib_pkey = listen_ib->sib_pkey;
1126                         ib->sib_flowinfo = listen_ib->sib_flowinfo;
1127                         ib->sib_addr = listen_ib->sib_addr;
1128                         ib->sib_sid = listen_ib->sib_sid;
1129                         ib->sib_scope_id = listen_ib->sib_scope_id;
1130                 }
1131                 ib->sib_sid_mask = cpu_to_be64(0xffffffffffffffffULL);
1132         }
1133         if (dst_addr) {
1134                 ib = (struct sockaddr_ib *)dst_addr;
1135                 ib->sib_family = AF_IB;
1136                 if (path) {
1137                         ib->sib_pkey = path->pkey;
1138                         ib->sib_flowinfo = path->flow_label;
1139                         memcpy(&ib->sib_addr, &path->dgid, 16);
1140                 }
1141         }
1142 }
1143
1144 static void cma_save_ip4_info(struct sockaddr_in *src_addr,
1145                               struct sockaddr_in *dst_addr,
1146                               struct cma_hdr *hdr,
1147                               __be16 local_port)
1148 {
1149         if (src_addr) {
1150                 *src_addr = (struct sockaddr_in) {
1151                         .sin_family = AF_INET,
1152                         .sin_addr.s_addr = hdr->dst_addr.ip4.addr,
1153                         .sin_port = local_port,
1154                 };
1155         }
1156
1157         if (dst_addr) {
1158                 *dst_addr = (struct sockaddr_in) {
1159                         .sin_family = AF_INET,
1160                         .sin_addr.s_addr = hdr->src_addr.ip4.addr,
1161                         .sin_port = hdr->port,
1162                 };
1163         }
1164 }
1165
1166 static void cma_save_ip6_info(struct sockaddr_in6 *src_addr,
1167                               struct sockaddr_in6 *dst_addr,
1168                               struct cma_hdr *hdr,
1169                               __be16 local_port)
1170 {
1171         if (src_addr) {
1172                 *src_addr = (struct sockaddr_in6) {
1173                         .sin6_family = AF_INET6,
1174                         .sin6_addr = hdr->dst_addr.ip6,
1175                         .sin6_port = local_port,
1176                 };
1177         }
1178
1179         if (dst_addr) {
1180                 *dst_addr = (struct sockaddr_in6) {
1181                         .sin6_family = AF_INET6,
1182                         .sin6_addr = hdr->src_addr.ip6,
1183                         .sin6_port = hdr->port,
1184                 };
1185         }
1186 }
1187
1188 static u16 cma_port_from_service_id(__be64 service_id)
1189 {
1190         return (u16)be64_to_cpu(service_id);
1191 }
1192
1193 static int cma_save_ip_info(struct sockaddr *src_addr,
1194                             struct sockaddr *dst_addr,
1195                             struct ib_cm_event *ib_event,
1196                             __be64 service_id)
1197 {
1198         struct cma_hdr *hdr;
1199         __be16 port;
1200
1201         hdr = ib_event->private_data;
1202         if (hdr->cma_version != CMA_VERSION)
1203                 return -EINVAL;
1204
1205         port = htons(cma_port_from_service_id(service_id));
1206
1207         switch (cma_get_ip_ver(hdr)) {
1208         case 4:
1209                 cma_save_ip4_info((struct sockaddr_in *)src_addr,
1210                                   (struct sockaddr_in *)dst_addr, hdr, port);
1211                 break;
1212         case 6:
1213                 cma_save_ip6_info((struct sockaddr_in6 *)src_addr,
1214                                   (struct sockaddr_in6 *)dst_addr, hdr, port);
1215                 break;
1216         default:
1217                 return -EAFNOSUPPORT;
1218         }
1219
1220         return 0;
1221 }
1222
1223 static int cma_save_net_info(struct sockaddr *src_addr,
1224                              struct sockaddr *dst_addr,
1225                              struct rdma_cm_id *listen_id,
1226                              struct ib_cm_event *ib_event,
1227                              sa_family_t sa_family, __be64 service_id)
1228 {
1229         if (sa_family == AF_IB) {
1230                 if (ib_event->event == IB_CM_REQ_RECEIVED)
1231                         cma_save_ib_info(src_addr, dst_addr, listen_id,
1232                                          ib_event->param.req_rcvd.primary_path);
1233                 else if (ib_event->event == IB_CM_SIDR_REQ_RECEIVED)
1234                         cma_save_ib_info(src_addr, dst_addr, listen_id, NULL);
1235                 return 0;
1236         }
1237
1238         return cma_save_ip_info(src_addr, dst_addr, ib_event, service_id);
1239 }
1240
1241 static int cma_save_req_info(const struct ib_cm_event *ib_event,
1242                              struct cma_req_info *req)
1243 {
1244         const struct ib_cm_req_event_param *req_param =
1245                 &ib_event->param.req_rcvd;
1246         const struct ib_cm_sidr_req_event_param *sidr_param =
1247                 &ib_event->param.sidr_req_rcvd;
1248
1249         switch (ib_event->event) {
1250         case IB_CM_REQ_RECEIVED:
1251                 req->device     = req_param->listen_id->device;
1252                 req->port       = req_param->port;
1253                 memcpy(&req->local_gid, &req_param->primary_path->sgid,
1254                        sizeof(req->local_gid));
1255                 req->has_gid    = true;
1256                 req->service_id = req_param->primary_path->service_id;
1257                 req->pkey       = be16_to_cpu(req_param->primary_path->pkey);
1258                 if (req->pkey != req_param->bth_pkey)
1259                         pr_warn_ratelimited("RDMA CMA: got different BTH P_Key (0x%x) and primary path P_Key (0x%x)\n"
1260                                             "RDMA CMA: in the future this may cause the request to be dropped\n",
1261                                             req_param->bth_pkey, req->pkey);
1262                 break;
1263         case IB_CM_SIDR_REQ_RECEIVED:
1264                 req->device     = sidr_param->listen_id->device;
1265                 req->port       = sidr_param->port;
1266                 req->has_gid    = false;
1267                 req->service_id = sidr_param->service_id;
1268                 req->pkey       = sidr_param->pkey;
1269                 if (req->pkey != sidr_param->bth_pkey)
1270                         pr_warn_ratelimited("RDMA CMA: got different BTH P_Key (0x%x) and SIDR request payload P_Key (0x%x)\n"
1271                                             "RDMA CMA: in the future this may cause the request to be dropped\n",
1272                                             sidr_param->bth_pkey, req->pkey);
1273                 break;
1274         default:
1275                 return -EINVAL;
1276         }
1277
1278         return 0;
1279 }
1280
1281 static bool validate_ipv4_net_dev(struct net_device *net_dev,
1282                                   const struct sockaddr_in *dst_addr,
1283                                   const struct sockaddr_in *src_addr)
1284 {
1285         __be32 daddr = dst_addr->sin_addr.s_addr,
1286                saddr = src_addr->sin_addr.s_addr;
1287         struct fib_result res;
1288         struct flowi4 fl4;
1289         int err;
1290         bool ret;
1291
1292         if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1293             ipv4_is_lbcast(daddr) || ipv4_is_zeronet(saddr) ||
1294             ipv4_is_zeronet(daddr) || ipv4_is_loopback(daddr) ||
1295             ipv4_is_loopback(saddr))
1296                 return false;
1297
1298         memset(&fl4, 0, sizeof(fl4));
1299         fl4.flowi4_iif = net_dev->ifindex;
1300         fl4.daddr = daddr;
1301         fl4.saddr = saddr;
1302
1303         rcu_read_lock();
1304         err = fib_lookup(dev_net(net_dev), &fl4, &res, 0);
1305         ret = err == 0 && FIB_RES_DEV(res) == net_dev;
1306         rcu_read_unlock();
1307
1308         return ret;
1309 }
1310
1311 static bool validate_ipv6_net_dev(struct net_device *net_dev,
1312                                   const struct sockaddr_in6 *dst_addr,
1313                                   const struct sockaddr_in6 *src_addr)
1314 {
1315 #if IS_ENABLED(CONFIG_IPV6)
1316         const int strict = ipv6_addr_type(&dst_addr->sin6_addr) &
1317                            IPV6_ADDR_LINKLOCAL;
1318         struct rt6_info *rt = rt6_lookup(dev_net(net_dev), &dst_addr->sin6_addr,
1319                                          &src_addr->sin6_addr, net_dev->ifindex,
1320                                          strict);
1321         bool ret;
1322
1323         if (!rt)
1324                 return false;
1325
1326         ret = rt->rt6i_idev->dev == net_dev;
1327         ip6_rt_put(rt);
1328
1329         return ret;
1330 #else
1331         return false;
1332 #endif
1333 }
1334
1335 static bool validate_net_dev(struct net_device *net_dev,
1336                              const struct sockaddr *daddr,
1337                              const struct sockaddr *saddr)
1338 {
1339         const struct sockaddr_in *daddr4 = (const struct sockaddr_in *)daddr;
1340         const struct sockaddr_in *saddr4 = (const struct sockaddr_in *)saddr;
1341         const struct sockaddr_in6 *daddr6 = (const struct sockaddr_in6 *)daddr;
1342         const struct sockaddr_in6 *saddr6 = (const struct sockaddr_in6 *)saddr;
1343
1344         switch (daddr->sa_family) {
1345         case AF_INET:
1346                 return saddr->sa_family == AF_INET &&
1347                        validate_ipv4_net_dev(net_dev, daddr4, saddr4);
1348
1349         case AF_INET6:
1350                 return saddr->sa_family == AF_INET6 &&
1351                        validate_ipv6_net_dev(net_dev, daddr6, saddr6);
1352
1353         default:
1354                 return false;
1355         }
1356 }
1357
1358 static struct net_device *cma_get_net_dev(struct ib_cm_event *ib_event,
1359                                           const struct cma_req_info *req)
1360 {
1361         struct sockaddr_storage listen_addr_storage, src_addr_storage;
1362         struct sockaddr *listen_addr = (struct sockaddr *)&listen_addr_storage,
1363                         *src_addr = (struct sockaddr *)&src_addr_storage;
1364         struct net_device *net_dev;
1365         const union ib_gid *gid = req->has_gid ? &req->local_gid : NULL;
1366         int err;
1367
1368         err = cma_save_ip_info(listen_addr, src_addr, ib_event,
1369                                req->service_id);
1370         if (err)
1371                 return ERR_PTR(err);
1372
1373         net_dev = ib_get_net_dev_by_params(req->device, req->port, req->pkey,
1374                                            gid, listen_addr);
1375         if (!net_dev)
1376                 return ERR_PTR(-ENODEV);
1377
1378         if (!validate_net_dev(net_dev, listen_addr, src_addr)) {
1379                 dev_put(net_dev);
1380                 return ERR_PTR(-EHOSTUNREACH);
1381         }
1382
1383         return net_dev;
1384 }
1385
1386 static enum rdma_port_space rdma_ps_from_service_id(__be64 service_id)
1387 {
1388         return (be64_to_cpu(service_id) >> 16) & 0xffff;
1389 }
1390
1391 static bool cma_match_private_data(struct rdma_id_private *id_priv,
1392                                    const struct cma_hdr *hdr)
1393 {
1394         struct sockaddr *addr = cma_src_addr(id_priv);
1395         __be32 ip4_addr;
1396         struct in6_addr ip6_addr;
1397
1398         if (cma_any_addr(addr) && !id_priv->afonly)
1399                 return true;
1400
1401         switch (addr->sa_family) {
1402         case AF_INET:
1403                 ip4_addr = ((struct sockaddr_in *)addr)->sin_addr.s_addr;
1404                 if (cma_get_ip_ver(hdr) != 4)
1405                         return false;
1406                 if (!cma_any_addr(addr) &&
1407                     hdr->dst_addr.ip4.addr != ip4_addr)
1408                         return false;
1409                 break;
1410         case AF_INET6:
1411                 ip6_addr = ((struct sockaddr_in6 *)addr)->sin6_addr;
1412                 if (cma_get_ip_ver(hdr) != 6)
1413                         return false;
1414                 if (!cma_any_addr(addr) &&
1415                     memcmp(&hdr->dst_addr.ip6, &ip6_addr, sizeof(ip6_addr)))
1416                         return false;
1417                 break;
1418         case AF_IB:
1419                 return true;
1420         default:
1421                 return false;
1422         }
1423
1424         return true;
1425 }
1426
1427 static bool cma_protocol_roce_dev_port(struct ib_device *device, int port_num)
1428 {
1429         enum rdma_link_layer ll = rdma_port_get_link_layer(device, port_num);
1430         enum rdma_transport_type transport =
1431                 rdma_node_get_transport(device->node_type);
1432
1433         return ll == IB_LINK_LAYER_ETHERNET && transport == RDMA_TRANSPORT_IB;
1434 }
1435
1436 static bool cma_protocol_roce(const struct rdma_cm_id *id)
1437 {
1438         struct ib_device *device = id->device;
1439         const int port_num = id->port_num ?: rdma_start_port(device);
1440
1441         return cma_protocol_roce_dev_port(device, port_num);
1442 }
1443
1444 static bool cma_match_net_dev(const struct rdma_cm_id *id,
1445                               const struct net_device *net_dev,
1446                               u8 port_num)
1447 {
1448         const struct rdma_addr *addr = &id->route.addr;
1449
1450         if (!net_dev)
1451                 /* This request is an AF_IB request or a RoCE request */
1452                 return (!id->port_num || id->port_num == port_num) &&
1453                        (addr->src_addr.ss_family == AF_IB ||
1454                         cma_protocol_roce_dev_port(id->device, port_num));
1455
1456         return !addr->dev_addr.bound_dev_if ||
1457                (net_eq(dev_net(net_dev), addr->dev_addr.net) &&
1458                 addr->dev_addr.bound_dev_if == net_dev->ifindex);
1459 }
1460
1461 static struct rdma_id_private *cma_find_listener(
1462                 const struct rdma_bind_list *bind_list,
1463                 const struct ib_cm_id *cm_id,
1464                 const struct ib_cm_event *ib_event,
1465                 const struct cma_req_info *req,
1466                 const struct net_device *net_dev)
1467 {
1468         struct rdma_id_private *id_priv, *id_priv_dev;
1469
1470         if (!bind_list)
1471                 return ERR_PTR(-EINVAL);
1472
1473         hlist_for_each_entry(id_priv, &bind_list->owners, node) {
1474                 if (cma_match_private_data(id_priv, ib_event->private_data)) {
1475                         if (id_priv->id.device == cm_id->device &&
1476                             cma_match_net_dev(&id_priv->id, net_dev, req->port))
1477                                 return id_priv;
1478                         list_for_each_entry(id_priv_dev,
1479                                             &id_priv->listen_list,
1480                                             listen_list) {
1481                                 if (id_priv_dev->id.device == cm_id->device &&
1482                                     cma_match_net_dev(&id_priv_dev->id, net_dev, req->port))
1483                                         return id_priv_dev;
1484                         }
1485                 }
1486         }
1487
1488         return ERR_PTR(-EINVAL);
1489 }
1490
1491 static struct rdma_id_private *cma_id_from_event(struct ib_cm_id *cm_id,
1492                                                  struct ib_cm_event *ib_event,
1493                                                  struct net_device **net_dev)
1494 {
1495         struct cma_req_info req;
1496         struct rdma_bind_list *bind_list;
1497         struct rdma_id_private *id_priv;
1498         int err;
1499
1500         err = cma_save_req_info(ib_event, &req);
1501         if (err)
1502                 return ERR_PTR(err);
1503
1504         *net_dev = cma_get_net_dev(ib_event, &req);
1505         if (IS_ERR(*net_dev)) {
1506                 if (PTR_ERR(*net_dev) == -EAFNOSUPPORT) {
1507                         /* Assuming the protocol is AF_IB */
1508                         *net_dev = NULL;
1509                 } else if (cma_protocol_roce_dev_port(req.device, req.port)) {
1510                         /* TODO find the net dev matching the request parameters
1511                          * through the RoCE GID table */
1512                         *net_dev = NULL;
1513                 } else {
1514                         return ERR_CAST(*net_dev);
1515                 }
1516         }
1517
1518         bind_list = cma_ps_find(*net_dev ? dev_net(*net_dev) : &init_net,
1519                                 rdma_ps_from_service_id(req.service_id),
1520                                 cma_port_from_service_id(req.service_id));
1521         id_priv = cma_find_listener(bind_list, cm_id, ib_event, &req, *net_dev);
1522         if (IS_ERR(id_priv) && *net_dev) {
1523                 dev_put(*net_dev);
1524                 *net_dev = NULL;
1525         }
1526
1527         return id_priv;
1528 }
1529
1530 static inline int cma_user_data_offset(struct rdma_id_private *id_priv)
1531 {
1532         return cma_family(id_priv) == AF_IB ? 0 : sizeof(struct cma_hdr);
1533 }
1534
1535 static void cma_cancel_route(struct rdma_id_private *id_priv)
1536 {
1537         if (rdma_cap_ib_sa(id_priv->id.device, id_priv->id.port_num)) {
1538                 if (id_priv->query)
1539                         ib_sa_cancel_query(id_priv->query_id, id_priv->query);
1540         }
1541 }
1542
1543 static void cma_cancel_listens(struct rdma_id_private *id_priv)
1544 {
1545         struct rdma_id_private *dev_id_priv;
1546
1547         /*
1548          * Remove from listen_any_list to prevent added devices from spawning
1549          * additional listen requests.
1550          */
1551         mutex_lock(&lock);
1552         list_del(&id_priv->list);
1553
1554         while (!list_empty(&id_priv->listen_list)) {
1555                 dev_id_priv = list_entry(id_priv->listen_list.next,
1556                                          struct rdma_id_private, listen_list);
1557                 /* sync with device removal to avoid duplicate destruction */
1558                 list_del_init(&dev_id_priv->list);
1559                 list_del(&dev_id_priv->listen_list);
1560                 mutex_unlock(&lock);
1561
1562                 rdma_destroy_id(&dev_id_priv->id);
1563                 mutex_lock(&lock);
1564         }
1565         mutex_unlock(&lock);
1566 }
1567
1568 static void cma_cancel_operation(struct rdma_id_private *id_priv,
1569                                  enum rdma_cm_state state)
1570 {
1571         switch (state) {
1572         case RDMA_CM_ADDR_QUERY:
1573                 rdma_addr_cancel(&id_priv->id.route.addr.dev_addr);
1574                 break;
1575         case RDMA_CM_ROUTE_QUERY:
1576                 cma_cancel_route(id_priv);
1577                 break;
1578         case RDMA_CM_LISTEN:
1579                 if (cma_any_addr(cma_src_addr(id_priv)) && !id_priv->cma_dev)
1580                         cma_cancel_listens(id_priv);
1581                 break;
1582         default:
1583                 break;
1584         }
1585 }
1586
1587 static void cma_release_port(struct rdma_id_private *id_priv)
1588 {
1589         struct rdma_bind_list *bind_list = id_priv->bind_list;
1590         struct net *net = id_priv->id.route.addr.dev_addr.net;
1591
1592         if (!bind_list)
1593                 return;
1594
1595         mutex_lock(&lock);
1596         hlist_del(&id_priv->node);
1597         if (hlist_empty(&bind_list->owners)) {
1598                 cma_ps_remove(net, bind_list->ps, bind_list->port);
1599                 kfree(bind_list);
1600         }
1601         mutex_unlock(&lock);
1602 }
1603
1604 static void cma_leave_mc_groups(struct rdma_id_private *id_priv)
1605 {
1606         struct cma_multicast *mc;
1607
1608         while (!list_empty(&id_priv->mc_list)) {
1609                 mc = container_of(id_priv->mc_list.next,
1610                                   struct cma_multicast, list);
1611                 list_del(&mc->list);
1612                 if (rdma_cap_ib_mcast(id_priv->cma_dev->device,
1613                                       id_priv->id.port_num)) {
1614                         ib_sa_free_multicast(mc->multicast.ib);
1615                         kfree(mc);
1616                 } else {
1617                         if (mc->igmp_joined) {
1618                                 struct rdma_dev_addr *dev_addr =
1619                                         &id_priv->id.route.addr.dev_addr;
1620                                 struct net_device *ndev = NULL;
1621
1622                                 if (dev_addr->bound_dev_if)
1623                                         ndev = dev_get_by_index(&init_net,
1624                                                                 dev_addr->bound_dev_if);
1625                                 if (ndev) {
1626                                         cma_igmp_send(ndev,
1627                                                       &mc->multicast.ib->rec.mgid,
1628                                                       false);
1629                                         dev_put(ndev);
1630                                 }
1631                         }
1632                         kref_put(&mc->mcref, release_mc);
1633                 }
1634         }
1635 }
1636
1637 void rdma_destroy_id(struct rdma_cm_id *id)
1638 {
1639         struct rdma_id_private *id_priv;
1640         enum rdma_cm_state state;
1641
1642         id_priv = container_of(id, struct rdma_id_private, id);
1643         state = cma_exch(id_priv, RDMA_CM_DESTROYING);
1644         cma_cancel_operation(id_priv, state);
1645
1646         /*
1647          * Wait for any active callback to finish.  New callbacks will find
1648          * the id_priv state set to destroying and abort.
1649          */
1650         mutex_lock(&id_priv->handler_mutex);
1651         mutex_unlock(&id_priv->handler_mutex);
1652
1653         if (id_priv->cma_dev) {
1654                 if (rdma_cap_ib_cm(id_priv->id.device, 1)) {
1655                         if (id_priv->cm_id.ib)
1656                                 ib_destroy_cm_id(id_priv->cm_id.ib);
1657                 } else if (rdma_cap_iw_cm(id_priv->id.device, 1)) {
1658                         if (id_priv->cm_id.iw)
1659                                 iw_destroy_cm_id(id_priv->cm_id.iw);
1660                 }
1661                 cma_leave_mc_groups(id_priv);
1662                 cma_release_dev(id_priv);
1663         }
1664
1665         cma_release_port(id_priv);
1666         cma_deref_id(id_priv);
1667         wait_for_completion(&id_priv->comp);
1668
1669         if (id_priv->internal_id)
1670                 cma_deref_id(id_priv->id.context);
1671
1672         kfree(id_priv->id.route.path_rec);
1673         put_net(id_priv->id.route.addr.dev_addr.net);
1674         kfree(id_priv);
1675 }
1676 EXPORT_SYMBOL(rdma_destroy_id);
1677
1678 static int cma_rep_recv(struct rdma_id_private *id_priv)
1679 {
1680         int ret;
1681
1682         ret = cma_modify_qp_rtr(id_priv, NULL);
1683         if (ret)
1684                 goto reject;
1685
1686         ret = cma_modify_qp_rts(id_priv, NULL);
1687         if (ret)
1688                 goto reject;
1689
1690         ret = ib_send_cm_rtu(id_priv->cm_id.ib, NULL, 0);
1691         if (ret)
1692                 goto reject;
1693
1694         return 0;
1695 reject:
1696         cma_modify_qp_err(id_priv);
1697         ib_send_cm_rej(id_priv->cm_id.ib, IB_CM_REJ_CONSUMER_DEFINED,
1698                        NULL, 0, NULL, 0);
1699         return ret;
1700 }
1701
1702 static void cma_set_rep_event_data(struct rdma_cm_event *event,
1703                                    struct ib_cm_rep_event_param *rep_data,
1704                                    void *private_data)
1705 {
1706         event->param.conn.private_data = private_data;
1707         event->param.conn.private_data_len = IB_CM_REP_PRIVATE_DATA_SIZE;
1708         event->param.conn.responder_resources = rep_data->responder_resources;
1709         event->param.conn.initiator_depth = rep_data->initiator_depth;
1710         event->param.conn.flow_control = rep_data->flow_control;
1711         event->param.conn.rnr_retry_count = rep_data->rnr_retry_count;
1712         event->param.conn.srq = rep_data->srq;
1713         event->param.conn.qp_num = rep_data->remote_qpn;
1714 }
1715
1716 static int cma_ib_handler(struct ib_cm_id *cm_id, struct ib_cm_event *ib_event)
1717 {
1718         struct rdma_id_private *id_priv = cm_id->context;
1719         struct rdma_cm_event event;
1720         int ret = 0;
1721
1722         mutex_lock(&id_priv->handler_mutex);
1723         if ((ib_event->event != IB_CM_TIMEWAIT_EXIT &&
1724              id_priv->state != RDMA_CM_CONNECT) ||
1725             (ib_event->event == IB_CM_TIMEWAIT_EXIT &&
1726              id_priv->state != RDMA_CM_DISCONNECT))
1727                 goto out;
1728
1729         memset(&event, 0, sizeof event);
1730         switch (ib_event->event) {
1731         case IB_CM_REQ_ERROR:
1732         case IB_CM_REP_ERROR:
1733                 event.event = RDMA_CM_EVENT_UNREACHABLE;
1734                 event.status = -ETIMEDOUT;
1735                 break;
1736         case IB_CM_REP_RECEIVED:
1737                 if (id_priv->id.qp) {
1738                         event.status = cma_rep_recv(id_priv);
1739                         event.event = event.status ? RDMA_CM_EVENT_CONNECT_ERROR :
1740                                                      RDMA_CM_EVENT_ESTABLISHED;
1741                 } else {
1742                         event.event = RDMA_CM_EVENT_CONNECT_RESPONSE;
1743                 }
1744                 cma_set_rep_event_data(&event, &ib_event->param.rep_rcvd,
1745                                        ib_event->private_data);
1746                 break;
1747         case IB_CM_RTU_RECEIVED:
1748         case IB_CM_USER_ESTABLISHED:
1749                 event.event = RDMA_CM_EVENT_ESTABLISHED;
1750                 break;
1751         case IB_CM_DREQ_ERROR:
1752                 event.status = -ETIMEDOUT; /* fall through */
1753         case IB_CM_DREQ_RECEIVED:
1754         case IB_CM_DREP_RECEIVED:
1755                 if (!cma_comp_exch(id_priv, RDMA_CM_CONNECT,
1756                                    RDMA_CM_DISCONNECT))
1757                         goto out;
1758                 event.event = RDMA_CM_EVENT_DISCONNECTED;
1759                 break;
1760         case IB_CM_TIMEWAIT_EXIT:
1761                 event.event = RDMA_CM_EVENT_TIMEWAIT_EXIT;
1762                 break;
1763         case IB_CM_MRA_RECEIVED:
1764                 /* ignore event */
1765                 goto out;
1766         case IB_CM_REJ_RECEIVED:
1767                 cma_modify_qp_err(id_priv);
1768                 event.status = ib_event->param.rej_rcvd.reason;
1769                 event.event = RDMA_CM_EVENT_REJECTED;
1770                 event.param.conn.private_data = ib_event->private_data;
1771                 event.param.conn.private_data_len = IB_CM_REJ_PRIVATE_DATA_SIZE;
1772                 break;
1773         default:
1774                 pr_err("RDMA CMA: unexpected IB CM event: %d\n",
1775                        ib_event->event);
1776                 goto out;
1777         }
1778
1779         ret = id_priv->id.event_handler(&id_priv->id, &event);
1780         if (ret) {
1781                 /* Destroy the CM ID by returning a non-zero value. */
1782                 id_priv->cm_id.ib = NULL;
1783                 cma_exch(id_priv, RDMA_CM_DESTROYING);
1784                 mutex_unlock(&id_priv->handler_mutex);
1785                 rdma_destroy_id(&id_priv->id);
1786                 return ret;
1787         }
1788 out:
1789         mutex_unlock(&id_priv->handler_mutex);
1790         return ret;
1791 }
1792
1793 static struct rdma_id_private *cma_new_conn_id(struct rdma_cm_id *listen_id,
1794                                                struct ib_cm_event *ib_event,
1795                                                struct net_device *net_dev)
1796 {
1797         struct rdma_id_private *id_priv;
1798         struct rdma_cm_id *id;
1799         struct rdma_route *rt;
1800         const sa_family_t ss_family = listen_id->route.addr.src_addr.ss_family;
1801         const __be64 service_id =
1802                       ib_event->param.req_rcvd.primary_path->service_id;
1803         int ret;
1804
1805         id = rdma_create_id(listen_id->route.addr.dev_addr.net,
1806                             listen_id->event_handler, listen_id->context,
1807                             listen_id->ps, ib_event->param.req_rcvd.qp_type);
1808         if (IS_ERR(id))
1809                 return NULL;
1810
1811         id_priv = container_of(id, struct rdma_id_private, id);
1812         if (cma_save_net_info((struct sockaddr *)&id->route.addr.src_addr,
1813                               (struct sockaddr *)&id->route.addr.dst_addr,
1814                               listen_id, ib_event, ss_family, service_id))
1815                 goto err;
1816
1817         rt = &id->route;
1818         rt->num_paths = ib_event->param.req_rcvd.alternate_path ? 2 : 1;
1819         rt->path_rec = kmalloc(sizeof *rt->path_rec * rt->num_paths,
1820                                GFP_KERNEL);
1821         if (!rt->path_rec)
1822                 goto err;
1823
1824         rt->path_rec[0] = *ib_event->param.req_rcvd.primary_path;
1825         if (rt->num_paths == 2)
1826                 rt->path_rec[1] = *ib_event->param.req_rcvd.alternate_path;
1827
1828         if (net_dev) {
1829                 ret = rdma_copy_addr(&rt->addr.dev_addr, net_dev, NULL);
1830                 if (ret)
1831                         goto err;
1832         } else {
1833                 if (!cma_protocol_roce(listen_id) &&
1834                     cma_any_addr(cma_src_addr(id_priv))) {
1835                         rt->addr.dev_addr.dev_type = ARPHRD_INFINIBAND;
1836                         rdma_addr_set_sgid(&rt->addr.dev_addr, &rt->path_rec[0].sgid);
1837                         ib_addr_set_pkey(&rt->addr.dev_addr, be16_to_cpu(rt->path_rec[0].pkey));
1838                 } else if (!cma_any_addr(cma_src_addr(id_priv))) {
1839                         ret = cma_translate_addr(cma_src_addr(id_priv), &rt->addr.dev_addr);
1840                         if (ret)
1841                                 goto err;
1842                 }
1843         }
1844         rdma_addr_set_dgid(&rt->addr.dev_addr, &rt->path_rec[0].dgid);
1845
1846         id_priv->state = RDMA_CM_CONNECT;
1847         return id_priv;
1848
1849 err:
1850         rdma_destroy_id(id);
1851         return NULL;
1852 }
1853
1854 static struct rdma_id_private *cma_new_udp_id(struct rdma_cm_id *listen_id,
1855                                               struct ib_cm_event *ib_event,
1856                                               struct net_device *net_dev)
1857 {
1858         struct rdma_id_private *id_priv;
1859         struct rdma_cm_id *id;
1860         const sa_family_t ss_family = listen_id->route.addr.src_addr.ss_family;
1861         struct net *net = listen_id->route.addr.dev_addr.net;
1862         int ret;
1863
1864         id = rdma_create_id(net, listen_id->event_handler, listen_id->context,
1865                             listen_id->ps, IB_QPT_UD);
1866         if (IS_ERR(id))
1867                 return NULL;
1868
1869         id_priv = container_of(id, struct rdma_id_private, id);
1870         if (cma_save_net_info((struct sockaddr *)&id->route.addr.src_addr,
1871                               (struct sockaddr *)&id->route.addr.dst_addr,
1872                               listen_id, ib_event, ss_family,
1873                               ib_event->param.sidr_req_rcvd.service_id))
1874                 goto err;
1875
1876         if (net_dev) {
1877                 ret = rdma_copy_addr(&id->route.addr.dev_addr, net_dev, NULL);
1878                 if (ret)
1879                         goto err;
1880         } else {
1881                 if (!cma_any_addr(cma_src_addr(id_priv))) {
1882                         ret = cma_translate_addr(cma_src_addr(id_priv),
1883                                                  &id->route.addr.dev_addr);
1884                         if (ret)
1885                                 goto err;
1886                 }
1887         }
1888
1889         id_priv->state = RDMA_CM_CONNECT;
1890         return id_priv;
1891 err:
1892         rdma_destroy_id(id);
1893         return NULL;
1894 }
1895
1896 static void cma_set_req_event_data(struct rdma_cm_event *event,
1897                                    struct ib_cm_req_event_param *req_data,
1898                                    void *private_data, int offset)
1899 {
1900         event->param.conn.private_data = private_data + offset;
1901         event->param.conn.private_data_len = IB_CM_REQ_PRIVATE_DATA_SIZE - offset;
1902         event->param.conn.responder_resources = req_data->responder_resources;
1903         event->param.conn.initiator_depth = req_data->initiator_depth;
1904         event->param.conn.flow_control = req_data->flow_control;
1905         event->param.conn.retry_count = req_data->retry_count;
1906         event->param.conn.rnr_retry_count = req_data->rnr_retry_count;
1907         event->param.conn.srq = req_data->srq;
1908         event->param.conn.qp_num = req_data->remote_qpn;
1909 }
1910
1911 static int cma_check_req_qp_type(struct rdma_cm_id *id, struct ib_cm_event *ib_event)
1912 {
1913         return (((ib_event->event == IB_CM_REQ_RECEIVED) &&
1914                  (ib_event->param.req_rcvd.qp_type == id->qp_type)) ||
1915                 ((ib_event->event == IB_CM_SIDR_REQ_RECEIVED) &&
1916                  (id->qp_type == IB_QPT_UD)) ||
1917                 (!id->qp_type));
1918 }
1919
1920 static int cma_req_handler(struct ib_cm_id *cm_id, struct ib_cm_event *ib_event)
1921 {
1922         struct rdma_id_private *listen_id, *conn_id = NULL;
1923         struct rdma_cm_event event;
1924         struct net_device *net_dev;
1925         int offset, ret;
1926
1927         listen_id = cma_id_from_event(cm_id, ib_event, &net_dev);
1928         if (IS_ERR(listen_id))
1929                 return PTR_ERR(listen_id);
1930
1931         if (!cma_check_req_qp_type(&listen_id->id, ib_event)) {
1932                 ret = -EINVAL;
1933                 goto net_dev_put;
1934         }
1935
1936         mutex_lock(&listen_id->handler_mutex);
1937         if (listen_id->state != RDMA_CM_LISTEN) {
1938                 ret = -ECONNABORTED;
1939                 goto err1;
1940         }
1941
1942         memset(&event, 0, sizeof event);
1943         offset = cma_user_data_offset(listen_id);
1944         event.event = RDMA_CM_EVENT_CONNECT_REQUEST;
1945         if (ib_event->event == IB_CM_SIDR_REQ_RECEIVED) {
1946                 conn_id = cma_new_udp_id(&listen_id->id, ib_event, net_dev);
1947                 event.param.ud.private_data = ib_event->private_data + offset;
1948                 event.param.ud.private_data_len =
1949                                 IB_CM_SIDR_REQ_PRIVATE_DATA_SIZE - offset;
1950         } else {
1951                 conn_id = cma_new_conn_id(&listen_id->id, ib_event, net_dev);
1952                 cma_set_req_event_data(&event, &ib_event->param.req_rcvd,
1953                                        ib_event->private_data, offset);
1954         }
1955         if (!conn_id) {
1956                 ret = -ENOMEM;
1957                 goto err1;
1958         }
1959
1960         mutex_lock_nested(&conn_id->handler_mutex, SINGLE_DEPTH_NESTING);
1961         ret = cma_acquire_dev(conn_id, listen_id);
1962         if (ret)
1963                 goto err2;
1964
1965         conn_id->cm_id.ib = cm_id;
1966         cm_id->context = conn_id;
1967         cm_id->cm_handler = cma_ib_handler;
1968
1969         /*
1970          * Protect against the user destroying conn_id from another thread
1971          * until we're done accessing it.
1972          */
1973         atomic_inc(&conn_id->refcount);
1974         ret = conn_id->id.event_handler(&conn_id->id, &event);
1975         if (ret)
1976                 goto err3;
1977         /*
1978          * Acquire mutex to prevent user executing rdma_destroy_id()
1979          * while we're accessing the cm_id.
1980          */
1981         mutex_lock(&lock);
1982         if (cma_comp(conn_id, RDMA_CM_CONNECT) &&
1983             (conn_id->id.qp_type != IB_QPT_UD))
1984                 ib_send_cm_mra(cm_id, CMA_CM_MRA_SETTING, NULL, 0);
1985         mutex_unlock(&lock);
1986         mutex_unlock(&conn_id->handler_mutex);
1987         mutex_unlock(&listen_id->handler_mutex);
1988         cma_deref_id(conn_id);
1989         if (net_dev)
1990                 dev_put(net_dev);
1991         return 0;
1992
1993 err3:
1994         cma_deref_id(conn_id);
1995         /* Destroy the CM ID by returning a non-zero value. */
1996         conn_id->cm_id.ib = NULL;
1997 err2:
1998         cma_exch(conn_id, RDMA_CM_DESTROYING);
1999         mutex_unlock(&conn_id->handler_mutex);
2000 err1:
2001         mutex_unlock(&listen_id->handler_mutex);
2002         if (conn_id)
2003                 rdma_destroy_id(&conn_id->id);
2004
2005 net_dev_put:
2006         if (net_dev)
2007                 dev_put(net_dev);
2008
2009         return ret;
2010 }
2011
2012 __be64 rdma_get_service_id(struct rdma_cm_id *id, struct sockaddr *addr)
2013 {
2014         if (addr->sa_family == AF_IB)
2015                 return ((struct sockaddr_ib *) addr)->sib_sid;
2016
2017         return cpu_to_be64(((u64)id->ps << 16) + be16_to_cpu(cma_port(addr)));
2018 }
2019 EXPORT_SYMBOL(rdma_get_service_id);
2020
2021 static int cma_iw_handler(struct iw_cm_id *iw_id, struct iw_cm_event *iw_event)
2022 {
2023         struct rdma_id_private *id_priv = iw_id->context;
2024         struct rdma_cm_event event;
2025         int ret = 0;
2026         struct sockaddr *laddr = (struct sockaddr *)&iw_event->local_addr;
2027         struct sockaddr *raddr = (struct sockaddr *)&iw_event->remote_addr;
2028
2029         mutex_lock(&id_priv->handler_mutex);
2030         if (id_priv->state != RDMA_CM_CONNECT)
2031                 goto out;
2032
2033         memset(&event, 0, sizeof event);
2034         switch (iw_event->event) {
2035         case IW_CM_EVENT_CLOSE:
2036                 event.event = RDMA_CM_EVENT_DISCONNECTED;
2037                 break;
2038         case IW_CM_EVENT_CONNECT_REPLY:
2039                 memcpy(cma_src_addr(id_priv), laddr,
2040                        rdma_addr_size(laddr));
2041                 memcpy(cma_dst_addr(id_priv), raddr,
2042                        rdma_addr_size(raddr));
2043                 switch (iw_event->status) {
2044                 case 0:
2045                         event.event = RDMA_CM_EVENT_ESTABLISHED;
2046                         event.param.conn.initiator_depth = iw_event->ird;
2047                         event.param.conn.responder_resources = iw_event->ord;
2048                         break;
2049                 case -ECONNRESET:
2050                 case -ECONNREFUSED:
2051                         event.event = RDMA_CM_EVENT_REJECTED;
2052                         break;
2053                 case -ETIMEDOUT:
2054                         event.event = RDMA_CM_EVENT_UNREACHABLE;
2055                         break;
2056                 default:
2057                         event.event = RDMA_CM_EVENT_CONNECT_ERROR;
2058                         break;
2059                 }
2060                 break;
2061         case IW_CM_EVENT_ESTABLISHED:
2062                 event.event = RDMA_CM_EVENT_ESTABLISHED;
2063                 event.param.conn.initiator_depth = iw_event->ird;
2064                 event.param.conn.responder_resources = iw_event->ord;
2065                 break;
2066         default:
2067                 BUG_ON(1);
2068         }
2069
2070         event.status = iw_event->status;
2071         event.param.conn.private_data = iw_event->private_data;
2072         event.param.conn.private_data_len = iw_event->private_data_len;
2073         ret = id_priv->id.event_handler(&id_priv->id, &event);
2074         if (ret) {
2075                 /* Destroy the CM ID by returning a non-zero value. */
2076                 id_priv->cm_id.iw = NULL;
2077                 cma_exch(id_priv, RDMA_CM_DESTROYING);
2078                 mutex_unlock(&id_priv->handler_mutex);
2079                 rdma_destroy_id(&id_priv->id);
2080                 return ret;
2081         }
2082
2083 out:
2084         mutex_unlock(&id_priv->handler_mutex);
2085         return ret;
2086 }
2087
2088 static int iw_conn_req_handler(struct iw_cm_id *cm_id,
2089                                struct iw_cm_event *iw_event)
2090 {
2091         struct rdma_cm_id *new_cm_id;
2092         struct rdma_id_private *listen_id, *conn_id;
2093         struct rdma_cm_event event;
2094         int ret = -ECONNABORTED;
2095         struct sockaddr *laddr = (struct sockaddr *)&iw_event->local_addr;
2096         struct sockaddr *raddr = (struct sockaddr *)&iw_event->remote_addr;
2097
2098         listen_id = cm_id->context;
2099
2100         mutex_lock(&listen_id->handler_mutex);
2101         if (listen_id->state != RDMA_CM_LISTEN)
2102                 goto out;
2103
2104         /* Create a new RDMA id for the new IW CM ID */
2105         new_cm_id = rdma_create_id(listen_id->id.route.addr.dev_addr.net,
2106                                    listen_id->id.event_handler,
2107                                    listen_id->id.context,
2108                                    RDMA_PS_TCP, IB_QPT_RC);
2109         if (IS_ERR(new_cm_id)) {
2110                 ret = -ENOMEM;
2111                 goto out;
2112         }
2113         conn_id = container_of(new_cm_id, struct rdma_id_private, id);
2114         mutex_lock_nested(&conn_id->handler_mutex, SINGLE_DEPTH_NESTING);
2115         conn_id->state = RDMA_CM_CONNECT;
2116
2117         ret = rdma_translate_ip(laddr, &conn_id->id.route.addr.dev_addr, NULL);
2118         if (ret) {
2119                 mutex_unlock(&conn_id->handler_mutex);
2120                 rdma_destroy_id(new_cm_id);
2121                 goto out;
2122         }
2123
2124         ret = cma_acquire_dev(conn_id, listen_id);
2125         if (ret) {
2126                 mutex_unlock(&conn_id->handler_mutex);
2127                 rdma_destroy_id(new_cm_id);
2128                 goto out;
2129         }
2130
2131         conn_id->cm_id.iw = cm_id;
2132         cm_id->context = conn_id;
2133         cm_id->cm_handler = cma_iw_handler;
2134
2135         memcpy(cma_src_addr(conn_id), laddr, rdma_addr_size(laddr));
2136         memcpy(cma_dst_addr(conn_id), raddr, rdma_addr_size(raddr));
2137
2138         memset(&event, 0, sizeof event);
2139         event.event = RDMA_CM_EVENT_CONNECT_REQUEST;
2140         event.param.conn.private_data = iw_event->private_data;
2141         event.param.conn.private_data_len = iw_event->private_data_len;
2142         event.param.conn.initiator_depth = iw_event->ird;
2143         event.param.conn.responder_resources = iw_event->ord;
2144
2145         /*
2146          * Protect against the user destroying conn_id from another thread
2147          * until we're done accessing it.
2148          */
2149         atomic_inc(&conn_id->refcount);
2150         ret = conn_id->id.event_handler(&conn_id->id, &event);
2151         if (ret) {
2152                 /* User wants to destroy the CM ID */
2153                 conn_id->cm_id.iw = NULL;
2154                 cma_exch(conn_id, RDMA_CM_DESTROYING);
2155                 mutex_unlock(&conn_id->handler_mutex);
2156                 cma_deref_id(conn_id);
2157                 rdma_destroy_id(&conn_id->id);
2158                 goto out;
2159         }
2160
2161         mutex_unlock(&conn_id->handler_mutex);
2162         cma_deref_id(conn_id);
2163
2164 out:
2165         mutex_unlock(&listen_id->handler_mutex);
2166         return ret;
2167 }
2168
2169 static int cma_ib_listen(struct rdma_id_private *id_priv)
2170 {
2171         struct sockaddr *addr;
2172         struct ib_cm_id *id;
2173         __be64 svc_id;
2174
2175         addr = cma_src_addr(id_priv);
2176         svc_id = rdma_get_service_id(&id_priv->id, addr);
2177         id = ib_cm_insert_listen(id_priv->id.device, cma_req_handler, svc_id);
2178         if (IS_ERR(id))
2179                 return PTR_ERR(id);
2180         id_priv->cm_id.ib = id;
2181
2182         return 0;
2183 }
2184
2185 static int cma_iw_listen(struct rdma_id_private *id_priv, int backlog)
2186 {
2187         int ret;
2188         struct iw_cm_id *id;
2189
2190         id = iw_create_cm_id(id_priv->id.device,
2191                              iw_conn_req_handler,
2192                              id_priv);
2193         if (IS_ERR(id))
2194                 return PTR_ERR(id);
2195
2196         id->tos = id_priv->tos;
2197         id_priv->cm_id.iw = id;
2198
2199         memcpy(&id_priv->cm_id.iw->local_addr, cma_src_addr(id_priv),
2200                rdma_addr_size(cma_src_addr(id_priv)));
2201
2202         ret = iw_cm_listen(id_priv->cm_id.iw, backlog);
2203
2204         if (ret) {
2205                 iw_destroy_cm_id(id_priv->cm_id.iw);
2206                 id_priv->cm_id.iw = NULL;
2207         }
2208
2209         return ret;
2210 }
2211
2212 static int cma_listen_handler(struct rdma_cm_id *id,
2213                               struct rdma_cm_event *event)
2214 {
2215         struct rdma_id_private *id_priv = id->context;
2216
2217         id->context = id_priv->id.context;
2218         id->event_handler = id_priv->id.event_handler;
2219         return id_priv->id.event_handler(id, event);
2220 }
2221
2222 static void cma_listen_on_dev(struct rdma_id_private *id_priv,
2223                               struct cma_device *cma_dev)
2224 {
2225         struct rdma_id_private *dev_id_priv;
2226         struct rdma_cm_id *id;
2227         struct net *net = id_priv->id.route.addr.dev_addr.net;
2228         int ret;
2229
2230         if (cma_family(id_priv) == AF_IB && !rdma_cap_ib_cm(cma_dev->device, 1))
2231                 return;
2232
2233         id = rdma_create_id(net, cma_listen_handler, id_priv, id_priv->id.ps,
2234                             id_priv->id.qp_type);
2235         if (IS_ERR(id))
2236                 return;
2237
2238         dev_id_priv = container_of(id, struct rdma_id_private, id);
2239
2240         dev_id_priv->state = RDMA_CM_ADDR_BOUND;
2241         memcpy(cma_src_addr(dev_id_priv), cma_src_addr(id_priv),
2242                rdma_addr_size(cma_src_addr(id_priv)));
2243
2244         _cma_attach_to_dev(dev_id_priv, cma_dev);
2245         list_add_tail(&dev_id_priv->listen_list, &id_priv->listen_list);
2246         atomic_inc(&id_priv->refcount);
2247         dev_id_priv->internal_id = 1;
2248         dev_id_priv->afonly = id_priv->afonly;
2249
2250         ret = rdma_listen(id, id_priv->backlog);
2251         if (ret)
2252                 pr_warn("RDMA CMA: cma_listen_on_dev, error %d, listening on device %s\n",
2253                         ret, cma_dev->device->name);
2254 }
2255
2256 static void cma_listen_on_all(struct rdma_id_private *id_priv)
2257 {
2258         struct cma_device *cma_dev;
2259
2260         mutex_lock(&lock);
2261         list_add_tail(&id_priv->list, &listen_any_list);
2262         list_for_each_entry(cma_dev, &dev_list, list)
2263                 cma_listen_on_dev(id_priv, cma_dev);
2264         mutex_unlock(&lock);
2265 }
2266
2267 void rdma_set_service_type(struct rdma_cm_id *id, int tos)
2268 {
2269         struct rdma_id_private *id_priv;
2270
2271         id_priv = container_of(id, struct rdma_id_private, id);
2272         id_priv->tos = (u8) tos;
2273 }
2274 EXPORT_SYMBOL(rdma_set_service_type);
2275
2276 static void cma_query_handler(int status, struct ib_sa_path_rec *path_rec,
2277                               void *context)
2278 {
2279         struct cma_work *work = context;
2280         struct rdma_route *route;
2281
2282         route = &work->id->id.route;
2283
2284         if (!status) {
2285                 route->num_paths = 1;
2286                 *route->path_rec = *path_rec;
2287         } else {
2288                 work->old_state = RDMA_CM_ROUTE_QUERY;
2289                 work->new_state = RDMA_CM_ADDR_RESOLVED;
2290                 work->event.event = RDMA_CM_EVENT_ROUTE_ERROR;
2291                 work->event.status = status;
2292         }
2293
2294         queue_work(cma_wq, &work->work);
2295 }
2296
2297 static int cma_query_ib_route(struct rdma_id_private *id_priv, int timeout_ms,
2298                               struct cma_work *work)
2299 {
2300         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
2301         struct ib_sa_path_rec path_rec;
2302         ib_sa_comp_mask comp_mask;
2303         struct sockaddr_in6 *sin6;
2304         struct sockaddr_ib *sib;
2305
2306         memset(&path_rec, 0, sizeof path_rec);
2307         rdma_addr_get_sgid(dev_addr, &path_rec.sgid);
2308         rdma_addr_get_dgid(dev_addr, &path_rec.dgid);
2309         path_rec.pkey = cpu_to_be16(ib_addr_get_pkey(dev_addr));
2310         path_rec.numb_path = 1;
2311         path_rec.reversible = 1;
2312         path_rec.service_id = rdma_get_service_id(&id_priv->id, cma_dst_addr(id_priv));
2313
2314         comp_mask = IB_SA_PATH_REC_DGID | IB_SA_PATH_REC_SGID |
2315                     IB_SA_PATH_REC_PKEY | IB_SA_PATH_REC_NUMB_PATH |
2316                     IB_SA_PATH_REC_REVERSIBLE | IB_SA_PATH_REC_SERVICE_ID;
2317
2318         switch (cma_family(id_priv)) {
2319         case AF_INET:
2320                 path_rec.qos_class = cpu_to_be16((u16) id_priv->tos);
2321                 comp_mask |= IB_SA_PATH_REC_QOS_CLASS;
2322                 break;
2323         case AF_INET6:
2324                 sin6 = (struct sockaddr_in6 *) cma_src_addr(id_priv);
2325                 path_rec.traffic_class = (u8) (be32_to_cpu(sin6->sin6_flowinfo) >> 20);
2326                 comp_mask |= IB_SA_PATH_REC_TRAFFIC_CLASS;
2327                 break;
2328         case AF_IB:
2329                 sib = (struct sockaddr_ib *) cma_src_addr(id_priv);
2330                 path_rec.traffic_class = (u8) (be32_to_cpu(sib->sib_flowinfo) >> 20);
2331                 comp_mask |= IB_SA_PATH_REC_TRAFFIC_CLASS;
2332                 break;
2333         }
2334
2335         id_priv->query_id = ib_sa_path_rec_get(&sa_client, id_priv->id.device,
2336                                                id_priv->id.port_num, &path_rec,
2337                                                comp_mask, timeout_ms,
2338                                                GFP_KERNEL, cma_query_handler,
2339                                                work, &id_priv->query);
2340
2341         return (id_priv->query_id < 0) ? id_priv->query_id : 0;
2342 }
2343
2344 static void cma_work_handler(struct work_struct *_work)
2345 {
2346         struct cma_work *work = container_of(_work, struct cma_work, work);
2347         struct rdma_id_private *id_priv = work->id;
2348         int destroy = 0;
2349
2350         mutex_lock(&id_priv->handler_mutex);
2351         if (!cma_comp_exch(id_priv, work->old_state, work->new_state))
2352                 goto out;
2353
2354         if (id_priv->id.event_handler(&id_priv->id, &work->event)) {
2355                 cma_exch(id_priv, RDMA_CM_DESTROYING);
2356                 destroy = 1;
2357         }
2358 out:
2359         mutex_unlock(&id_priv->handler_mutex);
2360         cma_deref_id(id_priv);
2361         if (destroy)
2362                 rdma_destroy_id(&id_priv->id);
2363         kfree(work);
2364 }
2365
2366 static void cma_ndev_work_handler(struct work_struct *_work)
2367 {
2368         struct cma_ndev_work *work = container_of(_work, struct cma_ndev_work, work);
2369         struct rdma_id_private *id_priv = work->id;
2370         int destroy = 0;
2371
2372         mutex_lock(&id_priv->handler_mutex);
2373         if (id_priv->state == RDMA_CM_DESTROYING ||
2374             id_priv->state == RDMA_CM_DEVICE_REMOVAL)
2375                 goto out;
2376
2377         if (id_priv->id.event_handler(&id_priv->id, &work->event)) {
2378                 cma_exch(id_priv, RDMA_CM_DESTROYING);
2379                 destroy = 1;
2380         }
2381
2382 out:
2383         mutex_unlock(&id_priv->handler_mutex);
2384         cma_deref_id(id_priv);
2385         if (destroy)
2386                 rdma_destroy_id(&id_priv->id);
2387         kfree(work);
2388 }
2389
2390 static int cma_resolve_ib_route(struct rdma_id_private *id_priv, int timeout_ms)
2391 {
2392         struct rdma_route *route = &id_priv->id.route;
2393         struct cma_work *work;
2394         int ret;
2395
2396         work = kzalloc(sizeof *work, GFP_KERNEL);
2397         if (!work)
2398                 return -ENOMEM;
2399
2400         work->id = id_priv;
2401         INIT_WORK(&work->work, cma_work_handler);
2402         work->old_state = RDMA_CM_ROUTE_QUERY;
2403         work->new_state = RDMA_CM_ROUTE_RESOLVED;
2404         work->event.event = RDMA_CM_EVENT_ROUTE_RESOLVED;
2405
2406         route->path_rec = kmalloc(sizeof *route->path_rec, GFP_KERNEL);
2407         if (!route->path_rec) {
2408                 ret = -ENOMEM;
2409                 goto err1;
2410         }
2411
2412         ret = cma_query_ib_route(id_priv, timeout_ms, work);
2413         if (ret)
2414                 goto err2;
2415
2416         return 0;
2417 err2:
2418         kfree(route->path_rec);
2419         route->path_rec = NULL;
2420 err1:
2421         kfree(work);
2422         return ret;
2423 }
2424
2425 int rdma_set_ib_paths(struct rdma_cm_id *id,
2426                       struct ib_sa_path_rec *path_rec, int num_paths)
2427 {
2428         struct rdma_id_private *id_priv;
2429         int ret;
2430
2431         id_priv = container_of(id, struct rdma_id_private, id);
2432         if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_RESOLVED,
2433                            RDMA_CM_ROUTE_RESOLVED))
2434                 return -EINVAL;
2435
2436         id->route.path_rec = kmemdup(path_rec, sizeof *path_rec * num_paths,
2437                                      GFP_KERNEL);
2438         if (!id->route.path_rec) {
2439                 ret = -ENOMEM;
2440                 goto err;
2441         }
2442
2443         id->route.num_paths = num_paths;
2444         return 0;
2445 err:
2446         cma_comp_exch(id_priv, RDMA_CM_ROUTE_RESOLVED, RDMA_CM_ADDR_RESOLVED);
2447         return ret;
2448 }
2449 EXPORT_SYMBOL(rdma_set_ib_paths);
2450
2451 static int cma_resolve_iw_route(struct rdma_id_private *id_priv, int timeout_ms)
2452 {
2453         struct cma_work *work;
2454
2455         work = kzalloc(sizeof *work, GFP_KERNEL);
2456         if (!work)
2457                 return -ENOMEM;
2458
2459         work->id = id_priv;
2460         INIT_WORK(&work->work, cma_work_handler);
2461         work->old_state = RDMA_CM_ROUTE_QUERY;
2462         work->new_state = RDMA_CM_ROUTE_RESOLVED;
2463         work->event.event = RDMA_CM_EVENT_ROUTE_RESOLVED;
2464         queue_work(cma_wq, &work->work);
2465         return 0;
2466 }
2467
2468 static int iboe_tos_to_sl(struct net_device *ndev, int tos)
2469 {
2470         int prio;
2471         struct net_device *dev;
2472
2473         prio = rt_tos2priority(tos);
2474         dev = ndev->priv_flags & IFF_802_1Q_VLAN ?
2475                 vlan_dev_real_dev(ndev) : ndev;
2476
2477         if (dev->num_tc)
2478                 return netdev_get_prio_tc_map(dev, prio);
2479
2480 #if IS_ENABLED(CONFIG_VLAN_8021Q)
2481         if (ndev->priv_flags & IFF_802_1Q_VLAN)
2482                 return (vlan_dev_get_egress_qos_mask(ndev, prio) &
2483                         VLAN_PRIO_MASK) >> VLAN_PRIO_SHIFT;
2484 #endif
2485         return 0;
2486 }
2487
2488 static enum ib_gid_type cma_route_gid_type(enum rdma_network_type network_type,
2489                                            unsigned long supported_gids,
2490                                            enum ib_gid_type default_gid)
2491 {
2492         if ((network_type == RDMA_NETWORK_IPV4 ||
2493              network_type == RDMA_NETWORK_IPV6) &&
2494             test_bit(IB_GID_TYPE_ROCE_UDP_ENCAP, &supported_gids))
2495                 return IB_GID_TYPE_ROCE_UDP_ENCAP;
2496
2497         return default_gid;
2498 }
2499
2500 static int cma_resolve_iboe_route(struct rdma_id_private *id_priv)
2501 {
2502         struct rdma_route *route = &id_priv->id.route;
2503         struct rdma_addr *addr = &route->addr;
2504         struct cma_work *work;
2505         int ret;
2506         struct net_device *ndev = NULL;
2507
2508
2509         work = kzalloc(sizeof *work, GFP_KERNEL);
2510         if (!work)
2511                 return -ENOMEM;
2512
2513         work->id = id_priv;
2514         INIT_WORK(&work->work, cma_work_handler);
2515
2516         route->path_rec = kzalloc(sizeof *route->path_rec, GFP_KERNEL);
2517         if (!route->path_rec) {
2518                 ret = -ENOMEM;
2519                 goto err1;
2520         }
2521
2522         route->num_paths = 1;
2523
2524         if (addr->dev_addr.bound_dev_if) {
2525                 unsigned long supported_gids;
2526
2527                 ndev = dev_get_by_index(&init_net, addr->dev_addr.bound_dev_if);
2528                 if (!ndev) {
2529                         ret = -ENODEV;
2530                         goto err2;
2531                 }
2532
2533                 if (ndev->flags & IFF_LOOPBACK) {
2534                         dev_put(ndev);
2535                         if (!id_priv->id.device->get_netdev) {
2536                                 ret = -EOPNOTSUPP;
2537                                 goto err2;
2538                         }
2539
2540                         ndev = id_priv->id.device->get_netdev(id_priv->id.device,
2541                                                               id_priv->id.port_num);
2542                         if (!ndev) {
2543                                 ret = -ENODEV;
2544                                 goto err2;
2545                         }
2546                 }
2547
2548                 route->path_rec->net = &init_net;
2549                 route->path_rec->ifindex = ndev->ifindex;
2550                 supported_gids = roce_gid_type_mask_support(id_priv->id.device,
2551                                                             id_priv->id.port_num);
2552                 route->path_rec->gid_type =
2553                         cma_route_gid_type(addr->dev_addr.network,
2554                                            supported_gids,
2555                                            id_priv->gid_type);
2556         }
2557         if (!ndev) {
2558                 ret = -ENODEV;
2559                 goto err2;
2560         }
2561
2562         memcpy(route->path_rec->dmac, addr->dev_addr.dst_dev_addr, ETH_ALEN);
2563
2564         rdma_ip2gid((struct sockaddr *)&id_priv->id.route.addr.src_addr,
2565                     &route->path_rec->sgid);
2566         rdma_ip2gid((struct sockaddr *)&id_priv->id.route.addr.dst_addr,
2567                     &route->path_rec->dgid);
2568
2569         /* Use the hint from IP Stack to select GID Type */
2570         if (route->path_rec->gid_type < ib_network_to_gid_type(addr->dev_addr.network))
2571                 route->path_rec->gid_type = ib_network_to_gid_type(addr->dev_addr.network);
2572         if (((struct sockaddr *)&id_priv->id.route.addr.dst_addr)->sa_family != AF_IB)
2573                 /* TODO: get the hoplimit from the inet/inet6 device */
2574                 route->path_rec->hop_limit = addr->dev_addr.hoplimit;
2575         else
2576                 route->path_rec->hop_limit = 1;
2577         route->path_rec->reversible = 1;
2578         route->path_rec->pkey = cpu_to_be16(0xffff);
2579         route->path_rec->mtu_selector = IB_SA_EQ;
2580         route->path_rec->sl = iboe_tos_to_sl(ndev, id_priv->tos);
2581         route->path_rec->mtu = iboe_get_mtu(ndev->mtu);
2582         route->path_rec->rate_selector = IB_SA_EQ;
2583         route->path_rec->rate = iboe_get_rate(ndev);
2584         dev_put(ndev);
2585         route->path_rec->packet_life_time_selector = IB_SA_EQ;
2586         route->path_rec->packet_life_time = CMA_IBOE_PACKET_LIFETIME;
2587         if (!route->path_rec->mtu) {
2588                 ret = -EINVAL;
2589                 goto err2;
2590         }
2591
2592         work->old_state = RDMA_CM_ROUTE_QUERY;
2593         work->new_state = RDMA_CM_ROUTE_RESOLVED;
2594         work->event.event = RDMA_CM_EVENT_ROUTE_RESOLVED;
2595         work->event.status = 0;
2596
2597         queue_work(cma_wq, &work->work);
2598
2599         return 0;
2600
2601 err2:
2602         kfree(route->path_rec);
2603         route->path_rec = NULL;
2604 err1:
2605         kfree(work);
2606         return ret;
2607 }
2608
2609 int rdma_resolve_route(struct rdma_cm_id *id, int timeout_ms)
2610 {
2611         struct rdma_id_private *id_priv;
2612         int ret;
2613
2614         id_priv = container_of(id, struct rdma_id_private, id);
2615         if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_RESOLVED, RDMA_CM_ROUTE_QUERY))
2616                 return -EINVAL;
2617
2618         atomic_inc(&id_priv->refcount);
2619         if (rdma_cap_ib_sa(id->device, id->port_num))
2620                 ret = cma_resolve_ib_route(id_priv, timeout_ms);
2621         else if (rdma_protocol_roce(id->device, id->port_num))
2622                 ret = cma_resolve_iboe_route(id_priv);
2623         else if (rdma_protocol_iwarp(id->device, id->port_num))
2624                 ret = cma_resolve_iw_route(id_priv, timeout_ms);
2625         else
2626                 ret = -ENOSYS;
2627
2628         if (ret)
2629                 goto err;
2630
2631         return 0;
2632 err:
2633         cma_comp_exch(id_priv, RDMA_CM_ROUTE_QUERY, RDMA_CM_ADDR_RESOLVED);
2634         cma_deref_id(id_priv);
2635         return ret;
2636 }
2637 EXPORT_SYMBOL(rdma_resolve_route);
2638
2639 static void cma_set_loopback(struct sockaddr *addr)
2640 {
2641         switch (addr->sa_family) {
2642         case AF_INET:
2643                 ((struct sockaddr_in *) addr)->sin_addr.s_addr = htonl(INADDR_LOOPBACK);
2644                 break;
2645         case AF_INET6:
2646                 ipv6_addr_set(&((struct sockaddr_in6 *) addr)->sin6_addr,
2647                               0, 0, 0, htonl(1));
2648                 break;
2649         default:
2650                 ib_addr_set(&((struct sockaddr_ib *) addr)->sib_addr,
2651                             0, 0, 0, htonl(1));
2652                 break;
2653         }
2654 }
2655
2656 static int cma_bind_loopback(struct rdma_id_private *id_priv)
2657 {
2658         struct cma_device *cma_dev, *cur_dev;
2659         struct ib_port_attr port_attr;
2660         union ib_gid gid;
2661         u16 pkey;
2662         int ret;
2663         u8 p;
2664
2665         cma_dev = NULL;
2666         mutex_lock(&lock);
2667         list_for_each_entry(cur_dev, &dev_list, list) {
2668                 if (cma_family(id_priv) == AF_IB &&
2669                     !rdma_cap_ib_cm(cur_dev->device, 1))
2670                         continue;
2671
2672                 if (!cma_dev)
2673                         cma_dev = cur_dev;
2674
2675                 for (p = 1; p <= cur_dev->device->phys_port_cnt; ++p) {
2676                         if (!ib_query_port(cur_dev->device, p, &port_attr) &&
2677                             port_attr.state == IB_PORT_ACTIVE) {
2678                                 cma_dev = cur_dev;
2679                                 goto port_found;
2680                         }
2681                 }
2682         }
2683
2684         if (!cma_dev) {
2685                 ret = -ENODEV;
2686                 goto out;
2687         }
2688
2689         p = 1;
2690
2691 port_found:
2692         ret = ib_get_cached_gid(cma_dev->device, p, 0, &gid, NULL);
2693         if (ret)
2694                 goto out;
2695
2696         ret = ib_get_cached_pkey(cma_dev->device, p, 0, &pkey);
2697         if (ret)
2698                 goto out;
2699
2700         id_priv->id.route.addr.dev_addr.dev_type =
2701                 (rdma_protocol_ib(cma_dev->device, p)) ?
2702                 ARPHRD_INFINIBAND : ARPHRD_ETHER;
2703
2704         rdma_addr_set_sgid(&id_priv->id.route.addr.dev_addr, &gid);
2705         ib_addr_set_pkey(&id_priv->id.route.addr.dev_addr, pkey);
2706         id_priv->id.port_num = p;
2707         cma_attach_to_dev(id_priv, cma_dev);
2708         cma_set_loopback(cma_src_addr(id_priv));
2709 out:
2710         mutex_unlock(&lock);
2711         return ret;
2712 }
2713
2714 static void addr_handler(int status, struct sockaddr *src_addr,
2715                          struct rdma_dev_addr *dev_addr, void *context)
2716 {
2717         struct rdma_id_private *id_priv = context;
2718         struct rdma_cm_event event;
2719
2720         memset(&event, 0, sizeof event);
2721         mutex_lock(&id_priv->handler_mutex);
2722         if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_QUERY,
2723                            RDMA_CM_ADDR_RESOLVED))
2724                 goto out;
2725
2726         memcpy(cma_src_addr(id_priv), src_addr, rdma_addr_size(src_addr));
2727         if (!status && !id_priv->cma_dev)
2728                 status = cma_acquire_dev(id_priv, NULL);
2729
2730         if (status) {
2731                 if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_RESOLVED,
2732                                    RDMA_CM_ADDR_BOUND))
2733                         goto out;
2734                 event.event = RDMA_CM_EVENT_ADDR_ERROR;
2735                 event.status = status;
2736         } else
2737                 event.event = RDMA_CM_EVENT_ADDR_RESOLVED;
2738
2739         if (id_priv->id.event_handler(&id_priv->id, &event)) {
2740                 cma_exch(id_priv, RDMA_CM_DESTROYING);
2741                 mutex_unlock(&id_priv->handler_mutex);
2742                 cma_deref_id(id_priv);
2743                 rdma_destroy_id(&id_priv->id);
2744                 return;
2745         }
2746 out:
2747         mutex_unlock(&id_priv->handler_mutex);
2748         cma_deref_id(id_priv);
2749 }
2750
2751 static int cma_resolve_loopback(struct rdma_id_private *id_priv)
2752 {
2753         struct cma_work *work;
2754         union ib_gid gid;
2755         int ret;
2756
2757         work = kzalloc(sizeof *work, GFP_KERNEL);
2758         if (!work)
2759                 return -ENOMEM;
2760
2761         if (!id_priv->cma_dev) {
2762                 ret = cma_bind_loopback(id_priv);
2763                 if (ret)
2764                         goto err;
2765         }
2766
2767         rdma_addr_get_sgid(&id_priv->id.route.addr.dev_addr, &gid);
2768         rdma_addr_set_dgid(&id_priv->id.route.addr.dev_addr, &gid);
2769
2770         work->id = id_priv;
2771         INIT_WORK(&work->work, cma_work_handler);
2772         work->old_state = RDMA_CM_ADDR_QUERY;
2773         work->new_state = RDMA_CM_ADDR_RESOLVED;
2774         work->event.event = RDMA_CM_EVENT_ADDR_RESOLVED;
2775         queue_work(cma_wq, &work->work);
2776         return 0;
2777 err:
2778         kfree(work);
2779         return ret;
2780 }
2781
2782 static int cma_resolve_ib_addr(struct rdma_id_private *id_priv)
2783 {
2784         struct cma_work *work;
2785         int ret;
2786
2787         work = kzalloc(sizeof *work, GFP_KERNEL);
2788         if (!work)
2789                 return -ENOMEM;
2790
2791         if (!id_priv->cma_dev) {
2792                 ret = cma_resolve_ib_dev(id_priv);
2793                 if (ret)
2794                         goto err;
2795         }
2796
2797         rdma_addr_set_dgid(&id_priv->id.route.addr.dev_addr, (union ib_gid *)
2798                 &(((struct sockaddr_ib *) &id_priv->id.route.addr.dst_addr)->sib_addr));
2799
2800         work->id = id_priv;
2801         INIT_WORK(&work->work, cma_work_handler);
2802         work->old_state = RDMA_CM_ADDR_QUERY;
2803         work->new_state = RDMA_CM_ADDR_RESOLVED;
2804         work->event.event = RDMA_CM_EVENT_ADDR_RESOLVED;
2805         queue_work(cma_wq, &work->work);
2806         return 0;
2807 err:
2808         kfree(work);
2809         return ret;
2810 }
2811
2812 static int cma_bind_addr(struct rdma_cm_id *id, struct sockaddr *src_addr,
2813                          struct sockaddr *dst_addr)
2814 {
2815         if (!src_addr || !src_addr->sa_family) {
2816                 src_addr = (struct sockaddr *) &id->route.addr.src_addr;
2817                 src_addr->sa_family = dst_addr->sa_family;
2818                 if (dst_addr->sa_family == AF_INET6) {
2819                         struct sockaddr_in6 *src_addr6 = (struct sockaddr_in6 *) src_addr;
2820                         struct sockaddr_in6 *dst_addr6 = (struct sockaddr_in6 *) dst_addr;
2821                         src_addr6->sin6_scope_id = dst_addr6->sin6_scope_id;
2822                         if (ipv6_addr_type(&dst_addr6->sin6_addr) & IPV6_ADDR_LINKLOCAL)
2823                                 id->route.addr.dev_addr.bound_dev_if = dst_addr6->sin6_scope_id;
2824                 } else if (dst_addr->sa_family == AF_IB) {
2825                         ((struct sockaddr_ib *) src_addr)->sib_pkey =
2826                                 ((struct sockaddr_ib *) dst_addr)->sib_pkey;
2827                 }
2828         }
2829         return rdma_bind_addr(id, src_addr);
2830 }
2831
2832 int rdma_resolve_addr(struct rdma_cm_id *id, struct sockaddr *src_addr,
2833                       struct sockaddr *dst_addr, int timeout_ms)
2834 {
2835         struct rdma_id_private *id_priv;
2836         int ret;
2837
2838         id_priv = container_of(id, struct rdma_id_private, id);
2839         if (id_priv->state == RDMA_CM_IDLE) {
2840                 ret = cma_bind_addr(id, src_addr, dst_addr);
2841                 if (ret)
2842                         return ret;
2843         }
2844
2845         if (cma_family(id_priv) != dst_addr->sa_family)
2846                 return -EINVAL;
2847
2848         if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_BOUND, RDMA_CM_ADDR_QUERY))
2849                 return -EINVAL;
2850
2851         atomic_inc(&id_priv->refcount);
2852         memcpy(cma_dst_addr(id_priv), dst_addr, rdma_addr_size(dst_addr));
2853         if (cma_any_addr(dst_addr)) {
2854                 ret = cma_resolve_loopback(id_priv);
2855         } else {
2856                 if (dst_addr->sa_family == AF_IB) {
2857                         ret = cma_resolve_ib_addr(id_priv);
2858                 } else {
2859                         ret = rdma_resolve_ip(&addr_client, cma_src_addr(id_priv),
2860                                               dst_addr, &id->route.addr.dev_addr,
2861                                               timeout_ms, addr_handler, id_priv);
2862                 }
2863         }
2864         if (ret)
2865                 goto err;
2866
2867         return 0;
2868 err:
2869         cma_comp_exch(id_priv, RDMA_CM_ADDR_QUERY, RDMA_CM_ADDR_BOUND);
2870         cma_deref_id(id_priv);
2871         return ret;
2872 }
2873 EXPORT_SYMBOL(rdma_resolve_addr);
2874
2875 int rdma_set_reuseaddr(struct rdma_cm_id *id, int reuse)
2876 {
2877         struct rdma_id_private *id_priv;
2878         unsigned long flags;
2879         int ret;
2880
2881         id_priv = container_of(id, struct rdma_id_private, id);
2882         spin_lock_irqsave(&id_priv->lock, flags);
2883         if (reuse || id_priv->state == RDMA_CM_IDLE) {
2884                 id_priv->reuseaddr = reuse;
2885                 ret = 0;
2886         } else {
2887                 ret = -EINVAL;
2888         }
2889         spin_unlock_irqrestore(&id_priv->lock, flags);
2890         return ret;
2891 }
2892 EXPORT_SYMBOL(rdma_set_reuseaddr);
2893
2894 int rdma_set_afonly(struct rdma_cm_id *id, int afonly)
2895 {
2896         struct rdma_id_private *id_priv;
2897         unsigned long flags;
2898         int ret;
2899
2900         id_priv = container_of(id, struct rdma_id_private, id);
2901         spin_lock_irqsave(&id_priv->lock, flags);
2902         if (id_priv->state == RDMA_CM_IDLE || id_priv->state == RDMA_CM_ADDR_BOUND) {
2903                 id_priv->options |= (1 << CMA_OPTION_AFONLY);
2904                 id_priv->afonly = afonly;
2905                 ret = 0;
2906         } else {
2907                 ret = -EINVAL;
2908         }
2909         spin_unlock_irqrestore(&id_priv->lock, flags);
2910         return ret;
2911 }
2912 EXPORT_SYMBOL(rdma_set_afonly);
2913
2914 static void cma_bind_port(struct rdma_bind_list *bind_list,
2915                           struct rdma_id_private *id_priv)
2916 {
2917         struct sockaddr *addr;
2918         struct sockaddr_ib *sib;
2919         u64 sid, mask;
2920         __be16 port;
2921
2922         addr = cma_src_addr(id_priv);
2923         port = htons(bind_list->port);
2924
2925         switch (addr->sa_family) {
2926         case AF_INET:
2927                 ((struct sockaddr_in *) addr)->sin_port = port;
2928                 break;
2929         case AF_INET6:
2930                 ((struct sockaddr_in6 *) addr)->sin6_port = port;
2931                 break;
2932         case AF_IB:
2933                 sib = (struct sockaddr_ib *) addr;
2934                 sid = be64_to_cpu(sib->sib_sid);
2935                 mask = be64_to_cpu(sib->sib_sid_mask);
2936                 sib->sib_sid = cpu_to_be64((sid & mask) | (u64) ntohs(port));
2937                 sib->sib_sid_mask = cpu_to_be64(~0ULL);
2938                 break;
2939         }
2940         id_priv->bind_list = bind_list;
2941         hlist_add_head(&id_priv->node, &bind_list->owners);
2942 }
2943
2944 static int cma_alloc_port(enum rdma_port_space ps,
2945                           struct rdma_id_private *id_priv, unsigned short snum)
2946 {
2947         struct rdma_bind_list *bind_list;
2948         int ret;
2949
2950         bind_list = kzalloc(sizeof *bind_list, GFP_KERNEL);
2951         if (!bind_list)
2952                 return -ENOMEM;
2953
2954         ret = cma_ps_alloc(id_priv->id.route.addr.dev_addr.net, ps, bind_list,
2955                            snum);
2956         if (ret < 0)
2957                 goto err;
2958
2959         bind_list->ps = ps;
2960         bind_list->port = (unsigned short)ret;
2961         cma_bind_port(bind_list, id_priv);
2962         return 0;
2963 err:
2964         kfree(bind_list);
2965         return ret == -ENOSPC ? -EADDRNOTAVAIL : ret;
2966 }
2967
2968 static int cma_alloc_any_port(enum rdma_port_space ps,
2969                               struct rdma_id_private *id_priv)
2970 {
2971         static unsigned int last_used_port;
2972         int low, high, remaining;
2973         unsigned int rover;
2974         struct net *net = id_priv->id.route.addr.dev_addr.net;
2975
2976         inet_get_local_port_range(net, &low, &high);
2977         remaining = (high - low) + 1;
2978         rover = prandom_u32() % remaining + low;
2979 retry:
2980         if (last_used_port != rover &&
2981             !cma_ps_find(net, ps, (unsigned short)rover)) {
2982                 int ret = cma_alloc_port(ps, id_priv, rover);
2983                 /*
2984                  * Remember previously used port number in order to avoid
2985                  * re-using same port immediately after it is closed.
2986                  */
2987                 if (!ret)
2988                         last_used_port = rover;
2989                 if (ret != -EADDRNOTAVAIL)
2990                         return ret;
2991         }
2992         if (--remaining) {
2993                 rover++;
2994                 if ((rover < low) || (rover > high))
2995                         rover = low;
2996                 goto retry;
2997         }
2998         return -EADDRNOTAVAIL;
2999 }
3000
3001 /*
3002  * Check that the requested port is available.  This is called when trying to
3003  * bind to a specific port, or when trying to listen on a bound port.  In
3004  * the latter case, the provided id_priv may already be on the bind_list, but
3005  * we still need to check that it's okay to start listening.
3006  */
3007 static int cma_check_port(struct rdma_bind_list *bind_list,
3008                           struct rdma_id_private *id_priv, uint8_t reuseaddr)
3009 {
3010         struct rdma_id_private *cur_id;
3011         struct sockaddr *addr, *cur_addr;
3012
3013         addr = cma_src_addr(id_priv);
3014         hlist_for_each_entry(cur_id, &bind_list->owners, node) {
3015                 if (id_priv == cur_id)
3016                         continue;
3017
3018                 if ((cur_id->state != RDMA_CM_LISTEN) && reuseaddr &&
3019                     cur_id->reuseaddr)
3020                         continue;
3021
3022                 cur_addr = cma_src_addr(cur_id);
3023                 if (id_priv->afonly && cur_id->afonly &&
3024                     (addr->sa_family != cur_addr->sa_family))
3025                         continue;
3026
3027                 if (cma_any_addr(addr) || cma_any_addr(cur_addr))
3028                         return -EADDRNOTAVAIL;
3029
3030                 if (!cma_addr_cmp(addr, cur_addr))
3031                         return -EADDRINUSE;
3032         }
3033         return 0;
3034 }
3035
3036 static int cma_use_port(enum rdma_port_space ps,
3037                         struct rdma_id_private *id_priv)
3038 {
3039         struct rdma_bind_list *bind_list;
3040         unsigned short snum;
3041         int ret;
3042
3043         snum = ntohs(cma_port(cma_src_addr(id_priv)));
3044         if (snum < PROT_SOCK && !capable(CAP_NET_BIND_SERVICE))
3045                 return -EACCES;
3046
3047         bind_list = cma_ps_find(id_priv->id.route.addr.dev_addr.net, ps, snum);
3048         if (!bind_list) {
3049                 ret = cma_alloc_port(ps, id_priv, snum);
3050         } else {
3051                 ret = cma_check_port(bind_list, id_priv, id_priv->reuseaddr);
3052                 if (!ret)
3053                         cma_bind_port(bind_list, id_priv);
3054         }
3055         return ret;
3056 }
3057
3058 static int cma_bind_listen(struct rdma_id_private *id_priv)
3059 {
3060         struct rdma_bind_list *bind_list = id_priv->bind_list;
3061         int ret = 0;
3062
3063         mutex_lock(&lock);
3064         if (bind_list->owners.first->next)
3065                 ret = cma_check_port(bind_list, id_priv, 0);
3066         mutex_unlock(&lock);
3067         return ret;
3068 }
3069
3070 static enum rdma_port_space cma_select_inet_ps(
3071                 struct rdma_id_private *id_priv)
3072 {
3073         switch (id_priv->id.ps) {
3074         case RDMA_PS_TCP:
3075         case RDMA_PS_UDP:
3076         case RDMA_PS_IPOIB:
3077         case RDMA_PS_IB:
3078                 return id_priv->id.ps;
3079         default:
3080
3081                 return 0;
3082         }
3083 }
3084
3085 static enum rdma_port_space cma_select_ib_ps(struct rdma_id_private *id_priv)
3086 {
3087         enum rdma_port_space ps = 0;
3088         struct sockaddr_ib *sib;
3089         u64 sid_ps, mask, sid;
3090
3091         sib = (struct sockaddr_ib *) cma_src_addr(id_priv);
3092         mask = be64_to_cpu(sib->sib_sid_mask) & RDMA_IB_IP_PS_MASK;
3093         sid = be64_to_cpu(sib->sib_sid) & mask;
3094
3095         if ((id_priv->id.ps == RDMA_PS_IB) && (sid == (RDMA_IB_IP_PS_IB & mask))) {
3096                 sid_ps = RDMA_IB_IP_PS_IB;
3097                 ps = RDMA_PS_IB;
3098         } else if (((id_priv->id.ps == RDMA_PS_IB) || (id_priv->id.ps == RDMA_PS_TCP)) &&
3099                    (sid == (RDMA_IB_IP_PS_TCP & mask))) {
3100                 sid_ps = RDMA_IB_IP_PS_TCP;
3101                 ps = RDMA_PS_TCP;
3102         } else if (((id_priv->id.ps == RDMA_PS_IB) || (id_priv->id.ps == RDMA_PS_UDP)) &&
3103                    (sid == (RDMA_IB_IP_PS_UDP & mask))) {
3104                 sid_ps = RDMA_IB_IP_PS_UDP;
3105                 ps = RDMA_PS_UDP;
3106         }
3107
3108         if (ps) {
3109                 sib->sib_sid = cpu_to_be64(sid_ps | ntohs(cma_port((struct sockaddr *) sib)));
3110                 sib->sib_sid_mask = cpu_to_be64(RDMA_IB_IP_PS_MASK |
3111                                                 be64_to_cpu(sib->sib_sid_mask));
3112         }
3113         return ps;
3114 }
3115
3116 static int cma_get_port(struct rdma_id_private *id_priv)
3117 {
3118         enum rdma_port_space ps;
3119         int ret;
3120
3121         if (cma_family(id_priv) != AF_IB)
3122                 ps = cma_select_inet_ps(id_priv);
3123         else
3124                 ps = cma_select_ib_ps(id_priv);
3125         if (!ps)
3126                 return -EPROTONOSUPPORT;
3127
3128         mutex_lock(&lock);
3129         if (cma_any_port(cma_src_addr(id_priv)))
3130                 ret = cma_alloc_any_port(ps, id_priv);
3131         else
3132                 ret = cma_use_port(ps, id_priv);
3133         mutex_unlock(&lock);
3134
3135         return ret;
3136 }
3137
3138 static int cma_check_linklocal(struct rdma_dev_addr *dev_addr,
3139                                struct sockaddr *addr)
3140 {
3141 #if IS_ENABLED(CONFIG_IPV6)
3142         struct sockaddr_in6 *sin6;
3143
3144         if (addr->sa_family != AF_INET6)
3145                 return 0;
3146
3147         sin6 = (struct sockaddr_in6 *) addr;
3148
3149         if (!(ipv6_addr_type(&sin6->sin6_addr) & IPV6_ADDR_LINKLOCAL))
3150                 return 0;
3151
3152         if (!sin6->sin6_scope_id)
3153                         return -EINVAL;
3154
3155         dev_addr->bound_dev_if = sin6->sin6_scope_id;
3156 #endif
3157         return 0;
3158 }
3159
3160 int rdma_listen(struct rdma_cm_id *id, int backlog)
3161 {
3162         struct rdma_id_private *id_priv;
3163         int ret;
3164
3165         id_priv = container_of(id, struct rdma_id_private, id);
3166         if (id_priv->state == RDMA_CM_IDLE) {
3167                 id->route.addr.src_addr.ss_family = AF_INET;
3168                 ret = rdma_bind_addr(id, cma_src_addr(id_priv));
3169                 if (ret)
3170                         return ret;
3171         }
3172
3173         if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_BOUND, RDMA_CM_LISTEN))
3174                 return -EINVAL;
3175
3176         if (id_priv->reuseaddr) {
3177                 ret = cma_bind_listen(id_priv);
3178                 if (ret)
3179                         goto err;
3180         }
3181
3182         id_priv->backlog = backlog;
3183         if (id->device) {
3184                 if (rdma_cap_ib_cm(id->device, 1)) {
3185                         ret = cma_ib_listen(id_priv);
3186                         if (ret)
3187                                 goto err;
3188                 } else if (rdma_cap_iw_cm(id->device, 1)) {
3189                         ret = cma_iw_listen(id_priv, backlog);
3190                         if (ret)
3191                                 goto err;
3192                 } else {
3193                         ret = -ENOSYS;
3194                         goto err;
3195                 }
3196         } else
3197                 cma_listen_on_all(id_priv);
3198
3199         return 0;
3200 err:
3201         id_priv->backlog = 0;
3202         cma_comp_exch(id_priv, RDMA_CM_LISTEN, RDMA_CM_ADDR_BOUND);
3203         return ret;
3204 }
3205 EXPORT_SYMBOL(rdma_listen);
3206
3207 int rdma_bind_addr(struct rdma_cm_id *id, struct sockaddr *addr)
3208 {
3209         struct rdma_id_private *id_priv;
3210         int ret;
3211
3212         if (addr->sa_family != AF_INET && addr->sa_family != AF_INET6 &&
3213             addr->sa_family != AF_IB)
3214                 return -EAFNOSUPPORT;
3215
3216         id_priv = container_of(id, struct rdma_id_private, id);
3217         if (!cma_comp_exch(id_priv, RDMA_CM_IDLE, RDMA_CM_ADDR_BOUND))
3218                 return -EINVAL;
3219
3220         ret = cma_check_linklocal(&id->route.addr.dev_addr, addr);
3221         if (ret)
3222                 goto err1;
3223
3224         memcpy(cma_src_addr(id_priv), addr, rdma_addr_size(addr));
3225         if (!cma_any_addr(addr)) {
3226                 ret = cma_translate_addr(addr, &id->route.addr.dev_addr);
3227                 if (ret)
3228                         goto err1;
3229
3230                 ret = cma_acquire_dev(id_priv, NULL);
3231                 if (ret)
3232                         goto err1;
3233         }
3234
3235         if (!(id_priv->options & (1 << CMA_OPTION_AFONLY))) {
3236                 if (addr->sa_family == AF_INET)
3237                         id_priv->afonly = 1;
3238 #if IS_ENABLED(CONFIG_IPV6)
3239                 else if (addr->sa_family == AF_INET6) {
3240                         struct net *net = id_priv->id.route.addr.dev_addr.net;
3241
3242                         id_priv->afonly = net->ipv6.sysctl.bindv6only;
3243                 }
3244 #endif
3245         }
3246         ret = cma_get_port(id_priv);
3247         if (ret)
3248                 goto err2;
3249
3250         return 0;
3251 err2:
3252         if (id_priv->cma_dev)
3253                 cma_release_dev(id_priv);
3254 err1:
3255         cma_comp_exch(id_priv, RDMA_CM_ADDR_BOUND, RDMA_CM_IDLE);
3256         return ret;
3257 }
3258 EXPORT_SYMBOL(rdma_bind_addr);
3259
3260 static int cma_format_hdr(void *hdr, struct rdma_id_private *id_priv)
3261 {
3262         struct cma_hdr *cma_hdr;
3263
3264         cma_hdr = hdr;
3265         cma_hdr->cma_version = CMA_VERSION;
3266         if (cma_family(id_priv) == AF_INET) {
3267                 struct sockaddr_in *src4, *dst4;
3268
3269                 src4 = (struct sockaddr_in *) cma_src_addr(id_priv);
3270                 dst4 = (struct sockaddr_in *) cma_dst_addr(id_priv);
3271
3272                 cma_set_ip_ver(cma_hdr, 4);
3273                 cma_hdr->src_addr.ip4.addr = src4->sin_addr.s_addr;
3274                 cma_hdr->dst_addr.ip4.addr = dst4->sin_addr.s_addr;
3275                 cma_hdr->port = src4->sin_port;
3276         } else if (cma_family(id_priv) == AF_INET6) {
3277                 struct sockaddr_in6 *src6, *dst6;
3278
3279                 src6 = (struct sockaddr_in6 *) cma_src_addr(id_priv);
3280                 dst6 = (struct sockaddr_in6 *) cma_dst_addr(id_priv);
3281
3282                 cma_set_ip_ver(cma_hdr, 6);
3283                 cma_hdr->src_addr.ip6 = src6->sin6_addr;
3284                 cma_hdr->dst_addr.ip6 = dst6->sin6_addr;
3285                 cma_hdr->port = src6->sin6_port;
3286         }
3287         return 0;
3288 }
3289
3290 static int cma_sidr_rep_handler(struct ib_cm_id *cm_id,
3291                                 struct ib_cm_event *ib_event)
3292 {
3293         struct rdma_id_private *id_priv = cm_id->context;
3294         struct rdma_cm_event event;
3295         struct ib_cm_sidr_rep_event_param *rep = &ib_event->param.sidr_rep_rcvd;
3296         int ret = 0;
3297
3298         mutex_lock(&id_priv->handler_mutex);
3299         if (id_priv->state != RDMA_CM_CONNECT)
3300                 goto out;
3301
3302         memset(&event, 0, sizeof event);
3303         switch (ib_event->event) {
3304         case IB_CM_SIDR_REQ_ERROR:
3305                 event.event = RDMA_CM_EVENT_UNREACHABLE;
3306                 event.status = -ETIMEDOUT;
3307                 break;
3308         case IB_CM_SIDR_REP_RECEIVED:
3309                 event.param.ud.private_data = ib_event->private_data;
3310                 event.param.ud.private_data_len = IB_CM_SIDR_REP_PRIVATE_DATA_SIZE;
3311                 if (rep->status != IB_SIDR_SUCCESS) {
3312                         event.event = RDMA_CM_EVENT_UNREACHABLE;
3313                         event.status = ib_event->param.sidr_rep_rcvd.status;
3314                         break;
3315                 }
3316                 ret = cma_set_qkey(id_priv, rep->qkey);
3317                 if (ret) {
3318                         event.event = RDMA_CM_EVENT_ADDR_ERROR;
3319                         event.status = ret;
3320                         break;
3321                 }
3322                 ib_init_ah_from_path(id_priv->id.device, id_priv->id.port_num,
3323                                      id_priv->id.route.path_rec,
3324                                      &event.param.ud.ah_attr);
3325                 event.param.ud.qp_num = rep->qpn;
3326                 event.param.ud.qkey = rep->qkey;
3327                 event.event = RDMA_CM_EVENT_ESTABLISHED;
3328                 event.status = 0;
3329                 break;
3330         default:
3331                 pr_err("RDMA CMA: unexpected IB CM event: %d\n",
3332                        ib_event->event);
3333                 goto out;
3334         }
3335
3336         ret = id_priv->id.event_handler(&id_priv->id, &event);
3337         if (ret) {
3338                 /* Destroy the CM ID by returning a non-zero value. */
3339                 id_priv->cm_id.ib = NULL;
3340                 cma_exch(id_priv, RDMA_CM_DESTROYING);
3341                 mutex_unlock(&id_priv->handler_mutex);
3342                 rdma_destroy_id(&id_priv->id);
3343                 return ret;
3344         }
3345 out:
3346         mutex_unlock(&id_priv->handler_mutex);
3347         return ret;
3348 }
3349
3350 static int cma_resolve_ib_udp(struct rdma_id_private *id_priv,
3351                               struct rdma_conn_param *conn_param)
3352 {
3353         struct ib_cm_sidr_req_param req;
3354         struct ib_cm_id *id;
3355         void *private_data;
3356         int offset, ret;
3357
3358         memset(&req, 0, sizeof req);
3359         offset = cma_user_data_offset(id_priv);
3360         req.private_data_len = offset + conn_param->private_data_len;
3361         if (req.private_data_len < conn_param->private_data_len)
3362                 return -EINVAL;
3363
3364         if (req.private_data_len) {
3365                 private_data = kzalloc(req.private_data_len, GFP_ATOMIC);
3366                 if (!private_data)
3367                         return -ENOMEM;
3368         } else {
3369                 private_data = NULL;
3370         }
3371
3372         if (conn_param->private_data && conn_param->private_data_len)
3373                 memcpy(private_data + offset, conn_param->private_data,
3374                        conn_param->private_data_len);
3375
3376         if (private_data) {
3377                 ret = cma_format_hdr(private_data, id_priv);
3378                 if (ret)
3379                         goto out;
3380                 req.private_data = private_data;
3381         }
3382
3383         id = ib_create_cm_id(id_priv->id.device, cma_sidr_rep_handler,
3384                              id_priv);
3385         if (IS_ERR(id)) {
3386                 ret = PTR_ERR(id);
3387                 goto out;
3388         }
3389         id_priv->cm_id.ib = id;
3390
3391         req.path = id_priv->id.route.path_rec;
3392         req.service_id = rdma_get_service_id(&id_priv->id, cma_dst_addr(id_priv));
3393         req.timeout_ms = 1 << (CMA_CM_RESPONSE_TIMEOUT - 8);
3394         req.max_cm_retries = CMA_MAX_CM_RETRIES;
3395
3396         ret = ib_send_cm_sidr_req(id_priv->cm_id.ib, &req);
3397         if (ret) {
3398                 ib_destroy_cm_id(id_priv->cm_id.ib);
3399                 id_priv->cm_id.ib = NULL;
3400         }
3401 out:
3402         kfree(private_data);
3403         return ret;
3404 }
3405
3406 static int cma_connect_ib(struct rdma_id_private *id_priv,
3407                           struct rdma_conn_param *conn_param)
3408 {
3409         struct ib_cm_req_param req;
3410         struct rdma_route *route;
3411         void *private_data;
3412         struct ib_cm_id *id;
3413         int offset, ret;
3414
3415         memset(&req, 0, sizeof req);
3416         offset = cma_user_data_offset(id_priv);
3417         req.private_data_len = offset + conn_param->private_data_len;
3418         if (req.private_data_len < conn_param->private_data_len)
3419                 return -EINVAL;
3420
3421         if (req.private_data_len) {
3422                 private_data = kzalloc(req.private_data_len, GFP_ATOMIC);
3423                 if (!private_data)
3424                         return -ENOMEM;
3425         } else {
3426                 private_data = NULL;
3427         }
3428
3429         if (conn_param->private_data && conn_param->private_data_len)
3430                 memcpy(private_data + offset, conn_param->private_data,
3431                        conn_param->private_data_len);
3432
3433         id = ib_create_cm_id(id_priv->id.device, cma_ib_handler, id_priv);
3434         if (IS_ERR(id)) {
3435                 ret = PTR_ERR(id);
3436                 goto out;
3437         }
3438         id_priv->cm_id.ib = id;
3439
3440         route = &id_priv->id.route;
3441         if (private_data) {
3442                 ret = cma_format_hdr(private_data, id_priv);
3443                 if (ret)
3444                         goto out;
3445                 req.private_data = private_data;
3446         }
3447
3448         req.primary_path = &route->path_rec[0];
3449         if (route->num_paths == 2)
3450                 req.alternate_path = &route->path_rec[1];
3451
3452         req.service_id = rdma_get_service_id(&id_priv->id, cma_dst_addr(id_priv));
3453         req.qp_num = id_priv->qp_num;
3454         req.qp_type = id_priv->id.qp_type;
3455         req.starting_psn = id_priv->seq_num;
3456         req.responder_resources = conn_param->responder_resources;
3457         req.initiator_depth = conn_param->initiator_depth;
3458         req.flow_control = conn_param->flow_control;
3459         req.retry_count = min_t(u8, 7, conn_param->retry_count);
3460         req.rnr_retry_count = min_t(u8, 7, conn_param->rnr_retry_count);
3461         req.remote_cm_response_timeout = CMA_CM_RESPONSE_TIMEOUT;
3462         req.local_cm_response_timeout = CMA_CM_RESPONSE_TIMEOUT;
3463         req.max_cm_retries = CMA_MAX_CM_RETRIES;
3464         req.srq = id_priv->srq ? 1 : 0;
3465
3466         ret = ib_send_cm_req(id_priv->cm_id.ib, &req);
3467 out:
3468         if (ret && !IS_ERR(id)) {
3469                 ib_destroy_cm_id(id);
3470                 id_priv->cm_id.ib = NULL;
3471         }
3472
3473         kfree(private_data);
3474         return ret;
3475 }
3476
3477 static int cma_connect_iw(struct rdma_id_private *id_priv,
3478                           struct rdma_conn_param *conn_param)
3479 {
3480         struct iw_cm_id *cm_id;
3481         int ret;
3482         struct iw_cm_conn_param iw_param;
3483
3484         cm_id = iw_create_cm_id(id_priv->id.device, cma_iw_handler, id_priv);
3485         if (IS_ERR(cm_id))
3486                 return PTR_ERR(cm_id);
3487
3488         cm_id->tos = id_priv->tos;
3489         id_priv->cm_id.iw = cm_id;
3490
3491         memcpy(&cm_id->local_addr, cma_src_addr(id_priv),
3492                rdma_addr_size(cma_src_addr(id_priv)));
3493         memcpy(&cm_id->remote_addr, cma_dst_addr(id_priv),
3494                rdma_addr_size(cma_dst_addr(id_priv)));
3495
3496         ret = cma_modify_qp_rtr(id_priv, conn_param);
3497         if (ret)
3498                 goto out;
3499
3500         if (conn_param) {
3501                 iw_param.ord = conn_param->initiator_depth;
3502                 iw_param.ird = conn_param->responder_resources;
3503                 iw_param.private_data = conn_param->private_data;
3504                 iw_param.private_data_len = conn_param->private_data_len;
3505                 iw_param.qpn = id_priv->id.qp ? id_priv->qp_num : conn_param->qp_num;
3506         } else {
3507                 memset(&iw_param, 0, sizeof iw_param);
3508                 iw_param.qpn = id_priv->qp_num;
3509         }
3510         ret = iw_cm_connect(cm_id, &iw_param);
3511 out:
3512         if (ret) {
3513                 iw_destroy_cm_id(cm_id);
3514                 id_priv->cm_id.iw = NULL;
3515         }
3516         return ret;
3517 }
3518
3519 int rdma_connect(struct rdma_cm_id *id, struct rdma_conn_param *conn_param)
3520 {
3521         struct rdma_id_private *id_priv;
3522         int ret;
3523
3524         id_priv = container_of(id, struct rdma_id_private, id);
3525         if (!cma_comp_exch(id_priv, RDMA_CM_ROUTE_RESOLVED, RDMA_CM_CONNECT))
3526                 return -EINVAL;
3527
3528         if (!id->qp) {
3529                 id_priv->qp_num = conn_param->qp_num;
3530                 id_priv->srq = conn_param->srq;
3531         }
3532
3533         if (rdma_cap_ib_cm(id->device, id->port_num)) {
3534                 if (id->qp_type == IB_QPT_UD)
3535                         ret = cma_resolve_ib_udp(id_priv, conn_param);
3536                 else
3537                         ret = cma_connect_ib(id_priv, conn_param);
3538         } else if (rdma_cap_iw_cm(id->device, id->port_num))
3539                 ret = cma_connect_iw(id_priv, conn_param);
3540         else
3541                 ret = -ENOSYS;
3542         if (ret)
3543                 goto err;
3544
3545         return 0;
3546 err:
3547         cma_comp_exch(id_priv, RDMA_CM_CONNECT, RDMA_CM_ROUTE_RESOLVED);
3548         return ret;
3549 }
3550 EXPORT_SYMBOL(rdma_connect);
3551
3552 static int cma_accept_ib(struct rdma_id_private *id_priv,
3553                          struct rdma_conn_param *conn_param)
3554 {
3555         struct ib_cm_rep_param rep;
3556         int ret;
3557
3558         ret = cma_modify_qp_rtr(id_priv, conn_param);
3559         if (ret)
3560                 goto out;
3561
3562         ret = cma_modify_qp_rts(id_priv, conn_param);
3563         if (ret)
3564                 goto out;
3565
3566         memset(&rep, 0, sizeof rep);
3567         rep.qp_num = id_priv->qp_num;
3568         rep.starting_psn = id_priv->seq_num;
3569         rep.private_data = conn_param->private_data;
3570         rep.private_data_len = conn_param->private_data_len;
3571         rep.responder_resources = conn_param->responder_resources;
3572         rep.initiator_depth = conn_param->initiator_depth;
3573         rep.failover_accepted = 0;
3574         rep.flow_control = conn_param->flow_control;
3575         rep.rnr_retry_count = min_t(u8, 7, conn_param->rnr_retry_count);
3576         rep.srq = id_priv->srq ? 1 : 0;
3577
3578         ret = ib_send_cm_rep(id_priv->cm_id.ib, &rep);
3579 out:
3580         return ret;
3581 }
3582
3583 static int cma_accept_iw(struct rdma_id_private *id_priv,
3584                   struct rdma_conn_param *conn_param)
3585 {
3586         struct iw_cm_conn_param iw_param;
3587         int ret;
3588
3589         ret = cma_modify_qp_rtr(id_priv, conn_param);
3590         if (ret)
3591                 return ret;
3592
3593         iw_param.ord = conn_param->initiator_depth;
3594         iw_param.ird = conn_param->responder_resources;
3595         iw_param.private_data = conn_param->private_data;
3596         iw_param.private_data_len = conn_param->private_data_len;
3597         if (id_priv->id.qp) {
3598                 iw_param.qpn = id_priv->qp_num;
3599         } else
3600                 iw_param.qpn = conn_param->qp_num;
3601
3602         return iw_cm_accept(id_priv->cm_id.iw, &iw_param);
3603 }
3604
3605 static int cma_send_sidr_rep(struct rdma_id_private *id_priv,
3606                              enum ib_cm_sidr_status status, u32 qkey,
3607                              const void *private_data, int private_data_len)
3608 {
3609         struct ib_cm_sidr_rep_param rep;
3610         int ret;
3611
3612         memset(&rep, 0, sizeof rep);
3613         rep.status = status;
3614         if (status == IB_SIDR_SUCCESS) {
3615                 ret = cma_set_qkey(id_priv, qkey);
3616                 if (ret)
3617                         return ret;
3618                 rep.qp_num = id_priv->qp_num;
3619                 rep.qkey = id_priv->qkey;
3620         }
3621         rep.private_data = private_data;
3622         rep.private_data_len = private_data_len;
3623
3624         return ib_send_cm_sidr_rep(id_priv->cm_id.ib, &rep);
3625 }
3626
3627 int rdma_accept(struct rdma_cm_id *id, struct rdma_conn_param *conn_param)
3628 {
3629         struct rdma_id_private *id_priv;
3630         int ret;
3631
3632         id_priv = container_of(id, struct rdma_id_private, id);
3633
3634         id_priv->owner = task_pid_nr(current);
3635
3636         if (!cma_comp(id_priv, RDMA_CM_CONNECT))
3637                 return -EINVAL;
3638
3639         if (!id->qp && conn_param) {
3640                 id_priv->qp_num = conn_param->qp_num;
3641                 id_priv->srq = conn_param->srq;
3642         }
3643
3644         if (rdma_cap_ib_cm(id->device, id->port_num)) {
3645                 if (id->qp_type == IB_QPT_UD) {
3646                         if (conn_param)
3647                                 ret = cma_send_sidr_rep(id_priv, IB_SIDR_SUCCESS,
3648                                                         conn_param->qkey,
3649                                                         conn_param->private_data,
3650                                                         conn_param->private_data_len);
3651                         else
3652                                 ret = cma_send_sidr_rep(id_priv, IB_SIDR_SUCCESS,
3653                                                         0, NULL, 0);
3654                 } else {
3655                         if (conn_param)
3656                                 ret = cma_accept_ib(id_priv, conn_param);
3657                         else
3658                                 ret = cma_rep_recv(id_priv);
3659                 }
3660         } else if (rdma_cap_iw_cm(id->device, id->port_num))
3661                 ret = cma_accept_iw(id_priv, conn_param);
3662         else
3663                 ret = -ENOSYS;
3664
3665         if (ret)
3666                 goto reject;
3667
3668         return 0;
3669 reject:
3670         cma_modify_qp_err(id_priv);
3671         rdma_reject(id, NULL, 0);
3672         return ret;
3673 }
3674 EXPORT_SYMBOL(rdma_accept);
3675
3676 int rdma_notify(struct rdma_cm_id *id, enum ib_event_type event)
3677 {
3678         struct rdma_id_private *id_priv;
3679         int ret;
3680
3681         id_priv = container_of(id, struct rdma_id_private, id);
3682         if (!id_priv->cm_id.ib)
3683                 return -EINVAL;
3684
3685         switch (id->device->node_type) {
3686         case RDMA_NODE_IB_CA:
3687                 ret = ib_cm_notify(id_priv->cm_id.ib, event);
3688                 break;
3689         default:
3690                 ret = 0;
3691                 break;
3692         }
3693         return ret;
3694 }
3695 EXPORT_SYMBOL(rdma_notify);
3696
3697 int rdma_reject(struct rdma_cm_id *id, const void *private_data,
3698                 u8 private_data_len)
3699 {
3700         struct rdma_id_private *id_priv;
3701         int ret;
3702
3703         id_priv = container_of(id, struct rdma_id_private, id);
3704         if (!id_priv->cm_id.ib)
3705                 return -EINVAL;
3706
3707         if (rdma_cap_ib_cm(id->device, id->port_num)) {
3708                 if (id->qp_type == IB_QPT_UD)
3709                         ret = cma_send_sidr_rep(id_priv, IB_SIDR_REJECT, 0,
3710                                                 private_data, private_data_len);
3711                 else
3712                         ret = ib_send_cm_rej(id_priv->cm_id.ib,
3713                                              IB_CM_REJ_CONSUMER_DEFINED, NULL,
3714                                              0, private_data, private_data_len);
3715         } else if (rdma_cap_iw_cm(id->device, id->port_num)) {
3716                 ret = iw_cm_reject(id_priv->cm_id.iw,
3717                                    private_data, private_data_len);
3718         } else
3719                 ret = -ENOSYS;
3720
3721         return ret;
3722 }
3723 EXPORT_SYMBOL(rdma_reject);
3724
3725 int rdma_disconnect(struct rdma_cm_id *id)
3726 {
3727         struct rdma_id_private *id_priv;
3728         int ret;
3729
3730         id_priv = container_of(id, struct rdma_id_private, id);
3731         if (!id_priv->cm_id.ib)
3732                 return -EINVAL;
3733
3734         if (rdma_cap_ib_cm(id->device, id->port_num)) {
3735                 ret = cma_modify_qp_err(id_priv);
3736                 if (ret)
3737                         goto out;
3738                 /* Initiate or respond to a disconnect. */
3739                 if (ib_send_cm_dreq(id_priv->cm_id.ib, NULL, 0))
3740                         ib_send_cm_drep(id_priv->cm_id.ib, NULL, 0);
3741         } else if (rdma_cap_iw_cm(id->device, id->port_num)) {
3742                 ret = iw_cm_disconnect(id_priv->cm_id.iw, 0);
3743         } else
3744                 ret = -EINVAL;
3745
3746 out:
3747         return ret;
3748 }
3749 EXPORT_SYMBOL(rdma_disconnect);
3750
3751 static int cma_ib_mc_handler(int status, struct ib_sa_multicast *multicast)
3752 {
3753         struct rdma_id_private *id_priv;
3754         struct cma_multicast *mc = multicast->context;
3755         struct rdma_cm_event event;
3756         int ret = 0;
3757
3758         id_priv = mc->id_priv;
3759         mutex_lock(&id_priv->handler_mutex);
3760         if (id_priv->state != RDMA_CM_ADDR_BOUND &&
3761             id_priv->state != RDMA_CM_ADDR_RESOLVED)
3762                 goto out;
3763
3764         if (!status)
3765                 status = cma_set_qkey(id_priv, be32_to_cpu(multicast->rec.qkey));
3766         mutex_lock(&id_priv->qp_mutex);
3767         if (!status && id_priv->id.qp)
3768                 status = ib_attach_mcast(id_priv->id.qp, &multicast->rec.mgid,
3769                                          be16_to_cpu(multicast->rec.mlid));
3770         mutex_unlock(&id_priv->qp_mutex);
3771
3772         memset(&event, 0, sizeof event);
3773         event.status = status;
3774         event.param.ud.private_data = mc->context;
3775         if (!status) {
3776                 struct rdma_dev_addr *dev_addr =
3777                         &id_priv->id.route.addr.dev_addr;
3778                 struct net_device *ndev =
3779                         dev_get_by_index(&init_net, dev_addr->bound_dev_if);
3780                 enum ib_gid_type gid_type =
3781                         id_priv->cma_dev->default_gid_type[id_priv->id.port_num -
3782                         rdma_start_port(id_priv->cma_dev->device)];
3783
3784                 event.event = RDMA_CM_EVENT_MULTICAST_JOIN;
3785                 ib_init_ah_from_mcmember(id_priv->id.device,
3786                                          id_priv->id.port_num, &multicast->rec,
3787                                          ndev, gid_type,
3788                                          &event.param.ud.ah_attr);
3789                 event.param.ud.qp_num = 0xFFFFFF;
3790                 event.param.ud.qkey = be32_to_cpu(multicast->rec.qkey);
3791                 if (ndev)
3792                         dev_put(ndev);
3793         } else
3794                 event.event = RDMA_CM_EVENT_MULTICAST_ERROR;
3795
3796         ret = id_priv->id.event_handler(&id_priv->id, &event);
3797         if (ret) {
3798                 cma_exch(id_priv, RDMA_CM_DESTROYING);
3799                 mutex_unlock(&id_priv->handler_mutex);
3800                 rdma_destroy_id(&id_priv->id);
3801                 return 0;
3802         }
3803
3804 out:
3805         mutex_unlock(&id_priv->handler_mutex);
3806         return 0;
3807 }
3808
3809 static void cma_set_mgid(struct rdma_id_private *id_priv,
3810                          struct sockaddr *addr, union ib_gid *mgid)
3811 {
3812         unsigned char mc_map[MAX_ADDR_LEN];
3813         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
3814         struct sockaddr_in *sin = (struct sockaddr_in *) addr;
3815         struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *) addr;
3816
3817         if (cma_any_addr(addr)) {
3818                 memset(mgid, 0, sizeof *mgid);
3819         } else if ((addr->sa_family == AF_INET6) &&
3820                    ((be32_to_cpu(sin6->sin6_addr.s6_addr32[0]) & 0xFFF0FFFF) ==
3821                                                                  0xFF10A01B)) {
3822                 /* IPv6 address is an SA assigned MGID. */
3823                 memcpy(mgid, &sin6->sin6_addr, sizeof *mgid);
3824         } else if (addr->sa_family == AF_IB) {
3825                 memcpy(mgid, &((struct sockaddr_ib *) addr)->sib_addr, sizeof *mgid);
3826         } else if ((addr->sa_family == AF_INET6)) {
3827                 ipv6_ib_mc_map(&sin6->sin6_addr, dev_addr->broadcast, mc_map);
3828                 if (id_priv->id.ps == RDMA_PS_UDP)
3829                         mc_map[7] = 0x01;       /* Use RDMA CM signature */
3830                 *mgid = *(union ib_gid *) (mc_map + 4);
3831         } else {
3832                 ip_ib_mc_map(sin->sin_addr.s_addr, dev_addr->broadcast, mc_map);
3833                 if (id_priv->id.ps == RDMA_PS_UDP)
3834                         mc_map[7] = 0x01;       /* Use RDMA CM signature */
3835                 *mgid = *(union ib_gid *) (mc_map + 4);
3836         }
3837 }
3838
3839 static void cma_query_sa_classport_info_cb(int status,
3840                                            struct ib_class_port_info *rec,
3841                                            void *context)
3842 {
3843         struct class_port_info_context *cb_ctx = context;
3844
3845         WARN_ON(!context);
3846
3847         if (status || !rec) {
3848                 pr_debug("RDMA CM: %s port %u failed query ClassPortInfo status: %d\n",
3849                          cb_ctx->device->name, cb_ctx->port_num, status);
3850                 goto out;
3851         }
3852
3853         memcpy(cb_ctx->class_port_info, rec, sizeof(struct ib_class_port_info));
3854
3855 out:
3856         complete(&cb_ctx->done);
3857 }
3858
3859 static int cma_query_sa_classport_info(struct ib_device *device, u8 port_num,
3860                                        struct ib_class_port_info *class_port_info)
3861 {
3862         struct class_port_info_context *cb_ctx;
3863         int ret;
3864
3865         cb_ctx = kmalloc(sizeof(*cb_ctx), GFP_KERNEL);
3866         if (!cb_ctx)
3867                 return -ENOMEM;
3868
3869         cb_ctx->device = device;
3870         cb_ctx->class_port_info = class_port_info;
3871         cb_ctx->port_num = port_num;
3872         init_completion(&cb_ctx->done);
3873
3874         ret = ib_sa_classport_info_rec_query(&sa_client, device, port_num,
3875                                              CMA_QUERY_CLASSPORT_INFO_TIMEOUT,
3876                                              GFP_KERNEL, cma_query_sa_classport_info_cb,
3877                                              cb_ctx, &cb_ctx->sa_query);
3878         if (ret < 0) {
3879                 pr_err("RDMA CM: %s port %u failed to send ClassPortInfo query, ret: %d\n",
3880                        device->name, port_num, ret);
3881                 goto out;
3882         }
3883
3884         wait_for_completion(&cb_ctx->done);
3885
3886 out:
3887         kfree(cb_ctx);
3888         return ret;
3889 }
3890
3891 static int cma_join_ib_multicast(struct rdma_id_private *id_priv,
3892                                  struct cma_multicast *mc)
3893 {
3894         struct ib_sa_mcmember_rec rec;
3895         struct ib_class_port_info class_port_info;
3896         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
3897         ib_sa_comp_mask comp_mask;
3898         int ret;
3899
3900         ib_addr_get_mgid(dev_addr, &rec.mgid);
3901         ret = ib_sa_get_mcmember_rec(id_priv->id.device, id_priv->id.port_num,
3902                                      &rec.mgid, &rec);
3903         if (ret)
3904                 return ret;
3905
3906         ret = cma_set_qkey(id_priv, 0);
3907         if (ret)
3908                 return ret;
3909
3910         cma_set_mgid(id_priv, (struct sockaddr *) &mc->addr, &rec.mgid);
3911         rec.qkey = cpu_to_be32(id_priv->qkey);
3912         rdma_addr_get_sgid(dev_addr, &rec.port_gid);
3913         rec.pkey = cpu_to_be16(ib_addr_get_pkey(dev_addr));
3914         rec.join_state = mc->join_state;
3915
3916         if (rec.join_state == BIT(SENDONLY_FULLMEMBER_JOIN)) {
3917                 ret = cma_query_sa_classport_info(id_priv->id.device,
3918                                                   id_priv->id.port_num,
3919                                                   &class_port_info);
3920
3921                 if (ret)
3922                         return ret;
3923
3924                 if (!(ib_get_cpi_capmask2(&class_port_info) &
3925                       IB_SA_CAP_MASK2_SENDONLY_FULL_MEM_SUPPORT)) {
3926                         pr_warn("RDMA CM: %s port %u Unable to multicast join\n"
3927                                 "RDMA CM: SM doesn't support Send Only Full Member option\n",
3928                                 id_priv->id.device->name, id_priv->id.port_num);
3929                         return -EOPNOTSUPP;
3930                 }
3931         }
3932
3933         comp_mask = IB_SA_MCMEMBER_REC_MGID | IB_SA_MCMEMBER_REC_PORT_GID |
3934                     IB_SA_MCMEMBER_REC_PKEY | IB_SA_MCMEMBER_REC_JOIN_STATE |
3935                     IB_SA_MCMEMBER_REC_QKEY | IB_SA_MCMEMBER_REC_SL |
3936                     IB_SA_MCMEMBER_REC_FLOW_LABEL |
3937                     IB_SA_MCMEMBER_REC_TRAFFIC_CLASS;
3938
3939         if (id_priv->id.ps == RDMA_PS_IPOIB)
3940                 comp_mask |= IB_SA_MCMEMBER_REC_RATE |
3941                              IB_SA_MCMEMBER_REC_RATE_SELECTOR |
3942                              IB_SA_MCMEMBER_REC_MTU_SELECTOR |
3943                              IB_SA_MCMEMBER_REC_MTU |
3944                              IB_SA_MCMEMBER_REC_HOP_LIMIT;
3945
3946         mc->multicast.ib = ib_sa_join_multicast(&sa_client, id_priv->id.device,
3947                                                 id_priv->id.port_num, &rec,
3948                                                 comp_mask, GFP_KERNEL,
3949                                                 cma_ib_mc_handler, mc);
3950         return PTR_ERR_OR_ZERO(mc->multicast.ib);
3951 }
3952
3953 static void iboe_mcast_work_handler(struct work_struct *work)
3954 {
3955         struct iboe_mcast_work *mw = container_of(work, struct iboe_mcast_work, work);
3956         struct cma_multicast *mc = mw->mc;
3957         struct ib_sa_multicast *m = mc->multicast.ib;
3958
3959         mc->multicast.ib->context = mc;
3960         cma_ib_mc_handler(0, m);
3961         kref_put(&mc->mcref, release_mc);
3962         kfree(mw);
3963 }
3964
3965 static void cma_iboe_set_mgid(struct sockaddr *addr, union ib_gid *mgid)
3966 {
3967         struct sockaddr_in *sin = (struct sockaddr_in *)addr;
3968         struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)addr;
3969
3970         if (cma_any_addr(addr)) {
3971                 memset(mgid, 0, sizeof *mgid);
3972         } else if (addr->sa_family == AF_INET6) {
3973                 memcpy(mgid, &sin6->sin6_addr, sizeof *mgid);
3974         } else {
3975                 mgid->raw[0] = 0xff;
3976                 mgid->raw[1] = 0x0e;
3977                 mgid->raw[2] = 0;
3978                 mgid->raw[3] = 0;
3979                 mgid->raw[4] = 0;
3980                 mgid->raw[5] = 0;
3981                 mgid->raw[6] = 0;
3982                 mgid->raw[7] = 0;
3983                 mgid->raw[8] = 0;
3984                 mgid->raw[9] = 0;
3985                 mgid->raw[10] = 0xff;
3986                 mgid->raw[11] = 0xff;
3987                 *(__be32 *)(&mgid->raw[12]) = sin->sin_addr.s_addr;
3988         }
3989 }
3990
3991 static int cma_iboe_join_multicast(struct rdma_id_private *id_priv,
3992                                    struct cma_multicast *mc)
3993 {
3994         struct iboe_mcast_work *work;
3995         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
3996         int err = 0;
3997         struct sockaddr *addr = (struct sockaddr *)&mc->addr;
3998         struct net_device *ndev = NULL;
3999         enum ib_gid_type gid_type;
4000         bool send_only;
4001
4002         send_only = mc->join_state == BIT(SENDONLY_FULLMEMBER_JOIN);
4003
4004         if (cma_zero_addr((struct sockaddr *)&mc->addr))
4005                 return -EINVAL;
4006
4007         work = kzalloc(sizeof *work, GFP_KERNEL);
4008         if (!work)
4009                 return -ENOMEM;
4010
4011         mc->multicast.ib = kzalloc(sizeof(struct ib_sa_multicast), GFP_KERNEL);
4012         if (!mc->multicast.ib) {
4013                 err = -ENOMEM;
4014                 goto out1;
4015         }
4016
4017         cma_iboe_set_mgid(addr, &mc->multicast.ib->rec.mgid);
4018
4019         mc->multicast.ib->rec.pkey = cpu_to_be16(0xffff);
4020         if (id_priv->id.ps == RDMA_PS_UDP)
4021                 mc->multicast.ib->rec.qkey = cpu_to_be32(RDMA_UDP_QKEY);
4022
4023         if (dev_addr->bound_dev_if)
4024                 ndev = dev_get_by_index(&init_net, dev_addr->bound_dev_if);
4025         if (!ndev) {
4026                 err = -ENODEV;
4027                 goto out2;
4028         }
4029         mc->multicast.ib->rec.rate = iboe_get_rate(ndev);
4030         mc->multicast.ib->rec.hop_limit = 1;
4031         mc->multicast.ib->rec.mtu = iboe_get_mtu(ndev->mtu);
4032
4033         gid_type = id_priv->cma_dev->default_gid_type[id_priv->id.port_num -
4034                    rdma_start_port(id_priv->cma_dev->device)];
4035         if (addr->sa_family == AF_INET) {
4036                 if (gid_type == IB_GID_TYPE_ROCE_UDP_ENCAP) {
4037                         mc->multicast.ib->rec.hop_limit = IPV6_DEFAULT_HOPLIMIT;
4038                         if (!send_only) {
4039                                 err = cma_igmp_send(ndev, &mc->multicast.ib->rec.mgid,
4040                                                     true);
4041                                 if (!err)
4042                                         mc->igmp_joined = true;
4043                         }
4044                 }
4045         } else {
4046                 if (gid_type == IB_GID_TYPE_ROCE_UDP_ENCAP)
4047                         err = -ENOTSUPP;
4048         }
4049         dev_put(ndev);
4050         if (err || !mc->multicast.ib->rec.mtu) {
4051                 if (!err)
4052                         err = -EINVAL;
4053                 goto out2;
4054         }
4055         rdma_ip2gid((struct sockaddr *)&id_priv->id.route.addr.src_addr,
4056                     &mc->multicast.ib->rec.port_gid);
4057         work->id = id_priv;
4058         work->mc = mc;
4059         INIT_WORK(&work->work, iboe_mcast_work_handler);
4060         kref_get(&mc->mcref);
4061         queue_work(cma_wq, &work->work);
4062
4063         return 0;
4064
4065 out2:
4066         kfree(mc->multicast.ib);
4067 out1:
4068         kfree(work);
4069         return err;
4070 }
4071
4072 int rdma_join_multicast(struct rdma_cm_id *id, struct sockaddr *addr,
4073                         u8 join_state, void *context)
4074 {
4075         struct rdma_id_private *id_priv;
4076         struct cma_multicast *mc;
4077         int ret;
4078
4079         id_priv = container_of(id, struct rdma_id_private, id);
4080         if (!cma_comp(id_priv, RDMA_CM_ADDR_BOUND) &&
4081             !cma_comp(id_priv, RDMA_CM_ADDR_RESOLVED))
4082                 return -EINVAL;
4083
4084         mc = kmalloc(sizeof *mc, GFP_KERNEL);
4085         if (!mc)
4086                 return -ENOMEM;
4087
4088         memcpy(&mc->addr, addr, rdma_addr_size(addr));
4089         mc->context = context;
4090         mc->id_priv = id_priv;
4091         mc->igmp_joined = false;
4092         mc->join_state = join_state;
4093         spin_lock(&id_priv->lock);
4094         list_add(&mc->list, &id_priv->mc_list);
4095         spin_unlock(&id_priv->lock);
4096
4097         if (rdma_protocol_roce(id->device, id->port_num)) {
4098                 kref_init(&mc->mcref);
4099                 ret = cma_iboe_join_multicast(id_priv, mc);
4100         } else if (rdma_cap_ib_mcast(id->device, id->port_num))
4101                 ret = cma_join_ib_multicast(id_priv, mc);
4102         else
4103                 ret = -ENOSYS;
4104
4105         if (ret) {
4106                 spin_lock_irq(&id_priv->lock);
4107                 list_del(&mc->list);
4108                 spin_unlock_irq(&id_priv->lock);
4109                 kfree(mc);
4110         }
4111         return ret;
4112 }
4113 EXPORT_SYMBOL(rdma_join_multicast);
4114
4115 void rdma_leave_multicast(struct rdma_cm_id *id, struct sockaddr *addr)
4116 {
4117         struct rdma_id_private *id_priv;
4118         struct cma_multicast *mc;
4119
4120         id_priv = container_of(id, struct rdma_id_private, id);
4121         spin_lock_irq(&id_priv->lock);
4122         list_for_each_entry(mc, &id_priv->mc_list, list) {
4123                 if (!memcmp(&mc->addr, addr, rdma_addr_size(addr))) {
4124                         list_del(&mc->list);
4125                         spin_unlock_irq(&id_priv->lock);
4126
4127                         if (id->qp)
4128                                 ib_detach_mcast(id->qp,
4129                                                 &mc->multicast.ib->rec.mgid,
4130                                                 be16_to_cpu(mc->multicast.ib->rec.mlid));
4131
4132                         BUG_ON(id_priv->cma_dev->device != id->device);
4133
4134                         if (rdma_cap_ib_mcast(id->device, id->port_num)) {
4135                                 ib_sa_free_multicast(mc->multicast.ib);
4136                                 kfree(mc);
4137                         } else if (rdma_protocol_roce(id->device, id->port_num)) {
4138                                 if (mc->igmp_joined) {
4139                                         struct rdma_dev_addr *dev_addr =
4140                                                 &id->route.addr.dev_addr;
4141                                         struct net_device *ndev = NULL;
4142
4143                                         if (dev_addr->bound_dev_if)
4144                                                 ndev = dev_get_by_index(&init_net,
4145                                                                         dev_addr->bound_dev_if);
4146                                         if (ndev) {
4147                                                 cma_igmp_send(ndev,
4148                                                               &mc->multicast.ib->rec.mgid,
4149                                                               false);
4150                                                 dev_put(ndev);
4151                                         }
4152                                         mc->igmp_joined = false;
4153                                 }
4154                                 kref_put(&mc->mcref, release_mc);
4155                         }
4156                         return;
4157                 }
4158         }
4159         spin_unlock_irq(&id_priv->lock);
4160 }
4161 EXPORT_SYMBOL(rdma_leave_multicast);
4162
4163 static int cma_netdev_change(struct net_device *ndev, struct rdma_id_private *id_priv)
4164 {
4165         struct rdma_dev_addr *dev_addr;
4166         struct cma_ndev_work *work;
4167
4168         dev_addr = &id_priv->id.route.addr.dev_addr;
4169
4170         if ((dev_addr->bound_dev_if == ndev->ifindex) &&
4171             (net_eq(dev_net(ndev), dev_addr->net)) &&
4172             memcmp(dev_addr->src_dev_addr, ndev->dev_addr, ndev->addr_len)) {
4173                 pr_info("RDMA CM addr change for ndev %s used by id %p\n",
4174                         ndev->name, &id_priv->id);
4175                 work = kzalloc(sizeof *work, GFP_KERNEL);
4176                 if (!work)
4177                         return -ENOMEM;
4178
4179                 INIT_WORK(&work->work, cma_ndev_work_handler);
4180                 work->id = id_priv;
4181                 work->event.event = RDMA_CM_EVENT_ADDR_CHANGE;
4182                 atomic_inc(&id_priv->refcount);
4183                 queue_work(cma_wq, &work->work);
4184         }
4185
4186         return 0;
4187 }
4188
4189 static int cma_netdev_callback(struct notifier_block *self, unsigned long event,
4190                                void *ptr)
4191 {
4192         struct net_device *ndev = netdev_notifier_info_to_dev(ptr);
4193         struct cma_device *cma_dev;
4194         struct rdma_id_private *id_priv;
4195         int ret = NOTIFY_DONE;
4196
4197         if (event != NETDEV_BONDING_FAILOVER)
4198                 return NOTIFY_DONE;
4199
4200         if (!(ndev->flags & IFF_MASTER) || !(ndev->priv_flags & IFF_BONDING))
4201                 return NOTIFY_DONE;
4202
4203         mutex_lock(&lock);
4204         list_for_each_entry(cma_dev, &dev_list, list)
4205                 list_for_each_entry(id_priv, &cma_dev->id_list, list) {
4206                         ret = cma_netdev_change(ndev, id_priv);
4207                         if (ret)
4208                                 goto out;
4209                 }
4210
4211 out:
4212         mutex_unlock(&lock);
4213         return ret;
4214 }
4215
4216 static struct notifier_block cma_nb = {
4217         .notifier_call = cma_netdev_callback
4218 };
4219
4220 static void cma_add_one(struct ib_device *device)
4221 {
4222         struct cma_device *cma_dev;
4223         struct rdma_id_private *id_priv;
4224         unsigned int i;
4225         unsigned long supported_gids = 0;
4226
4227         cma_dev = kmalloc(sizeof *cma_dev, GFP_KERNEL);
4228         if (!cma_dev)
4229                 return;
4230
4231         cma_dev->device = device;
4232         cma_dev->default_gid_type = kcalloc(device->phys_port_cnt,
4233                                             sizeof(*cma_dev->default_gid_type),
4234                                             GFP_KERNEL);
4235         if (!cma_dev->default_gid_type) {
4236                 kfree(cma_dev);
4237                 return;
4238         }
4239         for (i = rdma_start_port(device); i <= rdma_end_port(device); i++) {
4240                 supported_gids = roce_gid_type_mask_support(device, i);
4241                 WARN_ON(!supported_gids);
4242                 cma_dev->default_gid_type[i - rdma_start_port(device)] =
4243                         find_first_bit(&supported_gids, BITS_PER_LONG);
4244         }
4245
4246         init_completion(&cma_dev->comp);
4247         atomic_set(&cma_dev->refcount, 1);
4248         INIT_LIST_HEAD(&cma_dev->id_list);
4249         ib_set_client_data(device, &cma_client, cma_dev);
4250
4251         mutex_lock(&lock);
4252         list_add_tail(&cma_dev->list, &dev_list);
4253         list_for_each_entry(id_priv, &listen_any_list, list)
4254                 cma_listen_on_dev(id_priv, cma_dev);
4255         mutex_unlock(&lock);
4256 }
4257
4258 static int cma_remove_id_dev(struct rdma_id_private *id_priv)
4259 {
4260         struct rdma_cm_event event;
4261         enum rdma_cm_state state;
4262         int ret = 0;
4263
4264         /* Record that we want to remove the device */
4265         state = cma_exch(id_priv, RDMA_CM_DEVICE_REMOVAL);
4266         if (state == RDMA_CM_DESTROYING)
4267                 return 0;
4268
4269         cma_cancel_operation(id_priv, state);
4270         mutex_lock(&id_priv->handler_mutex);
4271
4272         /* Check for destruction from another callback. */
4273         if (!cma_comp(id_priv, RDMA_CM_DEVICE_REMOVAL))
4274                 goto out;
4275
4276         memset(&event, 0, sizeof event);
4277         event.event = RDMA_CM_EVENT_DEVICE_REMOVAL;
4278         ret = id_priv->id.event_handler(&id_priv->id, &event);
4279 out:
4280         mutex_unlock(&id_priv->handler_mutex);
4281         return ret;
4282 }
4283
4284 static void cma_process_remove(struct cma_device *cma_dev)
4285 {
4286         struct rdma_id_private *id_priv;
4287         int ret;
4288
4289         mutex_lock(&lock);
4290         while (!list_empty(&cma_dev->id_list)) {
4291                 id_priv = list_entry(cma_dev->id_list.next,
4292                                      struct rdma_id_private, list);
4293
4294                 list_del(&id_priv->listen_list);
4295                 list_del_init(&id_priv->list);
4296                 atomic_inc(&id_priv->refcount);
4297                 mutex_unlock(&lock);
4298
4299                 ret = id_priv->internal_id ? 1 : cma_remove_id_dev(id_priv);
4300                 cma_deref_id(id_priv);
4301                 if (ret)
4302                         rdma_destroy_id(&id_priv->id);
4303
4304                 mutex_lock(&lock);
4305         }
4306         mutex_unlock(&lock);
4307
4308         cma_deref_dev(cma_dev);
4309         wait_for_completion(&cma_dev->comp);
4310 }
4311
4312 static void cma_remove_one(struct ib_device *device, void *client_data)
4313 {
4314         struct cma_device *cma_dev = client_data;
4315
4316         if (!cma_dev)
4317                 return;
4318
4319         mutex_lock(&lock);
4320         list_del(&cma_dev->list);
4321         mutex_unlock(&lock);
4322
4323         cma_process_remove(cma_dev);
4324         kfree(cma_dev->default_gid_type);
4325         kfree(cma_dev);
4326 }
4327
4328 static int cma_get_id_stats(struct sk_buff *skb, struct netlink_callback *cb)
4329 {
4330         struct nlmsghdr *nlh;
4331         struct rdma_cm_id_stats *id_stats;
4332         struct rdma_id_private *id_priv;
4333         struct rdma_cm_id *id = NULL;
4334         struct cma_device *cma_dev;
4335         int i_dev = 0, i_id = 0;
4336
4337         /*
4338          * We export all of the IDs as a sequence of messages.  Each
4339          * ID gets its own netlink message.
4340          */
4341         mutex_lock(&lock);
4342
4343         list_for_each_entry(cma_dev, &dev_list, list) {
4344                 if (i_dev < cb->args[0]) {
4345                         i_dev++;
4346                         continue;
4347                 }
4348
4349                 i_id = 0;
4350                 list_for_each_entry(id_priv, &cma_dev->id_list, list) {
4351                         if (i_id < cb->args[1]) {
4352                                 i_id++;
4353                                 continue;
4354                         }
4355
4356                         id_stats = ibnl_put_msg(skb, &nlh, cb->nlh->nlmsg_seq,
4357                                                 sizeof *id_stats, RDMA_NL_RDMA_CM,
4358                                                 RDMA_NL_RDMA_CM_ID_STATS,
4359                                                 NLM_F_MULTI);
4360                         if (!id_stats)
4361                                 goto out;
4362
4363                         memset(id_stats, 0, sizeof *id_stats);
4364                         id = &id_priv->id;
4365                         id_stats->node_type = id->route.addr.dev_addr.dev_type;
4366                         id_stats->port_num = id->port_num;
4367                         id_stats->bound_dev_if =
4368                                 id->route.addr.dev_addr.bound_dev_if;
4369
4370                         if (ibnl_put_attr(skb, nlh,
4371                                           rdma_addr_size(cma_src_addr(id_priv)),
4372                                           cma_src_addr(id_priv),
4373                                           RDMA_NL_RDMA_CM_ATTR_SRC_ADDR))
4374                                 goto out;
4375                         if (ibnl_put_attr(skb, nlh,
4376                                           rdma_addr_size(cma_src_addr(id_priv)),
4377                                           cma_dst_addr(id_priv),
4378                                           RDMA_NL_RDMA_CM_ATTR_DST_ADDR))
4379                                 goto out;
4380
4381                         id_stats->pid           = id_priv->owner;
4382                         id_stats->port_space    = id->ps;
4383                         id_stats->cm_state      = id_priv->state;
4384                         id_stats->qp_num        = id_priv->qp_num;
4385                         id_stats->qp_type       = id->qp_type;
4386
4387                         i_id++;
4388                 }
4389
4390                 cb->args[1] = 0;
4391                 i_dev++;
4392         }
4393
4394 out:
4395         mutex_unlock(&lock);
4396         cb->args[0] = i_dev;
4397         cb->args[1] = i_id;
4398
4399         return skb->len;
4400 }
4401
4402 static const struct ibnl_client_cbs cma_cb_table[] = {
4403         [RDMA_NL_RDMA_CM_ID_STATS] = { .dump = cma_get_id_stats,
4404                                        .module = THIS_MODULE },
4405 };
4406
4407 static int cma_init_net(struct net *net)
4408 {
4409         struct cma_pernet *pernet = cma_pernet(net);
4410
4411         idr_init(&pernet->tcp_ps);
4412         idr_init(&pernet->udp_ps);
4413         idr_init(&pernet->ipoib_ps);
4414         idr_init(&pernet->ib_ps);
4415
4416         return 0;
4417 }
4418
4419 static void cma_exit_net(struct net *net)
4420 {
4421         struct cma_pernet *pernet = cma_pernet(net);
4422
4423         idr_destroy(&pernet->tcp_ps);
4424         idr_destroy(&pernet->udp_ps);
4425         idr_destroy(&pernet->ipoib_ps);
4426         idr_destroy(&pernet->ib_ps);
4427 }
4428
4429 static struct pernet_operations cma_pernet_operations = {
4430         .init = cma_init_net,
4431         .exit = cma_exit_net,
4432         .id = &cma_pernet_id,
4433         .size = sizeof(struct cma_pernet),
4434 };
4435
4436 static int __init cma_init(void)
4437 {
4438         int ret;
4439
4440         cma_wq = alloc_ordered_workqueue("rdma_cm", WQ_MEM_RECLAIM);
4441         if (!cma_wq)
4442                 return -ENOMEM;
4443
4444         ret = register_pernet_subsys(&cma_pernet_operations);
4445         if (ret)
4446                 goto err_wq;
4447
4448         ib_sa_register_client(&sa_client);
4449         rdma_addr_register_client(&addr_client);
4450         register_netdevice_notifier(&cma_nb);
4451
4452         ret = ib_register_client(&cma_client);
4453         if (ret)
4454                 goto err;
4455
4456         if (ibnl_add_client(RDMA_NL_RDMA_CM, ARRAY_SIZE(cma_cb_table),
4457                             cma_cb_table))
4458                 pr_warn("RDMA CMA: failed to add netlink callback\n");
4459         cma_configfs_init();
4460
4461         return 0;
4462
4463 err:
4464         unregister_netdevice_notifier(&cma_nb);
4465         rdma_addr_unregister_client(&addr_client);
4466         ib_sa_unregister_client(&sa_client);
4467 err_wq:
4468         destroy_workqueue(cma_wq);
4469         return ret;
4470 }
4471
4472 static void __exit cma_cleanup(void)
4473 {
4474         cma_configfs_exit();
4475         ibnl_remove_client(RDMA_NL_RDMA_CM);
4476         ib_unregister_client(&cma_client);
4477         unregister_netdevice_notifier(&cma_nb);
4478         rdma_addr_unregister_client(&addr_client);
4479         ib_sa_unregister_client(&sa_client);
4480         unregister_pernet_subsys(&cma_pernet_operations);
4481         destroy_workqueue(cma_wq);
4482 }
4483
4484 module_init(cma_init);
4485 module_exit(cma_cleanup);